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" 33c7a03b52SJosef Bacik #include "uuid-tree.h" 347572dec8SJosef Bacik #include "ioctl.h" 3567707479SJosef Bacik #include "relocation.h" 362fc6822cSJosef Bacik #include "scrub.h" 3779154b1bSChris Mason 38956504a3SJosef Bacik static struct kmem_cache *btrfs_trans_handle_cachep; 39956504a3SJosef Bacik 40fc7cbcd4SDavid Sterba #define BTRFS_ROOT_TRANS_TAG 0 410f7d52f4SChris Mason 4261c047b5SQu Wenruo /* 4361c047b5SQu Wenruo * Transaction states and transitions 4461c047b5SQu Wenruo * 4561c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 4661c047b5SQu Wenruo * | 4761c047b5SQu Wenruo * | To next stage: 4861c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 4961c047b5SQu Wenruo * V 5061c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 5161c047b5SQu Wenruo * | 5261c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 5361c047b5SQu Wenruo * | start_transaction() variants. 5461c047b5SQu Wenruo * | 5561c047b5SQu Wenruo * | To next stage: 5661c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 5761c047b5SQu Wenruo * | transaction N 5861c047b5SQu Wenruo * V 5961c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 6061c047b5SQu Wenruo * | 6161c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 6261c047b5SQu Wenruo * | is one 6361c047b5SQu Wenruo * | 6461c047b5SQu Wenruo * | Then one of the following happes: 6561c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 6661c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 6761c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 6861c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 6961c047b5SQu Wenruo * | 7061c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 7161c047b5SQu Wenruo * | to this running transaction. 7261c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 7361c047b5SQu Wenruo * | transaction N+1. 7461c047b5SQu Wenruo * | 7561c047b5SQu Wenruo * | To next stage: 7661c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 7761c047b5SQu Wenruo * | haven been released. 7861c047b5SQu Wenruo * V 7961c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 8061c047b5SQu Wenruo * | 8161c047b5SQu Wenruo * | The heavy lifting transaction work is started. 8261c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 8361c047b5SQu Wenruo * | snapshots, running qgroups. 8461c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 8561c047b5SQu Wenruo * | trees. 8661c047b5SQu Wenruo * | 8761c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 8861c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 8961c047b5SQu Wenruo * | 9061c047b5SQu Wenruo * | To next stage: 9161c047b5SQu Wenruo * | Until all supporting trees are updated. 9261c047b5SQu Wenruo * V 9361c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 9461c047b5SQu Wenruo * | Transaction N+1 9561c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 9661c047b5SQu Wenruo * | need to write them back to disk and update | 9761c047b5SQu Wenruo * | super blocks. | 9861c047b5SQu Wenruo * | | 9961c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 10061c047b5SQu Wenruo * | start. | 10161c047b5SQu Wenruo * | All new start_transaction() calls will be | 10261c047b5SQu Wenruo * | attached to transid N+1. | 10361c047b5SQu Wenruo * | | 10461c047b5SQu Wenruo * | To next stage: | 10561c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 10661c047b5SQu Wenruo * | written to block devices | 10761c047b5SQu Wenruo * V | 10861c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 10961c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 11061c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 11161c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 11261c047b5SQu Wenruo */ 113e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1144a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 115bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 116bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1174a9d8bdeSMiao Xie __TRANS_ATTACH | 118a6d155d2SFilipe Manana __TRANS_JOIN | 119a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 120bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1214a9d8bdeSMiao Xie __TRANS_ATTACH | 1224a9d8bdeSMiao Xie __TRANS_JOIN | 123a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 124a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 125d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 126d0c2f4faSFilipe Manana __TRANS_ATTACH | 127d0c2f4faSFilipe Manana __TRANS_JOIN | 128d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 129d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 130bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1314a9d8bdeSMiao Xie __TRANS_ATTACH | 1324a9d8bdeSMiao Xie __TRANS_JOIN | 133a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 134a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1354a9d8bdeSMiao Xie }; 1364a9d8bdeSMiao Xie 137724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 13879154b1bSChris Mason { 1399b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1409b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 141a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1425c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1435c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 14481f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 14581f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1461262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 147ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 148ab8d0fc4SJeff Mahoney "pending csums is %llu", 1491262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1507785a663SFilipe Manana /* 1517785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1527785a663SFilipe Manana * because the transaction was aborted and a commit did not 1537785a663SFilipe Manana * happen (things failed before writing the new superblock 1547785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1557785a663SFilipe Manana * discard the physical locations of the block groups. 1567785a663SFilipe Manana */ 1577785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 15832da5386SDavid Sterba struct btrfs_block_group *cache; 1597785a663SFilipe Manana 1607785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 16132da5386SDavid Sterba struct btrfs_block_group, 1627785a663SFilipe Manana bg_list); 1637785a663SFilipe Manana list_del_init(&cache->bg_list); 1646b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1657785a663SFilipe Manana btrfs_put_block_group(cache); 1667785a663SFilipe Manana } 167bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1684b5faeacSDavid Sterba kfree(transaction); 16979154b1bSChris Mason } 17078fae27eSChris Mason } 17179154b1bSChris Mason 172889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 173817d52f8SJosef Bacik { 174889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 17516916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1769e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 1779e351cc8SJosef Bacik 178dfba78dcSFilipe Manana /* 179dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 180dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 181dfba78dcSFilipe Manana */ 182dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 183dfba78dcSFilipe Manana 1849e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 185d96b3424SFilipe Manana 186d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 187d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 188d96b3424SFilipe Manana 189889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1909e351cc8SJosef Bacik dirty_list) { 1919e351cc8SJosef Bacik list_del_init(&root->dirty_list); 192817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 193817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 19441e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 195370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1969e351cc8SJosef Bacik } 1972b9dbef2SJosef Bacik 1982b9dbef2SJosef Bacik /* We can free old roots now. */ 199889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 200889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 201889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 2022b9dbef2SJosef Bacik struct btrfs_root, root_list); 2032b9dbef2SJosef Bacik list_del_init(&root->root_list); 204889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 205889bfa39SJosef Bacik btrfs_free_log(trans, root); 2062b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 207889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 2082b9dbef2SJosef Bacik } 209889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 21027d56e62SJosef Bacik 2119e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 212817d52f8SJosef Bacik } 213817d52f8SJosef Bacik 2140860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2150860adfdSMiao Xie unsigned int type) 2160860adfdSMiao Xie { 2170860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2180860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2190860adfdSMiao Xie } 2200860adfdSMiao Xie 2210860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2220860adfdSMiao Xie unsigned int type) 2230860adfdSMiao Xie { 2240860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2250860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2260860adfdSMiao Xie } 2270860adfdSMiao Xie 2280860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2290860adfdSMiao Xie unsigned int type) 2300860adfdSMiao Xie { 2310860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2320860adfdSMiao Xie } 2330860adfdSMiao Xie 2340860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2350860adfdSMiao Xie { 2360860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 237178260b2SMiao Xie } 238178260b2SMiao Xie 239d352ac68SChris Mason /* 24079bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 24179bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 24279bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 24379bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 24479bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 245fb6dea26SJosef Bacik */ 246fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 247fb6dea26SJosef Bacik { 248fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 249fb6dea26SJosef Bacik 250fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 251fb6dea26SJosef Bacik return; 252fb6dea26SJosef Bacik 253fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 25463f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 255fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 256fb6dea26SJosef Bacik } 257fb6dea26SJosef Bacik 258fb6dea26SJosef Bacik /* 259d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 260d352ac68SChris Mason */ 2612ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2622ff7e61eSJeff Mahoney unsigned int type) 26379154b1bSChris Mason { 26479154b1bSChris Mason struct btrfs_transaction *cur_trans; 265a4abeea4SJosef Bacik 26619ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 267d43317dcSChris Mason loop: 26849b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 26984961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 27019ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 27149b25e05SJeff Mahoney return -EROFS; 27249b25e05SJeff Mahoney } 27349b25e05SJeff Mahoney 27419ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 275a4abeea4SJosef Bacik if (cur_trans) { 276bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 27719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 27849b25e05SJeff Mahoney return cur_trans->aborted; 279871383beSDavid Sterba } 2804a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 281178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 282178260b2SMiao Xie return -EBUSY; 283178260b2SMiao Xie } 2849b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 285a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2860860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 28719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 288e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2895a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 290a4abeea4SJosef Bacik return 0; 291a4abeea4SJosef Bacik } 29219ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 293a4abeea4SJosef Bacik 294354aa0fbSMiao Xie /* 295354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 296354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 297354aa0fbSMiao Xie */ 298354aa0fbSMiao Xie if (type == TRANS_ATTACH) 299354aa0fbSMiao Xie return -ENOENT; 300354aa0fbSMiao Xie 3014a9d8bdeSMiao Xie /* 3024a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3034a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3044a9d8bdeSMiao Xie */ 3054a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3064a9d8bdeSMiao Xie 3074b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 308db5b493aSTsutomu Itoh if (!cur_trans) 309db5b493aSTsutomu Itoh return -ENOMEM; 310d43317dcSChris Mason 311e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 3125a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 313e1489b4fSIoannis Angelakopoulos 31419ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 31519ae4e81SJan Schmidt if (fs_info->running_transaction) { 316d43317dcSChris Mason /* 317d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3184a9d8bdeSMiao Xie * to redo the checks above 319d43317dcSChris Mason */ 3205a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 321e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3224b5faeacSDavid Sterba kfree(cur_trans); 323d43317dcSChris Mason goto loop; 32484961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 325e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3265a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 327e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3284b5faeacSDavid Sterba kfree(cur_trans); 3297b8b92afSJosef Bacik return -EROFS; 330a4abeea4SJosef Bacik } 331d43317dcSChris Mason 332ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 33348778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 33448778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 33513c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3360860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 33779154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 33879154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3394a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 340a4abeea4SJosef Bacik /* 341a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 342a4abeea4SJosef Bacik * commit the transaction. 343a4abeea4SJosef Bacik */ 3449b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3453204d33cSJosef Bacik cur_trans->flags = 0; 346afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 34756bec294SChris Mason 348a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 349a099d0fdSAlexandru Moise 3505c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3513368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 352d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 35320b297d6SJan Schmidt 35420b297d6SJan Schmidt /* 35520b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 35620b297d6SJan Schmidt * the log must never go across transaction boundaries. 35720b297d6SJan Schmidt */ 35820b297d6SJan Schmidt smp_mb(); 35931b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3605d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 36131b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3625d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 363fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 36420b297d6SJan Schmidt 36556bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 36656bec294SChris Mason 3673063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 368bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3699e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 370ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3711bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3722b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3731bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 374ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 375e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3762b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 37719ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 378c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 37935da5a7eSDavid Sterba IO_TREE_TRANS_DIRTY_PAGES); 380fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 38135da5a7eSDavid Sterba IO_TREE_FS_PINNED_EXTENTS); 38219ae4e81SJan Schmidt fs_info->generation++; 38319ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 38419ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 38549b25e05SJeff Mahoney cur_trans->aborted = 0; 38619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 38715ee9bc7SJosef Bacik 38879154b1bSChris Mason return 0; 38979154b1bSChris Mason } 39079154b1bSChris Mason 391d352ac68SChris Mason /* 39292a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 39392a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 39492a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 39592a7cc42SQu Wenruo * transaction commits. 396d352ac68SChris Mason */ 3977585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3986426c7adSQu Wenruo struct btrfs_root *root, 3996426c7adSQu Wenruo int force) 4006702ed49SChris Mason { 4010b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 40203a7e111SJosef Bacik int ret = 0; 4030b246afaSJeff Mahoney 40492a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4056426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4064d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4075d4f98a2SYan Zheng 4087585717fSChris Mason /* 40927cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4107585717fSChris Mason * we have the reloc mutex held now, so there 4117585717fSChris Mason * is only one writer in this function 4127585717fSChris Mason */ 41327cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4147585717fSChris Mason 41527cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4167585717fSChris Mason * they find our root->last_trans update 4177585717fSChris Mason */ 4187585717fSChris Mason smp_wmb(); 4197585717fSChris Mason 420fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4216426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 422fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 423a4abeea4SJosef Bacik return 0; 424a4abeea4SJosef Bacik } 425fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4266702ed49SChris Mason (unsigned long)root->root_key.objectid, 4276702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 428fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4297585717fSChris Mason root->last_trans = trans->transid; 4307585717fSChris Mason 4317585717fSChris Mason /* this is pretty tricky. We don't want to 4327585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4337585717fSChris Mason * unless we're really doing the first setup for this root in 4347585717fSChris Mason * this transaction. 4357585717fSChris Mason * 4367585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4377585717fSChris Mason * if we want to take the expensive mutex. 4387585717fSChris Mason * 4397585717fSChris Mason * But, we have to set root->last_trans before we 4407585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4417585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 44227cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4437585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4447585717fSChris Mason * 4457585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4467585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4477585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4487585717fSChris Mason * done before we pop in the zero below 4497585717fSChris Mason */ 45003a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 451c7548af6SChris Mason smp_mb__before_atomic(); 45227cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4536702ed49SChris Mason } 45403a7e111SJosef Bacik return ret; 4556702ed49SChris Mason } 4565d4f98a2SYan Zheng 4577585717fSChris Mason 4582b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4592b9dbef2SJosef Bacik struct btrfs_root *root) 4602b9dbef2SJosef Bacik { 4610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4622b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4632b9dbef2SJosef Bacik 4642b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4652b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4662b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4672b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4682b9dbef2SJosef Bacik 4692b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 470fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 471fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4722b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4732b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 474fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4752b9dbef2SJosef Bacik } 4762b9dbef2SJosef Bacik 4777585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4787585717fSChris Mason struct btrfs_root *root) 4797585717fSChris Mason { 4800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4811409e6ccSJosef Bacik int ret; 4820b246afaSJeff Mahoney 48392a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4847585717fSChris Mason return 0; 4857585717fSChris Mason 4867585717fSChris Mason /* 48727cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4887585717fSChris Mason * and barriers 4897585717fSChris Mason */ 4907585717fSChris Mason smp_rmb(); 4917585717fSChris Mason if (root->last_trans == trans->transid && 49227cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4937585717fSChris Mason return 0; 4947585717fSChris Mason 4950b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4961409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4970b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4987585717fSChris Mason 4991409e6ccSJosef Bacik return ret; 5007585717fSChris Mason } 5017585717fSChris Mason 5024a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5034a9d8bdeSMiao Xie { 5043296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 505501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 506bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5074a9d8bdeSMiao Xie } 5084a9d8bdeSMiao Xie 509d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 510d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 511d352ac68SChris Mason * transaction might not be fully on disk. 512d352ac68SChris Mason */ 5132ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 51479154b1bSChris Mason { 515f9295749SChris Mason struct btrfs_transaction *cur_trans; 51679154b1bSChris Mason 5170b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5180b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5194a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5209b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5210b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52272d63ed6SLi Zefan 5233e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5240b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 525501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 526bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 527724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 528a4abeea4SJosef Bacik } else { 5290b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 530f9295749SChris Mason } 53137d1aeeeSChris Mason } 53237d1aeeeSChris Mason 5332ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 53437d1aeeeSChris Mason { 5350b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 536a4abeea4SJosef Bacik return 0; 537a4abeea4SJosef Bacik 53892e2f7e3SNikolay Borisov if (type == TRANS_START) 539a4abeea4SJosef Bacik return 1; 540a4abeea4SJosef Bacik 541a22285a6SYan, Zheng return 0; 542a22285a6SYan, Zheng } 543a22285a6SYan, Zheng 54420dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 54520dd2cbfSMiao Xie { 5460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5470b246afaSJeff Mahoney 5480b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 54992a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 55020dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 55120dd2cbfSMiao Xie root->reloc_root) 55220dd2cbfSMiao Xie return false; 55320dd2cbfSMiao Xie 55420dd2cbfSMiao Xie return true; 55520dd2cbfSMiao Xie } 55620dd2cbfSMiao Xie 55708e007d2SMiao Xie static struct btrfs_trans_handle * 5585aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 559003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 560003d7c59SJeff Mahoney bool enforce_qgroups) 561a22285a6SYan, Zheng { 5620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 563ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 564a22285a6SYan, Zheng struct btrfs_trans_handle *h; 565a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 566b5009945SJosef Bacik u64 num_bytes = 0; 567c5567237SArne Jansen u64 qgroup_reserved = 0; 56820dd2cbfSMiao Xie bool reloc_reserved = false; 5699c343784SJosef Bacik bool do_chunk_alloc = false; 57020dd2cbfSMiao Xie int ret; 571acce952bSliubo 57284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 573acce952bSliubo return ERR_PTR(-EROFS); 5742a1eb461SJosef Bacik 57546c4e71eSFilipe Manana if (current->journal_info) { 5760860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5772a1eb461SJosef Bacik h = current->journal_info; 578b50fff81SDavid Sterba refcount_inc(&h->use_count); 579b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5802a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5812a1eb461SJosef Bacik h->block_rsv = NULL; 5822a1eb461SJosef Bacik goto got_it; 5832a1eb461SJosef Bacik } 584b5009945SJosef Bacik 585b5009945SJosef Bacik /* 586b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 587b5009945SJosef Bacik * the appropriate flushing if need be. 588b5009945SJosef Bacik */ 589003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 590ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 591ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 592ba2c4d4eSJosef Bacik 5930b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 594733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 595003d7c59SJeff Mahoney enforce_qgroups); 596c5567237SArne Jansen if (ret) 597c5567237SArne Jansen return ERR_PTR(ret); 598c5567237SArne Jansen 599ba2c4d4eSJosef Bacik /* 600ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 601ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 6020f69d1f4SFilipe Manana * accomplish this by simply assuming we'll do num_items worth 6030f69d1f4SFilipe Manana * of delayed refs updates in this trans handle, and refill that 6040f69d1f4SFilipe Manana * amount for whatever is missing in the reserve. 605ba2c4d4eSJosef Bacik */ 6062bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6077f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 608e4773b57SFilipe Manana !btrfs_block_rsv_full(delayed_refs_rsv)) { 6090f69d1f4SFilipe Manana delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, 6100f69d1f4SFilipe Manana num_items); 6110f69d1f4SFilipe Manana num_bytes += delayed_refs_bytes; 612ba2c4d4eSJosef Bacik } 613ba2c4d4eSJosef Bacik 61420dd2cbfSMiao Xie /* 61520dd2cbfSMiao Xie * Do the reservation for the relocation root creation 61620dd2cbfSMiao Xie */ 617ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6180b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 61920dd2cbfSMiao Xie reloc_reserved = true; 62020dd2cbfSMiao Xie } 62120dd2cbfSMiao Xie 6229270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 623ba2c4d4eSJosef Bacik if (ret) 624ba2c4d4eSJosef Bacik goto reserve_fail; 625ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 626ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 627ba2c4d4eSJosef Bacik delayed_refs_bytes); 628ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 629ba2c4d4eSJosef Bacik } 6309c343784SJosef Bacik 6319c343784SJosef Bacik if (rsv->space_info->force_alloc) 6329c343784SJosef Bacik do_chunk_alloc = true; 633ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 634748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 635ba2c4d4eSJosef Bacik /* 636ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 637ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 638ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 639ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 640ba2c4d4eSJosef Bacik * here. 641ba2c4d4eSJosef Bacik */ 642ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 643b5009945SJosef Bacik if (ret) 644843fcf35SMiao Xie goto reserve_fail; 645b5009945SJosef Bacik } 646a22285a6SYan, Zheng again: 647f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 648843fcf35SMiao Xie if (!h) { 649843fcf35SMiao Xie ret = -ENOMEM; 650843fcf35SMiao Xie goto alloc_fail; 651843fcf35SMiao Xie } 652a22285a6SYan, Zheng 65398114659SJosef Bacik /* 65498114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 65598114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 65698114659SJosef Bacik * because we're already holding a ref. We need this because we could 65798114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 65898114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 659354aa0fbSMiao Xie * 660354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 661354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 66298114659SJosef Bacik */ 6630860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6640b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 665b2b5ef5cSJan Kara 6662ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6672ff7e61eSJeff Mahoney wait_current_trans(fs_info); 668a22285a6SYan, Zheng 669a4abeea4SJosef Bacik do { 6702ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 671178260b2SMiao Xie if (ret == -EBUSY) { 6722ff7e61eSJeff Mahoney wait_current_trans(fs_info); 673a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 674a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 675178260b2SMiao Xie ret = -ENOENT; 676178260b2SMiao Xie } 677a4abeea4SJosef Bacik } while (ret == -EBUSY); 678a4abeea4SJosef Bacik 679a43f7f82SLiu Bo if (ret < 0) 680843fcf35SMiao Xie goto join_fail; 6810f7d52f4SChris Mason 6820b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 683a22285a6SYan, Zheng 684a22285a6SYan, Zheng h->transid = cur_trans->transid; 685a22285a6SYan, Zheng h->transaction = cur_trans; 686b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 68764b63580SJeff Mahoney h->fs_info = root->fs_info; 6887174109cSQu Wenruo 689a698d075SMiao Xie h->type = type; 690ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 691b7ec40d7SChris Mason 692a22285a6SYan, Zheng smp_mb(); 6933296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 6942ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 695abdd2e80SFilipe Manana current->journal_info = h; 6963a45bb20SJeff Mahoney btrfs_commit_transaction(h); 697a22285a6SYan, Zheng goto again; 698a22285a6SYan, Zheng } 6999ed74f2dSJosef Bacik 700b5009945SJosef Bacik if (num_bytes) { 7010b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7022bcc0328SLiu Bo h->transid, num_bytes, 1); 7030b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 704b5009945SJosef Bacik h->bytes_reserved = num_bytes; 70520dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 706a22285a6SYan, Zheng } 707a22285a6SYan, Zheng 7082a1eb461SJosef Bacik got_it: 709bcf3a3e7SNikolay Borisov if (!current->journal_info) 710a22285a6SYan, Zheng current->journal_info = h; 711fcc99734SQu Wenruo 712fcc99734SQu Wenruo /* 7139c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7149c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7159c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7169c343784SJosef Bacik * value here. 7179c343784SJosef Bacik */ 7189c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7199c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7209c343784SJosef Bacik 7219c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7229c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7239c343784SJosef Bacik } 7249c343784SJosef Bacik 7259c343784SJosef Bacik /* 726fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 727fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 728fcc99734SQu Wenruo * transaction. 729fcc99734SQu Wenruo * 730fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 731fcc99734SQu Wenruo * or we can deadlock. 732fcc99734SQu Wenruo */ 73368075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 73468075ea8SJosef Bacik if (ret) { 73568075ea8SJosef Bacik /* 73668075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 73768075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 73868075ea8SJosef Bacik * not just freed. 73968075ea8SJosef Bacik */ 74068075ea8SJosef Bacik btrfs_end_transaction(h); 74168075ea8SJosef Bacik return ERR_PTR(ret); 74268075ea8SJosef Bacik } 743fcc99734SQu Wenruo 74479154b1bSChris Mason return h; 745843fcf35SMiao Xie 746843fcf35SMiao Xie join_fail: 7470860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7480b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 749843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 750843fcf35SMiao Xie alloc_fail: 751843fcf35SMiao Xie if (num_bytes) 7522ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 75363f018beSNikolay Borisov num_bytes, NULL); 754843fcf35SMiao Xie reserve_fail: 755733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 756843fcf35SMiao Xie return ERR_PTR(ret); 75779154b1bSChris Mason } 75879154b1bSChris Mason 759f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7605aed1dd8SAlexandru Moise unsigned int num_items) 761f9295749SChris Mason { 76208e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 763003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 764f9295749SChris Mason } 765003d7c59SJeff Mahoney 7668eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7678eab77ffSFilipe Manana struct btrfs_root *root, 7687f9fe614SJosef Bacik unsigned int num_items) 7698eab77ffSFilipe Manana { 7707f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7717f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7728eab77ffSFilipe Manana } 7738407aa46SMiao Xie 7747a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 775f9295749SChris Mason { 776003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 777003d7c59SJeff Mahoney true); 778f9295749SChris Mason } 779f9295749SChris Mason 7808d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7810af3d00bSJosef Bacik { 782575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 783003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7840af3d00bSJosef Bacik } 7850af3d00bSJosef Bacik 786d4edf39bSMiao Xie /* 787a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 788a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 789a6d155d2SFilipe Manana */ 790a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 791a6d155d2SFilipe Manana { 792a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 793a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 794a6d155d2SFilipe Manana } 795a6d155d2SFilipe Manana 796a6d155d2SFilipe Manana /* 797d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 798d4edf39bSMiao Xie * 799d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 800d4edf39bSMiao Xie * don't want to start a new one. 801d4edf39bSMiao Xie * 802d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 803d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 804d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 805d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 806d4edf39bSMiao Xie * invoke 807d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 808d4edf39bSMiao Xie */ 809354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 81060376ce4SJosef Bacik { 811575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 812003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 81360376ce4SJosef Bacik } 81460376ce4SJosef Bacik 815d4edf39bSMiao Xie /* 81690b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 817d4edf39bSMiao Xie * 81852042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 819d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 820d4edf39bSMiao Xie * complete. 821d4edf39bSMiao Xie */ 822d4edf39bSMiao Xie struct btrfs_trans_handle * 823d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 824d4edf39bSMiao Xie { 825d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 826d4edf39bSMiao Xie 827575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 828003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 829*b28ff3a7SFilipe Manana if (trans == ERR_PTR(-ENOENT)) { 830*b28ff3a7SFilipe Manana int ret; 831*b28ff3a7SFilipe Manana 832*b28ff3a7SFilipe Manana ret = btrfs_wait_for_commit(root->fs_info, 0); 833*b28ff3a7SFilipe Manana if (ret) 834*b28ff3a7SFilipe Manana return ERR_PTR(ret); 835*b28ff3a7SFilipe Manana } 836d4edf39bSMiao Xie 837d4edf39bSMiao Xie return trans; 838d4edf39bSMiao Xie } 839d4edf39bSMiao Xie 840d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 841d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 842d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 84389ce8a63SChris Mason { 8445fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8455fd76bf3SOmar Sandoval u64 transid = commit->transid; 8465fd76bf3SOmar Sandoval bool put = false; 8475fd76bf3SOmar Sandoval 8483e738c53SIoannis Angelakopoulos /* 8493e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8503e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8513e738c53SIoannis Angelakopoulos */ 8523e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8533e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8543e738c53SIoannis Angelakopoulos else 8553e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8563e738c53SIoannis Angelakopoulos 8575fd76bf3SOmar Sandoval while (1) { 858d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8595fd76bf3SOmar Sandoval if (put) 8605fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8615fd76bf3SOmar Sandoval 8625fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8635fd76bf3SOmar Sandoval break; 8645fd76bf3SOmar Sandoval 8655fd76bf3SOmar Sandoval /* 8665fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8675fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8685fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8695fd76bf3SOmar Sandoval * transaction. Wait for those. 8705fd76bf3SOmar Sandoval */ 8715fd76bf3SOmar Sandoval 8725fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8735fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8745fd76bf3SOmar Sandoval struct btrfs_transaction, 8755fd76bf3SOmar Sandoval list); 8765fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8775fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8785fd76bf3SOmar Sandoval break; 8795fd76bf3SOmar Sandoval } 8805fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8815fd76bf3SOmar Sandoval put = true; 8825fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8835fd76bf3SOmar Sandoval } 88489ce8a63SChris Mason } 88589ce8a63SChris Mason 8862ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 88746204592SSage Weil { 88846204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8898cd2807fSMiao Xie int ret = 0; 89046204592SSage Weil 89146204592SSage Weil if (transid) { 8920b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 893a4abeea4SJosef Bacik goto out; 89446204592SSage Weil 89546204592SSage Weil /* find specified transaction */ 8960b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8970b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 89846204592SSage Weil if (t->transid == transid) { 89946204592SSage Weil cur_trans = t; 9009b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 9018cd2807fSMiao Xie ret = 0; 90246204592SSage Weil break; 90346204592SSage Weil } 9048cd2807fSMiao Xie if (t->transid > transid) { 9058cd2807fSMiao Xie ret = 0; 90646204592SSage Weil break; 90746204592SSage Weil } 9088cd2807fSMiao Xie } 9090b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 91042383020SSage Weil 91142383020SSage Weil /* 91242383020SSage Weil * The specified transaction doesn't exist, or we 91342383020SSage Weil * raced with btrfs_commit_transaction 91442383020SSage Weil */ 91542383020SSage Weil if (!cur_trans) { 9160b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 91742383020SSage Weil ret = -EINVAL; 9188cd2807fSMiao Xie goto out; 91942383020SSage Weil } 92046204592SSage Weil } else { 92146204592SSage Weil /* find newest transaction that is committing | committed */ 9220b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9230b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 92446204592SSage Weil list) { 9254a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9264a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9273473f3c0SJosef Bacik break; 92846204592SSage Weil cur_trans = t; 9299b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 93046204592SSage Weil break; 93146204592SSage Weil } 93246204592SSage Weil } 9330b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 93446204592SSage Weil if (!cur_trans) 935a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 93646204592SSage Weil } 93746204592SSage Weil 938d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 939bf7ecbe9SFilipe Manana ret = cur_trans->aborted; 940724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 941a4abeea4SJosef Bacik out: 94246204592SSage Weil return ret; 94346204592SSage Weil } 94446204592SSage Weil 9452ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 94637d1aeeeSChris Mason { 9472ff7e61eSJeff Mahoney wait_current_trans(fs_info); 94837d1aeeeSChris Mason } 94937d1aeeeSChris Mason 950a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9518929ecfaSYan, Zheng { 9528929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9538929ecfaSYan, Zheng 9543296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 955e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 956a2633b6aSNikolay Borisov return true; 9578929ecfaSYan, Zheng 95804fb3285SFilipe Manana if (btrfs_check_space_for_delayed_refs(trans->fs_info)) 95904fb3285SFilipe Manana return true; 96004fb3285SFilipe Manana 96104fb3285SFilipe Manana return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50); 9628929ecfaSYan, Zheng } 9638929ecfaSYan, Zheng 964dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 965dc60c525SNikolay Borisov 9660e34693fSNikolay Borisov { 967dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 968dc60c525SNikolay Borisov 9690e34693fSNikolay Borisov if (!trans->block_rsv) { 9700e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9710e34693fSNikolay Borisov return; 9720e34693fSNikolay Borisov } 9730e34693fSNikolay Borisov 9740e34693fSNikolay Borisov if (!trans->bytes_reserved) 9750e34693fSNikolay Borisov return; 9760e34693fSNikolay Borisov 9770e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9780e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9790e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9800e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 98163f018beSNikolay Borisov trans->bytes_reserved, NULL); 9820e34693fSNikolay Borisov trans->bytes_reserved = 0; 9830e34693fSNikolay Borisov } 9840e34693fSNikolay Borisov 98589ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9863a45bb20SJeff Mahoney int throttle) 98779154b1bSChris Mason { 9883a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9898929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9904edc2ca3SDave Jones int err = 0; 991d6e4a428SChris Mason 992b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 993b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9942a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9952a1eb461SJosef Bacik return 0; 9962a1eb461SJosef Bacik } 9972a1eb461SJosef Bacik 998dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9994c13d758SJosef Bacik trans->block_rsv = NULL; 1000c5567237SArne Jansen 10016c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 1002ea658badSJosef Bacik 10034fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 10044fbcdf66SFilipe Manana 10050860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 10060b246afaSJeff Mahoney sb_end_intwrite(info->sb); 10076df7881aSJosef Bacik 10088929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 100913c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 101013c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10110860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 101289ce8a63SChris Mason 1013093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1014e1489b4fSIoannis Angelakopoulos 10155a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1016e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1017e1489b4fSIoannis Angelakopoulos 1018724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10199ed74f2dSJosef Bacik 10209ed74f2dSJosef Bacik if (current->journal_info == trans) 10219ed74f2dSJosef Bacik current->journal_info = NULL; 1022ab78c84dSChris Mason 102324bbcf04SYan, Zheng if (throttle) 10242ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 102524bbcf04SYan, Zheng 102684961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10274e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1028fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1029fbabd4a3SJosef Bacik err = trans->aborted; 1030fbabd4a3SJosef Bacik else 1031fbabd4a3SJosef Bacik err = -EROFS; 10324e121c06SJosef Bacik } 103349b25e05SJeff Mahoney 10344edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10354edc2ca3SDave Jones return err; 103679154b1bSChris Mason } 103779154b1bSChris Mason 10383a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 103989ce8a63SChris Mason { 10403a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 104189ce8a63SChris Mason } 104289ce8a63SChris Mason 10433a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 104489ce8a63SChris Mason { 10453a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 104616cdcec7SMiao Xie } 104716cdcec7SMiao Xie 1048d352ac68SChris Mason /* 1049d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1050d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1051690587d1SChris Mason * those extents are sent to disk but does not wait on them 1052d352ac68SChris Mason */ 10532ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10548cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 105579154b1bSChris Mason { 1056777e6bd7SChris Mason int err = 0; 10577c4452b9SChris Mason int werr = 0; 10580b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1059e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1060777e6bd7SChris Mason u64 start = 0; 10615f39d397SChris Mason u64 end; 10627c4452b9SChris Mason 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 } 1098690587d1SChris Mason return werr; 1099690587d1SChris Mason } 1100690587d1SChris Mason 1101690587d1SChris Mason /* 1102690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1103690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1104690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1105690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1106690587d1SChris Mason */ 1107bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1108bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1109690587d1SChris Mason { 1110690587d1SChris Mason int err = 0; 1111690587d1SChris Mason int werr = 0; 11120b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1113e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1114690587d1SChris Mason u64 start = 0; 1115690587d1SChris Mason u64 end; 1116690587d1SChris Mason 11171728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1118e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1119663dfbb0SFilipe Manana /* 1120663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1121663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1122663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1123663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1124663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 112541e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1126663dfbb0SFilipe Manana */ 1127663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1128bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1129663dfbb0SFilipe Manana if (err == -ENOMEM) 1130663dfbb0SFilipe Manana err = 0; 1131663dfbb0SFilipe Manana if (!err) 11321728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1133777e6bd7SChris Mason if (err) 1134777e6bd7SChris Mason werr = err; 1135e38e2ed7SFilipe Manana free_extent_state(cached_state); 1136e38e2ed7SFilipe Manana cached_state = NULL; 1137777e6bd7SChris Mason cond_resched(); 11381728366eSJosef Bacik start = end + 1; 1139777e6bd7SChris Mason } 11407c4452b9SChris Mason if (err) 11417c4452b9SChris Mason werr = err; 1142bf89d38fSJeff Mahoney return werr; 1143bf89d38fSJeff Mahoney } 1144656f30dbSFilipe Manana 1145b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1146bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1147bf89d38fSJeff Mahoney { 1148bf89d38fSJeff Mahoney bool errors = false; 1149bf89d38fSJeff Mahoney int err; 1150bf89d38fSJeff Mahoney 1151bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1152bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1153bf89d38fSJeff Mahoney errors = true; 1154bf89d38fSJeff Mahoney 1155bf89d38fSJeff Mahoney if (errors && !err) 1156bf89d38fSJeff Mahoney err = -EIO; 1157bf89d38fSJeff Mahoney return err; 1158bf89d38fSJeff Mahoney } 1159bf89d38fSJeff Mahoney 1160bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1161bf89d38fSJeff Mahoney { 1162bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1163bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1164bf89d38fSJeff Mahoney bool errors = false; 1165bf89d38fSJeff Mahoney int err; 1166bf89d38fSJeff Mahoney 1167bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1168bf89d38fSJeff Mahoney 1169bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1170656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11710b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1172656f30dbSFilipe Manana errors = true; 1173656f30dbSFilipe Manana 1174656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11750b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1176656f30dbSFilipe Manana errors = true; 1177656f30dbSFilipe Manana 1178bf89d38fSJeff Mahoney if (errors && !err) 1179bf89d38fSJeff Mahoney err = -EIO; 1180bf89d38fSJeff Mahoney return err; 118179154b1bSChris Mason } 118279154b1bSChris Mason 1183690587d1SChris Mason /* 1184c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1185c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1186c9b577c0SNikolay Borisov * log commit. 1187c9b577c0SNikolay Borisov * 1188c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1189690587d1SChris Mason */ 119070458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1191d0c803c4SChris Mason { 1192663dfbb0SFilipe Manana int ret; 1193c9b577c0SNikolay Borisov int ret2; 1194c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 119570458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1196c9b577c0SNikolay Borisov struct blk_plug plug; 1197663dfbb0SFilipe Manana 1198c9b577c0SNikolay Borisov blk_start_plug(&plug); 1199c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1200c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1201c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1202c9b577c0SNikolay Borisov 120341e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1204663dfbb0SFilipe Manana 1205c9b577c0SNikolay Borisov if (ret) 1206663dfbb0SFilipe Manana return ret; 1207c9b577c0SNikolay Borisov else if (ret2) 1208c9b577c0SNikolay Borisov return ret2; 1209c9b577c0SNikolay Borisov else 1210c9b577c0SNikolay Borisov return 0; 1211d0c803c4SChris Mason } 1212d0c803c4SChris Mason 1213d352ac68SChris Mason /* 1214d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1215d352ac68SChris Mason * 1216d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1217d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1218d352ac68SChris Mason * allocation tree. 1219d352ac68SChris Mason * 1220d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1221d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1222d352ac68SChris Mason */ 12230b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 122479154b1bSChris Mason struct btrfs_root *root) 122579154b1bSChris Mason { 122679154b1bSChris Mason int ret; 12270b86a832SChris Mason u64 old_root_bytenr; 122886b9f2ecSYan, Zheng u64 old_root_used; 12290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12300b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 123179154b1bSChris Mason 123286b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 123356bec294SChris Mason 123479154b1bSChris Mason while (1) { 12350b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 123686b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1237ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 123879154b1bSChris Mason break; 123987ef2bb4SChris Mason 12405d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 124179154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12420b86a832SChris Mason &root->root_key, 12430b86a832SChris Mason &root->root_item); 124449b25e05SJeff Mahoney if (ret) 124549b25e05SJeff Mahoney return ret; 124656bec294SChris Mason 124786b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1248e7070be1SJosef Bacik } 1249276e680dSYan Zheng 12500b86a832SChris Mason return 0; 12510b86a832SChris Mason } 12520b86a832SChris Mason 1253d352ac68SChris Mason /* 1254d352ac68SChris Mason * update all the cowonly tree roots on disk 125549b25e05SJeff Mahoney * 125649b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 125749b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 125849b25e05SJeff Mahoney * to clean up the delayed refs. 1259d352ac68SChris Mason */ 12609386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12610b86a832SChris Mason { 12629386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1263ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12641bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12650b86a832SChris Mason struct list_head *next; 126684234f3aSYan Zheng struct extent_buffer *eb; 126756bec294SChris Mason int ret; 126884234f3aSYan Zheng 1269dfba78dcSFilipe Manana /* 1270dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1271dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1272dfba78dcSFilipe Manana */ 1273dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1274dfba78dcSFilipe Manana 127584234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 127649b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12779631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 127884234f3aSYan Zheng btrfs_tree_unlock(eb); 127984234f3aSYan Zheng free_extent_buffer(eb); 12800b86a832SChris Mason 128149b25e05SJeff Mahoney if (ret) 128249b25e05SJeff Mahoney return ret; 128349b25e05SJeff Mahoney 1284196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1285c16ce190SJosef Bacik if (ret) 1286c16ce190SJosef Bacik return ret; 12872b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1288c16ce190SJosef Bacik if (ret) 1289c16ce190SJosef Bacik return ret; 1290280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1291c16ce190SJosef Bacik if (ret) 1292c16ce190SJosef Bacik return ret; 1293546adb0dSJan Schmidt 1294bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1295dcdf7f6dSJosef Bacik if (ret) 1296dcdf7f6dSJosef Bacik return ret; 1297dcdf7f6dSJosef Bacik 1298ea526d18SJosef Bacik again: 12990b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 13002ff7e61eSJeff Mahoney struct btrfs_root *root; 13010b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 13020b86a832SChris Mason list_del_init(next); 13030b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1304e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 130587ef2bb4SChris Mason 13069e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13079e351cc8SJosef Bacik &trans->transaction->switch_commits); 130849b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 130949b25e05SJeff Mahoney if (ret) 131049b25e05SJeff Mahoney return ret; 1311488bc2a2SJosef Bacik } 1312488bc2a2SJosef Bacik 1313488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1314c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1315ea526d18SJosef Bacik if (ret) 1316ea526d18SJosef Bacik return ret; 1317276e680dSYan Zheng 13181bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13195742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1320ea526d18SJosef Bacik if (ret) 1321ea526d18SJosef Bacik return ret; 1322488bc2a2SJosef Bacik 1323488bc2a2SJosef Bacik /* 1324488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1325488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1326488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1327488bc2a2SJosef Bacik * everything gets run. 1328488bc2a2SJosef Bacik */ 1329c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1330ea526d18SJosef Bacik if (ret) 1331ea526d18SJosef Bacik return ret; 1332ea526d18SJosef Bacik } 1333ea526d18SJosef Bacik 1334ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1335ea526d18SJosef Bacik goto again; 1336ea526d18SJosef Bacik 13379f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13389f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13399f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13408dabb742SStefan Behrens 134179154b1bSChris Mason return 0; 134279154b1bSChris Mason } 134379154b1bSChris Mason 1344d352ac68SChris Mason /* 1345b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1346b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1347b4be6aefSJosef Bacik */ 1348b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1349b4be6aefSJosef Bacik { 1350b4be6aefSJosef Bacik /* 1351b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1352b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1353b4be6aefSJosef Bacik * up. 1354b4be6aefSJosef Bacik */ 1355b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1356b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1357b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1358b4be6aefSJosef Bacik struct btrfs_root, 1359b4be6aefSJosef Bacik root_list); 1360b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1361b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1362b4be6aefSJosef Bacik return; 1363b4be6aefSJosef Bacik } 1364b4be6aefSJosef Bacik } 1365b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1366b4be6aefSJosef Bacik 1367b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1368b4be6aefSJosef Bacik } 1369b4be6aefSJosef Bacik 1370b4be6aefSJosef Bacik /* 1371d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1372d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1373d352ac68SChris Mason * be deleted 1374d352ac68SChris Mason */ 1375cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13765eda7b5eSChris Mason { 13770b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13780b246afaSJeff Mahoney 13790b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1380dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1381dc9492c1SJosef Bacik btrfs_grab_root(root); 1382b4be6aefSJosef Bacik 1383b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1384b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1385b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1386b4be6aefSJosef Bacik else 13870b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1388dc9492c1SJosef Bacik } 13890b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13905eda7b5eSChris Mason } 13915eda7b5eSChris Mason 1392d352ac68SChris Mason /* 1393dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1394dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1395d352ac68SChris Mason */ 13967e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13970f7d52f4SChris Mason { 13987e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1399fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1400fc7cbcd4SDavid Sterba int i; 1401fc7cbcd4SDavid Sterba int ret; 140254aa1f4dSChris Mason 1403dfba78dcSFilipe Manana /* 1404dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1405dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1406dfba78dcSFilipe Manana */ 1407dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1408dfba78dcSFilipe Manana 1409fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1410fc7cbcd4SDavid Sterba while (1) { 1411fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1412fc7cbcd4SDavid Sterba (void **)gang, 0, 1413fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1414fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1415fc7cbcd4SDavid Sterba if (ret == 0) 1416fc7cbcd4SDavid Sterba break; 1417fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1418fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1419fc7cbcd4SDavid Sterba int ret2; 14204f4317c1SJosef Bacik 1421dfba78dcSFilipe Manana /* 1422dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1423dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1424dfba78dcSFilipe Manana */ 1425dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1426dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1427dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1428dfba78dcSFilipe Manana 1429fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14302619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14310f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1432fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 143331153d81SYan Zheng 1434e02119d5SChris Mason btrfs_free_log(trans, root); 1435fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1436fc7cbcd4SDavid Sterba if (ret2) 1437fc7cbcd4SDavid Sterba return ret2; 1438e02119d5SChris Mason 1439fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 144027cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1441c7548af6SChris Mason smp_mb__after_atomic(); 1442f1ebcc74SLiu Bo 1443978d910dSYan Zheng if (root->commit_root != root->node) { 14449e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14459e351cc8SJosef Bacik &trans->transaction->switch_commits); 1446fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1447fc7cbcd4SDavid Sterba root->node); 1448978d910dSYan Zheng } 144931153d81SYan Zheng 1450fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1451fc7cbcd4SDavid Sterba &root->root_key, 1452fc7cbcd4SDavid Sterba &root->root_item); 1453fc7cbcd4SDavid Sterba if (ret2) 1454fc7cbcd4SDavid Sterba return ret2; 1455fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1456733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14570f7d52f4SChris Mason } 1458fc7cbcd4SDavid Sterba } 1459fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14604f4317c1SJosef Bacik return 0; 14610f7d52f4SChris Mason } 14620f7d52f4SChris Mason 1463d352ac68SChris Mason /* 1464de78b51aSEric Sandeen * defrag a given btree. 1465de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1466d352ac68SChris Mason */ 1467de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1468e9d0b13bSChris Mason { 1469e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1470e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14718929ecfaSYan, Zheng int ret; 1472e9d0b13bSChris Mason 147327cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1474e9d0b13bSChris Mason return 0; 14758929ecfaSYan, Zheng 14766b80053dSChris Mason while (1) { 14778929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14786819703fSDavid Sterba if (IS_ERR(trans)) { 14796819703fSDavid Sterba ret = PTR_ERR(trans); 14806819703fSDavid Sterba break; 14816819703fSDavid Sterba } 14828929ecfaSYan, Zheng 1483de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14848929ecfaSYan, Zheng 14853a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14862ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1487e9d0b13bSChris Mason cond_resched(); 1488e9d0b13bSChris Mason 1489ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1490e9d0b13bSChris Mason break; 1491210549ebSDavid Sterba 1492ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1493ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1494210549ebSDavid Sterba ret = -EAGAIN; 1495210549ebSDavid Sterba break; 1496210549ebSDavid Sterba } 1497e9d0b13bSChris Mason } 149827cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14998929ecfaSYan, Zheng return ret; 1500e9d0b13bSChris Mason } 1501e9d0b13bSChris Mason 1502d352ac68SChris Mason /* 15036426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 15046426c7adSQu Wenruo * 15056426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 15066426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 15076426c7adSQu Wenruo * qgroup works. 15086426c7adSQu Wenruo */ 15096426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15106426c7adSQu Wenruo struct btrfs_root *src, 15116426c7adSQu Wenruo struct btrfs_root *parent, 15126426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15136426c7adSQu Wenruo u64 dst_objectid) 15146426c7adSQu Wenruo { 15156426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15166426c7adSQu Wenruo int ret; 15176426c7adSQu Wenruo 15186426c7adSQu Wenruo /* 15196426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15206426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15216426c7adSQu Wenruo * kick in anyway. 15226426c7adSQu Wenruo */ 15239ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15246426c7adSQu Wenruo return 0; 15256426c7adSQu Wenruo 15266426c7adSQu Wenruo /* 152752042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15284d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15294d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15304d31778aSQu Wenruo * item. 15314d31778aSQu Wenruo */ 15321c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15331c442d22SJosef Bacik if (ret) 15341c442d22SJosef Bacik return ret; 15354d31778aSQu Wenruo 15364d31778aSQu Wenruo /* 15372a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15382a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15392a4d84c1SJosef Bacik * 15402a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15412a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15422a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15432a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15442a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15452a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15462a4d84c1SJosef Bacik */ 15472a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15482a4d84c1SJosef Bacik if (ret) { 15492a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 155044365827SNaohiro Aota return ret; 15512a4d84c1SJosef Bacik } 15522a4d84c1SJosef Bacik 15537e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15546426c7adSQu Wenruo if (ret) 15556426c7adSQu Wenruo goto out; 1556460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15576426c7adSQu Wenruo if (ret < 0) 15586426c7adSQu Wenruo goto out; 15596426c7adSQu Wenruo 15606426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1561a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15626426c7adSQu Wenruo inherit); 15636426c7adSQu Wenruo if (ret < 0) 15646426c7adSQu Wenruo goto out; 15656426c7adSQu Wenruo 15666426c7adSQu Wenruo /* 15676426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15686426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15696426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15706426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15716426c7adSQu Wenruo * 2) write all btree blocks onto disk 15726426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15736426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15746426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15756426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15766426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15776426c7adSQu Wenruo */ 15789386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15796426c7adSQu Wenruo if (ret) 15806426c7adSQu Wenruo goto out; 1581889bfa39SJosef Bacik switch_commit_roots(trans); 158270458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15836426c7adSQu Wenruo if (ret) 1584f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15856426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15866426c7adSQu Wenruo 15876426c7adSQu Wenruo out: 15886426c7adSQu Wenruo /* 15896426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15906426c7adSQu Wenruo * last_trans == cur_transid. 15916426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15926426c7adSQu Wenruo * insert_dir_item() 15936426c7adSQu Wenruo */ 15946426c7adSQu Wenruo if (!ret) 15951c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15966426c7adSQu Wenruo return ret; 15976426c7adSQu Wenruo } 15986426c7adSQu Wenruo 15996426c7adSQu Wenruo /* 1600d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1601aec8030aSMiao Xie * transaction commit. This does the actual creation. 1602aec8030aSMiao Xie * 1603aec8030aSMiao Xie * Note: 1604aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1605aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1606aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1607d352ac68SChris Mason */ 160880b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16093063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16103063d29fSChris Mason { 161108d50ca3SNikolay Borisov 161208d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16133063d29fSChris Mason struct btrfs_key key; 161480b6794dSChris Mason struct btrfs_root_item *new_root_item; 16153063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16163063d29fSChris Mason struct btrfs_root *root = pending->root; 16176bdb72deSSage Weil struct btrfs_root *parent_root; 161898c9942aSLiu Bo struct btrfs_block_rsv *rsv; 1619ab3c5c18SSweet Tea Dorminy struct inode *parent_inode = pending->dir; 162042874b3dSMiao Xie struct btrfs_path *path; 162142874b3dSMiao Xie struct btrfs_dir_item *dir_item; 16223063d29fSChris Mason struct extent_buffer *tmp; 1623925baeddSChris Mason struct extent_buffer *old; 162495582b00SDeepa Dinamani struct timespec64 cur_time; 1625aec8030aSMiao Xie int ret = 0; 1626d68fc57bSYan, Zheng u64 to_reserve = 0; 16276bdb72deSSage Weil u64 index = 0; 1628a22285a6SYan, Zheng u64 objectid; 1629b83cc969SLi Zefan u64 root_flags; 1630ab3c5c18SSweet Tea Dorminy unsigned int nofs_flags; 1631ab3c5c18SSweet Tea Dorminy struct fscrypt_name fname; 16323063d29fSChris Mason 16338546b570SDavid Sterba ASSERT(pending->path); 16348546b570SDavid Sterba path = pending->path; 163542874b3dSMiao Xie 1636b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1637b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1638a22285a6SYan, Zheng 1639ab3c5c18SSweet Tea Dorminy /* 1640ab3c5c18SSweet Tea Dorminy * We're inside a transaction and must make sure that any potential 1641ab3c5c18SSweet Tea Dorminy * allocations with GFP_KERNEL in fscrypt won't recurse back to 1642ab3c5c18SSweet Tea Dorminy * filesystem. 1643ab3c5c18SSweet Tea Dorminy */ 1644ab3c5c18SSweet Tea Dorminy nofs_flags = memalloc_nofs_save(); 1645ab3c5c18SSweet Tea Dorminy pending->error = fscrypt_setup_filename(parent_inode, 1646ab3c5c18SSweet Tea Dorminy &pending->dentry->d_name, 0, 1647ab3c5c18SSweet Tea Dorminy &fname); 1648ab3c5c18SSweet Tea Dorminy memalloc_nofs_restore(nofs_flags); 1649ab3c5c18SSweet Tea Dorminy if (pending->error) 1650ab3c5c18SSweet Tea Dorminy goto free_pending; 1651ab3c5c18SSweet Tea Dorminy 1652543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1653aec8030aSMiao Xie if (pending->error) 1654ab3c5c18SSweet Tea Dorminy goto free_fname; 16553063d29fSChris Mason 1656d6726335SQu Wenruo /* 1657d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1658d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1659d6726335SQu Wenruo */ 1660d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1661d6726335SQu Wenruo 1662147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1663d68fc57bSYan, Zheng 1664d68fc57bSYan, Zheng if (to_reserve > 0) { 16659270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1666aec8030aSMiao Xie &pending->block_rsv, 166708e007d2SMiao Xie to_reserve, 166808e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1669aec8030aSMiao Xie if (pending->error) 1670d6726335SQu Wenruo goto clear_skip_qgroup; 1671d68fc57bSYan, Zheng } 1672d68fc57bSYan, Zheng 16733063d29fSChris Mason key.objectid = objectid; 1674a22285a6SYan, Zheng key.offset = (u64)-1; 1675a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16763063d29fSChris Mason 16776fa9700eSMiao Xie rsv = trans->block_rsv; 1678a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16792382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16800b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 168188d3a5aaSJosef Bacik trans->transid, 168288d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1683a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1684f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1685f0118cb6SJosef Bacik if (ret) 1686f0118cb6SJosef Bacik goto fail; 1687c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 168804b285f3SDeepa Dinamani 16896bdb72deSSage Weil /* 16906bdb72deSSage Weil * insert the directory item 16916bdb72deSSage Weil */ 1692877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 1693df9f2782SFilipe Manana if (ret) { 1694df9f2782SFilipe Manana btrfs_abort_transaction(trans, ret); 1695df9f2782SFilipe Manana goto fail; 1696df9f2782SFilipe Manana } 169742874b3dSMiao Xie 169842874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 169942874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 17004a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 17016db75318SSweet Tea Dorminy &fname.disk_name, 0); 170242874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1703fe66a05aSChris Mason pending->error = -EEXIST; 1704aec8030aSMiao Xie goto dir_item_existed; 170542874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 170642874b3dSMiao Xie ret = PTR_ERR(dir_item); 170766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17088732d44fSMiao Xie goto fail; 170979787eaaSJeff Mahoney } 171042874b3dSMiao Xie btrfs_release_path(path); 17116bdb72deSSage Weil 1712e999376fSChris Mason /* 1713e999376fSChris Mason * pull in the delayed directory update 1714e999376fSChris Mason * and the delayed inode item 1715e999376fSChris Mason * otherwise we corrupt the FS during 1716e999376fSChris Mason * snapshot 1717e999376fSChris Mason */ 1718e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 17198732d44fSMiao Xie if (ret) { /* Transaction aborted */ 172066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17218732d44fSMiao Xie goto fail; 17228732d44fSMiao Xie } 1723e999376fSChris Mason 1724f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1725f0118cb6SJosef Bacik if (ret) { 1726f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1727f0118cb6SJosef Bacik goto fail; 1728f0118cb6SJosef Bacik } 17296bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17306bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 173108fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17326bdb72deSSage Weil 1733b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1734b83cc969SLi Zefan if (pending->readonly) 1735b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1736b83cc969SLi Zefan else 1737b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1738b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1739b83cc969SLi Zefan 17408ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17418ea05e3aSAlexander Block trans->transid); 1742807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17438ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17448ea05e3aSAlexander Block BTRFS_UUID_SIZE); 174570023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 174670023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 174770023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17488ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17498ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17508ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17518ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 175270023da2SStefan Behrens } 17533cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17543cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 175570023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17568ea05e3aSAlexander Block 1757925baeddSChris Mason old = btrfs_lock_root_node(root); 17589631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17599631e4ccSJosef Bacik BTRFS_NESTING_COW); 176079787eaaSJeff Mahoney if (ret) { 176179787eaaSJeff Mahoney btrfs_tree_unlock(old); 176279787eaaSJeff Mahoney free_extent_buffer(old); 176366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17648732d44fSMiao Xie goto fail; 176579787eaaSJeff Mahoney } 176649b25e05SJeff Mahoney 176749b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 176879787eaaSJeff Mahoney /* clean up in any case */ 1769925baeddSChris Mason btrfs_tree_unlock(old); 1770925baeddSChris Mason free_extent_buffer(old); 17718732d44fSMiao Xie if (ret) { 177266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17738732d44fSMiao Xie goto fail; 17748732d44fSMiao Xie } 1775f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 177627cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1777f1ebcc74SLiu Bo smp_wmb(); 1778f1ebcc74SLiu Bo 17795d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1780a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1781a22285a6SYan, Zheng key.offset = trans->transid; 1782a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1783925baeddSChris Mason btrfs_tree_unlock(tmp); 17843063d29fSChris Mason free_extent_buffer(tmp); 17858732d44fSMiao Xie if (ret) { 178666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17878732d44fSMiao Xie goto fail; 17888732d44fSMiao Xie } 17890660b5afSChris Mason 1790a22285a6SYan, Zheng /* 1791a22285a6SYan, Zheng * insert root back/forward references 1792a22285a6SYan, Zheng */ 17936025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1794a22285a6SYan, Zheng parent_root->root_key.objectid, 17954a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 17966db75318SSweet Tea Dorminy &fname.disk_name); 17978732d44fSMiao Xie if (ret) { 179866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17998732d44fSMiao Xie goto fail; 18008732d44fSMiao Xie } 1801a22285a6SYan, Zheng 1802a22285a6SYan, Zheng key.offset = (u64)-1; 18032dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 180479787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 180579787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 18062d892ccdSFilipe Manana pending->snap = NULL; 180766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18088732d44fSMiao Xie goto fail; 180979787eaaSJeff Mahoney } 1810d68fc57bSYan, Zheng 181149b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 18128732d44fSMiao Xie if (ret) { 181366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18148732d44fSMiao Xie goto fail; 18158732d44fSMiao Xie } 1816361048f5SMiao Xie 18176426c7adSQu Wenruo /* 18186426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 18196426c7adSQu Wenruo * snapshot hack to do fast snapshot. 18206426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 18216426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 18226426c7adSQu Wenruo */ 18236426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18246426c7adSQu Wenruo pending->inherit, objectid); 18256426c7adSQu Wenruo if (ret < 0) 18266426c7adSQu Wenruo goto fail; 18276426c7adSQu Wenruo 18286db75318SSweet Tea Dorminy ret = btrfs_insert_dir_item(trans, &fname.disk_name, 18296db75318SSweet Tea Dorminy BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, 18306db75318SSweet Tea Dorminy index); 183142874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18329c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18338732d44fSMiao Xie if (ret) { 183466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18358732d44fSMiao Xie goto fail; 18368732d44fSMiao Xie } 183742874b3dSMiao Xie 18386ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 18396db75318SSweet Tea Dorminy fname.disk_name.len * 2); 1840c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1841c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1842729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1843dd5f9615SStefan Behrens if (ret) { 184466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1845dd5f9615SStefan Behrens goto fail; 1846dd5f9615SStefan Behrens } 1847807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1848807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1849cdb345a8SLu Fengqi objectid); 1850dd5f9615SStefan Behrens if (ret) { 185166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1852dd5f9615SStefan Behrens goto fail; 1853dd5f9615SStefan Behrens } 1854dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1855cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1856dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1857dd5f9615SStefan Behrens objectid); 1858dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 185966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1860dd5f9615SStefan Behrens goto fail; 1861dd5f9615SStefan Behrens } 1862dd5f9615SStefan Behrens } 1863d6726335SQu Wenruo 18643063d29fSChris Mason fail: 1865aec8030aSMiao Xie pending->error = ret; 1866aec8030aSMiao Xie dir_item_existed: 186798c9942aSLiu Bo trans->block_rsv = rsv; 18682382c5ccSLiu Bo trans->bytes_reserved = 0; 1869d6726335SQu Wenruo clear_skip_qgroup: 1870d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 1871ab3c5c18SSweet Tea Dorminy free_fname: 1872ab3c5c18SSweet Tea Dorminy fscrypt_free_filename(&fname); 1873ab3c5c18SSweet Tea Dorminy free_pending: 18746fa9700eSMiao Xie kfree(new_root_item); 1875b0c0ea63SDavid Sterba pending->root_item = NULL; 187642874b3dSMiao Xie btrfs_free_path(path); 18778546b570SDavid Sterba pending->path = NULL; 18788546b570SDavid Sterba 187949b25e05SJeff Mahoney return ret; 18803063d29fSChris Mason } 18813063d29fSChris Mason 1882d352ac68SChris Mason /* 1883d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1884d352ac68SChris Mason */ 188508d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18863063d29fSChris Mason { 1887aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18883063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1889aec8030aSMiao Xie int ret = 0; 18903de4586cSChris Mason 1891aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1892aec8030aSMiao Xie list_del(&pending->list); 189308d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1894aec8030aSMiao Xie if (ret) 1895aec8030aSMiao Xie break; 1896aec8030aSMiao Xie } 1897aec8030aSMiao Xie return ret; 18983de4586cSChris Mason } 18993de4586cSChris Mason 19002ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 19015d4f98a2SYan Zheng { 19025d4f98a2SYan Zheng struct btrfs_root_item *root_item; 19035d4f98a2SYan Zheng struct btrfs_super_block *super; 19045d4f98a2SYan Zheng 19050b246afaSJeff Mahoney super = fs_info->super_copy; 19065d4f98a2SYan Zheng 19070b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1908093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1909093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1910093e037cSDavid Sterba super->chunk_root_level = root_item->level; 19115d4f98a2SYan Zheng 19120b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1913093e037cSDavid Sterba super->root = root_item->bytenr; 1914093e037cSDavid Sterba super->generation = root_item->generation; 1915093e037cSDavid Sterba super->root_level = root_item->level; 19160b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1917093e037cSDavid Sterba super->cache_generation = root_item->generation; 191894846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 191994846229SBoris Burkov super->cache_generation = 0; 19200b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1921093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 19225d4f98a2SYan Zheng } 19235d4f98a2SYan Zheng 1924f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1925f36f3042SChris Mason { 19264a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1927f36f3042SChris Mason int ret = 0; 19284a9d8bdeSMiao Xie 1929a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19304a9d8bdeSMiao Xie trans = info->running_transaction; 19314a9d8bdeSMiao Xie if (trans) 19324a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1933a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1934f36f3042SChris Mason return ret; 1935f36f3042SChris Mason } 1936f36f3042SChris Mason 19378929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19388929ecfaSYan, Zheng { 19394a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19408929ecfaSYan, Zheng int ret = 0; 19414a9d8bdeSMiao Xie 1942a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19434a9d8bdeSMiao Xie trans = info->running_transaction; 19444a9d8bdeSMiao Xie if (trans) 19454a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1946a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19478929ecfaSYan, Zheng return ret; 19488929ecfaSYan, Zheng } 19498929ecfaSYan, Zheng 1950fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1951bb9c12c9SSage Weil { 19523a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1953bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1954bb9c12c9SSage Weil 1955fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1956fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1957fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1958bb9c12c9SSage Weil 1959bb9c12c9SSage Weil /* take transaction reference */ 1960bb9c12c9SSage Weil cur_trans = trans->transaction; 19619b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1962bb9c12c9SSage Weil 19633a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19646fc4e354SSage Weil 19656fc4e354SSage Weil /* 1966ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1967ae5d29d4SDavid Sterba * subsequent transaction joins 1968ae5d29d4SDavid Sterba */ 19693e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1970ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1971ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1972ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1973724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1974bb9c12c9SSage Weil } 1975bb9c12c9SSage Weil 197697cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 197749b25e05SJeff Mahoney { 197897cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 197949b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 198049b25e05SJeff Mahoney 1981b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 198249b25e05SJeff Mahoney 198366642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19847b8b92afSJosef Bacik 19850b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 198666b6135bSLiu Bo 198725d8c284SMiao Xie /* 198825d8c284SMiao Xie * If the transaction is removed from the list, it means this 198925d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 199025d8c284SMiao Xie * to call the cleanup function. 199125d8c284SMiao Xie */ 199225d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 199366b6135bSLiu Bo 19940b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19954a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19960b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1997e1489b4fSIoannis Angelakopoulos 1998e1489b4fSIoannis Angelakopoulos /* 1999e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 2000e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 2001e1489b4fSIoannis Angelakopoulos */ 2002e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2003f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 2004f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 2005f094ac32SLiu Bo 20060b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 2007d7096fc3SJosef Bacik } 2008061dde82SFilipe Manana 2009061dde82SFilipe Manana /* 2010061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 2011061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 2012061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 2013061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 2014061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 2015061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 2016061dde82SFilipe Manana */ 2017061dde82SFilipe Manana list_del_init(&cur_trans->list); 2018061dde82SFilipe Manana 20190b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 202049b25e05SJeff Mahoney 20212ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 202249b25e05SJeff Mahoney 20230b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20240b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20250b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20260b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20274a9d8bdeSMiao Xie 2028e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20290b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2030724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2031724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 203249b25e05SJeff Mahoney 20332e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 203449b25e05SJeff Mahoney 203549b25e05SJeff Mahoney if (current->journal_info == trans) 203649b25e05SJeff Mahoney current->journal_info = NULL; 20372d82a40aSFilipe Manana 20382d82a40aSFilipe Manana /* 20392d82a40aSFilipe Manana * If relocation is running, we can't cancel scrub because that will 20402d82a40aSFilipe Manana * result in a deadlock. Before relocating a block group, relocation 20412d82a40aSFilipe Manana * pauses scrub, then starts and commits a transaction before unpausing 20422d82a40aSFilipe Manana * scrub. If the transaction commit is being done by the relocation 20432d82a40aSFilipe Manana * task or triggered by another task and the relocation task is waiting 20442d82a40aSFilipe Manana * for the commit, and we end up here due to an error in the commit 20452d82a40aSFilipe Manana * path, then calling btrfs_scrub_cancel() will deadlock, as we are 20462d82a40aSFilipe Manana * asking for scrub to stop while having it asked to be paused higher 20472d82a40aSFilipe Manana * above in relocation code. 20482d82a40aSFilipe Manana */ 20492d82a40aSFilipe Manana if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 20500b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 205149b25e05SJeff Mahoney 205249b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 205349b25e05SJeff Mahoney } 205449b25e05SJeff Mahoney 2055c7cc64a9SDavid Sterba /* 2056c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2057c7cc64a9SDavid Sterba * transaction and remove them from the list 2058c7cc64a9SDavid Sterba */ 2059c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2060c7cc64a9SDavid Sterba { 2061c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 206232da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2063c7cc64a9SDavid Sterba 2064c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2065c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2066c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2067c7cc64a9SDavid Sterba } 2068c7cc64a9SDavid Sterba } 2069c7cc64a9SDavid Sterba 207088090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 207182436617SMiao Xie { 2072ce8ea7ccSJosef Bacik /* 2073a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2074ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2075ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2076ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2077ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2078ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2079ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2080a0f0cf83SFilipe Manana * 2081a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2082a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2083a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2084a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2085a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2086a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2087a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2088ce8ea7ccSJosef Bacik */ 208988090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2090a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 209182436617SMiao Xie return 0; 209282436617SMiao Xie } 209382436617SMiao Xie 209488090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 209582436617SMiao Xie { 209688090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20976374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 209882436617SMiao Xie } 209982436617SMiao Xie 210028b21c55SFilipe Manana /* 210128b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 210228b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 210328b21c55SFilipe Manana * and while holding fs_info->trans_lock. 210428b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 210528b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 210628b21c55SFilipe Manana * returns an error. 210728b21c55SFilipe Manana */ 210828b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 210928b21c55SFilipe Manana { 211028b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 211128b21c55SFilipe Manana 211228b21c55SFilipe Manana if (!trans->pending_snapshot) 211328b21c55SFilipe Manana return; 211428b21c55SFilipe Manana 211528b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 211628b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 211728b21c55SFilipe Manana 211828b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 211928b21c55SFilipe Manana } 212028b21c55SFilipe Manana 2121e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2122e55958c8SIoannis Angelakopoulos { 2123e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2124e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2125e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2126e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2127e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2128e55958c8SIoannis Angelakopoulos } 2129e55958c8SIoannis Angelakopoulos 21303a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 213179154b1bSChris Mason { 21323a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 213349b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 21348fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 213525287e0aSMiao Xie int ret; 2136e55958c8SIoannis Angelakopoulos ktime_t start_time; 2137e55958c8SIoannis Angelakopoulos ktime_t interval; 213879154b1bSChris Mason 213935b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21403e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 214135b814f3SNikolay Borisov 2142c52cc7b7SJosef Bacik clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); 2143c52cc7b7SJosef Bacik 21448d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2145bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 214625287e0aSMiao Xie ret = cur_trans->aborted; 21473e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 214825287e0aSMiao Xie } 214949b25e05SJeff Mahoney 2150f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2151f45c752bSJosef Bacik trans->block_rsv = NULL; 2152f45c752bSJosef Bacik 2153e19eb11fSJosef Bacik /* 2154e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2155e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2156e19eb11fSJosef Bacik */ 2157e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2158e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2159e19eb11fSJosef Bacik /* 2160e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2161e19eb11fSJosef Bacik * Any running threads may add more while we are here. 216256bec294SChris Mason */ 2163c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21643e738c53SIoannis Angelakopoulos if (ret) 21653e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2166e19eb11fSJosef Bacik } 216756bec294SChris Mason 21686c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2169ea658badSJosef Bacik 21703204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21711bbc621eSChris Mason int run_it = 0; 21721bbc621eSChris Mason 21731bbc621eSChris Mason /* this mutex is also taken before trying to set 21741bbc621eSChris Mason * block groups readonly. We need to make sure 21751bbc621eSChris Mason * that nobody has set a block group readonly 21761bbc621eSChris Mason * after a extents from that block group have been 21771bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21781bbc621eSChris Mason * will wait for the transaction to commit if it 21793204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21801bbc621eSChris Mason * 21813204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21823204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21831bbc621eSChris Mason * hurt to have more than one go through, but there's no 21841bbc621eSChris Mason * real advantage to it either. 21851bbc621eSChris Mason */ 21860b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21873204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21883204d33cSJosef Bacik &cur_trans->flags)) 21891bbc621eSChris Mason run_it = 1; 21900b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21911bbc621eSChris Mason 2192f9cacae3SNikolay Borisov if (run_it) { 219321217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21943e738c53SIoannis Angelakopoulos if (ret) 21953e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2196f9cacae3SNikolay Borisov } 2197f9cacae3SNikolay Borisov } 21981bbc621eSChris Mason 21990b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 22004a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2201d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2202d0c2f4faSFilipe Manana 220328b21c55SFilipe Manana add_pending_snapshot(trans); 220428b21c55SFilipe Manana 22050b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 22069b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2207ccd467d6SChris Mason 2208d0c2f4faSFilipe Manana if (trans->in_fsync) 2209d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 22103e738c53SIoannis Angelakopoulos 22113e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 22123e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2213d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2214d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 221515ee9bc7SJosef Bacik 2216bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2217b4924a0fSLiu Bo ret = cur_trans->aborted; 2218b4924a0fSLiu Bo 2219724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 222015ee9bc7SJosef Bacik 222149b25e05SJeff Mahoney return ret; 222279154b1bSChris Mason } 22234313b399SChris Mason 22244a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 22250b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 22263e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2227bb9c12c9SSage Weil 22280b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2229d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2230d0c2f4faSFilipe Manana 2231d0c2f4faSFilipe Manana if (trans->in_fsync) 2232d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2233d0c2f4faSFilipe Manana 2234ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2235ccd467d6SChris Mason struct btrfs_transaction, list); 2236d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 22379b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 22380b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2239ccd467d6SChris Mason 2240d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2241d0c2f4faSFilipe Manana 2242bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2243ccd467d6SChris Mason 2244724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22451f9b8c8fSFilipe Manana if (ret) 2246e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2247a4abeea4SJosef Bacik } else { 22480b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2249ccd467d6SChris Mason } 2250a4abeea4SJosef Bacik } else { 22510b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2252cb2d3dadSFilipe Manana /* 2253cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2254cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2255cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2256cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2257cb2d3dadSFilipe Manana */ 225884961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2259cb2d3dadSFilipe Manana ret = -EROFS; 2260e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2261cb2d3dadSFilipe Manana } 2262ccd467d6SChris Mason } 226315ee9bc7SJosef Bacik 2264e55958c8SIoannis Angelakopoulos /* 2265e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2266e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2267e55958c8SIoannis Angelakopoulos */ 2268e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2269e55958c8SIoannis Angelakopoulos 22700860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22710860adfdSMiao Xie 227288090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 227382436617SMiao Xie if (ret) 2274e1489b4fSIoannis Angelakopoulos goto lockdep_release; 227582436617SMiao Xie 2276e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 227749b25e05SJeff Mahoney if (ret) 2278e1489b4fSIoannis Angelakopoulos goto lockdep_release; 227916cdcec7SMiao Xie 22805a9ba670SIoannis Angelakopoulos /* 22815a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22825a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22835a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22845a9ba670SIoannis Angelakopoulos */ 22855a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 22865a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2287581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2288581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2289ed3b3d31SChris Mason 2290581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2291e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2292e1489b4fSIoannis Angelakopoulos if (ret) { 2293e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2294ca469637SMiao Xie goto cleanup_transaction; 2295e1489b4fSIoannis Angelakopoulos } 2296ca469637SMiao Xie 229788090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2298cb7ab021SWang Shilong 229948778179SFilipe Manana /* 230048778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 230148778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 230248778179SFilipe Manana * extents complete we lose logged data after a power failure. 230348778179SFilipe Manana */ 23048b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 230548778179SFilipe Manana wait_event(cur_trans->pending_wait, 230648778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 230748778179SFilipe Manana 23082ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2309ed0ca140SJosef Bacik /* 2310ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2311ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 23124a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2313ed0ca140SJosef Bacik */ 23140b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 231528b21c55SFilipe Manana add_pending_snapshot(trans); 23164a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 23170b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2318e1489b4fSIoannis Angelakopoulos 2319e1489b4fSIoannis Angelakopoulos /* 2320e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2321e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2322e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2323e1489b4fSIoannis Angelakopoulos */ 2324e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2325e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2326ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2327ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 232815ee9bc7SJosef Bacik 2329fdfbf020SJosef Bacik /* 23303e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 23313e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 23323e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 23333e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 23343e738c53SIoannis Angelakopoulos */ 23353e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 23363e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 23373e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23383e738c53SIoannis Angelakopoulos 23393e738c53SIoannis Angelakopoulos /* 2340fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2341fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2342fdfbf020SJosef Bacik * kthread could do the work. 2343fdfbf020SJosef Bacik */ 2344fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2345fdfbf020SJosef Bacik 2346bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23472cba30f1SMiao Xie ret = cur_trans->aborted; 23483e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23496cf7f77eSWang Shilong goto scrub_continue; 23502cba30f1SMiao Xie } 23517585717fSChris Mason /* 23527585717fSChris Mason * the reloc mutex makes sure that we stop 23537585717fSChris Mason * the balancing code from coming in and moving 23547585717fSChris Mason * extents around in the middle of the commit 23557585717fSChris Mason */ 23560b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23577585717fSChris Mason 235842874b3dSMiao Xie /* 235942874b3dSMiao Xie * We needn't worry about the delayed items because we will 236042874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 236142874b3dSMiao Xie * core function of the snapshot creation. 236242874b3dSMiao Xie */ 236308d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 236456e9f6eaSDavid Sterba if (ret) 236556e9f6eaSDavid Sterba goto unlock_reloc; 23663063d29fSChris Mason 236742874b3dSMiao Xie /* 236842874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 236942874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 237042874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 237142874b3dSMiao Xie * them. 237242874b3dSMiao Xie * 237342874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 237442874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 237542874b3dSMiao Xie * the nodes and leaves. 237642874b3dSMiao Xie */ 2377e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 237856e9f6eaSDavid Sterba if (ret) 237956e9f6eaSDavid Sterba goto unlock_reloc; 238016cdcec7SMiao Xie 2381c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 238256e9f6eaSDavid Sterba if (ret) 238356e9f6eaSDavid Sterba goto unlock_reloc; 238456bec294SChris Mason 2385e999376fSChris Mason /* 2386e999376fSChris Mason * make sure none of the code above managed to slip in a 2387e999376fSChris Mason * delayed item 2388e999376fSChris Mason */ 2389ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2390e999376fSChris Mason 23912c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2392dc17ff8fSChris Mason 23937e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 239456e9f6eaSDavid Sterba if (ret) 2395dfba78dcSFilipe Manana goto unlock_reloc; 239654aa1f4dSChris Mason 23975d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2398e02119d5SChris Mason * safe to free the root of tree log roots 2399e02119d5SChris Mason */ 24000b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2401e02119d5SChris Mason 24020ed4792aSQu Wenruo /* 24030ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 24040ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 24050ed4792aSQu Wenruo */ 2406460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 240756e9f6eaSDavid Sterba if (ret < 0) 2408dfba78dcSFilipe Manana goto unlock_reloc; 24090ed4792aSQu Wenruo 24109386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 241156e9f6eaSDavid Sterba if (ret) 2412dfba78dcSFilipe Manana goto unlock_reloc; 241354aa1f4dSChris Mason 24142cba30f1SMiao Xie /* 24152cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 24162cba30f1SMiao Xie * update ->aborted, check it. 24172cba30f1SMiao Xie */ 2418bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 24192cba30f1SMiao Xie ret = cur_trans->aborted; 2420dfba78dcSFilipe Manana goto unlock_reloc; 24212cba30f1SMiao Xie } 24222cba30f1SMiao Xie 24230b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 24245f39d397SChris Mason 24250b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 24260b246afaSJeff Mahoney fs_info->tree_root->node); 24270b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 24289e351cc8SJosef Bacik &cur_trans->switch_commits); 24295d4f98a2SYan Zheng 24300b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 24310b246afaSJeff Mahoney fs_info->chunk_root->node); 24320b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 24339e351cc8SJosef Bacik &cur_trans->switch_commits); 24349e351cc8SJosef Bacik 2435f7238e50SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2436f7238e50SJosef Bacik btrfs_set_root_node(&fs_info->block_group_root->root_item, 2437f7238e50SJosef Bacik fs_info->block_group_root->node); 2438f7238e50SJosef Bacik list_add_tail(&fs_info->block_group_root->dirty_list, 2439f7238e50SJosef Bacik &cur_trans->switch_commits); 2440f7238e50SJosef Bacik } 2441f7238e50SJosef Bacik 2442889bfa39SJosef Bacik switch_commit_roots(trans); 24435d4f98a2SYan Zheng 2444ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 24451bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 24462ff7e61eSJeff Mahoney update_super_roots(fs_info); 2447e02119d5SChris Mason 24480b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 24490b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 24500b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 24510b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2452ccd467d6SChris Mason 2453bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2454935e5cc9SMiao Xie 24550b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 24560b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2457656f30dbSFilipe Manana 24584fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24594fbcdf66SFilipe Manana 2460dfba78dcSFilipe Manana /* 2461dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2462dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2463dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2464dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2465dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2466dfba78dcSFilipe Manana * writing its superblock. 2467dfba78dcSFilipe Manana */ 2468dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2469dfba78dcSFilipe Manana 24700b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 24714a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 24720b246afaSJeff Mahoney fs_info->running_transaction = NULL; 24730b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 24740b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2475b7ec40d7SChris Mason 24760b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 24773e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2478e6dcd2dcSChris Mason 2479b7625f46SQu Wenruo /* If we have features changed, wake up the cleaner to update sysfs. */ 2480b7625f46SQu Wenruo if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) && 2481b7625f46SQu Wenruo fs_info->cleaner_kthread) 2482b7625f46SQu Wenruo wake_up_process(fs_info->cleaner_kthread); 2483b7625f46SQu Wenruo 248470458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 248549b25e05SJeff Mahoney if (ret) { 24860b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 248708748810SDavid Sterba "Error while writing out transaction"); 24880b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 24896cf7f77eSWang Shilong goto scrub_continue; 249049b25e05SJeff Mahoney } 249149b25e05SJeff Mahoney 2492eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2493e02119d5SChris Mason /* 2494e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2495e02119d5SChris Mason * to go about their business 2496e02119d5SChris Mason */ 24970b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2498c1f32b7cSAnand Jain if (ret) 2499c1f32b7cSAnand Jain goto scrub_continue; 2500e02119d5SChris Mason 2501d0c2f4faSFilipe Manana /* 2502d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2503d0c2f4faSFilipe Manana * which can change it. 2504d0c2f4faSFilipe Manana */ 2505d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2506d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 25073e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2508d0c2f4faSFilipe Manana 25095ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 25104313b399SChris Mason 25113204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 25120b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 251313212b54SZhao Lei 25140b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 25154a9d8bdeSMiao Xie /* 25164a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 25174a9d8bdeSMiao Xie * which can change it. 25184a9d8bdeSMiao Xie */ 25194a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 25202c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 25213e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25223de4586cSChris Mason 25230b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 252413c5a93eSJosef Bacik list_del_init(&cur_trans->list); 25250b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2526a4abeea4SJosef Bacik 2527724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2528724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 252958176a96SJosef Bacik 25300860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 25310b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2532b2b5ef5cSJan Kara 25332e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 25341abe9b8aSliubo 2535e55958c8SIoannis Angelakopoulos interval = ktime_get_ns() - start_time; 2536e55958c8SIoannis Angelakopoulos 25372ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2538a2de733cSArne Jansen 25399ed74f2dSJosef Bacik if (current->journal_info == trans) 25409ed74f2dSJosef Bacik current->journal_info = NULL; 25419ed74f2dSJosef Bacik 25422c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 254324bbcf04SYan, Zheng 2544e55958c8SIoannis Angelakopoulos update_commit_stats(fs_info, interval); 2545e55958c8SIoannis Angelakopoulos 254679154b1bSChris Mason return ret; 254749b25e05SJeff Mahoney 254856e9f6eaSDavid Sterba unlock_reloc: 254956e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 25503e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 25516cf7f77eSWang Shilong scrub_continue: 25523e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 25533e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25542ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 255549b25e05SJeff Mahoney cleanup_transaction: 2556dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2557c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 25584fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 25590e721106SJosef Bacik trans->block_rsv = NULL; 25600b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 256149b25e05SJeff Mahoney if (current->journal_info == trans) 256249b25e05SJeff Mahoney current->journal_info = NULL; 256397cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 256449b25e05SJeff Mahoney 256549b25e05SJeff Mahoney return ret; 2566e1489b4fSIoannis Angelakopoulos 2567e1489b4fSIoannis Angelakopoulos lockdep_release: 25685a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2569e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2570e1489b4fSIoannis Angelakopoulos goto cleanup_transaction; 25713e738c53SIoannis Angelakopoulos 25723e738c53SIoannis Angelakopoulos lockdep_trans_commit_start_release: 25733e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 25743e738c53SIoannis Angelakopoulos btrfs_end_transaction(trans); 25753e738c53SIoannis Angelakopoulos return ret; 257679154b1bSChris Mason } 257779154b1bSChris Mason 2578d352ac68SChris Mason /* 25799d1a2a3aSDavid Sterba * return < 0 if error 25809d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 25819d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 25829d1a2a3aSDavid Sterba * 25839d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 25849d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 25859d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 25869d1a2a3aSDavid Sterba * few seconds later. 2587d352ac68SChris Mason */ 258833c44184SJosef Bacik int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2589e9d0b13bSChris Mason { 259033c44184SJosef Bacik struct btrfs_root *root; 25919d1a2a3aSDavid Sterba int ret; 2592e9d0b13bSChris Mason 2593a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 25949d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 25959d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 25969d1a2a3aSDavid Sterba return 0; 25979d1a2a3aSDavid Sterba } 25989d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 25999d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2600cfad392bSJosef Bacik list_del_init(&root->root_list); 2601a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 26025d4f98a2SYan Zheng 26034fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 260476dda93cSYan, Zheng 260516cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 260616cdcec7SMiao Xie 260776dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 260876dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 26090078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 261076dda93cSYan, Zheng else 26110078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 261232471dc2SDavid Sterba 2613dc9492c1SJosef Bacik btrfs_put_root(root); 26146596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2615e9d0b13bSChris Mason } 2616572d9ab7SDavid Sterba 2617fccf0c84SJosef Bacik /* 2618fccf0c84SJosef Bacik * We only mark the transaction aborted and then set the file system read-only. 2619fccf0c84SJosef Bacik * This will prevent new transactions from starting or trying to join this 2620fccf0c84SJosef Bacik * one. 2621fccf0c84SJosef Bacik * 2622fccf0c84SJosef Bacik * This means that error recovery at the call site is limited to freeing 2623fccf0c84SJosef Bacik * any local memory allocations and passing the error code up without 2624fccf0c84SJosef Bacik * further cleanup. The transaction should complete as it normally would 2625fccf0c84SJosef Bacik * in the call path but will return -EIO. 2626fccf0c84SJosef Bacik * 2627fccf0c84SJosef Bacik * We'll complete the cleanup in btrfs_end_transaction and 2628fccf0c84SJosef Bacik * btrfs_commit_transaction. 2629fccf0c84SJosef Bacik */ 2630fccf0c84SJosef Bacik void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, 2631fccf0c84SJosef Bacik const char *function, 2632fccf0c84SJosef Bacik unsigned int line, int errno, bool first_hit) 2633fccf0c84SJosef Bacik { 2634fccf0c84SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 2635fccf0c84SJosef Bacik 2636fccf0c84SJosef Bacik WRITE_ONCE(trans->aborted, errno); 2637fccf0c84SJosef Bacik WRITE_ONCE(trans->transaction->aborted, errno); 2638fccf0c84SJosef Bacik if (first_hit && errno == -ENOSPC) 2639fccf0c84SJosef Bacik btrfs_dump_space_info_for_trans_abort(fs_info); 2640fccf0c84SJosef Bacik /* Wake up anybody who may be waiting on this transaction */ 2641fccf0c84SJosef Bacik wake_up(&fs_info->transaction_wait); 2642fccf0c84SJosef Bacik wake_up(&fs_info->transaction_blocked_wait); 2643fccf0c84SJosef Bacik __btrfs_handle_fs_error(fs_info, function, line, errno, NULL); 2644fccf0c84SJosef Bacik } 2645fccf0c84SJosef Bacik 2646956504a3SJosef Bacik int __init btrfs_transaction_init(void) 2647956504a3SJosef Bacik { 2648956504a3SJosef Bacik btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2649956504a3SJosef Bacik sizeof(struct btrfs_trans_handle), 0, 2650956504a3SJosef Bacik SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2651956504a3SJosef Bacik if (!btrfs_trans_handle_cachep) 2652956504a3SJosef Bacik return -ENOMEM; 2653956504a3SJosef Bacik return 0; 2654956504a3SJosef Bacik } 2655956504a3SJosef Bacik 2656956504a3SJosef Bacik void __cold btrfs_transaction_exit(void) 2657956504a3SJosef Bacik { 2658956504a3SJosef Bacik kmem_cache_destroy(btrfs_trans_handle_cachep); 2659956504a3SJosef Bacik } 2660