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> 9ab3c5c18SSweet Tea Dorminy #include <linux/sched/mm.h> 10d3c2fdcfSChris Mason #include <linux/writeback.h> 115f39d397SChris Mason #include <linux/pagemap.h> 125f2cc086SChris Mason #include <linux/blkdev.h> 138ea05e3aSAlexander Block #include <linux/uuid.h> 14e55958c8SIoannis Angelakopoulos #include <linux/timekeeping.h> 15602cbe91SDavid Sterba #include "misc.h" 1679154b1bSChris Mason #include "ctree.h" 1779154b1bSChris Mason #include "disk-io.h" 1879154b1bSChris Mason #include "transaction.h" 19925baeddSChris Mason #include "locking.h" 20e02119d5SChris Mason #include "tree-log.h" 21733f4fbbSStefan Behrens #include "volumes.h" 228dabb742SStefan Behrens #include "dev-replace.h" 23fcebe456SJosef Bacik #include "qgroup.h" 24aac0023cSJosef Bacik #include "block-group.h" 259c343784SJosef Bacik #include "space-info.h" 26d3575156SNaohiro Aota #include "zoned.h" 27c7f13d42SJosef Bacik #include "fs.h" 2807e81dc9SJosef Bacik #include "accessors.h" 29a0231804SJosef Bacik #include "extent-tree.h" 3045c40c8fSJosef Bacik #include "root-tree.h" 31*59b818e0SJosef Bacik #include "defrag.h" 3279154b1bSChris Mason 33956504a3SJosef Bacik static struct kmem_cache *btrfs_trans_handle_cachep; 34956504a3SJosef Bacik 35fc7cbcd4SDavid Sterba #define BTRFS_ROOT_TRANS_TAG 0 360f7d52f4SChris Mason 3761c047b5SQu Wenruo /* 3861c047b5SQu Wenruo * Transaction states and transitions 3961c047b5SQu Wenruo * 4061c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 4161c047b5SQu Wenruo * | 4261c047b5SQu Wenruo * | To next stage: 4361c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 4461c047b5SQu Wenruo * V 4561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 4661c047b5SQu Wenruo * | 4761c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 4861c047b5SQu Wenruo * | start_transaction() variants. 4961c047b5SQu Wenruo * | 5061c047b5SQu Wenruo * | To next stage: 5161c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 5261c047b5SQu Wenruo * | transaction N 5361c047b5SQu Wenruo * V 5461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 5561c047b5SQu Wenruo * | 5661c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 5761c047b5SQu Wenruo * | is one 5861c047b5SQu Wenruo * | 5961c047b5SQu Wenruo * | Then one of the following happes: 6061c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 6161c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 6261c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 6361c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 6461c047b5SQu Wenruo * | 6561c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 6661c047b5SQu Wenruo * | to this running transaction. 6761c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 6861c047b5SQu Wenruo * | transaction N+1. 6961c047b5SQu Wenruo * | 7061c047b5SQu Wenruo * | To next stage: 7161c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 7261c047b5SQu Wenruo * | haven been released. 7361c047b5SQu Wenruo * V 7461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 7561c047b5SQu Wenruo * | 7661c047b5SQu Wenruo * | The heavy lifting transaction work is started. 7761c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 7861c047b5SQu Wenruo * | snapshots, running qgroups. 7961c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 8061c047b5SQu Wenruo * | trees. 8161c047b5SQu Wenruo * | 8261c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 8361c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 8461c047b5SQu Wenruo * | 8561c047b5SQu Wenruo * | To next stage: 8661c047b5SQu Wenruo * | Until all supporting trees are updated. 8761c047b5SQu Wenruo * V 8861c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 8961c047b5SQu Wenruo * | Transaction N+1 9061c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 9161c047b5SQu Wenruo * | need to write them back to disk and update | 9261c047b5SQu Wenruo * | super blocks. | 9361c047b5SQu Wenruo * | | 9461c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 9561c047b5SQu Wenruo * | start. | 9661c047b5SQu Wenruo * | All new start_transaction() calls will be | 9761c047b5SQu Wenruo * | attached to transid N+1. | 9861c047b5SQu Wenruo * | | 9961c047b5SQu Wenruo * | To next stage: | 10061c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 10161c047b5SQu Wenruo * | written to block devices | 10261c047b5SQu Wenruo * V | 10361c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 10461c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 10561c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 10661c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 10761c047b5SQu Wenruo */ 108e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1094a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 110bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 111bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1124a9d8bdeSMiao Xie __TRANS_ATTACH | 113a6d155d2SFilipe Manana __TRANS_JOIN | 114a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 115bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1164a9d8bdeSMiao Xie __TRANS_ATTACH | 1174a9d8bdeSMiao Xie __TRANS_JOIN | 118a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 119a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 120d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 121d0c2f4faSFilipe Manana __TRANS_ATTACH | 122d0c2f4faSFilipe Manana __TRANS_JOIN | 123d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 124d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 125bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1264a9d8bdeSMiao Xie __TRANS_ATTACH | 1274a9d8bdeSMiao Xie __TRANS_JOIN | 128a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 129a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1304a9d8bdeSMiao Xie }; 1314a9d8bdeSMiao Xie 132724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 13379154b1bSChris Mason { 1349b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1359b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 136a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1375c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1385c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 13981f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 14081f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1411262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 142ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 143ab8d0fc4SJeff Mahoney "pending csums is %llu", 1441262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1457785a663SFilipe Manana /* 1467785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1477785a663SFilipe Manana * because the transaction was aborted and a commit did not 1487785a663SFilipe Manana * happen (things failed before writing the new superblock 1497785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1507785a663SFilipe Manana * discard the physical locations of the block groups. 1517785a663SFilipe Manana */ 1527785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 15332da5386SDavid Sterba struct btrfs_block_group *cache; 1547785a663SFilipe Manana 1557785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 15632da5386SDavid Sterba struct btrfs_block_group, 1577785a663SFilipe Manana bg_list); 1587785a663SFilipe Manana list_del_init(&cache->bg_list); 1596b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1607785a663SFilipe Manana btrfs_put_block_group(cache); 1617785a663SFilipe Manana } 162bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1634b5faeacSDavid Sterba kfree(transaction); 16479154b1bSChris Mason } 16578fae27eSChris Mason } 16679154b1bSChris Mason 167889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 168817d52f8SJosef Bacik { 169889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 17016916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1719e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 1729e351cc8SJosef Bacik 173dfba78dcSFilipe Manana /* 174dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 175dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 176dfba78dcSFilipe Manana */ 177dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 178dfba78dcSFilipe Manana 1799e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 180d96b3424SFilipe Manana 181d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 182d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 183d96b3424SFilipe Manana 184889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1859e351cc8SJosef Bacik dirty_list) { 1869e351cc8SJosef Bacik list_del_init(&root->dirty_list); 187817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 188817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 18941e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 190370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1919e351cc8SJosef Bacik } 1922b9dbef2SJosef Bacik 1932b9dbef2SJosef Bacik /* We can free old roots now. */ 194889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 195889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 196889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1972b9dbef2SJosef Bacik struct btrfs_root, root_list); 1982b9dbef2SJosef Bacik list_del_init(&root->root_list); 199889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 200889bfa39SJosef Bacik btrfs_free_log(trans, root); 2012b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 202889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 2032b9dbef2SJosef Bacik } 204889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 20527d56e62SJosef Bacik 2069e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 207817d52f8SJosef Bacik } 208817d52f8SJosef Bacik 2090860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2100860adfdSMiao Xie unsigned int type) 2110860adfdSMiao Xie { 2120860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2130860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2140860adfdSMiao Xie } 2150860adfdSMiao Xie 2160860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2170860adfdSMiao Xie unsigned int type) 2180860adfdSMiao Xie { 2190860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2200860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2210860adfdSMiao Xie } 2220860adfdSMiao Xie 2230860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2240860adfdSMiao Xie unsigned int type) 2250860adfdSMiao Xie { 2260860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2270860adfdSMiao Xie } 2280860adfdSMiao Xie 2290860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2300860adfdSMiao Xie { 2310860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 232178260b2SMiao Xie } 233178260b2SMiao Xie 234d352ac68SChris Mason /* 23579bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 23679bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 23779bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 23879bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 23979bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 240fb6dea26SJosef Bacik */ 241fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 242fb6dea26SJosef Bacik { 243fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 244fb6dea26SJosef Bacik 245fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 246fb6dea26SJosef Bacik return; 247fb6dea26SJosef Bacik 248fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 24963f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 250fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 251fb6dea26SJosef Bacik } 252fb6dea26SJosef Bacik 253fb6dea26SJosef Bacik /* 254d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 255d352ac68SChris Mason */ 2562ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2572ff7e61eSJeff Mahoney unsigned int type) 25879154b1bSChris Mason { 25979154b1bSChris Mason struct btrfs_transaction *cur_trans; 260a4abeea4SJosef Bacik 26119ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 262d43317dcSChris Mason loop: 26349b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 26484961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 26519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26649b25e05SJeff Mahoney return -EROFS; 26749b25e05SJeff Mahoney } 26849b25e05SJeff Mahoney 26919ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 270a4abeea4SJosef Bacik if (cur_trans) { 271bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 27219ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 27349b25e05SJeff Mahoney return cur_trans->aborted; 274871383beSDavid Sterba } 2754a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 276178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 277178260b2SMiao Xie return -EBUSY; 278178260b2SMiao Xie } 2799b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 280a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2810860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 28219ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 283e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2845a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 285a4abeea4SJosef Bacik return 0; 286a4abeea4SJosef Bacik } 28719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 288a4abeea4SJosef Bacik 289354aa0fbSMiao Xie /* 290354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 291354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 292354aa0fbSMiao Xie */ 293354aa0fbSMiao Xie if (type == TRANS_ATTACH) 294354aa0fbSMiao Xie return -ENOENT; 295354aa0fbSMiao Xie 2964a9d8bdeSMiao Xie /* 2974a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 2984a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 2994a9d8bdeSMiao Xie */ 3004a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3014a9d8bdeSMiao Xie 3024b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 303db5b493aSTsutomu Itoh if (!cur_trans) 304db5b493aSTsutomu Itoh return -ENOMEM; 305d43317dcSChris Mason 306e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 3075a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 308e1489b4fSIoannis Angelakopoulos 30919ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 31019ae4e81SJan Schmidt if (fs_info->running_transaction) { 311d43317dcSChris Mason /* 312d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3134a9d8bdeSMiao Xie * to redo the checks above 314d43317dcSChris Mason */ 3155a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 316e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3174b5faeacSDavid Sterba kfree(cur_trans); 318d43317dcSChris Mason goto loop; 31984961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 320e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3215a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 322e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3234b5faeacSDavid Sterba kfree(cur_trans); 3247b8b92afSJosef Bacik return -EROFS; 325a4abeea4SJosef Bacik } 326d43317dcSChris Mason 327ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 32848778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 32948778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 33013c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3310860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 33279154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 33379154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3344a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 335a4abeea4SJosef Bacik /* 336a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 337a4abeea4SJosef Bacik * commit the transaction. 338a4abeea4SJosef Bacik */ 3399b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3403204d33cSJosef Bacik cur_trans->flags = 0; 341afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 34256bec294SChris Mason 343a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 344a099d0fdSAlexandru Moise 3455c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3463368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 347d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 34820b297d6SJan Schmidt 34920b297d6SJan Schmidt /* 35020b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 35120b297d6SJan Schmidt * the log must never go across transaction boundaries. 35220b297d6SJan Schmidt */ 35320b297d6SJan Schmidt smp_mb(); 35431b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3555d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 35631b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3575d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 358fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 35920b297d6SJan Schmidt 36056bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 36156bec294SChris Mason 3623063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 363bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3649e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 365ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3661bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3672b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3681bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 369ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 370e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3712b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 372d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 373d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 37419ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 375c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 376efb0645bSJosef Bacik IO_TREE_TRANS_DIRTY_PAGES, NULL); 377fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 378fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 37919ae4e81SJan Schmidt fs_info->generation++; 38019ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 38119ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 38249b25e05SJeff Mahoney cur_trans->aborted = 0; 38319ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 38415ee9bc7SJosef Bacik 38579154b1bSChris Mason return 0; 38679154b1bSChris Mason } 38779154b1bSChris Mason 388d352ac68SChris Mason /* 38992a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 39092a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 39192a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 39292a7cc42SQu Wenruo * transaction commits. 393d352ac68SChris Mason */ 3947585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3956426c7adSQu Wenruo struct btrfs_root *root, 3966426c7adSQu Wenruo int force) 3976702ed49SChris Mason { 3980b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 39903a7e111SJosef Bacik int ret = 0; 4000b246afaSJeff Mahoney 40192a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4026426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4034d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4045d4f98a2SYan Zheng 4057585717fSChris Mason /* 40627cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4077585717fSChris Mason * we have the reloc mutex held now, so there 4087585717fSChris Mason * is only one writer in this function 4097585717fSChris Mason */ 41027cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4117585717fSChris Mason 41227cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4137585717fSChris Mason * they find our root->last_trans update 4147585717fSChris Mason */ 4157585717fSChris Mason smp_wmb(); 4167585717fSChris Mason 417fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4186426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 419fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 420a4abeea4SJosef Bacik return 0; 421a4abeea4SJosef Bacik } 422fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4236702ed49SChris Mason (unsigned long)root->root_key.objectid, 4246702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 425fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4267585717fSChris Mason root->last_trans = trans->transid; 4277585717fSChris Mason 4287585717fSChris Mason /* this is pretty tricky. We don't want to 4297585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4307585717fSChris Mason * unless we're really doing the first setup for this root in 4317585717fSChris Mason * this transaction. 4327585717fSChris Mason * 4337585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4347585717fSChris Mason * if we want to take the expensive mutex. 4357585717fSChris Mason * 4367585717fSChris Mason * But, we have to set root->last_trans before we 4377585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4387585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 43927cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4407585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4417585717fSChris Mason * 4427585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4437585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4447585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4457585717fSChris Mason * done before we pop in the zero below 4467585717fSChris Mason */ 44703a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 448c7548af6SChris Mason smp_mb__before_atomic(); 44927cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4506702ed49SChris Mason } 45103a7e111SJosef Bacik return ret; 4526702ed49SChris Mason } 4535d4f98a2SYan Zheng 4547585717fSChris Mason 4552b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4562b9dbef2SJosef Bacik struct btrfs_root *root) 4572b9dbef2SJosef Bacik { 4580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4592b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4602b9dbef2SJosef Bacik 4612b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4622b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4632b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4642b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4652b9dbef2SJosef Bacik 4662b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 467fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 468fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4692b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4702b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 471fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4722b9dbef2SJosef Bacik } 4732b9dbef2SJosef Bacik 4747585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4757585717fSChris Mason struct btrfs_root *root) 4767585717fSChris Mason { 4770b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4781409e6ccSJosef Bacik int ret; 4790b246afaSJeff Mahoney 48092a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4817585717fSChris Mason return 0; 4827585717fSChris Mason 4837585717fSChris Mason /* 48427cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4857585717fSChris Mason * and barriers 4867585717fSChris Mason */ 4877585717fSChris Mason smp_rmb(); 4887585717fSChris Mason if (root->last_trans == trans->transid && 48927cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4907585717fSChris Mason return 0; 4917585717fSChris Mason 4920b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4931409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4940b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4957585717fSChris Mason 4961409e6ccSJosef Bacik return ret; 4977585717fSChris Mason } 4987585717fSChris Mason 4994a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5004a9d8bdeSMiao Xie { 5013296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 502501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 503bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5044a9d8bdeSMiao Xie } 5054a9d8bdeSMiao Xie 506d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 507d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 508d352ac68SChris Mason * transaction might not be fully on disk. 509d352ac68SChris Mason */ 5102ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 51179154b1bSChris Mason { 512f9295749SChris Mason struct btrfs_transaction *cur_trans; 51379154b1bSChris Mason 5140b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5150b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5164a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5179b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5180b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51972d63ed6SLi Zefan 5203e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5210b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 522501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 523bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 524724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 525a4abeea4SJosef Bacik } else { 5260b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 527f9295749SChris Mason } 52837d1aeeeSChris Mason } 52937d1aeeeSChris Mason 5302ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 53137d1aeeeSChris Mason { 5320b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 533a4abeea4SJosef Bacik return 0; 534a4abeea4SJosef Bacik 53592e2f7e3SNikolay Borisov if (type == TRANS_START) 536a4abeea4SJosef Bacik return 1; 537a4abeea4SJosef Bacik 538a22285a6SYan, Zheng return 0; 539a22285a6SYan, Zheng } 540a22285a6SYan, Zheng 54120dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 54220dd2cbfSMiao Xie { 5430b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5440b246afaSJeff Mahoney 5450b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 54692a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 54720dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 54820dd2cbfSMiao Xie root->reloc_root) 54920dd2cbfSMiao Xie return false; 55020dd2cbfSMiao Xie 55120dd2cbfSMiao Xie return true; 55220dd2cbfSMiao Xie } 55320dd2cbfSMiao Xie 55408e007d2SMiao Xie static struct btrfs_trans_handle * 5555aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 556003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 557003d7c59SJeff Mahoney bool enforce_qgroups) 558a22285a6SYan, Zheng { 5590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 560ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 561a22285a6SYan, Zheng struct btrfs_trans_handle *h; 562a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 563b5009945SJosef Bacik u64 num_bytes = 0; 564c5567237SArne Jansen u64 qgroup_reserved = 0; 56520dd2cbfSMiao Xie bool reloc_reserved = false; 5669c343784SJosef Bacik bool do_chunk_alloc = false; 56720dd2cbfSMiao Xie int ret; 568acce952bSliubo 56984961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 570acce952bSliubo return ERR_PTR(-EROFS); 5712a1eb461SJosef Bacik 57246c4e71eSFilipe Manana if (current->journal_info) { 5730860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5742a1eb461SJosef Bacik h = current->journal_info; 575b50fff81SDavid Sterba refcount_inc(&h->use_count); 576b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5772a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5782a1eb461SJosef Bacik h->block_rsv = NULL; 5792a1eb461SJosef Bacik goto got_it; 5802a1eb461SJosef Bacik } 581b5009945SJosef Bacik 582b5009945SJosef Bacik /* 583b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 584b5009945SJosef Bacik * the appropriate flushing if need be. 585b5009945SJosef Bacik */ 586003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 587ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 588ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 589ba2c4d4eSJosef Bacik 5900b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 591733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 592003d7c59SJeff Mahoney enforce_qgroups); 593c5567237SArne Jansen if (ret) 594c5567237SArne Jansen return ERR_PTR(ret); 595c5567237SArne Jansen 596ba2c4d4eSJosef Bacik /* 597ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 598ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 599ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 600ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 601ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 602ba2c4d4eSJosef Bacik */ 6032bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6047f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 605748f553cSDavid Sterba btrfs_block_rsv_full(delayed_refs_rsv) == 0) { 606ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 607ba2c4d4eSJosef Bacik num_bytes <<= 1; 608ba2c4d4eSJosef Bacik } 609ba2c4d4eSJosef Bacik 61020dd2cbfSMiao Xie /* 61120dd2cbfSMiao Xie * Do the reservation for the relocation root creation 61220dd2cbfSMiao Xie */ 613ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6140b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 61520dd2cbfSMiao Xie reloc_reserved = true; 61620dd2cbfSMiao Xie } 61720dd2cbfSMiao Xie 6189270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 619ba2c4d4eSJosef Bacik if (ret) 620ba2c4d4eSJosef Bacik goto reserve_fail; 621ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 622ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 623ba2c4d4eSJosef Bacik delayed_refs_bytes); 624ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 625ba2c4d4eSJosef Bacik } 6269c343784SJosef Bacik 6279c343784SJosef Bacik if (rsv->space_info->force_alloc) 6289c343784SJosef Bacik do_chunk_alloc = true; 629ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 630748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 631ba2c4d4eSJosef Bacik /* 632ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 633ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 634ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 635ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 636ba2c4d4eSJosef Bacik * here. 637ba2c4d4eSJosef Bacik */ 638ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 639b5009945SJosef Bacik if (ret) 640843fcf35SMiao Xie goto reserve_fail; 641b5009945SJosef Bacik } 642a22285a6SYan, Zheng again: 643f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 644843fcf35SMiao Xie if (!h) { 645843fcf35SMiao Xie ret = -ENOMEM; 646843fcf35SMiao Xie goto alloc_fail; 647843fcf35SMiao Xie } 648a22285a6SYan, Zheng 64998114659SJosef Bacik /* 65098114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 65198114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 65298114659SJosef Bacik * because we're already holding a ref. We need this because we could 65398114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 65498114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 655354aa0fbSMiao Xie * 656354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 657354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 65898114659SJosef Bacik */ 6590860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6600b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 661b2b5ef5cSJan Kara 6622ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6632ff7e61eSJeff Mahoney wait_current_trans(fs_info); 664a22285a6SYan, Zheng 665a4abeea4SJosef Bacik do { 6662ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 667178260b2SMiao Xie if (ret == -EBUSY) { 6682ff7e61eSJeff Mahoney wait_current_trans(fs_info); 669a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 670a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 671178260b2SMiao Xie ret = -ENOENT; 672178260b2SMiao Xie } 673a4abeea4SJosef Bacik } while (ret == -EBUSY); 674a4abeea4SJosef Bacik 675a43f7f82SLiu Bo if (ret < 0) 676843fcf35SMiao Xie goto join_fail; 6770f7d52f4SChris Mason 6780b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 679a22285a6SYan, Zheng 680a22285a6SYan, Zheng h->transid = cur_trans->transid; 681a22285a6SYan, Zheng h->transaction = cur_trans; 682b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 68364b63580SJeff Mahoney h->fs_info = root->fs_info; 6847174109cSQu Wenruo 685a698d075SMiao Xie h->type = type; 686ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 687b7ec40d7SChris Mason 688a22285a6SYan, Zheng smp_mb(); 6893296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 6902ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 691abdd2e80SFilipe Manana current->journal_info = h; 6923a45bb20SJeff Mahoney btrfs_commit_transaction(h); 693a22285a6SYan, Zheng goto again; 694a22285a6SYan, Zheng } 6959ed74f2dSJosef Bacik 696b5009945SJosef Bacik if (num_bytes) { 6970b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 6982bcc0328SLiu Bo h->transid, num_bytes, 1); 6990b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 700b5009945SJosef Bacik h->bytes_reserved = num_bytes; 70120dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 702a22285a6SYan, Zheng } 703a22285a6SYan, Zheng 7042a1eb461SJosef Bacik got_it: 705bcf3a3e7SNikolay Borisov if (!current->journal_info) 706a22285a6SYan, Zheng current->journal_info = h; 707fcc99734SQu Wenruo 708fcc99734SQu Wenruo /* 7099c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7109c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7119c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7129c343784SJosef Bacik * value here. 7139c343784SJosef Bacik */ 7149c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7159c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7169c343784SJosef Bacik 7179c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7189c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7199c343784SJosef Bacik } 7209c343784SJosef Bacik 7219c343784SJosef Bacik /* 722fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 723fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 724fcc99734SQu Wenruo * transaction. 725fcc99734SQu Wenruo * 726fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 727fcc99734SQu Wenruo * or we can deadlock. 728fcc99734SQu Wenruo */ 72968075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 73068075ea8SJosef Bacik if (ret) { 73168075ea8SJosef Bacik /* 73268075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 73368075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 73468075ea8SJosef Bacik * not just freed. 73568075ea8SJosef Bacik */ 73668075ea8SJosef Bacik btrfs_end_transaction(h); 73768075ea8SJosef Bacik return ERR_PTR(ret); 73868075ea8SJosef Bacik } 739fcc99734SQu Wenruo 74079154b1bSChris Mason return h; 741843fcf35SMiao Xie 742843fcf35SMiao Xie join_fail: 7430860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7440b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 745843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 746843fcf35SMiao Xie alloc_fail: 747843fcf35SMiao Xie if (num_bytes) 7482ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 74963f018beSNikolay Borisov num_bytes, NULL); 750843fcf35SMiao Xie reserve_fail: 751733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 752843fcf35SMiao Xie return ERR_PTR(ret); 75379154b1bSChris Mason } 75479154b1bSChris Mason 755f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7565aed1dd8SAlexandru Moise unsigned int num_items) 757f9295749SChris Mason { 75808e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 759003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 760f9295749SChris Mason } 761003d7c59SJeff Mahoney 7628eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7638eab77ffSFilipe Manana struct btrfs_root *root, 7647f9fe614SJosef Bacik unsigned int num_items) 7658eab77ffSFilipe Manana { 7667f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7677f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7688eab77ffSFilipe Manana } 7698407aa46SMiao Xie 7707a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 771f9295749SChris Mason { 772003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 773003d7c59SJeff Mahoney true); 774f9295749SChris Mason } 775f9295749SChris Mason 7768d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7770af3d00bSJosef Bacik { 778575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 779003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7800af3d00bSJosef Bacik } 7810af3d00bSJosef Bacik 782d4edf39bSMiao Xie /* 783a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 784a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 785a6d155d2SFilipe Manana */ 786a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 787a6d155d2SFilipe Manana { 788a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 789a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 790a6d155d2SFilipe Manana } 791a6d155d2SFilipe Manana 792a6d155d2SFilipe Manana /* 793d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 794d4edf39bSMiao Xie * 795d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 796d4edf39bSMiao Xie * don't want to start a new one. 797d4edf39bSMiao Xie * 798d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 799d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 800d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 801d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 802d4edf39bSMiao Xie * invoke 803d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 804d4edf39bSMiao Xie */ 805354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 80660376ce4SJosef Bacik { 807575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 808003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 80960376ce4SJosef Bacik } 81060376ce4SJosef Bacik 811d4edf39bSMiao Xie /* 81290b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 813d4edf39bSMiao Xie * 81452042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 815d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 816d4edf39bSMiao Xie * complete. 817d4edf39bSMiao Xie */ 818d4edf39bSMiao Xie struct btrfs_trans_handle * 819d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 820d4edf39bSMiao Xie { 821d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 822d4edf39bSMiao Xie 823575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 824003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8258d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8262ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 827d4edf39bSMiao Xie 828d4edf39bSMiao Xie return trans; 829d4edf39bSMiao Xie } 830d4edf39bSMiao Xie 831d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 832d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 833d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 83489ce8a63SChris Mason { 8355fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8365fd76bf3SOmar Sandoval u64 transid = commit->transid; 8375fd76bf3SOmar Sandoval bool put = false; 8385fd76bf3SOmar Sandoval 8393e738c53SIoannis Angelakopoulos /* 8403e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8413e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8423e738c53SIoannis Angelakopoulos */ 8433e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8443e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8453e738c53SIoannis Angelakopoulos else 8463e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8473e738c53SIoannis Angelakopoulos 8485fd76bf3SOmar Sandoval while (1) { 849d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8505fd76bf3SOmar Sandoval if (put) 8515fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8525fd76bf3SOmar Sandoval 8535fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8545fd76bf3SOmar Sandoval break; 8555fd76bf3SOmar Sandoval 8565fd76bf3SOmar Sandoval /* 8575fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8585fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8595fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8605fd76bf3SOmar Sandoval * transaction. Wait for those. 8615fd76bf3SOmar Sandoval */ 8625fd76bf3SOmar Sandoval 8635fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8645fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8655fd76bf3SOmar Sandoval struct btrfs_transaction, 8665fd76bf3SOmar Sandoval list); 8675fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8685fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8695fd76bf3SOmar Sandoval break; 8705fd76bf3SOmar Sandoval } 8715fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8725fd76bf3SOmar Sandoval put = true; 8735fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8745fd76bf3SOmar Sandoval } 87589ce8a63SChris Mason } 87689ce8a63SChris Mason 8772ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 87846204592SSage Weil { 87946204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8808cd2807fSMiao Xie int ret = 0; 88146204592SSage Weil 88246204592SSage Weil if (transid) { 8830b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 884a4abeea4SJosef Bacik goto out; 88546204592SSage Weil 88646204592SSage Weil /* find specified transaction */ 8870b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8880b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 88946204592SSage Weil if (t->transid == transid) { 89046204592SSage Weil cur_trans = t; 8919b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8928cd2807fSMiao Xie ret = 0; 89346204592SSage Weil break; 89446204592SSage Weil } 8958cd2807fSMiao Xie if (t->transid > transid) { 8968cd2807fSMiao Xie ret = 0; 89746204592SSage Weil break; 89846204592SSage Weil } 8998cd2807fSMiao Xie } 9000b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 90142383020SSage Weil 90242383020SSage Weil /* 90342383020SSage Weil * The specified transaction doesn't exist, or we 90442383020SSage Weil * raced with btrfs_commit_transaction 90542383020SSage Weil */ 90642383020SSage Weil if (!cur_trans) { 9070b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 90842383020SSage Weil ret = -EINVAL; 9098cd2807fSMiao Xie goto out; 91042383020SSage Weil } 91146204592SSage Weil } else { 91246204592SSage Weil /* find newest transaction that is committing | committed */ 9130b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9140b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 91546204592SSage Weil list) { 9164a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9174a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9183473f3c0SJosef Bacik break; 91946204592SSage Weil cur_trans = t; 9209b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 92146204592SSage Weil break; 92246204592SSage Weil } 92346204592SSage Weil } 9240b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 92546204592SSage Weil if (!cur_trans) 926a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 92746204592SSage Weil } 92846204592SSage Weil 929d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 930724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 931a4abeea4SJosef Bacik out: 93246204592SSage Weil return ret; 93346204592SSage Weil } 93446204592SSage Weil 9352ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 93637d1aeeeSChris Mason { 9372ff7e61eSJeff Mahoney wait_current_trans(fs_info); 93837d1aeeeSChris Mason } 93937d1aeeeSChris Mason 9408a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9418929ecfaSYan, Zheng { 9422ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9430b246afaSJeff Mahoney 94464403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9458a8f4deaSNikolay Borisov return true; 94636ba022aSJosef Bacik 9472ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9488929ecfaSYan, Zheng } 9498929ecfaSYan, Zheng 950a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9518929ecfaSYan, Zheng { 9528929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9538929ecfaSYan, Zheng 9543296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 955e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 956a2633b6aSNikolay Borisov return true; 9578929ecfaSYan, Zheng 9582ff7e61eSJeff Mahoney return should_end_transaction(trans); 9598929ecfaSYan, Zheng } 9608929ecfaSYan, Zheng 961dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 962dc60c525SNikolay Borisov 9630e34693fSNikolay Borisov { 964dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 965dc60c525SNikolay Borisov 9660e34693fSNikolay Borisov if (!trans->block_rsv) { 9670e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9680e34693fSNikolay Borisov return; 9690e34693fSNikolay Borisov } 9700e34693fSNikolay Borisov 9710e34693fSNikolay Borisov if (!trans->bytes_reserved) 9720e34693fSNikolay Borisov return; 9730e34693fSNikolay Borisov 9740e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9750e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9760e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9770e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 97863f018beSNikolay Borisov trans->bytes_reserved, NULL); 9790e34693fSNikolay Borisov trans->bytes_reserved = 0; 9800e34693fSNikolay Borisov } 9810e34693fSNikolay Borisov 98289ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9833a45bb20SJeff Mahoney int throttle) 98479154b1bSChris Mason { 9853a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9868929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9874edc2ca3SDave Jones int err = 0; 988d6e4a428SChris Mason 989b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 990b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9912a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9922a1eb461SJosef Bacik return 0; 9932a1eb461SJosef Bacik } 9942a1eb461SJosef Bacik 995dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9964c13d758SJosef Bacik trans->block_rsv = NULL; 997c5567237SArne Jansen 9986c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 999ea658badSJosef Bacik 10004fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 10014fbcdf66SFilipe Manana 10020860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 10030b246afaSJeff Mahoney sb_end_intwrite(info->sb); 10046df7881aSJosef Bacik 10058929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 100613c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 100713c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10080860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 100989ce8a63SChris Mason 1010093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1011e1489b4fSIoannis Angelakopoulos 10125a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1013e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1014e1489b4fSIoannis Angelakopoulos 1015724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10169ed74f2dSJosef Bacik 10179ed74f2dSJosef Bacik if (current->journal_info == trans) 10189ed74f2dSJosef Bacik current->journal_info = NULL; 1019ab78c84dSChris Mason 102024bbcf04SYan, Zheng if (throttle) 10212ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 102224bbcf04SYan, Zheng 102384961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10244e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1025fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1026fbabd4a3SJosef Bacik err = trans->aborted; 1027fbabd4a3SJosef Bacik else 1028fbabd4a3SJosef Bacik err = -EROFS; 10294e121c06SJosef Bacik } 103049b25e05SJeff Mahoney 10314edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10324edc2ca3SDave Jones return err; 103379154b1bSChris Mason } 103479154b1bSChris Mason 10353a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 103689ce8a63SChris Mason { 10373a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 103889ce8a63SChris Mason } 103989ce8a63SChris Mason 10403a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 104189ce8a63SChris Mason { 10423a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 104316cdcec7SMiao Xie } 104416cdcec7SMiao Xie 1045d352ac68SChris Mason /* 1046d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1047d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1048690587d1SChris Mason * those extents are sent to disk but does not wait on them 1049d352ac68SChris Mason */ 10502ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10518cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 105279154b1bSChris Mason { 1053777e6bd7SChris Mason int err = 0; 10547c4452b9SChris Mason int werr = 0; 10550b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1056e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1057777e6bd7SChris Mason u64 start = 0; 10585f39d397SChris Mason u64 end; 10597c4452b9SChris Mason 10606300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10611728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1062e6138876SJosef Bacik mark, &cached_state)) { 1063663dfbb0SFilipe Manana bool wait_writeback = false; 1064663dfbb0SFilipe Manana 1065663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1066663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1067210aa277SDavid Sterba mark, &cached_state); 1068663dfbb0SFilipe Manana /* 1069663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1070663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1071663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1072663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1073bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1074bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1075bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1076bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1077bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1078663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 107941e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1080663dfbb0SFilipe Manana */ 1081663dfbb0SFilipe Manana if (err == -ENOMEM) { 1082663dfbb0SFilipe Manana err = 0; 1083663dfbb0SFilipe Manana wait_writeback = true; 1084663dfbb0SFilipe Manana } 1085663dfbb0SFilipe Manana if (!err) 10861728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10877c4452b9SChris Mason if (err) 10887c4452b9SChris Mason werr = err; 1089663dfbb0SFilipe Manana else if (wait_writeback) 1090663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1091e38e2ed7SFilipe Manana free_extent_state(cached_state); 1092663dfbb0SFilipe Manana cached_state = NULL; 10931728366eSJosef Bacik cond_resched(); 10941728366eSJosef Bacik start = end + 1; 10957c4452b9SChris Mason } 10966300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1097690587d1SChris Mason return werr; 1098690587d1SChris Mason } 1099690587d1SChris Mason 1100690587d1SChris Mason /* 1101690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1102690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1103690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1104690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1105690587d1SChris Mason */ 1106bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1107bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1108690587d1SChris Mason { 1109690587d1SChris Mason int err = 0; 1110690587d1SChris Mason int werr = 0; 11110b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1112e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1113690587d1SChris Mason u64 start = 0; 1114690587d1SChris Mason u64 end; 1115690587d1SChris Mason 11161728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1117e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1118663dfbb0SFilipe Manana /* 1119663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1120663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1121663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1122663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1123663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 112441e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1125663dfbb0SFilipe Manana */ 1126663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1127bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1128663dfbb0SFilipe Manana if (err == -ENOMEM) 1129663dfbb0SFilipe Manana err = 0; 1130663dfbb0SFilipe Manana if (!err) 11311728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1132777e6bd7SChris Mason if (err) 1133777e6bd7SChris Mason werr = err; 1134e38e2ed7SFilipe Manana free_extent_state(cached_state); 1135e38e2ed7SFilipe Manana cached_state = NULL; 1136777e6bd7SChris Mason cond_resched(); 11371728366eSJosef Bacik start = end + 1; 1138777e6bd7SChris Mason } 11397c4452b9SChris Mason if (err) 11407c4452b9SChris Mason werr = err; 1141bf89d38fSJeff Mahoney return werr; 1142bf89d38fSJeff Mahoney } 1143656f30dbSFilipe Manana 1144b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1145bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1146bf89d38fSJeff Mahoney { 1147bf89d38fSJeff Mahoney bool errors = false; 1148bf89d38fSJeff Mahoney int err; 1149bf89d38fSJeff Mahoney 1150bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1151bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1152bf89d38fSJeff Mahoney errors = true; 1153bf89d38fSJeff Mahoney 1154bf89d38fSJeff Mahoney if (errors && !err) 1155bf89d38fSJeff Mahoney err = -EIO; 1156bf89d38fSJeff Mahoney return err; 1157bf89d38fSJeff Mahoney } 1158bf89d38fSJeff Mahoney 1159bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1160bf89d38fSJeff Mahoney { 1161bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1162bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1163bf89d38fSJeff Mahoney bool errors = false; 1164bf89d38fSJeff Mahoney int err; 1165bf89d38fSJeff Mahoney 1166bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1167bf89d38fSJeff Mahoney 1168bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1169656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11700b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1171656f30dbSFilipe Manana errors = true; 1172656f30dbSFilipe Manana 1173656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11740b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1175656f30dbSFilipe Manana errors = true; 1176656f30dbSFilipe Manana 1177bf89d38fSJeff Mahoney if (errors && !err) 1178bf89d38fSJeff Mahoney err = -EIO; 1179bf89d38fSJeff Mahoney return err; 118079154b1bSChris Mason } 118179154b1bSChris Mason 1182690587d1SChris Mason /* 1183c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1184c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1185c9b577c0SNikolay Borisov * log commit. 1186c9b577c0SNikolay Borisov * 1187c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1188690587d1SChris Mason */ 118970458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1190d0c803c4SChris Mason { 1191663dfbb0SFilipe Manana int ret; 1192c9b577c0SNikolay Borisov int ret2; 1193c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 119470458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1195c9b577c0SNikolay Borisov struct blk_plug plug; 1196663dfbb0SFilipe Manana 1197c9b577c0SNikolay Borisov blk_start_plug(&plug); 1198c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1199c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1200c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1201c9b577c0SNikolay Borisov 120241e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1203663dfbb0SFilipe Manana 1204c9b577c0SNikolay Borisov if (ret) 1205663dfbb0SFilipe Manana return ret; 1206c9b577c0SNikolay Borisov else if (ret2) 1207c9b577c0SNikolay Borisov return ret2; 1208c9b577c0SNikolay Borisov else 1209c9b577c0SNikolay Borisov return 0; 1210d0c803c4SChris Mason } 1211d0c803c4SChris Mason 1212d352ac68SChris Mason /* 1213d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1214d352ac68SChris Mason * 1215d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1216d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1217d352ac68SChris Mason * allocation tree. 1218d352ac68SChris Mason * 1219d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1220d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1221d352ac68SChris Mason */ 12220b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 122379154b1bSChris Mason struct btrfs_root *root) 122479154b1bSChris Mason { 122579154b1bSChris Mason int ret; 12260b86a832SChris Mason u64 old_root_bytenr; 122786b9f2ecSYan, Zheng u64 old_root_used; 12280b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12290b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 123079154b1bSChris Mason 123186b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 123256bec294SChris Mason 123379154b1bSChris Mason while (1) { 12340b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 123586b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1236ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 123779154b1bSChris Mason break; 123887ef2bb4SChris Mason 12395d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 124079154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12410b86a832SChris Mason &root->root_key, 12420b86a832SChris Mason &root->root_item); 124349b25e05SJeff Mahoney if (ret) 124449b25e05SJeff Mahoney return ret; 124556bec294SChris Mason 124686b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1247e7070be1SJosef Bacik } 1248276e680dSYan Zheng 12490b86a832SChris Mason return 0; 12500b86a832SChris Mason } 12510b86a832SChris Mason 1252d352ac68SChris Mason /* 1253d352ac68SChris Mason * update all the cowonly tree roots on disk 125449b25e05SJeff Mahoney * 125549b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 125649b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 125749b25e05SJeff Mahoney * to clean up the delayed refs. 1258d352ac68SChris Mason */ 12599386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12600b86a832SChris Mason { 12619386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1262ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12631bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12640b86a832SChris Mason struct list_head *next; 126584234f3aSYan Zheng struct extent_buffer *eb; 126656bec294SChris Mason int ret; 126784234f3aSYan Zheng 1268dfba78dcSFilipe Manana /* 1269dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1270dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1271dfba78dcSFilipe Manana */ 1272dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1273dfba78dcSFilipe Manana 127484234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 127549b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12769631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 127784234f3aSYan Zheng btrfs_tree_unlock(eb); 127884234f3aSYan Zheng free_extent_buffer(eb); 12790b86a832SChris Mason 128049b25e05SJeff Mahoney if (ret) 128149b25e05SJeff Mahoney return ret; 128249b25e05SJeff Mahoney 1283196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1284c16ce190SJosef Bacik if (ret) 1285c16ce190SJosef Bacik return ret; 12862b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1287c16ce190SJosef Bacik if (ret) 1288c16ce190SJosef Bacik return ret; 1289280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1290c16ce190SJosef Bacik if (ret) 1291c16ce190SJosef Bacik return ret; 1292546adb0dSJan Schmidt 1293bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1294dcdf7f6dSJosef Bacik if (ret) 1295dcdf7f6dSJosef Bacik return ret; 1296dcdf7f6dSJosef Bacik 1297ea526d18SJosef Bacik again: 12980b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12992ff7e61eSJeff Mahoney struct btrfs_root *root; 13000b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 13010b86a832SChris Mason list_del_init(next); 13020b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1303e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 130487ef2bb4SChris Mason 13059e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13069e351cc8SJosef Bacik &trans->transaction->switch_commits); 130749b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 130849b25e05SJeff Mahoney if (ret) 130949b25e05SJeff Mahoney return ret; 1310488bc2a2SJosef Bacik } 1311488bc2a2SJosef Bacik 1312488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1313c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1314ea526d18SJosef Bacik if (ret) 1315ea526d18SJosef Bacik return ret; 1316276e680dSYan Zheng 13171bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13185742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1319ea526d18SJosef Bacik if (ret) 1320ea526d18SJosef Bacik return ret; 1321488bc2a2SJosef Bacik 1322488bc2a2SJosef Bacik /* 1323488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1324488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1325488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1326488bc2a2SJosef Bacik * everything gets run. 1327488bc2a2SJosef Bacik */ 1328c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1329ea526d18SJosef Bacik if (ret) 1330ea526d18SJosef Bacik return ret; 1331ea526d18SJosef Bacik } 1332ea526d18SJosef Bacik 1333ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1334ea526d18SJosef Bacik goto again; 1335ea526d18SJosef Bacik 13369f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13379f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13389f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13398dabb742SStefan Behrens 134079154b1bSChris Mason return 0; 134179154b1bSChris Mason } 134279154b1bSChris Mason 1343d352ac68SChris Mason /* 1344b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1345b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1346b4be6aefSJosef Bacik */ 1347b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1348b4be6aefSJosef Bacik { 1349b4be6aefSJosef Bacik /* 1350b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1351b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1352b4be6aefSJosef Bacik * up. 1353b4be6aefSJosef Bacik */ 1354b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1355b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1356b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1357b4be6aefSJosef Bacik struct btrfs_root, 1358b4be6aefSJosef Bacik root_list); 1359b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1360b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1361b4be6aefSJosef Bacik return; 1362b4be6aefSJosef Bacik } 1363b4be6aefSJosef Bacik } 1364b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1365b4be6aefSJosef Bacik 1366b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1367b4be6aefSJosef Bacik } 1368b4be6aefSJosef Bacik 1369b4be6aefSJosef Bacik /* 1370d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1371d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1372d352ac68SChris Mason * be deleted 1373d352ac68SChris Mason */ 1374cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13755eda7b5eSChris Mason { 13760b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13770b246afaSJeff Mahoney 13780b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1379dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1380dc9492c1SJosef Bacik btrfs_grab_root(root); 1381b4be6aefSJosef Bacik 1382b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1383b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1384b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1385b4be6aefSJosef Bacik else 13860b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1387dc9492c1SJosef Bacik } 13880b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13895eda7b5eSChris Mason } 13905eda7b5eSChris Mason 1391d352ac68SChris Mason /* 1392dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1393dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1394d352ac68SChris Mason */ 13957e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13960f7d52f4SChris Mason { 13977e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1398fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1399fc7cbcd4SDavid Sterba int i; 1400fc7cbcd4SDavid Sterba int ret; 140154aa1f4dSChris Mason 1402dfba78dcSFilipe Manana /* 1403dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1404dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1405dfba78dcSFilipe Manana */ 1406dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1407dfba78dcSFilipe Manana 1408fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1409fc7cbcd4SDavid Sterba while (1) { 1410fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1411fc7cbcd4SDavid Sterba (void **)gang, 0, 1412fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1413fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1414fc7cbcd4SDavid Sterba if (ret == 0) 1415fc7cbcd4SDavid Sterba break; 1416fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1417fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1418fc7cbcd4SDavid Sterba int ret2; 14194f4317c1SJosef Bacik 1420dfba78dcSFilipe Manana /* 1421dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1422dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1423dfba78dcSFilipe Manana */ 1424dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1425dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1426dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1427dfba78dcSFilipe Manana 1428fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14292619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14300f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1431fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 143231153d81SYan Zheng 1433e02119d5SChris Mason btrfs_free_log(trans, root); 1434fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1435fc7cbcd4SDavid Sterba if (ret2) 1436fc7cbcd4SDavid Sterba return ret2; 1437e02119d5SChris Mason 1438fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 143927cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1440c7548af6SChris Mason smp_mb__after_atomic(); 1441f1ebcc74SLiu Bo 1442978d910dSYan Zheng if (root->commit_root != root->node) { 14439e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14449e351cc8SJosef Bacik &trans->transaction->switch_commits); 1445fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1446fc7cbcd4SDavid Sterba root->node); 1447978d910dSYan Zheng } 144831153d81SYan Zheng 1449fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1450fc7cbcd4SDavid Sterba &root->root_key, 1451fc7cbcd4SDavid Sterba &root->root_item); 1452fc7cbcd4SDavid Sterba if (ret2) 1453fc7cbcd4SDavid Sterba return ret2; 1454fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1455733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14560f7d52f4SChris Mason } 1457fc7cbcd4SDavid Sterba } 1458fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14594f4317c1SJosef Bacik return 0; 14600f7d52f4SChris Mason } 14610f7d52f4SChris Mason 1462d352ac68SChris Mason /* 1463de78b51aSEric Sandeen * defrag a given btree. 1464de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1465d352ac68SChris Mason */ 1466de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1467e9d0b13bSChris Mason { 1468e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1469e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14708929ecfaSYan, Zheng int ret; 1471e9d0b13bSChris Mason 147227cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1473e9d0b13bSChris Mason return 0; 14748929ecfaSYan, Zheng 14756b80053dSChris Mason while (1) { 14768929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14776819703fSDavid Sterba if (IS_ERR(trans)) { 14786819703fSDavid Sterba ret = PTR_ERR(trans); 14796819703fSDavid Sterba break; 14806819703fSDavid Sterba } 14818929ecfaSYan, Zheng 1482de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14838929ecfaSYan, Zheng 14843a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14852ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1486e9d0b13bSChris Mason cond_resched(); 1487e9d0b13bSChris Mason 1488ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1489e9d0b13bSChris Mason break; 1490210549ebSDavid Sterba 1491ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1492ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1493210549ebSDavid Sterba ret = -EAGAIN; 1494210549ebSDavid Sterba break; 1495210549ebSDavid Sterba } 1496e9d0b13bSChris Mason } 149727cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14988929ecfaSYan, Zheng return ret; 1499e9d0b13bSChris Mason } 1500e9d0b13bSChris Mason 1501d352ac68SChris Mason /* 15026426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 15036426c7adSQu Wenruo * 15046426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 15056426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 15066426c7adSQu Wenruo * qgroup works. 15076426c7adSQu Wenruo */ 15086426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15096426c7adSQu Wenruo struct btrfs_root *src, 15106426c7adSQu Wenruo struct btrfs_root *parent, 15116426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15126426c7adSQu Wenruo u64 dst_objectid) 15136426c7adSQu Wenruo { 15146426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15156426c7adSQu Wenruo int ret; 15166426c7adSQu Wenruo 15176426c7adSQu Wenruo /* 15186426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15196426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15206426c7adSQu Wenruo * kick in anyway. 15216426c7adSQu Wenruo */ 15229ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15236426c7adSQu Wenruo return 0; 15246426c7adSQu Wenruo 15256426c7adSQu Wenruo /* 152652042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15274d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15284d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15294d31778aSQu Wenruo * item. 15304d31778aSQu Wenruo */ 15311c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15321c442d22SJosef Bacik if (ret) 15331c442d22SJosef Bacik return ret; 15344d31778aSQu Wenruo 15354d31778aSQu Wenruo /* 15362a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15372a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15382a4d84c1SJosef Bacik * 15392a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15402a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15412a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15422a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15432a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15442a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15452a4d84c1SJosef Bacik */ 15462a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15472a4d84c1SJosef Bacik if (ret) { 15482a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 154944365827SNaohiro Aota return ret; 15502a4d84c1SJosef Bacik } 15512a4d84c1SJosef Bacik 15527e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15536426c7adSQu Wenruo if (ret) 15546426c7adSQu Wenruo goto out; 1555460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15566426c7adSQu Wenruo if (ret < 0) 15576426c7adSQu Wenruo goto out; 15586426c7adSQu Wenruo 15596426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1560a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15616426c7adSQu Wenruo inherit); 15626426c7adSQu Wenruo if (ret < 0) 15636426c7adSQu Wenruo goto out; 15646426c7adSQu Wenruo 15656426c7adSQu Wenruo /* 15666426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15676426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15686426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15696426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15706426c7adSQu Wenruo * 2) write all btree blocks onto disk 15716426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15726426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15736426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15746426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15756426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15766426c7adSQu Wenruo */ 15779386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15786426c7adSQu Wenruo if (ret) 15796426c7adSQu Wenruo goto out; 1580889bfa39SJosef Bacik switch_commit_roots(trans); 158170458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15826426c7adSQu Wenruo if (ret) 1583f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15846426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15856426c7adSQu Wenruo 15866426c7adSQu Wenruo out: 15876426c7adSQu Wenruo /* 15886426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15896426c7adSQu Wenruo * last_trans == cur_transid. 15906426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15916426c7adSQu Wenruo * insert_dir_item() 15926426c7adSQu Wenruo */ 15936426c7adSQu Wenruo if (!ret) 15941c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15956426c7adSQu Wenruo return ret; 15966426c7adSQu Wenruo } 15976426c7adSQu Wenruo 15986426c7adSQu Wenruo /* 1599d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1600aec8030aSMiao Xie * transaction commit. This does the actual creation. 1601aec8030aSMiao Xie * 1602aec8030aSMiao Xie * Note: 1603aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1604aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1605aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1606d352ac68SChris Mason */ 160780b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16083063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16093063d29fSChris Mason { 161008d50ca3SNikolay Borisov 161108d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16123063d29fSChris Mason struct btrfs_key key; 161380b6794dSChris Mason struct btrfs_root_item *new_root_item; 16143063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16153063d29fSChris Mason struct btrfs_root *root = pending->root; 16166bdb72deSSage Weil struct btrfs_root *parent_root; 161798c9942aSLiu Bo struct btrfs_block_rsv *rsv; 1618ab3c5c18SSweet Tea Dorminy struct inode *parent_inode = pending->dir; 161942874b3dSMiao Xie struct btrfs_path *path; 162042874b3dSMiao Xie struct btrfs_dir_item *dir_item; 16213063d29fSChris Mason struct extent_buffer *tmp; 1622925baeddSChris Mason struct extent_buffer *old; 162395582b00SDeepa Dinamani struct timespec64 cur_time; 1624aec8030aSMiao Xie int ret = 0; 1625d68fc57bSYan, Zheng u64 to_reserve = 0; 16266bdb72deSSage Weil u64 index = 0; 1627a22285a6SYan, Zheng u64 objectid; 1628b83cc969SLi Zefan u64 root_flags; 1629ab3c5c18SSweet Tea Dorminy unsigned int nofs_flags; 1630ab3c5c18SSweet Tea Dorminy struct fscrypt_name fname; 16313063d29fSChris Mason 16328546b570SDavid Sterba ASSERT(pending->path); 16338546b570SDavid Sterba path = pending->path; 163442874b3dSMiao Xie 1635b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1636b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1637a22285a6SYan, Zheng 1638ab3c5c18SSweet Tea Dorminy /* 1639ab3c5c18SSweet Tea Dorminy * We're inside a transaction and must make sure that any potential 1640ab3c5c18SSweet Tea Dorminy * allocations with GFP_KERNEL in fscrypt won't recurse back to 1641ab3c5c18SSweet Tea Dorminy * filesystem. 1642ab3c5c18SSweet Tea Dorminy */ 1643ab3c5c18SSweet Tea Dorminy nofs_flags = memalloc_nofs_save(); 1644ab3c5c18SSweet Tea Dorminy pending->error = fscrypt_setup_filename(parent_inode, 1645ab3c5c18SSweet Tea Dorminy &pending->dentry->d_name, 0, 1646ab3c5c18SSweet Tea Dorminy &fname); 1647ab3c5c18SSweet Tea Dorminy memalloc_nofs_restore(nofs_flags); 1648ab3c5c18SSweet Tea Dorminy if (pending->error) 1649ab3c5c18SSweet Tea Dorminy goto free_pending; 1650ab3c5c18SSweet Tea Dorminy 1651543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1652aec8030aSMiao Xie if (pending->error) 1653ab3c5c18SSweet Tea Dorminy goto free_fname; 16543063d29fSChris Mason 1655d6726335SQu Wenruo /* 1656d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1657d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1658d6726335SQu Wenruo */ 1659d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1660d6726335SQu Wenruo 1661147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1662d68fc57bSYan, Zheng 1663d68fc57bSYan, Zheng if (to_reserve > 0) { 16649270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1665aec8030aSMiao Xie &pending->block_rsv, 166608e007d2SMiao Xie to_reserve, 166708e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1668aec8030aSMiao Xie if (pending->error) 1669d6726335SQu Wenruo goto clear_skip_qgroup; 1670d68fc57bSYan, Zheng } 1671d68fc57bSYan, Zheng 16723063d29fSChris Mason key.objectid = objectid; 1673a22285a6SYan, Zheng key.offset = (u64)-1; 1674a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16753063d29fSChris Mason 16766fa9700eSMiao Xie rsv = trans->block_rsv; 1677a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16782382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16790b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 168088d3a5aaSJosef Bacik trans->transid, 168188d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1682a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1683f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1684f0118cb6SJosef Bacik if (ret) 1685f0118cb6SJosef Bacik goto fail; 1686c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 168704b285f3SDeepa Dinamani 16886bdb72deSSage Weil /* 16896bdb72deSSage Weil * insert the directory item 16906bdb72deSSage Weil */ 1691877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 169249b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 169342874b3dSMiao Xie 169442874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 169542874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16964a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 16976db75318SSweet Tea Dorminy &fname.disk_name, 0); 169842874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1699fe66a05aSChris Mason pending->error = -EEXIST; 1700aec8030aSMiao Xie goto dir_item_existed; 170142874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 170242874b3dSMiao Xie ret = PTR_ERR(dir_item); 170366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17048732d44fSMiao Xie goto fail; 170579787eaaSJeff Mahoney } 170642874b3dSMiao Xie btrfs_release_path(path); 17076bdb72deSSage Weil 1708e999376fSChris Mason /* 1709e999376fSChris Mason * pull in the delayed directory update 1710e999376fSChris Mason * and the delayed inode item 1711e999376fSChris Mason * otherwise we corrupt the FS during 1712e999376fSChris Mason * snapshot 1713e999376fSChris Mason */ 1714e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 17158732d44fSMiao Xie if (ret) { /* Transaction aborted */ 171666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17178732d44fSMiao Xie goto fail; 17188732d44fSMiao Xie } 1719e999376fSChris Mason 1720f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1721f0118cb6SJosef Bacik if (ret) { 1722f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1723f0118cb6SJosef Bacik goto fail; 1724f0118cb6SJosef Bacik } 17256bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17266bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 172708fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17286bdb72deSSage Weil 1729b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1730b83cc969SLi Zefan if (pending->readonly) 1731b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1732b83cc969SLi Zefan else 1733b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1734b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1735b83cc969SLi Zefan 17368ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17378ea05e3aSAlexander Block trans->transid); 1738807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17398ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17408ea05e3aSAlexander Block BTRFS_UUID_SIZE); 174170023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 174270023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 174370023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17448ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17458ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17468ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17478ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 174870023da2SStefan Behrens } 17493cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17503cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 175170023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17528ea05e3aSAlexander Block 1753925baeddSChris Mason old = btrfs_lock_root_node(root); 17549631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17559631e4ccSJosef Bacik BTRFS_NESTING_COW); 175679787eaaSJeff Mahoney if (ret) { 175779787eaaSJeff Mahoney btrfs_tree_unlock(old); 175879787eaaSJeff Mahoney free_extent_buffer(old); 175966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17608732d44fSMiao Xie goto fail; 176179787eaaSJeff Mahoney } 176249b25e05SJeff Mahoney 176349b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 176479787eaaSJeff Mahoney /* clean up in any case */ 1765925baeddSChris Mason btrfs_tree_unlock(old); 1766925baeddSChris Mason free_extent_buffer(old); 17678732d44fSMiao Xie if (ret) { 176866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17698732d44fSMiao Xie goto fail; 17708732d44fSMiao Xie } 1771f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 177227cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1773f1ebcc74SLiu Bo smp_wmb(); 1774f1ebcc74SLiu Bo 17755d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1776a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1777a22285a6SYan, Zheng key.offset = trans->transid; 1778a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1779925baeddSChris Mason btrfs_tree_unlock(tmp); 17803063d29fSChris Mason free_extent_buffer(tmp); 17818732d44fSMiao Xie if (ret) { 178266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17838732d44fSMiao Xie goto fail; 17848732d44fSMiao Xie } 17850660b5afSChris Mason 1786a22285a6SYan, Zheng /* 1787a22285a6SYan, Zheng * insert root back/forward references 1788a22285a6SYan, Zheng */ 17896025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1790a22285a6SYan, Zheng parent_root->root_key.objectid, 17914a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 17926db75318SSweet Tea Dorminy &fname.disk_name); 17938732d44fSMiao Xie if (ret) { 179466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17958732d44fSMiao Xie goto fail; 17968732d44fSMiao Xie } 1797a22285a6SYan, Zheng 1798a22285a6SYan, Zheng key.offset = (u64)-1; 17992dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 180079787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 180179787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 18022d892ccdSFilipe Manana pending->snap = NULL; 180366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18048732d44fSMiao Xie goto fail; 180579787eaaSJeff Mahoney } 1806d68fc57bSYan, Zheng 180749b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 18088732d44fSMiao Xie if (ret) { 180966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18108732d44fSMiao Xie goto fail; 18118732d44fSMiao Xie } 1812361048f5SMiao Xie 18136426c7adSQu Wenruo /* 18146426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 18156426c7adSQu Wenruo * snapshot hack to do fast snapshot. 18166426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 18176426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 18186426c7adSQu Wenruo */ 18196426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18206426c7adSQu Wenruo pending->inherit, objectid); 18216426c7adSQu Wenruo if (ret < 0) 18226426c7adSQu Wenruo goto fail; 18236426c7adSQu Wenruo 18246db75318SSweet Tea Dorminy ret = btrfs_insert_dir_item(trans, &fname.disk_name, 18256db75318SSweet Tea Dorminy BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, 18266db75318SSweet Tea Dorminy index); 182742874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18289c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18298732d44fSMiao Xie if (ret) { 183066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18318732d44fSMiao Xie goto fail; 18328732d44fSMiao Xie } 183342874b3dSMiao Xie 18346ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 18356db75318SSweet Tea Dorminy fname.disk_name.len * 2); 1836c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1837c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1838729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1839dd5f9615SStefan Behrens if (ret) { 184066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1841dd5f9615SStefan Behrens goto fail; 1842dd5f9615SStefan Behrens } 1843807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1844807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1845cdb345a8SLu Fengqi objectid); 1846dd5f9615SStefan Behrens if (ret) { 184766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1848dd5f9615SStefan Behrens goto fail; 1849dd5f9615SStefan Behrens } 1850dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1851cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1852dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1853dd5f9615SStefan Behrens objectid); 1854dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 185566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1856dd5f9615SStefan Behrens goto fail; 1857dd5f9615SStefan Behrens } 1858dd5f9615SStefan Behrens } 1859d6726335SQu Wenruo 18603063d29fSChris Mason fail: 1861aec8030aSMiao Xie pending->error = ret; 1862aec8030aSMiao Xie dir_item_existed: 186398c9942aSLiu Bo trans->block_rsv = rsv; 18642382c5ccSLiu Bo trans->bytes_reserved = 0; 1865d6726335SQu Wenruo clear_skip_qgroup: 1866d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 1867ab3c5c18SSweet Tea Dorminy free_fname: 1868ab3c5c18SSweet Tea Dorminy fscrypt_free_filename(&fname); 1869ab3c5c18SSweet Tea Dorminy free_pending: 18706fa9700eSMiao Xie kfree(new_root_item); 1871b0c0ea63SDavid Sterba pending->root_item = NULL; 187242874b3dSMiao Xie btrfs_free_path(path); 18738546b570SDavid Sterba pending->path = NULL; 18748546b570SDavid Sterba 187549b25e05SJeff Mahoney return ret; 18763063d29fSChris Mason } 18773063d29fSChris Mason 1878d352ac68SChris Mason /* 1879d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1880d352ac68SChris Mason */ 188108d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18823063d29fSChris Mason { 1883aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18843063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1885aec8030aSMiao Xie int ret = 0; 18863de4586cSChris Mason 1887aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1888aec8030aSMiao Xie list_del(&pending->list); 188908d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1890aec8030aSMiao Xie if (ret) 1891aec8030aSMiao Xie break; 1892aec8030aSMiao Xie } 1893aec8030aSMiao Xie return ret; 18943de4586cSChris Mason } 18953de4586cSChris Mason 18962ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18975d4f98a2SYan Zheng { 18985d4f98a2SYan Zheng struct btrfs_root_item *root_item; 18995d4f98a2SYan Zheng struct btrfs_super_block *super; 19005d4f98a2SYan Zheng 19010b246afaSJeff Mahoney super = fs_info->super_copy; 19025d4f98a2SYan Zheng 19030b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1904093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1905093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1906093e037cSDavid Sterba super->chunk_root_level = root_item->level; 19075d4f98a2SYan Zheng 19080b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1909093e037cSDavid Sterba super->root = root_item->bytenr; 1910093e037cSDavid Sterba super->generation = root_item->generation; 1911093e037cSDavid Sterba super->root_level = root_item->level; 19120b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1913093e037cSDavid Sterba super->cache_generation = root_item->generation; 191494846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 191594846229SBoris Burkov super->cache_generation = 0; 19160b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1917093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 19185d4f98a2SYan Zheng } 19195d4f98a2SYan Zheng 1920f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1921f36f3042SChris Mason { 19224a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1923f36f3042SChris Mason int ret = 0; 19244a9d8bdeSMiao Xie 1925a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19264a9d8bdeSMiao Xie trans = info->running_transaction; 19274a9d8bdeSMiao Xie if (trans) 19284a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1929a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1930f36f3042SChris Mason return ret; 1931f36f3042SChris Mason } 1932f36f3042SChris Mason 19338929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19348929ecfaSYan, Zheng { 19354a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19368929ecfaSYan, Zheng int ret = 0; 19374a9d8bdeSMiao Xie 1938a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19394a9d8bdeSMiao Xie trans = info->running_transaction; 19404a9d8bdeSMiao Xie if (trans) 19414a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1942a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19438929ecfaSYan, Zheng return ret; 19448929ecfaSYan, Zheng } 19458929ecfaSYan, Zheng 1946fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1947bb9c12c9SSage Weil { 19483a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1949bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1950bb9c12c9SSage Weil 1951fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1952fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1953fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1954bb9c12c9SSage Weil 1955bb9c12c9SSage Weil /* take transaction reference */ 1956bb9c12c9SSage Weil cur_trans = trans->transaction; 19579b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1958bb9c12c9SSage Weil 19593a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19606fc4e354SSage Weil 19616fc4e354SSage Weil /* 1962ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1963ae5d29d4SDavid Sterba * subsequent transaction joins 1964ae5d29d4SDavid Sterba */ 19653e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1966ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1967ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1968ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1969724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1970bb9c12c9SSage Weil } 1971bb9c12c9SSage Weil 197297cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 197349b25e05SJeff Mahoney { 197497cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 197549b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 197649b25e05SJeff Mahoney 1977b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 197849b25e05SJeff Mahoney 197966642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19807b8b92afSJosef Bacik 19810b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 198266b6135bSLiu Bo 198325d8c284SMiao Xie /* 198425d8c284SMiao Xie * If the transaction is removed from the list, it means this 198525d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 198625d8c284SMiao Xie * to call the cleanup function. 198725d8c284SMiao Xie */ 198825d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 198966b6135bSLiu Bo 19900b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19914a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19920b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1993e1489b4fSIoannis Angelakopoulos 1994e1489b4fSIoannis Angelakopoulos /* 1995e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 1996e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 1997e1489b4fSIoannis Angelakopoulos */ 1998e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 1999f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 2000f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 2001f094ac32SLiu Bo 20020b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 2003d7096fc3SJosef Bacik } 2004061dde82SFilipe Manana 2005061dde82SFilipe Manana /* 2006061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 2007061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 2008061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 2009061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 2010061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 2011061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 2012061dde82SFilipe Manana */ 2013061dde82SFilipe Manana list_del_init(&cur_trans->list); 2014061dde82SFilipe Manana 20150b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 201649b25e05SJeff Mahoney 20172ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 201849b25e05SJeff Mahoney 20190b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20200b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20210b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20220b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20234a9d8bdeSMiao Xie 2024e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20250b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2026724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2027724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 202849b25e05SJeff Mahoney 20292e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 203049b25e05SJeff Mahoney 203149b25e05SJeff Mahoney if (current->journal_info == trans) 203249b25e05SJeff Mahoney current->journal_info = NULL; 20330b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 203449b25e05SJeff Mahoney 203549b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 203649b25e05SJeff Mahoney } 203749b25e05SJeff Mahoney 2038c7cc64a9SDavid Sterba /* 2039c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2040c7cc64a9SDavid Sterba * transaction and remove them from the list 2041c7cc64a9SDavid Sterba */ 2042c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2043c7cc64a9SDavid Sterba { 2044c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 204532da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2046c7cc64a9SDavid Sterba 2047c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2048c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2049c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2050c7cc64a9SDavid Sterba } 2051c7cc64a9SDavid Sterba } 2052c7cc64a9SDavid Sterba 205388090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 205482436617SMiao Xie { 2055ce8ea7ccSJosef Bacik /* 2056a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2057ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2058ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2059ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2060ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2061ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2062ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2063a0f0cf83SFilipe Manana * 2064a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2065a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2066a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2067a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2068a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2069a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2070a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2071ce8ea7ccSJosef Bacik */ 207288090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2073a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 207482436617SMiao Xie return 0; 207582436617SMiao Xie } 207682436617SMiao Xie 207788090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 207882436617SMiao Xie { 207988090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20806374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 208182436617SMiao Xie } 208282436617SMiao Xie 208328b21c55SFilipe Manana /* 208428b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 208528b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 208628b21c55SFilipe Manana * and while holding fs_info->trans_lock. 208728b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 208828b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 208928b21c55SFilipe Manana * returns an error. 209028b21c55SFilipe Manana */ 209128b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 209228b21c55SFilipe Manana { 209328b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 209428b21c55SFilipe Manana 209528b21c55SFilipe Manana if (!trans->pending_snapshot) 209628b21c55SFilipe Manana return; 209728b21c55SFilipe Manana 209828b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 209928b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 210028b21c55SFilipe Manana 210128b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 210228b21c55SFilipe Manana } 210328b21c55SFilipe Manana 2104e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2105e55958c8SIoannis Angelakopoulos { 2106e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2107e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2108e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2109e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2110e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2111e55958c8SIoannis Angelakopoulos } 2112e55958c8SIoannis Angelakopoulos 21133a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 211479154b1bSChris Mason { 21153a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 211649b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 21178fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 211825287e0aSMiao Xie int ret; 2119e55958c8SIoannis Angelakopoulos ktime_t start_time; 2120e55958c8SIoannis Angelakopoulos ktime_t interval; 212179154b1bSChris Mason 212235b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21233e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 212435b814f3SNikolay Borisov 2125c52cc7b7SJosef Bacik clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); 2126c52cc7b7SJosef Bacik 21278d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2128bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 212925287e0aSMiao Xie ret = cur_trans->aborted; 21303e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 213125287e0aSMiao Xie } 213249b25e05SJeff Mahoney 2133f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2134f45c752bSJosef Bacik trans->block_rsv = NULL; 2135f45c752bSJosef Bacik 2136e19eb11fSJosef Bacik /* 2137e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2138e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2139e19eb11fSJosef Bacik */ 2140e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2141e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2142e19eb11fSJosef Bacik /* 2143e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2144e19eb11fSJosef Bacik * Any running threads may add more while we are here. 214556bec294SChris Mason */ 2146c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21473e738c53SIoannis Angelakopoulos if (ret) 21483e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2149e19eb11fSJosef Bacik } 215056bec294SChris Mason 21516c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2152ea658badSJosef Bacik 21533204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21541bbc621eSChris Mason int run_it = 0; 21551bbc621eSChris Mason 21561bbc621eSChris Mason /* this mutex is also taken before trying to set 21571bbc621eSChris Mason * block groups readonly. We need to make sure 21581bbc621eSChris Mason * that nobody has set a block group readonly 21591bbc621eSChris Mason * after a extents from that block group have been 21601bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21611bbc621eSChris Mason * will wait for the transaction to commit if it 21623204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21631bbc621eSChris Mason * 21643204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21653204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21661bbc621eSChris Mason * hurt to have more than one go through, but there's no 21671bbc621eSChris Mason * real advantage to it either. 21681bbc621eSChris Mason */ 21690b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21703204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21713204d33cSJosef Bacik &cur_trans->flags)) 21721bbc621eSChris Mason run_it = 1; 21730b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21741bbc621eSChris Mason 2175f9cacae3SNikolay Borisov if (run_it) { 217621217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21773e738c53SIoannis Angelakopoulos if (ret) 21783e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2179f9cacae3SNikolay Borisov } 2180f9cacae3SNikolay Borisov } 21811bbc621eSChris Mason 21820b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21834a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2184d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2185d0c2f4faSFilipe Manana 218628b21c55SFilipe Manana add_pending_snapshot(trans); 218728b21c55SFilipe Manana 21880b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21899b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2190ccd467d6SChris Mason 2191d0c2f4faSFilipe Manana if (trans->in_fsync) 2192d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 21933e738c53SIoannis Angelakopoulos 21943e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 21953e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2196d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2197d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 219815ee9bc7SJosef Bacik 2199bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2200b4924a0fSLiu Bo ret = cur_trans->aborted; 2201b4924a0fSLiu Bo 2202724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 220315ee9bc7SJosef Bacik 220449b25e05SJeff Mahoney return ret; 220579154b1bSChris Mason } 22064313b399SChris Mason 22074a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 22080b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 22093e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2210bb9c12c9SSage Weil 22110b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2212d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2213d0c2f4faSFilipe Manana 2214d0c2f4faSFilipe Manana if (trans->in_fsync) 2215d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2216d0c2f4faSFilipe Manana 2217ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2218ccd467d6SChris Mason struct btrfs_transaction, list); 2219d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 22209b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 22210b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2222ccd467d6SChris Mason 2223d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2224d0c2f4faSFilipe Manana 2225bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2226ccd467d6SChris Mason 2227724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22281f9b8c8fSFilipe Manana if (ret) 2229e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2230a4abeea4SJosef Bacik } else { 22310b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2232ccd467d6SChris Mason } 2233a4abeea4SJosef Bacik } else { 22340b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2235cb2d3dadSFilipe Manana /* 2236cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2237cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2238cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2239cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2240cb2d3dadSFilipe Manana */ 224184961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2242cb2d3dadSFilipe Manana ret = -EROFS; 2243e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2244cb2d3dadSFilipe Manana } 2245ccd467d6SChris Mason } 224615ee9bc7SJosef Bacik 2247e55958c8SIoannis Angelakopoulos /* 2248e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2249e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2250e55958c8SIoannis Angelakopoulos */ 2251e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2252e55958c8SIoannis Angelakopoulos 22530860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22540860adfdSMiao Xie 225588090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 225682436617SMiao Xie if (ret) 2257e1489b4fSIoannis Angelakopoulos goto lockdep_release; 225882436617SMiao Xie 2259e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 226049b25e05SJeff Mahoney if (ret) 2261e1489b4fSIoannis Angelakopoulos goto lockdep_release; 226216cdcec7SMiao Xie 22635a9ba670SIoannis Angelakopoulos /* 22645a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22655a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22665a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22675a9ba670SIoannis Angelakopoulos */ 22685a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 22695a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2270581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2271581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2272ed3b3d31SChris Mason 2273581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2274e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2275e1489b4fSIoannis Angelakopoulos if (ret) { 2276e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2277ca469637SMiao Xie goto cleanup_transaction; 2278e1489b4fSIoannis Angelakopoulos } 2279ca469637SMiao Xie 228088090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2281cb7ab021SWang Shilong 228248778179SFilipe Manana /* 228348778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 228448778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 228548778179SFilipe Manana * extents complete we lose logged data after a power failure. 228648778179SFilipe Manana */ 22878b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 228848778179SFilipe Manana wait_event(cur_trans->pending_wait, 228948778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 229048778179SFilipe Manana 22912ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2292ed0ca140SJosef Bacik /* 2293ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2294ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22954a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2296ed0ca140SJosef Bacik */ 22970b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 229828b21c55SFilipe Manana add_pending_snapshot(trans); 22994a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 23000b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2301e1489b4fSIoannis Angelakopoulos 2302e1489b4fSIoannis Angelakopoulos /* 2303e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2304e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2305e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2306e1489b4fSIoannis Angelakopoulos */ 2307e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2308e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2309ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2310ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 231115ee9bc7SJosef Bacik 2312fdfbf020SJosef Bacik /* 23133e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 23143e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 23153e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 23163e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 23173e738c53SIoannis Angelakopoulos */ 23183e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 23193e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 23203e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23213e738c53SIoannis Angelakopoulos 23223e738c53SIoannis Angelakopoulos /* 2323fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2324fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2325fdfbf020SJosef Bacik * kthread could do the work. 2326fdfbf020SJosef Bacik */ 2327fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2328fdfbf020SJosef Bacik 2329bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23302cba30f1SMiao Xie ret = cur_trans->aborted; 23313e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23326cf7f77eSWang Shilong goto scrub_continue; 23332cba30f1SMiao Xie } 23347585717fSChris Mason /* 23357585717fSChris Mason * the reloc mutex makes sure that we stop 23367585717fSChris Mason * the balancing code from coming in and moving 23377585717fSChris Mason * extents around in the middle of the commit 23387585717fSChris Mason */ 23390b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23407585717fSChris Mason 234142874b3dSMiao Xie /* 234242874b3dSMiao Xie * We needn't worry about the delayed items because we will 234342874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 234442874b3dSMiao Xie * core function of the snapshot creation. 234542874b3dSMiao Xie */ 234608d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 234756e9f6eaSDavid Sterba if (ret) 234856e9f6eaSDavid Sterba goto unlock_reloc; 23493063d29fSChris Mason 235042874b3dSMiao Xie /* 235142874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 235242874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 235342874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 235442874b3dSMiao Xie * them. 235542874b3dSMiao Xie * 235642874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 235742874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 235842874b3dSMiao Xie * the nodes and leaves. 235942874b3dSMiao Xie */ 2360e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 236156e9f6eaSDavid Sterba if (ret) 236256e9f6eaSDavid Sterba goto unlock_reloc; 236316cdcec7SMiao Xie 2364c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 236556e9f6eaSDavid Sterba if (ret) 236656e9f6eaSDavid Sterba goto unlock_reloc; 236756bec294SChris Mason 2368e999376fSChris Mason /* 2369e999376fSChris Mason * make sure none of the code above managed to slip in a 2370e999376fSChris Mason * delayed item 2371e999376fSChris Mason */ 2372ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2373e999376fSChris Mason 23742c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2375dc17ff8fSChris Mason 23767e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 237756e9f6eaSDavid Sterba if (ret) 2378dfba78dcSFilipe Manana goto unlock_reloc; 237954aa1f4dSChris Mason 23805d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2381e02119d5SChris Mason * safe to free the root of tree log roots 2382e02119d5SChris Mason */ 23830b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2384e02119d5SChris Mason 23850ed4792aSQu Wenruo /* 23860ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 23870ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 23880ed4792aSQu Wenruo */ 2389460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 239056e9f6eaSDavid Sterba if (ret < 0) 2391dfba78dcSFilipe Manana goto unlock_reloc; 23920ed4792aSQu Wenruo 23939386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 239456e9f6eaSDavid Sterba if (ret) 2395dfba78dcSFilipe Manana goto unlock_reloc; 239654aa1f4dSChris Mason 23972cba30f1SMiao Xie /* 23982cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 23992cba30f1SMiao Xie * update ->aborted, check it. 24002cba30f1SMiao Xie */ 2401bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 24022cba30f1SMiao Xie ret = cur_trans->aborted; 2403dfba78dcSFilipe Manana goto unlock_reloc; 24042cba30f1SMiao Xie } 24052cba30f1SMiao Xie 24060b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 24075f39d397SChris Mason 24080b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 24090b246afaSJeff Mahoney fs_info->tree_root->node); 24100b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 24119e351cc8SJosef Bacik &cur_trans->switch_commits); 24125d4f98a2SYan Zheng 24130b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 24140b246afaSJeff Mahoney fs_info->chunk_root->node); 24150b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 24169e351cc8SJosef Bacik &cur_trans->switch_commits); 24179e351cc8SJosef Bacik 2418f7238e50SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2419f7238e50SJosef Bacik btrfs_set_root_node(&fs_info->block_group_root->root_item, 2420f7238e50SJosef Bacik fs_info->block_group_root->node); 2421f7238e50SJosef Bacik list_add_tail(&fs_info->block_group_root->dirty_list, 2422f7238e50SJosef Bacik &cur_trans->switch_commits); 2423f7238e50SJosef Bacik } 2424f7238e50SJosef Bacik 2425889bfa39SJosef Bacik switch_commit_roots(trans); 24265d4f98a2SYan Zheng 2427ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 24281bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 24292ff7e61eSJeff Mahoney update_super_roots(fs_info); 2430e02119d5SChris Mason 24310b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 24320b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 24330b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 24340b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2435ccd467d6SChris Mason 2436bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2437935e5cc9SMiao Xie 24380b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 24390b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2440656f30dbSFilipe Manana 24414fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24424fbcdf66SFilipe Manana 2443dfba78dcSFilipe Manana /* 2444dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2445dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2446dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2447dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2448dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2449dfba78dcSFilipe Manana * writing its superblock. 2450dfba78dcSFilipe Manana */ 2451dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2452dfba78dcSFilipe Manana 24530b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 24544a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 24550b246afaSJeff Mahoney fs_info->running_transaction = NULL; 24560b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 24570b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2458b7ec40d7SChris Mason 24590b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 24603e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2461e6dcd2dcSChris Mason 246270458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 246349b25e05SJeff Mahoney if (ret) { 24640b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 246508748810SDavid Sterba "Error while writing out transaction"); 24660b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 24676cf7f77eSWang Shilong goto scrub_continue; 246849b25e05SJeff Mahoney } 246949b25e05SJeff Mahoney 2470d3575156SNaohiro Aota /* 2471d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2472d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2473d3575156SNaohiro Aota * buffers. 2474d3575156SNaohiro Aota */ 2475d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2476d3575156SNaohiro Aota 2477eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2478e02119d5SChris Mason /* 2479e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2480e02119d5SChris Mason * to go about their business 2481e02119d5SChris Mason */ 24820b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2483c1f32b7cSAnand Jain if (ret) 2484c1f32b7cSAnand Jain goto scrub_continue; 2485e02119d5SChris Mason 2486d0c2f4faSFilipe Manana /* 2487d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2488d0c2f4faSFilipe Manana * which can change it. 2489d0c2f4faSFilipe Manana */ 2490d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2491d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 24923e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2493d0c2f4faSFilipe Manana 24945ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 24954313b399SChris Mason 24963204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 24970b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 249813212b54SZhao Lei 24990b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 25004a9d8bdeSMiao Xie /* 25014a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 25024a9d8bdeSMiao Xie * which can change it. 25034a9d8bdeSMiao Xie */ 25044a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 25052c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 25063e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25073de4586cSChris Mason 25080b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 250913c5a93eSJosef Bacik list_del_init(&cur_trans->list); 25100b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2511a4abeea4SJosef Bacik 2512724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2513724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 251458176a96SJosef Bacik 25150860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 25160b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2517b2b5ef5cSJan Kara 25182e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 25191abe9b8aSliubo 2520e55958c8SIoannis Angelakopoulos interval = ktime_get_ns() - start_time; 2521e55958c8SIoannis Angelakopoulos 25222ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2523a2de733cSArne Jansen 25249ed74f2dSJosef Bacik if (current->journal_info == trans) 25259ed74f2dSJosef Bacik current->journal_info = NULL; 25269ed74f2dSJosef Bacik 25272c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 252824bbcf04SYan, Zheng 2529e55958c8SIoannis Angelakopoulos update_commit_stats(fs_info, interval); 2530e55958c8SIoannis Angelakopoulos 253179154b1bSChris Mason return ret; 253249b25e05SJeff Mahoney 253356e9f6eaSDavid Sterba unlock_reloc: 253456e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 25353e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 25366cf7f77eSWang Shilong scrub_continue: 25373e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 25383e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25392ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 254049b25e05SJeff Mahoney cleanup_transaction: 2541dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2542c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 25434fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 25440e721106SJosef Bacik trans->block_rsv = NULL; 25450b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 254649b25e05SJeff Mahoney if (current->journal_info == trans) 254749b25e05SJeff Mahoney current->journal_info = NULL; 254897cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 254949b25e05SJeff Mahoney 255049b25e05SJeff Mahoney return ret; 2551e1489b4fSIoannis Angelakopoulos 2552e1489b4fSIoannis Angelakopoulos lockdep_release: 25535a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2554e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2555e1489b4fSIoannis Angelakopoulos goto cleanup_transaction; 25563e738c53SIoannis Angelakopoulos 25573e738c53SIoannis Angelakopoulos lockdep_trans_commit_start_release: 25583e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 25593e738c53SIoannis Angelakopoulos btrfs_end_transaction(trans); 25603e738c53SIoannis Angelakopoulos return ret; 256179154b1bSChris Mason } 256279154b1bSChris Mason 2563d352ac68SChris Mason /* 25649d1a2a3aSDavid Sterba * return < 0 if error 25659d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 25669d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 25679d1a2a3aSDavid Sterba * 25689d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 25699d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 25709d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 25719d1a2a3aSDavid Sterba * few seconds later. 2572d352ac68SChris Mason */ 257333c44184SJosef Bacik int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2574e9d0b13bSChris Mason { 257533c44184SJosef Bacik struct btrfs_root *root; 25769d1a2a3aSDavid Sterba int ret; 2577e9d0b13bSChris Mason 2578a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 25799d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 25809d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 25819d1a2a3aSDavid Sterba return 0; 25829d1a2a3aSDavid Sterba } 25839d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 25849d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2585cfad392bSJosef Bacik list_del_init(&root->root_list); 2586a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 25875d4f98a2SYan Zheng 25884fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 258976dda93cSYan, Zheng 259016cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 259116cdcec7SMiao Xie 259276dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 259376dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 25940078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 259576dda93cSYan, Zheng else 25960078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 259732471dc2SDavid Sterba 2598dc9492c1SJosef Bacik btrfs_put_root(root); 25996596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2600e9d0b13bSChris Mason } 2601572d9ab7SDavid Sterba 2602956504a3SJosef Bacik int __init btrfs_transaction_init(void) 2603956504a3SJosef Bacik { 2604956504a3SJosef Bacik btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2605956504a3SJosef Bacik sizeof(struct btrfs_trans_handle), 0, 2606956504a3SJosef Bacik SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2607956504a3SJosef Bacik if (!btrfs_trans_handle_cachep) 2608956504a3SJosef Bacik return -ENOMEM; 2609956504a3SJosef Bacik return 0; 2610956504a3SJosef Bacik } 2611956504a3SJosef Bacik 2612956504a3SJosef Bacik void __cold btrfs_transaction_exit(void) 2613956504a3SJosef Bacik { 2614956504a3SJosef Bacik kmem_cache_destroy(btrfs_trans_handle_cachep); 2615956504a3SJosef Bacik } 2616