1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 679154b1bSChris Mason #include <linux/fs.h> 75a0e3ad6STejun Heo #include <linux/slab.h> 834088780SChris Mason #include <linux/sched.h> 9d3c2fdcfSChris Mason #include <linux/writeback.h> 105f39d397SChris Mason #include <linux/pagemap.h> 115f2cc086SChris Mason #include <linux/blkdev.h> 128ea05e3aSAlexander Block #include <linux/uuid.h> 13e55958c8SIoannis Angelakopoulos #include <linux/timekeeping.h> 14602cbe91SDavid Sterba #include "misc.h" 1579154b1bSChris Mason #include "ctree.h" 1679154b1bSChris Mason #include "disk-io.h" 1779154b1bSChris Mason #include "transaction.h" 18925baeddSChris Mason #include "locking.h" 19e02119d5SChris Mason #include "tree-log.h" 20733f4fbbSStefan Behrens #include "volumes.h" 218dabb742SStefan Behrens #include "dev-replace.h" 22fcebe456SJosef Bacik #include "qgroup.h" 23aac0023cSJosef Bacik #include "block-group.h" 249c343784SJosef Bacik #include "space-info.h" 25d3575156SNaohiro Aota #include "zoned.h" 2679154b1bSChris Mason 27fc7cbcd4SDavid Sterba #define BTRFS_ROOT_TRANS_TAG 0 280f7d52f4SChris Mason 2961c047b5SQu Wenruo /* 3061c047b5SQu Wenruo * Transaction states and transitions 3161c047b5SQu Wenruo * 3261c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 3361c047b5SQu Wenruo * | 3461c047b5SQu Wenruo * | To next stage: 3561c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 3661c047b5SQu Wenruo * V 3761c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 3861c047b5SQu Wenruo * | 3961c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 4061c047b5SQu Wenruo * | start_transaction() variants. 4161c047b5SQu Wenruo * | 4261c047b5SQu Wenruo * | To next stage: 4361c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 4461c047b5SQu Wenruo * | transaction N 4561c047b5SQu Wenruo * V 4661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 4761c047b5SQu Wenruo * | 4861c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 4961c047b5SQu Wenruo * | is one 5061c047b5SQu Wenruo * | 5161c047b5SQu Wenruo * | Then one of the following happes: 5261c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 5361c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 5461c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 5561c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 5661c047b5SQu Wenruo * | 5761c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 5861c047b5SQu Wenruo * | to this running transaction. 5961c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 6061c047b5SQu Wenruo * | transaction N+1. 6161c047b5SQu Wenruo * | 6261c047b5SQu Wenruo * | To next stage: 6361c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 6461c047b5SQu Wenruo * | haven been released. 6561c047b5SQu Wenruo * V 6661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 6761c047b5SQu Wenruo * | 6861c047b5SQu Wenruo * | The heavy lifting transaction work is started. 6961c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 7061c047b5SQu Wenruo * | snapshots, running qgroups. 7161c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 7261c047b5SQu Wenruo * | trees. 7361c047b5SQu Wenruo * | 7461c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 7561c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 7661c047b5SQu Wenruo * | 7761c047b5SQu Wenruo * | To next stage: 7861c047b5SQu Wenruo * | Until all supporting trees are updated. 7961c047b5SQu Wenruo * V 8061c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 8161c047b5SQu Wenruo * | Transaction N+1 8261c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 8361c047b5SQu Wenruo * | need to write them back to disk and update | 8461c047b5SQu Wenruo * | super blocks. | 8561c047b5SQu Wenruo * | | 8661c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 8761c047b5SQu Wenruo * | start. | 8861c047b5SQu Wenruo * | All new start_transaction() calls will be | 8961c047b5SQu Wenruo * | attached to transid N+1. | 9061c047b5SQu Wenruo * | | 9161c047b5SQu Wenruo * | To next stage: | 9261c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 9361c047b5SQu Wenruo * | written to block devices | 9461c047b5SQu Wenruo * V | 9561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 9661c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 9761c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 9861c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 9961c047b5SQu Wenruo */ 100e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1014a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 102bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 103bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1044a9d8bdeSMiao Xie __TRANS_ATTACH | 105a6d155d2SFilipe Manana __TRANS_JOIN | 106a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 107bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1084a9d8bdeSMiao Xie __TRANS_ATTACH | 1094a9d8bdeSMiao Xie __TRANS_JOIN | 110a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 111a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 112d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 113d0c2f4faSFilipe Manana __TRANS_ATTACH | 114d0c2f4faSFilipe Manana __TRANS_JOIN | 115d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 116d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 117bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1184a9d8bdeSMiao Xie __TRANS_ATTACH | 1194a9d8bdeSMiao Xie __TRANS_JOIN | 120a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 121a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1224a9d8bdeSMiao Xie }; 1234a9d8bdeSMiao Xie 124724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 12579154b1bSChris Mason { 1269b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1279b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 128a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1295c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1305c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 13181f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 13281f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1331262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 134ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 135ab8d0fc4SJeff Mahoney "pending csums is %llu", 1361262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1377785a663SFilipe Manana /* 1387785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1397785a663SFilipe Manana * because the transaction was aborted and a commit did not 1407785a663SFilipe Manana * happen (things failed before writing the new superblock 1417785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1427785a663SFilipe Manana * discard the physical locations of the block groups. 1437785a663SFilipe Manana */ 1447785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 14532da5386SDavid Sterba struct btrfs_block_group *cache; 1467785a663SFilipe Manana 1477785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 14832da5386SDavid Sterba struct btrfs_block_group, 1497785a663SFilipe Manana bg_list); 1507785a663SFilipe Manana list_del_init(&cache->bg_list); 1516b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1527785a663SFilipe Manana btrfs_put_block_group(cache); 1537785a663SFilipe Manana } 154bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1554b5faeacSDavid Sterba kfree(transaction); 15679154b1bSChris Mason } 15778fae27eSChris Mason } 15879154b1bSChris Mason 159889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 160817d52f8SJosef Bacik { 161889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 16216916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1639e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 1649e351cc8SJosef Bacik 165dfba78dcSFilipe Manana /* 166dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 167dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 168dfba78dcSFilipe Manana */ 169dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 170dfba78dcSFilipe Manana 1719e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 172d96b3424SFilipe Manana 173d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 174d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 175d96b3424SFilipe Manana 176889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1779e351cc8SJosef Bacik dirty_list) { 1789e351cc8SJosef Bacik list_del_init(&root->dirty_list); 179817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 180817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 18141e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 182370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1839e351cc8SJosef Bacik } 1842b9dbef2SJosef Bacik 1852b9dbef2SJosef Bacik /* We can free old roots now. */ 186889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 187889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 188889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1892b9dbef2SJosef Bacik struct btrfs_root, root_list); 1902b9dbef2SJosef Bacik list_del_init(&root->root_list); 191889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 192889bfa39SJosef Bacik btrfs_free_log(trans, root); 1932b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 194889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1952b9dbef2SJosef Bacik } 196889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 19727d56e62SJosef Bacik 1989e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 199817d52f8SJosef Bacik } 200817d52f8SJosef Bacik 2010860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2020860adfdSMiao Xie unsigned int type) 2030860adfdSMiao Xie { 2040860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2050860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2060860adfdSMiao Xie } 2070860adfdSMiao Xie 2080860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2090860adfdSMiao Xie unsigned int type) 2100860adfdSMiao Xie { 2110860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2120860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2130860adfdSMiao Xie } 2140860adfdSMiao Xie 2150860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2160860adfdSMiao Xie unsigned int type) 2170860adfdSMiao Xie { 2180860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2190860adfdSMiao Xie } 2200860adfdSMiao Xie 2210860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2220860adfdSMiao Xie { 2230860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 224178260b2SMiao Xie } 225178260b2SMiao Xie 226d352ac68SChris Mason /* 22779bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 22879bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 22979bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 23079bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 23179bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 232fb6dea26SJosef Bacik */ 233fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 234fb6dea26SJosef Bacik { 235fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 236fb6dea26SJosef Bacik 237fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 238fb6dea26SJosef Bacik return; 239fb6dea26SJosef Bacik 240fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 24163f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 242fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 243fb6dea26SJosef Bacik } 244fb6dea26SJosef Bacik 245fb6dea26SJosef Bacik /* 246d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 247d352ac68SChris Mason */ 2482ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2492ff7e61eSJeff Mahoney unsigned int type) 25079154b1bSChris Mason { 25179154b1bSChris Mason struct btrfs_transaction *cur_trans; 252a4abeea4SJosef Bacik 25319ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 254d43317dcSChris Mason loop: 25549b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 25684961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 25719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 25849b25e05SJeff Mahoney return -EROFS; 25949b25e05SJeff Mahoney } 26049b25e05SJeff Mahoney 26119ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 262a4abeea4SJosef Bacik if (cur_trans) { 263bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 26419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26549b25e05SJeff Mahoney return cur_trans->aborted; 266871383beSDavid Sterba } 2674a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 268178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 269178260b2SMiao Xie return -EBUSY; 270178260b2SMiao Xie } 2719b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 272a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2730860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 27419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 275e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2765a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 277a4abeea4SJosef Bacik return 0; 278a4abeea4SJosef Bacik } 27919ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 280a4abeea4SJosef Bacik 281354aa0fbSMiao Xie /* 282354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 283354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 284354aa0fbSMiao Xie */ 285354aa0fbSMiao Xie if (type == TRANS_ATTACH) 286354aa0fbSMiao Xie return -ENOENT; 287354aa0fbSMiao Xie 2884a9d8bdeSMiao Xie /* 2894a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 2904a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 2914a9d8bdeSMiao Xie */ 2924a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 2934a9d8bdeSMiao Xie 2944b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 295db5b493aSTsutomu Itoh if (!cur_trans) 296db5b493aSTsutomu Itoh return -ENOMEM; 297d43317dcSChris Mason 298e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2995a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 300e1489b4fSIoannis Angelakopoulos 30119ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 30219ae4e81SJan Schmidt if (fs_info->running_transaction) { 303d43317dcSChris Mason /* 304d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3054a9d8bdeSMiao Xie * to redo the checks above 306d43317dcSChris Mason */ 3075a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 308e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3094b5faeacSDavid Sterba kfree(cur_trans); 310d43317dcSChris Mason goto loop; 31184961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 312e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3135a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 314e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3154b5faeacSDavid Sterba kfree(cur_trans); 3167b8b92afSJosef Bacik return -EROFS; 317a4abeea4SJosef Bacik } 318d43317dcSChris Mason 319ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 32048778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 32148778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 32213c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3230860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 32479154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 32579154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3264a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 327a4abeea4SJosef Bacik /* 328a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 329a4abeea4SJosef Bacik * commit the transaction. 330a4abeea4SJosef Bacik */ 3319b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3323204d33cSJosef Bacik cur_trans->flags = 0; 333afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 33456bec294SChris Mason 335a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 336a099d0fdSAlexandru Moise 3375c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3383368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 339d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 34020b297d6SJan Schmidt 34120b297d6SJan Schmidt /* 34220b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 34320b297d6SJan Schmidt * the log must never go across transaction boundaries. 34420b297d6SJan Schmidt */ 34520b297d6SJan Schmidt smp_mb(); 34631b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3475d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 34831b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3495d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 350fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 35120b297d6SJan Schmidt 35256bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 35356bec294SChris Mason 3543063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 355bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3569e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 357ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3581bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3592b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3601bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 361ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 362e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3632b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 364d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 365d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 36619ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 367c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 36843eb5f29SQu Wenruo IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); 369fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 370fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 37119ae4e81SJan Schmidt fs_info->generation++; 37219ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 37319ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 37449b25e05SJeff Mahoney cur_trans->aborted = 0; 37519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 37615ee9bc7SJosef Bacik 37779154b1bSChris Mason return 0; 37879154b1bSChris Mason } 37979154b1bSChris Mason 380d352ac68SChris Mason /* 38192a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 38292a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 38392a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 38492a7cc42SQu Wenruo * transaction commits. 385d352ac68SChris Mason */ 3867585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3876426c7adSQu Wenruo struct btrfs_root *root, 3886426c7adSQu Wenruo int force) 3896702ed49SChris Mason { 3900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 39103a7e111SJosef Bacik int ret = 0; 3920b246afaSJeff Mahoney 39392a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 3946426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 3954d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 3965d4f98a2SYan Zheng 3977585717fSChris Mason /* 39827cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 3997585717fSChris Mason * we have the reloc mutex held now, so there 4007585717fSChris Mason * is only one writer in this function 4017585717fSChris Mason */ 40227cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4037585717fSChris Mason 40427cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4057585717fSChris Mason * they find our root->last_trans update 4067585717fSChris Mason */ 4077585717fSChris Mason smp_wmb(); 4087585717fSChris Mason 409fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4106426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 411fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 412a4abeea4SJosef Bacik return 0; 413a4abeea4SJosef Bacik } 414fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4156702ed49SChris Mason (unsigned long)root->root_key.objectid, 4166702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 417fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4187585717fSChris Mason root->last_trans = trans->transid; 4197585717fSChris Mason 4207585717fSChris Mason /* this is pretty tricky. We don't want to 4217585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4227585717fSChris Mason * unless we're really doing the first setup for this root in 4237585717fSChris Mason * this transaction. 4247585717fSChris Mason * 4257585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4267585717fSChris Mason * if we want to take the expensive mutex. 4277585717fSChris Mason * 4287585717fSChris Mason * But, we have to set root->last_trans before we 4297585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4307585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 43127cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4327585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4337585717fSChris Mason * 4347585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4357585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4367585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4377585717fSChris Mason * done before we pop in the zero below 4387585717fSChris Mason */ 43903a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 440c7548af6SChris Mason smp_mb__before_atomic(); 44127cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4426702ed49SChris Mason } 44303a7e111SJosef Bacik return ret; 4446702ed49SChris Mason } 4455d4f98a2SYan Zheng 4467585717fSChris Mason 4472b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4482b9dbef2SJosef Bacik struct btrfs_root *root) 4492b9dbef2SJosef Bacik { 4500b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4512b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4522b9dbef2SJosef Bacik 4532b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4542b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4552b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4562b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4572b9dbef2SJosef Bacik 4582b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 459fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 460fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4612b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4622b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 463fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4642b9dbef2SJosef Bacik } 4652b9dbef2SJosef Bacik 4667585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4677585717fSChris Mason struct btrfs_root *root) 4687585717fSChris Mason { 4690b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4701409e6ccSJosef Bacik int ret; 4710b246afaSJeff Mahoney 47292a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4737585717fSChris Mason return 0; 4747585717fSChris Mason 4757585717fSChris Mason /* 47627cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4777585717fSChris Mason * and barriers 4787585717fSChris Mason */ 4797585717fSChris Mason smp_rmb(); 4807585717fSChris Mason if (root->last_trans == trans->transid && 48127cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4827585717fSChris Mason return 0; 4837585717fSChris Mason 4840b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4851409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4860b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4877585717fSChris Mason 4881409e6ccSJosef Bacik return ret; 4897585717fSChris Mason } 4907585717fSChris Mason 4914a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 4924a9d8bdeSMiao Xie { 4933296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 494501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 495bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 4964a9d8bdeSMiao Xie } 4974a9d8bdeSMiao Xie 498d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 499d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 500d352ac68SChris Mason * transaction might not be fully on disk. 501d352ac68SChris Mason */ 5022ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 50379154b1bSChris Mason { 504f9295749SChris Mason struct btrfs_transaction *cur_trans; 50579154b1bSChris Mason 5060b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5070b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5084a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5099b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5100b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51172d63ed6SLi Zefan 5123e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5130b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 514501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 515bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 516724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 517a4abeea4SJosef Bacik } else { 5180b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 519f9295749SChris Mason } 52037d1aeeeSChris Mason } 52137d1aeeeSChris Mason 5222ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 52337d1aeeeSChris Mason { 5240b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 525a4abeea4SJosef Bacik return 0; 526a4abeea4SJosef Bacik 52792e2f7e3SNikolay Borisov if (type == TRANS_START) 528a4abeea4SJosef Bacik return 1; 529a4abeea4SJosef Bacik 530a22285a6SYan, Zheng return 0; 531a22285a6SYan, Zheng } 532a22285a6SYan, Zheng 53320dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 53420dd2cbfSMiao Xie { 5350b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5360b246afaSJeff Mahoney 5370b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 53892a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 53920dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 54020dd2cbfSMiao Xie root->reloc_root) 54120dd2cbfSMiao Xie return false; 54220dd2cbfSMiao Xie 54320dd2cbfSMiao Xie return true; 54420dd2cbfSMiao Xie } 54520dd2cbfSMiao Xie 54608e007d2SMiao Xie static struct btrfs_trans_handle * 5475aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 548003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 549003d7c59SJeff Mahoney bool enforce_qgroups) 550a22285a6SYan, Zheng { 5510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 552ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 553a22285a6SYan, Zheng struct btrfs_trans_handle *h; 554a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 555b5009945SJosef Bacik u64 num_bytes = 0; 556c5567237SArne Jansen u64 qgroup_reserved = 0; 55720dd2cbfSMiao Xie bool reloc_reserved = false; 5589c343784SJosef Bacik bool do_chunk_alloc = false; 55920dd2cbfSMiao Xie int ret; 560acce952bSliubo 56184961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 562acce952bSliubo return ERR_PTR(-EROFS); 5632a1eb461SJosef Bacik 56446c4e71eSFilipe Manana if (current->journal_info) { 5650860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5662a1eb461SJosef Bacik h = current->journal_info; 567b50fff81SDavid Sterba refcount_inc(&h->use_count); 568b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5692a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5702a1eb461SJosef Bacik h->block_rsv = NULL; 5712a1eb461SJosef Bacik goto got_it; 5722a1eb461SJosef Bacik } 573b5009945SJosef Bacik 574b5009945SJosef Bacik /* 575b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 576b5009945SJosef Bacik * the appropriate flushing if need be. 577b5009945SJosef Bacik */ 578003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 579ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 580ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 581ba2c4d4eSJosef Bacik 5820b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 583733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 584003d7c59SJeff Mahoney enforce_qgroups); 585c5567237SArne Jansen if (ret) 586c5567237SArne Jansen return ERR_PTR(ret); 587c5567237SArne Jansen 588ba2c4d4eSJosef Bacik /* 589ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 590ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 591ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 592ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 593ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 594ba2c4d4eSJosef Bacik */ 5952bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 5967f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 597748f553cSDavid Sterba btrfs_block_rsv_full(delayed_refs_rsv) == 0) { 598ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 599ba2c4d4eSJosef Bacik num_bytes <<= 1; 600ba2c4d4eSJosef Bacik } 601ba2c4d4eSJosef Bacik 60220dd2cbfSMiao Xie /* 60320dd2cbfSMiao Xie * Do the reservation for the relocation root creation 60420dd2cbfSMiao Xie */ 605ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6060b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 60720dd2cbfSMiao Xie reloc_reserved = true; 60820dd2cbfSMiao Xie } 60920dd2cbfSMiao Xie 6109270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 611ba2c4d4eSJosef Bacik if (ret) 612ba2c4d4eSJosef Bacik goto reserve_fail; 613ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 614ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 615ba2c4d4eSJosef Bacik delayed_refs_bytes); 616ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 617ba2c4d4eSJosef Bacik } 6189c343784SJosef Bacik 6199c343784SJosef Bacik if (rsv->space_info->force_alloc) 6209c343784SJosef Bacik do_chunk_alloc = true; 621ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 622748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 623ba2c4d4eSJosef Bacik /* 624ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 625ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 626ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 627ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 628ba2c4d4eSJosef Bacik * here. 629ba2c4d4eSJosef Bacik */ 630ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 631b5009945SJosef Bacik if (ret) 632843fcf35SMiao Xie goto reserve_fail; 633b5009945SJosef Bacik } 634a22285a6SYan, Zheng again: 635f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 636843fcf35SMiao Xie if (!h) { 637843fcf35SMiao Xie ret = -ENOMEM; 638843fcf35SMiao Xie goto alloc_fail; 639843fcf35SMiao Xie } 640a22285a6SYan, Zheng 64198114659SJosef Bacik /* 64298114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 64398114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 64498114659SJosef Bacik * because we're already holding a ref. We need this because we could 64598114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 64698114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 647354aa0fbSMiao Xie * 648354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 649354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 65098114659SJosef Bacik */ 6510860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6520b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 653b2b5ef5cSJan Kara 6542ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6552ff7e61eSJeff Mahoney wait_current_trans(fs_info); 656a22285a6SYan, Zheng 657a4abeea4SJosef Bacik do { 6582ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 659178260b2SMiao Xie if (ret == -EBUSY) { 6602ff7e61eSJeff Mahoney wait_current_trans(fs_info); 661a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 662a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 663178260b2SMiao Xie ret = -ENOENT; 664178260b2SMiao Xie } 665a4abeea4SJosef Bacik } while (ret == -EBUSY); 666a4abeea4SJosef Bacik 667a43f7f82SLiu Bo if (ret < 0) 668843fcf35SMiao Xie goto join_fail; 6690f7d52f4SChris Mason 6700b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 671a22285a6SYan, Zheng 672a22285a6SYan, Zheng h->transid = cur_trans->transid; 673a22285a6SYan, Zheng h->transaction = cur_trans; 674b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 67564b63580SJeff Mahoney h->fs_info = root->fs_info; 6767174109cSQu Wenruo 677a698d075SMiao Xie h->type = type; 678ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 679b7ec40d7SChris Mason 680a22285a6SYan, Zheng smp_mb(); 6813296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 6822ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 683abdd2e80SFilipe Manana current->journal_info = h; 6843a45bb20SJeff Mahoney btrfs_commit_transaction(h); 685a22285a6SYan, Zheng goto again; 686a22285a6SYan, Zheng } 6879ed74f2dSJosef Bacik 688b5009945SJosef Bacik if (num_bytes) { 6890b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 6902bcc0328SLiu Bo h->transid, num_bytes, 1); 6910b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 692b5009945SJosef Bacik h->bytes_reserved = num_bytes; 69320dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 694a22285a6SYan, Zheng } 695a22285a6SYan, Zheng 6962a1eb461SJosef Bacik got_it: 697bcf3a3e7SNikolay Borisov if (!current->journal_info) 698a22285a6SYan, Zheng current->journal_info = h; 699fcc99734SQu Wenruo 700fcc99734SQu Wenruo /* 7019c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7029c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7039c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7049c343784SJosef Bacik * value here. 7059c343784SJosef Bacik */ 7069c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7079c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7089c343784SJosef Bacik 7099c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7109c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7119c343784SJosef Bacik } 7129c343784SJosef Bacik 7139c343784SJosef Bacik /* 714fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 715fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 716fcc99734SQu Wenruo * transaction. 717fcc99734SQu Wenruo * 718fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 719fcc99734SQu Wenruo * or we can deadlock. 720fcc99734SQu Wenruo */ 72168075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 72268075ea8SJosef Bacik if (ret) { 72368075ea8SJosef Bacik /* 72468075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 72568075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 72668075ea8SJosef Bacik * not just freed. 72768075ea8SJosef Bacik */ 72868075ea8SJosef Bacik btrfs_end_transaction(h); 72968075ea8SJosef Bacik return ERR_PTR(ret); 73068075ea8SJosef Bacik } 731fcc99734SQu Wenruo 73279154b1bSChris Mason return h; 733843fcf35SMiao Xie 734843fcf35SMiao Xie join_fail: 7350860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7360b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 737843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 738843fcf35SMiao Xie alloc_fail: 739843fcf35SMiao Xie if (num_bytes) 7402ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 74163f018beSNikolay Borisov num_bytes, NULL); 742843fcf35SMiao Xie reserve_fail: 743733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 744843fcf35SMiao Xie return ERR_PTR(ret); 74579154b1bSChris Mason } 74679154b1bSChris Mason 747f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7485aed1dd8SAlexandru Moise unsigned int num_items) 749f9295749SChris Mason { 75008e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 751003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 752f9295749SChris Mason } 753003d7c59SJeff Mahoney 7548eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7558eab77ffSFilipe Manana struct btrfs_root *root, 7567f9fe614SJosef Bacik unsigned int num_items) 7578eab77ffSFilipe Manana { 7587f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7597f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7608eab77ffSFilipe Manana } 7618407aa46SMiao Xie 7627a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 763f9295749SChris Mason { 764003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 765003d7c59SJeff Mahoney true); 766f9295749SChris Mason } 767f9295749SChris Mason 7688d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7690af3d00bSJosef Bacik { 770575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 771003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7720af3d00bSJosef Bacik } 7730af3d00bSJosef Bacik 774d4edf39bSMiao Xie /* 775a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 776a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 777a6d155d2SFilipe Manana */ 778a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 779a6d155d2SFilipe Manana { 780a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 781a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 782a6d155d2SFilipe Manana } 783a6d155d2SFilipe Manana 784a6d155d2SFilipe Manana /* 785d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 786d4edf39bSMiao Xie * 787d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 788d4edf39bSMiao Xie * don't want to start a new one. 789d4edf39bSMiao Xie * 790d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 791d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 792d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 793d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 794d4edf39bSMiao Xie * invoke 795d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 796d4edf39bSMiao Xie */ 797354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 79860376ce4SJosef Bacik { 799575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 800003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 80160376ce4SJosef Bacik } 80260376ce4SJosef Bacik 803d4edf39bSMiao Xie /* 80490b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 805d4edf39bSMiao Xie * 80652042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 807d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 808d4edf39bSMiao Xie * complete. 809d4edf39bSMiao Xie */ 810d4edf39bSMiao Xie struct btrfs_trans_handle * 811d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 812d4edf39bSMiao Xie { 813d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 814d4edf39bSMiao Xie 815575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 816003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8178d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8182ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 819d4edf39bSMiao Xie 820d4edf39bSMiao Xie return trans; 821d4edf39bSMiao Xie } 822d4edf39bSMiao Xie 823d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 824d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 825d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 82689ce8a63SChris Mason { 8275fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8285fd76bf3SOmar Sandoval u64 transid = commit->transid; 8295fd76bf3SOmar Sandoval bool put = false; 8305fd76bf3SOmar Sandoval 8313e738c53SIoannis Angelakopoulos /* 8323e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8333e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8343e738c53SIoannis Angelakopoulos */ 8353e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8363e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8373e738c53SIoannis Angelakopoulos else 8383e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8393e738c53SIoannis Angelakopoulos 8405fd76bf3SOmar Sandoval while (1) { 841d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8425fd76bf3SOmar Sandoval if (put) 8435fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8445fd76bf3SOmar Sandoval 8455fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8465fd76bf3SOmar Sandoval break; 8475fd76bf3SOmar Sandoval 8485fd76bf3SOmar Sandoval /* 8495fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8505fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8515fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8525fd76bf3SOmar Sandoval * transaction. Wait for those. 8535fd76bf3SOmar Sandoval */ 8545fd76bf3SOmar Sandoval 8555fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8565fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8575fd76bf3SOmar Sandoval struct btrfs_transaction, 8585fd76bf3SOmar Sandoval list); 8595fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8605fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8615fd76bf3SOmar Sandoval break; 8625fd76bf3SOmar Sandoval } 8635fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8645fd76bf3SOmar Sandoval put = true; 8655fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8665fd76bf3SOmar Sandoval } 86789ce8a63SChris Mason } 86889ce8a63SChris Mason 8692ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 87046204592SSage Weil { 87146204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8728cd2807fSMiao Xie int ret = 0; 87346204592SSage Weil 87446204592SSage Weil if (transid) { 8750b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 876a4abeea4SJosef Bacik goto out; 87746204592SSage Weil 87846204592SSage Weil /* find specified transaction */ 8790b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8800b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 88146204592SSage Weil if (t->transid == transid) { 88246204592SSage Weil cur_trans = t; 8839b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8848cd2807fSMiao Xie ret = 0; 88546204592SSage Weil break; 88646204592SSage Weil } 8878cd2807fSMiao Xie if (t->transid > transid) { 8888cd2807fSMiao Xie ret = 0; 88946204592SSage Weil break; 89046204592SSage Weil } 8918cd2807fSMiao Xie } 8920b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 89342383020SSage Weil 89442383020SSage Weil /* 89542383020SSage Weil * The specified transaction doesn't exist, or we 89642383020SSage Weil * raced with btrfs_commit_transaction 89742383020SSage Weil */ 89842383020SSage Weil if (!cur_trans) { 8990b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 90042383020SSage Weil ret = -EINVAL; 9018cd2807fSMiao Xie goto out; 90242383020SSage Weil } 90346204592SSage Weil } else { 90446204592SSage Weil /* find newest transaction that is committing | committed */ 9050b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9060b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 90746204592SSage Weil list) { 9084a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9094a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9103473f3c0SJosef Bacik break; 91146204592SSage Weil cur_trans = t; 9129b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 91346204592SSage Weil break; 91446204592SSage Weil } 91546204592SSage Weil } 9160b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 91746204592SSage Weil if (!cur_trans) 918a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 91946204592SSage Weil } 92046204592SSage Weil 921d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 922724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 923a4abeea4SJosef Bacik out: 92446204592SSage Weil return ret; 92546204592SSage Weil } 92646204592SSage Weil 9272ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 92837d1aeeeSChris Mason { 9292ff7e61eSJeff Mahoney wait_current_trans(fs_info); 93037d1aeeeSChris Mason } 93137d1aeeeSChris Mason 9328a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9338929ecfaSYan, Zheng { 9342ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9350b246afaSJeff Mahoney 93664403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9378a8f4deaSNikolay Borisov return true; 93836ba022aSJosef Bacik 9392ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9408929ecfaSYan, Zheng } 9418929ecfaSYan, Zheng 942a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9438929ecfaSYan, Zheng { 9448929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9458929ecfaSYan, Zheng 9463296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 947e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 948a2633b6aSNikolay Borisov return true; 9498929ecfaSYan, Zheng 9502ff7e61eSJeff Mahoney return should_end_transaction(trans); 9518929ecfaSYan, Zheng } 9528929ecfaSYan, Zheng 953dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 954dc60c525SNikolay Borisov 9550e34693fSNikolay Borisov { 956dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 957dc60c525SNikolay Borisov 9580e34693fSNikolay Borisov if (!trans->block_rsv) { 9590e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9600e34693fSNikolay Borisov return; 9610e34693fSNikolay Borisov } 9620e34693fSNikolay Borisov 9630e34693fSNikolay Borisov if (!trans->bytes_reserved) 9640e34693fSNikolay Borisov return; 9650e34693fSNikolay Borisov 9660e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9670e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9680e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9690e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 97063f018beSNikolay Borisov trans->bytes_reserved, NULL); 9710e34693fSNikolay Borisov trans->bytes_reserved = 0; 9720e34693fSNikolay Borisov } 9730e34693fSNikolay Borisov 97489ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9753a45bb20SJeff Mahoney int throttle) 97679154b1bSChris Mason { 9773a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9788929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9794edc2ca3SDave Jones int err = 0; 980d6e4a428SChris Mason 981b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 982b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9832a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9842a1eb461SJosef Bacik return 0; 9852a1eb461SJosef Bacik } 9862a1eb461SJosef Bacik 987dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9884c13d758SJosef Bacik trans->block_rsv = NULL; 989c5567237SArne Jansen 9906c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 991ea658badSJosef Bacik 9924fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9934fbcdf66SFilipe Manana 9940860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9950b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9966df7881aSJosef Bacik 9978929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 99813c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 99913c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10000860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 100189ce8a63SChris Mason 1002093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1003e1489b4fSIoannis Angelakopoulos 10045a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1005e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1006e1489b4fSIoannis Angelakopoulos 1007724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10089ed74f2dSJosef Bacik 10099ed74f2dSJosef Bacik if (current->journal_info == trans) 10109ed74f2dSJosef Bacik current->journal_info = NULL; 1011ab78c84dSChris Mason 101224bbcf04SYan, Zheng if (throttle) 10132ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 101424bbcf04SYan, Zheng 101584961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10164e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1017fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1018fbabd4a3SJosef Bacik err = trans->aborted; 1019fbabd4a3SJosef Bacik else 1020fbabd4a3SJosef Bacik err = -EROFS; 10214e121c06SJosef Bacik } 102249b25e05SJeff Mahoney 10234edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10244edc2ca3SDave Jones return err; 102579154b1bSChris Mason } 102679154b1bSChris Mason 10273a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 102889ce8a63SChris Mason { 10293a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 103089ce8a63SChris Mason } 103189ce8a63SChris Mason 10323a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 103389ce8a63SChris Mason { 10343a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 103516cdcec7SMiao Xie } 103616cdcec7SMiao Xie 1037d352ac68SChris Mason /* 1038d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1039d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1040690587d1SChris Mason * those extents are sent to disk but does not wait on them 1041d352ac68SChris Mason */ 10422ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10438cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 104479154b1bSChris Mason { 1045777e6bd7SChris Mason int err = 0; 10467c4452b9SChris Mason int werr = 0; 10470b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1048e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1049777e6bd7SChris Mason u64 start = 0; 10505f39d397SChris Mason u64 end; 10517c4452b9SChris Mason 10526300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10531728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1054e6138876SJosef Bacik mark, &cached_state)) { 1055663dfbb0SFilipe Manana bool wait_writeback = false; 1056663dfbb0SFilipe Manana 1057663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1058663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1059210aa277SDavid Sterba mark, &cached_state); 1060663dfbb0SFilipe Manana /* 1061663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1062663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1063663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1064663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1065bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1066bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1067bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1068bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1069bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1070663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 107141e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1072663dfbb0SFilipe Manana */ 1073663dfbb0SFilipe Manana if (err == -ENOMEM) { 1074663dfbb0SFilipe Manana err = 0; 1075663dfbb0SFilipe Manana wait_writeback = true; 1076663dfbb0SFilipe Manana } 1077663dfbb0SFilipe Manana if (!err) 10781728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10797c4452b9SChris Mason if (err) 10807c4452b9SChris Mason werr = err; 1081663dfbb0SFilipe Manana else if (wait_writeback) 1082663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1083e38e2ed7SFilipe Manana free_extent_state(cached_state); 1084663dfbb0SFilipe Manana cached_state = NULL; 10851728366eSJosef Bacik cond_resched(); 10861728366eSJosef Bacik start = end + 1; 10877c4452b9SChris Mason } 10886300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1089690587d1SChris Mason return werr; 1090690587d1SChris Mason } 1091690587d1SChris Mason 1092690587d1SChris Mason /* 1093690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1094690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1095690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1096690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1097690587d1SChris Mason */ 1098bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1099bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1100690587d1SChris Mason { 1101690587d1SChris Mason int err = 0; 1102690587d1SChris Mason int werr = 0; 11030b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1104e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1105690587d1SChris Mason u64 start = 0; 1106690587d1SChris Mason u64 end; 1107690587d1SChris Mason 11081728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1109e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1110663dfbb0SFilipe Manana /* 1111663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1112663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1113663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1114663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1115663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 111641e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1117663dfbb0SFilipe Manana */ 1118663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1119*bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1120663dfbb0SFilipe Manana if (err == -ENOMEM) 1121663dfbb0SFilipe Manana err = 0; 1122663dfbb0SFilipe Manana if (!err) 11231728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1124777e6bd7SChris Mason if (err) 1125777e6bd7SChris Mason werr = err; 1126e38e2ed7SFilipe Manana free_extent_state(cached_state); 1127e38e2ed7SFilipe Manana cached_state = NULL; 1128777e6bd7SChris Mason cond_resched(); 11291728366eSJosef Bacik start = end + 1; 1130777e6bd7SChris Mason } 11317c4452b9SChris Mason if (err) 11327c4452b9SChris Mason werr = err; 1133bf89d38fSJeff Mahoney return werr; 1134bf89d38fSJeff Mahoney } 1135656f30dbSFilipe Manana 1136b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1137bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1138bf89d38fSJeff Mahoney { 1139bf89d38fSJeff Mahoney bool errors = false; 1140bf89d38fSJeff Mahoney int err; 1141bf89d38fSJeff Mahoney 1142bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1143bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1144bf89d38fSJeff Mahoney errors = true; 1145bf89d38fSJeff Mahoney 1146bf89d38fSJeff Mahoney if (errors && !err) 1147bf89d38fSJeff Mahoney err = -EIO; 1148bf89d38fSJeff Mahoney return err; 1149bf89d38fSJeff Mahoney } 1150bf89d38fSJeff Mahoney 1151bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1152bf89d38fSJeff Mahoney { 1153bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1154bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1155bf89d38fSJeff Mahoney bool errors = false; 1156bf89d38fSJeff Mahoney int err; 1157bf89d38fSJeff Mahoney 1158bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1159bf89d38fSJeff Mahoney 1160bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1161656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11620b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1163656f30dbSFilipe Manana errors = true; 1164656f30dbSFilipe Manana 1165656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11660b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1167656f30dbSFilipe Manana errors = true; 1168656f30dbSFilipe Manana 1169bf89d38fSJeff Mahoney if (errors && !err) 1170bf89d38fSJeff Mahoney err = -EIO; 1171bf89d38fSJeff Mahoney return err; 117279154b1bSChris Mason } 117379154b1bSChris Mason 1174690587d1SChris Mason /* 1175c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1176c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1177c9b577c0SNikolay Borisov * log commit. 1178c9b577c0SNikolay Borisov * 1179c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1180690587d1SChris Mason */ 118170458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1182d0c803c4SChris Mason { 1183663dfbb0SFilipe Manana int ret; 1184c9b577c0SNikolay Borisov int ret2; 1185c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 118670458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1187c9b577c0SNikolay Borisov struct blk_plug plug; 1188663dfbb0SFilipe Manana 1189c9b577c0SNikolay Borisov blk_start_plug(&plug); 1190c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1191c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1192c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1193c9b577c0SNikolay Borisov 119441e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1195663dfbb0SFilipe Manana 1196c9b577c0SNikolay Borisov if (ret) 1197663dfbb0SFilipe Manana return ret; 1198c9b577c0SNikolay Borisov else if (ret2) 1199c9b577c0SNikolay Borisov return ret2; 1200c9b577c0SNikolay Borisov else 1201c9b577c0SNikolay Borisov return 0; 1202d0c803c4SChris Mason } 1203d0c803c4SChris Mason 1204d352ac68SChris Mason /* 1205d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1206d352ac68SChris Mason * 1207d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1208d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1209d352ac68SChris Mason * allocation tree. 1210d352ac68SChris Mason * 1211d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1212d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1213d352ac68SChris Mason */ 12140b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 121579154b1bSChris Mason struct btrfs_root *root) 121679154b1bSChris Mason { 121779154b1bSChris Mason int ret; 12180b86a832SChris Mason u64 old_root_bytenr; 121986b9f2ecSYan, Zheng u64 old_root_used; 12200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12210b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 122279154b1bSChris Mason 122386b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 122456bec294SChris Mason 122579154b1bSChris Mason while (1) { 12260b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 122786b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1228ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 122979154b1bSChris Mason break; 123087ef2bb4SChris Mason 12315d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 123279154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12330b86a832SChris Mason &root->root_key, 12340b86a832SChris Mason &root->root_item); 123549b25e05SJeff Mahoney if (ret) 123649b25e05SJeff Mahoney return ret; 123756bec294SChris Mason 123886b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1239e7070be1SJosef Bacik } 1240276e680dSYan Zheng 12410b86a832SChris Mason return 0; 12420b86a832SChris Mason } 12430b86a832SChris Mason 1244d352ac68SChris Mason /* 1245d352ac68SChris Mason * update all the cowonly tree roots on disk 124649b25e05SJeff Mahoney * 124749b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 124849b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 124949b25e05SJeff Mahoney * to clean up the delayed refs. 1250d352ac68SChris Mason */ 12519386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12520b86a832SChris Mason { 12539386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1254ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12551bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12560b86a832SChris Mason struct list_head *next; 125784234f3aSYan Zheng struct extent_buffer *eb; 125856bec294SChris Mason int ret; 125984234f3aSYan Zheng 1260dfba78dcSFilipe Manana /* 1261dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1262dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1263dfba78dcSFilipe Manana */ 1264dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1265dfba78dcSFilipe Manana 126684234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 126749b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12689631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 126984234f3aSYan Zheng btrfs_tree_unlock(eb); 127084234f3aSYan Zheng free_extent_buffer(eb); 12710b86a832SChris Mason 127249b25e05SJeff Mahoney if (ret) 127349b25e05SJeff Mahoney return ret; 127449b25e05SJeff Mahoney 1275196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1276c16ce190SJosef Bacik if (ret) 1277c16ce190SJosef Bacik return ret; 12782b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1279c16ce190SJosef Bacik if (ret) 1280c16ce190SJosef Bacik return ret; 1281280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1282c16ce190SJosef Bacik if (ret) 1283c16ce190SJosef Bacik return ret; 1284546adb0dSJan Schmidt 1285bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1286dcdf7f6dSJosef Bacik if (ret) 1287dcdf7f6dSJosef Bacik return ret; 1288dcdf7f6dSJosef Bacik 1289ea526d18SJosef Bacik again: 12900b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12912ff7e61eSJeff Mahoney struct btrfs_root *root; 12920b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12930b86a832SChris Mason list_del_init(next); 12940b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1295e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 129687ef2bb4SChris Mason 12979e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 12989e351cc8SJosef Bacik &trans->transaction->switch_commits); 129949b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 130049b25e05SJeff Mahoney if (ret) 130149b25e05SJeff Mahoney return ret; 1302488bc2a2SJosef Bacik } 1303488bc2a2SJosef Bacik 1304488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1305c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1306ea526d18SJosef Bacik if (ret) 1307ea526d18SJosef Bacik return ret; 1308276e680dSYan Zheng 13091bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13105742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1311ea526d18SJosef Bacik if (ret) 1312ea526d18SJosef Bacik return ret; 1313488bc2a2SJosef Bacik 1314488bc2a2SJosef Bacik /* 1315488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1316488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1317488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1318488bc2a2SJosef Bacik * everything gets run. 1319488bc2a2SJosef Bacik */ 1320c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1321ea526d18SJosef Bacik if (ret) 1322ea526d18SJosef Bacik return ret; 1323ea526d18SJosef Bacik } 1324ea526d18SJosef Bacik 1325ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1326ea526d18SJosef Bacik goto again; 1327ea526d18SJosef Bacik 13289f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13299f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13309f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13318dabb742SStefan Behrens 133279154b1bSChris Mason return 0; 133379154b1bSChris Mason } 133479154b1bSChris Mason 1335d352ac68SChris Mason /* 1336b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1337b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1338b4be6aefSJosef Bacik */ 1339b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1340b4be6aefSJosef Bacik { 1341b4be6aefSJosef Bacik /* 1342b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1343b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1344b4be6aefSJosef Bacik * up. 1345b4be6aefSJosef Bacik */ 1346b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1347b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1348b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1349b4be6aefSJosef Bacik struct btrfs_root, 1350b4be6aefSJosef Bacik root_list); 1351b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1352b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1353b4be6aefSJosef Bacik return; 1354b4be6aefSJosef Bacik } 1355b4be6aefSJosef Bacik } 1356b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1357b4be6aefSJosef Bacik 1358b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1359b4be6aefSJosef Bacik } 1360b4be6aefSJosef Bacik 1361b4be6aefSJosef Bacik /* 1362d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1363d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1364d352ac68SChris Mason * be deleted 1365d352ac68SChris Mason */ 1366cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13675eda7b5eSChris Mason { 13680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13690b246afaSJeff Mahoney 13700b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1371dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1372dc9492c1SJosef Bacik btrfs_grab_root(root); 1373b4be6aefSJosef Bacik 1374b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1375b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1376b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1377b4be6aefSJosef Bacik else 13780b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1379dc9492c1SJosef Bacik } 13800b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13815eda7b5eSChris Mason } 13825eda7b5eSChris Mason 1383d352ac68SChris Mason /* 1384dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1385dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1386d352ac68SChris Mason */ 13877e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13880f7d52f4SChris Mason { 13897e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1390fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1391fc7cbcd4SDavid Sterba int i; 1392fc7cbcd4SDavid Sterba int ret; 139354aa1f4dSChris Mason 1394dfba78dcSFilipe Manana /* 1395dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1396dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1397dfba78dcSFilipe Manana */ 1398dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1399dfba78dcSFilipe Manana 1400fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1401fc7cbcd4SDavid Sterba while (1) { 1402fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1403fc7cbcd4SDavid Sterba (void **)gang, 0, 1404fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1405fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1406fc7cbcd4SDavid Sterba if (ret == 0) 1407fc7cbcd4SDavid Sterba break; 1408fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1409fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1410fc7cbcd4SDavid Sterba int ret2; 14114f4317c1SJosef Bacik 1412dfba78dcSFilipe Manana /* 1413dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1414dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1415dfba78dcSFilipe Manana */ 1416dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1417dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1418dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1419dfba78dcSFilipe Manana 1420fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14212619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14220f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1423fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 142431153d81SYan Zheng 1425e02119d5SChris Mason btrfs_free_log(trans, root); 1426fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1427fc7cbcd4SDavid Sterba if (ret2) 1428fc7cbcd4SDavid Sterba return ret2; 1429e02119d5SChris Mason 1430fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 143127cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1432c7548af6SChris Mason smp_mb__after_atomic(); 1433f1ebcc74SLiu Bo 1434978d910dSYan Zheng if (root->commit_root != root->node) { 14359e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14369e351cc8SJosef Bacik &trans->transaction->switch_commits); 1437fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1438fc7cbcd4SDavid Sterba root->node); 1439978d910dSYan Zheng } 144031153d81SYan Zheng 1441fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1442fc7cbcd4SDavid Sterba &root->root_key, 1443fc7cbcd4SDavid Sterba &root->root_item); 1444fc7cbcd4SDavid Sterba if (ret2) 1445fc7cbcd4SDavid Sterba return ret2; 1446fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1447733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14480f7d52f4SChris Mason } 1449fc7cbcd4SDavid Sterba } 1450fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14514f4317c1SJosef Bacik return 0; 14520f7d52f4SChris Mason } 14530f7d52f4SChris Mason 1454d352ac68SChris Mason /* 1455de78b51aSEric Sandeen * defrag a given btree. 1456de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1457d352ac68SChris Mason */ 1458de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1459e9d0b13bSChris Mason { 1460e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1461e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14628929ecfaSYan, Zheng int ret; 1463e9d0b13bSChris Mason 146427cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1465e9d0b13bSChris Mason return 0; 14668929ecfaSYan, Zheng 14676b80053dSChris Mason while (1) { 14688929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14696819703fSDavid Sterba if (IS_ERR(trans)) { 14706819703fSDavid Sterba ret = PTR_ERR(trans); 14716819703fSDavid Sterba break; 14726819703fSDavid Sterba } 14738929ecfaSYan, Zheng 1474de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14758929ecfaSYan, Zheng 14763a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14772ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1478e9d0b13bSChris Mason cond_resched(); 1479e9d0b13bSChris Mason 1480ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1481e9d0b13bSChris Mason break; 1482210549ebSDavid Sterba 1483ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1484ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1485210549ebSDavid Sterba ret = -EAGAIN; 1486210549ebSDavid Sterba break; 1487210549ebSDavid Sterba } 1488e9d0b13bSChris Mason } 148927cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14908929ecfaSYan, Zheng return ret; 1491e9d0b13bSChris Mason } 1492e9d0b13bSChris Mason 1493d352ac68SChris Mason /* 14946426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14956426c7adSQu Wenruo * 14966426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14976426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 14986426c7adSQu Wenruo * qgroup works. 14996426c7adSQu Wenruo */ 15006426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15016426c7adSQu Wenruo struct btrfs_root *src, 15026426c7adSQu Wenruo struct btrfs_root *parent, 15036426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15046426c7adSQu Wenruo u64 dst_objectid) 15056426c7adSQu Wenruo { 15066426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15076426c7adSQu Wenruo int ret; 15086426c7adSQu Wenruo 15096426c7adSQu Wenruo /* 15106426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15116426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15126426c7adSQu Wenruo * kick in anyway. 15136426c7adSQu Wenruo */ 15149ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15156426c7adSQu Wenruo return 0; 15166426c7adSQu Wenruo 15176426c7adSQu Wenruo /* 151852042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15194d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15204d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15214d31778aSQu Wenruo * item. 15224d31778aSQu Wenruo */ 15231c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15241c442d22SJosef Bacik if (ret) 15251c442d22SJosef Bacik return ret; 15264d31778aSQu Wenruo 15274d31778aSQu Wenruo /* 15282a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15292a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15302a4d84c1SJosef Bacik * 15312a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15322a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15332a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15342a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15352a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15362a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15372a4d84c1SJosef Bacik */ 15382a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15392a4d84c1SJosef Bacik if (ret) { 15402a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 154144365827SNaohiro Aota return ret; 15422a4d84c1SJosef Bacik } 15432a4d84c1SJosef Bacik 15447e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15456426c7adSQu Wenruo if (ret) 15466426c7adSQu Wenruo goto out; 1547460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15486426c7adSQu Wenruo if (ret < 0) 15496426c7adSQu Wenruo goto out; 15506426c7adSQu Wenruo 15516426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1552a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15536426c7adSQu Wenruo inherit); 15546426c7adSQu Wenruo if (ret < 0) 15556426c7adSQu Wenruo goto out; 15566426c7adSQu Wenruo 15576426c7adSQu Wenruo /* 15586426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15596426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15606426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15616426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15626426c7adSQu Wenruo * 2) write all btree blocks onto disk 15636426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15646426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15656426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15666426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15676426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15686426c7adSQu Wenruo */ 15699386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15706426c7adSQu Wenruo if (ret) 15716426c7adSQu Wenruo goto out; 1572889bfa39SJosef Bacik switch_commit_roots(trans); 157370458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15746426c7adSQu Wenruo if (ret) 1575f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15766426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15776426c7adSQu Wenruo 15786426c7adSQu Wenruo out: 15796426c7adSQu Wenruo /* 15806426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15816426c7adSQu Wenruo * last_trans == cur_transid. 15826426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15836426c7adSQu Wenruo * insert_dir_item() 15846426c7adSQu Wenruo */ 15856426c7adSQu Wenruo if (!ret) 15861c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15876426c7adSQu Wenruo return ret; 15886426c7adSQu Wenruo } 15896426c7adSQu Wenruo 15906426c7adSQu Wenruo /* 1591d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1592aec8030aSMiao Xie * transaction commit. This does the actual creation. 1593aec8030aSMiao Xie * 1594aec8030aSMiao Xie * Note: 1595aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1596aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1597aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1598d352ac68SChris Mason */ 159980b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16003063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16013063d29fSChris Mason { 160208d50ca3SNikolay Borisov 160308d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16043063d29fSChris Mason struct btrfs_key key; 160580b6794dSChris Mason struct btrfs_root_item *new_root_item; 16063063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16073063d29fSChris Mason struct btrfs_root *root = pending->root; 16086bdb72deSSage Weil struct btrfs_root *parent_root; 160998c9942aSLiu Bo struct btrfs_block_rsv *rsv; 16106bdb72deSSage Weil struct inode *parent_inode; 161142874b3dSMiao Xie struct btrfs_path *path; 161242874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1613a22285a6SYan, Zheng struct dentry *dentry; 16143063d29fSChris Mason struct extent_buffer *tmp; 1615925baeddSChris Mason struct extent_buffer *old; 161695582b00SDeepa Dinamani struct timespec64 cur_time; 1617aec8030aSMiao Xie int ret = 0; 1618d68fc57bSYan, Zheng u64 to_reserve = 0; 16196bdb72deSSage Weil u64 index = 0; 1620a22285a6SYan, Zheng u64 objectid; 1621b83cc969SLi Zefan u64 root_flags; 16223063d29fSChris Mason 16238546b570SDavid Sterba ASSERT(pending->path); 16248546b570SDavid Sterba path = pending->path; 162542874b3dSMiao Xie 1626b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1627b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1628a22285a6SYan, Zheng 1629543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1630aec8030aSMiao Xie if (pending->error) 16316fa9700eSMiao Xie goto no_free_objectid; 16323063d29fSChris Mason 1633d6726335SQu Wenruo /* 1634d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1635d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1636d6726335SQu Wenruo */ 1637d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1638d6726335SQu Wenruo 1639147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1640d68fc57bSYan, Zheng 1641d68fc57bSYan, Zheng if (to_reserve > 0) { 16429270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1643aec8030aSMiao Xie &pending->block_rsv, 164408e007d2SMiao Xie to_reserve, 164508e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1646aec8030aSMiao Xie if (pending->error) 1647d6726335SQu Wenruo goto clear_skip_qgroup; 1648d68fc57bSYan, Zheng } 1649d68fc57bSYan, Zheng 16503063d29fSChris Mason key.objectid = objectid; 1651a22285a6SYan, Zheng key.offset = (u64)-1; 1652a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16533063d29fSChris Mason 16546fa9700eSMiao Xie rsv = trans->block_rsv; 1655a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16562382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16570b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 165888d3a5aaSJosef Bacik trans->transid, 165988d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1660a22285a6SYan, Zheng dentry = pending->dentry; 1661e9662f70SMiao Xie parent_inode = pending->dir; 1662a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1663f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1664f0118cb6SJosef Bacik if (ret) 1665f0118cb6SJosef Bacik goto fail; 1666c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 166704b285f3SDeepa Dinamani 16686bdb72deSSage Weil /* 16696bdb72deSSage Weil * insert the directory item 16706bdb72deSSage Weil */ 1671877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 167249b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 167342874b3dSMiao Xie 167442874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 167542874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16764a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 167742874b3dSMiao Xie dentry->d_name.name, 167842874b3dSMiao Xie dentry->d_name.len, 0); 167942874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1680fe66a05aSChris Mason pending->error = -EEXIST; 1681aec8030aSMiao Xie goto dir_item_existed; 168242874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 168342874b3dSMiao Xie ret = PTR_ERR(dir_item); 168466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16858732d44fSMiao Xie goto fail; 168679787eaaSJeff Mahoney } 168742874b3dSMiao Xie btrfs_release_path(path); 16886bdb72deSSage Weil 1689e999376fSChris Mason /* 1690e999376fSChris Mason * pull in the delayed directory update 1691e999376fSChris Mason * and the delayed inode item 1692e999376fSChris Mason * otherwise we corrupt the FS during 1693e999376fSChris Mason * snapshot 1694e999376fSChris Mason */ 1695e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 16968732d44fSMiao Xie if (ret) { /* Transaction aborted */ 169766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16988732d44fSMiao Xie goto fail; 16998732d44fSMiao Xie } 1700e999376fSChris Mason 1701f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1702f0118cb6SJosef Bacik if (ret) { 1703f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1704f0118cb6SJosef Bacik goto fail; 1705f0118cb6SJosef Bacik } 17066bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17076bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 170808fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17096bdb72deSSage Weil 1710b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1711b83cc969SLi Zefan if (pending->readonly) 1712b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1713b83cc969SLi Zefan else 1714b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1715b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1716b83cc969SLi Zefan 17178ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17188ea05e3aSAlexander Block trans->transid); 1719807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17208ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17218ea05e3aSAlexander Block BTRFS_UUID_SIZE); 172270023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 172370023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 172470023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17258ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17268ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17278ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17288ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 172970023da2SStefan Behrens } 17303cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17313cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 173270023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17338ea05e3aSAlexander Block 1734925baeddSChris Mason old = btrfs_lock_root_node(root); 17359631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17369631e4ccSJosef Bacik BTRFS_NESTING_COW); 173779787eaaSJeff Mahoney if (ret) { 173879787eaaSJeff Mahoney btrfs_tree_unlock(old); 173979787eaaSJeff Mahoney free_extent_buffer(old); 174066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17418732d44fSMiao Xie goto fail; 174279787eaaSJeff Mahoney } 174349b25e05SJeff Mahoney 174449b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 174579787eaaSJeff Mahoney /* clean up in any case */ 1746925baeddSChris Mason btrfs_tree_unlock(old); 1747925baeddSChris Mason free_extent_buffer(old); 17488732d44fSMiao Xie if (ret) { 174966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17508732d44fSMiao Xie goto fail; 17518732d44fSMiao Xie } 1752f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 175327cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1754f1ebcc74SLiu Bo smp_wmb(); 1755f1ebcc74SLiu Bo 17565d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1757a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1758a22285a6SYan, Zheng key.offset = trans->transid; 1759a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1760925baeddSChris Mason btrfs_tree_unlock(tmp); 17613063d29fSChris Mason free_extent_buffer(tmp); 17628732d44fSMiao Xie if (ret) { 176366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17648732d44fSMiao Xie goto fail; 17658732d44fSMiao Xie } 17660660b5afSChris Mason 1767a22285a6SYan, Zheng /* 1768a22285a6SYan, Zheng * insert root back/forward references 1769a22285a6SYan, Zheng */ 17706025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1771a22285a6SYan, Zheng parent_root->root_key.objectid, 17724a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1773a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 17748732d44fSMiao Xie if (ret) { 177566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17768732d44fSMiao Xie goto fail; 17778732d44fSMiao Xie } 1778a22285a6SYan, Zheng 1779a22285a6SYan, Zheng key.offset = (u64)-1; 17802dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 178179787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 178279787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17832d892ccdSFilipe Manana pending->snap = NULL; 178466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17858732d44fSMiao Xie goto fail; 178679787eaaSJeff Mahoney } 1787d68fc57bSYan, Zheng 178849b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17898732d44fSMiao Xie if (ret) { 179066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17918732d44fSMiao Xie goto fail; 17928732d44fSMiao Xie } 1793361048f5SMiao Xie 17946426c7adSQu Wenruo /* 17956426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17966426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17976426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 17986426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 17996426c7adSQu Wenruo */ 18006426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18016426c7adSQu Wenruo pending->inherit, objectid); 18026426c7adSQu Wenruo if (ret < 0) 18036426c7adSQu Wenruo goto fail; 18046426c7adSQu Wenruo 1805684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1806684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1807684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 180842874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18099c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18108732d44fSMiao Xie if (ret) { 181166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18128732d44fSMiao Xie goto fail; 18138732d44fSMiao Xie } 181442874b3dSMiao Xie 18156ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 181642874b3dSMiao Xie dentry->d_name.len * 2); 1817c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1818c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1819729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1820dd5f9615SStefan Behrens if (ret) { 182166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1822dd5f9615SStefan Behrens goto fail; 1823dd5f9615SStefan Behrens } 1824807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1825807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1826cdb345a8SLu Fengqi objectid); 1827dd5f9615SStefan Behrens if (ret) { 182866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1829dd5f9615SStefan Behrens goto fail; 1830dd5f9615SStefan Behrens } 1831dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1832cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1833dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1834dd5f9615SStefan Behrens objectid); 1835dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 183666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1837dd5f9615SStefan Behrens goto fail; 1838dd5f9615SStefan Behrens } 1839dd5f9615SStefan Behrens } 1840d6726335SQu Wenruo 18413063d29fSChris Mason fail: 1842aec8030aSMiao Xie pending->error = ret; 1843aec8030aSMiao Xie dir_item_existed: 184498c9942aSLiu Bo trans->block_rsv = rsv; 18452382c5ccSLiu Bo trans->bytes_reserved = 0; 1846d6726335SQu Wenruo clear_skip_qgroup: 1847d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 18486fa9700eSMiao Xie no_free_objectid: 18496fa9700eSMiao Xie kfree(new_root_item); 1850b0c0ea63SDavid Sterba pending->root_item = NULL; 185142874b3dSMiao Xie btrfs_free_path(path); 18528546b570SDavid Sterba pending->path = NULL; 18538546b570SDavid Sterba 185449b25e05SJeff Mahoney return ret; 18553063d29fSChris Mason } 18563063d29fSChris Mason 1857d352ac68SChris Mason /* 1858d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1859d352ac68SChris Mason */ 186008d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18613063d29fSChris Mason { 1862aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18633063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1864aec8030aSMiao Xie int ret = 0; 18653de4586cSChris Mason 1866aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1867aec8030aSMiao Xie list_del(&pending->list); 186808d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1869aec8030aSMiao Xie if (ret) 1870aec8030aSMiao Xie break; 1871aec8030aSMiao Xie } 1872aec8030aSMiao Xie return ret; 18733de4586cSChris Mason } 18743de4586cSChris Mason 18752ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18765d4f98a2SYan Zheng { 18775d4f98a2SYan Zheng struct btrfs_root_item *root_item; 18785d4f98a2SYan Zheng struct btrfs_super_block *super; 18795d4f98a2SYan Zheng 18800b246afaSJeff Mahoney super = fs_info->super_copy; 18815d4f98a2SYan Zheng 18820b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1883093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1884093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1885093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18865d4f98a2SYan Zheng 18870b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1888093e037cSDavid Sterba super->root = root_item->bytenr; 1889093e037cSDavid Sterba super->generation = root_item->generation; 1890093e037cSDavid Sterba super->root_level = root_item->level; 18910b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1892093e037cSDavid Sterba super->cache_generation = root_item->generation; 189394846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 189494846229SBoris Burkov super->cache_generation = 0; 18950b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1896093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18975d4f98a2SYan Zheng } 18985d4f98a2SYan Zheng 1899f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1900f36f3042SChris Mason { 19014a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1902f36f3042SChris Mason int ret = 0; 19034a9d8bdeSMiao Xie 1904a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19054a9d8bdeSMiao Xie trans = info->running_transaction; 19064a9d8bdeSMiao Xie if (trans) 19074a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1908a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1909f36f3042SChris Mason return ret; 1910f36f3042SChris Mason } 1911f36f3042SChris Mason 19128929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19138929ecfaSYan, Zheng { 19144a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19158929ecfaSYan, Zheng int ret = 0; 19164a9d8bdeSMiao Xie 1917a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19184a9d8bdeSMiao Xie trans = info->running_transaction; 19194a9d8bdeSMiao Xie if (trans) 19204a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1921a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19228929ecfaSYan, Zheng return ret; 19238929ecfaSYan, Zheng } 19248929ecfaSYan, Zheng 1925fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1926bb9c12c9SSage Weil { 19273a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1928bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1929bb9c12c9SSage Weil 1930fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1931fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1932fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1933bb9c12c9SSage Weil 1934bb9c12c9SSage Weil /* take transaction reference */ 1935bb9c12c9SSage Weil cur_trans = trans->transaction; 19369b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1937bb9c12c9SSage Weil 19383a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19396fc4e354SSage Weil 19406fc4e354SSage Weil /* 1941ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1942ae5d29d4SDavid Sterba * subsequent transaction joins 1943ae5d29d4SDavid Sterba */ 19443e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1945ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1946ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1947ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1948724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1949bb9c12c9SSage Weil } 1950bb9c12c9SSage Weil 195197cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 195249b25e05SJeff Mahoney { 195397cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 195449b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 195549b25e05SJeff Mahoney 1956b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 195749b25e05SJeff Mahoney 195866642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19597b8b92afSJosef Bacik 19600b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 196166b6135bSLiu Bo 196225d8c284SMiao Xie /* 196325d8c284SMiao Xie * If the transaction is removed from the list, it means this 196425d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 196525d8c284SMiao Xie * to call the cleanup function. 196625d8c284SMiao Xie */ 196725d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 196866b6135bSLiu Bo 19690b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19704a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19710b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1972e1489b4fSIoannis Angelakopoulos 1973e1489b4fSIoannis Angelakopoulos /* 1974e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 1975e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 1976e1489b4fSIoannis Angelakopoulos */ 1977e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 1978f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1979f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1980f094ac32SLiu Bo 19810b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1982d7096fc3SJosef Bacik } 1983061dde82SFilipe Manana 1984061dde82SFilipe Manana /* 1985061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 1986061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 1987061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 1988061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 1989061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 1990061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 1991061dde82SFilipe Manana */ 1992061dde82SFilipe Manana list_del_init(&cur_trans->list); 1993061dde82SFilipe Manana 19940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 199549b25e05SJeff Mahoney 19962ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 199749b25e05SJeff Mahoney 19980b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 19990b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20000b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20010b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20024a9d8bdeSMiao Xie 2003e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20040b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2005724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2006724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 200749b25e05SJeff Mahoney 20082e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 200949b25e05SJeff Mahoney 201049b25e05SJeff Mahoney if (current->journal_info == trans) 201149b25e05SJeff Mahoney current->journal_info = NULL; 20120b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 201349b25e05SJeff Mahoney 201449b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 201549b25e05SJeff Mahoney } 201649b25e05SJeff Mahoney 2017c7cc64a9SDavid Sterba /* 2018c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2019c7cc64a9SDavid Sterba * transaction and remove them from the list 2020c7cc64a9SDavid Sterba */ 2021c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2022c7cc64a9SDavid Sterba { 2023c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 202432da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2025c7cc64a9SDavid Sterba 2026c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2027c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2028c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2029c7cc64a9SDavid Sterba } 2030c7cc64a9SDavid Sterba } 2031c7cc64a9SDavid Sterba 203288090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 203382436617SMiao Xie { 2034ce8ea7ccSJosef Bacik /* 2035a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2036ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2037ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2038ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2039ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2040ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2041ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2042a0f0cf83SFilipe Manana * 2043a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2044a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2045a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2046a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2047a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2048a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2049a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2050ce8ea7ccSJosef Bacik */ 205188090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2052a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 205382436617SMiao Xie return 0; 205482436617SMiao Xie } 205582436617SMiao Xie 205688090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 205782436617SMiao Xie { 205888090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20596374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 206082436617SMiao Xie } 206182436617SMiao Xie 206228b21c55SFilipe Manana /* 206328b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 206428b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 206528b21c55SFilipe Manana * and while holding fs_info->trans_lock. 206628b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 206728b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 206828b21c55SFilipe Manana * returns an error. 206928b21c55SFilipe Manana */ 207028b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 207128b21c55SFilipe Manana { 207228b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 207328b21c55SFilipe Manana 207428b21c55SFilipe Manana if (!trans->pending_snapshot) 207528b21c55SFilipe Manana return; 207628b21c55SFilipe Manana 207728b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 207828b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 207928b21c55SFilipe Manana 208028b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 208128b21c55SFilipe Manana } 208228b21c55SFilipe Manana 2083e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2084e55958c8SIoannis Angelakopoulos { 2085e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2086e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2087e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2088e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2089e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2090e55958c8SIoannis Angelakopoulos } 2091e55958c8SIoannis Angelakopoulos 20923a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 209379154b1bSChris Mason { 20943a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 209549b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20968fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 209725287e0aSMiao Xie int ret; 2098e55958c8SIoannis Angelakopoulos ktime_t start_time; 2099e55958c8SIoannis Angelakopoulos ktime_t interval; 210079154b1bSChris Mason 210135b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21023e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 210335b814f3SNikolay Borisov 21048d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2105bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 210625287e0aSMiao Xie ret = cur_trans->aborted; 21073e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 210825287e0aSMiao Xie } 210949b25e05SJeff Mahoney 2110f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2111f45c752bSJosef Bacik trans->block_rsv = NULL; 2112f45c752bSJosef Bacik 2113e19eb11fSJosef Bacik /* 2114e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2115e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2116e19eb11fSJosef Bacik */ 2117e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2118e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2119e19eb11fSJosef Bacik /* 2120e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2121e19eb11fSJosef Bacik * Any running threads may add more while we are here. 212256bec294SChris Mason */ 2123c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21243e738c53SIoannis Angelakopoulos if (ret) 21253e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2126e19eb11fSJosef Bacik } 212756bec294SChris Mason 21286c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2129ea658badSJosef Bacik 21303204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21311bbc621eSChris Mason int run_it = 0; 21321bbc621eSChris Mason 21331bbc621eSChris Mason /* this mutex is also taken before trying to set 21341bbc621eSChris Mason * block groups readonly. We need to make sure 21351bbc621eSChris Mason * that nobody has set a block group readonly 21361bbc621eSChris Mason * after a extents from that block group have been 21371bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21381bbc621eSChris Mason * will wait for the transaction to commit if it 21393204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21401bbc621eSChris Mason * 21413204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21423204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21431bbc621eSChris Mason * hurt to have more than one go through, but there's no 21441bbc621eSChris Mason * real advantage to it either. 21451bbc621eSChris Mason */ 21460b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21473204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21483204d33cSJosef Bacik &cur_trans->flags)) 21491bbc621eSChris Mason run_it = 1; 21500b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21511bbc621eSChris Mason 2152f9cacae3SNikolay Borisov if (run_it) { 215321217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21543e738c53SIoannis Angelakopoulos if (ret) 21553e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2156f9cacae3SNikolay Borisov } 2157f9cacae3SNikolay Borisov } 21581bbc621eSChris Mason 21590b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21604a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2161d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2162d0c2f4faSFilipe Manana 216328b21c55SFilipe Manana add_pending_snapshot(trans); 216428b21c55SFilipe Manana 21650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21669b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2167ccd467d6SChris Mason 2168d0c2f4faSFilipe Manana if (trans->in_fsync) 2169d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 21703e738c53SIoannis Angelakopoulos 21713e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 21723e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2173d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2174d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 217515ee9bc7SJosef Bacik 2176bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2177b4924a0fSLiu Bo ret = cur_trans->aborted; 2178b4924a0fSLiu Bo 2179724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 218015ee9bc7SJosef Bacik 218149b25e05SJeff Mahoney return ret; 218279154b1bSChris Mason } 21834313b399SChris Mason 21844a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21850b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 21863e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2187bb9c12c9SSage Weil 21880b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2189d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2190d0c2f4faSFilipe Manana 2191d0c2f4faSFilipe Manana if (trans->in_fsync) 2192d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2193d0c2f4faSFilipe Manana 2194ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2195ccd467d6SChris Mason struct btrfs_transaction, list); 2196d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 21979b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 21980b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2199ccd467d6SChris Mason 2200d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2201d0c2f4faSFilipe Manana 2202bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2203ccd467d6SChris Mason 2204724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22051f9b8c8fSFilipe Manana if (ret) 2206e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2207a4abeea4SJosef Bacik } else { 22080b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2209ccd467d6SChris Mason } 2210a4abeea4SJosef Bacik } else { 22110b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2212cb2d3dadSFilipe Manana /* 2213cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2214cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2215cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2216cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2217cb2d3dadSFilipe Manana */ 221884961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2219cb2d3dadSFilipe Manana ret = -EROFS; 2220e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2221cb2d3dadSFilipe Manana } 2222ccd467d6SChris Mason } 222315ee9bc7SJosef Bacik 2224e55958c8SIoannis Angelakopoulos /* 2225e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2226e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2227e55958c8SIoannis Angelakopoulos */ 2228e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2229e55958c8SIoannis Angelakopoulos 22300860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22310860adfdSMiao Xie 223288090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 223382436617SMiao Xie if (ret) 2234e1489b4fSIoannis Angelakopoulos goto lockdep_release; 223582436617SMiao Xie 2236e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 223749b25e05SJeff Mahoney if (ret) 2238e1489b4fSIoannis Angelakopoulos goto lockdep_release; 223916cdcec7SMiao Xie 22405a9ba670SIoannis Angelakopoulos /* 22415a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22425a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22435a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22445a9ba670SIoannis Angelakopoulos */ 22455a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 22465a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2247581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2248581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2249ed3b3d31SChris Mason 2250581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2251e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2252e1489b4fSIoannis Angelakopoulos if (ret) { 2253e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2254ca469637SMiao Xie goto cleanup_transaction; 2255e1489b4fSIoannis Angelakopoulos } 2256ca469637SMiao Xie 225788090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2258cb7ab021SWang Shilong 225948778179SFilipe Manana /* 226048778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 226148778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 226248778179SFilipe Manana * extents complete we lose logged data after a power failure. 226348778179SFilipe Manana */ 22648b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 226548778179SFilipe Manana wait_event(cur_trans->pending_wait, 226648778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 226748778179SFilipe Manana 22682ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2269ed0ca140SJosef Bacik /* 2270ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2271ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22724a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2273ed0ca140SJosef Bacik */ 22740b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 227528b21c55SFilipe Manana add_pending_snapshot(trans); 22764a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 22770b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2278e1489b4fSIoannis Angelakopoulos 2279e1489b4fSIoannis Angelakopoulos /* 2280e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2281e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2282e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2283e1489b4fSIoannis Angelakopoulos */ 2284e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2285e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2286ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2287ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 228815ee9bc7SJosef Bacik 2289fdfbf020SJosef Bacik /* 22903e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 22913e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 22923e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 22933e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 22943e738c53SIoannis Angelakopoulos */ 22953e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 22963e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 22973e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 22983e738c53SIoannis Angelakopoulos 22993e738c53SIoannis Angelakopoulos /* 2300fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2301fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2302fdfbf020SJosef Bacik * kthread could do the work. 2303fdfbf020SJosef Bacik */ 2304fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2305fdfbf020SJosef Bacik 2306bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23072cba30f1SMiao Xie ret = cur_trans->aborted; 23083e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23096cf7f77eSWang Shilong goto scrub_continue; 23102cba30f1SMiao Xie } 23117585717fSChris Mason /* 23127585717fSChris Mason * the reloc mutex makes sure that we stop 23137585717fSChris Mason * the balancing code from coming in and moving 23147585717fSChris Mason * extents around in the middle of the commit 23157585717fSChris Mason */ 23160b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23177585717fSChris Mason 231842874b3dSMiao Xie /* 231942874b3dSMiao Xie * We needn't worry about the delayed items because we will 232042874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 232142874b3dSMiao Xie * core function of the snapshot creation. 232242874b3dSMiao Xie */ 232308d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 232456e9f6eaSDavid Sterba if (ret) 232556e9f6eaSDavid Sterba goto unlock_reloc; 23263063d29fSChris Mason 232742874b3dSMiao Xie /* 232842874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 232942874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 233042874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 233142874b3dSMiao Xie * them. 233242874b3dSMiao Xie * 233342874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 233442874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 233542874b3dSMiao Xie * the nodes and leaves. 233642874b3dSMiao Xie */ 2337e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 233856e9f6eaSDavid Sterba if (ret) 233956e9f6eaSDavid Sterba goto unlock_reloc; 234016cdcec7SMiao Xie 2341c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 234256e9f6eaSDavid Sterba if (ret) 234356e9f6eaSDavid Sterba goto unlock_reloc; 234456bec294SChris Mason 2345e999376fSChris Mason /* 2346e999376fSChris Mason * make sure none of the code above managed to slip in a 2347e999376fSChris Mason * delayed item 2348e999376fSChris Mason */ 2349ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2350e999376fSChris Mason 23512c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2352dc17ff8fSChris Mason 23537e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 235456e9f6eaSDavid Sterba if (ret) 2355dfba78dcSFilipe Manana goto unlock_reloc; 235654aa1f4dSChris Mason 23573818aea2SQu Wenruo /* 23587e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 23597e1876acSDavid Sterba * before the next transaction. 23603818aea2SQu Wenruo */ 23610b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 23623818aea2SQu Wenruo 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 2585572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2586572d9ab7SDavid Sterba { 2587572d9ab7SDavid Sterba unsigned long prev; 2588572d9ab7SDavid Sterba unsigned long bit; 2589572d9ab7SDavid Sterba 25906c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2591572d9ab7SDavid Sterba if (!prev) 2592572d9ab7SDavid Sterba return; 2593572d9ab7SDavid Sterba 2594d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2595d51033d0SDavid Sterba if (prev & bit) 2596d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2597d51033d0SDavid Sterba prev &= ~bit; 2598d51033d0SDavid Sterba 2599572d9ab7SDavid Sterba if (prev) 2600572d9ab7SDavid Sterba btrfs_warn(fs_info, 2601572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2602572d9ab7SDavid Sterba } 2603