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> 13e55958c8SIoannis Angelakopoulos #include <linux/timekeeping.h> 14602cbe91SDavid Sterba #include "misc.h" 1579154b1bSChris Mason #include "ctree.h" 1679154b1bSChris Mason #include "disk-io.h" 1779154b1bSChris Mason #include "transaction.h" 18925baeddSChris Mason #include "locking.h" 19e02119d5SChris Mason #include "tree-log.h" 20733f4fbbSStefan Behrens #include "volumes.h" 218dabb742SStefan Behrens #include "dev-replace.h" 22fcebe456SJosef Bacik #include "qgroup.h" 23aac0023cSJosef Bacik #include "block-group.h" 249c343784SJosef Bacik #include "space-info.h" 25d3575156SNaohiro Aota #include "zoned.h" 2679154b1bSChris Mason 27*956504a3SJosef Bacik static struct kmem_cache *btrfs_trans_handle_cachep; 28*956504a3SJosef Bacik 29fc7cbcd4SDavid Sterba #define BTRFS_ROOT_TRANS_TAG 0 300f7d52f4SChris Mason 3161c047b5SQu Wenruo /* 3261c047b5SQu Wenruo * Transaction states and transitions 3361c047b5SQu Wenruo * 3461c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 3561c047b5SQu Wenruo * | 3661c047b5SQu Wenruo * | To next stage: 3761c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 3861c047b5SQu Wenruo * V 3961c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 4061c047b5SQu Wenruo * | 4161c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 4261c047b5SQu Wenruo * | start_transaction() variants. 4361c047b5SQu Wenruo * | 4461c047b5SQu Wenruo * | To next stage: 4561c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 4661c047b5SQu Wenruo * | transaction N 4761c047b5SQu Wenruo * V 4861c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 4961c047b5SQu Wenruo * | 5061c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 5161c047b5SQu Wenruo * | is one 5261c047b5SQu Wenruo * | 5361c047b5SQu Wenruo * | Then one of the following happes: 5461c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 5561c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 5661c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 5761c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 5861c047b5SQu Wenruo * | 5961c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 6061c047b5SQu Wenruo * | to this running transaction. 6161c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 6261c047b5SQu Wenruo * | transaction N+1. 6361c047b5SQu Wenruo * | 6461c047b5SQu Wenruo * | To next stage: 6561c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 6661c047b5SQu Wenruo * | haven been released. 6761c047b5SQu Wenruo * V 6861c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 6961c047b5SQu Wenruo * | 7061c047b5SQu Wenruo * | The heavy lifting transaction work is started. 7161c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 7261c047b5SQu Wenruo * | snapshots, running qgroups. 7361c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 7461c047b5SQu Wenruo * | trees. 7561c047b5SQu Wenruo * | 7661c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 7761c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 7861c047b5SQu Wenruo * | 7961c047b5SQu Wenruo * | To next stage: 8061c047b5SQu Wenruo * | Until all supporting trees are updated. 8161c047b5SQu Wenruo * V 8261c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 8361c047b5SQu Wenruo * | Transaction N+1 8461c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 8561c047b5SQu Wenruo * | need to write them back to disk and update | 8661c047b5SQu Wenruo * | super blocks. | 8761c047b5SQu Wenruo * | | 8861c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 8961c047b5SQu Wenruo * | start. | 9061c047b5SQu Wenruo * | All new start_transaction() calls will be | 9161c047b5SQu Wenruo * | attached to transid N+1. | 9261c047b5SQu Wenruo * | | 9361c047b5SQu Wenruo * | To next stage: | 9461c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 9561c047b5SQu Wenruo * | written to block devices | 9661c047b5SQu Wenruo * V | 9761c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 9861c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 9961c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 10061c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 10161c047b5SQu Wenruo */ 102e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1034a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 104bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 105bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1064a9d8bdeSMiao Xie __TRANS_ATTACH | 107a6d155d2SFilipe Manana __TRANS_JOIN | 108a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 109bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1104a9d8bdeSMiao Xie __TRANS_ATTACH | 1114a9d8bdeSMiao Xie __TRANS_JOIN | 112a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 113a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 114d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 115d0c2f4faSFilipe Manana __TRANS_ATTACH | 116d0c2f4faSFilipe Manana __TRANS_JOIN | 117d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 118d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 119bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1204a9d8bdeSMiao Xie __TRANS_ATTACH | 1214a9d8bdeSMiao Xie __TRANS_JOIN | 122a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 123a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1244a9d8bdeSMiao Xie }; 1254a9d8bdeSMiao Xie 126724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 12779154b1bSChris Mason { 1289b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1299b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 130a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1315c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1325c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 13381f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 13481f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1351262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 136ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 137ab8d0fc4SJeff Mahoney "pending csums is %llu", 1381262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1397785a663SFilipe Manana /* 1407785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1417785a663SFilipe Manana * because the transaction was aborted and a commit did not 1427785a663SFilipe Manana * happen (things failed before writing the new superblock 1437785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1447785a663SFilipe Manana * discard the physical locations of the block groups. 1457785a663SFilipe Manana */ 1467785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 14732da5386SDavid Sterba struct btrfs_block_group *cache; 1487785a663SFilipe Manana 1497785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 15032da5386SDavid Sterba struct btrfs_block_group, 1517785a663SFilipe Manana bg_list); 1527785a663SFilipe Manana list_del_init(&cache->bg_list); 1536b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1547785a663SFilipe Manana btrfs_put_block_group(cache); 1557785a663SFilipe Manana } 156bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1574b5faeacSDavid Sterba kfree(transaction); 15879154b1bSChris Mason } 15978fae27eSChris Mason } 16079154b1bSChris Mason 161889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 162817d52f8SJosef Bacik { 163889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 16416916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1659e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 1669e351cc8SJosef Bacik 167dfba78dcSFilipe Manana /* 168dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 169dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 170dfba78dcSFilipe Manana */ 171dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 172dfba78dcSFilipe Manana 1739e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 174d96b3424SFilipe Manana 175d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 176d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 177d96b3424SFilipe Manana 178889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1799e351cc8SJosef Bacik dirty_list) { 1809e351cc8SJosef Bacik list_del_init(&root->dirty_list); 181817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 182817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 18341e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 184370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1859e351cc8SJosef Bacik } 1862b9dbef2SJosef Bacik 1872b9dbef2SJosef Bacik /* We can free old roots now. */ 188889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 189889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 190889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1912b9dbef2SJosef Bacik struct btrfs_root, root_list); 1922b9dbef2SJosef Bacik list_del_init(&root->root_list); 193889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 194889bfa39SJosef Bacik btrfs_free_log(trans, root); 1952b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 196889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1972b9dbef2SJosef Bacik } 198889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 19927d56e62SJosef Bacik 2009e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 201817d52f8SJosef Bacik } 202817d52f8SJosef Bacik 2030860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2040860adfdSMiao Xie unsigned int type) 2050860adfdSMiao Xie { 2060860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2070860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2080860adfdSMiao Xie } 2090860adfdSMiao Xie 2100860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2110860adfdSMiao Xie unsigned int type) 2120860adfdSMiao Xie { 2130860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2140860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2150860adfdSMiao Xie } 2160860adfdSMiao Xie 2170860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2180860adfdSMiao Xie unsigned int type) 2190860adfdSMiao Xie { 2200860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2210860adfdSMiao Xie } 2220860adfdSMiao Xie 2230860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2240860adfdSMiao Xie { 2250860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 226178260b2SMiao Xie } 227178260b2SMiao Xie 228d352ac68SChris Mason /* 22979bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 23079bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 23179bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 23279bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 23379bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 234fb6dea26SJosef Bacik */ 235fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 236fb6dea26SJosef Bacik { 237fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 238fb6dea26SJosef Bacik 239fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 240fb6dea26SJosef Bacik return; 241fb6dea26SJosef Bacik 242fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 24363f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 244fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 245fb6dea26SJosef Bacik } 246fb6dea26SJosef Bacik 247fb6dea26SJosef Bacik /* 248d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 249d352ac68SChris Mason */ 2502ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2512ff7e61eSJeff Mahoney unsigned int type) 25279154b1bSChris Mason { 25379154b1bSChris Mason struct btrfs_transaction *cur_trans; 254a4abeea4SJosef Bacik 25519ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 256d43317dcSChris Mason loop: 25749b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 25884961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 25919ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26049b25e05SJeff Mahoney return -EROFS; 26149b25e05SJeff Mahoney } 26249b25e05SJeff Mahoney 26319ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 264a4abeea4SJosef Bacik if (cur_trans) { 265bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 26619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26749b25e05SJeff Mahoney return cur_trans->aborted; 268871383beSDavid Sterba } 2694a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 270178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 271178260b2SMiao Xie return -EBUSY; 272178260b2SMiao Xie } 2739b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 274a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2750860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 27619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 277e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2785a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 279a4abeea4SJosef Bacik return 0; 280a4abeea4SJosef Bacik } 28119ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 282a4abeea4SJosef Bacik 283354aa0fbSMiao Xie /* 284354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 285354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 286354aa0fbSMiao Xie */ 287354aa0fbSMiao Xie if (type == TRANS_ATTACH) 288354aa0fbSMiao Xie return -ENOENT; 289354aa0fbSMiao Xie 2904a9d8bdeSMiao Xie /* 2914a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 2924a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 2934a9d8bdeSMiao Xie */ 2944a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 2954a9d8bdeSMiao Xie 2964b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 297db5b493aSTsutomu Itoh if (!cur_trans) 298db5b493aSTsutomu Itoh return -ENOMEM; 299d43317dcSChris Mason 300e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 3015a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 302e1489b4fSIoannis Angelakopoulos 30319ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 30419ae4e81SJan Schmidt if (fs_info->running_transaction) { 305d43317dcSChris Mason /* 306d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3074a9d8bdeSMiao Xie * to redo the checks above 308d43317dcSChris Mason */ 3095a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 310e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3114b5faeacSDavid Sterba kfree(cur_trans); 312d43317dcSChris Mason goto loop; 31384961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 314e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3155a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 316e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3174b5faeacSDavid Sterba kfree(cur_trans); 3187b8b92afSJosef Bacik return -EROFS; 319a4abeea4SJosef Bacik } 320d43317dcSChris Mason 321ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 32248778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 32348778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 32413c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3250860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 32679154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 32779154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3284a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 329a4abeea4SJosef Bacik /* 330a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 331a4abeea4SJosef Bacik * commit the transaction. 332a4abeea4SJosef Bacik */ 3339b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3343204d33cSJosef Bacik cur_trans->flags = 0; 335afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 33656bec294SChris Mason 337a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 338a099d0fdSAlexandru Moise 3395c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3403368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 341d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 34220b297d6SJan Schmidt 34320b297d6SJan Schmidt /* 34420b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 34520b297d6SJan Schmidt * the log must never go across transaction boundaries. 34620b297d6SJan Schmidt */ 34720b297d6SJan Schmidt smp_mb(); 34831b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3495d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 35031b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3515d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 352fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 35320b297d6SJan Schmidt 35456bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 35556bec294SChris Mason 3563063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 357bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3589e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 359ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3601bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3612b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3621bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 363ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 364e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3652b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 366d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 367d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 36819ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 369c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 370efb0645bSJosef Bacik IO_TREE_TRANS_DIRTY_PAGES, NULL); 371fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 372fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 37319ae4e81SJan Schmidt fs_info->generation++; 37419ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 37519ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 37649b25e05SJeff Mahoney cur_trans->aborted = 0; 37719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 37815ee9bc7SJosef Bacik 37979154b1bSChris Mason return 0; 38079154b1bSChris Mason } 38179154b1bSChris Mason 382d352ac68SChris Mason /* 38392a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 38492a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 38592a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 38692a7cc42SQu Wenruo * transaction commits. 387d352ac68SChris Mason */ 3887585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3896426c7adSQu Wenruo struct btrfs_root *root, 3906426c7adSQu Wenruo int force) 3916702ed49SChris Mason { 3920b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 39303a7e111SJosef Bacik int ret = 0; 3940b246afaSJeff Mahoney 39592a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 3966426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 3974d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 3985d4f98a2SYan Zheng 3997585717fSChris Mason /* 40027cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4017585717fSChris Mason * we have the reloc mutex held now, so there 4027585717fSChris Mason * is only one writer in this function 4037585717fSChris Mason */ 40427cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4057585717fSChris Mason 40627cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4077585717fSChris Mason * they find our root->last_trans update 4087585717fSChris Mason */ 4097585717fSChris Mason smp_wmb(); 4107585717fSChris Mason 411fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4126426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 413fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 414a4abeea4SJosef Bacik return 0; 415a4abeea4SJosef Bacik } 416fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4176702ed49SChris Mason (unsigned long)root->root_key.objectid, 4186702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 419fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4207585717fSChris Mason root->last_trans = trans->transid; 4217585717fSChris Mason 4227585717fSChris Mason /* this is pretty tricky. We don't want to 4237585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4247585717fSChris Mason * unless we're really doing the first setup for this root in 4257585717fSChris Mason * this transaction. 4267585717fSChris Mason * 4277585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4287585717fSChris Mason * if we want to take the expensive mutex. 4297585717fSChris Mason * 4307585717fSChris Mason * But, we have to set root->last_trans before we 4317585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4327585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 43327cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4347585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4357585717fSChris Mason * 4367585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4377585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4387585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4397585717fSChris Mason * done before we pop in the zero below 4407585717fSChris Mason */ 44103a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 442c7548af6SChris Mason smp_mb__before_atomic(); 44327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4446702ed49SChris Mason } 44503a7e111SJosef Bacik return ret; 4466702ed49SChris Mason } 4475d4f98a2SYan Zheng 4487585717fSChris Mason 4492b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4502b9dbef2SJosef Bacik struct btrfs_root *root) 4512b9dbef2SJosef Bacik { 4520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4532b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4542b9dbef2SJosef Bacik 4552b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4562b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4572b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4582b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4592b9dbef2SJosef Bacik 4602b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 461fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 462fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4632b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4642b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 465fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4662b9dbef2SJosef Bacik } 4672b9dbef2SJosef Bacik 4687585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4697585717fSChris Mason struct btrfs_root *root) 4707585717fSChris Mason { 4710b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4721409e6ccSJosef Bacik int ret; 4730b246afaSJeff Mahoney 47492a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4757585717fSChris Mason return 0; 4767585717fSChris Mason 4777585717fSChris Mason /* 47827cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4797585717fSChris Mason * and barriers 4807585717fSChris Mason */ 4817585717fSChris Mason smp_rmb(); 4827585717fSChris Mason if (root->last_trans == trans->transid && 48327cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4847585717fSChris Mason return 0; 4857585717fSChris Mason 4860b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4871409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4880b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4897585717fSChris Mason 4901409e6ccSJosef Bacik return ret; 4917585717fSChris Mason } 4927585717fSChris Mason 4934a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 4944a9d8bdeSMiao Xie { 4953296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 496501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 497bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 4984a9d8bdeSMiao Xie } 4994a9d8bdeSMiao Xie 500d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 501d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 502d352ac68SChris Mason * transaction might not be fully on disk. 503d352ac68SChris Mason */ 5042ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 50579154b1bSChris Mason { 506f9295749SChris Mason struct btrfs_transaction *cur_trans; 50779154b1bSChris Mason 5080b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5090b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5104a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5119b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5120b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51372d63ed6SLi Zefan 5143e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5150b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 516501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 517bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 518724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 519a4abeea4SJosef Bacik } else { 5200b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 521f9295749SChris Mason } 52237d1aeeeSChris Mason } 52337d1aeeeSChris Mason 5242ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 52537d1aeeeSChris Mason { 5260b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 527a4abeea4SJosef Bacik return 0; 528a4abeea4SJosef Bacik 52992e2f7e3SNikolay Borisov if (type == TRANS_START) 530a4abeea4SJosef Bacik return 1; 531a4abeea4SJosef Bacik 532a22285a6SYan, Zheng return 0; 533a22285a6SYan, Zheng } 534a22285a6SYan, Zheng 53520dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 53620dd2cbfSMiao Xie { 5370b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5380b246afaSJeff Mahoney 5390b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 54092a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 54120dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 54220dd2cbfSMiao Xie root->reloc_root) 54320dd2cbfSMiao Xie return false; 54420dd2cbfSMiao Xie 54520dd2cbfSMiao Xie return true; 54620dd2cbfSMiao Xie } 54720dd2cbfSMiao Xie 54808e007d2SMiao Xie static struct btrfs_trans_handle * 5495aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 550003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 551003d7c59SJeff Mahoney bool enforce_qgroups) 552a22285a6SYan, Zheng { 5530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 554ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 555a22285a6SYan, Zheng struct btrfs_trans_handle *h; 556a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 557b5009945SJosef Bacik u64 num_bytes = 0; 558c5567237SArne Jansen u64 qgroup_reserved = 0; 55920dd2cbfSMiao Xie bool reloc_reserved = false; 5609c343784SJosef Bacik bool do_chunk_alloc = false; 56120dd2cbfSMiao Xie int ret; 562acce952bSliubo 56384961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 564acce952bSliubo return ERR_PTR(-EROFS); 5652a1eb461SJosef Bacik 56646c4e71eSFilipe Manana if (current->journal_info) { 5670860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5682a1eb461SJosef Bacik h = current->journal_info; 569b50fff81SDavid Sterba refcount_inc(&h->use_count); 570b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5712a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5722a1eb461SJosef Bacik h->block_rsv = NULL; 5732a1eb461SJosef Bacik goto got_it; 5742a1eb461SJosef Bacik } 575b5009945SJosef Bacik 576b5009945SJosef Bacik /* 577b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 578b5009945SJosef Bacik * the appropriate flushing if need be. 579b5009945SJosef Bacik */ 580003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 581ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 582ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 583ba2c4d4eSJosef Bacik 5840b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 585733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 586003d7c59SJeff Mahoney enforce_qgroups); 587c5567237SArne Jansen if (ret) 588c5567237SArne Jansen return ERR_PTR(ret); 589c5567237SArne Jansen 590ba2c4d4eSJosef Bacik /* 591ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 592ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 593ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 594ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 595ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 596ba2c4d4eSJosef Bacik */ 5972bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 5987f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 599748f553cSDavid Sterba btrfs_block_rsv_full(delayed_refs_rsv) == 0) { 600ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 601ba2c4d4eSJosef Bacik num_bytes <<= 1; 602ba2c4d4eSJosef Bacik } 603ba2c4d4eSJosef Bacik 60420dd2cbfSMiao Xie /* 60520dd2cbfSMiao Xie * Do the reservation for the relocation root creation 60620dd2cbfSMiao Xie */ 607ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6080b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 60920dd2cbfSMiao Xie reloc_reserved = true; 61020dd2cbfSMiao Xie } 61120dd2cbfSMiao Xie 6129270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 613ba2c4d4eSJosef Bacik if (ret) 614ba2c4d4eSJosef Bacik goto reserve_fail; 615ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 616ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 617ba2c4d4eSJosef Bacik delayed_refs_bytes); 618ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 619ba2c4d4eSJosef Bacik } 6209c343784SJosef Bacik 6219c343784SJosef Bacik if (rsv->space_info->force_alloc) 6229c343784SJosef Bacik do_chunk_alloc = true; 623ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 624748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 625ba2c4d4eSJosef Bacik /* 626ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 627ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 628ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 629ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 630ba2c4d4eSJosef Bacik * here. 631ba2c4d4eSJosef Bacik */ 632ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 633b5009945SJosef Bacik if (ret) 634843fcf35SMiao Xie goto reserve_fail; 635b5009945SJosef Bacik } 636a22285a6SYan, Zheng again: 637f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 638843fcf35SMiao Xie if (!h) { 639843fcf35SMiao Xie ret = -ENOMEM; 640843fcf35SMiao Xie goto alloc_fail; 641843fcf35SMiao Xie } 642a22285a6SYan, Zheng 64398114659SJosef Bacik /* 64498114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 64598114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 64698114659SJosef Bacik * because we're already holding a ref. We need this because we could 64798114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 64898114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 649354aa0fbSMiao Xie * 650354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 651354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 65298114659SJosef Bacik */ 6530860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6540b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 655b2b5ef5cSJan Kara 6562ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6572ff7e61eSJeff Mahoney wait_current_trans(fs_info); 658a22285a6SYan, Zheng 659a4abeea4SJosef Bacik do { 6602ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 661178260b2SMiao Xie if (ret == -EBUSY) { 6622ff7e61eSJeff Mahoney wait_current_trans(fs_info); 663a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 664a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 665178260b2SMiao Xie ret = -ENOENT; 666178260b2SMiao Xie } 667a4abeea4SJosef Bacik } while (ret == -EBUSY); 668a4abeea4SJosef Bacik 669a43f7f82SLiu Bo if (ret < 0) 670843fcf35SMiao Xie goto join_fail; 6710f7d52f4SChris Mason 6720b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 673a22285a6SYan, Zheng 674a22285a6SYan, Zheng h->transid = cur_trans->transid; 675a22285a6SYan, Zheng h->transaction = cur_trans; 676b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 67764b63580SJeff Mahoney h->fs_info = root->fs_info; 6787174109cSQu Wenruo 679a698d075SMiao Xie h->type = type; 680ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 681b7ec40d7SChris Mason 682a22285a6SYan, Zheng smp_mb(); 6833296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 6842ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 685abdd2e80SFilipe Manana current->journal_info = h; 6863a45bb20SJeff Mahoney btrfs_commit_transaction(h); 687a22285a6SYan, Zheng goto again; 688a22285a6SYan, Zheng } 6899ed74f2dSJosef Bacik 690b5009945SJosef Bacik if (num_bytes) { 6910b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 6922bcc0328SLiu Bo h->transid, num_bytes, 1); 6930b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 694b5009945SJosef Bacik h->bytes_reserved = num_bytes; 69520dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 696a22285a6SYan, Zheng } 697a22285a6SYan, Zheng 6982a1eb461SJosef Bacik got_it: 699bcf3a3e7SNikolay Borisov if (!current->journal_info) 700a22285a6SYan, Zheng current->journal_info = h; 701fcc99734SQu Wenruo 702fcc99734SQu Wenruo /* 7039c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7049c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7059c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7069c343784SJosef Bacik * value here. 7079c343784SJosef Bacik */ 7089c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7099c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7109c343784SJosef Bacik 7119c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7129c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7139c343784SJosef Bacik } 7149c343784SJosef Bacik 7159c343784SJosef Bacik /* 716fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 717fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 718fcc99734SQu Wenruo * transaction. 719fcc99734SQu Wenruo * 720fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 721fcc99734SQu Wenruo * or we can deadlock. 722fcc99734SQu Wenruo */ 72368075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 72468075ea8SJosef Bacik if (ret) { 72568075ea8SJosef Bacik /* 72668075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 72768075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 72868075ea8SJosef Bacik * not just freed. 72968075ea8SJosef Bacik */ 73068075ea8SJosef Bacik btrfs_end_transaction(h); 73168075ea8SJosef Bacik return ERR_PTR(ret); 73268075ea8SJosef Bacik } 733fcc99734SQu Wenruo 73479154b1bSChris Mason return h; 735843fcf35SMiao Xie 736843fcf35SMiao Xie join_fail: 7370860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7380b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 739843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 740843fcf35SMiao Xie alloc_fail: 741843fcf35SMiao Xie if (num_bytes) 7422ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 74363f018beSNikolay Borisov num_bytes, NULL); 744843fcf35SMiao Xie reserve_fail: 745733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 746843fcf35SMiao Xie return ERR_PTR(ret); 74779154b1bSChris Mason } 74879154b1bSChris Mason 749f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7505aed1dd8SAlexandru Moise unsigned int num_items) 751f9295749SChris Mason { 75208e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 753003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 754f9295749SChris Mason } 755003d7c59SJeff Mahoney 7568eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7578eab77ffSFilipe Manana struct btrfs_root *root, 7587f9fe614SJosef Bacik unsigned int num_items) 7598eab77ffSFilipe Manana { 7607f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7617f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7628eab77ffSFilipe Manana } 7638407aa46SMiao Xie 7647a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 765f9295749SChris Mason { 766003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 767003d7c59SJeff Mahoney true); 768f9295749SChris Mason } 769f9295749SChris Mason 7708d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7710af3d00bSJosef Bacik { 772575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 773003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7740af3d00bSJosef Bacik } 7750af3d00bSJosef Bacik 776d4edf39bSMiao Xie /* 777a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 778a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 779a6d155d2SFilipe Manana */ 780a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 781a6d155d2SFilipe Manana { 782a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 783a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 784a6d155d2SFilipe Manana } 785a6d155d2SFilipe Manana 786a6d155d2SFilipe Manana /* 787d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 788d4edf39bSMiao Xie * 789d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 790d4edf39bSMiao Xie * don't want to start a new one. 791d4edf39bSMiao Xie * 792d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 793d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 794d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 795d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 796d4edf39bSMiao Xie * invoke 797d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 798d4edf39bSMiao Xie */ 799354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 80060376ce4SJosef Bacik { 801575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 802003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 80360376ce4SJosef Bacik } 80460376ce4SJosef Bacik 805d4edf39bSMiao Xie /* 80690b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 807d4edf39bSMiao Xie * 80852042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 809d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 810d4edf39bSMiao Xie * complete. 811d4edf39bSMiao Xie */ 812d4edf39bSMiao Xie struct btrfs_trans_handle * 813d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 814d4edf39bSMiao Xie { 815d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 816d4edf39bSMiao Xie 817575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 818003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8198d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8202ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 821d4edf39bSMiao Xie 822d4edf39bSMiao Xie return trans; 823d4edf39bSMiao Xie } 824d4edf39bSMiao Xie 825d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 826d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 827d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 82889ce8a63SChris Mason { 8295fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8305fd76bf3SOmar Sandoval u64 transid = commit->transid; 8315fd76bf3SOmar Sandoval bool put = false; 8325fd76bf3SOmar Sandoval 8333e738c53SIoannis Angelakopoulos /* 8343e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8353e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8363e738c53SIoannis Angelakopoulos */ 8373e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8383e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8393e738c53SIoannis Angelakopoulos else 8403e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8413e738c53SIoannis Angelakopoulos 8425fd76bf3SOmar Sandoval while (1) { 843d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8445fd76bf3SOmar Sandoval if (put) 8455fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8465fd76bf3SOmar Sandoval 8475fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8485fd76bf3SOmar Sandoval break; 8495fd76bf3SOmar Sandoval 8505fd76bf3SOmar Sandoval /* 8515fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8525fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8535fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8545fd76bf3SOmar Sandoval * transaction. Wait for those. 8555fd76bf3SOmar Sandoval */ 8565fd76bf3SOmar Sandoval 8575fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8585fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8595fd76bf3SOmar Sandoval struct btrfs_transaction, 8605fd76bf3SOmar Sandoval list); 8615fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8625fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8635fd76bf3SOmar Sandoval break; 8645fd76bf3SOmar Sandoval } 8655fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8665fd76bf3SOmar Sandoval put = true; 8675fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8685fd76bf3SOmar Sandoval } 86989ce8a63SChris Mason } 87089ce8a63SChris Mason 8712ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 87246204592SSage Weil { 87346204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8748cd2807fSMiao Xie int ret = 0; 87546204592SSage Weil 87646204592SSage Weil if (transid) { 8770b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 878a4abeea4SJosef Bacik goto out; 87946204592SSage Weil 88046204592SSage Weil /* find specified transaction */ 8810b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8820b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 88346204592SSage Weil if (t->transid == transid) { 88446204592SSage Weil cur_trans = t; 8859b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8868cd2807fSMiao Xie ret = 0; 88746204592SSage Weil break; 88846204592SSage Weil } 8898cd2807fSMiao Xie if (t->transid > transid) { 8908cd2807fSMiao Xie ret = 0; 89146204592SSage Weil break; 89246204592SSage Weil } 8938cd2807fSMiao Xie } 8940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 89542383020SSage Weil 89642383020SSage Weil /* 89742383020SSage Weil * The specified transaction doesn't exist, or we 89842383020SSage Weil * raced with btrfs_commit_transaction 89942383020SSage Weil */ 90042383020SSage Weil if (!cur_trans) { 9010b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 90242383020SSage Weil ret = -EINVAL; 9038cd2807fSMiao Xie goto out; 90442383020SSage Weil } 90546204592SSage Weil } else { 90646204592SSage Weil /* find newest transaction that is committing | committed */ 9070b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9080b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 90946204592SSage Weil list) { 9104a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9114a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9123473f3c0SJosef Bacik break; 91346204592SSage Weil cur_trans = t; 9149b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 91546204592SSage Weil break; 91646204592SSage Weil } 91746204592SSage Weil } 9180b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 91946204592SSage Weil if (!cur_trans) 920a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 92146204592SSage Weil } 92246204592SSage Weil 923d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 924724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 925a4abeea4SJosef Bacik out: 92646204592SSage Weil return ret; 92746204592SSage Weil } 92846204592SSage Weil 9292ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 93037d1aeeeSChris Mason { 9312ff7e61eSJeff Mahoney wait_current_trans(fs_info); 93237d1aeeeSChris Mason } 93337d1aeeeSChris Mason 9348a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9358929ecfaSYan, Zheng { 9362ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9370b246afaSJeff Mahoney 93864403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9398a8f4deaSNikolay Borisov return true; 94036ba022aSJosef Bacik 9412ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9428929ecfaSYan, Zheng } 9438929ecfaSYan, Zheng 944a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9458929ecfaSYan, Zheng { 9468929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9478929ecfaSYan, Zheng 9483296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 949e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 950a2633b6aSNikolay Borisov return true; 9518929ecfaSYan, Zheng 9522ff7e61eSJeff Mahoney return should_end_transaction(trans); 9538929ecfaSYan, Zheng } 9548929ecfaSYan, Zheng 955dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 956dc60c525SNikolay Borisov 9570e34693fSNikolay Borisov { 958dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 959dc60c525SNikolay Borisov 9600e34693fSNikolay Borisov if (!trans->block_rsv) { 9610e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9620e34693fSNikolay Borisov return; 9630e34693fSNikolay Borisov } 9640e34693fSNikolay Borisov 9650e34693fSNikolay Borisov if (!trans->bytes_reserved) 9660e34693fSNikolay Borisov return; 9670e34693fSNikolay Borisov 9680e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9690e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9700e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9710e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 97263f018beSNikolay Borisov trans->bytes_reserved, NULL); 9730e34693fSNikolay Borisov trans->bytes_reserved = 0; 9740e34693fSNikolay Borisov } 9750e34693fSNikolay Borisov 97689ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9773a45bb20SJeff Mahoney int throttle) 97879154b1bSChris Mason { 9793a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9808929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9814edc2ca3SDave Jones int err = 0; 982d6e4a428SChris Mason 983b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 984b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9852a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9862a1eb461SJosef Bacik return 0; 9872a1eb461SJosef Bacik } 9882a1eb461SJosef Bacik 989dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9904c13d758SJosef Bacik trans->block_rsv = NULL; 991c5567237SArne Jansen 9926c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 993ea658badSJosef Bacik 9944fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9954fbcdf66SFilipe Manana 9960860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9970b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9986df7881aSJosef Bacik 9998929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 100013c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 100113c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10020860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 100389ce8a63SChris Mason 1004093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1005e1489b4fSIoannis Angelakopoulos 10065a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1007e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1008e1489b4fSIoannis Angelakopoulos 1009724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10109ed74f2dSJosef Bacik 10119ed74f2dSJosef Bacik if (current->journal_info == trans) 10129ed74f2dSJosef Bacik current->journal_info = NULL; 1013ab78c84dSChris Mason 101424bbcf04SYan, Zheng if (throttle) 10152ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 101624bbcf04SYan, Zheng 101784961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10184e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1019fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1020fbabd4a3SJosef Bacik err = trans->aborted; 1021fbabd4a3SJosef Bacik else 1022fbabd4a3SJosef Bacik err = -EROFS; 10234e121c06SJosef Bacik } 102449b25e05SJeff Mahoney 10254edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10264edc2ca3SDave Jones return err; 102779154b1bSChris Mason } 102879154b1bSChris Mason 10293a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 103089ce8a63SChris Mason { 10313a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 103289ce8a63SChris Mason } 103389ce8a63SChris Mason 10343a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 103589ce8a63SChris Mason { 10363a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 103716cdcec7SMiao Xie } 103816cdcec7SMiao Xie 1039d352ac68SChris Mason /* 1040d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1041d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1042690587d1SChris Mason * those extents are sent to disk but does not wait on them 1043d352ac68SChris Mason */ 10442ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10458cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 104679154b1bSChris Mason { 1047777e6bd7SChris Mason int err = 0; 10487c4452b9SChris Mason int werr = 0; 10490b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1050e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1051777e6bd7SChris Mason u64 start = 0; 10525f39d397SChris Mason u64 end; 10537c4452b9SChris Mason 10546300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10551728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1056e6138876SJosef Bacik mark, &cached_state)) { 1057663dfbb0SFilipe Manana bool wait_writeback = false; 1058663dfbb0SFilipe Manana 1059663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1060663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1061210aa277SDavid Sterba mark, &cached_state); 1062663dfbb0SFilipe Manana /* 1063663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1064663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1065663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1066663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1067bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1068bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1069bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1070bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1071bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1072663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 107341e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1074663dfbb0SFilipe Manana */ 1075663dfbb0SFilipe Manana if (err == -ENOMEM) { 1076663dfbb0SFilipe Manana err = 0; 1077663dfbb0SFilipe Manana wait_writeback = true; 1078663dfbb0SFilipe Manana } 1079663dfbb0SFilipe Manana if (!err) 10801728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10817c4452b9SChris Mason if (err) 10827c4452b9SChris Mason werr = err; 1083663dfbb0SFilipe Manana else if (wait_writeback) 1084663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1085e38e2ed7SFilipe Manana free_extent_state(cached_state); 1086663dfbb0SFilipe Manana cached_state = NULL; 10871728366eSJosef Bacik cond_resched(); 10881728366eSJosef Bacik start = end + 1; 10897c4452b9SChris Mason } 10906300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1091690587d1SChris Mason return werr; 1092690587d1SChris Mason } 1093690587d1SChris Mason 1094690587d1SChris Mason /* 1095690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1096690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1097690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1098690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1099690587d1SChris Mason */ 1100bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1101bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1102690587d1SChris Mason { 1103690587d1SChris Mason int err = 0; 1104690587d1SChris Mason int werr = 0; 11050b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1106e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1107690587d1SChris Mason u64 start = 0; 1108690587d1SChris Mason u64 end; 1109690587d1SChris Mason 11101728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1111e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1112663dfbb0SFilipe Manana /* 1113663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1114663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1115663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1116663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1117663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 111841e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1119663dfbb0SFilipe Manana */ 1120663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1121bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1122663dfbb0SFilipe Manana if (err == -ENOMEM) 1123663dfbb0SFilipe Manana err = 0; 1124663dfbb0SFilipe Manana if (!err) 11251728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1126777e6bd7SChris Mason if (err) 1127777e6bd7SChris Mason werr = err; 1128e38e2ed7SFilipe Manana free_extent_state(cached_state); 1129e38e2ed7SFilipe Manana cached_state = NULL; 1130777e6bd7SChris Mason cond_resched(); 11311728366eSJosef Bacik start = end + 1; 1132777e6bd7SChris Mason } 11337c4452b9SChris Mason if (err) 11347c4452b9SChris Mason werr = err; 1135bf89d38fSJeff Mahoney return werr; 1136bf89d38fSJeff Mahoney } 1137656f30dbSFilipe Manana 1138b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1139bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1140bf89d38fSJeff Mahoney { 1141bf89d38fSJeff Mahoney bool errors = false; 1142bf89d38fSJeff Mahoney int err; 1143bf89d38fSJeff Mahoney 1144bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1145bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1146bf89d38fSJeff Mahoney errors = true; 1147bf89d38fSJeff Mahoney 1148bf89d38fSJeff Mahoney if (errors && !err) 1149bf89d38fSJeff Mahoney err = -EIO; 1150bf89d38fSJeff Mahoney return err; 1151bf89d38fSJeff Mahoney } 1152bf89d38fSJeff Mahoney 1153bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1154bf89d38fSJeff Mahoney { 1155bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1156bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1157bf89d38fSJeff Mahoney bool errors = false; 1158bf89d38fSJeff Mahoney int err; 1159bf89d38fSJeff Mahoney 1160bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1161bf89d38fSJeff Mahoney 1162bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1163656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11640b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1165656f30dbSFilipe Manana errors = true; 1166656f30dbSFilipe Manana 1167656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11680b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1169656f30dbSFilipe Manana errors = true; 1170656f30dbSFilipe Manana 1171bf89d38fSJeff Mahoney if (errors && !err) 1172bf89d38fSJeff Mahoney err = -EIO; 1173bf89d38fSJeff Mahoney return err; 117479154b1bSChris Mason } 117579154b1bSChris Mason 1176690587d1SChris Mason /* 1177c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1178c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1179c9b577c0SNikolay Borisov * log commit. 1180c9b577c0SNikolay Borisov * 1181c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1182690587d1SChris Mason */ 118370458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1184d0c803c4SChris Mason { 1185663dfbb0SFilipe Manana int ret; 1186c9b577c0SNikolay Borisov int ret2; 1187c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 118870458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1189c9b577c0SNikolay Borisov struct blk_plug plug; 1190663dfbb0SFilipe Manana 1191c9b577c0SNikolay Borisov blk_start_plug(&plug); 1192c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1193c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1194c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1195c9b577c0SNikolay Borisov 119641e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1197663dfbb0SFilipe Manana 1198c9b577c0SNikolay Borisov if (ret) 1199663dfbb0SFilipe Manana return ret; 1200c9b577c0SNikolay Borisov else if (ret2) 1201c9b577c0SNikolay Borisov return ret2; 1202c9b577c0SNikolay Borisov else 1203c9b577c0SNikolay Borisov return 0; 1204d0c803c4SChris Mason } 1205d0c803c4SChris Mason 1206d352ac68SChris Mason /* 1207d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1208d352ac68SChris Mason * 1209d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1210d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1211d352ac68SChris Mason * allocation tree. 1212d352ac68SChris Mason * 1213d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1214d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1215d352ac68SChris Mason */ 12160b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 121779154b1bSChris Mason struct btrfs_root *root) 121879154b1bSChris Mason { 121979154b1bSChris Mason int ret; 12200b86a832SChris Mason u64 old_root_bytenr; 122186b9f2ecSYan, Zheng u64 old_root_used; 12220b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12230b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 122479154b1bSChris Mason 122586b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 122656bec294SChris Mason 122779154b1bSChris Mason while (1) { 12280b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 122986b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1230ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 123179154b1bSChris Mason break; 123287ef2bb4SChris Mason 12335d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 123479154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12350b86a832SChris Mason &root->root_key, 12360b86a832SChris Mason &root->root_item); 123749b25e05SJeff Mahoney if (ret) 123849b25e05SJeff Mahoney return ret; 123956bec294SChris Mason 124086b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1241e7070be1SJosef Bacik } 1242276e680dSYan Zheng 12430b86a832SChris Mason return 0; 12440b86a832SChris Mason } 12450b86a832SChris Mason 1246d352ac68SChris Mason /* 1247d352ac68SChris Mason * update all the cowonly tree roots on disk 124849b25e05SJeff Mahoney * 124949b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 125049b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 125149b25e05SJeff Mahoney * to clean up the delayed refs. 1252d352ac68SChris Mason */ 12539386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12540b86a832SChris Mason { 12559386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1256ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12571bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12580b86a832SChris Mason struct list_head *next; 125984234f3aSYan Zheng struct extent_buffer *eb; 126056bec294SChris Mason int ret; 126184234f3aSYan Zheng 1262dfba78dcSFilipe Manana /* 1263dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1264dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1265dfba78dcSFilipe Manana */ 1266dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1267dfba78dcSFilipe Manana 126884234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 126949b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12709631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 127184234f3aSYan Zheng btrfs_tree_unlock(eb); 127284234f3aSYan Zheng free_extent_buffer(eb); 12730b86a832SChris Mason 127449b25e05SJeff Mahoney if (ret) 127549b25e05SJeff Mahoney return ret; 127649b25e05SJeff Mahoney 1277196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1278c16ce190SJosef Bacik if (ret) 1279c16ce190SJosef Bacik return ret; 12802b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1281c16ce190SJosef Bacik if (ret) 1282c16ce190SJosef Bacik return ret; 1283280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1284c16ce190SJosef Bacik if (ret) 1285c16ce190SJosef Bacik return ret; 1286546adb0dSJan Schmidt 1287bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1288dcdf7f6dSJosef Bacik if (ret) 1289dcdf7f6dSJosef Bacik return ret; 1290dcdf7f6dSJosef Bacik 1291ea526d18SJosef Bacik again: 12920b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12932ff7e61eSJeff Mahoney struct btrfs_root *root; 12940b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12950b86a832SChris Mason list_del_init(next); 12960b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1297e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 129887ef2bb4SChris Mason 12999e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13009e351cc8SJosef Bacik &trans->transaction->switch_commits); 130149b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 130249b25e05SJeff Mahoney if (ret) 130349b25e05SJeff Mahoney return ret; 1304488bc2a2SJosef Bacik } 1305488bc2a2SJosef Bacik 1306488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1307c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1308ea526d18SJosef Bacik if (ret) 1309ea526d18SJosef Bacik return ret; 1310276e680dSYan Zheng 13111bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13125742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1313ea526d18SJosef Bacik if (ret) 1314ea526d18SJosef Bacik return ret; 1315488bc2a2SJosef Bacik 1316488bc2a2SJosef Bacik /* 1317488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1318488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1319488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1320488bc2a2SJosef Bacik * everything gets run. 1321488bc2a2SJosef Bacik */ 1322c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1323ea526d18SJosef Bacik if (ret) 1324ea526d18SJosef Bacik return ret; 1325ea526d18SJosef Bacik } 1326ea526d18SJosef Bacik 1327ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1328ea526d18SJosef Bacik goto again; 1329ea526d18SJosef Bacik 13309f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13319f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13329f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13338dabb742SStefan Behrens 133479154b1bSChris Mason return 0; 133579154b1bSChris Mason } 133679154b1bSChris Mason 1337d352ac68SChris Mason /* 1338b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1339b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1340b4be6aefSJosef Bacik */ 1341b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1342b4be6aefSJosef Bacik { 1343b4be6aefSJosef Bacik /* 1344b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1345b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1346b4be6aefSJosef Bacik * up. 1347b4be6aefSJosef Bacik */ 1348b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1349b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1350b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1351b4be6aefSJosef Bacik struct btrfs_root, 1352b4be6aefSJosef Bacik root_list); 1353b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1354b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1355b4be6aefSJosef Bacik return; 1356b4be6aefSJosef Bacik } 1357b4be6aefSJosef Bacik } 1358b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1359b4be6aefSJosef Bacik 1360b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1361b4be6aefSJosef Bacik } 1362b4be6aefSJosef Bacik 1363b4be6aefSJosef Bacik /* 1364d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1365d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1366d352ac68SChris Mason * be deleted 1367d352ac68SChris Mason */ 1368cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13695eda7b5eSChris Mason { 13700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13710b246afaSJeff Mahoney 13720b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1373dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1374dc9492c1SJosef Bacik btrfs_grab_root(root); 1375b4be6aefSJosef Bacik 1376b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1377b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1378b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1379b4be6aefSJosef Bacik else 13800b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1381dc9492c1SJosef Bacik } 13820b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13835eda7b5eSChris Mason } 13845eda7b5eSChris Mason 1385d352ac68SChris Mason /* 1386dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1387dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1388d352ac68SChris Mason */ 13897e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13900f7d52f4SChris Mason { 13917e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1392fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1393fc7cbcd4SDavid Sterba int i; 1394fc7cbcd4SDavid Sterba int ret; 139554aa1f4dSChris Mason 1396dfba78dcSFilipe Manana /* 1397dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1398dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1399dfba78dcSFilipe Manana */ 1400dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1401dfba78dcSFilipe Manana 1402fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1403fc7cbcd4SDavid Sterba while (1) { 1404fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1405fc7cbcd4SDavid Sterba (void **)gang, 0, 1406fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1407fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1408fc7cbcd4SDavid Sterba if (ret == 0) 1409fc7cbcd4SDavid Sterba break; 1410fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1411fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1412fc7cbcd4SDavid Sterba int ret2; 14134f4317c1SJosef Bacik 1414dfba78dcSFilipe Manana /* 1415dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1416dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1417dfba78dcSFilipe Manana */ 1418dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1419dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1420dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1421dfba78dcSFilipe Manana 1422fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14232619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14240f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1425fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 142631153d81SYan Zheng 1427e02119d5SChris Mason btrfs_free_log(trans, root); 1428fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1429fc7cbcd4SDavid Sterba if (ret2) 1430fc7cbcd4SDavid Sterba return ret2; 1431e02119d5SChris Mason 1432fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 143327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1434c7548af6SChris Mason smp_mb__after_atomic(); 1435f1ebcc74SLiu Bo 1436978d910dSYan Zheng if (root->commit_root != root->node) { 14379e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14389e351cc8SJosef Bacik &trans->transaction->switch_commits); 1439fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1440fc7cbcd4SDavid Sterba root->node); 1441978d910dSYan Zheng } 144231153d81SYan Zheng 1443fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1444fc7cbcd4SDavid Sterba &root->root_key, 1445fc7cbcd4SDavid Sterba &root->root_item); 1446fc7cbcd4SDavid Sterba if (ret2) 1447fc7cbcd4SDavid Sterba return ret2; 1448fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1449733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14500f7d52f4SChris Mason } 1451fc7cbcd4SDavid Sterba } 1452fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14534f4317c1SJosef Bacik return 0; 14540f7d52f4SChris Mason } 14550f7d52f4SChris Mason 1456d352ac68SChris Mason /* 1457de78b51aSEric Sandeen * defrag a given btree. 1458de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1459d352ac68SChris Mason */ 1460de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1461e9d0b13bSChris Mason { 1462e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1463e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14648929ecfaSYan, Zheng int ret; 1465e9d0b13bSChris Mason 146627cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1467e9d0b13bSChris Mason return 0; 14688929ecfaSYan, Zheng 14696b80053dSChris Mason while (1) { 14708929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14716819703fSDavid Sterba if (IS_ERR(trans)) { 14726819703fSDavid Sterba ret = PTR_ERR(trans); 14736819703fSDavid Sterba break; 14746819703fSDavid Sterba } 14758929ecfaSYan, Zheng 1476de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14778929ecfaSYan, Zheng 14783a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14792ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1480e9d0b13bSChris Mason cond_resched(); 1481e9d0b13bSChris Mason 1482ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1483e9d0b13bSChris Mason break; 1484210549ebSDavid Sterba 1485ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1486ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1487210549ebSDavid Sterba ret = -EAGAIN; 1488210549ebSDavid Sterba break; 1489210549ebSDavid Sterba } 1490e9d0b13bSChris Mason } 149127cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14928929ecfaSYan, Zheng return ret; 1493e9d0b13bSChris Mason } 1494e9d0b13bSChris Mason 1495d352ac68SChris Mason /* 14966426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14976426c7adSQu Wenruo * 14986426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14996426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 15006426c7adSQu Wenruo * qgroup works. 15016426c7adSQu Wenruo */ 15026426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15036426c7adSQu Wenruo struct btrfs_root *src, 15046426c7adSQu Wenruo struct btrfs_root *parent, 15056426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15066426c7adSQu Wenruo u64 dst_objectid) 15076426c7adSQu Wenruo { 15086426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15096426c7adSQu Wenruo int ret; 15106426c7adSQu Wenruo 15116426c7adSQu Wenruo /* 15126426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15136426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15146426c7adSQu Wenruo * kick in anyway. 15156426c7adSQu Wenruo */ 15169ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15176426c7adSQu Wenruo return 0; 15186426c7adSQu Wenruo 15196426c7adSQu Wenruo /* 152052042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15214d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15224d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15234d31778aSQu Wenruo * item. 15244d31778aSQu Wenruo */ 15251c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15261c442d22SJosef Bacik if (ret) 15271c442d22SJosef Bacik return ret; 15284d31778aSQu Wenruo 15294d31778aSQu Wenruo /* 15302a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15312a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15322a4d84c1SJosef Bacik * 15332a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15342a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15352a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15362a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15372a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15382a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15392a4d84c1SJosef Bacik */ 15402a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15412a4d84c1SJosef Bacik if (ret) { 15422a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 154344365827SNaohiro Aota return ret; 15442a4d84c1SJosef Bacik } 15452a4d84c1SJosef Bacik 15467e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15476426c7adSQu Wenruo if (ret) 15486426c7adSQu Wenruo goto out; 1549460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15506426c7adSQu Wenruo if (ret < 0) 15516426c7adSQu Wenruo goto out; 15526426c7adSQu Wenruo 15536426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1554a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15556426c7adSQu Wenruo inherit); 15566426c7adSQu Wenruo if (ret < 0) 15576426c7adSQu Wenruo goto out; 15586426c7adSQu Wenruo 15596426c7adSQu Wenruo /* 15606426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15616426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15626426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15636426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15646426c7adSQu Wenruo * 2) write all btree blocks onto disk 15656426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15666426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15676426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15686426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15696426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15706426c7adSQu Wenruo */ 15719386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15726426c7adSQu Wenruo if (ret) 15736426c7adSQu Wenruo goto out; 1574889bfa39SJosef Bacik switch_commit_roots(trans); 157570458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15766426c7adSQu Wenruo if (ret) 1577f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15786426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15796426c7adSQu Wenruo 15806426c7adSQu Wenruo out: 15816426c7adSQu Wenruo /* 15826426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15836426c7adSQu Wenruo * last_trans == cur_transid. 15846426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15856426c7adSQu Wenruo * insert_dir_item() 15866426c7adSQu Wenruo */ 15876426c7adSQu Wenruo if (!ret) 15881c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15896426c7adSQu Wenruo return ret; 15906426c7adSQu Wenruo } 15916426c7adSQu Wenruo 15926426c7adSQu Wenruo /* 1593d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1594aec8030aSMiao Xie * transaction commit. This does the actual creation. 1595aec8030aSMiao Xie * 1596aec8030aSMiao Xie * Note: 1597aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1598aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1599aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1600d352ac68SChris Mason */ 160180b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16023063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16033063d29fSChris Mason { 160408d50ca3SNikolay Borisov 160508d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16063063d29fSChris Mason struct btrfs_key key; 160780b6794dSChris Mason struct btrfs_root_item *new_root_item; 16083063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16093063d29fSChris Mason struct btrfs_root *root = pending->root; 16106bdb72deSSage Weil struct btrfs_root *parent_root; 161198c9942aSLiu Bo struct btrfs_block_rsv *rsv; 16126bdb72deSSage Weil struct inode *parent_inode; 161342874b3dSMiao Xie struct btrfs_path *path; 161442874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1615a22285a6SYan, Zheng struct dentry *dentry; 16163063d29fSChris Mason struct extent_buffer *tmp; 1617925baeddSChris Mason struct extent_buffer *old; 161895582b00SDeepa Dinamani struct timespec64 cur_time; 1619aec8030aSMiao Xie int ret = 0; 1620d68fc57bSYan, Zheng u64 to_reserve = 0; 16216bdb72deSSage Weil u64 index = 0; 1622a22285a6SYan, Zheng u64 objectid; 1623b83cc969SLi Zefan u64 root_flags; 16243063d29fSChris Mason 16258546b570SDavid Sterba ASSERT(pending->path); 16268546b570SDavid Sterba path = pending->path; 162742874b3dSMiao Xie 1628b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1629b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1630a22285a6SYan, Zheng 1631543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1632aec8030aSMiao Xie if (pending->error) 16336fa9700eSMiao Xie goto no_free_objectid; 16343063d29fSChris Mason 1635d6726335SQu Wenruo /* 1636d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1637d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1638d6726335SQu Wenruo */ 1639d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1640d6726335SQu Wenruo 1641147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1642d68fc57bSYan, Zheng 1643d68fc57bSYan, Zheng if (to_reserve > 0) { 16449270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1645aec8030aSMiao Xie &pending->block_rsv, 164608e007d2SMiao Xie to_reserve, 164708e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1648aec8030aSMiao Xie if (pending->error) 1649d6726335SQu Wenruo goto clear_skip_qgroup; 1650d68fc57bSYan, Zheng } 1651d68fc57bSYan, Zheng 16523063d29fSChris Mason key.objectid = objectid; 1653a22285a6SYan, Zheng key.offset = (u64)-1; 1654a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16553063d29fSChris Mason 16566fa9700eSMiao Xie rsv = trans->block_rsv; 1657a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16582382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16590b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 166088d3a5aaSJosef Bacik trans->transid, 166188d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1662a22285a6SYan, Zheng dentry = pending->dentry; 1663e9662f70SMiao Xie parent_inode = pending->dir; 1664a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1665f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1666f0118cb6SJosef Bacik if (ret) 1667f0118cb6SJosef Bacik goto fail; 1668c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 166904b285f3SDeepa Dinamani 16706bdb72deSSage Weil /* 16716bdb72deSSage Weil * insert the directory item 16726bdb72deSSage Weil */ 1673877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 167449b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 167542874b3dSMiao Xie 167642874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 167742874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16784a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 167942874b3dSMiao Xie dentry->d_name.name, 168042874b3dSMiao Xie dentry->d_name.len, 0); 168142874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1682fe66a05aSChris Mason pending->error = -EEXIST; 1683aec8030aSMiao Xie goto dir_item_existed; 168442874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 168542874b3dSMiao Xie ret = PTR_ERR(dir_item); 168666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16878732d44fSMiao Xie goto fail; 168879787eaaSJeff Mahoney } 168942874b3dSMiao Xie btrfs_release_path(path); 16906bdb72deSSage Weil 1691e999376fSChris Mason /* 1692e999376fSChris Mason * pull in the delayed directory update 1693e999376fSChris Mason * and the delayed inode item 1694e999376fSChris Mason * otherwise we corrupt the FS during 1695e999376fSChris Mason * snapshot 1696e999376fSChris Mason */ 1697e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 16988732d44fSMiao Xie if (ret) { /* Transaction aborted */ 169966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17008732d44fSMiao Xie goto fail; 17018732d44fSMiao Xie } 1702e999376fSChris Mason 1703f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1704f0118cb6SJosef Bacik if (ret) { 1705f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1706f0118cb6SJosef Bacik goto fail; 1707f0118cb6SJosef Bacik } 17086bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17096bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 171008fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17116bdb72deSSage Weil 1712b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1713b83cc969SLi Zefan if (pending->readonly) 1714b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1715b83cc969SLi Zefan else 1716b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1717b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1718b83cc969SLi Zefan 17198ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17208ea05e3aSAlexander Block trans->transid); 1721807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17228ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17238ea05e3aSAlexander Block BTRFS_UUID_SIZE); 172470023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 172570023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 172670023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17278ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17288ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17298ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17308ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 173170023da2SStefan Behrens } 17323cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17333cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 173470023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17358ea05e3aSAlexander Block 1736925baeddSChris Mason old = btrfs_lock_root_node(root); 17379631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17389631e4ccSJosef Bacik BTRFS_NESTING_COW); 173979787eaaSJeff Mahoney if (ret) { 174079787eaaSJeff Mahoney btrfs_tree_unlock(old); 174179787eaaSJeff Mahoney free_extent_buffer(old); 174266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17438732d44fSMiao Xie goto fail; 174479787eaaSJeff Mahoney } 174549b25e05SJeff Mahoney 174649b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 174779787eaaSJeff Mahoney /* clean up in any case */ 1748925baeddSChris Mason btrfs_tree_unlock(old); 1749925baeddSChris Mason free_extent_buffer(old); 17508732d44fSMiao Xie if (ret) { 175166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17528732d44fSMiao Xie goto fail; 17538732d44fSMiao Xie } 1754f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 175527cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1756f1ebcc74SLiu Bo smp_wmb(); 1757f1ebcc74SLiu Bo 17585d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1759a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1760a22285a6SYan, Zheng key.offset = trans->transid; 1761a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1762925baeddSChris Mason btrfs_tree_unlock(tmp); 17633063d29fSChris Mason free_extent_buffer(tmp); 17648732d44fSMiao Xie if (ret) { 176566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17668732d44fSMiao Xie goto fail; 17678732d44fSMiao Xie } 17680660b5afSChris Mason 1769a22285a6SYan, Zheng /* 1770a22285a6SYan, Zheng * insert root back/forward references 1771a22285a6SYan, Zheng */ 17726025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1773a22285a6SYan, Zheng parent_root->root_key.objectid, 17744a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1775a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 17768732d44fSMiao Xie if (ret) { 177766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17788732d44fSMiao Xie goto fail; 17798732d44fSMiao Xie } 1780a22285a6SYan, Zheng 1781a22285a6SYan, Zheng key.offset = (u64)-1; 17822dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 178379787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 178479787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17852d892ccdSFilipe Manana pending->snap = NULL; 178666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17878732d44fSMiao Xie goto fail; 178879787eaaSJeff Mahoney } 1789d68fc57bSYan, Zheng 179049b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17918732d44fSMiao Xie if (ret) { 179266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17938732d44fSMiao Xie goto fail; 17948732d44fSMiao Xie } 1795361048f5SMiao Xie 17966426c7adSQu Wenruo /* 17976426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17986426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17996426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 18006426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 18016426c7adSQu Wenruo */ 18026426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18036426c7adSQu Wenruo pending->inherit, objectid); 18046426c7adSQu Wenruo if (ret < 0) 18056426c7adSQu Wenruo goto fail; 18066426c7adSQu Wenruo 1807684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1808684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1809684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 181042874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18119c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18128732d44fSMiao Xie if (ret) { 181366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18148732d44fSMiao Xie goto fail; 18158732d44fSMiao Xie } 181642874b3dSMiao Xie 18176ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 181842874b3dSMiao Xie dentry->d_name.len * 2); 1819c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1820c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1821729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1822dd5f9615SStefan Behrens if (ret) { 182366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1824dd5f9615SStefan Behrens goto fail; 1825dd5f9615SStefan Behrens } 1826807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1827807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1828cdb345a8SLu Fengqi objectid); 1829dd5f9615SStefan Behrens if (ret) { 183066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1831dd5f9615SStefan Behrens goto fail; 1832dd5f9615SStefan Behrens } 1833dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1834cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1835dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1836dd5f9615SStefan Behrens objectid); 1837dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 183866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1839dd5f9615SStefan Behrens goto fail; 1840dd5f9615SStefan Behrens } 1841dd5f9615SStefan Behrens } 1842d6726335SQu Wenruo 18433063d29fSChris Mason fail: 1844aec8030aSMiao Xie pending->error = ret; 1845aec8030aSMiao Xie dir_item_existed: 184698c9942aSLiu Bo trans->block_rsv = rsv; 18472382c5ccSLiu Bo trans->bytes_reserved = 0; 1848d6726335SQu Wenruo clear_skip_qgroup: 1849d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 18506fa9700eSMiao Xie no_free_objectid: 18516fa9700eSMiao Xie kfree(new_root_item); 1852b0c0ea63SDavid Sterba pending->root_item = NULL; 185342874b3dSMiao Xie btrfs_free_path(path); 18548546b570SDavid Sterba pending->path = NULL; 18558546b570SDavid Sterba 185649b25e05SJeff Mahoney return ret; 18573063d29fSChris Mason } 18583063d29fSChris Mason 1859d352ac68SChris Mason /* 1860d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1861d352ac68SChris Mason */ 186208d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18633063d29fSChris Mason { 1864aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18653063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1866aec8030aSMiao Xie int ret = 0; 18673de4586cSChris Mason 1868aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1869aec8030aSMiao Xie list_del(&pending->list); 187008d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1871aec8030aSMiao Xie if (ret) 1872aec8030aSMiao Xie break; 1873aec8030aSMiao Xie } 1874aec8030aSMiao Xie return ret; 18753de4586cSChris Mason } 18763de4586cSChris Mason 18772ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18785d4f98a2SYan Zheng { 18795d4f98a2SYan Zheng struct btrfs_root_item *root_item; 18805d4f98a2SYan Zheng struct btrfs_super_block *super; 18815d4f98a2SYan Zheng 18820b246afaSJeff Mahoney super = fs_info->super_copy; 18835d4f98a2SYan Zheng 18840b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1885093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1886093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1887093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18885d4f98a2SYan Zheng 18890b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1890093e037cSDavid Sterba super->root = root_item->bytenr; 1891093e037cSDavid Sterba super->generation = root_item->generation; 1892093e037cSDavid Sterba super->root_level = root_item->level; 18930b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1894093e037cSDavid Sterba super->cache_generation = root_item->generation; 189594846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 189694846229SBoris Burkov super->cache_generation = 0; 18970b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1898093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18995d4f98a2SYan Zheng } 19005d4f98a2SYan Zheng 1901f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1902f36f3042SChris Mason { 19034a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1904f36f3042SChris Mason int ret = 0; 19054a9d8bdeSMiao Xie 1906a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19074a9d8bdeSMiao Xie trans = info->running_transaction; 19084a9d8bdeSMiao Xie if (trans) 19094a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1910a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1911f36f3042SChris Mason return ret; 1912f36f3042SChris Mason } 1913f36f3042SChris Mason 19148929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19158929ecfaSYan, Zheng { 19164a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19178929ecfaSYan, Zheng int ret = 0; 19184a9d8bdeSMiao Xie 1919a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19204a9d8bdeSMiao Xie trans = info->running_transaction; 19214a9d8bdeSMiao Xie if (trans) 19224a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1923a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19248929ecfaSYan, Zheng return ret; 19258929ecfaSYan, Zheng } 19268929ecfaSYan, Zheng 1927fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1928bb9c12c9SSage Weil { 19293a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1930bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1931bb9c12c9SSage Weil 1932fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1933fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1934fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1935bb9c12c9SSage Weil 1936bb9c12c9SSage Weil /* take transaction reference */ 1937bb9c12c9SSage Weil cur_trans = trans->transaction; 19389b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1939bb9c12c9SSage Weil 19403a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19416fc4e354SSage Weil 19426fc4e354SSage Weil /* 1943ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1944ae5d29d4SDavid Sterba * subsequent transaction joins 1945ae5d29d4SDavid Sterba */ 19463e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1947ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1948ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1949ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1950724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1951bb9c12c9SSage Weil } 1952bb9c12c9SSage Weil 195397cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 195449b25e05SJeff Mahoney { 195597cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 195649b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 195749b25e05SJeff Mahoney 1958b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 195949b25e05SJeff Mahoney 196066642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19617b8b92afSJosef Bacik 19620b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 196366b6135bSLiu Bo 196425d8c284SMiao Xie /* 196525d8c284SMiao Xie * If the transaction is removed from the list, it means this 196625d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 196725d8c284SMiao Xie * to call the cleanup function. 196825d8c284SMiao Xie */ 196925d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 197066b6135bSLiu Bo 19710b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19724a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19730b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1974e1489b4fSIoannis Angelakopoulos 1975e1489b4fSIoannis Angelakopoulos /* 1976e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 1977e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 1978e1489b4fSIoannis Angelakopoulos */ 1979e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 1980f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1981f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1982f094ac32SLiu Bo 19830b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1984d7096fc3SJosef Bacik } 1985061dde82SFilipe Manana 1986061dde82SFilipe Manana /* 1987061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 1988061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 1989061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 1990061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 1991061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 1992061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 1993061dde82SFilipe Manana */ 1994061dde82SFilipe Manana list_del_init(&cur_trans->list); 1995061dde82SFilipe Manana 19960b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 199749b25e05SJeff Mahoney 19982ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 199949b25e05SJeff Mahoney 20000b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20010b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20020b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20030b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20044a9d8bdeSMiao Xie 2005e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20060b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2007724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2008724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 200949b25e05SJeff Mahoney 20102e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 201149b25e05SJeff Mahoney 201249b25e05SJeff Mahoney if (current->journal_info == trans) 201349b25e05SJeff Mahoney current->journal_info = NULL; 20140b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 201549b25e05SJeff Mahoney 201649b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 201749b25e05SJeff Mahoney } 201849b25e05SJeff Mahoney 2019c7cc64a9SDavid Sterba /* 2020c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2021c7cc64a9SDavid Sterba * transaction and remove them from the list 2022c7cc64a9SDavid Sterba */ 2023c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2024c7cc64a9SDavid Sterba { 2025c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 202632da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2027c7cc64a9SDavid Sterba 2028c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2029c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2030c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2031c7cc64a9SDavid Sterba } 2032c7cc64a9SDavid Sterba } 2033c7cc64a9SDavid Sterba 203488090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 203582436617SMiao Xie { 2036ce8ea7ccSJosef Bacik /* 2037a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2038ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2039ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2040ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2041ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2042ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2043ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2044a0f0cf83SFilipe Manana * 2045a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2046a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2047a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2048a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2049a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2050a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2051a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2052ce8ea7ccSJosef Bacik */ 205388090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2054a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 205582436617SMiao Xie return 0; 205682436617SMiao Xie } 205782436617SMiao Xie 205888090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 205982436617SMiao Xie { 206088090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20616374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 206282436617SMiao Xie } 206382436617SMiao Xie 206428b21c55SFilipe Manana /* 206528b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 206628b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 206728b21c55SFilipe Manana * and while holding fs_info->trans_lock. 206828b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 206928b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 207028b21c55SFilipe Manana * returns an error. 207128b21c55SFilipe Manana */ 207228b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 207328b21c55SFilipe Manana { 207428b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 207528b21c55SFilipe Manana 207628b21c55SFilipe Manana if (!trans->pending_snapshot) 207728b21c55SFilipe Manana return; 207828b21c55SFilipe Manana 207928b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 208028b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 208128b21c55SFilipe Manana 208228b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 208328b21c55SFilipe Manana } 208428b21c55SFilipe Manana 2085e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2086e55958c8SIoannis Angelakopoulos { 2087e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2088e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2089e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2090e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2091e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2092e55958c8SIoannis Angelakopoulos } 2093e55958c8SIoannis Angelakopoulos 20943a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 209579154b1bSChris Mason { 20963a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 209749b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20988fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 209925287e0aSMiao Xie int ret; 2100e55958c8SIoannis Angelakopoulos ktime_t start_time; 2101e55958c8SIoannis Angelakopoulos ktime_t interval; 210279154b1bSChris Mason 210335b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21043e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 210535b814f3SNikolay Borisov 21068d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2107bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 210825287e0aSMiao Xie ret = cur_trans->aborted; 21093e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 211025287e0aSMiao Xie } 211149b25e05SJeff Mahoney 2112f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2113f45c752bSJosef Bacik trans->block_rsv = NULL; 2114f45c752bSJosef Bacik 2115e19eb11fSJosef Bacik /* 2116e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2117e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2118e19eb11fSJosef Bacik */ 2119e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2120e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2121e19eb11fSJosef Bacik /* 2122e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2123e19eb11fSJosef Bacik * Any running threads may add more while we are here. 212456bec294SChris Mason */ 2125c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21263e738c53SIoannis Angelakopoulos if (ret) 21273e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2128e19eb11fSJosef Bacik } 212956bec294SChris Mason 21306c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2131ea658badSJosef Bacik 21323204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21331bbc621eSChris Mason int run_it = 0; 21341bbc621eSChris Mason 21351bbc621eSChris Mason /* this mutex is also taken before trying to set 21361bbc621eSChris Mason * block groups readonly. We need to make sure 21371bbc621eSChris Mason * that nobody has set a block group readonly 21381bbc621eSChris Mason * after a extents from that block group have been 21391bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21401bbc621eSChris Mason * will wait for the transaction to commit if it 21413204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21421bbc621eSChris Mason * 21433204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21443204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21451bbc621eSChris Mason * hurt to have more than one go through, but there's no 21461bbc621eSChris Mason * real advantage to it either. 21471bbc621eSChris Mason */ 21480b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21493204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21503204d33cSJosef Bacik &cur_trans->flags)) 21511bbc621eSChris Mason run_it = 1; 21520b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21531bbc621eSChris Mason 2154f9cacae3SNikolay Borisov if (run_it) { 215521217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21563e738c53SIoannis Angelakopoulos if (ret) 21573e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2158f9cacae3SNikolay Borisov } 2159f9cacae3SNikolay Borisov } 21601bbc621eSChris Mason 21610b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21624a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2163d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2164d0c2f4faSFilipe Manana 216528b21c55SFilipe Manana add_pending_snapshot(trans); 216628b21c55SFilipe Manana 21670b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21689b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2169ccd467d6SChris Mason 2170d0c2f4faSFilipe Manana if (trans->in_fsync) 2171d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 21723e738c53SIoannis Angelakopoulos 21733e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 21743e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2175d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2176d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 217715ee9bc7SJosef Bacik 2178bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2179b4924a0fSLiu Bo ret = cur_trans->aborted; 2180b4924a0fSLiu Bo 2181724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 218215ee9bc7SJosef Bacik 218349b25e05SJeff Mahoney return ret; 218479154b1bSChris Mason } 21854313b399SChris Mason 21864a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21870b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 21883e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2189bb9c12c9SSage Weil 21900b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2191d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2192d0c2f4faSFilipe Manana 2193d0c2f4faSFilipe Manana if (trans->in_fsync) 2194d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2195d0c2f4faSFilipe Manana 2196ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2197ccd467d6SChris Mason struct btrfs_transaction, list); 2198d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 21999b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 22000b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2201ccd467d6SChris Mason 2202d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2203d0c2f4faSFilipe Manana 2204bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2205ccd467d6SChris Mason 2206724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22071f9b8c8fSFilipe Manana if (ret) 2208e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2209a4abeea4SJosef Bacik } else { 22100b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2211ccd467d6SChris Mason } 2212a4abeea4SJosef Bacik } else { 22130b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2214cb2d3dadSFilipe Manana /* 2215cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2216cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2217cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2218cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2219cb2d3dadSFilipe Manana */ 222084961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2221cb2d3dadSFilipe Manana ret = -EROFS; 2222e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2223cb2d3dadSFilipe Manana } 2224ccd467d6SChris Mason } 222515ee9bc7SJosef Bacik 2226e55958c8SIoannis Angelakopoulos /* 2227e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2228e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2229e55958c8SIoannis Angelakopoulos */ 2230e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2231e55958c8SIoannis Angelakopoulos 22320860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22330860adfdSMiao Xie 223488090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 223582436617SMiao Xie if (ret) 2236e1489b4fSIoannis Angelakopoulos goto lockdep_release; 223782436617SMiao Xie 2238e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 223949b25e05SJeff Mahoney if (ret) 2240e1489b4fSIoannis Angelakopoulos goto lockdep_release; 224116cdcec7SMiao Xie 22425a9ba670SIoannis Angelakopoulos /* 22435a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22445a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22455a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22465a9ba670SIoannis Angelakopoulos */ 22475a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 22485a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2249581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2250581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2251ed3b3d31SChris Mason 2252581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2253e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2254e1489b4fSIoannis Angelakopoulos if (ret) { 2255e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2256ca469637SMiao Xie goto cleanup_transaction; 2257e1489b4fSIoannis Angelakopoulos } 2258ca469637SMiao Xie 225988090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2260cb7ab021SWang Shilong 226148778179SFilipe Manana /* 226248778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 226348778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 226448778179SFilipe Manana * extents complete we lose logged data after a power failure. 226548778179SFilipe Manana */ 22668b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 226748778179SFilipe Manana wait_event(cur_trans->pending_wait, 226848778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 226948778179SFilipe Manana 22702ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2271ed0ca140SJosef Bacik /* 2272ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2273ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22744a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2275ed0ca140SJosef Bacik */ 22760b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 227728b21c55SFilipe Manana add_pending_snapshot(trans); 22784a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 22790b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2280e1489b4fSIoannis Angelakopoulos 2281e1489b4fSIoannis Angelakopoulos /* 2282e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2283e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2284e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2285e1489b4fSIoannis Angelakopoulos */ 2286e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2287e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2288ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2289ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 229015ee9bc7SJosef Bacik 2291fdfbf020SJosef Bacik /* 22923e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 22933e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 22943e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 22953e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 22963e738c53SIoannis Angelakopoulos */ 22973e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 22983e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 22993e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23003e738c53SIoannis Angelakopoulos 23013e738c53SIoannis Angelakopoulos /* 2302fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2303fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2304fdfbf020SJosef Bacik * kthread could do the work. 2305fdfbf020SJosef Bacik */ 2306fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2307fdfbf020SJosef Bacik 2308bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23092cba30f1SMiao Xie ret = cur_trans->aborted; 23103e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23116cf7f77eSWang Shilong goto scrub_continue; 23122cba30f1SMiao Xie } 23137585717fSChris Mason /* 23147585717fSChris Mason * the reloc mutex makes sure that we stop 23157585717fSChris Mason * the balancing code from coming in and moving 23167585717fSChris Mason * extents around in the middle of the commit 23177585717fSChris Mason */ 23180b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23197585717fSChris Mason 232042874b3dSMiao Xie /* 232142874b3dSMiao Xie * We needn't worry about the delayed items because we will 232242874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 232342874b3dSMiao Xie * core function of the snapshot creation. 232442874b3dSMiao Xie */ 232508d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 232656e9f6eaSDavid Sterba if (ret) 232756e9f6eaSDavid Sterba goto unlock_reloc; 23283063d29fSChris Mason 232942874b3dSMiao Xie /* 233042874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 233142874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 233242874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 233342874b3dSMiao Xie * them. 233442874b3dSMiao Xie * 233542874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 233642874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 233742874b3dSMiao Xie * the nodes and leaves. 233842874b3dSMiao Xie */ 2339e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 234056e9f6eaSDavid Sterba if (ret) 234156e9f6eaSDavid Sterba goto unlock_reloc; 234216cdcec7SMiao Xie 2343c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 234456e9f6eaSDavid Sterba if (ret) 234556e9f6eaSDavid Sterba goto unlock_reloc; 234656bec294SChris Mason 2347e999376fSChris Mason /* 2348e999376fSChris Mason * make sure none of the code above managed to slip in a 2349e999376fSChris Mason * delayed item 2350e999376fSChris Mason */ 2351ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2352e999376fSChris Mason 23532c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2354dc17ff8fSChris Mason 23557e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 235656e9f6eaSDavid Sterba if (ret) 2357dfba78dcSFilipe Manana goto unlock_reloc; 235854aa1f4dSChris Mason 23593818aea2SQu Wenruo /* 23607e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 23617e1876acSDavid Sterba * before the next transaction. 23623818aea2SQu Wenruo */ 23630b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 23643818aea2SQu Wenruo 23655d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2366e02119d5SChris Mason * safe to free the root of tree log roots 2367e02119d5SChris Mason */ 23680b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2369e02119d5SChris Mason 23700ed4792aSQu Wenruo /* 23710ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 23720ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 23730ed4792aSQu Wenruo */ 2374460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 237556e9f6eaSDavid Sterba if (ret < 0) 2376dfba78dcSFilipe Manana goto unlock_reloc; 23770ed4792aSQu Wenruo 23789386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 237956e9f6eaSDavid Sterba if (ret) 2380dfba78dcSFilipe Manana goto unlock_reloc; 238154aa1f4dSChris Mason 23822cba30f1SMiao Xie /* 23832cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 23842cba30f1SMiao Xie * update ->aborted, check it. 23852cba30f1SMiao Xie */ 2386bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23872cba30f1SMiao Xie ret = cur_trans->aborted; 2388dfba78dcSFilipe Manana goto unlock_reloc; 23892cba30f1SMiao Xie } 23902cba30f1SMiao Xie 23910b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 23925f39d397SChris Mason 23930b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 23940b246afaSJeff Mahoney fs_info->tree_root->node); 23950b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 23969e351cc8SJosef Bacik &cur_trans->switch_commits); 23975d4f98a2SYan Zheng 23980b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 23990b246afaSJeff Mahoney fs_info->chunk_root->node); 24000b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 24019e351cc8SJosef Bacik &cur_trans->switch_commits); 24029e351cc8SJosef Bacik 2403f7238e50SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2404f7238e50SJosef Bacik btrfs_set_root_node(&fs_info->block_group_root->root_item, 2405f7238e50SJosef Bacik fs_info->block_group_root->node); 2406f7238e50SJosef Bacik list_add_tail(&fs_info->block_group_root->dirty_list, 2407f7238e50SJosef Bacik &cur_trans->switch_commits); 2408f7238e50SJosef Bacik } 2409f7238e50SJosef Bacik 2410889bfa39SJosef Bacik switch_commit_roots(trans); 24115d4f98a2SYan Zheng 2412ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 24131bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 24142ff7e61eSJeff Mahoney update_super_roots(fs_info); 2415e02119d5SChris Mason 24160b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 24170b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 24180b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 24190b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2420ccd467d6SChris Mason 2421bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2422935e5cc9SMiao Xie 24230b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 24240b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2425656f30dbSFilipe Manana 24264fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24274fbcdf66SFilipe Manana 2428dfba78dcSFilipe Manana /* 2429dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2430dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2431dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2432dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2433dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2434dfba78dcSFilipe Manana * writing its superblock. 2435dfba78dcSFilipe Manana */ 2436dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2437dfba78dcSFilipe Manana 24380b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 24394a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 24400b246afaSJeff Mahoney fs_info->running_transaction = NULL; 24410b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 24420b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2443b7ec40d7SChris Mason 24440b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 24453e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2446e6dcd2dcSChris Mason 244770458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 244849b25e05SJeff Mahoney if (ret) { 24490b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 245008748810SDavid Sterba "Error while writing out transaction"); 24510b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 24526cf7f77eSWang Shilong goto scrub_continue; 245349b25e05SJeff Mahoney } 245449b25e05SJeff Mahoney 2455d3575156SNaohiro Aota /* 2456d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2457d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2458d3575156SNaohiro Aota * buffers. 2459d3575156SNaohiro Aota */ 2460d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2461d3575156SNaohiro Aota 2462eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2463e02119d5SChris Mason /* 2464e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2465e02119d5SChris Mason * to go about their business 2466e02119d5SChris Mason */ 24670b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2468c1f32b7cSAnand Jain if (ret) 2469c1f32b7cSAnand Jain goto scrub_continue; 2470e02119d5SChris Mason 2471d0c2f4faSFilipe Manana /* 2472d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2473d0c2f4faSFilipe Manana * which can change it. 2474d0c2f4faSFilipe Manana */ 2475d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2476d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 24773e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2478d0c2f4faSFilipe Manana 24795ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 24804313b399SChris Mason 24813204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 24820b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 248313212b54SZhao Lei 24840b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 24854a9d8bdeSMiao Xie /* 24864a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 24874a9d8bdeSMiao Xie * which can change it. 24884a9d8bdeSMiao Xie */ 24894a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 24902c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 24913e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 24923de4586cSChris Mason 24930b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 249413c5a93eSJosef Bacik list_del_init(&cur_trans->list); 24950b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2496a4abeea4SJosef Bacik 2497724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2498724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 249958176a96SJosef Bacik 25000860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 25010b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2502b2b5ef5cSJan Kara 25032e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 25041abe9b8aSliubo 2505e55958c8SIoannis Angelakopoulos interval = ktime_get_ns() - start_time; 2506e55958c8SIoannis Angelakopoulos 25072ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2508a2de733cSArne Jansen 25099ed74f2dSJosef Bacik if (current->journal_info == trans) 25109ed74f2dSJosef Bacik current->journal_info = NULL; 25119ed74f2dSJosef Bacik 25122c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 251324bbcf04SYan, Zheng 2514e55958c8SIoannis Angelakopoulos update_commit_stats(fs_info, interval); 2515e55958c8SIoannis Angelakopoulos 251679154b1bSChris Mason return ret; 251749b25e05SJeff Mahoney 251856e9f6eaSDavid Sterba unlock_reloc: 251956e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 25203e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 25216cf7f77eSWang Shilong scrub_continue: 25223e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 25233e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25242ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 252549b25e05SJeff Mahoney cleanup_transaction: 2526dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2527c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 25284fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 25290e721106SJosef Bacik trans->block_rsv = NULL; 25300b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 253149b25e05SJeff Mahoney if (current->journal_info == trans) 253249b25e05SJeff Mahoney current->journal_info = NULL; 253397cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 253449b25e05SJeff Mahoney 253549b25e05SJeff Mahoney return ret; 2536e1489b4fSIoannis Angelakopoulos 2537e1489b4fSIoannis Angelakopoulos lockdep_release: 25385a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2539e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2540e1489b4fSIoannis Angelakopoulos goto cleanup_transaction; 25413e738c53SIoannis Angelakopoulos 25423e738c53SIoannis Angelakopoulos lockdep_trans_commit_start_release: 25433e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 25443e738c53SIoannis Angelakopoulos btrfs_end_transaction(trans); 25453e738c53SIoannis Angelakopoulos return ret; 254679154b1bSChris Mason } 254779154b1bSChris Mason 2548d352ac68SChris Mason /* 25499d1a2a3aSDavid Sterba * return < 0 if error 25509d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 25519d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 25529d1a2a3aSDavid Sterba * 25539d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 25549d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 25559d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 25569d1a2a3aSDavid Sterba * few seconds later. 2557d352ac68SChris Mason */ 255833c44184SJosef Bacik int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2559e9d0b13bSChris Mason { 256033c44184SJosef Bacik struct btrfs_root *root; 25619d1a2a3aSDavid Sterba int ret; 2562e9d0b13bSChris Mason 2563a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 25649d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 25659d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 25669d1a2a3aSDavid Sterba return 0; 25679d1a2a3aSDavid Sterba } 25689d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 25699d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2570cfad392bSJosef Bacik list_del_init(&root->root_list); 2571a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 25725d4f98a2SYan Zheng 25734fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 257476dda93cSYan, Zheng 257516cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 257616cdcec7SMiao Xie 257776dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 257876dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 25790078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 258076dda93cSYan, Zheng else 25810078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 258232471dc2SDavid Sterba 2583dc9492c1SJosef Bacik btrfs_put_root(root); 25846596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2585e9d0b13bSChris Mason } 2586572d9ab7SDavid Sterba 2587572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2588572d9ab7SDavid Sterba { 2589572d9ab7SDavid Sterba unsigned long prev; 2590572d9ab7SDavid Sterba unsigned long bit; 2591572d9ab7SDavid Sterba 25926c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2593572d9ab7SDavid Sterba if (!prev) 2594572d9ab7SDavid Sterba return; 2595572d9ab7SDavid Sterba 2596d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2597d51033d0SDavid Sterba if (prev & bit) 2598d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2599d51033d0SDavid Sterba prev &= ~bit; 2600d51033d0SDavid Sterba 2601572d9ab7SDavid Sterba if (prev) 2602572d9ab7SDavid Sterba btrfs_warn(fs_info, 2603572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2604572d9ab7SDavid Sterba } 2605*956504a3SJosef Bacik 2606*956504a3SJosef Bacik int __init btrfs_transaction_init(void) 2607*956504a3SJosef Bacik { 2608*956504a3SJosef Bacik btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2609*956504a3SJosef Bacik sizeof(struct btrfs_trans_handle), 0, 2610*956504a3SJosef Bacik SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2611*956504a3SJosef Bacik if (!btrfs_trans_handle_cachep) 2612*956504a3SJosef Bacik return -ENOMEM; 2613*956504a3SJosef Bacik return 0; 2614*956504a3SJosef Bacik } 2615*956504a3SJosef Bacik 2616*956504a3SJosef Bacik void __cold btrfs_transaction_exit(void) 2617*956504a3SJosef Bacik { 2618*956504a3SJosef Bacik kmem_cache_destroy(btrfs_trans_handle_cachep); 2619*956504a3SJosef Bacik } 2620