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" 3159b818e0SJosef Bacik #include "defrag.h" 32f2b39277SJosef Bacik #include "dir-item.h" 33*c7a03b52SJosef Bacik #include "uuid-tree.h" 3479154b1bSChris Mason 35956504a3SJosef Bacik static struct kmem_cache *btrfs_trans_handle_cachep; 36956504a3SJosef Bacik 37fc7cbcd4SDavid Sterba #define BTRFS_ROOT_TRANS_TAG 0 380f7d52f4SChris Mason 3961c047b5SQu Wenruo /* 4061c047b5SQu Wenruo * Transaction states and transitions 4161c047b5SQu Wenruo * 4261c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 4361c047b5SQu Wenruo * | 4461c047b5SQu Wenruo * | To next stage: 4561c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 4661c047b5SQu Wenruo * V 4761c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 4861c047b5SQu Wenruo * | 4961c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 5061c047b5SQu Wenruo * | start_transaction() variants. 5161c047b5SQu Wenruo * | 5261c047b5SQu Wenruo * | To next stage: 5361c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 5461c047b5SQu Wenruo * | transaction N 5561c047b5SQu Wenruo * V 5661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 5761c047b5SQu Wenruo * | 5861c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 5961c047b5SQu Wenruo * | is one 6061c047b5SQu Wenruo * | 6161c047b5SQu Wenruo * | Then one of the following happes: 6261c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 6361c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 6461c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 6561c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 6661c047b5SQu Wenruo * | 6761c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 6861c047b5SQu Wenruo * | to this running transaction. 6961c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 7061c047b5SQu Wenruo * | transaction N+1. 7161c047b5SQu Wenruo * | 7261c047b5SQu Wenruo * | To next stage: 7361c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 7461c047b5SQu Wenruo * | haven been released. 7561c047b5SQu Wenruo * V 7661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 7761c047b5SQu Wenruo * | 7861c047b5SQu Wenruo * | The heavy lifting transaction work is started. 7961c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 8061c047b5SQu Wenruo * | snapshots, running qgroups. 8161c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 8261c047b5SQu Wenruo * | trees. 8361c047b5SQu Wenruo * | 8461c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 8561c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 8661c047b5SQu Wenruo * | 8761c047b5SQu Wenruo * | To next stage: 8861c047b5SQu Wenruo * | Until all supporting trees are updated. 8961c047b5SQu Wenruo * V 9061c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 9161c047b5SQu Wenruo * | Transaction N+1 9261c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 9361c047b5SQu Wenruo * | need to write them back to disk and update | 9461c047b5SQu Wenruo * | super blocks. | 9561c047b5SQu Wenruo * | | 9661c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 9761c047b5SQu Wenruo * | start. | 9861c047b5SQu Wenruo * | All new start_transaction() calls will be | 9961c047b5SQu Wenruo * | attached to transid N+1. | 10061c047b5SQu Wenruo * | | 10161c047b5SQu Wenruo * | To next stage: | 10261c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 10361c047b5SQu Wenruo * | written to block devices | 10461c047b5SQu Wenruo * V | 10561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 10661c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 10761c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 10861c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 10961c047b5SQu Wenruo */ 110e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1114a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 112bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 113bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1144a9d8bdeSMiao Xie __TRANS_ATTACH | 115a6d155d2SFilipe Manana __TRANS_JOIN | 116a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 117bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1184a9d8bdeSMiao Xie __TRANS_ATTACH | 1194a9d8bdeSMiao Xie __TRANS_JOIN | 120a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 121a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 122d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 123d0c2f4faSFilipe Manana __TRANS_ATTACH | 124d0c2f4faSFilipe Manana __TRANS_JOIN | 125d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 126d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 127bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1284a9d8bdeSMiao Xie __TRANS_ATTACH | 1294a9d8bdeSMiao Xie __TRANS_JOIN | 130a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 131a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1324a9d8bdeSMiao Xie }; 1334a9d8bdeSMiao Xie 134724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 13579154b1bSChris Mason { 1369b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1379b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 138a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1395c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1405c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 14181f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 14281f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1431262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 144ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 145ab8d0fc4SJeff Mahoney "pending csums is %llu", 1461262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1477785a663SFilipe Manana /* 1487785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1497785a663SFilipe Manana * because the transaction was aborted and a commit did not 1507785a663SFilipe Manana * happen (things failed before writing the new superblock 1517785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1527785a663SFilipe Manana * discard the physical locations of the block groups. 1537785a663SFilipe Manana */ 1547785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 15532da5386SDavid Sterba struct btrfs_block_group *cache; 1567785a663SFilipe Manana 1577785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 15832da5386SDavid Sterba struct btrfs_block_group, 1597785a663SFilipe Manana bg_list); 1607785a663SFilipe Manana list_del_init(&cache->bg_list); 1616b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1627785a663SFilipe Manana btrfs_put_block_group(cache); 1637785a663SFilipe Manana } 164bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1654b5faeacSDavid Sterba kfree(transaction); 16679154b1bSChris Mason } 16778fae27eSChris Mason } 16879154b1bSChris Mason 169889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 170817d52f8SJosef Bacik { 171889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 17216916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1739e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 1749e351cc8SJosef Bacik 175dfba78dcSFilipe Manana /* 176dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 177dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 178dfba78dcSFilipe Manana */ 179dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 180dfba78dcSFilipe Manana 1819e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 182d96b3424SFilipe Manana 183d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 184d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 185d96b3424SFilipe Manana 186889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1879e351cc8SJosef Bacik dirty_list) { 1889e351cc8SJosef Bacik list_del_init(&root->dirty_list); 189817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 190817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 19141e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 192370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1939e351cc8SJosef Bacik } 1942b9dbef2SJosef Bacik 1952b9dbef2SJosef Bacik /* We can free old roots now. */ 196889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 197889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 198889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1992b9dbef2SJosef Bacik struct btrfs_root, root_list); 2002b9dbef2SJosef Bacik list_del_init(&root->root_list); 201889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 202889bfa39SJosef Bacik btrfs_free_log(trans, root); 2032b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 204889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 2052b9dbef2SJosef Bacik } 206889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 20727d56e62SJosef Bacik 2089e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 209817d52f8SJosef Bacik } 210817d52f8SJosef Bacik 2110860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2120860adfdSMiao Xie unsigned int type) 2130860adfdSMiao Xie { 2140860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2150860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2160860adfdSMiao Xie } 2170860adfdSMiao Xie 2180860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2190860adfdSMiao Xie unsigned int type) 2200860adfdSMiao Xie { 2210860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2220860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2230860adfdSMiao Xie } 2240860adfdSMiao Xie 2250860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2260860adfdSMiao Xie unsigned int type) 2270860adfdSMiao Xie { 2280860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2290860adfdSMiao Xie } 2300860adfdSMiao Xie 2310860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2320860adfdSMiao Xie { 2330860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 234178260b2SMiao Xie } 235178260b2SMiao Xie 236d352ac68SChris Mason /* 23779bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 23879bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 23979bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 24079bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 24179bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 242fb6dea26SJosef Bacik */ 243fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 244fb6dea26SJosef Bacik { 245fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 246fb6dea26SJosef Bacik 247fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 248fb6dea26SJosef Bacik return; 249fb6dea26SJosef Bacik 250fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 25163f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 252fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 253fb6dea26SJosef Bacik } 254fb6dea26SJosef Bacik 255fb6dea26SJosef Bacik /* 256d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 257d352ac68SChris Mason */ 2582ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2592ff7e61eSJeff Mahoney unsigned int type) 26079154b1bSChris Mason { 26179154b1bSChris Mason struct btrfs_transaction *cur_trans; 262a4abeea4SJosef Bacik 26319ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 264d43317dcSChris Mason loop: 26549b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 26684961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 26719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26849b25e05SJeff Mahoney return -EROFS; 26949b25e05SJeff Mahoney } 27049b25e05SJeff Mahoney 27119ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 272a4abeea4SJosef Bacik if (cur_trans) { 273bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 27419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 27549b25e05SJeff Mahoney return cur_trans->aborted; 276871383beSDavid Sterba } 2774a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 278178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 279178260b2SMiao Xie return -EBUSY; 280178260b2SMiao Xie } 2819b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 282a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2830860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 28419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 285e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2865a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 287a4abeea4SJosef Bacik return 0; 288a4abeea4SJosef Bacik } 28919ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 290a4abeea4SJosef Bacik 291354aa0fbSMiao Xie /* 292354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 293354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 294354aa0fbSMiao Xie */ 295354aa0fbSMiao Xie if (type == TRANS_ATTACH) 296354aa0fbSMiao Xie return -ENOENT; 297354aa0fbSMiao Xie 2984a9d8bdeSMiao Xie /* 2994a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3004a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3014a9d8bdeSMiao Xie */ 3024a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3034a9d8bdeSMiao Xie 3044b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 305db5b493aSTsutomu Itoh if (!cur_trans) 306db5b493aSTsutomu Itoh return -ENOMEM; 307d43317dcSChris Mason 308e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 3095a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 310e1489b4fSIoannis Angelakopoulos 31119ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 31219ae4e81SJan Schmidt if (fs_info->running_transaction) { 313d43317dcSChris Mason /* 314d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3154a9d8bdeSMiao Xie * to redo the checks above 316d43317dcSChris Mason */ 3175a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 318e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3194b5faeacSDavid Sterba kfree(cur_trans); 320d43317dcSChris Mason goto loop; 32184961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 322e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3235a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 324e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3254b5faeacSDavid Sterba kfree(cur_trans); 3267b8b92afSJosef Bacik return -EROFS; 327a4abeea4SJosef Bacik } 328d43317dcSChris Mason 329ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 33048778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 33148778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 33213c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3330860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 33479154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 33579154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3364a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 337a4abeea4SJosef Bacik /* 338a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 339a4abeea4SJosef Bacik * commit the transaction. 340a4abeea4SJosef Bacik */ 3419b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3423204d33cSJosef Bacik cur_trans->flags = 0; 343afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 34456bec294SChris Mason 345a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 346a099d0fdSAlexandru Moise 3475c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3483368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 349d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 35020b297d6SJan Schmidt 35120b297d6SJan Schmidt /* 35220b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 35320b297d6SJan Schmidt * the log must never go across transaction boundaries. 35420b297d6SJan Schmidt */ 35520b297d6SJan Schmidt smp_mb(); 35631b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3575d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 35831b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3595d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 360fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 36120b297d6SJan Schmidt 36256bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 36356bec294SChris Mason 3643063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 365bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3669e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 367ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3681bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3692b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3701bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 371ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 372e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3732b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 374d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 375d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 37619ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 377c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 378efb0645bSJosef Bacik IO_TREE_TRANS_DIRTY_PAGES, NULL); 379fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 380fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 38119ae4e81SJan Schmidt fs_info->generation++; 38219ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 38319ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 38449b25e05SJeff Mahoney cur_trans->aborted = 0; 38519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 38615ee9bc7SJosef Bacik 38779154b1bSChris Mason return 0; 38879154b1bSChris Mason } 38979154b1bSChris Mason 390d352ac68SChris Mason /* 39192a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 39292a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 39392a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 39492a7cc42SQu Wenruo * transaction commits. 395d352ac68SChris Mason */ 3967585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3976426c7adSQu Wenruo struct btrfs_root *root, 3986426c7adSQu Wenruo int force) 3996702ed49SChris Mason { 4000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 40103a7e111SJosef Bacik int ret = 0; 4020b246afaSJeff Mahoney 40392a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4046426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4054d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4065d4f98a2SYan Zheng 4077585717fSChris Mason /* 40827cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4097585717fSChris Mason * we have the reloc mutex held now, so there 4107585717fSChris Mason * is only one writer in this function 4117585717fSChris Mason */ 41227cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4137585717fSChris Mason 41427cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4157585717fSChris Mason * they find our root->last_trans update 4167585717fSChris Mason */ 4177585717fSChris Mason smp_wmb(); 4187585717fSChris Mason 419fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4206426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 421fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 422a4abeea4SJosef Bacik return 0; 423a4abeea4SJosef Bacik } 424fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4256702ed49SChris Mason (unsigned long)root->root_key.objectid, 4266702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 427fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4287585717fSChris Mason root->last_trans = trans->transid; 4297585717fSChris Mason 4307585717fSChris Mason /* this is pretty tricky. We don't want to 4317585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4327585717fSChris Mason * unless we're really doing the first setup for this root in 4337585717fSChris Mason * this transaction. 4347585717fSChris Mason * 4357585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4367585717fSChris Mason * if we want to take the expensive mutex. 4377585717fSChris Mason * 4387585717fSChris Mason * But, we have to set root->last_trans before we 4397585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4407585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 44127cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4427585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4437585717fSChris Mason * 4447585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4457585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4467585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4477585717fSChris Mason * done before we pop in the zero below 4487585717fSChris Mason */ 44903a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 450c7548af6SChris Mason smp_mb__before_atomic(); 45127cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4526702ed49SChris Mason } 45303a7e111SJosef Bacik return ret; 4546702ed49SChris Mason } 4555d4f98a2SYan Zheng 4567585717fSChris Mason 4572b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4582b9dbef2SJosef Bacik struct btrfs_root *root) 4592b9dbef2SJosef Bacik { 4600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4612b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4622b9dbef2SJosef Bacik 4632b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4642b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4652b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4662b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4672b9dbef2SJosef Bacik 4682b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 469fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 470fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4712b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4722b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 473fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4742b9dbef2SJosef Bacik } 4752b9dbef2SJosef Bacik 4767585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4777585717fSChris Mason struct btrfs_root *root) 4787585717fSChris Mason { 4790b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4801409e6ccSJosef Bacik int ret; 4810b246afaSJeff Mahoney 48292a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4837585717fSChris Mason return 0; 4847585717fSChris Mason 4857585717fSChris Mason /* 48627cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4877585717fSChris Mason * and barriers 4887585717fSChris Mason */ 4897585717fSChris Mason smp_rmb(); 4907585717fSChris Mason if (root->last_trans == trans->transid && 49127cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4927585717fSChris Mason return 0; 4937585717fSChris Mason 4940b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4951409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4960b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4977585717fSChris Mason 4981409e6ccSJosef Bacik return ret; 4997585717fSChris Mason } 5007585717fSChris Mason 5014a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5024a9d8bdeSMiao Xie { 5033296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 504501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 505bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5064a9d8bdeSMiao Xie } 5074a9d8bdeSMiao Xie 508d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 509d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 510d352ac68SChris Mason * transaction might not be fully on disk. 511d352ac68SChris Mason */ 5122ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 51379154b1bSChris Mason { 514f9295749SChris Mason struct btrfs_transaction *cur_trans; 51579154b1bSChris Mason 5160b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5170b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5184a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5199b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5200b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52172d63ed6SLi Zefan 5223e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5230b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 524501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 525bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 526724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 527a4abeea4SJosef Bacik } else { 5280b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 529f9295749SChris Mason } 53037d1aeeeSChris Mason } 53137d1aeeeSChris Mason 5322ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 53337d1aeeeSChris Mason { 5340b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 535a4abeea4SJosef Bacik return 0; 536a4abeea4SJosef Bacik 53792e2f7e3SNikolay Borisov if (type == TRANS_START) 538a4abeea4SJosef Bacik return 1; 539a4abeea4SJosef Bacik 540a22285a6SYan, Zheng return 0; 541a22285a6SYan, Zheng } 542a22285a6SYan, Zheng 54320dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 54420dd2cbfSMiao Xie { 5450b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5460b246afaSJeff Mahoney 5470b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 54892a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 54920dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 55020dd2cbfSMiao Xie root->reloc_root) 55120dd2cbfSMiao Xie return false; 55220dd2cbfSMiao Xie 55320dd2cbfSMiao Xie return true; 55420dd2cbfSMiao Xie } 55520dd2cbfSMiao Xie 55608e007d2SMiao Xie static struct btrfs_trans_handle * 5575aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 558003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 559003d7c59SJeff Mahoney bool enforce_qgroups) 560a22285a6SYan, Zheng { 5610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 562ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 563a22285a6SYan, Zheng struct btrfs_trans_handle *h; 564a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 565b5009945SJosef Bacik u64 num_bytes = 0; 566c5567237SArne Jansen u64 qgroup_reserved = 0; 56720dd2cbfSMiao Xie bool reloc_reserved = false; 5689c343784SJosef Bacik bool do_chunk_alloc = false; 56920dd2cbfSMiao Xie int ret; 570acce952bSliubo 57184961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 572acce952bSliubo return ERR_PTR(-EROFS); 5732a1eb461SJosef Bacik 57446c4e71eSFilipe Manana if (current->journal_info) { 5750860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5762a1eb461SJosef Bacik h = current->journal_info; 577b50fff81SDavid Sterba refcount_inc(&h->use_count); 578b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5792a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5802a1eb461SJosef Bacik h->block_rsv = NULL; 5812a1eb461SJosef Bacik goto got_it; 5822a1eb461SJosef Bacik } 583b5009945SJosef Bacik 584b5009945SJosef Bacik /* 585b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 586b5009945SJosef Bacik * the appropriate flushing if need be. 587b5009945SJosef Bacik */ 588003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 589ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 590ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 591ba2c4d4eSJosef Bacik 5920b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 593733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 594003d7c59SJeff Mahoney enforce_qgroups); 595c5567237SArne Jansen if (ret) 596c5567237SArne Jansen return ERR_PTR(ret); 597c5567237SArne Jansen 598ba2c4d4eSJosef Bacik /* 599ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 600ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 601ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 602ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 603ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 604ba2c4d4eSJosef Bacik */ 6052bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6067f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 607748f553cSDavid Sterba btrfs_block_rsv_full(delayed_refs_rsv) == 0) { 608ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 609ba2c4d4eSJosef Bacik num_bytes <<= 1; 610ba2c4d4eSJosef Bacik } 611ba2c4d4eSJosef Bacik 61220dd2cbfSMiao Xie /* 61320dd2cbfSMiao Xie * Do the reservation for the relocation root creation 61420dd2cbfSMiao Xie */ 615ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6160b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 61720dd2cbfSMiao Xie reloc_reserved = true; 61820dd2cbfSMiao Xie } 61920dd2cbfSMiao Xie 6209270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 621ba2c4d4eSJosef Bacik if (ret) 622ba2c4d4eSJosef Bacik goto reserve_fail; 623ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 624ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 625ba2c4d4eSJosef Bacik delayed_refs_bytes); 626ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 627ba2c4d4eSJosef Bacik } 6289c343784SJosef Bacik 6299c343784SJosef Bacik if (rsv->space_info->force_alloc) 6309c343784SJosef Bacik do_chunk_alloc = true; 631ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 632748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 633ba2c4d4eSJosef Bacik /* 634ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 635ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 636ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 637ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 638ba2c4d4eSJosef Bacik * here. 639ba2c4d4eSJosef Bacik */ 640ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 641b5009945SJosef Bacik if (ret) 642843fcf35SMiao Xie goto reserve_fail; 643b5009945SJosef Bacik } 644a22285a6SYan, Zheng again: 645f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 646843fcf35SMiao Xie if (!h) { 647843fcf35SMiao Xie ret = -ENOMEM; 648843fcf35SMiao Xie goto alloc_fail; 649843fcf35SMiao Xie } 650a22285a6SYan, Zheng 65198114659SJosef Bacik /* 65298114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 65398114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 65498114659SJosef Bacik * because we're already holding a ref. We need this because we could 65598114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 65698114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 657354aa0fbSMiao Xie * 658354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 659354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 66098114659SJosef Bacik */ 6610860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6620b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 663b2b5ef5cSJan Kara 6642ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6652ff7e61eSJeff Mahoney wait_current_trans(fs_info); 666a22285a6SYan, Zheng 667a4abeea4SJosef Bacik do { 6682ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 669178260b2SMiao Xie if (ret == -EBUSY) { 6702ff7e61eSJeff Mahoney wait_current_trans(fs_info); 671a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 672a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 673178260b2SMiao Xie ret = -ENOENT; 674178260b2SMiao Xie } 675a4abeea4SJosef Bacik } while (ret == -EBUSY); 676a4abeea4SJosef Bacik 677a43f7f82SLiu Bo if (ret < 0) 678843fcf35SMiao Xie goto join_fail; 6790f7d52f4SChris Mason 6800b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 681a22285a6SYan, Zheng 682a22285a6SYan, Zheng h->transid = cur_trans->transid; 683a22285a6SYan, Zheng h->transaction = cur_trans; 684b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 68564b63580SJeff Mahoney h->fs_info = root->fs_info; 6867174109cSQu Wenruo 687a698d075SMiao Xie h->type = type; 688ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 689b7ec40d7SChris Mason 690a22285a6SYan, Zheng smp_mb(); 6913296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 6922ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 693abdd2e80SFilipe Manana current->journal_info = h; 6943a45bb20SJeff Mahoney btrfs_commit_transaction(h); 695a22285a6SYan, Zheng goto again; 696a22285a6SYan, Zheng } 6979ed74f2dSJosef Bacik 698b5009945SJosef Bacik if (num_bytes) { 6990b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7002bcc0328SLiu Bo h->transid, num_bytes, 1); 7010b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 702b5009945SJosef Bacik h->bytes_reserved = num_bytes; 70320dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 704a22285a6SYan, Zheng } 705a22285a6SYan, Zheng 7062a1eb461SJosef Bacik got_it: 707bcf3a3e7SNikolay Borisov if (!current->journal_info) 708a22285a6SYan, Zheng current->journal_info = h; 709fcc99734SQu Wenruo 710fcc99734SQu Wenruo /* 7119c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7129c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7139c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7149c343784SJosef Bacik * value here. 7159c343784SJosef Bacik */ 7169c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7179c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7189c343784SJosef Bacik 7199c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7209c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7219c343784SJosef Bacik } 7229c343784SJosef Bacik 7239c343784SJosef Bacik /* 724fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 725fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 726fcc99734SQu Wenruo * transaction. 727fcc99734SQu Wenruo * 728fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 729fcc99734SQu Wenruo * or we can deadlock. 730fcc99734SQu Wenruo */ 73168075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 73268075ea8SJosef Bacik if (ret) { 73368075ea8SJosef Bacik /* 73468075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 73568075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 73668075ea8SJosef Bacik * not just freed. 73768075ea8SJosef Bacik */ 73868075ea8SJosef Bacik btrfs_end_transaction(h); 73968075ea8SJosef Bacik return ERR_PTR(ret); 74068075ea8SJosef Bacik } 741fcc99734SQu Wenruo 74279154b1bSChris Mason return h; 743843fcf35SMiao Xie 744843fcf35SMiao Xie join_fail: 7450860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7460b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 747843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 748843fcf35SMiao Xie alloc_fail: 749843fcf35SMiao Xie if (num_bytes) 7502ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 75163f018beSNikolay Borisov num_bytes, NULL); 752843fcf35SMiao Xie reserve_fail: 753733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 754843fcf35SMiao Xie return ERR_PTR(ret); 75579154b1bSChris Mason } 75679154b1bSChris Mason 757f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7585aed1dd8SAlexandru Moise unsigned int num_items) 759f9295749SChris Mason { 76008e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 761003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 762f9295749SChris Mason } 763003d7c59SJeff Mahoney 7648eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7658eab77ffSFilipe Manana struct btrfs_root *root, 7667f9fe614SJosef Bacik unsigned int num_items) 7678eab77ffSFilipe Manana { 7687f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7697f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7708eab77ffSFilipe Manana } 7718407aa46SMiao Xie 7727a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 773f9295749SChris Mason { 774003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 775003d7c59SJeff Mahoney true); 776f9295749SChris Mason } 777f9295749SChris Mason 7788d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7790af3d00bSJosef Bacik { 780575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 781003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7820af3d00bSJosef Bacik } 7830af3d00bSJosef Bacik 784d4edf39bSMiao Xie /* 785a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 786a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 787a6d155d2SFilipe Manana */ 788a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 789a6d155d2SFilipe Manana { 790a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 791a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 792a6d155d2SFilipe Manana } 793a6d155d2SFilipe Manana 794a6d155d2SFilipe Manana /* 795d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 796d4edf39bSMiao Xie * 797d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 798d4edf39bSMiao Xie * don't want to start a new one. 799d4edf39bSMiao Xie * 800d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 801d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 802d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 803d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 804d4edf39bSMiao Xie * invoke 805d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 806d4edf39bSMiao Xie */ 807354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 80860376ce4SJosef Bacik { 809575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 810003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 81160376ce4SJosef Bacik } 81260376ce4SJosef Bacik 813d4edf39bSMiao Xie /* 81490b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 815d4edf39bSMiao Xie * 81652042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 817d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 818d4edf39bSMiao Xie * complete. 819d4edf39bSMiao Xie */ 820d4edf39bSMiao Xie struct btrfs_trans_handle * 821d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 822d4edf39bSMiao Xie { 823d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 824d4edf39bSMiao Xie 825575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 826003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8278d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8282ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 829d4edf39bSMiao Xie 830d4edf39bSMiao Xie return trans; 831d4edf39bSMiao Xie } 832d4edf39bSMiao Xie 833d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 834d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 835d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 83689ce8a63SChris Mason { 8375fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8385fd76bf3SOmar Sandoval u64 transid = commit->transid; 8395fd76bf3SOmar Sandoval bool put = false; 8405fd76bf3SOmar Sandoval 8413e738c53SIoannis Angelakopoulos /* 8423e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8433e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8443e738c53SIoannis Angelakopoulos */ 8453e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8463e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8473e738c53SIoannis Angelakopoulos else 8483e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8493e738c53SIoannis Angelakopoulos 8505fd76bf3SOmar Sandoval while (1) { 851d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8525fd76bf3SOmar Sandoval if (put) 8535fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8545fd76bf3SOmar Sandoval 8555fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8565fd76bf3SOmar Sandoval break; 8575fd76bf3SOmar Sandoval 8585fd76bf3SOmar Sandoval /* 8595fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8605fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8615fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8625fd76bf3SOmar Sandoval * transaction. Wait for those. 8635fd76bf3SOmar Sandoval */ 8645fd76bf3SOmar Sandoval 8655fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8665fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8675fd76bf3SOmar Sandoval struct btrfs_transaction, 8685fd76bf3SOmar Sandoval list); 8695fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8705fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8715fd76bf3SOmar Sandoval break; 8725fd76bf3SOmar Sandoval } 8735fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8745fd76bf3SOmar Sandoval put = true; 8755fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8765fd76bf3SOmar Sandoval } 87789ce8a63SChris Mason } 87889ce8a63SChris Mason 8792ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 88046204592SSage Weil { 88146204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8828cd2807fSMiao Xie int ret = 0; 88346204592SSage Weil 88446204592SSage Weil if (transid) { 8850b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 886a4abeea4SJosef Bacik goto out; 88746204592SSage Weil 88846204592SSage Weil /* find specified transaction */ 8890b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8900b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 89146204592SSage Weil if (t->transid == transid) { 89246204592SSage Weil cur_trans = t; 8939b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8948cd2807fSMiao Xie ret = 0; 89546204592SSage Weil break; 89646204592SSage Weil } 8978cd2807fSMiao Xie if (t->transid > transid) { 8988cd2807fSMiao Xie ret = 0; 89946204592SSage Weil break; 90046204592SSage Weil } 9018cd2807fSMiao Xie } 9020b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 90342383020SSage Weil 90442383020SSage Weil /* 90542383020SSage Weil * The specified transaction doesn't exist, or we 90642383020SSage Weil * raced with btrfs_commit_transaction 90742383020SSage Weil */ 90842383020SSage Weil if (!cur_trans) { 9090b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 91042383020SSage Weil ret = -EINVAL; 9118cd2807fSMiao Xie goto out; 91242383020SSage Weil } 91346204592SSage Weil } else { 91446204592SSage Weil /* find newest transaction that is committing | committed */ 9150b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9160b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 91746204592SSage Weil list) { 9184a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9194a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9203473f3c0SJosef Bacik break; 92146204592SSage Weil cur_trans = t; 9229b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 92346204592SSage Weil break; 92446204592SSage Weil } 92546204592SSage Weil } 9260b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 92746204592SSage Weil if (!cur_trans) 928a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 92946204592SSage Weil } 93046204592SSage Weil 931d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 932724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 933a4abeea4SJosef Bacik out: 93446204592SSage Weil return ret; 93546204592SSage Weil } 93646204592SSage Weil 9372ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 93837d1aeeeSChris Mason { 9392ff7e61eSJeff Mahoney wait_current_trans(fs_info); 94037d1aeeeSChris Mason } 94137d1aeeeSChris Mason 9428a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9438929ecfaSYan, Zheng { 9442ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9450b246afaSJeff Mahoney 94664403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9478a8f4deaSNikolay Borisov return true; 94836ba022aSJosef Bacik 9492ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9508929ecfaSYan, Zheng } 9518929ecfaSYan, Zheng 952a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9538929ecfaSYan, Zheng { 9548929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9558929ecfaSYan, Zheng 9563296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 957e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 958a2633b6aSNikolay Borisov return true; 9598929ecfaSYan, Zheng 9602ff7e61eSJeff Mahoney return should_end_transaction(trans); 9618929ecfaSYan, Zheng } 9628929ecfaSYan, Zheng 963dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 964dc60c525SNikolay Borisov 9650e34693fSNikolay Borisov { 966dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 967dc60c525SNikolay Borisov 9680e34693fSNikolay Borisov if (!trans->block_rsv) { 9690e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9700e34693fSNikolay Borisov return; 9710e34693fSNikolay Borisov } 9720e34693fSNikolay Borisov 9730e34693fSNikolay Borisov if (!trans->bytes_reserved) 9740e34693fSNikolay Borisov return; 9750e34693fSNikolay Borisov 9760e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9770e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9780e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9790e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 98063f018beSNikolay Borisov trans->bytes_reserved, NULL); 9810e34693fSNikolay Borisov trans->bytes_reserved = 0; 9820e34693fSNikolay Borisov } 9830e34693fSNikolay Borisov 98489ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9853a45bb20SJeff Mahoney int throttle) 98679154b1bSChris Mason { 9873a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9888929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9894edc2ca3SDave Jones int err = 0; 990d6e4a428SChris Mason 991b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 992b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9932a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9942a1eb461SJosef Bacik return 0; 9952a1eb461SJosef Bacik } 9962a1eb461SJosef Bacik 997dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9984c13d758SJosef Bacik trans->block_rsv = NULL; 999c5567237SArne Jansen 10006c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 1001ea658badSJosef Bacik 10024fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 10034fbcdf66SFilipe Manana 10040860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 10050b246afaSJeff Mahoney sb_end_intwrite(info->sb); 10066df7881aSJosef Bacik 10078929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 100813c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 100913c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10100860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 101189ce8a63SChris Mason 1012093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1013e1489b4fSIoannis Angelakopoulos 10145a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1015e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1016e1489b4fSIoannis Angelakopoulos 1017724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10189ed74f2dSJosef Bacik 10199ed74f2dSJosef Bacik if (current->journal_info == trans) 10209ed74f2dSJosef Bacik current->journal_info = NULL; 1021ab78c84dSChris Mason 102224bbcf04SYan, Zheng if (throttle) 10232ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 102424bbcf04SYan, Zheng 102584961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10264e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1027fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1028fbabd4a3SJosef Bacik err = trans->aborted; 1029fbabd4a3SJosef Bacik else 1030fbabd4a3SJosef Bacik err = -EROFS; 10314e121c06SJosef Bacik } 103249b25e05SJeff Mahoney 10334edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10344edc2ca3SDave Jones return err; 103579154b1bSChris Mason } 103679154b1bSChris Mason 10373a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 103889ce8a63SChris Mason { 10393a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 104089ce8a63SChris Mason } 104189ce8a63SChris Mason 10423a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 104389ce8a63SChris Mason { 10443a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 104516cdcec7SMiao Xie } 104616cdcec7SMiao Xie 1047d352ac68SChris Mason /* 1048d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1049d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1050690587d1SChris Mason * those extents are sent to disk but does not wait on them 1051d352ac68SChris Mason */ 10522ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10538cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 105479154b1bSChris Mason { 1055777e6bd7SChris Mason int err = 0; 10567c4452b9SChris Mason int werr = 0; 10570b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1058e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1059777e6bd7SChris Mason u64 start = 0; 10605f39d397SChris Mason u64 end; 10617c4452b9SChris Mason 10626300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10631728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1064e6138876SJosef Bacik mark, &cached_state)) { 1065663dfbb0SFilipe Manana bool wait_writeback = false; 1066663dfbb0SFilipe Manana 1067663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1068663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1069210aa277SDavid Sterba mark, &cached_state); 1070663dfbb0SFilipe Manana /* 1071663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1072663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1073663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1074663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1075bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1076bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1077bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1078bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1079bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1080663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 108141e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1082663dfbb0SFilipe Manana */ 1083663dfbb0SFilipe Manana if (err == -ENOMEM) { 1084663dfbb0SFilipe Manana err = 0; 1085663dfbb0SFilipe Manana wait_writeback = true; 1086663dfbb0SFilipe Manana } 1087663dfbb0SFilipe Manana if (!err) 10881728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10897c4452b9SChris Mason if (err) 10907c4452b9SChris Mason werr = err; 1091663dfbb0SFilipe Manana else if (wait_writeback) 1092663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1093e38e2ed7SFilipe Manana free_extent_state(cached_state); 1094663dfbb0SFilipe Manana cached_state = NULL; 10951728366eSJosef Bacik cond_resched(); 10961728366eSJosef Bacik start = end + 1; 10977c4452b9SChris Mason } 10986300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1099690587d1SChris Mason return werr; 1100690587d1SChris Mason } 1101690587d1SChris Mason 1102690587d1SChris Mason /* 1103690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1104690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1105690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1106690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1107690587d1SChris Mason */ 1108bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1109bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1110690587d1SChris Mason { 1111690587d1SChris Mason int err = 0; 1112690587d1SChris Mason int werr = 0; 11130b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1114e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1115690587d1SChris Mason u64 start = 0; 1116690587d1SChris Mason u64 end; 1117690587d1SChris Mason 11181728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1119e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1120663dfbb0SFilipe Manana /* 1121663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1122663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1123663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1124663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1125663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 112641e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1127663dfbb0SFilipe Manana */ 1128663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1129bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1130663dfbb0SFilipe Manana if (err == -ENOMEM) 1131663dfbb0SFilipe Manana err = 0; 1132663dfbb0SFilipe Manana if (!err) 11331728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1134777e6bd7SChris Mason if (err) 1135777e6bd7SChris Mason werr = err; 1136e38e2ed7SFilipe Manana free_extent_state(cached_state); 1137e38e2ed7SFilipe Manana cached_state = NULL; 1138777e6bd7SChris Mason cond_resched(); 11391728366eSJosef Bacik start = end + 1; 1140777e6bd7SChris Mason } 11417c4452b9SChris Mason if (err) 11427c4452b9SChris Mason werr = err; 1143bf89d38fSJeff Mahoney return werr; 1144bf89d38fSJeff Mahoney } 1145656f30dbSFilipe Manana 1146b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1147bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1148bf89d38fSJeff Mahoney { 1149bf89d38fSJeff Mahoney bool errors = false; 1150bf89d38fSJeff Mahoney int err; 1151bf89d38fSJeff Mahoney 1152bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1153bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1154bf89d38fSJeff Mahoney errors = true; 1155bf89d38fSJeff Mahoney 1156bf89d38fSJeff Mahoney if (errors && !err) 1157bf89d38fSJeff Mahoney err = -EIO; 1158bf89d38fSJeff Mahoney return err; 1159bf89d38fSJeff Mahoney } 1160bf89d38fSJeff Mahoney 1161bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1162bf89d38fSJeff Mahoney { 1163bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1164bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1165bf89d38fSJeff Mahoney bool errors = false; 1166bf89d38fSJeff Mahoney int err; 1167bf89d38fSJeff Mahoney 1168bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1169bf89d38fSJeff Mahoney 1170bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1171656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11720b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1173656f30dbSFilipe Manana errors = true; 1174656f30dbSFilipe Manana 1175656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11760b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1177656f30dbSFilipe Manana errors = true; 1178656f30dbSFilipe Manana 1179bf89d38fSJeff Mahoney if (errors && !err) 1180bf89d38fSJeff Mahoney err = -EIO; 1181bf89d38fSJeff Mahoney return err; 118279154b1bSChris Mason } 118379154b1bSChris Mason 1184690587d1SChris Mason /* 1185c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1186c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1187c9b577c0SNikolay Borisov * log commit. 1188c9b577c0SNikolay Borisov * 1189c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1190690587d1SChris Mason */ 119170458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1192d0c803c4SChris Mason { 1193663dfbb0SFilipe Manana int ret; 1194c9b577c0SNikolay Borisov int ret2; 1195c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 119670458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1197c9b577c0SNikolay Borisov struct blk_plug plug; 1198663dfbb0SFilipe Manana 1199c9b577c0SNikolay Borisov blk_start_plug(&plug); 1200c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1201c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1202c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1203c9b577c0SNikolay Borisov 120441e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1205663dfbb0SFilipe Manana 1206c9b577c0SNikolay Borisov if (ret) 1207663dfbb0SFilipe Manana return ret; 1208c9b577c0SNikolay Borisov else if (ret2) 1209c9b577c0SNikolay Borisov return ret2; 1210c9b577c0SNikolay Borisov else 1211c9b577c0SNikolay Borisov return 0; 1212d0c803c4SChris Mason } 1213d0c803c4SChris Mason 1214d352ac68SChris Mason /* 1215d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1216d352ac68SChris Mason * 1217d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1218d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1219d352ac68SChris Mason * allocation tree. 1220d352ac68SChris Mason * 1221d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1222d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1223d352ac68SChris Mason */ 12240b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 122579154b1bSChris Mason struct btrfs_root *root) 122679154b1bSChris Mason { 122779154b1bSChris Mason int ret; 12280b86a832SChris Mason u64 old_root_bytenr; 122986b9f2ecSYan, Zheng u64 old_root_used; 12300b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12310b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 123279154b1bSChris Mason 123386b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 123456bec294SChris Mason 123579154b1bSChris Mason while (1) { 12360b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 123786b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1238ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 123979154b1bSChris Mason break; 124087ef2bb4SChris Mason 12415d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 124279154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12430b86a832SChris Mason &root->root_key, 12440b86a832SChris Mason &root->root_item); 124549b25e05SJeff Mahoney if (ret) 124649b25e05SJeff Mahoney return ret; 124756bec294SChris Mason 124886b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1249e7070be1SJosef Bacik } 1250276e680dSYan Zheng 12510b86a832SChris Mason return 0; 12520b86a832SChris Mason } 12530b86a832SChris Mason 1254d352ac68SChris Mason /* 1255d352ac68SChris Mason * update all the cowonly tree roots on disk 125649b25e05SJeff Mahoney * 125749b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 125849b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 125949b25e05SJeff Mahoney * to clean up the delayed refs. 1260d352ac68SChris Mason */ 12619386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12620b86a832SChris Mason { 12639386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1264ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12651bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12660b86a832SChris Mason struct list_head *next; 126784234f3aSYan Zheng struct extent_buffer *eb; 126856bec294SChris Mason int ret; 126984234f3aSYan Zheng 1270dfba78dcSFilipe Manana /* 1271dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1272dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1273dfba78dcSFilipe Manana */ 1274dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1275dfba78dcSFilipe Manana 127684234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 127749b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12789631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 127984234f3aSYan Zheng btrfs_tree_unlock(eb); 128084234f3aSYan Zheng free_extent_buffer(eb); 12810b86a832SChris Mason 128249b25e05SJeff Mahoney if (ret) 128349b25e05SJeff Mahoney return ret; 128449b25e05SJeff Mahoney 1285196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1286c16ce190SJosef Bacik if (ret) 1287c16ce190SJosef Bacik return ret; 12882b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1289c16ce190SJosef Bacik if (ret) 1290c16ce190SJosef Bacik return ret; 1291280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1292c16ce190SJosef Bacik if (ret) 1293c16ce190SJosef Bacik return ret; 1294546adb0dSJan Schmidt 1295bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1296dcdf7f6dSJosef Bacik if (ret) 1297dcdf7f6dSJosef Bacik return ret; 1298dcdf7f6dSJosef Bacik 1299ea526d18SJosef Bacik again: 13000b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 13012ff7e61eSJeff Mahoney struct btrfs_root *root; 13020b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 13030b86a832SChris Mason list_del_init(next); 13040b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1305e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 130687ef2bb4SChris Mason 13079e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13089e351cc8SJosef Bacik &trans->transaction->switch_commits); 130949b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 131049b25e05SJeff Mahoney if (ret) 131149b25e05SJeff Mahoney return ret; 1312488bc2a2SJosef Bacik } 1313488bc2a2SJosef Bacik 1314488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1315c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1316ea526d18SJosef Bacik if (ret) 1317ea526d18SJosef Bacik return ret; 1318276e680dSYan Zheng 13191bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13205742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1321ea526d18SJosef Bacik if (ret) 1322ea526d18SJosef Bacik return ret; 1323488bc2a2SJosef Bacik 1324488bc2a2SJosef Bacik /* 1325488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1326488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1327488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1328488bc2a2SJosef Bacik * everything gets run. 1329488bc2a2SJosef Bacik */ 1330c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1331ea526d18SJosef Bacik if (ret) 1332ea526d18SJosef Bacik return ret; 1333ea526d18SJosef Bacik } 1334ea526d18SJosef Bacik 1335ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1336ea526d18SJosef Bacik goto again; 1337ea526d18SJosef Bacik 13389f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13399f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13409f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13418dabb742SStefan Behrens 134279154b1bSChris Mason return 0; 134379154b1bSChris Mason } 134479154b1bSChris Mason 1345d352ac68SChris Mason /* 1346b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1347b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1348b4be6aefSJosef Bacik */ 1349b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1350b4be6aefSJosef Bacik { 1351b4be6aefSJosef Bacik /* 1352b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1353b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1354b4be6aefSJosef Bacik * up. 1355b4be6aefSJosef Bacik */ 1356b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1357b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1358b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1359b4be6aefSJosef Bacik struct btrfs_root, 1360b4be6aefSJosef Bacik root_list); 1361b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1362b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1363b4be6aefSJosef Bacik return; 1364b4be6aefSJosef Bacik } 1365b4be6aefSJosef Bacik } 1366b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1367b4be6aefSJosef Bacik 1368b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1369b4be6aefSJosef Bacik } 1370b4be6aefSJosef Bacik 1371b4be6aefSJosef Bacik /* 1372d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1373d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1374d352ac68SChris Mason * be deleted 1375d352ac68SChris Mason */ 1376cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13775eda7b5eSChris Mason { 13780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13790b246afaSJeff Mahoney 13800b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1381dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1382dc9492c1SJosef Bacik btrfs_grab_root(root); 1383b4be6aefSJosef Bacik 1384b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1385b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1386b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1387b4be6aefSJosef Bacik else 13880b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1389dc9492c1SJosef Bacik } 13900b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13915eda7b5eSChris Mason } 13925eda7b5eSChris Mason 1393d352ac68SChris Mason /* 1394dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1395dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1396d352ac68SChris Mason */ 13977e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13980f7d52f4SChris Mason { 13997e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1400fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1401fc7cbcd4SDavid Sterba int i; 1402fc7cbcd4SDavid Sterba int ret; 140354aa1f4dSChris Mason 1404dfba78dcSFilipe Manana /* 1405dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1406dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1407dfba78dcSFilipe Manana */ 1408dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1409dfba78dcSFilipe Manana 1410fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1411fc7cbcd4SDavid Sterba while (1) { 1412fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1413fc7cbcd4SDavid Sterba (void **)gang, 0, 1414fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1415fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1416fc7cbcd4SDavid Sterba if (ret == 0) 1417fc7cbcd4SDavid Sterba break; 1418fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1419fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1420fc7cbcd4SDavid Sterba int ret2; 14214f4317c1SJosef Bacik 1422dfba78dcSFilipe Manana /* 1423dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1424dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1425dfba78dcSFilipe Manana */ 1426dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1427dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1428dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1429dfba78dcSFilipe Manana 1430fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14312619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14320f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1433fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 143431153d81SYan Zheng 1435e02119d5SChris Mason btrfs_free_log(trans, root); 1436fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1437fc7cbcd4SDavid Sterba if (ret2) 1438fc7cbcd4SDavid Sterba return ret2; 1439e02119d5SChris Mason 1440fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 144127cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1442c7548af6SChris Mason smp_mb__after_atomic(); 1443f1ebcc74SLiu Bo 1444978d910dSYan Zheng if (root->commit_root != root->node) { 14459e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14469e351cc8SJosef Bacik &trans->transaction->switch_commits); 1447fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1448fc7cbcd4SDavid Sterba root->node); 1449978d910dSYan Zheng } 145031153d81SYan Zheng 1451fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1452fc7cbcd4SDavid Sterba &root->root_key, 1453fc7cbcd4SDavid Sterba &root->root_item); 1454fc7cbcd4SDavid Sterba if (ret2) 1455fc7cbcd4SDavid Sterba return ret2; 1456fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1457733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14580f7d52f4SChris Mason } 1459fc7cbcd4SDavid Sterba } 1460fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14614f4317c1SJosef Bacik return 0; 14620f7d52f4SChris Mason } 14630f7d52f4SChris Mason 1464d352ac68SChris Mason /* 1465de78b51aSEric Sandeen * defrag a given btree. 1466de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1467d352ac68SChris Mason */ 1468de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1469e9d0b13bSChris Mason { 1470e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1471e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14728929ecfaSYan, Zheng int ret; 1473e9d0b13bSChris Mason 147427cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1475e9d0b13bSChris Mason return 0; 14768929ecfaSYan, Zheng 14776b80053dSChris Mason while (1) { 14788929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14796819703fSDavid Sterba if (IS_ERR(trans)) { 14806819703fSDavid Sterba ret = PTR_ERR(trans); 14816819703fSDavid Sterba break; 14826819703fSDavid Sterba } 14838929ecfaSYan, Zheng 1484de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14858929ecfaSYan, Zheng 14863a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14872ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1488e9d0b13bSChris Mason cond_resched(); 1489e9d0b13bSChris Mason 1490ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1491e9d0b13bSChris Mason break; 1492210549ebSDavid Sterba 1493ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1494ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1495210549ebSDavid Sterba ret = -EAGAIN; 1496210549ebSDavid Sterba break; 1497210549ebSDavid Sterba } 1498e9d0b13bSChris Mason } 149927cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 15008929ecfaSYan, Zheng return ret; 1501e9d0b13bSChris Mason } 1502e9d0b13bSChris Mason 1503d352ac68SChris Mason /* 15046426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 15056426c7adSQu Wenruo * 15066426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 15076426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 15086426c7adSQu Wenruo * qgroup works. 15096426c7adSQu Wenruo */ 15106426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15116426c7adSQu Wenruo struct btrfs_root *src, 15126426c7adSQu Wenruo struct btrfs_root *parent, 15136426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15146426c7adSQu Wenruo u64 dst_objectid) 15156426c7adSQu Wenruo { 15166426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15176426c7adSQu Wenruo int ret; 15186426c7adSQu Wenruo 15196426c7adSQu Wenruo /* 15206426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15216426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15226426c7adSQu Wenruo * kick in anyway. 15236426c7adSQu Wenruo */ 15249ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15256426c7adSQu Wenruo return 0; 15266426c7adSQu Wenruo 15276426c7adSQu Wenruo /* 152852042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15294d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15304d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15314d31778aSQu Wenruo * item. 15324d31778aSQu Wenruo */ 15331c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15341c442d22SJosef Bacik if (ret) 15351c442d22SJosef Bacik return ret; 15364d31778aSQu Wenruo 15374d31778aSQu Wenruo /* 15382a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15392a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15402a4d84c1SJosef Bacik * 15412a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15422a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15432a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15442a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15452a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15462a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15472a4d84c1SJosef Bacik */ 15482a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15492a4d84c1SJosef Bacik if (ret) { 15502a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 155144365827SNaohiro Aota return ret; 15522a4d84c1SJosef Bacik } 15532a4d84c1SJosef Bacik 15547e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15556426c7adSQu Wenruo if (ret) 15566426c7adSQu Wenruo goto out; 1557460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15586426c7adSQu Wenruo if (ret < 0) 15596426c7adSQu Wenruo goto out; 15606426c7adSQu Wenruo 15616426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1562a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15636426c7adSQu Wenruo inherit); 15646426c7adSQu Wenruo if (ret < 0) 15656426c7adSQu Wenruo goto out; 15666426c7adSQu Wenruo 15676426c7adSQu Wenruo /* 15686426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15696426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15706426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15716426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15726426c7adSQu Wenruo * 2) write all btree blocks onto disk 15736426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15746426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15756426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15766426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15776426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15786426c7adSQu Wenruo */ 15799386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15806426c7adSQu Wenruo if (ret) 15816426c7adSQu Wenruo goto out; 1582889bfa39SJosef Bacik switch_commit_roots(trans); 158370458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15846426c7adSQu Wenruo if (ret) 1585f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15866426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15876426c7adSQu Wenruo 15886426c7adSQu Wenruo out: 15896426c7adSQu Wenruo /* 15906426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15916426c7adSQu Wenruo * last_trans == cur_transid. 15926426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15936426c7adSQu Wenruo * insert_dir_item() 15946426c7adSQu Wenruo */ 15956426c7adSQu Wenruo if (!ret) 15961c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15976426c7adSQu Wenruo return ret; 15986426c7adSQu Wenruo } 15996426c7adSQu Wenruo 16006426c7adSQu Wenruo /* 1601d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1602aec8030aSMiao Xie * transaction commit. This does the actual creation. 1603aec8030aSMiao Xie * 1604aec8030aSMiao Xie * Note: 1605aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1606aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1607aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1608d352ac68SChris Mason */ 160980b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16103063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16113063d29fSChris Mason { 161208d50ca3SNikolay Borisov 161308d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16143063d29fSChris Mason struct btrfs_key key; 161580b6794dSChris Mason struct btrfs_root_item *new_root_item; 16163063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16173063d29fSChris Mason struct btrfs_root *root = pending->root; 16186bdb72deSSage Weil struct btrfs_root *parent_root; 161998c9942aSLiu Bo struct btrfs_block_rsv *rsv; 1620ab3c5c18SSweet Tea Dorminy struct inode *parent_inode = pending->dir; 162142874b3dSMiao Xie struct btrfs_path *path; 162242874b3dSMiao Xie struct btrfs_dir_item *dir_item; 16233063d29fSChris Mason struct extent_buffer *tmp; 1624925baeddSChris Mason struct extent_buffer *old; 162595582b00SDeepa Dinamani struct timespec64 cur_time; 1626aec8030aSMiao Xie int ret = 0; 1627d68fc57bSYan, Zheng u64 to_reserve = 0; 16286bdb72deSSage Weil u64 index = 0; 1629a22285a6SYan, Zheng u64 objectid; 1630b83cc969SLi Zefan u64 root_flags; 1631ab3c5c18SSweet Tea Dorminy unsigned int nofs_flags; 1632ab3c5c18SSweet Tea Dorminy struct fscrypt_name fname; 16333063d29fSChris Mason 16348546b570SDavid Sterba ASSERT(pending->path); 16358546b570SDavid Sterba path = pending->path; 163642874b3dSMiao Xie 1637b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1638b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1639a22285a6SYan, Zheng 1640ab3c5c18SSweet Tea Dorminy /* 1641ab3c5c18SSweet Tea Dorminy * We're inside a transaction and must make sure that any potential 1642ab3c5c18SSweet Tea Dorminy * allocations with GFP_KERNEL in fscrypt won't recurse back to 1643ab3c5c18SSweet Tea Dorminy * filesystem. 1644ab3c5c18SSweet Tea Dorminy */ 1645ab3c5c18SSweet Tea Dorminy nofs_flags = memalloc_nofs_save(); 1646ab3c5c18SSweet Tea Dorminy pending->error = fscrypt_setup_filename(parent_inode, 1647ab3c5c18SSweet Tea Dorminy &pending->dentry->d_name, 0, 1648ab3c5c18SSweet Tea Dorminy &fname); 1649ab3c5c18SSweet Tea Dorminy memalloc_nofs_restore(nofs_flags); 1650ab3c5c18SSweet Tea Dorminy if (pending->error) 1651ab3c5c18SSweet Tea Dorminy goto free_pending; 1652ab3c5c18SSweet Tea Dorminy 1653543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1654aec8030aSMiao Xie if (pending->error) 1655ab3c5c18SSweet Tea Dorminy goto free_fname; 16563063d29fSChris Mason 1657d6726335SQu Wenruo /* 1658d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1659d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1660d6726335SQu Wenruo */ 1661d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1662d6726335SQu Wenruo 1663147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1664d68fc57bSYan, Zheng 1665d68fc57bSYan, Zheng if (to_reserve > 0) { 16669270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1667aec8030aSMiao Xie &pending->block_rsv, 166808e007d2SMiao Xie to_reserve, 166908e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1670aec8030aSMiao Xie if (pending->error) 1671d6726335SQu Wenruo goto clear_skip_qgroup; 1672d68fc57bSYan, Zheng } 1673d68fc57bSYan, Zheng 16743063d29fSChris Mason key.objectid = objectid; 1675a22285a6SYan, Zheng key.offset = (u64)-1; 1676a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16773063d29fSChris Mason 16786fa9700eSMiao Xie rsv = trans->block_rsv; 1679a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16802382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16810b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 168288d3a5aaSJosef Bacik trans->transid, 168388d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1684a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1685f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1686f0118cb6SJosef Bacik if (ret) 1687f0118cb6SJosef Bacik goto fail; 1688c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 168904b285f3SDeepa Dinamani 16906bdb72deSSage Weil /* 16916bdb72deSSage Weil * insert the directory item 16926bdb72deSSage Weil */ 1693877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 169449b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 169542874b3dSMiao Xie 169642874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 169742874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16984a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 16996db75318SSweet Tea Dorminy &fname.disk_name, 0); 170042874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1701fe66a05aSChris Mason pending->error = -EEXIST; 1702aec8030aSMiao Xie goto dir_item_existed; 170342874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 170442874b3dSMiao Xie ret = PTR_ERR(dir_item); 170566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17068732d44fSMiao Xie goto fail; 170779787eaaSJeff Mahoney } 170842874b3dSMiao Xie btrfs_release_path(path); 17096bdb72deSSage Weil 1710e999376fSChris Mason /* 1711e999376fSChris Mason * pull in the delayed directory update 1712e999376fSChris Mason * and the delayed inode item 1713e999376fSChris Mason * otherwise we corrupt the FS during 1714e999376fSChris Mason * snapshot 1715e999376fSChris Mason */ 1716e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 17178732d44fSMiao Xie if (ret) { /* Transaction aborted */ 171866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17198732d44fSMiao Xie goto fail; 17208732d44fSMiao Xie } 1721e999376fSChris Mason 1722f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1723f0118cb6SJosef Bacik if (ret) { 1724f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1725f0118cb6SJosef Bacik goto fail; 1726f0118cb6SJosef Bacik } 17276bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17286bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 172908fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17306bdb72deSSage Weil 1731b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1732b83cc969SLi Zefan if (pending->readonly) 1733b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1734b83cc969SLi Zefan else 1735b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1736b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1737b83cc969SLi Zefan 17388ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17398ea05e3aSAlexander Block trans->transid); 1740807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17418ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17428ea05e3aSAlexander Block BTRFS_UUID_SIZE); 174370023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 174470023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 174570023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17468ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17478ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17488ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17498ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 175070023da2SStefan Behrens } 17513cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17523cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 175370023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17548ea05e3aSAlexander Block 1755925baeddSChris Mason old = btrfs_lock_root_node(root); 17569631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17579631e4ccSJosef Bacik BTRFS_NESTING_COW); 175879787eaaSJeff Mahoney if (ret) { 175979787eaaSJeff Mahoney btrfs_tree_unlock(old); 176079787eaaSJeff Mahoney free_extent_buffer(old); 176166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17628732d44fSMiao Xie goto fail; 176379787eaaSJeff Mahoney } 176449b25e05SJeff Mahoney 176549b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 176679787eaaSJeff Mahoney /* clean up in any case */ 1767925baeddSChris Mason btrfs_tree_unlock(old); 1768925baeddSChris Mason free_extent_buffer(old); 17698732d44fSMiao Xie if (ret) { 177066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17718732d44fSMiao Xie goto fail; 17728732d44fSMiao Xie } 1773f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 177427cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1775f1ebcc74SLiu Bo smp_wmb(); 1776f1ebcc74SLiu Bo 17775d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1778a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1779a22285a6SYan, Zheng key.offset = trans->transid; 1780a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1781925baeddSChris Mason btrfs_tree_unlock(tmp); 17823063d29fSChris Mason free_extent_buffer(tmp); 17838732d44fSMiao Xie if (ret) { 178466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17858732d44fSMiao Xie goto fail; 17868732d44fSMiao Xie } 17870660b5afSChris Mason 1788a22285a6SYan, Zheng /* 1789a22285a6SYan, Zheng * insert root back/forward references 1790a22285a6SYan, Zheng */ 17916025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1792a22285a6SYan, Zheng parent_root->root_key.objectid, 17934a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 17946db75318SSweet Tea Dorminy &fname.disk_name); 17958732d44fSMiao Xie if (ret) { 179666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17978732d44fSMiao Xie goto fail; 17988732d44fSMiao Xie } 1799a22285a6SYan, Zheng 1800a22285a6SYan, Zheng key.offset = (u64)-1; 18012dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 180279787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 180379787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 18042d892ccdSFilipe Manana pending->snap = NULL; 180566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18068732d44fSMiao Xie goto fail; 180779787eaaSJeff Mahoney } 1808d68fc57bSYan, Zheng 180949b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 18108732d44fSMiao Xie if (ret) { 181166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18128732d44fSMiao Xie goto fail; 18138732d44fSMiao Xie } 1814361048f5SMiao Xie 18156426c7adSQu Wenruo /* 18166426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 18176426c7adSQu Wenruo * snapshot hack to do fast snapshot. 18186426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 18196426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 18206426c7adSQu Wenruo */ 18216426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18226426c7adSQu Wenruo pending->inherit, objectid); 18236426c7adSQu Wenruo if (ret < 0) 18246426c7adSQu Wenruo goto fail; 18256426c7adSQu Wenruo 18266db75318SSweet Tea Dorminy ret = btrfs_insert_dir_item(trans, &fname.disk_name, 18276db75318SSweet Tea Dorminy BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, 18286db75318SSweet Tea Dorminy index); 182942874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18309c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18318732d44fSMiao Xie if (ret) { 183266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18338732d44fSMiao Xie goto fail; 18348732d44fSMiao Xie } 183542874b3dSMiao Xie 18366ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 18376db75318SSweet Tea Dorminy fname.disk_name.len * 2); 1838c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1839c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1840729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1841dd5f9615SStefan Behrens if (ret) { 184266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1843dd5f9615SStefan Behrens goto fail; 1844dd5f9615SStefan Behrens } 1845807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1846807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1847cdb345a8SLu Fengqi objectid); 1848dd5f9615SStefan Behrens if (ret) { 184966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1850dd5f9615SStefan Behrens goto fail; 1851dd5f9615SStefan Behrens } 1852dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1853cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1854dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1855dd5f9615SStefan Behrens objectid); 1856dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 185766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1858dd5f9615SStefan Behrens goto fail; 1859dd5f9615SStefan Behrens } 1860dd5f9615SStefan Behrens } 1861d6726335SQu Wenruo 18623063d29fSChris Mason fail: 1863aec8030aSMiao Xie pending->error = ret; 1864aec8030aSMiao Xie dir_item_existed: 186598c9942aSLiu Bo trans->block_rsv = rsv; 18662382c5ccSLiu Bo trans->bytes_reserved = 0; 1867d6726335SQu Wenruo clear_skip_qgroup: 1868d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 1869ab3c5c18SSweet Tea Dorminy free_fname: 1870ab3c5c18SSweet Tea Dorminy fscrypt_free_filename(&fname); 1871ab3c5c18SSweet Tea Dorminy free_pending: 18726fa9700eSMiao Xie kfree(new_root_item); 1873b0c0ea63SDavid Sterba pending->root_item = NULL; 187442874b3dSMiao Xie btrfs_free_path(path); 18758546b570SDavid Sterba pending->path = NULL; 18768546b570SDavid Sterba 187749b25e05SJeff Mahoney return ret; 18783063d29fSChris Mason } 18793063d29fSChris Mason 1880d352ac68SChris Mason /* 1881d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1882d352ac68SChris Mason */ 188308d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18843063d29fSChris Mason { 1885aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18863063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1887aec8030aSMiao Xie int ret = 0; 18883de4586cSChris Mason 1889aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1890aec8030aSMiao Xie list_del(&pending->list); 189108d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1892aec8030aSMiao Xie if (ret) 1893aec8030aSMiao Xie break; 1894aec8030aSMiao Xie } 1895aec8030aSMiao Xie return ret; 18963de4586cSChris Mason } 18973de4586cSChris Mason 18982ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18995d4f98a2SYan Zheng { 19005d4f98a2SYan Zheng struct btrfs_root_item *root_item; 19015d4f98a2SYan Zheng struct btrfs_super_block *super; 19025d4f98a2SYan Zheng 19030b246afaSJeff Mahoney super = fs_info->super_copy; 19045d4f98a2SYan Zheng 19050b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1906093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1907093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1908093e037cSDavid Sterba super->chunk_root_level = root_item->level; 19095d4f98a2SYan Zheng 19100b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1911093e037cSDavid Sterba super->root = root_item->bytenr; 1912093e037cSDavid Sterba super->generation = root_item->generation; 1913093e037cSDavid Sterba super->root_level = root_item->level; 19140b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1915093e037cSDavid Sterba super->cache_generation = root_item->generation; 191694846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 191794846229SBoris Burkov super->cache_generation = 0; 19180b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1919093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 19205d4f98a2SYan Zheng } 19215d4f98a2SYan Zheng 1922f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1923f36f3042SChris Mason { 19244a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1925f36f3042SChris Mason int ret = 0; 19264a9d8bdeSMiao Xie 1927a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19284a9d8bdeSMiao Xie trans = info->running_transaction; 19294a9d8bdeSMiao Xie if (trans) 19304a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1931a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1932f36f3042SChris Mason return ret; 1933f36f3042SChris Mason } 1934f36f3042SChris Mason 19358929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19368929ecfaSYan, Zheng { 19374a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19388929ecfaSYan, Zheng int ret = 0; 19394a9d8bdeSMiao Xie 1940a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19414a9d8bdeSMiao Xie trans = info->running_transaction; 19424a9d8bdeSMiao Xie if (trans) 19434a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1944a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19458929ecfaSYan, Zheng return ret; 19468929ecfaSYan, Zheng } 19478929ecfaSYan, Zheng 1948fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1949bb9c12c9SSage Weil { 19503a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1951bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1952bb9c12c9SSage Weil 1953fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1954fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1955fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1956bb9c12c9SSage Weil 1957bb9c12c9SSage Weil /* take transaction reference */ 1958bb9c12c9SSage Weil cur_trans = trans->transaction; 19599b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1960bb9c12c9SSage Weil 19613a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19626fc4e354SSage Weil 19636fc4e354SSage Weil /* 1964ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1965ae5d29d4SDavid Sterba * subsequent transaction joins 1966ae5d29d4SDavid Sterba */ 19673e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1968ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1969ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1970ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1971724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1972bb9c12c9SSage Weil } 1973bb9c12c9SSage Weil 197497cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 197549b25e05SJeff Mahoney { 197697cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 197749b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 197849b25e05SJeff Mahoney 1979b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 198049b25e05SJeff Mahoney 198166642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19827b8b92afSJosef Bacik 19830b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 198466b6135bSLiu Bo 198525d8c284SMiao Xie /* 198625d8c284SMiao Xie * If the transaction is removed from the list, it means this 198725d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 198825d8c284SMiao Xie * to call the cleanup function. 198925d8c284SMiao Xie */ 199025d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 199166b6135bSLiu Bo 19920b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19934a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1995e1489b4fSIoannis Angelakopoulos 1996e1489b4fSIoannis Angelakopoulos /* 1997e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 1998e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 1999e1489b4fSIoannis Angelakopoulos */ 2000e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2001f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 2002f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 2003f094ac32SLiu Bo 20040b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 2005d7096fc3SJosef Bacik } 2006061dde82SFilipe Manana 2007061dde82SFilipe Manana /* 2008061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 2009061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 2010061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 2011061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 2012061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 2013061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 2014061dde82SFilipe Manana */ 2015061dde82SFilipe Manana list_del_init(&cur_trans->list); 2016061dde82SFilipe Manana 20170b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 201849b25e05SJeff Mahoney 20192ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 202049b25e05SJeff Mahoney 20210b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20220b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20230b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20240b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20254a9d8bdeSMiao Xie 2026e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20270b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2028724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2029724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 203049b25e05SJeff Mahoney 20312e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 203249b25e05SJeff Mahoney 203349b25e05SJeff Mahoney if (current->journal_info == trans) 203449b25e05SJeff Mahoney current->journal_info = NULL; 20350b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 203649b25e05SJeff Mahoney 203749b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 203849b25e05SJeff Mahoney } 203949b25e05SJeff Mahoney 2040c7cc64a9SDavid Sterba /* 2041c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2042c7cc64a9SDavid Sterba * transaction and remove them from the list 2043c7cc64a9SDavid Sterba */ 2044c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2045c7cc64a9SDavid Sterba { 2046c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 204732da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2048c7cc64a9SDavid Sterba 2049c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2050c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2051c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2052c7cc64a9SDavid Sterba } 2053c7cc64a9SDavid Sterba } 2054c7cc64a9SDavid Sterba 205588090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 205682436617SMiao Xie { 2057ce8ea7ccSJosef Bacik /* 2058a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2059ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2060ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2061ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2062ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2063ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2064ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2065a0f0cf83SFilipe Manana * 2066a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2067a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2068a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2069a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2070a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2071a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2072a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2073ce8ea7ccSJosef Bacik */ 207488090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2075a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 207682436617SMiao Xie return 0; 207782436617SMiao Xie } 207882436617SMiao Xie 207988090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 208082436617SMiao Xie { 208188090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20826374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 208382436617SMiao Xie } 208482436617SMiao Xie 208528b21c55SFilipe Manana /* 208628b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 208728b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 208828b21c55SFilipe Manana * and while holding fs_info->trans_lock. 208928b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 209028b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 209128b21c55SFilipe Manana * returns an error. 209228b21c55SFilipe Manana */ 209328b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 209428b21c55SFilipe Manana { 209528b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 209628b21c55SFilipe Manana 209728b21c55SFilipe Manana if (!trans->pending_snapshot) 209828b21c55SFilipe Manana return; 209928b21c55SFilipe Manana 210028b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 210128b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 210228b21c55SFilipe Manana 210328b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 210428b21c55SFilipe Manana } 210528b21c55SFilipe Manana 2106e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2107e55958c8SIoannis Angelakopoulos { 2108e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2109e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2110e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2111e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2112e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2113e55958c8SIoannis Angelakopoulos } 2114e55958c8SIoannis Angelakopoulos 21153a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 211679154b1bSChris Mason { 21173a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 211849b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 21198fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 212025287e0aSMiao Xie int ret; 2121e55958c8SIoannis Angelakopoulos ktime_t start_time; 2122e55958c8SIoannis Angelakopoulos ktime_t interval; 212379154b1bSChris Mason 212435b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21253e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 212635b814f3SNikolay Borisov 2127c52cc7b7SJosef Bacik clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); 2128c52cc7b7SJosef Bacik 21298d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2130bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 213125287e0aSMiao Xie ret = cur_trans->aborted; 21323e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 213325287e0aSMiao Xie } 213449b25e05SJeff Mahoney 2135f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2136f45c752bSJosef Bacik trans->block_rsv = NULL; 2137f45c752bSJosef Bacik 2138e19eb11fSJosef Bacik /* 2139e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2140e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2141e19eb11fSJosef Bacik */ 2142e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2143e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2144e19eb11fSJosef Bacik /* 2145e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2146e19eb11fSJosef Bacik * Any running threads may add more while we are here. 214756bec294SChris Mason */ 2148c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21493e738c53SIoannis Angelakopoulos if (ret) 21503e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2151e19eb11fSJosef Bacik } 215256bec294SChris Mason 21536c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2154ea658badSJosef Bacik 21553204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21561bbc621eSChris Mason int run_it = 0; 21571bbc621eSChris Mason 21581bbc621eSChris Mason /* this mutex is also taken before trying to set 21591bbc621eSChris Mason * block groups readonly. We need to make sure 21601bbc621eSChris Mason * that nobody has set a block group readonly 21611bbc621eSChris Mason * after a extents from that block group have been 21621bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21631bbc621eSChris Mason * will wait for the transaction to commit if it 21643204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21651bbc621eSChris Mason * 21663204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21673204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21681bbc621eSChris Mason * hurt to have more than one go through, but there's no 21691bbc621eSChris Mason * real advantage to it either. 21701bbc621eSChris Mason */ 21710b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21723204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21733204d33cSJosef Bacik &cur_trans->flags)) 21741bbc621eSChris Mason run_it = 1; 21750b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21761bbc621eSChris Mason 2177f9cacae3SNikolay Borisov if (run_it) { 217821217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21793e738c53SIoannis Angelakopoulos if (ret) 21803e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2181f9cacae3SNikolay Borisov } 2182f9cacae3SNikolay Borisov } 21831bbc621eSChris Mason 21840b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21854a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2186d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2187d0c2f4faSFilipe Manana 218828b21c55SFilipe Manana add_pending_snapshot(trans); 218928b21c55SFilipe Manana 21900b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21919b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2192ccd467d6SChris Mason 2193d0c2f4faSFilipe Manana if (trans->in_fsync) 2194d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 21953e738c53SIoannis Angelakopoulos 21963e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 21973e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2198d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2199d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 220015ee9bc7SJosef Bacik 2201bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2202b4924a0fSLiu Bo ret = cur_trans->aborted; 2203b4924a0fSLiu Bo 2204724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 220515ee9bc7SJosef Bacik 220649b25e05SJeff Mahoney return ret; 220779154b1bSChris Mason } 22084313b399SChris Mason 22094a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 22100b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 22113e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2212bb9c12c9SSage Weil 22130b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2214d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2215d0c2f4faSFilipe Manana 2216d0c2f4faSFilipe Manana if (trans->in_fsync) 2217d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2218d0c2f4faSFilipe Manana 2219ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2220ccd467d6SChris Mason struct btrfs_transaction, list); 2221d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 22229b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 22230b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2224ccd467d6SChris Mason 2225d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2226d0c2f4faSFilipe Manana 2227bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2228ccd467d6SChris Mason 2229724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22301f9b8c8fSFilipe Manana if (ret) 2231e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2232a4abeea4SJosef Bacik } else { 22330b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2234ccd467d6SChris Mason } 2235a4abeea4SJosef Bacik } else { 22360b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2237cb2d3dadSFilipe Manana /* 2238cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2239cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2240cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2241cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2242cb2d3dadSFilipe Manana */ 224384961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2244cb2d3dadSFilipe Manana ret = -EROFS; 2245e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2246cb2d3dadSFilipe Manana } 2247ccd467d6SChris Mason } 224815ee9bc7SJosef Bacik 2249e55958c8SIoannis Angelakopoulos /* 2250e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2251e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2252e55958c8SIoannis Angelakopoulos */ 2253e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2254e55958c8SIoannis Angelakopoulos 22550860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22560860adfdSMiao Xie 225788090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 225882436617SMiao Xie if (ret) 2259e1489b4fSIoannis Angelakopoulos goto lockdep_release; 226082436617SMiao Xie 2261e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 226249b25e05SJeff Mahoney if (ret) 2263e1489b4fSIoannis Angelakopoulos goto lockdep_release; 226416cdcec7SMiao Xie 22655a9ba670SIoannis Angelakopoulos /* 22665a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22675a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22685a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22695a9ba670SIoannis Angelakopoulos */ 22705a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 22715a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2272581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2273581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2274ed3b3d31SChris Mason 2275581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2276e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2277e1489b4fSIoannis Angelakopoulos if (ret) { 2278e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2279ca469637SMiao Xie goto cleanup_transaction; 2280e1489b4fSIoannis Angelakopoulos } 2281ca469637SMiao Xie 228288090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2283cb7ab021SWang Shilong 228448778179SFilipe Manana /* 228548778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 228648778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 228748778179SFilipe Manana * extents complete we lose logged data after a power failure. 228848778179SFilipe Manana */ 22898b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 229048778179SFilipe Manana wait_event(cur_trans->pending_wait, 229148778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 229248778179SFilipe Manana 22932ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2294ed0ca140SJosef Bacik /* 2295ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2296ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22974a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2298ed0ca140SJosef Bacik */ 22990b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 230028b21c55SFilipe Manana add_pending_snapshot(trans); 23014a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 23020b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2303e1489b4fSIoannis Angelakopoulos 2304e1489b4fSIoannis Angelakopoulos /* 2305e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2306e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2307e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2308e1489b4fSIoannis Angelakopoulos */ 2309e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2310e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2311ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2312ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 231315ee9bc7SJosef Bacik 2314fdfbf020SJosef Bacik /* 23153e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 23163e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 23173e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 23183e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 23193e738c53SIoannis Angelakopoulos */ 23203e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 23213e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 23223e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23233e738c53SIoannis Angelakopoulos 23243e738c53SIoannis Angelakopoulos /* 2325fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2326fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2327fdfbf020SJosef Bacik * kthread could do the work. 2328fdfbf020SJosef Bacik */ 2329fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2330fdfbf020SJosef Bacik 2331bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23322cba30f1SMiao Xie ret = cur_trans->aborted; 23333e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23346cf7f77eSWang Shilong goto scrub_continue; 23352cba30f1SMiao Xie } 23367585717fSChris Mason /* 23377585717fSChris Mason * the reloc mutex makes sure that we stop 23387585717fSChris Mason * the balancing code from coming in and moving 23397585717fSChris Mason * extents around in the middle of the commit 23407585717fSChris Mason */ 23410b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23427585717fSChris Mason 234342874b3dSMiao Xie /* 234442874b3dSMiao Xie * We needn't worry about the delayed items because we will 234542874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 234642874b3dSMiao Xie * core function of the snapshot creation. 234742874b3dSMiao Xie */ 234808d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 234956e9f6eaSDavid Sterba if (ret) 235056e9f6eaSDavid Sterba goto unlock_reloc; 23513063d29fSChris Mason 235242874b3dSMiao Xie /* 235342874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 235442874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 235542874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 235642874b3dSMiao Xie * them. 235742874b3dSMiao Xie * 235842874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 235942874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 236042874b3dSMiao Xie * the nodes and leaves. 236142874b3dSMiao Xie */ 2362e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 236356e9f6eaSDavid Sterba if (ret) 236456e9f6eaSDavid Sterba goto unlock_reloc; 236516cdcec7SMiao Xie 2366c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 236756e9f6eaSDavid Sterba if (ret) 236856e9f6eaSDavid Sterba goto unlock_reloc; 236956bec294SChris Mason 2370e999376fSChris Mason /* 2371e999376fSChris Mason * make sure none of the code above managed to slip in a 2372e999376fSChris Mason * delayed item 2373e999376fSChris Mason */ 2374ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2375e999376fSChris Mason 23762c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2377dc17ff8fSChris Mason 23787e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 237956e9f6eaSDavid Sterba if (ret) 2380dfba78dcSFilipe Manana goto unlock_reloc; 238154aa1f4dSChris Mason 23825d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2383e02119d5SChris Mason * safe to free the root of tree log roots 2384e02119d5SChris Mason */ 23850b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2386e02119d5SChris Mason 23870ed4792aSQu Wenruo /* 23880ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 23890ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 23900ed4792aSQu Wenruo */ 2391460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 239256e9f6eaSDavid Sterba if (ret < 0) 2393dfba78dcSFilipe Manana goto unlock_reloc; 23940ed4792aSQu Wenruo 23959386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 239656e9f6eaSDavid Sterba if (ret) 2397dfba78dcSFilipe Manana goto unlock_reloc; 239854aa1f4dSChris Mason 23992cba30f1SMiao Xie /* 24002cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 24012cba30f1SMiao Xie * update ->aborted, check it. 24022cba30f1SMiao Xie */ 2403bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 24042cba30f1SMiao Xie ret = cur_trans->aborted; 2405dfba78dcSFilipe Manana goto unlock_reloc; 24062cba30f1SMiao Xie } 24072cba30f1SMiao Xie 24080b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 24095f39d397SChris Mason 24100b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 24110b246afaSJeff Mahoney fs_info->tree_root->node); 24120b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 24139e351cc8SJosef Bacik &cur_trans->switch_commits); 24145d4f98a2SYan Zheng 24150b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 24160b246afaSJeff Mahoney fs_info->chunk_root->node); 24170b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 24189e351cc8SJosef Bacik &cur_trans->switch_commits); 24199e351cc8SJosef Bacik 2420f7238e50SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2421f7238e50SJosef Bacik btrfs_set_root_node(&fs_info->block_group_root->root_item, 2422f7238e50SJosef Bacik fs_info->block_group_root->node); 2423f7238e50SJosef Bacik list_add_tail(&fs_info->block_group_root->dirty_list, 2424f7238e50SJosef Bacik &cur_trans->switch_commits); 2425f7238e50SJosef Bacik } 2426f7238e50SJosef Bacik 2427889bfa39SJosef Bacik switch_commit_roots(trans); 24285d4f98a2SYan Zheng 2429ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 24301bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 24312ff7e61eSJeff Mahoney update_super_roots(fs_info); 2432e02119d5SChris Mason 24330b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 24340b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 24350b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 24360b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2437ccd467d6SChris Mason 2438bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2439935e5cc9SMiao Xie 24400b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 24410b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2442656f30dbSFilipe Manana 24434fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24444fbcdf66SFilipe Manana 2445dfba78dcSFilipe Manana /* 2446dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2447dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2448dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2449dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2450dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2451dfba78dcSFilipe Manana * writing its superblock. 2452dfba78dcSFilipe Manana */ 2453dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2454dfba78dcSFilipe Manana 24550b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 24564a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 24570b246afaSJeff Mahoney fs_info->running_transaction = NULL; 24580b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 24590b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2460b7ec40d7SChris Mason 24610b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 24623e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2463e6dcd2dcSChris Mason 246470458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 246549b25e05SJeff Mahoney if (ret) { 24660b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 246708748810SDavid Sterba "Error while writing out transaction"); 24680b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 24696cf7f77eSWang Shilong goto scrub_continue; 247049b25e05SJeff Mahoney } 247149b25e05SJeff Mahoney 2472d3575156SNaohiro Aota /* 2473d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2474d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2475d3575156SNaohiro Aota * buffers. 2476d3575156SNaohiro Aota */ 2477d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2478d3575156SNaohiro Aota 2479eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2480e02119d5SChris Mason /* 2481e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2482e02119d5SChris Mason * to go about their business 2483e02119d5SChris Mason */ 24840b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2485c1f32b7cSAnand Jain if (ret) 2486c1f32b7cSAnand Jain goto scrub_continue; 2487e02119d5SChris Mason 2488d0c2f4faSFilipe Manana /* 2489d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2490d0c2f4faSFilipe Manana * which can change it. 2491d0c2f4faSFilipe Manana */ 2492d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2493d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 24943e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2495d0c2f4faSFilipe Manana 24965ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 24974313b399SChris Mason 24983204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 24990b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 250013212b54SZhao Lei 25010b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 25024a9d8bdeSMiao Xie /* 25034a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 25044a9d8bdeSMiao Xie * which can change it. 25054a9d8bdeSMiao Xie */ 25064a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 25072c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 25083e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25093de4586cSChris Mason 25100b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 251113c5a93eSJosef Bacik list_del_init(&cur_trans->list); 25120b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2513a4abeea4SJosef Bacik 2514724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2515724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 251658176a96SJosef Bacik 25170860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 25180b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2519b2b5ef5cSJan Kara 25202e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 25211abe9b8aSliubo 2522e55958c8SIoannis Angelakopoulos interval = ktime_get_ns() - start_time; 2523e55958c8SIoannis Angelakopoulos 25242ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2525a2de733cSArne Jansen 25269ed74f2dSJosef Bacik if (current->journal_info == trans) 25279ed74f2dSJosef Bacik current->journal_info = NULL; 25289ed74f2dSJosef Bacik 25292c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 253024bbcf04SYan, Zheng 2531e55958c8SIoannis Angelakopoulos update_commit_stats(fs_info, interval); 2532e55958c8SIoannis Angelakopoulos 253379154b1bSChris Mason return ret; 253449b25e05SJeff Mahoney 253556e9f6eaSDavid Sterba unlock_reloc: 253656e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 25373e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 25386cf7f77eSWang Shilong scrub_continue: 25393e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 25403e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25412ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 254249b25e05SJeff Mahoney cleanup_transaction: 2543dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2544c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 25454fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 25460e721106SJosef Bacik trans->block_rsv = NULL; 25470b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 254849b25e05SJeff Mahoney if (current->journal_info == trans) 254949b25e05SJeff Mahoney current->journal_info = NULL; 255097cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 255149b25e05SJeff Mahoney 255249b25e05SJeff Mahoney return ret; 2553e1489b4fSIoannis Angelakopoulos 2554e1489b4fSIoannis Angelakopoulos lockdep_release: 25555a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2556e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2557e1489b4fSIoannis Angelakopoulos goto cleanup_transaction; 25583e738c53SIoannis Angelakopoulos 25593e738c53SIoannis Angelakopoulos lockdep_trans_commit_start_release: 25603e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 25613e738c53SIoannis Angelakopoulos btrfs_end_transaction(trans); 25623e738c53SIoannis Angelakopoulos return ret; 256379154b1bSChris Mason } 256479154b1bSChris Mason 2565d352ac68SChris Mason /* 25669d1a2a3aSDavid Sterba * return < 0 if error 25679d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 25689d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 25699d1a2a3aSDavid Sterba * 25709d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 25719d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 25729d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 25739d1a2a3aSDavid Sterba * few seconds later. 2574d352ac68SChris Mason */ 257533c44184SJosef Bacik int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2576e9d0b13bSChris Mason { 257733c44184SJosef Bacik struct btrfs_root *root; 25789d1a2a3aSDavid Sterba int ret; 2579e9d0b13bSChris Mason 2580a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 25819d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 25829d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 25839d1a2a3aSDavid Sterba return 0; 25849d1a2a3aSDavid Sterba } 25859d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 25869d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2587cfad392bSJosef Bacik list_del_init(&root->root_list); 2588a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 25895d4f98a2SYan Zheng 25904fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 259176dda93cSYan, Zheng 259216cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 259316cdcec7SMiao Xie 259476dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 259576dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 25960078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 259776dda93cSYan, Zheng else 25980078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 259932471dc2SDavid Sterba 2600dc9492c1SJosef Bacik btrfs_put_root(root); 26016596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2602e9d0b13bSChris Mason } 2603572d9ab7SDavid Sterba 2604956504a3SJosef Bacik int __init btrfs_transaction_init(void) 2605956504a3SJosef Bacik { 2606956504a3SJosef Bacik btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2607956504a3SJosef Bacik sizeof(struct btrfs_trans_handle), 0, 2608956504a3SJosef Bacik SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2609956504a3SJosef Bacik if (!btrfs_trans_handle_cachep) 2610956504a3SJosef Bacik return -ENOMEM; 2611956504a3SJosef Bacik return 0; 2612956504a3SJosef Bacik } 2613956504a3SJosef Bacik 2614956504a3SJosef Bacik void __cold btrfs_transaction_exit(void) 2615956504a3SJosef Bacik { 2616956504a3SJosef Bacik kmem_cache_destroy(btrfs_trans_handle_cachep); 2617956504a3SJosef Bacik } 2618