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" 26c7f13d42SJosef Bacik #include "fs.h" 27*07e81dc9SJosef Bacik #include "accessors.h" 2879154b1bSChris Mason 29956504a3SJosef Bacik static struct kmem_cache *btrfs_trans_handle_cachep; 30956504a3SJosef Bacik 31fc7cbcd4SDavid Sterba #define BTRFS_ROOT_TRANS_TAG 0 320f7d52f4SChris Mason 3361c047b5SQu Wenruo /* 3461c047b5SQu Wenruo * Transaction states and transitions 3561c047b5SQu Wenruo * 3661c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 3761c047b5SQu Wenruo * | 3861c047b5SQu Wenruo * | To next stage: 3961c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 4061c047b5SQu Wenruo * V 4161c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 4261c047b5SQu Wenruo * | 4361c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 4461c047b5SQu Wenruo * | start_transaction() variants. 4561c047b5SQu Wenruo * | 4661c047b5SQu Wenruo * | To next stage: 4761c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 4861c047b5SQu Wenruo * | transaction N 4961c047b5SQu Wenruo * V 5061c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 5161c047b5SQu Wenruo * | 5261c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 5361c047b5SQu Wenruo * | is one 5461c047b5SQu Wenruo * | 5561c047b5SQu Wenruo * | Then one of the following happes: 5661c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 5761c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 5861c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 5961c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 6061c047b5SQu Wenruo * | 6161c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 6261c047b5SQu Wenruo * | to this running transaction. 6361c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 6461c047b5SQu Wenruo * | transaction N+1. 6561c047b5SQu Wenruo * | 6661c047b5SQu Wenruo * | To next stage: 6761c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 6861c047b5SQu Wenruo * | haven been released. 6961c047b5SQu Wenruo * V 7061c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 7161c047b5SQu Wenruo * | 7261c047b5SQu Wenruo * | The heavy lifting transaction work is started. 7361c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 7461c047b5SQu Wenruo * | snapshots, running qgroups. 7561c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 7661c047b5SQu Wenruo * | trees. 7761c047b5SQu Wenruo * | 7861c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 7961c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 8061c047b5SQu Wenruo * | 8161c047b5SQu Wenruo * | To next stage: 8261c047b5SQu Wenruo * | Until all supporting trees are updated. 8361c047b5SQu Wenruo * V 8461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 8561c047b5SQu Wenruo * | Transaction N+1 8661c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 8761c047b5SQu Wenruo * | need to write them back to disk and update | 8861c047b5SQu Wenruo * | super blocks. | 8961c047b5SQu Wenruo * | | 9061c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 9161c047b5SQu Wenruo * | start. | 9261c047b5SQu Wenruo * | All new start_transaction() calls will be | 9361c047b5SQu Wenruo * | attached to transid N+1. | 9461c047b5SQu Wenruo * | | 9561c047b5SQu Wenruo * | To next stage: | 9661c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 9761c047b5SQu Wenruo * | written to block devices | 9861c047b5SQu Wenruo * V | 9961c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 10061c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 10161c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 10261c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 10361c047b5SQu Wenruo */ 104e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1054a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 106bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 107bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1084a9d8bdeSMiao Xie __TRANS_ATTACH | 109a6d155d2SFilipe Manana __TRANS_JOIN | 110a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 111bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1124a9d8bdeSMiao Xie __TRANS_ATTACH | 1134a9d8bdeSMiao Xie __TRANS_JOIN | 114a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 115a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 116d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 117d0c2f4faSFilipe Manana __TRANS_ATTACH | 118d0c2f4faSFilipe Manana __TRANS_JOIN | 119d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 120d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 121bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1224a9d8bdeSMiao Xie __TRANS_ATTACH | 1234a9d8bdeSMiao Xie __TRANS_JOIN | 124a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 125a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1264a9d8bdeSMiao Xie }; 1274a9d8bdeSMiao Xie 128724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 12979154b1bSChris Mason { 1309b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1319b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 132a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1335c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1345c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 13581f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 13681f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1371262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 138ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 139ab8d0fc4SJeff Mahoney "pending csums is %llu", 1401262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1417785a663SFilipe Manana /* 1427785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1437785a663SFilipe Manana * because the transaction was aborted and a commit did not 1447785a663SFilipe Manana * happen (things failed before writing the new superblock 1457785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1467785a663SFilipe Manana * discard the physical locations of the block groups. 1477785a663SFilipe Manana */ 1487785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 14932da5386SDavid Sterba struct btrfs_block_group *cache; 1507785a663SFilipe Manana 1517785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 15232da5386SDavid Sterba struct btrfs_block_group, 1537785a663SFilipe Manana bg_list); 1547785a663SFilipe Manana list_del_init(&cache->bg_list); 1556b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1567785a663SFilipe Manana btrfs_put_block_group(cache); 1577785a663SFilipe Manana } 158bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1594b5faeacSDavid Sterba kfree(transaction); 16079154b1bSChris Mason } 16178fae27eSChris Mason } 16279154b1bSChris Mason 163889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 164817d52f8SJosef Bacik { 165889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 16616916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1679e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 1689e351cc8SJosef Bacik 169dfba78dcSFilipe Manana /* 170dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 171dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 172dfba78dcSFilipe Manana */ 173dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 174dfba78dcSFilipe Manana 1759e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 176d96b3424SFilipe Manana 177d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 178d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 179d96b3424SFilipe Manana 180889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1819e351cc8SJosef Bacik dirty_list) { 1829e351cc8SJosef Bacik list_del_init(&root->dirty_list); 183817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 184817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 18541e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 186370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1879e351cc8SJosef Bacik } 1882b9dbef2SJosef Bacik 1892b9dbef2SJosef Bacik /* We can free old roots now. */ 190889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 191889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 192889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1932b9dbef2SJosef Bacik struct btrfs_root, root_list); 1942b9dbef2SJosef Bacik list_del_init(&root->root_list); 195889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 196889bfa39SJosef Bacik btrfs_free_log(trans, root); 1972b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 198889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1992b9dbef2SJosef Bacik } 200889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 20127d56e62SJosef Bacik 2029e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 203817d52f8SJosef Bacik } 204817d52f8SJosef Bacik 2050860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2060860adfdSMiao Xie unsigned int type) 2070860adfdSMiao Xie { 2080860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2090860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2100860adfdSMiao Xie } 2110860adfdSMiao Xie 2120860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2130860adfdSMiao Xie unsigned int type) 2140860adfdSMiao Xie { 2150860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2160860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2170860adfdSMiao Xie } 2180860adfdSMiao Xie 2190860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2200860adfdSMiao Xie unsigned int type) 2210860adfdSMiao Xie { 2220860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2230860adfdSMiao Xie } 2240860adfdSMiao Xie 2250860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2260860adfdSMiao Xie { 2270860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 228178260b2SMiao Xie } 229178260b2SMiao Xie 230d352ac68SChris Mason /* 23179bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 23279bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 23379bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 23479bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 23579bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 236fb6dea26SJosef Bacik */ 237fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 238fb6dea26SJosef Bacik { 239fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 240fb6dea26SJosef Bacik 241fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 242fb6dea26SJosef Bacik return; 243fb6dea26SJosef Bacik 244fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 24563f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 246fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 247fb6dea26SJosef Bacik } 248fb6dea26SJosef Bacik 249fb6dea26SJosef Bacik /* 250d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 251d352ac68SChris Mason */ 2522ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2532ff7e61eSJeff Mahoney unsigned int type) 25479154b1bSChris Mason { 25579154b1bSChris Mason struct btrfs_transaction *cur_trans; 256a4abeea4SJosef Bacik 25719ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 258d43317dcSChris Mason loop: 25949b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 26084961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 26119ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26249b25e05SJeff Mahoney return -EROFS; 26349b25e05SJeff Mahoney } 26449b25e05SJeff Mahoney 26519ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 266a4abeea4SJosef Bacik if (cur_trans) { 267bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 26819ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26949b25e05SJeff Mahoney return cur_trans->aborted; 270871383beSDavid Sterba } 2714a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 272178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 273178260b2SMiao Xie return -EBUSY; 274178260b2SMiao Xie } 2759b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 276a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2770860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 27819ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 279e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2805a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 281a4abeea4SJosef Bacik return 0; 282a4abeea4SJosef Bacik } 28319ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 284a4abeea4SJosef Bacik 285354aa0fbSMiao Xie /* 286354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 287354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 288354aa0fbSMiao Xie */ 289354aa0fbSMiao Xie if (type == TRANS_ATTACH) 290354aa0fbSMiao Xie return -ENOENT; 291354aa0fbSMiao Xie 2924a9d8bdeSMiao Xie /* 2934a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 2944a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 2954a9d8bdeSMiao Xie */ 2964a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 2974a9d8bdeSMiao Xie 2984b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 299db5b493aSTsutomu Itoh if (!cur_trans) 300db5b493aSTsutomu Itoh return -ENOMEM; 301d43317dcSChris Mason 302e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 3035a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 304e1489b4fSIoannis Angelakopoulos 30519ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 30619ae4e81SJan Schmidt if (fs_info->running_transaction) { 307d43317dcSChris Mason /* 308d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3094a9d8bdeSMiao Xie * to redo the checks above 310d43317dcSChris Mason */ 3115a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 312e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3134b5faeacSDavid Sterba kfree(cur_trans); 314d43317dcSChris Mason goto loop; 31584961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 316e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3175a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 318e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3194b5faeacSDavid Sterba kfree(cur_trans); 3207b8b92afSJosef Bacik return -EROFS; 321a4abeea4SJosef Bacik } 322d43317dcSChris Mason 323ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 32448778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 32548778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 32613c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3270860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 32879154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 32979154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3304a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 331a4abeea4SJosef Bacik /* 332a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 333a4abeea4SJosef Bacik * commit the transaction. 334a4abeea4SJosef Bacik */ 3359b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3363204d33cSJosef Bacik cur_trans->flags = 0; 337afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 33856bec294SChris Mason 339a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 340a099d0fdSAlexandru Moise 3415c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3423368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 343d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 34420b297d6SJan Schmidt 34520b297d6SJan Schmidt /* 34620b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 34720b297d6SJan Schmidt * the log must never go across transaction boundaries. 34820b297d6SJan Schmidt */ 34920b297d6SJan Schmidt smp_mb(); 35031b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3515d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 35231b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3535d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 354fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 35520b297d6SJan Schmidt 35656bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 35756bec294SChris Mason 3583063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 359bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3609e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 361ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3621bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3632b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3641bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 365ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 366e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3672b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 368d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 369d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 37019ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 371c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 372efb0645bSJosef Bacik IO_TREE_TRANS_DIRTY_PAGES, NULL); 373fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 374fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 37519ae4e81SJan Schmidt fs_info->generation++; 37619ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 37719ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 37849b25e05SJeff Mahoney cur_trans->aborted = 0; 37919ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 38015ee9bc7SJosef Bacik 38179154b1bSChris Mason return 0; 38279154b1bSChris Mason } 38379154b1bSChris Mason 384d352ac68SChris Mason /* 38592a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 38692a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 38792a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 38892a7cc42SQu Wenruo * transaction commits. 389d352ac68SChris Mason */ 3907585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3916426c7adSQu Wenruo struct btrfs_root *root, 3926426c7adSQu Wenruo int force) 3936702ed49SChris Mason { 3940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 39503a7e111SJosef Bacik int ret = 0; 3960b246afaSJeff Mahoney 39792a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 3986426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 3994d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4005d4f98a2SYan Zheng 4017585717fSChris Mason /* 40227cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4037585717fSChris Mason * we have the reloc mutex held now, so there 4047585717fSChris Mason * is only one writer in this function 4057585717fSChris Mason */ 40627cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4077585717fSChris Mason 40827cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4097585717fSChris Mason * they find our root->last_trans update 4107585717fSChris Mason */ 4117585717fSChris Mason smp_wmb(); 4127585717fSChris Mason 413fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4146426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 415fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 416a4abeea4SJosef Bacik return 0; 417a4abeea4SJosef Bacik } 418fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4196702ed49SChris Mason (unsigned long)root->root_key.objectid, 4206702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 421fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4227585717fSChris Mason root->last_trans = trans->transid; 4237585717fSChris Mason 4247585717fSChris Mason /* this is pretty tricky. We don't want to 4257585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4267585717fSChris Mason * unless we're really doing the first setup for this root in 4277585717fSChris Mason * this transaction. 4287585717fSChris Mason * 4297585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4307585717fSChris Mason * if we want to take the expensive mutex. 4317585717fSChris Mason * 4327585717fSChris Mason * But, we have to set root->last_trans before we 4337585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4347585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 43527cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4367585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4377585717fSChris Mason * 4387585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4397585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4407585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4417585717fSChris Mason * done before we pop in the zero below 4427585717fSChris Mason */ 44303a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 444c7548af6SChris Mason smp_mb__before_atomic(); 44527cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4466702ed49SChris Mason } 44703a7e111SJosef Bacik return ret; 4486702ed49SChris Mason } 4495d4f98a2SYan Zheng 4507585717fSChris Mason 4512b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4522b9dbef2SJosef Bacik struct btrfs_root *root) 4532b9dbef2SJosef Bacik { 4540b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4552b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4562b9dbef2SJosef Bacik 4572b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4582b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4592b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4602b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4612b9dbef2SJosef Bacik 4622b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 463fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 464fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4652b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4662b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 467fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4682b9dbef2SJosef Bacik } 4692b9dbef2SJosef Bacik 4707585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4717585717fSChris Mason struct btrfs_root *root) 4727585717fSChris Mason { 4730b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4741409e6ccSJosef Bacik int ret; 4750b246afaSJeff Mahoney 47692a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4777585717fSChris Mason return 0; 4787585717fSChris Mason 4797585717fSChris Mason /* 48027cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4817585717fSChris Mason * and barriers 4827585717fSChris Mason */ 4837585717fSChris Mason smp_rmb(); 4847585717fSChris Mason if (root->last_trans == trans->transid && 48527cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4867585717fSChris Mason return 0; 4877585717fSChris Mason 4880b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4891409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4900b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4917585717fSChris Mason 4921409e6ccSJosef Bacik return ret; 4937585717fSChris Mason } 4947585717fSChris Mason 4954a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 4964a9d8bdeSMiao Xie { 4973296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 498501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 499bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5004a9d8bdeSMiao Xie } 5014a9d8bdeSMiao Xie 502d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 503d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 504d352ac68SChris Mason * transaction might not be fully on disk. 505d352ac68SChris Mason */ 5062ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 50779154b1bSChris Mason { 508f9295749SChris Mason struct btrfs_transaction *cur_trans; 50979154b1bSChris Mason 5100b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5110b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5124a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5139b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5140b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51572d63ed6SLi Zefan 5163e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5170b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 518501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 519bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 520724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 521a4abeea4SJosef Bacik } else { 5220b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 523f9295749SChris Mason } 52437d1aeeeSChris Mason } 52537d1aeeeSChris Mason 5262ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 52737d1aeeeSChris Mason { 5280b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 529a4abeea4SJosef Bacik return 0; 530a4abeea4SJosef Bacik 53192e2f7e3SNikolay Borisov if (type == TRANS_START) 532a4abeea4SJosef Bacik return 1; 533a4abeea4SJosef Bacik 534a22285a6SYan, Zheng return 0; 535a22285a6SYan, Zheng } 536a22285a6SYan, Zheng 53720dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 53820dd2cbfSMiao Xie { 5390b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5400b246afaSJeff Mahoney 5410b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 54292a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 54320dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 54420dd2cbfSMiao Xie root->reloc_root) 54520dd2cbfSMiao Xie return false; 54620dd2cbfSMiao Xie 54720dd2cbfSMiao Xie return true; 54820dd2cbfSMiao Xie } 54920dd2cbfSMiao Xie 55008e007d2SMiao Xie static struct btrfs_trans_handle * 5515aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 552003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 553003d7c59SJeff Mahoney bool enforce_qgroups) 554a22285a6SYan, Zheng { 5550b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 556ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 557a22285a6SYan, Zheng struct btrfs_trans_handle *h; 558a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 559b5009945SJosef Bacik u64 num_bytes = 0; 560c5567237SArne Jansen u64 qgroup_reserved = 0; 56120dd2cbfSMiao Xie bool reloc_reserved = false; 5629c343784SJosef Bacik bool do_chunk_alloc = false; 56320dd2cbfSMiao Xie int ret; 564acce952bSliubo 56584961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 566acce952bSliubo return ERR_PTR(-EROFS); 5672a1eb461SJosef Bacik 56846c4e71eSFilipe Manana if (current->journal_info) { 5690860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5702a1eb461SJosef Bacik h = current->journal_info; 571b50fff81SDavid Sterba refcount_inc(&h->use_count); 572b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5732a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5742a1eb461SJosef Bacik h->block_rsv = NULL; 5752a1eb461SJosef Bacik goto got_it; 5762a1eb461SJosef Bacik } 577b5009945SJosef Bacik 578b5009945SJosef Bacik /* 579b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 580b5009945SJosef Bacik * the appropriate flushing if need be. 581b5009945SJosef Bacik */ 582003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 583ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 584ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 585ba2c4d4eSJosef Bacik 5860b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 587733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 588003d7c59SJeff Mahoney enforce_qgroups); 589c5567237SArne Jansen if (ret) 590c5567237SArne Jansen return ERR_PTR(ret); 591c5567237SArne Jansen 592ba2c4d4eSJosef Bacik /* 593ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 594ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 595ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 596ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 597ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 598ba2c4d4eSJosef Bacik */ 5992bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6007f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 601748f553cSDavid Sterba btrfs_block_rsv_full(delayed_refs_rsv) == 0) { 602ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 603ba2c4d4eSJosef Bacik num_bytes <<= 1; 604ba2c4d4eSJosef Bacik } 605ba2c4d4eSJosef Bacik 60620dd2cbfSMiao Xie /* 60720dd2cbfSMiao Xie * Do the reservation for the relocation root creation 60820dd2cbfSMiao Xie */ 609ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6100b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 61120dd2cbfSMiao Xie reloc_reserved = true; 61220dd2cbfSMiao Xie } 61320dd2cbfSMiao Xie 6149270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 615ba2c4d4eSJosef Bacik if (ret) 616ba2c4d4eSJosef Bacik goto reserve_fail; 617ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 618ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 619ba2c4d4eSJosef Bacik delayed_refs_bytes); 620ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 621ba2c4d4eSJosef Bacik } 6229c343784SJosef Bacik 6239c343784SJosef Bacik if (rsv->space_info->force_alloc) 6249c343784SJosef Bacik do_chunk_alloc = true; 625ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 626748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 627ba2c4d4eSJosef Bacik /* 628ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 629ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 630ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 631ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 632ba2c4d4eSJosef Bacik * here. 633ba2c4d4eSJosef Bacik */ 634ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 635b5009945SJosef Bacik if (ret) 636843fcf35SMiao Xie goto reserve_fail; 637b5009945SJosef Bacik } 638a22285a6SYan, Zheng again: 639f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 640843fcf35SMiao Xie if (!h) { 641843fcf35SMiao Xie ret = -ENOMEM; 642843fcf35SMiao Xie goto alloc_fail; 643843fcf35SMiao Xie } 644a22285a6SYan, Zheng 64598114659SJosef Bacik /* 64698114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 64798114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 64898114659SJosef Bacik * because we're already holding a ref. We need this because we could 64998114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 65098114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 651354aa0fbSMiao Xie * 652354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 653354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 65498114659SJosef Bacik */ 6550860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6560b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 657b2b5ef5cSJan Kara 6582ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6592ff7e61eSJeff Mahoney wait_current_trans(fs_info); 660a22285a6SYan, Zheng 661a4abeea4SJosef Bacik do { 6622ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 663178260b2SMiao Xie if (ret == -EBUSY) { 6642ff7e61eSJeff Mahoney wait_current_trans(fs_info); 665a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 666a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 667178260b2SMiao Xie ret = -ENOENT; 668178260b2SMiao Xie } 669a4abeea4SJosef Bacik } while (ret == -EBUSY); 670a4abeea4SJosef Bacik 671a43f7f82SLiu Bo if (ret < 0) 672843fcf35SMiao Xie goto join_fail; 6730f7d52f4SChris Mason 6740b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 675a22285a6SYan, Zheng 676a22285a6SYan, Zheng h->transid = cur_trans->transid; 677a22285a6SYan, Zheng h->transaction = cur_trans; 678b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 67964b63580SJeff Mahoney h->fs_info = root->fs_info; 6807174109cSQu Wenruo 681a698d075SMiao Xie h->type = type; 682ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 683b7ec40d7SChris Mason 684a22285a6SYan, Zheng smp_mb(); 6853296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 6862ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 687abdd2e80SFilipe Manana current->journal_info = h; 6883a45bb20SJeff Mahoney btrfs_commit_transaction(h); 689a22285a6SYan, Zheng goto again; 690a22285a6SYan, Zheng } 6919ed74f2dSJosef Bacik 692b5009945SJosef Bacik if (num_bytes) { 6930b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 6942bcc0328SLiu Bo h->transid, num_bytes, 1); 6950b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 696b5009945SJosef Bacik h->bytes_reserved = num_bytes; 69720dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 698a22285a6SYan, Zheng } 699a22285a6SYan, Zheng 7002a1eb461SJosef Bacik got_it: 701bcf3a3e7SNikolay Borisov if (!current->journal_info) 702a22285a6SYan, Zheng current->journal_info = h; 703fcc99734SQu Wenruo 704fcc99734SQu Wenruo /* 7059c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7069c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7079c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7089c343784SJosef Bacik * value here. 7099c343784SJosef Bacik */ 7109c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7119c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7129c343784SJosef Bacik 7139c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7149c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7159c343784SJosef Bacik } 7169c343784SJosef Bacik 7179c343784SJosef Bacik /* 718fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 719fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 720fcc99734SQu Wenruo * transaction. 721fcc99734SQu Wenruo * 722fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 723fcc99734SQu Wenruo * or we can deadlock. 724fcc99734SQu Wenruo */ 72568075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 72668075ea8SJosef Bacik if (ret) { 72768075ea8SJosef Bacik /* 72868075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 72968075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 73068075ea8SJosef Bacik * not just freed. 73168075ea8SJosef Bacik */ 73268075ea8SJosef Bacik btrfs_end_transaction(h); 73368075ea8SJosef Bacik return ERR_PTR(ret); 73468075ea8SJosef Bacik } 735fcc99734SQu Wenruo 73679154b1bSChris Mason return h; 737843fcf35SMiao Xie 738843fcf35SMiao Xie join_fail: 7390860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7400b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 741843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 742843fcf35SMiao Xie alloc_fail: 743843fcf35SMiao Xie if (num_bytes) 7442ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 74563f018beSNikolay Borisov num_bytes, NULL); 746843fcf35SMiao Xie reserve_fail: 747733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 748843fcf35SMiao Xie return ERR_PTR(ret); 74979154b1bSChris Mason } 75079154b1bSChris Mason 751f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7525aed1dd8SAlexandru Moise unsigned int num_items) 753f9295749SChris Mason { 75408e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 755003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 756f9295749SChris Mason } 757003d7c59SJeff Mahoney 7588eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7598eab77ffSFilipe Manana struct btrfs_root *root, 7607f9fe614SJosef Bacik unsigned int num_items) 7618eab77ffSFilipe Manana { 7627f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7637f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7648eab77ffSFilipe Manana } 7658407aa46SMiao Xie 7667a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 767f9295749SChris Mason { 768003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 769003d7c59SJeff Mahoney true); 770f9295749SChris Mason } 771f9295749SChris Mason 7728d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7730af3d00bSJosef Bacik { 774575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 775003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7760af3d00bSJosef Bacik } 7770af3d00bSJosef Bacik 778d4edf39bSMiao Xie /* 779a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 780a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 781a6d155d2SFilipe Manana */ 782a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 783a6d155d2SFilipe Manana { 784a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 785a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 786a6d155d2SFilipe Manana } 787a6d155d2SFilipe Manana 788a6d155d2SFilipe Manana /* 789d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 790d4edf39bSMiao Xie * 791d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 792d4edf39bSMiao Xie * don't want to start a new one. 793d4edf39bSMiao Xie * 794d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 795d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 796d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 797d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 798d4edf39bSMiao Xie * invoke 799d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 800d4edf39bSMiao Xie */ 801354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 80260376ce4SJosef Bacik { 803575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 804003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 80560376ce4SJosef Bacik } 80660376ce4SJosef Bacik 807d4edf39bSMiao Xie /* 80890b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 809d4edf39bSMiao Xie * 81052042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 811d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 812d4edf39bSMiao Xie * complete. 813d4edf39bSMiao Xie */ 814d4edf39bSMiao Xie struct btrfs_trans_handle * 815d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 816d4edf39bSMiao Xie { 817d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 818d4edf39bSMiao Xie 819575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 820003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8218d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8222ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 823d4edf39bSMiao Xie 824d4edf39bSMiao Xie return trans; 825d4edf39bSMiao Xie } 826d4edf39bSMiao Xie 827d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 828d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 829d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 83089ce8a63SChris Mason { 8315fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8325fd76bf3SOmar Sandoval u64 transid = commit->transid; 8335fd76bf3SOmar Sandoval bool put = false; 8345fd76bf3SOmar Sandoval 8353e738c53SIoannis Angelakopoulos /* 8363e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8373e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8383e738c53SIoannis Angelakopoulos */ 8393e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8403e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8413e738c53SIoannis Angelakopoulos else 8423e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8433e738c53SIoannis Angelakopoulos 8445fd76bf3SOmar Sandoval while (1) { 845d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8465fd76bf3SOmar Sandoval if (put) 8475fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8485fd76bf3SOmar Sandoval 8495fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8505fd76bf3SOmar Sandoval break; 8515fd76bf3SOmar Sandoval 8525fd76bf3SOmar Sandoval /* 8535fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8545fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8555fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8565fd76bf3SOmar Sandoval * transaction. Wait for those. 8575fd76bf3SOmar Sandoval */ 8585fd76bf3SOmar Sandoval 8595fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8605fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8615fd76bf3SOmar Sandoval struct btrfs_transaction, 8625fd76bf3SOmar Sandoval list); 8635fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8645fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8655fd76bf3SOmar Sandoval break; 8665fd76bf3SOmar Sandoval } 8675fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8685fd76bf3SOmar Sandoval put = true; 8695fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8705fd76bf3SOmar Sandoval } 87189ce8a63SChris Mason } 87289ce8a63SChris Mason 8732ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 87446204592SSage Weil { 87546204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8768cd2807fSMiao Xie int ret = 0; 87746204592SSage Weil 87846204592SSage Weil if (transid) { 8790b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 880a4abeea4SJosef Bacik goto out; 88146204592SSage Weil 88246204592SSage Weil /* find specified transaction */ 8830b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8840b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 88546204592SSage Weil if (t->transid == transid) { 88646204592SSage Weil cur_trans = t; 8879b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8888cd2807fSMiao Xie ret = 0; 88946204592SSage Weil break; 89046204592SSage Weil } 8918cd2807fSMiao Xie if (t->transid > transid) { 8928cd2807fSMiao Xie ret = 0; 89346204592SSage Weil break; 89446204592SSage Weil } 8958cd2807fSMiao Xie } 8960b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 89742383020SSage Weil 89842383020SSage Weil /* 89942383020SSage Weil * The specified transaction doesn't exist, or we 90042383020SSage Weil * raced with btrfs_commit_transaction 90142383020SSage Weil */ 90242383020SSage Weil if (!cur_trans) { 9030b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 90442383020SSage Weil ret = -EINVAL; 9058cd2807fSMiao Xie goto out; 90642383020SSage Weil } 90746204592SSage Weil } else { 90846204592SSage Weil /* find newest transaction that is committing | committed */ 9090b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9100b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 91146204592SSage Weil list) { 9124a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9134a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9143473f3c0SJosef Bacik break; 91546204592SSage Weil cur_trans = t; 9169b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 91746204592SSage Weil break; 91846204592SSage Weil } 91946204592SSage Weil } 9200b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 92146204592SSage Weil if (!cur_trans) 922a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 92346204592SSage Weil } 92446204592SSage Weil 925d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 926724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 927a4abeea4SJosef Bacik out: 92846204592SSage Weil return ret; 92946204592SSage Weil } 93046204592SSage Weil 9312ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 93237d1aeeeSChris Mason { 9332ff7e61eSJeff Mahoney wait_current_trans(fs_info); 93437d1aeeeSChris Mason } 93537d1aeeeSChris Mason 9368a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9378929ecfaSYan, Zheng { 9382ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9390b246afaSJeff Mahoney 94064403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9418a8f4deaSNikolay Borisov return true; 94236ba022aSJosef Bacik 9432ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9448929ecfaSYan, Zheng } 9458929ecfaSYan, Zheng 946a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9478929ecfaSYan, Zheng { 9488929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9498929ecfaSYan, Zheng 9503296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 951e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 952a2633b6aSNikolay Borisov return true; 9538929ecfaSYan, Zheng 9542ff7e61eSJeff Mahoney return should_end_transaction(trans); 9558929ecfaSYan, Zheng } 9568929ecfaSYan, Zheng 957dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 958dc60c525SNikolay Borisov 9590e34693fSNikolay Borisov { 960dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 961dc60c525SNikolay Borisov 9620e34693fSNikolay Borisov if (!trans->block_rsv) { 9630e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9640e34693fSNikolay Borisov return; 9650e34693fSNikolay Borisov } 9660e34693fSNikolay Borisov 9670e34693fSNikolay Borisov if (!trans->bytes_reserved) 9680e34693fSNikolay Borisov return; 9690e34693fSNikolay Borisov 9700e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9710e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9720e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9730e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 97463f018beSNikolay Borisov trans->bytes_reserved, NULL); 9750e34693fSNikolay Borisov trans->bytes_reserved = 0; 9760e34693fSNikolay Borisov } 9770e34693fSNikolay Borisov 97889ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9793a45bb20SJeff Mahoney int throttle) 98079154b1bSChris Mason { 9813a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9828929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9834edc2ca3SDave Jones int err = 0; 984d6e4a428SChris Mason 985b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 986b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9872a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9882a1eb461SJosef Bacik return 0; 9892a1eb461SJosef Bacik } 9902a1eb461SJosef Bacik 991dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9924c13d758SJosef Bacik trans->block_rsv = NULL; 993c5567237SArne Jansen 9946c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 995ea658badSJosef Bacik 9964fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9974fbcdf66SFilipe Manana 9980860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9990b246afaSJeff Mahoney sb_end_intwrite(info->sb); 10006df7881aSJosef Bacik 10018929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 100213c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 100313c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10040860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 100589ce8a63SChris Mason 1006093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1007e1489b4fSIoannis Angelakopoulos 10085a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1009e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1010e1489b4fSIoannis Angelakopoulos 1011724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10129ed74f2dSJosef Bacik 10139ed74f2dSJosef Bacik if (current->journal_info == trans) 10149ed74f2dSJosef Bacik current->journal_info = NULL; 1015ab78c84dSChris Mason 101624bbcf04SYan, Zheng if (throttle) 10172ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 101824bbcf04SYan, Zheng 101984961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10204e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1021fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1022fbabd4a3SJosef Bacik err = trans->aborted; 1023fbabd4a3SJosef Bacik else 1024fbabd4a3SJosef Bacik err = -EROFS; 10254e121c06SJosef Bacik } 102649b25e05SJeff Mahoney 10274edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10284edc2ca3SDave Jones return err; 102979154b1bSChris Mason } 103079154b1bSChris Mason 10313a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 103289ce8a63SChris Mason { 10333a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 103489ce8a63SChris Mason } 103589ce8a63SChris Mason 10363a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 103789ce8a63SChris Mason { 10383a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 103916cdcec7SMiao Xie } 104016cdcec7SMiao Xie 1041d352ac68SChris Mason /* 1042d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1043d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1044690587d1SChris Mason * those extents are sent to disk but does not wait on them 1045d352ac68SChris Mason */ 10462ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10478cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 104879154b1bSChris Mason { 1049777e6bd7SChris Mason int err = 0; 10507c4452b9SChris Mason int werr = 0; 10510b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1052e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1053777e6bd7SChris Mason u64 start = 0; 10545f39d397SChris Mason u64 end; 10557c4452b9SChris Mason 10566300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10571728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1058e6138876SJosef Bacik mark, &cached_state)) { 1059663dfbb0SFilipe Manana bool wait_writeback = false; 1060663dfbb0SFilipe Manana 1061663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1062663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1063210aa277SDavid Sterba mark, &cached_state); 1064663dfbb0SFilipe Manana /* 1065663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1066663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1067663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1068663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1069bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1070bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1071bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1072bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1073bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1074663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 107541e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1076663dfbb0SFilipe Manana */ 1077663dfbb0SFilipe Manana if (err == -ENOMEM) { 1078663dfbb0SFilipe Manana err = 0; 1079663dfbb0SFilipe Manana wait_writeback = true; 1080663dfbb0SFilipe Manana } 1081663dfbb0SFilipe Manana if (!err) 10821728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10837c4452b9SChris Mason if (err) 10847c4452b9SChris Mason werr = err; 1085663dfbb0SFilipe Manana else if (wait_writeback) 1086663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1087e38e2ed7SFilipe Manana free_extent_state(cached_state); 1088663dfbb0SFilipe Manana cached_state = NULL; 10891728366eSJosef Bacik cond_resched(); 10901728366eSJosef Bacik start = end + 1; 10917c4452b9SChris Mason } 10926300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1093690587d1SChris Mason return werr; 1094690587d1SChris Mason } 1095690587d1SChris Mason 1096690587d1SChris Mason /* 1097690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1098690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1099690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1100690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1101690587d1SChris Mason */ 1102bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1103bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1104690587d1SChris Mason { 1105690587d1SChris Mason int err = 0; 1106690587d1SChris Mason int werr = 0; 11070b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1108e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1109690587d1SChris Mason u64 start = 0; 1110690587d1SChris Mason u64 end; 1111690587d1SChris Mason 11121728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1113e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1114663dfbb0SFilipe Manana /* 1115663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1116663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1117663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1118663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1119663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 112041e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1121663dfbb0SFilipe Manana */ 1122663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1123bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1124663dfbb0SFilipe Manana if (err == -ENOMEM) 1125663dfbb0SFilipe Manana err = 0; 1126663dfbb0SFilipe Manana if (!err) 11271728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1128777e6bd7SChris Mason if (err) 1129777e6bd7SChris Mason werr = err; 1130e38e2ed7SFilipe Manana free_extent_state(cached_state); 1131e38e2ed7SFilipe Manana cached_state = NULL; 1132777e6bd7SChris Mason cond_resched(); 11331728366eSJosef Bacik start = end + 1; 1134777e6bd7SChris Mason } 11357c4452b9SChris Mason if (err) 11367c4452b9SChris Mason werr = err; 1137bf89d38fSJeff Mahoney return werr; 1138bf89d38fSJeff Mahoney } 1139656f30dbSFilipe Manana 1140b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1141bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1142bf89d38fSJeff Mahoney { 1143bf89d38fSJeff Mahoney bool errors = false; 1144bf89d38fSJeff Mahoney int err; 1145bf89d38fSJeff Mahoney 1146bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1147bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1148bf89d38fSJeff Mahoney errors = true; 1149bf89d38fSJeff Mahoney 1150bf89d38fSJeff Mahoney if (errors && !err) 1151bf89d38fSJeff Mahoney err = -EIO; 1152bf89d38fSJeff Mahoney return err; 1153bf89d38fSJeff Mahoney } 1154bf89d38fSJeff Mahoney 1155bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1156bf89d38fSJeff Mahoney { 1157bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1158bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1159bf89d38fSJeff Mahoney bool errors = false; 1160bf89d38fSJeff Mahoney int err; 1161bf89d38fSJeff Mahoney 1162bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1163bf89d38fSJeff Mahoney 1164bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1165656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11660b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1167656f30dbSFilipe Manana errors = true; 1168656f30dbSFilipe Manana 1169656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11700b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1171656f30dbSFilipe Manana errors = true; 1172656f30dbSFilipe Manana 1173bf89d38fSJeff Mahoney if (errors && !err) 1174bf89d38fSJeff Mahoney err = -EIO; 1175bf89d38fSJeff Mahoney return err; 117679154b1bSChris Mason } 117779154b1bSChris Mason 1178690587d1SChris Mason /* 1179c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1180c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1181c9b577c0SNikolay Borisov * log commit. 1182c9b577c0SNikolay Borisov * 1183c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1184690587d1SChris Mason */ 118570458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1186d0c803c4SChris Mason { 1187663dfbb0SFilipe Manana int ret; 1188c9b577c0SNikolay Borisov int ret2; 1189c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 119070458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1191c9b577c0SNikolay Borisov struct blk_plug plug; 1192663dfbb0SFilipe Manana 1193c9b577c0SNikolay Borisov blk_start_plug(&plug); 1194c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1195c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1196c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1197c9b577c0SNikolay Borisov 119841e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1199663dfbb0SFilipe Manana 1200c9b577c0SNikolay Borisov if (ret) 1201663dfbb0SFilipe Manana return ret; 1202c9b577c0SNikolay Borisov else if (ret2) 1203c9b577c0SNikolay Borisov return ret2; 1204c9b577c0SNikolay Borisov else 1205c9b577c0SNikolay Borisov return 0; 1206d0c803c4SChris Mason } 1207d0c803c4SChris Mason 1208d352ac68SChris Mason /* 1209d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1210d352ac68SChris Mason * 1211d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1212d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1213d352ac68SChris Mason * allocation tree. 1214d352ac68SChris Mason * 1215d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1216d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1217d352ac68SChris Mason */ 12180b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 121979154b1bSChris Mason struct btrfs_root *root) 122079154b1bSChris Mason { 122179154b1bSChris Mason int ret; 12220b86a832SChris Mason u64 old_root_bytenr; 122386b9f2ecSYan, Zheng u64 old_root_used; 12240b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12250b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 122679154b1bSChris Mason 122786b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 122856bec294SChris Mason 122979154b1bSChris Mason while (1) { 12300b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 123186b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1232ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 123379154b1bSChris Mason break; 123487ef2bb4SChris Mason 12355d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 123679154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12370b86a832SChris Mason &root->root_key, 12380b86a832SChris Mason &root->root_item); 123949b25e05SJeff Mahoney if (ret) 124049b25e05SJeff Mahoney return ret; 124156bec294SChris Mason 124286b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1243e7070be1SJosef Bacik } 1244276e680dSYan Zheng 12450b86a832SChris Mason return 0; 12460b86a832SChris Mason } 12470b86a832SChris Mason 1248d352ac68SChris Mason /* 1249d352ac68SChris Mason * update all the cowonly tree roots on disk 125049b25e05SJeff Mahoney * 125149b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 125249b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 125349b25e05SJeff Mahoney * to clean up the delayed refs. 1254d352ac68SChris Mason */ 12559386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12560b86a832SChris Mason { 12579386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1258ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12591bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12600b86a832SChris Mason struct list_head *next; 126184234f3aSYan Zheng struct extent_buffer *eb; 126256bec294SChris Mason int ret; 126384234f3aSYan Zheng 1264dfba78dcSFilipe Manana /* 1265dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1266dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1267dfba78dcSFilipe Manana */ 1268dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1269dfba78dcSFilipe Manana 127084234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 127149b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12729631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 127384234f3aSYan Zheng btrfs_tree_unlock(eb); 127484234f3aSYan Zheng free_extent_buffer(eb); 12750b86a832SChris Mason 127649b25e05SJeff Mahoney if (ret) 127749b25e05SJeff Mahoney return ret; 127849b25e05SJeff Mahoney 1279196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1280c16ce190SJosef Bacik if (ret) 1281c16ce190SJosef Bacik return ret; 12822b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1283c16ce190SJosef Bacik if (ret) 1284c16ce190SJosef Bacik return ret; 1285280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1286c16ce190SJosef Bacik if (ret) 1287c16ce190SJosef Bacik return ret; 1288546adb0dSJan Schmidt 1289bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1290dcdf7f6dSJosef Bacik if (ret) 1291dcdf7f6dSJosef Bacik return ret; 1292dcdf7f6dSJosef Bacik 1293ea526d18SJosef Bacik again: 12940b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12952ff7e61eSJeff Mahoney struct btrfs_root *root; 12960b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12970b86a832SChris Mason list_del_init(next); 12980b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1299e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 130087ef2bb4SChris Mason 13019e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13029e351cc8SJosef Bacik &trans->transaction->switch_commits); 130349b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 130449b25e05SJeff Mahoney if (ret) 130549b25e05SJeff Mahoney return ret; 1306488bc2a2SJosef Bacik } 1307488bc2a2SJosef Bacik 1308488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1309c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1310ea526d18SJosef Bacik if (ret) 1311ea526d18SJosef Bacik return ret; 1312276e680dSYan Zheng 13131bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13145742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1315ea526d18SJosef Bacik if (ret) 1316ea526d18SJosef Bacik return ret; 1317488bc2a2SJosef Bacik 1318488bc2a2SJosef Bacik /* 1319488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1320488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1321488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1322488bc2a2SJosef Bacik * everything gets run. 1323488bc2a2SJosef Bacik */ 1324c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1325ea526d18SJosef Bacik if (ret) 1326ea526d18SJosef Bacik return ret; 1327ea526d18SJosef Bacik } 1328ea526d18SJosef Bacik 1329ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1330ea526d18SJosef Bacik goto again; 1331ea526d18SJosef Bacik 13329f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13339f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13349f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13358dabb742SStefan Behrens 133679154b1bSChris Mason return 0; 133779154b1bSChris Mason } 133879154b1bSChris Mason 1339d352ac68SChris Mason /* 1340b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1341b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1342b4be6aefSJosef Bacik */ 1343b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1344b4be6aefSJosef Bacik { 1345b4be6aefSJosef Bacik /* 1346b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1347b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1348b4be6aefSJosef Bacik * up. 1349b4be6aefSJosef Bacik */ 1350b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1351b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1352b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1353b4be6aefSJosef Bacik struct btrfs_root, 1354b4be6aefSJosef Bacik root_list); 1355b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1356b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1357b4be6aefSJosef Bacik return; 1358b4be6aefSJosef Bacik } 1359b4be6aefSJosef Bacik } 1360b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1361b4be6aefSJosef Bacik 1362b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1363b4be6aefSJosef Bacik } 1364b4be6aefSJosef Bacik 1365b4be6aefSJosef Bacik /* 1366d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1367d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1368d352ac68SChris Mason * be deleted 1369d352ac68SChris Mason */ 1370cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13715eda7b5eSChris Mason { 13720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13730b246afaSJeff Mahoney 13740b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1375dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1376dc9492c1SJosef Bacik btrfs_grab_root(root); 1377b4be6aefSJosef Bacik 1378b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1379b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1380b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1381b4be6aefSJosef Bacik else 13820b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1383dc9492c1SJosef Bacik } 13840b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13855eda7b5eSChris Mason } 13865eda7b5eSChris Mason 1387d352ac68SChris Mason /* 1388dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1389dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1390d352ac68SChris Mason */ 13917e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13920f7d52f4SChris Mason { 13937e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1394fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1395fc7cbcd4SDavid Sterba int i; 1396fc7cbcd4SDavid Sterba int ret; 139754aa1f4dSChris Mason 1398dfba78dcSFilipe Manana /* 1399dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1400dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1401dfba78dcSFilipe Manana */ 1402dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1403dfba78dcSFilipe Manana 1404fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1405fc7cbcd4SDavid Sterba while (1) { 1406fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1407fc7cbcd4SDavid Sterba (void **)gang, 0, 1408fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1409fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1410fc7cbcd4SDavid Sterba if (ret == 0) 1411fc7cbcd4SDavid Sterba break; 1412fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1413fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1414fc7cbcd4SDavid Sterba int ret2; 14154f4317c1SJosef Bacik 1416dfba78dcSFilipe Manana /* 1417dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1418dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1419dfba78dcSFilipe Manana */ 1420dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1421dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1422dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1423dfba78dcSFilipe Manana 1424fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14252619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14260f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1427fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 142831153d81SYan Zheng 1429e02119d5SChris Mason btrfs_free_log(trans, root); 1430fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1431fc7cbcd4SDavid Sterba if (ret2) 1432fc7cbcd4SDavid Sterba return ret2; 1433e02119d5SChris Mason 1434fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 143527cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1436c7548af6SChris Mason smp_mb__after_atomic(); 1437f1ebcc74SLiu Bo 1438978d910dSYan Zheng if (root->commit_root != root->node) { 14399e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14409e351cc8SJosef Bacik &trans->transaction->switch_commits); 1441fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1442fc7cbcd4SDavid Sterba root->node); 1443978d910dSYan Zheng } 144431153d81SYan Zheng 1445fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1446fc7cbcd4SDavid Sterba &root->root_key, 1447fc7cbcd4SDavid Sterba &root->root_item); 1448fc7cbcd4SDavid Sterba if (ret2) 1449fc7cbcd4SDavid Sterba return ret2; 1450fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1451733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14520f7d52f4SChris Mason } 1453fc7cbcd4SDavid Sterba } 1454fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14554f4317c1SJosef Bacik return 0; 14560f7d52f4SChris Mason } 14570f7d52f4SChris Mason 1458d352ac68SChris Mason /* 1459de78b51aSEric Sandeen * defrag a given btree. 1460de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1461d352ac68SChris Mason */ 1462de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1463e9d0b13bSChris Mason { 1464e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1465e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14668929ecfaSYan, Zheng int ret; 1467e9d0b13bSChris Mason 146827cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1469e9d0b13bSChris Mason return 0; 14708929ecfaSYan, Zheng 14716b80053dSChris Mason while (1) { 14728929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14736819703fSDavid Sterba if (IS_ERR(trans)) { 14746819703fSDavid Sterba ret = PTR_ERR(trans); 14756819703fSDavid Sterba break; 14766819703fSDavid Sterba } 14778929ecfaSYan, Zheng 1478de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14798929ecfaSYan, Zheng 14803a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14812ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1482e9d0b13bSChris Mason cond_resched(); 1483e9d0b13bSChris Mason 1484ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1485e9d0b13bSChris Mason break; 1486210549ebSDavid Sterba 1487ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1488ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1489210549ebSDavid Sterba ret = -EAGAIN; 1490210549ebSDavid Sterba break; 1491210549ebSDavid Sterba } 1492e9d0b13bSChris Mason } 149327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14948929ecfaSYan, Zheng return ret; 1495e9d0b13bSChris Mason } 1496e9d0b13bSChris Mason 1497d352ac68SChris Mason /* 14986426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14996426c7adSQu Wenruo * 15006426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 15016426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 15026426c7adSQu Wenruo * qgroup works. 15036426c7adSQu Wenruo */ 15046426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15056426c7adSQu Wenruo struct btrfs_root *src, 15066426c7adSQu Wenruo struct btrfs_root *parent, 15076426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15086426c7adSQu Wenruo u64 dst_objectid) 15096426c7adSQu Wenruo { 15106426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15116426c7adSQu Wenruo int ret; 15126426c7adSQu Wenruo 15136426c7adSQu Wenruo /* 15146426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15156426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15166426c7adSQu Wenruo * kick in anyway. 15176426c7adSQu Wenruo */ 15189ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15196426c7adSQu Wenruo return 0; 15206426c7adSQu Wenruo 15216426c7adSQu Wenruo /* 152252042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15234d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15244d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15254d31778aSQu Wenruo * item. 15264d31778aSQu Wenruo */ 15271c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15281c442d22SJosef Bacik if (ret) 15291c442d22SJosef Bacik return ret; 15304d31778aSQu Wenruo 15314d31778aSQu Wenruo /* 15322a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15332a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15342a4d84c1SJosef Bacik * 15352a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15362a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15372a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15382a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15392a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15402a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15412a4d84c1SJosef Bacik */ 15422a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15432a4d84c1SJosef Bacik if (ret) { 15442a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 154544365827SNaohiro Aota return ret; 15462a4d84c1SJosef Bacik } 15472a4d84c1SJosef Bacik 15487e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15496426c7adSQu Wenruo if (ret) 15506426c7adSQu Wenruo goto out; 1551460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15526426c7adSQu Wenruo if (ret < 0) 15536426c7adSQu Wenruo goto out; 15546426c7adSQu Wenruo 15556426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1556a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15576426c7adSQu Wenruo inherit); 15586426c7adSQu Wenruo if (ret < 0) 15596426c7adSQu Wenruo goto out; 15606426c7adSQu Wenruo 15616426c7adSQu Wenruo /* 15626426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15636426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15646426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15656426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15666426c7adSQu Wenruo * 2) write all btree blocks onto disk 15676426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15686426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15696426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15706426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15716426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15726426c7adSQu Wenruo */ 15739386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15746426c7adSQu Wenruo if (ret) 15756426c7adSQu Wenruo goto out; 1576889bfa39SJosef Bacik switch_commit_roots(trans); 157770458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15786426c7adSQu Wenruo if (ret) 1579f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15806426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15816426c7adSQu Wenruo 15826426c7adSQu Wenruo out: 15836426c7adSQu Wenruo /* 15846426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15856426c7adSQu Wenruo * last_trans == cur_transid. 15866426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15876426c7adSQu Wenruo * insert_dir_item() 15886426c7adSQu Wenruo */ 15896426c7adSQu Wenruo if (!ret) 15901c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15916426c7adSQu Wenruo return ret; 15926426c7adSQu Wenruo } 15936426c7adSQu Wenruo 15946426c7adSQu Wenruo /* 1595d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1596aec8030aSMiao Xie * transaction commit. This does the actual creation. 1597aec8030aSMiao Xie * 1598aec8030aSMiao Xie * Note: 1599aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1600aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1601aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1602d352ac68SChris Mason */ 160380b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16043063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16053063d29fSChris Mason { 160608d50ca3SNikolay Borisov 160708d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16083063d29fSChris Mason struct btrfs_key key; 160980b6794dSChris Mason struct btrfs_root_item *new_root_item; 16103063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16113063d29fSChris Mason struct btrfs_root *root = pending->root; 16126bdb72deSSage Weil struct btrfs_root *parent_root; 161398c9942aSLiu Bo struct btrfs_block_rsv *rsv; 16146bdb72deSSage Weil struct inode *parent_inode; 161542874b3dSMiao Xie struct btrfs_path *path; 161642874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1617a22285a6SYan, Zheng struct dentry *dentry; 16183063d29fSChris Mason struct extent_buffer *tmp; 1619925baeddSChris Mason struct extent_buffer *old; 162095582b00SDeepa Dinamani struct timespec64 cur_time; 1621aec8030aSMiao Xie int ret = 0; 1622d68fc57bSYan, Zheng u64 to_reserve = 0; 16236bdb72deSSage Weil u64 index = 0; 1624a22285a6SYan, Zheng u64 objectid; 1625b83cc969SLi Zefan u64 root_flags; 16263063d29fSChris Mason 16278546b570SDavid Sterba ASSERT(pending->path); 16288546b570SDavid Sterba path = pending->path; 162942874b3dSMiao Xie 1630b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1631b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1632a22285a6SYan, Zheng 1633543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1634aec8030aSMiao Xie if (pending->error) 16356fa9700eSMiao Xie goto no_free_objectid; 16363063d29fSChris Mason 1637d6726335SQu Wenruo /* 1638d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1639d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1640d6726335SQu Wenruo */ 1641d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1642d6726335SQu Wenruo 1643147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1644d68fc57bSYan, Zheng 1645d68fc57bSYan, Zheng if (to_reserve > 0) { 16469270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1647aec8030aSMiao Xie &pending->block_rsv, 164808e007d2SMiao Xie to_reserve, 164908e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1650aec8030aSMiao Xie if (pending->error) 1651d6726335SQu Wenruo goto clear_skip_qgroup; 1652d68fc57bSYan, Zheng } 1653d68fc57bSYan, Zheng 16543063d29fSChris Mason key.objectid = objectid; 1655a22285a6SYan, Zheng key.offset = (u64)-1; 1656a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16573063d29fSChris Mason 16586fa9700eSMiao Xie rsv = trans->block_rsv; 1659a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16602382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16610b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 166288d3a5aaSJosef Bacik trans->transid, 166388d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1664a22285a6SYan, Zheng dentry = pending->dentry; 1665e9662f70SMiao Xie parent_inode = pending->dir; 1666a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1667f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1668f0118cb6SJosef Bacik if (ret) 1669f0118cb6SJosef Bacik goto fail; 1670c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 167104b285f3SDeepa Dinamani 16726bdb72deSSage Weil /* 16736bdb72deSSage Weil * insert the directory item 16746bdb72deSSage Weil */ 1675877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 167649b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 167742874b3dSMiao Xie 167842874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 167942874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16804a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 168142874b3dSMiao Xie dentry->d_name.name, 168242874b3dSMiao Xie dentry->d_name.len, 0); 168342874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1684fe66a05aSChris Mason pending->error = -EEXIST; 1685aec8030aSMiao Xie goto dir_item_existed; 168642874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 168742874b3dSMiao Xie ret = PTR_ERR(dir_item); 168866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16898732d44fSMiao Xie goto fail; 169079787eaaSJeff Mahoney } 169142874b3dSMiao Xie btrfs_release_path(path); 16926bdb72deSSage Weil 1693e999376fSChris Mason /* 1694e999376fSChris Mason * pull in the delayed directory update 1695e999376fSChris Mason * and the delayed inode item 1696e999376fSChris Mason * otherwise we corrupt the FS during 1697e999376fSChris Mason * snapshot 1698e999376fSChris Mason */ 1699e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 17008732d44fSMiao Xie if (ret) { /* Transaction aborted */ 170166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17028732d44fSMiao Xie goto fail; 17038732d44fSMiao Xie } 1704e999376fSChris Mason 1705f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1706f0118cb6SJosef Bacik if (ret) { 1707f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1708f0118cb6SJosef Bacik goto fail; 1709f0118cb6SJosef Bacik } 17106bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17116bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 171208fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17136bdb72deSSage Weil 1714b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1715b83cc969SLi Zefan if (pending->readonly) 1716b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1717b83cc969SLi Zefan else 1718b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1719b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1720b83cc969SLi Zefan 17218ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17228ea05e3aSAlexander Block trans->transid); 1723807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17248ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17258ea05e3aSAlexander Block BTRFS_UUID_SIZE); 172670023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 172770023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 172870023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17298ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17308ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17318ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17328ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 173370023da2SStefan Behrens } 17343cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17353cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 173670023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17378ea05e3aSAlexander Block 1738925baeddSChris Mason old = btrfs_lock_root_node(root); 17399631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17409631e4ccSJosef Bacik BTRFS_NESTING_COW); 174179787eaaSJeff Mahoney if (ret) { 174279787eaaSJeff Mahoney btrfs_tree_unlock(old); 174379787eaaSJeff Mahoney free_extent_buffer(old); 174466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17458732d44fSMiao Xie goto fail; 174679787eaaSJeff Mahoney } 174749b25e05SJeff Mahoney 174849b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 174979787eaaSJeff Mahoney /* clean up in any case */ 1750925baeddSChris Mason btrfs_tree_unlock(old); 1751925baeddSChris Mason free_extent_buffer(old); 17528732d44fSMiao Xie if (ret) { 175366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17548732d44fSMiao Xie goto fail; 17558732d44fSMiao Xie } 1756f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 175727cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1758f1ebcc74SLiu Bo smp_wmb(); 1759f1ebcc74SLiu Bo 17605d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1761a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1762a22285a6SYan, Zheng key.offset = trans->transid; 1763a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1764925baeddSChris Mason btrfs_tree_unlock(tmp); 17653063d29fSChris Mason free_extent_buffer(tmp); 17668732d44fSMiao Xie if (ret) { 176766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17688732d44fSMiao Xie goto fail; 17698732d44fSMiao Xie } 17700660b5afSChris Mason 1771a22285a6SYan, Zheng /* 1772a22285a6SYan, Zheng * insert root back/forward references 1773a22285a6SYan, Zheng */ 17746025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1775a22285a6SYan, Zheng parent_root->root_key.objectid, 17764a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1777a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 17788732d44fSMiao Xie if (ret) { 177966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17808732d44fSMiao Xie goto fail; 17818732d44fSMiao Xie } 1782a22285a6SYan, Zheng 1783a22285a6SYan, Zheng key.offset = (u64)-1; 17842dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 178579787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 178679787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17872d892ccdSFilipe Manana pending->snap = NULL; 178866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17898732d44fSMiao Xie goto fail; 179079787eaaSJeff Mahoney } 1791d68fc57bSYan, Zheng 179249b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17938732d44fSMiao Xie if (ret) { 179466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17958732d44fSMiao Xie goto fail; 17968732d44fSMiao Xie } 1797361048f5SMiao Xie 17986426c7adSQu Wenruo /* 17996426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 18006426c7adSQu Wenruo * snapshot hack to do fast snapshot. 18016426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 18026426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 18036426c7adSQu Wenruo */ 18046426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18056426c7adSQu Wenruo pending->inherit, objectid); 18066426c7adSQu Wenruo if (ret < 0) 18076426c7adSQu Wenruo goto fail; 18086426c7adSQu Wenruo 1809684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1810684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1811684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 181242874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18139c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18148732d44fSMiao Xie if (ret) { 181566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18168732d44fSMiao Xie goto fail; 18178732d44fSMiao Xie } 181842874b3dSMiao Xie 18196ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 182042874b3dSMiao Xie dentry->d_name.len * 2); 1821c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1822c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1823729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1824dd5f9615SStefan Behrens if (ret) { 182566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1826dd5f9615SStefan Behrens goto fail; 1827dd5f9615SStefan Behrens } 1828807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1829807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1830cdb345a8SLu Fengqi objectid); 1831dd5f9615SStefan Behrens if (ret) { 183266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1833dd5f9615SStefan Behrens goto fail; 1834dd5f9615SStefan Behrens } 1835dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1836cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1837dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1838dd5f9615SStefan Behrens objectid); 1839dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 184066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1841dd5f9615SStefan Behrens goto fail; 1842dd5f9615SStefan Behrens } 1843dd5f9615SStefan Behrens } 1844d6726335SQu Wenruo 18453063d29fSChris Mason fail: 1846aec8030aSMiao Xie pending->error = ret; 1847aec8030aSMiao Xie dir_item_existed: 184898c9942aSLiu Bo trans->block_rsv = rsv; 18492382c5ccSLiu Bo trans->bytes_reserved = 0; 1850d6726335SQu Wenruo clear_skip_qgroup: 1851d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 18526fa9700eSMiao Xie no_free_objectid: 18536fa9700eSMiao Xie kfree(new_root_item); 1854b0c0ea63SDavid Sterba pending->root_item = NULL; 185542874b3dSMiao Xie btrfs_free_path(path); 18568546b570SDavid Sterba pending->path = NULL; 18578546b570SDavid Sterba 185849b25e05SJeff Mahoney return ret; 18593063d29fSChris Mason } 18603063d29fSChris Mason 1861d352ac68SChris Mason /* 1862d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1863d352ac68SChris Mason */ 186408d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18653063d29fSChris Mason { 1866aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18673063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1868aec8030aSMiao Xie int ret = 0; 18693de4586cSChris Mason 1870aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1871aec8030aSMiao Xie list_del(&pending->list); 187208d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1873aec8030aSMiao Xie if (ret) 1874aec8030aSMiao Xie break; 1875aec8030aSMiao Xie } 1876aec8030aSMiao Xie return ret; 18773de4586cSChris Mason } 18783de4586cSChris Mason 18792ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18805d4f98a2SYan Zheng { 18815d4f98a2SYan Zheng struct btrfs_root_item *root_item; 18825d4f98a2SYan Zheng struct btrfs_super_block *super; 18835d4f98a2SYan Zheng 18840b246afaSJeff Mahoney super = fs_info->super_copy; 18855d4f98a2SYan Zheng 18860b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1887093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1888093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1889093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18905d4f98a2SYan Zheng 18910b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1892093e037cSDavid Sterba super->root = root_item->bytenr; 1893093e037cSDavid Sterba super->generation = root_item->generation; 1894093e037cSDavid Sterba super->root_level = root_item->level; 18950b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1896093e037cSDavid Sterba super->cache_generation = root_item->generation; 189794846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 189894846229SBoris Burkov super->cache_generation = 0; 18990b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1900093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 19015d4f98a2SYan Zheng } 19025d4f98a2SYan Zheng 1903f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1904f36f3042SChris Mason { 19054a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1906f36f3042SChris Mason int ret = 0; 19074a9d8bdeSMiao Xie 1908a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19094a9d8bdeSMiao Xie trans = info->running_transaction; 19104a9d8bdeSMiao Xie if (trans) 19114a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1912a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1913f36f3042SChris Mason return ret; 1914f36f3042SChris Mason } 1915f36f3042SChris Mason 19168929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19178929ecfaSYan, Zheng { 19184a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19198929ecfaSYan, Zheng int ret = 0; 19204a9d8bdeSMiao Xie 1921a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19224a9d8bdeSMiao Xie trans = info->running_transaction; 19234a9d8bdeSMiao Xie if (trans) 19244a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1925a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19268929ecfaSYan, Zheng return ret; 19278929ecfaSYan, Zheng } 19288929ecfaSYan, Zheng 1929fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1930bb9c12c9SSage Weil { 19313a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1932bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1933bb9c12c9SSage Weil 1934fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1935fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1936fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1937bb9c12c9SSage Weil 1938bb9c12c9SSage Weil /* take transaction reference */ 1939bb9c12c9SSage Weil cur_trans = trans->transaction; 19409b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1941bb9c12c9SSage Weil 19423a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19436fc4e354SSage Weil 19446fc4e354SSage Weil /* 1945ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1946ae5d29d4SDavid Sterba * subsequent transaction joins 1947ae5d29d4SDavid Sterba */ 19483e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1949ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1950ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1951ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1952724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1953bb9c12c9SSage Weil } 1954bb9c12c9SSage Weil 195597cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 195649b25e05SJeff Mahoney { 195797cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 195849b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 195949b25e05SJeff Mahoney 1960b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 196149b25e05SJeff Mahoney 196266642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19637b8b92afSJosef Bacik 19640b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 196566b6135bSLiu Bo 196625d8c284SMiao Xie /* 196725d8c284SMiao Xie * If the transaction is removed from the list, it means this 196825d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 196925d8c284SMiao Xie * to call the cleanup function. 197025d8c284SMiao Xie */ 197125d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 197266b6135bSLiu Bo 19730b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19744a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19750b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1976e1489b4fSIoannis Angelakopoulos 1977e1489b4fSIoannis Angelakopoulos /* 1978e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 1979e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 1980e1489b4fSIoannis Angelakopoulos */ 1981e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 1982f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1983f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1984f094ac32SLiu Bo 19850b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1986d7096fc3SJosef Bacik } 1987061dde82SFilipe Manana 1988061dde82SFilipe Manana /* 1989061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 1990061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 1991061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 1992061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 1993061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 1994061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 1995061dde82SFilipe Manana */ 1996061dde82SFilipe Manana list_del_init(&cur_trans->list); 1997061dde82SFilipe Manana 19980b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 199949b25e05SJeff Mahoney 20002ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 200149b25e05SJeff Mahoney 20020b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20030b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20040b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20050b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20064a9d8bdeSMiao Xie 2007e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20080b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2009724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2010724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 201149b25e05SJeff Mahoney 20122e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 201349b25e05SJeff Mahoney 201449b25e05SJeff Mahoney if (current->journal_info == trans) 201549b25e05SJeff Mahoney current->journal_info = NULL; 20160b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 201749b25e05SJeff Mahoney 201849b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 201949b25e05SJeff Mahoney } 202049b25e05SJeff Mahoney 2021c7cc64a9SDavid Sterba /* 2022c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2023c7cc64a9SDavid Sterba * transaction and remove them from the list 2024c7cc64a9SDavid Sterba */ 2025c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2026c7cc64a9SDavid Sterba { 2027c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 202832da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2029c7cc64a9SDavid Sterba 2030c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2031c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2032c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2033c7cc64a9SDavid Sterba } 2034c7cc64a9SDavid Sterba } 2035c7cc64a9SDavid Sterba 203688090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 203782436617SMiao Xie { 2038ce8ea7ccSJosef Bacik /* 2039a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2040ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2041ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2042ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2043ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2044ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2045ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2046a0f0cf83SFilipe Manana * 2047a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2048a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2049a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2050a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2051a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2052a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2053a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2054ce8ea7ccSJosef Bacik */ 205588090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2056a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 205782436617SMiao Xie return 0; 205882436617SMiao Xie } 205982436617SMiao Xie 206088090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 206182436617SMiao Xie { 206288090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20636374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 206482436617SMiao Xie } 206582436617SMiao Xie 206628b21c55SFilipe Manana /* 206728b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 206828b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 206928b21c55SFilipe Manana * and while holding fs_info->trans_lock. 207028b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 207128b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 207228b21c55SFilipe Manana * returns an error. 207328b21c55SFilipe Manana */ 207428b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 207528b21c55SFilipe Manana { 207628b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 207728b21c55SFilipe Manana 207828b21c55SFilipe Manana if (!trans->pending_snapshot) 207928b21c55SFilipe Manana return; 208028b21c55SFilipe Manana 208128b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 208228b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 208328b21c55SFilipe Manana 208428b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 208528b21c55SFilipe Manana } 208628b21c55SFilipe Manana 2087e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2088e55958c8SIoannis Angelakopoulos { 2089e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2090e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2091e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2092e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2093e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2094e55958c8SIoannis Angelakopoulos } 2095e55958c8SIoannis Angelakopoulos 20963a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 209779154b1bSChris Mason { 20983a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 209949b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 21008fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 210125287e0aSMiao Xie int ret; 2102e55958c8SIoannis Angelakopoulos ktime_t start_time; 2103e55958c8SIoannis Angelakopoulos ktime_t interval; 210479154b1bSChris Mason 210535b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21063e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 210735b814f3SNikolay Borisov 2108c52cc7b7SJosef Bacik clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); 2109c52cc7b7SJosef Bacik 21108d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2111bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 211225287e0aSMiao Xie ret = cur_trans->aborted; 21133e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 211425287e0aSMiao Xie } 211549b25e05SJeff Mahoney 2116f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2117f45c752bSJosef Bacik trans->block_rsv = NULL; 2118f45c752bSJosef Bacik 2119e19eb11fSJosef Bacik /* 2120e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2121e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2122e19eb11fSJosef Bacik */ 2123e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2124e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2125e19eb11fSJosef Bacik /* 2126e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2127e19eb11fSJosef Bacik * Any running threads may add more while we are here. 212856bec294SChris Mason */ 2129c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21303e738c53SIoannis Angelakopoulos if (ret) 21313e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2132e19eb11fSJosef Bacik } 213356bec294SChris Mason 21346c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2135ea658badSJosef Bacik 21363204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21371bbc621eSChris Mason int run_it = 0; 21381bbc621eSChris Mason 21391bbc621eSChris Mason /* this mutex is also taken before trying to set 21401bbc621eSChris Mason * block groups readonly. We need to make sure 21411bbc621eSChris Mason * that nobody has set a block group readonly 21421bbc621eSChris Mason * after a extents from that block group have been 21431bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21441bbc621eSChris Mason * will wait for the transaction to commit if it 21453204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21461bbc621eSChris Mason * 21473204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21483204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21491bbc621eSChris Mason * hurt to have more than one go through, but there's no 21501bbc621eSChris Mason * real advantage to it either. 21511bbc621eSChris Mason */ 21520b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21533204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21543204d33cSJosef Bacik &cur_trans->flags)) 21551bbc621eSChris Mason run_it = 1; 21560b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21571bbc621eSChris Mason 2158f9cacae3SNikolay Borisov if (run_it) { 215921217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21603e738c53SIoannis Angelakopoulos if (ret) 21613e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2162f9cacae3SNikolay Borisov } 2163f9cacae3SNikolay Borisov } 21641bbc621eSChris Mason 21650b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21664a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2167d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2168d0c2f4faSFilipe Manana 216928b21c55SFilipe Manana add_pending_snapshot(trans); 217028b21c55SFilipe Manana 21710b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21729b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2173ccd467d6SChris Mason 2174d0c2f4faSFilipe Manana if (trans->in_fsync) 2175d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 21763e738c53SIoannis Angelakopoulos 21773e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 21783e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2179d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2180d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 218115ee9bc7SJosef Bacik 2182bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2183b4924a0fSLiu Bo ret = cur_trans->aborted; 2184b4924a0fSLiu Bo 2185724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 218615ee9bc7SJosef Bacik 218749b25e05SJeff Mahoney return ret; 218879154b1bSChris Mason } 21894313b399SChris Mason 21904a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21910b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 21923e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2193bb9c12c9SSage Weil 21940b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2195d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2196d0c2f4faSFilipe Manana 2197d0c2f4faSFilipe Manana if (trans->in_fsync) 2198d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2199d0c2f4faSFilipe Manana 2200ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2201ccd467d6SChris Mason struct btrfs_transaction, list); 2202d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 22039b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 22040b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2205ccd467d6SChris Mason 2206d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2207d0c2f4faSFilipe Manana 2208bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2209ccd467d6SChris Mason 2210724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22111f9b8c8fSFilipe Manana if (ret) 2212e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2213a4abeea4SJosef Bacik } else { 22140b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2215ccd467d6SChris Mason } 2216a4abeea4SJosef Bacik } else { 22170b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2218cb2d3dadSFilipe Manana /* 2219cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2220cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2221cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2222cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2223cb2d3dadSFilipe Manana */ 222484961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2225cb2d3dadSFilipe Manana ret = -EROFS; 2226e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2227cb2d3dadSFilipe Manana } 2228ccd467d6SChris Mason } 222915ee9bc7SJosef Bacik 2230e55958c8SIoannis Angelakopoulos /* 2231e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2232e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2233e55958c8SIoannis Angelakopoulos */ 2234e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2235e55958c8SIoannis Angelakopoulos 22360860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22370860adfdSMiao Xie 223888090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 223982436617SMiao Xie if (ret) 2240e1489b4fSIoannis Angelakopoulos goto lockdep_release; 224182436617SMiao Xie 2242e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 224349b25e05SJeff Mahoney if (ret) 2244e1489b4fSIoannis Angelakopoulos goto lockdep_release; 224516cdcec7SMiao Xie 22465a9ba670SIoannis Angelakopoulos /* 22475a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22485a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22495a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22505a9ba670SIoannis Angelakopoulos */ 22515a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 22525a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2253581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2254581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2255ed3b3d31SChris Mason 2256581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2257e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2258e1489b4fSIoannis Angelakopoulos if (ret) { 2259e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2260ca469637SMiao Xie goto cleanup_transaction; 2261e1489b4fSIoannis Angelakopoulos } 2262ca469637SMiao Xie 226388090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2264cb7ab021SWang Shilong 226548778179SFilipe Manana /* 226648778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 226748778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 226848778179SFilipe Manana * extents complete we lose logged data after a power failure. 226948778179SFilipe Manana */ 22708b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 227148778179SFilipe Manana wait_event(cur_trans->pending_wait, 227248778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 227348778179SFilipe Manana 22742ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2275ed0ca140SJosef Bacik /* 2276ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2277ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22784a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2279ed0ca140SJosef Bacik */ 22800b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 228128b21c55SFilipe Manana add_pending_snapshot(trans); 22824a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 22830b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2284e1489b4fSIoannis Angelakopoulos 2285e1489b4fSIoannis Angelakopoulos /* 2286e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2287e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2288e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2289e1489b4fSIoannis Angelakopoulos */ 2290e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2291e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2292ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2293ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 229415ee9bc7SJosef Bacik 2295fdfbf020SJosef Bacik /* 22963e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 22973e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 22983e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 22993e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 23003e738c53SIoannis Angelakopoulos */ 23013e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 23023e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 23033e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23043e738c53SIoannis Angelakopoulos 23053e738c53SIoannis Angelakopoulos /* 2306fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2307fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2308fdfbf020SJosef Bacik * kthread could do the work. 2309fdfbf020SJosef Bacik */ 2310fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2311fdfbf020SJosef Bacik 2312bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23132cba30f1SMiao Xie ret = cur_trans->aborted; 23143e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23156cf7f77eSWang Shilong goto scrub_continue; 23162cba30f1SMiao Xie } 23177585717fSChris Mason /* 23187585717fSChris Mason * the reloc mutex makes sure that we stop 23197585717fSChris Mason * the balancing code from coming in and moving 23207585717fSChris Mason * extents around in the middle of the commit 23217585717fSChris Mason */ 23220b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23237585717fSChris Mason 232442874b3dSMiao Xie /* 232542874b3dSMiao Xie * We needn't worry about the delayed items because we will 232642874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 232742874b3dSMiao Xie * core function of the snapshot creation. 232842874b3dSMiao Xie */ 232908d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 233056e9f6eaSDavid Sterba if (ret) 233156e9f6eaSDavid Sterba goto unlock_reloc; 23323063d29fSChris Mason 233342874b3dSMiao Xie /* 233442874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 233542874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 233642874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 233742874b3dSMiao Xie * them. 233842874b3dSMiao Xie * 233942874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 234042874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 234142874b3dSMiao Xie * the nodes and leaves. 234242874b3dSMiao Xie */ 2343e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 234456e9f6eaSDavid Sterba if (ret) 234556e9f6eaSDavid Sterba goto unlock_reloc; 234616cdcec7SMiao Xie 2347c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 234856e9f6eaSDavid Sterba if (ret) 234956e9f6eaSDavid Sterba goto unlock_reloc; 235056bec294SChris Mason 2351e999376fSChris Mason /* 2352e999376fSChris Mason * make sure none of the code above managed to slip in a 2353e999376fSChris Mason * delayed item 2354e999376fSChris Mason */ 2355ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2356e999376fSChris Mason 23572c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2358dc17ff8fSChris Mason 23597e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 236056e9f6eaSDavid Sterba if (ret) 2361dfba78dcSFilipe Manana goto unlock_reloc; 236254aa1f4dSChris Mason 23635d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2364e02119d5SChris Mason * safe to free the root of tree log roots 2365e02119d5SChris Mason */ 23660b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2367e02119d5SChris Mason 23680ed4792aSQu Wenruo /* 23690ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 23700ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 23710ed4792aSQu Wenruo */ 2372460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 237356e9f6eaSDavid Sterba if (ret < 0) 2374dfba78dcSFilipe Manana goto unlock_reloc; 23750ed4792aSQu Wenruo 23769386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 237756e9f6eaSDavid Sterba if (ret) 2378dfba78dcSFilipe Manana goto unlock_reloc; 237954aa1f4dSChris Mason 23802cba30f1SMiao Xie /* 23812cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 23822cba30f1SMiao Xie * update ->aborted, check it. 23832cba30f1SMiao Xie */ 2384bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23852cba30f1SMiao Xie ret = cur_trans->aborted; 2386dfba78dcSFilipe Manana goto unlock_reloc; 23872cba30f1SMiao Xie } 23882cba30f1SMiao Xie 23890b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 23905f39d397SChris Mason 23910b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 23920b246afaSJeff Mahoney fs_info->tree_root->node); 23930b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 23949e351cc8SJosef Bacik &cur_trans->switch_commits); 23955d4f98a2SYan Zheng 23960b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 23970b246afaSJeff Mahoney fs_info->chunk_root->node); 23980b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 23999e351cc8SJosef Bacik &cur_trans->switch_commits); 24009e351cc8SJosef Bacik 2401f7238e50SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2402f7238e50SJosef Bacik btrfs_set_root_node(&fs_info->block_group_root->root_item, 2403f7238e50SJosef Bacik fs_info->block_group_root->node); 2404f7238e50SJosef Bacik list_add_tail(&fs_info->block_group_root->dirty_list, 2405f7238e50SJosef Bacik &cur_trans->switch_commits); 2406f7238e50SJosef Bacik } 2407f7238e50SJosef Bacik 2408889bfa39SJosef Bacik switch_commit_roots(trans); 24095d4f98a2SYan Zheng 2410ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 24111bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 24122ff7e61eSJeff Mahoney update_super_roots(fs_info); 2413e02119d5SChris Mason 24140b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 24150b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 24160b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 24170b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2418ccd467d6SChris Mason 2419bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2420935e5cc9SMiao Xie 24210b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 24220b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2423656f30dbSFilipe Manana 24244fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24254fbcdf66SFilipe Manana 2426dfba78dcSFilipe Manana /* 2427dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2428dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2429dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2430dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2431dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2432dfba78dcSFilipe Manana * writing its superblock. 2433dfba78dcSFilipe Manana */ 2434dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2435dfba78dcSFilipe Manana 24360b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 24374a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 24380b246afaSJeff Mahoney fs_info->running_transaction = NULL; 24390b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 24400b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2441b7ec40d7SChris Mason 24420b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 24433e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2444e6dcd2dcSChris Mason 244570458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 244649b25e05SJeff Mahoney if (ret) { 24470b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 244808748810SDavid Sterba "Error while writing out transaction"); 24490b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 24506cf7f77eSWang Shilong goto scrub_continue; 245149b25e05SJeff Mahoney } 245249b25e05SJeff Mahoney 2453d3575156SNaohiro Aota /* 2454d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2455d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2456d3575156SNaohiro Aota * buffers. 2457d3575156SNaohiro Aota */ 2458d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2459d3575156SNaohiro Aota 2460eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2461e02119d5SChris Mason /* 2462e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2463e02119d5SChris Mason * to go about their business 2464e02119d5SChris Mason */ 24650b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2466c1f32b7cSAnand Jain if (ret) 2467c1f32b7cSAnand Jain goto scrub_continue; 2468e02119d5SChris Mason 2469d0c2f4faSFilipe Manana /* 2470d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2471d0c2f4faSFilipe Manana * which can change it. 2472d0c2f4faSFilipe Manana */ 2473d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2474d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 24753e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2476d0c2f4faSFilipe Manana 24775ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 24784313b399SChris Mason 24793204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 24800b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 248113212b54SZhao Lei 24820b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 24834a9d8bdeSMiao Xie /* 24844a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 24854a9d8bdeSMiao Xie * which can change it. 24864a9d8bdeSMiao Xie */ 24874a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 24882c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 24893e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 24903de4586cSChris Mason 24910b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 249213c5a93eSJosef Bacik list_del_init(&cur_trans->list); 24930b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2494a4abeea4SJosef Bacik 2495724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2496724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 249758176a96SJosef Bacik 24980860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 24990b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2500b2b5ef5cSJan Kara 25012e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 25021abe9b8aSliubo 2503e55958c8SIoannis Angelakopoulos interval = ktime_get_ns() - start_time; 2504e55958c8SIoannis Angelakopoulos 25052ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2506a2de733cSArne Jansen 25079ed74f2dSJosef Bacik if (current->journal_info == trans) 25089ed74f2dSJosef Bacik current->journal_info = NULL; 25099ed74f2dSJosef Bacik 25102c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 251124bbcf04SYan, Zheng 2512e55958c8SIoannis Angelakopoulos update_commit_stats(fs_info, interval); 2513e55958c8SIoannis Angelakopoulos 251479154b1bSChris Mason return ret; 251549b25e05SJeff Mahoney 251656e9f6eaSDavid Sterba unlock_reloc: 251756e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 25183e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 25196cf7f77eSWang Shilong scrub_continue: 25203e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 25213e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25222ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 252349b25e05SJeff Mahoney cleanup_transaction: 2524dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2525c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 25264fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 25270e721106SJosef Bacik trans->block_rsv = NULL; 25280b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 252949b25e05SJeff Mahoney if (current->journal_info == trans) 253049b25e05SJeff Mahoney current->journal_info = NULL; 253197cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 253249b25e05SJeff Mahoney 253349b25e05SJeff Mahoney return ret; 2534e1489b4fSIoannis Angelakopoulos 2535e1489b4fSIoannis Angelakopoulos lockdep_release: 25365a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2537e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2538e1489b4fSIoannis Angelakopoulos goto cleanup_transaction; 25393e738c53SIoannis Angelakopoulos 25403e738c53SIoannis Angelakopoulos lockdep_trans_commit_start_release: 25413e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 25423e738c53SIoannis Angelakopoulos btrfs_end_transaction(trans); 25433e738c53SIoannis Angelakopoulos return ret; 254479154b1bSChris Mason } 254579154b1bSChris Mason 2546d352ac68SChris Mason /* 25479d1a2a3aSDavid Sterba * return < 0 if error 25489d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 25499d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 25509d1a2a3aSDavid Sterba * 25519d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 25529d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 25539d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 25549d1a2a3aSDavid Sterba * few seconds later. 2555d352ac68SChris Mason */ 255633c44184SJosef Bacik int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2557e9d0b13bSChris Mason { 255833c44184SJosef Bacik struct btrfs_root *root; 25599d1a2a3aSDavid Sterba int ret; 2560e9d0b13bSChris Mason 2561a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 25629d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 25639d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 25649d1a2a3aSDavid Sterba return 0; 25659d1a2a3aSDavid Sterba } 25669d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 25679d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2568cfad392bSJosef Bacik list_del_init(&root->root_list); 2569a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 25705d4f98a2SYan Zheng 25714fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 257276dda93cSYan, Zheng 257316cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 257416cdcec7SMiao Xie 257576dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 257676dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 25770078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 257876dda93cSYan, Zheng else 25790078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 258032471dc2SDavid Sterba 2581dc9492c1SJosef Bacik btrfs_put_root(root); 25826596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2583e9d0b13bSChris Mason } 2584572d9ab7SDavid Sterba 2585956504a3SJosef Bacik int __init btrfs_transaction_init(void) 2586956504a3SJosef Bacik { 2587956504a3SJosef Bacik btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2588956504a3SJosef Bacik sizeof(struct btrfs_trans_handle), 0, 2589956504a3SJosef Bacik SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2590956504a3SJosef Bacik if (!btrfs_trans_handle_cachep) 2591956504a3SJosef Bacik return -ENOMEM; 2592956504a3SJosef Bacik return 0; 2593956504a3SJosef Bacik } 2594956504a3SJosef Bacik 2595956504a3SJosef Bacik void __cold btrfs_transaction_exit(void) 2596956504a3SJosef Bacik { 2597956504a3SJosef Bacik kmem_cache_destroy(btrfs_trans_handle_cachep); 2598956504a3SJosef Bacik } 2599