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 /* 295*4490e803SFilipe Manana * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the 296*4490e803SFilipe Manana * current transaction, and commit it. If there is no transaction, just 297*4490e803SFilipe Manana * return ENOENT. 298354aa0fbSMiao Xie */ 299*4490e803SFilipe Manana if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART) 300354aa0fbSMiao Xie return -ENOENT; 301354aa0fbSMiao Xie 3024a9d8bdeSMiao Xie /* 3034a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3044a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3054a9d8bdeSMiao Xie */ 3064a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3074a9d8bdeSMiao Xie 3084b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 309db5b493aSTsutomu Itoh if (!cur_trans) 310db5b493aSTsutomu Itoh return -ENOMEM; 311d43317dcSChris Mason 312e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 3135a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 314e1489b4fSIoannis Angelakopoulos 31519ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 31619ae4e81SJan Schmidt if (fs_info->running_transaction) { 317d43317dcSChris Mason /* 318d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3194a9d8bdeSMiao Xie * to redo the checks above 320d43317dcSChris Mason */ 3215a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 322e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3234b5faeacSDavid Sterba kfree(cur_trans); 324d43317dcSChris Mason goto loop; 32584961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 326e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3275a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 328e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3294b5faeacSDavid Sterba kfree(cur_trans); 3307b8b92afSJosef Bacik return -EROFS; 331a4abeea4SJosef Bacik } 332d43317dcSChris Mason 333ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 33448778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 33548778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 33613c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3370860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 33879154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 33979154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3404a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 341a4abeea4SJosef Bacik /* 342a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 343a4abeea4SJosef Bacik * commit the transaction. 344a4abeea4SJosef Bacik */ 3459b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3463204d33cSJosef Bacik cur_trans->flags = 0; 347afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 34856bec294SChris Mason 349a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 350a099d0fdSAlexandru Moise 3515c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3523368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 353d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 35420b297d6SJan Schmidt 35520b297d6SJan Schmidt /* 35620b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 35720b297d6SJan Schmidt * the log must never go across transaction boundaries. 35820b297d6SJan Schmidt */ 35920b297d6SJan Schmidt smp_mb(); 36031b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3615d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 36231b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3635d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 364fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 36520b297d6SJan Schmidt 36656bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 36756bec294SChris Mason 3683063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 369bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3709e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 371ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3721bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3732b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3741bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 375ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 376e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3772b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 37819ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 379c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 38035da5a7eSDavid Sterba IO_TREE_TRANS_DIRTY_PAGES); 381fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 38235da5a7eSDavid Sterba IO_TREE_FS_PINNED_EXTENTS); 38319ae4e81SJan Schmidt fs_info->generation++; 38419ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 38519ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 38649b25e05SJeff Mahoney cur_trans->aborted = 0; 38719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 38815ee9bc7SJosef Bacik 38979154b1bSChris Mason return 0; 39079154b1bSChris Mason } 39179154b1bSChris Mason 392d352ac68SChris Mason /* 39392a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 39492a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 39592a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 39692a7cc42SQu Wenruo * transaction commits. 397d352ac68SChris Mason */ 3987585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3996426c7adSQu Wenruo struct btrfs_root *root, 4006426c7adSQu Wenruo int force) 4016702ed49SChris Mason { 4020b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 40303a7e111SJosef Bacik int ret = 0; 4040b246afaSJeff Mahoney 40592a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4066426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4074d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4085d4f98a2SYan Zheng 4097585717fSChris Mason /* 41027cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4117585717fSChris Mason * we have the reloc mutex held now, so there 4127585717fSChris Mason * is only one writer in this function 4137585717fSChris Mason */ 41427cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4157585717fSChris Mason 41627cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4177585717fSChris Mason * they find our root->last_trans update 4187585717fSChris Mason */ 4197585717fSChris Mason smp_wmb(); 4207585717fSChris Mason 421fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4226426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 423fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 424a4abeea4SJosef Bacik return 0; 425a4abeea4SJosef Bacik } 426fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4276702ed49SChris Mason (unsigned long)root->root_key.objectid, 4286702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 429fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4307585717fSChris Mason root->last_trans = trans->transid; 4317585717fSChris Mason 4327585717fSChris Mason /* this is pretty tricky. We don't want to 4337585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4347585717fSChris Mason * unless we're really doing the first setup for this root in 4357585717fSChris Mason * this transaction. 4367585717fSChris Mason * 4377585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4387585717fSChris Mason * if we want to take the expensive mutex. 4397585717fSChris Mason * 4407585717fSChris Mason * But, we have to set root->last_trans before we 4417585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4427585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 44327cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4447585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4457585717fSChris Mason * 4467585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4477585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4487585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4497585717fSChris Mason * done before we pop in the zero below 4507585717fSChris Mason */ 45103a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 452c7548af6SChris Mason smp_mb__before_atomic(); 45327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4546702ed49SChris Mason } 45503a7e111SJosef Bacik return ret; 4566702ed49SChris Mason } 4575d4f98a2SYan Zheng 4587585717fSChris Mason 4592b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4602b9dbef2SJosef Bacik struct btrfs_root *root) 4612b9dbef2SJosef Bacik { 4620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4632b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4642b9dbef2SJosef Bacik 4652b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4662b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4672b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4682b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4692b9dbef2SJosef Bacik 4702b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 471fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 472fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4732b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4742b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 475fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4762b9dbef2SJosef Bacik } 4772b9dbef2SJosef Bacik 4787585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4797585717fSChris Mason struct btrfs_root *root) 4807585717fSChris Mason { 4810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4821409e6ccSJosef Bacik int ret; 4830b246afaSJeff Mahoney 48492a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4857585717fSChris Mason return 0; 4867585717fSChris Mason 4877585717fSChris Mason /* 48827cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4897585717fSChris Mason * and barriers 4907585717fSChris Mason */ 4917585717fSChris Mason smp_rmb(); 4927585717fSChris Mason if (root->last_trans == trans->transid && 49327cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4947585717fSChris Mason return 0; 4957585717fSChris Mason 4960b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4971409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4980b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4997585717fSChris Mason 5001409e6ccSJosef Bacik return ret; 5017585717fSChris Mason } 5027585717fSChris Mason 5034a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5044a9d8bdeSMiao Xie { 5053296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 506501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 507bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5084a9d8bdeSMiao Xie } 5094a9d8bdeSMiao Xie 510d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 511d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 512d352ac68SChris Mason * transaction might not be fully on disk. 513d352ac68SChris Mason */ 5142ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 51579154b1bSChris Mason { 516f9295749SChris Mason struct btrfs_transaction *cur_trans; 51779154b1bSChris Mason 5180b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5190b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5204a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5219b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5220b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52372d63ed6SLi Zefan 5243e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5250b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 526501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 527bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 528724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 529a4abeea4SJosef Bacik } else { 5300b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 531f9295749SChris Mason } 53237d1aeeeSChris Mason } 53337d1aeeeSChris Mason 5342ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 53537d1aeeeSChris Mason { 5360b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 537a4abeea4SJosef Bacik return 0; 538a4abeea4SJosef Bacik 53992e2f7e3SNikolay Borisov if (type == TRANS_START) 540a4abeea4SJosef Bacik return 1; 541a4abeea4SJosef Bacik 542a22285a6SYan, Zheng return 0; 543a22285a6SYan, Zheng } 544a22285a6SYan, Zheng 54520dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 54620dd2cbfSMiao Xie { 5470b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5480b246afaSJeff Mahoney 5490b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 55092a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 55120dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 55220dd2cbfSMiao Xie root->reloc_root) 55320dd2cbfSMiao Xie return false; 55420dd2cbfSMiao Xie 55520dd2cbfSMiao Xie return true; 55620dd2cbfSMiao Xie } 55720dd2cbfSMiao Xie 55808e007d2SMiao Xie static struct btrfs_trans_handle * 5595aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 560003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 561003d7c59SJeff Mahoney bool enforce_qgroups) 562a22285a6SYan, Zheng { 5630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 564ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 565a22285a6SYan, Zheng struct btrfs_trans_handle *h; 566a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 567b5009945SJosef Bacik u64 num_bytes = 0; 568c5567237SArne Jansen u64 qgroup_reserved = 0; 56920dd2cbfSMiao Xie bool reloc_reserved = false; 5709c343784SJosef Bacik bool do_chunk_alloc = false; 57120dd2cbfSMiao Xie int ret; 572acce952bSliubo 57384961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 574acce952bSliubo return ERR_PTR(-EROFS); 5752a1eb461SJosef Bacik 57646c4e71eSFilipe Manana if (current->journal_info) { 5770860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5782a1eb461SJosef Bacik h = current->journal_info; 579b50fff81SDavid Sterba refcount_inc(&h->use_count); 580b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5812a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5822a1eb461SJosef Bacik h->block_rsv = NULL; 5832a1eb461SJosef Bacik goto got_it; 5842a1eb461SJosef Bacik } 585b5009945SJosef Bacik 586b5009945SJosef Bacik /* 587b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 588b5009945SJosef Bacik * the appropriate flushing if need be. 589b5009945SJosef Bacik */ 590003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 591ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 592ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 593ba2c4d4eSJosef Bacik 5940b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 595a6496849SBoris Burkov /* 596a6496849SBoris Burkov * Use prealloc for now, as there might be a currently running 597a6496849SBoris Burkov * transaction that could free this reserved space prematurely 598a6496849SBoris Burkov * by committing. 599a6496849SBoris Burkov */ 600a6496849SBoris Burkov ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserved, 601a6496849SBoris Burkov enforce_qgroups, false); 602c5567237SArne Jansen if (ret) 603c5567237SArne Jansen return ERR_PTR(ret); 604c5567237SArne Jansen 605ba2c4d4eSJosef Bacik /* 606ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 607ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 6080f69d1f4SFilipe Manana * accomplish this by simply assuming we'll do num_items worth 6090f69d1f4SFilipe Manana * of delayed refs updates in this trans handle, and refill that 6100f69d1f4SFilipe Manana * amount for whatever is missing in the reserve. 611ba2c4d4eSJosef Bacik */ 6122bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6137f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 614e4773b57SFilipe Manana !btrfs_block_rsv_full(delayed_refs_rsv)) { 6150f69d1f4SFilipe Manana delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, 6160f69d1f4SFilipe Manana num_items); 6170f69d1f4SFilipe Manana num_bytes += delayed_refs_bytes; 618ba2c4d4eSJosef Bacik } 619ba2c4d4eSJosef Bacik 62020dd2cbfSMiao Xie /* 62120dd2cbfSMiao Xie * Do the reservation for the relocation root creation 62220dd2cbfSMiao Xie */ 623ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6240b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 62520dd2cbfSMiao Xie reloc_reserved = true; 62620dd2cbfSMiao Xie } 62720dd2cbfSMiao Xie 6289270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 629ba2c4d4eSJosef Bacik if (ret) 630ba2c4d4eSJosef Bacik goto reserve_fail; 631ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 632ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 633ba2c4d4eSJosef Bacik delayed_refs_bytes); 634ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 635ba2c4d4eSJosef Bacik } 6369c343784SJosef Bacik 6379c343784SJosef Bacik if (rsv->space_info->force_alloc) 6389c343784SJosef Bacik do_chunk_alloc = true; 639ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 640748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 641ba2c4d4eSJosef Bacik /* 642ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 643ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 644ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 645ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 646ba2c4d4eSJosef Bacik * here. 647ba2c4d4eSJosef Bacik */ 648ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 649b5009945SJosef Bacik if (ret) 650843fcf35SMiao Xie goto reserve_fail; 651b5009945SJosef Bacik } 652a22285a6SYan, Zheng again: 653f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 654843fcf35SMiao Xie if (!h) { 655843fcf35SMiao Xie ret = -ENOMEM; 656843fcf35SMiao Xie goto alloc_fail; 657843fcf35SMiao Xie } 658a22285a6SYan, Zheng 65998114659SJosef Bacik /* 66098114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 66198114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 66298114659SJosef Bacik * because we're already holding a ref. We need this because we could 66398114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 66498114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 665354aa0fbSMiao Xie * 666354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 667354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 66898114659SJosef Bacik */ 6690860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6700b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 671b2b5ef5cSJan Kara 6722ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6732ff7e61eSJeff Mahoney wait_current_trans(fs_info); 674a22285a6SYan, Zheng 675a4abeea4SJosef Bacik do { 6762ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 677178260b2SMiao Xie if (ret == -EBUSY) { 6782ff7e61eSJeff Mahoney wait_current_trans(fs_info); 679a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 680a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 681178260b2SMiao Xie ret = -ENOENT; 682178260b2SMiao Xie } 683a4abeea4SJosef Bacik } while (ret == -EBUSY); 684a4abeea4SJosef Bacik 685a43f7f82SLiu Bo if (ret < 0) 686843fcf35SMiao Xie goto join_fail; 6870f7d52f4SChris Mason 6880b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 689a22285a6SYan, Zheng 690a22285a6SYan, Zheng h->transid = cur_trans->transid; 691a22285a6SYan, Zheng h->transaction = cur_trans; 692b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 69364b63580SJeff Mahoney h->fs_info = root->fs_info; 6947174109cSQu Wenruo 695a698d075SMiao Xie h->type = type; 696ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 697b7ec40d7SChris Mason 698a22285a6SYan, Zheng smp_mb(); 6993296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 7002ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 701abdd2e80SFilipe Manana current->journal_info = h; 7023a45bb20SJeff Mahoney btrfs_commit_transaction(h); 703a22285a6SYan, Zheng goto again; 704a22285a6SYan, Zheng } 7059ed74f2dSJosef Bacik 706b5009945SJosef Bacik if (num_bytes) { 7070b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7082bcc0328SLiu Bo h->transid, num_bytes, 1); 7090b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 710b5009945SJosef Bacik h->bytes_reserved = num_bytes; 71120dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 712a22285a6SYan, Zheng } 713a22285a6SYan, Zheng 714a6496849SBoris Burkov /* 715a6496849SBoris Burkov * Now that we have found a transaction to be a part of, convert the 716a6496849SBoris Burkov * qgroup reservation from prealloc to pertrans. A different transaction 717a6496849SBoris Burkov * can't race in and free our pertrans out from under us. 718a6496849SBoris Burkov */ 719a6496849SBoris Burkov if (qgroup_reserved) 720a6496849SBoris Burkov btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved); 721a6496849SBoris Burkov 7222a1eb461SJosef Bacik got_it: 723bcf3a3e7SNikolay Borisov if (!current->journal_info) 724a22285a6SYan, Zheng current->journal_info = h; 725fcc99734SQu Wenruo 726fcc99734SQu Wenruo /* 7279c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7289c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7299c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7309c343784SJosef Bacik * value here. 7319c343784SJosef Bacik */ 7329c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7339c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7349c343784SJosef Bacik 7359c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7369c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7379c343784SJosef Bacik } 7389c343784SJosef Bacik 7399c343784SJosef Bacik /* 740fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 741fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 742fcc99734SQu Wenruo * transaction. 743fcc99734SQu Wenruo * 744fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 745fcc99734SQu Wenruo * or we can deadlock. 746fcc99734SQu Wenruo */ 74768075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 74868075ea8SJosef Bacik if (ret) { 74968075ea8SJosef Bacik /* 75068075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 75168075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 75268075ea8SJosef Bacik * not just freed. 75368075ea8SJosef Bacik */ 75468075ea8SJosef Bacik btrfs_end_transaction(h); 75568075ea8SJosef Bacik return ERR_PTR(ret); 75668075ea8SJosef Bacik } 757fcc99734SQu Wenruo 75879154b1bSChris Mason return h; 759843fcf35SMiao Xie 760843fcf35SMiao Xie join_fail: 7610860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7620b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 763843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 764843fcf35SMiao Xie alloc_fail: 765843fcf35SMiao Xie if (num_bytes) 7662ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 76763f018beSNikolay Borisov num_bytes, NULL); 768843fcf35SMiao Xie reserve_fail: 769a6496849SBoris Burkov btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved); 770843fcf35SMiao Xie return ERR_PTR(ret); 77179154b1bSChris Mason } 77279154b1bSChris Mason 773f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7745aed1dd8SAlexandru Moise unsigned int num_items) 775f9295749SChris Mason { 77608e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 777003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 778f9295749SChris Mason } 779003d7c59SJeff Mahoney 7808eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7818eab77ffSFilipe Manana struct btrfs_root *root, 7827f9fe614SJosef Bacik unsigned int num_items) 7838eab77ffSFilipe Manana { 7847f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7857f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7868eab77ffSFilipe Manana } 7878407aa46SMiao Xie 7887a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 789f9295749SChris Mason { 790003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 791003d7c59SJeff Mahoney true); 792f9295749SChris Mason } 793f9295749SChris Mason 7948d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7950af3d00bSJosef Bacik { 796575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 797003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7980af3d00bSJosef Bacik } 7990af3d00bSJosef Bacik 800d4edf39bSMiao Xie /* 801a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 802a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 803a6d155d2SFilipe Manana */ 804a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 805a6d155d2SFilipe Manana { 806a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 807a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 808a6d155d2SFilipe Manana } 809a6d155d2SFilipe Manana 810a6d155d2SFilipe Manana /* 811d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 812d4edf39bSMiao Xie * 813d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 814d4edf39bSMiao Xie * don't want to start a new one. 815d4edf39bSMiao Xie * 816d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 817d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 818d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 819d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 820d4edf39bSMiao Xie * invoke 821d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 822d4edf39bSMiao Xie */ 823354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 82460376ce4SJosef Bacik { 825575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 826003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 82760376ce4SJosef Bacik } 82860376ce4SJosef Bacik 829d4edf39bSMiao Xie /* 83090b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 831d4edf39bSMiao Xie * 83252042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 833d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 834d4edf39bSMiao Xie * complete. 835d4edf39bSMiao Xie */ 836d4edf39bSMiao Xie struct btrfs_trans_handle * 837d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 838d4edf39bSMiao Xie { 839d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 840d4edf39bSMiao Xie 841575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 842003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 843b28ff3a7SFilipe Manana if (trans == ERR_PTR(-ENOENT)) { 844b28ff3a7SFilipe Manana int ret; 845b28ff3a7SFilipe Manana 846b28ff3a7SFilipe Manana ret = btrfs_wait_for_commit(root->fs_info, 0); 847b28ff3a7SFilipe Manana if (ret) 848b28ff3a7SFilipe Manana return ERR_PTR(ret); 849b28ff3a7SFilipe Manana } 850d4edf39bSMiao Xie 851d4edf39bSMiao Xie return trans; 852d4edf39bSMiao Xie } 853d4edf39bSMiao Xie 854d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 855d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 856d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 85789ce8a63SChris Mason { 8585fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8595fd76bf3SOmar Sandoval u64 transid = commit->transid; 8605fd76bf3SOmar Sandoval bool put = false; 8615fd76bf3SOmar Sandoval 8623e738c53SIoannis Angelakopoulos /* 8633e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8643e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8653e738c53SIoannis Angelakopoulos */ 8663e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8673e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8683e738c53SIoannis Angelakopoulos else 8693e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8703e738c53SIoannis Angelakopoulos 8715fd76bf3SOmar Sandoval while (1) { 872d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8735fd76bf3SOmar Sandoval if (put) 8745fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8755fd76bf3SOmar Sandoval 8765fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8775fd76bf3SOmar Sandoval break; 8785fd76bf3SOmar Sandoval 8795fd76bf3SOmar Sandoval /* 8805fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8815fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8825fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8835fd76bf3SOmar Sandoval * transaction. Wait for those. 8845fd76bf3SOmar Sandoval */ 8855fd76bf3SOmar Sandoval 8865fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8875fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8885fd76bf3SOmar Sandoval struct btrfs_transaction, 8895fd76bf3SOmar Sandoval list); 8905fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8915fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8925fd76bf3SOmar Sandoval break; 8935fd76bf3SOmar Sandoval } 8945fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8955fd76bf3SOmar Sandoval put = true; 8965fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8975fd76bf3SOmar Sandoval } 89889ce8a63SChris Mason } 89989ce8a63SChris Mason 9002ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 90146204592SSage Weil { 90246204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 9038cd2807fSMiao Xie int ret = 0; 90446204592SSage Weil 90546204592SSage Weil if (transid) { 9060b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 907a4abeea4SJosef Bacik goto out; 90846204592SSage Weil 90946204592SSage Weil /* find specified transaction */ 9100b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9110b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 91246204592SSage Weil if (t->transid == transid) { 91346204592SSage Weil cur_trans = t; 9149b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 9158cd2807fSMiao Xie ret = 0; 91646204592SSage Weil break; 91746204592SSage Weil } 9188cd2807fSMiao Xie if (t->transid > transid) { 9198cd2807fSMiao Xie ret = 0; 92046204592SSage Weil break; 92146204592SSage Weil } 9228cd2807fSMiao Xie } 9230b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 92442383020SSage Weil 92542383020SSage Weil /* 92642383020SSage Weil * The specified transaction doesn't exist, or we 92742383020SSage Weil * raced with btrfs_commit_transaction 92842383020SSage Weil */ 92942383020SSage Weil if (!cur_trans) { 9300b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 93142383020SSage Weil ret = -EINVAL; 9328cd2807fSMiao Xie goto out; 93342383020SSage Weil } 93446204592SSage Weil } else { 93546204592SSage Weil /* find newest transaction that is committing | committed */ 9360b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9370b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 93846204592SSage Weil list) { 9394a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9404a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9413473f3c0SJosef Bacik break; 94246204592SSage Weil cur_trans = t; 9439b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 94446204592SSage Weil break; 94546204592SSage Weil } 94646204592SSage Weil } 9470b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 94846204592SSage Weil if (!cur_trans) 949a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 95046204592SSage Weil } 95146204592SSage Weil 952d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 953bf7ecbe9SFilipe Manana ret = cur_trans->aborted; 954724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 955a4abeea4SJosef Bacik out: 95646204592SSage Weil return ret; 95746204592SSage Weil } 95846204592SSage Weil 9592ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 96037d1aeeeSChris Mason { 9612ff7e61eSJeff Mahoney wait_current_trans(fs_info); 96237d1aeeeSChris Mason } 96337d1aeeeSChris Mason 964a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9658929ecfaSYan, Zheng { 9668929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9678929ecfaSYan, Zheng 9683296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 969e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 970a2633b6aSNikolay Borisov return true; 9718929ecfaSYan, Zheng 97204fb3285SFilipe Manana if (btrfs_check_space_for_delayed_refs(trans->fs_info)) 97304fb3285SFilipe Manana return true; 97404fb3285SFilipe Manana 97504fb3285SFilipe Manana return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50); 9768929ecfaSYan, Zheng } 9778929ecfaSYan, Zheng 978dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 979dc60c525SNikolay Borisov 9800e34693fSNikolay Borisov { 981dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 982dc60c525SNikolay Borisov 9830e34693fSNikolay Borisov if (!trans->block_rsv) { 9840e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9850e34693fSNikolay Borisov return; 9860e34693fSNikolay Borisov } 9870e34693fSNikolay Borisov 9880e34693fSNikolay Borisov if (!trans->bytes_reserved) 9890e34693fSNikolay Borisov return; 9900e34693fSNikolay Borisov 9910e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9920e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9930e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9940e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 99563f018beSNikolay Borisov trans->bytes_reserved, NULL); 9960e34693fSNikolay Borisov trans->bytes_reserved = 0; 9970e34693fSNikolay Borisov } 9980e34693fSNikolay Borisov 99989ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 10003a45bb20SJeff Mahoney int throttle) 100179154b1bSChris Mason { 10023a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 10038929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 10044edc2ca3SDave Jones int err = 0; 1005d6e4a428SChris Mason 1006b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 1007b50fff81SDavid Sterba refcount_dec(&trans->use_count); 10082a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 10092a1eb461SJosef Bacik return 0; 10102a1eb461SJosef Bacik } 10112a1eb461SJosef Bacik 1012dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 10134c13d758SJosef Bacik trans->block_rsv = NULL; 1014c5567237SArne Jansen 10156c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 1016ea658badSJosef Bacik 10174fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 10184fbcdf66SFilipe Manana 10190860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 10200b246afaSJeff Mahoney sb_end_intwrite(info->sb); 10216df7881aSJosef Bacik 10228929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 102313c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 102413c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10250860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 102689ce8a63SChris Mason 1027093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1028e1489b4fSIoannis Angelakopoulos 10295a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1030e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1031e1489b4fSIoannis Angelakopoulos 1032724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10339ed74f2dSJosef Bacik 10349ed74f2dSJosef Bacik if (current->journal_info == trans) 10359ed74f2dSJosef Bacik current->journal_info = NULL; 1036ab78c84dSChris Mason 103724bbcf04SYan, Zheng if (throttle) 10382ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 103924bbcf04SYan, Zheng 104084961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10414e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1042fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1043fbabd4a3SJosef Bacik err = trans->aborted; 1044fbabd4a3SJosef Bacik else 1045fbabd4a3SJosef Bacik err = -EROFS; 10464e121c06SJosef Bacik } 104749b25e05SJeff Mahoney 10484edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10494edc2ca3SDave Jones return err; 105079154b1bSChris Mason } 105179154b1bSChris Mason 10523a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 105389ce8a63SChris Mason { 10543a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 105589ce8a63SChris Mason } 105689ce8a63SChris Mason 10573a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 105889ce8a63SChris Mason { 10593a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 106016cdcec7SMiao Xie } 106116cdcec7SMiao Xie 1062d352ac68SChris Mason /* 1063d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1064d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1065690587d1SChris Mason * those extents are sent to disk but does not wait on them 1066d352ac68SChris Mason */ 10672ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10688cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 106979154b1bSChris Mason { 1070777e6bd7SChris Mason int err = 0; 10717c4452b9SChris Mason int werr = 0; 10720b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1073e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1074777e6bd7SChris Mason u64 start = 0; 10755f39d397SChris Mason u64 end; 10767c4452b9SChris Mason 1077e5860f82SFilipe Manana while (find_first_extent_bit(dirty_pages, start, &start, &end, 1078e6138876SJosef Bacik mark, &cached_state)) { 1079663dfbb0SFilipe Manana bool wait_writeback = false; 1080663dfbb0SFilipe Manana 1081663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1082663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1083210aa277SDavid Sterba mark, &cached_state); 1084663dfbb0SFilipe Manana /* 1085663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1086663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1087663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1088663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1089bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1090bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1091bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1092bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1093bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1094663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 109541e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1096663dfbb0SFilipe Manana */ 1097663dfbb0SFilipe Manana if (err == -ENOMEM) { 1098663dfbb0SFilipe Manana err = 0; 1099663dfbb0SFilipe Manana wait_writeback = true; 1100663dfbb0SFilipe Manana } 1101663dfbb0SFilipe Manana if (!err) 11021728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 11037c4452b9SChris Mason if (err) 11047c4452b9SChris Mason werr = err; 1105663dfbb0SFilipe Manana else if (wait_writeback) 1106663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1107e38e2ed7SFilipe Manana free_extent_state(cached_state); 1108663dfbb0SFilipe Manana cached_state = NULL; 11091728366eSJosef Bacik cond_resched(); 11101728366eSJosef Bacik start = end + 1; 11117c4452b9SChris Mason } 1112690587d1SChris Mason return werr; 1113690587d1SChris Mason } 1114690587d1SChris Mason 1115690587d1SChris Mason /* 1116690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1117690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1118690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1119690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1120690587d1SChris Mason */ 1121bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1122bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1123690587d1SChris Mason { 1124690587d1SChris Mason int err = 0; 1125690587d1SChris Mason int werr = 0; 11260b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1127e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1128690587d1SChris Mason u64 start = 0; 1129690587d1SChris Mason u64 end; 1130690587d1SChris Mason 1131e5860f82SFilipe Manana while (find_first_extent_bit(dirty_pages, start, &start, &end, 1132e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1133663dfbb0SFilipe Manana /* 1134663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1135663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1136663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1137663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1138663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 113941e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1140663dfbb0SFilipe Manana */ 1141663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1142bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1143663dfbb0SFilipe Manana if (err == -ENOMEM) 1144663dfbb0SFilipe Manana err = 0; 1145663dfbb0SFilipe Manana if (!err) 11461728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1147777e6bd7SChris Mason if (err) 1148777e6bd7SChris Mason werr = err; 1149e38e2ed7SFilipe Manana free_extent_state(cached_state); 1150e38e2ed7SFilipe Manana cached_state = NULL; 1151777e6bd7SChris Mason cond_resched(); 11521728366eSJosef Bacik start = end + 1; 1153777e6bd7SChris Mason } 11547c4452b9SChris Mason if (err) 11557c4452b9SChris Mason werr = err; 1156bf89d38fSJeff Mahoney return werr; 1157bf89d38fSJeff Mahoney } 1158656f30dbSFilipe Manana 1159b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1160bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1161bf89d38fSJeff Mahoney { 1162bf89d38fSJeff Mahoney bool errors = false; 1163bf89d38fSJeff Mahoney int err; 1164bf89d38fSJeff Mahoney 1165bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1166bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1167bf89d38fSJeff Mahoney errors = true; 1168bf89d38fSJeff Mahoney 1169bf89d38fSJeff Mahoney if (errors && !err) 1170bf89d38fSJeff Mahoney err = -EIO; 1171bf89d38fSJeff Mahoney return err; 1172bf89d38fSJeff Mahoney } 1173bf89d38fSJeff Mahoney 1174bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1175bf89d38fSJeff Mahoney { 1176bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1177bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1178bf89d38fSJeff Mahoney bool errors = false; 1179bf89d38fSJeff Mahoney int err; 1180bf89d38fSJeff Mahoney 1181bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1182bf89d38fSJeff Mahoney 1183bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1184656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11850b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1186656f30dbSFilipe Manana errors = true; 1187656f30dbSFilipe Manana 1188656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11890b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1190656f30dbSFilipe Manana errors = true; 1191656f30dbSFilipe Manana 1192bf89d38fSJeff Mahoney if (errors && !err) 1193bf89d38fSJeff Mahoney err = -EIO; 1194bf89d38fSJeff Mahoney return err; 119579154b1bSChris Mason } 119679154b1bSChris Mason 1197690587d1SChris Mason /* 1198c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1199c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1200c9b577c0SNikolay Borisov * log commit. 1201c9b577c0SNikolay Borisov * 1202c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1203690587d1SChris Mason */ 120470458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1205d0c803c4SChris Mason { 1206663dfbb0SFilipe Manana int ret; 1207c9b577c0SNikolay Borisov int ret2; 1208c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 120970458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1210c9b577c0SNikolay Borisov struct blk_plug plug; 1211663dfbb0SFilipe Manana 1212c9b577c0SNikolay Borisov blk_start_plug(&plug); 1213c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1214c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1215c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1216c9b577c0SNikolay Borisov 121741e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1218663dfbb0SFilipe Manana 1219c9b577c0SNikolay Borisov if (ret) 1220663dfbb0SFilipe Manana return ret; 1221c9b577c0SNikolay Borisov else if (ret2) 1222c9b577c0SNikolay Borisov return ret2; 1223c9b577c0SNikolay Borisov else 1224c9b577c0SNikolay Borisov return 0; 1225d0c803c4SChris Mason } 1226d0c803c4SChris Mason 1227d352ac68SChris Mason /* 1228d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1229d352ac68SChris Mason * 1230d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1231d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1232d352ac68SChris Mason * allocation tree. 1233d352ac68SChris Mason * 1234d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1235d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1236d352ac68SChris Mason */ 12370b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 123879154b1bSChris Mason struct btrfs_root *root) 123979154b1bSChris Mason { 124079154b1bSChris Mason int ret; 12410b86a832SChris Mason u64 old_root_bytenr; 124286b9f2ecSYan, Zheng u64 old_root_used; 12430b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12440b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 124579154b1bSChris Mason 124686b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 124756bec294SChris Mason 124879154b1bSChris Mason while (1) { 12490b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 125086b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1251ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 125279154b1bSChris Mason break; 125387ef2bb4SChris Mason 12545d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 125579154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12560b86a832SChris Mason &root->root_key, 12570b86a832SChris Mason &root->root_item); 125849b25e05SJeff Mahoney if (ret) 125949b25e05SJeff Mahoney return ret; 126056bec294SChris Mason 126186b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1262e7070be1SJosef Bacik } 1263276e680dSYan Zheng 12640b86a832SChris Mason return 0; 12650b86a832SChris Mason } 12660b86a832SChris Mason 1267d352ac68SChris Mason /* 1268d352ac68SChris Mason * update all the cowonly tree roots on disk 126949b25e05SJeff Mahoney * 127049b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 127149b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 127249b25e05SJeff Mahoney * to clean up the delayed refs. 1273d352ac68SChris Mason */ 12749386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12750b86a832SChris Mason { 12769386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1277ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12781bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12790b86a832SChris Mason struct list_head *next; 128084234f3aSYan Zheng struct extent_buffer *eb; 128156bec294SChris Mason int ret; 128284234f3aSYan Zheng 1283dfba78dcSFilipe Manana /* 1284dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1285dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1286dfba78dcSFilipe Manana */ 1287dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1288dfba78dcSFilipe Manana 128984234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 129049b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12919631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 129284234f3aSYan Zheng btrfs_tree_unlock(eb); 129384234f3aSYan Zheng free_extent_buffer(eb); 12940b86a832SChris Mason 129549b25e05SJeff Mahoney if (ret) 129649b25e05SJeff Mahoney return ret; 129749b25e05SJeff Mahoney 1298196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1299c16ce190SJosef Bacik if (ret) 1300c16ce190SJosef Bacik return ret; 13012b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1302c16ce190SJosef Bacik if (ret) 1303c16ce190SJosef Bacik return ret; 1304280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1305c16ce190SJosef Bacik if (ret) 1306c16ce190SJosef Bacik return ret; 1307546adb0dSJan Schmidt 1308bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1309dcdf7f6dSJosef Bacik if (ret) 1310dcdf7f6dSJosef Bacik return ret; 1311dcdf7f6dSJosef Bacik 1312ea526d18SJosef Bacik again: 13130b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 13142ff7e61eSJeff Mahoney struct btrfs_root *root; 13150b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 13160b86a832SChris Mason list_del_init(next); 13170b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1318e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 131987ef2bb4SChris Mason 13209e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13219e351cc8SJosef Bacik &trans->transaction->switch_commits); 132249b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 132349b25e05SJeff Mahoney if (ret) 132449b25e05SJeff Mahoney return ret; 1325488bc2a2SJosef Bacik } 1326488bc2a2SJosef Bacik 1327488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1328c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1329ea526d18SJosef Bacik if (ret) 1330ea526d18SJosef Bacik return ret; 1331276e680dSYan Zheng 13321bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13335742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1334ea526d18SJosef Bacik if (ret) 1335ea526d18SJosef Bacik return ret; 1336488bc2a2SJosef Bacik 1337488bc2a2SJosef Bacik /* 1338488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1339488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1340488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1341488bc2a2SJosef Bacik * everything gets run. 1342488bc2a2SJosef Bacik */ 1343c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1344ea526d18SJosef Bacik if (ret) 1345ea526d18SJosef Bacik return ret; 1346ea526d18SJosef Bacik } 1347ea526d18SJosef Bacik 1348ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1349ea526d18SJosef Bacik goto again; 1350ea526d18SJosef Bacik 13519f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13529f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13539f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13548dabb742SStefan Behrens 135579154b1bSChris Mason return 0; 135679154b1bSChris Mason } 135779154b1bSChris Mason 1358d352ac68SChris Mason /* 1359b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1360b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1361b4be6aefSJosef Bacik */ 1362b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1363b4be6aefSJosef Bacik { 1364b4be6aefSJosef Bacik /* 1365b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1366b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1367b4be6aefSJosef Bacik * up. 1368b4be6aefSJosef Bacik */ 1369b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1370b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1371b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1372b4be6aefSJosef Bacik struct btrfs_root, 1373b4be6aefSJosef Bacik root_list); 1374b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1375b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1376b4be6aefSJosef Bacik return; 1377b4be6aefSJosef Bacik } 1378b4be6aefSJosef Bacik } 1379b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1380b4be6aefSJosef Bacik 1381b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1382b4be6aefSJosef Bacik } 1383b4be6aefSJosef Bacik 1384b4be6aefSJosef Bacik /* 1385d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1386d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1387d352ac68SChris Mason * be deleted 1388d352ac68SChris Mason */ 1389cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13905eda7b5eSChris Mason { 13910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13920b246afaSJeff Mahoney 13930b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1394dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1395dc9492c1SJosef Bacik btrfs_grab_root(root); 1396b4be6aefSJosef Bacik 1397b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1398b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1399b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1400b4be6aefSJosef Bacik else 14010b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1402dc9492c1SJosef Bacik } 14030b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 14045eda7b5eSChris Mason } 14055eda7b5eSChris Mason 1406d352ac68SChris Mason /* 1407dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1408dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1409d352ac68SChris Mason */ 14107e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 14110f7d52f4SChris Mason { 14127e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1413fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1414fc7cbcd4SDavid Sterba int i; 1415fc7cbcd4SDavid Sterba int ret; 141654aa1f4dSChris Mason 1417dfba78dcSFilipe Manana /* 1418dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1419dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1420dfba78dcSFilipe Manana */ 1421dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1422dfba78dcSFilipe Manana 1423fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1424fc7cbcd4SDavid Sterba while (1) { 1425fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1426fc7cbcd4SDavid Sterba (void **)gang, 0, 1427fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1428fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1429fc7cbcd4SDavid Sterba if (ret == 0) 1430fc7cbcd4SDavid Sterba break; 1431fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1432fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1433fc7cbcd4SDavid Sterba int ret2; 14344f4317c1SJosef Bacik 1435dfba78dcSFilipe Manana /* 1436dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1437dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1438dfba78dcSFilipe Manana */ 1439dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1440dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1441dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1442dfba78dcSFilipe Manana 1443fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14442619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14450f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1446fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 144731153d81SYan Zheng 1448e02119d5SChris Mason btrfs_free_log(trans, root); 1449fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1450fc7cbcd4SDavid Sterba if (ret2) 1451fc7cbcd4SDavid Sterba return ret2; 1452e02119d5SChris Mason 1453fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 145427cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1455c7548af6SChris Mason smp_mb__after_atomic(); 1456f1ebcc74SLiu Bo 1457978d910dSYan Zheng if (root->commit_root != root->node) { 14589e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14599e351cc8SJosef Bacik &trans->transaction->switch_commits); 1460fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1461fc7cbcd4SDavid Sterba root->node); 1462978d910dSYan Zheng } 146331153d81SYan Zheng 1464fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1465fc7cbcd4SDavid Sterba &root->root_key, 1466fc7cbcd4SDavid Sterba &root->root_item); 1467fc7cbcd4SDavid Sterba if (ret2) 1468fc7cbcd4SDavid Sterba return ret2; 1469fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1470733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14710f7d52f4SChris Mason } 1472fc7cbcd4SDavid Sterba } 1473fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14744f4317c1SJosef Bacik return 0; 14750f7d52f4SChris Mason } 14760f7d52f4SChris Mason 1477d352ac68SChris Mason /* 1478de78b51aSEric Sandeen * defrag a given btree. 1479de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1480d352ac68SChris Mason */ 1481de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1482e9d0b13bSChris Mason { 1483e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1484e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14858929ecfaSYan, Zheng int ret; 1486e9d0b13bSChris Mason 148727cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1488e9d0b13bSChris Mason return 0; 14898929ecfaSYan, Zheng 14906b80053dSChris Mason while (1) { 14918929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14926819703fSDavid Sterba if (IS_ERR(trans)) { 14936819703fSDavid Sterba ret = PTR_ERR(trans); 14946819703fSDavid Sterba break; 14956819703fSDavid Sterba } 14968929ecfaSYan, Zheng 1497de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14988929ecfaSYan, Zheng 14993a45bb20SJeff Mahoney btrfs_end_transaction(trans); 15002ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1501e9d0b13bSChris Mason cond_resched(); 1502e9d0b13bSChris Mason 1503ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1504e9d0b13bSChris Mason break; 1505210549ebSDavid Sterba 1506ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1507ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1508210549ebSDavid Sterba ret = -EAGAIN; 1509210549ebSDavid Sterba break; 1510210549ebSDavid Sterba } 1511e9d0b13bSChris Mason } 151227cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 15138929ecfaSYan, Zheng return ret; 1514e9d0b13bSChris Mason } 1515e9d0b13bSChris Mason 1516d352ac68SChris Mason /* 15176426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 15186426c7adSQu Wenruo * 15196426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 15206426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 15216426c7adSQu Wenruo * qgroup works. 15226426c7adSQu Wenruo */ 15236426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15246426c7adSQu Wenruo struct btrfs_root *src, 15256426c7adSQu Wenruo struct btrfs_root *parent, 15266426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15276426c7adSQu Wenruo u64 dst_objectid) 15286426c7adSQu Wenruo { 15296426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15306426c7adSQu Wenruo int ret; 15316426c7adSQu Wenruo 15326426c7adSQu Wenruo /* 15336426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15346426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15356426c7adSQu Wenruo * kick in anyway. 15366426c7adSQu Wenruo */ 15379ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15386426c7adSQu Wenruo return 0; 15396426c7adSQu Wenruo 15406426c7adSQu Wenruo /* 154152042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15424d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15434d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15444d31778aSQu Wenruo * item. 15454d31778aSQu Wenruo */ 15461c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15471c442d22SJosef Bacik if (ret) 15481c442d22SJosef Bacik return ret; 15494d31778aSQu Wenruo 15504d31778aSQu Wenruo /* 15512a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15522a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15532a4d84c1SJosef Bacik * 15542a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15552a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15562a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15572a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15582a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15592a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15602a4d84c1SJosef Bacik */ 15612a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15622a4d84c1SJosef Bacik if (ret) { 15632a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 156444365827SNaohiro Aota return ret; 15652a4d84c1SJosef Bacik } 15662a4d84c1SJosef Bacik 15677e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15686426c7adSQu Wenruo if (ret) 15696426c7adSQu Wenruo goto out; 1570460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15716426c7adSQu Wenruo if (ret < 0) 15726426c7adSQu Wenruo goto out; 15736426c7adSQu Wenruo 15746426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1575a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15766426c7adSQu Wenruo inherit); 15776426c7adSQu Wenruo if (ret < 0) 15786426c7adSQu Wenruo goto out; 15796426c7adSQu Wenruo 15806426c7adSQu Wenruo /* 15816426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15826426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15836426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15846426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15856426c7adSQu Wenruo * 2) write all btree blocks onto disk 15866426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15876426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15886426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15896426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15906426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15916426c7adSQu Wenruo */ 15929386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15936426c7adSQu Wenruo if (ret) 15946426c7adSQu Wenruo goto out; 1595889bfa39SJosef Bacik switch_commit_roots(trans); 159670458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15976426c7adSQu Wenruo if (ret) 1598f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15996426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 16006426c7adSQu Wenruo 16016426c7adSQu Wenruo out: 16026426c7adSQu Wenruo /* 16036426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 16046426c7adSQu Wenruo * last_trans == cur_transid. 16056426c7adSQu Wenruo * Or it won't be committed again onto disk after later 16066426c7adSQu Wenruo * insert_dir_item() 16076426c7adSQu Wenruo */ 16086426c7adSQu Wenruo if (!ret) 16091c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 16106426c7adSQu Wenruo return ret; 16116426c7adSQu Wenruo } 16126426c7adSQu Wenruo 16136426c7adSQu Wenruo /* 1614d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1615aec8030aSMiao Xie * transaction commit. This does the actual creation. 1616aec8030aSMiao Xie * 1617aec8030aSMiao Xie * Note: 1618aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1619aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1620aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1621d352ac68SChris Mason */ 162280b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16233063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16243063d29fSChris Mason { 162508d50ca3SNikolay Borisov 162608d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16273063d29fSChris Mason struct btrfs_key key; 162880b6794dSChris Mason struct btrfs_root_item *new_root_item; 16293063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16303063d29fSChris Mason struct btrfs_root *root = pending->root; 16316bdb72deSSage Weil struct btrfs_root *parent_root; 163298c9942aSLiu Bo struct btrfs_block_rsv *rsv; 1633ab3c5c18SSweet Tea Dorminy struct inode *parent_inode = pending->dir; 163442874b3dSMiao Xie struct btrfs_path *path; 163542874b3dSMiao Xie struct btrfs_dir_item *dir_item; 16363063d29fSChris Mason struct extent_buffer *tmp; 1637925baeddSChris Mason struct extent_buffer *old; 163895582b00SDeepa Dinamani struct timespec64 cur_time; 1639aec8030aSMiao Xie int ret = 0; 1640d68fc57bSYan, Zheng u64 to_reserve = 0; 16416bdb72deSSage Weil u64 index = 0; 1642a22285a6SYan, Zheng u64 objectid; 1643b83cc969SLi Zefan u64 root_flags; 1644ab3c5c18SSweet Tea Dorminy unsigned int nofs_flags; 1645ab3c5c18SSweet Tea Dorminy struct fscrypt_name fname; 16463063d29fSChris Mason 16478546b570SDavid Sterba ASSERT(pending->path); 16488546b570SDavid Sterba path = pending->path; 164942874b3dSMiao Xie 1650b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1651b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1652a22285a6SYan, Zheng 1653ab3c5c18SSweet Tea Dorminy /* 1654ab3c5c18SSweet Tea Dorminy * We're inside a transaction and must make sure that any potential 1655ab3c5c18SSweet Tea Dorminy * allocations with GFP_KERNEL in fscrypt won't recurse back to 1656ab3c5c18SSweet Tea Dorminy * filesystem. 1657ab3c5c18SSweet Tea Dorminy */ 1658ab3c5c18SSweet Tea Dorminy nofs_flags = memalloc_nofs_save(); 1659ab3c5c18SSweet Tea Dorminy pending->error = fscrypt_setup_filename(parent_inode, 1660ab3c5c18SSweet Tea Dorminy &pending->dentry->d_name, 0, 1661ab3c5c18SSweet Tea Dorminy &fname); 1662ab3c5c18SSweet Tea Dorminy memalloc_nofs_restore(nofs_flags); 1663ab3c5c18SSweet Tea Dorminy if (pending->error) 1664ab3c5c18SSweet Tea Dorminy goto free_pending; 1665ab3c5c18SSweet Tea Dorminy 1666543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1667aec8030aSMiao Xie if (pending->error) 1668ab3c5c18SSweet Tea Dorminy goto free_fname; 16693063d29fSChris Mason 1670d6726335SQu Wenruo /* 1671d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1672d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1673d6726335SQu Wenruo */ 1674d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1675d6726335SQu Wenruo 1676147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1677d68fc57bSYan, Zheng 1678d68fc57bSYan, Zheng if (to_reserve > 0) { 16799270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1680aec8030aSMiao Xie &pending->block_rsv, 168108e007d2SMiao Xie to_reserve, 168208e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1683aec8030aSMiao Xie if (pending->error) 1684d6726335SQu Wenruo goto clear_skip_qgroup; 1685d68fc57bSYan, Zheng } 1686d68fc57bSYan, Zheng 16873063d29fSChris Mason key.objectid = objectid; 1688a22285a6SYan, Zheng key.offset = (u64)-1; 1689a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16903063d29fSChris Mason 16916fa9700eSMiao Xie rsv = trans->block_rsv; 1692a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16932382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16940b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 169588d3a5aaSJosef Bacik trans->transid, 169688d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1697a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1698f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1699f0118cb6SJosef Bacik if (ret) 1700f0118cb6SJosef Bacik goto fail; 1701c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 170204b285f3SDeepa Dinamani 17036bdb72deSSage Weil /* 17046bdb72deSSage Weil * insert the directory item 17056bdb72deSSage Weil */ 1706877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 1707df9f2782SFilipe Manana if (ret) { 1708df9f2782SFilipe Manana btrfs_abort_transaction(trans, ret); 1709df9f2782SFilipe Manana goto fail; 1710df9f2782SFilipe Manana } 171142874b3dSMiao Xie 171242874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 171342874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 17144a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 17156db75318SSweet Tea Dorminy &fname.disk_name, 0); 171642874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1717fe66a05aSChris Mason pending->error = -EEXIST; 1718aec8030aSMiao Xie goto dir_item_existed; 171942874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 172042874b3dSMiao Xie ret = PTR_ERR(dir_item); 172166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17228732d44fSMiao Xie goto fail; 172379787eaaSJeff Mahoney } 172442874b3dSMiao Xie btrfs_release_path(path); 17256bdb72deSSage Weil 1726e999376fSChris Mason /* 1727e999376fSChris Mason * pull in the delayed directory update 1728e999376fSChris Mason * and the delayed inode item 1729e999376fSChris Mason * otherwise we corrupt the FS during 1730e999376fSChris Mason * snapshot 1731e999376fSChris Mason */ 1732e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 17338732d44fSMiao Xie if (ret) { /* Transaction aborted */ 173466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17358732d44fSMiao Xie goto fail; 17368732d44fSMiao Xie } 1737e999376fSChris Mason 1738f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1739f0118cb6SJosef Bacik if (ret) { 1740f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1741f0118cb6SJosef Bacik goto fail; 1742f0118cb6SJosef Bacik } 17436bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17446bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 174508fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17466bdb72deSSage Weil 1747b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1748b83cc969SLi Zefan if (pending->readonly) 1749b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1750b83cc969SLi Zefan else 1751b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1752b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1753b83cc969SLi Zefan 17548ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17558ea05e3aSAlexander Block trans->transid); 1756807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17578ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17588ea05e3aSAlexander Block BTRFS_UUID_SIZE); 175970023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 176070023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 176170023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17628ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17638ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17648ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17658ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 176670023da2SStefan Behrens } 17673cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17683cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 176970023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17708ea05e3aSAlexander Block 1771925baeddSChris Mason old = btrfs_lock_root_node(root); 17729631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17739631e4ccSJosef Bacik BTRFS_NESTING_COW); 177479787eaaSJeff Mahoney if (ret) { 177579787eaaSJeff Mahoney btrfs_tree_unlock(old); 177679787eaaSJeff Mahoney free_extent_buffer(old); 177766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17788732d44fSMiao Xie goto fail; 177979787eaaSJeff Mahoney } 178049b25e05SJeff Mahoney 178149b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 178279787eaaSJeff Mahoney /* clean up in any case */ 1783925baeddSChris Mason btrfs_tree_unlock(old); 1784925baeddSChris Mason free_extent_buffer(old); 17858732d44fSMiao Xie if (ret) { 178666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17878732d44fSMiao Xie goto fail; 17888732d44fSMiao Xie } 1789f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 179027cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1791f1ebcc74SLiu Bo smp_wmb(); 1792f1ebcc74SLiu Bo 17935d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1794a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1795a22285a6SYan, Zheng key.offset = trans->transid; 1796a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1797925baeddSChris Mason btrfs_tree_unlock(tmp); 17983063d29fSChris Mason free_extent_buffer(tmp); 17998732d44fSMiao Xie if (ret) { 180066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18018732d44fSMiao Xie goto fail; 18028732d44fSMiao Xie } 18030660b5afSChris Mason 1804a22285a6SYan, Zheng /* 1805a22285a6SYan, Zheng * insert root back/forward references 1806a22285a6SYan, Zheng */ 18076025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1808a22285a6SYan, Zheng parent_root->root_key.objectid, 18094a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 18106db75318SSweet Tea Dorminy &fname.disk_name); 18118732d44fSMiao Xie if (ret) { 181266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18138732d44fSMiao Xie goto fail; 18148732d44fSMiao Xie } 1815a22285a6SYan, Zheng 1816a22285a6SYan, Zheng key.offset = (u64)-1; 18172dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 181879787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 181979787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 18202d892ccdSFilipe Manana pending->snap = NULL; 182166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18228732d44fSMiao Xie goto fail; 182379787eaaSJeff Mahoney } 1824d68fc57bSYan, Zheng 182549b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 18268732d44fSMiao Xie if (ret) { 182766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18288732d44fSMiao Xie goto fail; 18298732d44fSMiao Xie } 1830361048f5SMiao Xie 18316426c7adSQu Wenruo /* 18326426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 18336426c7adSQu Wenruo * snapshot hack to do fast snapshot. 18346426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 18356426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 18366426c7adSQu Wenruo */ 18376426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18386426c7adSQu Wenruo pending->inherit, objectid); 18396426c7adSQu Wenruo if (ret < 0) 18406426c7adSQu Wenruo goto fail; 18416426c7adSQu Wenruo 18426db75318SSweet Tea Dorminy ret = btrfs_insert_dir_item(trans, &fname.disk_name, 18436db75318SSweet Tea Dorminy BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, 18446db75318SSweet Tea Dorminy index); 184542874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18469c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18478732d44fSMiao Xie if (ret) { 184866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18498732d44fSMiao Xie goto fail; 18508732d44fSMiao Xie } 185142874b3dSMiao Xie 18526ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 18536db75318SSweet Tea Dorminy fname.disk_name.len * 2); 1854c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1855c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1856729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1857dd5f9615SStefan Behrens if (ret) { 185866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1859dd5f9615SStefan Behrens goto fail; 1860dd5f9615SStefan Behrens } 1861807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1862807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1863cdb345a8SLu Fengqi objectid); 1864dd5f9615SStefan Behrens if (ret) { 186566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1866dd5f9615SStefan Behrens goto fail; 1867dd5f9615SStefan Behrens } 1868dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1869cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1870dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1871dd5f9615SStefan Behrens objectid); 1872dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 187366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1874dd5f9615SStefan Behrens goto fail; 1875dd5f9615SStefan Behrens } 1876dd5f9615SStefan Behrens } 1877d6726335SQu Wenruo 18783063d29fSChris Mason fail: 1879aec8030aSMiao Xie pending->error = ret; 1880aec8030aSMiao Xie dir_item_existed: 188198c9942aSLiu Bo trans->block_rsv = rsv; 18822382c5ccSLiu Bo trans->bytes_reserved = 0; 1883d6726335SQu Wenruo clear_skip_qgroup: 1884d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 1885ab3c5c18SSweet Tea Dorminy free_fname: 1886ab3c5c18SSweet Tea Dorminy fscrypt_free_filename(&fname); 1887ab3c5c18SSweet Tea Dorminy free_pending: 18886fa9700eSMiao Xie kfree(new_root_item); 1889b0c0ea63SDavid Sterba pending->root_item = NULL; 189042874b3dSMiao Xie btrfs_free_path(path); 18918546b570SDavid Sterba pending->path = NULL; 18928546b570SDavid Sterba 189349b25e05SJeff Mahoney return ret; 18943063d29fSChris Mason } 18953063d29fSChris Mason 1896d352ac68SChris Mason /* 1897d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1898d352ac68SChris Mason */ 189908d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 19003063d29fSChris Mason { 1901aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 19023063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1903aec8030aSMiao Xie int ret = 0; 19043de4586cSChris Mason 1905aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1906aec8030aSMiao Xie list_del(&pending->list); 190708d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1908aec8030aSMiao Xie if (ret) 1909aec8030aSMiao Xie break; 1910aec8030aSMiao Xie } 1911aec8030aSMiao Xie return ret; 19123de4586cSChris Mason } 19133de4586cSChris Mason 19142ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 19155d4f98a2SYan Zheng { 19165d4f98a2SYan Zheng struct btrfs_root_item *root_item; 19175d4f98a2SYan Zheng struct btrfs_super_block *super; 19185d4f98a2SYan Zheng 19190b246afaSJeff Mahoney super = fs_info->super_copy; 19205d4f98a2SYan Zheng 19210b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1922093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1923093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1924093e037cSDavid Sterba super->chunk_root_level = root_item->level; 19255d4f98a2SYan Zheng 19260b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1927093e037cSDavid Sterba super->root = root_item->bytenr; 1928093e037cSDavid Sterba super->generation = root_item->generation; 1929093e037cSDavid Sterba super->root_level = root_item->level; 19300b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1931093e037cSDavid Sterba super->cache_generation = root_item->generation; 193294846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 193394846229SBoris Burkov super->cache_generation = 0; 19340b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1935093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 19365d4f98a2SYan Zheng } 19375d4f98a2SYan Zheng 1938f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1939f36f3042SChris Mason { 19404a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1941f36f3042SChris Mason int ret = 0; 19424a9d8bdeSMiao Xie 1943a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19444a9d8bdeSMiao Xie trans = info->running_transaction; 19454a9d8bdeSMiao Xie if (trans) 19464a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1947a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1948f36f3042SChris Mason return ret; 1949f36f3042SChris Mason } 1950f36f3042SChris Mason 19518929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19528929ecfaSYan, Zheng { 19534a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19548929ecfaSYan, Zheng int ret = 0; 19554a9d8bdeSMiao Xie 1956a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19574a9d8bdeSMiao Xie trans = info->running_transaction; 19584a9d8bdeSMiao Xie if (trans) 19594a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1960a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19618929ecfaSYan, Zheng return ret; 19628929ecfaSYan, Zheng } 19638929ecfaSYan, Zheng 1964fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1965bb9c12c9SSage Weil { 19663a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1967bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1968bb9c12c9SSage Weil 1969fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1970fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1971fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1972bb9c12c9SSage Weil 1973bb9c12c9SSage Weil /* take transaction reference */ 1974bb9c12c9SSage Weil cur_trans = trans->transaction; 19759b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1976bb9c12c9SSage Weil 19773a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19786fc4e354SSage Weil 19796fc4e354SSage Weil /* 1980ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1981ae5d29d4SDavid Sterba * subsequent transaction joins 1982ae5d29d4SDavid Sterba */ 19833e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1984ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1985ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1986ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1987724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1988bb9c12c9SSage Weil } 1989bb9c12c9SSage Weil 199097cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 199149b25e05SJeff Mahoney { 199297cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 199349b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 199449b25e05SJeff Mahoney 1995b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 199649b25e05SJeff Mahoney 199766642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19987b8b92afSJosef Bacik 19990b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 200066b6135bSLiu Bo 200125d8c284SMiao Xie /* 200225d8c284SMiao Xie * If the transaction is removed from the list, it means this 200325d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 200425d8c284SMiao Xie * to call the cleanup function. 200525d8c284SMiao Xie */ 200625d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 200766b6135bSLiu Bo 20080b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 20094a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 20100b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2011e1489b4fSIoannis Angelakopoulos 2012e1489b4fSIoannis Angelakopoulos /* 2013e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 2014e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 2015e1489b4fSIoannis Angelakopoulos */ 2016e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2017f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 2018f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 2019f094ac32SLiu Bo 20200b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 2021d7096fc3SJosef Bacik } 2022061dde82SFilipe Manana 2023061dde82SFilipe Manana /* 2024061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 2025061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 2026061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 2027061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 2028061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 2029061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 2030061dde82SFilipe Manana */ 2031061dde82SFilipe Manana list_del_init(&cur_trans->list); 2032061dde82SFilipe Manana 20330b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 203449b25e05SJeff Mahoney 20352ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 203649b25e05SJeff Mahoney 20370b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20380b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20390b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20400b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20414a9d8bdeSMiao Xie 2042e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20430b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2044724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2045724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 204649b25e05SJeff Mahoney 20472e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 204849b25e05SJeff Mahoney 204949b25e05SJeff Mahoney if (current->journal_info == trans) 205049b25e05SJeff Mahoney current->journal_info = NULL; 20512d82a40aSFilipe Manana 20522d82a40aSFilipe Manana /* 20532d82a40aSFilipe Manana * If relocation is running, we can't cancel scrub because that will 20542d82a40aSFilipe Manana * result in a deadlock. Before relocating a block group, relocation 20552d82a40aSFilipe Manana * pauses scrub, then starts and commits a transaction before unpausing 20562d82a40aSFilipe Manana * scrub. If the transaction commit is being done by the relocation 20572d82a40aSFilipe Manana * task or triggered by another task and the relocation task is waiting 20582d82a40aSFilipe Manana * for the commit, and we end up here due to an error in the commit 20592d82a40aSFilipe Manana * path, then calling btrfs_scrub_cancel() will deadlock, as we are 20602d82a40aSFilipe Manana * asking for scrub to stop while having it asked to be paused higher 20612d82a40aSFilipe Manana * above in relocation code. 20622d82a40aSFilipe Manana */ 20632d82a40aSFilipe Manana if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 20640b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 206549b25e05SJeff Mahoney 206649b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 206749b25e05SJeff Mahoney } 206849b25e05SJeff Mahoney 2069c7cc64a9SDavid Sterba /* 2070c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2071c7cc64a9SDavid Sterba * transaction and remove them from the list 2072c7cc64a9SDavid Sterba */ 2073c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2074c7cc64a9SDavid Sterba { 2075c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 207632da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2077c7cc64a9SDavid Sterba 2078c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2079c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2080c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2081c7cc64a9SDavid Sterba } 2082c7cc64a9SDavid Sterba } 2083c7cc64a9SDavid Sterba 208488090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 208582436617SMiao Xie { 2086ce8ea7ccSJosef Bacik /* 2087a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2088ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2089ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2090ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2091ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2092ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2093ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2094a0f0cf83SFilipe Manana * 2095a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2096a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2097a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2098a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2099a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2100a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2101a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2102ce8ea7ccSJosef Bacik */ 210388090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2104a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 210582436617SMiao Xie return 0; 210682436617SMiao Xie } 210782436617SMiao Xie 210888090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 210982436617SMiao Xie { 211088090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 21116374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 211282436617SMiao Xie } 211382436617SMiao Xie 211428b21c55SFilipe Manana /* 211528b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 211628b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 211728b21c55SFilipe Manana * and while holding fs_info->trans_lock. 211828b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 211928b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 212028b21c55SFilipe Manana * returns an error. 212128b21c55SFilipe Manana */ 212228b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 212328b21c55SFilipe Manana { 212428b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 212528b21c55SFilipe Manana 212628b21c55SFilipe Manana if (!trans->pending_snapshot) 212728b21c55SFilipe Manana return; 212828b21c55SFilipe Manana 212928b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 213028b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 213128b21c55SFilipe Manana 213228b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 213328b21c55SFilipe Manana } 213428b21c55SFilipe Manana 2135e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2136e55958c8SIoannis Angelakopoulos { 2137e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2138e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2139e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2140e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2141e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2142e55958c8SIoannis Angelakopoulos } 2143e55958c8SIoannis Angelakopoulos 21443a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 214579154b1bSChris Mason { 21463a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 214749b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 21488fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 214925287e0aSMiao Xie int ret; 2150e55958c8SIoannis Angelakopoulos ktime_t start_time; 2151e55958c8SIoannis Angelakopoulos ktime_t interval; 215279154b1bSChris Mason 215335b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21543e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 215535b814f3SNikolay Borisov 2156c52cc7b7SJosef Bacik clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); 2157c52cc7b7SJosef Bacik 21588d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2159bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 216025287e0aSMiao Xie ret = cur_trans->aborted; 21613e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 216225287e0aSMiao Xie } 216349b25e05SJeff Mahoney 2164f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2165f45c752bSJosef Bacik trans->block_rsv = NULL; 2166f45c752bSJosef Bacik 2167e19eb11fSJosef Bacik /* 2168e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2169e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2170e19eb11fSJosef Bacik */ 2171e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2172e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2173e19eb11fSJosef Bacik /* 2174e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2175e19eb11fSJosef Bacik * Any running threads may add more while we are here. 217656bec294SChris Mason */ 2177c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21783e738c53SIoannis Angelakopoulos if (ret) 21793e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2180e19eb11fSJosef Bacik } 218156bec294SChris Mason 21826c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2183ea658badSJosef Bacik 21843204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21851bbc621eSChris Mason int run_it = 0; 21861bbc621eSChris Mason 21871bbc621eSChris Mason /* this mutex is also taken before trying to set 21881bbc621eSChris Mason * block groups readonly. We need to make sure 21891bbc621eSChris Mason * that nobody has set a block group readonly 21901bbc621eSChris Mason * after a extents from that block group have been 21911bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21921bbc621eSChris Mason * will wait for the transaction to commit if it 21933204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21941bbc621eSChris Mason * 21953204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21963204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21971bbc621eSChris Mason * hurt to have more than one go through, but there's no 21981bbc621eSChris Mason * real advantage to it either. 21991bbc621eSChris Mason */ 22000b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 22013204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 22023204d33cSJosef Bacik &cur_trans->flags)) 22031bbc621eSChris Mason run_it = 1; 22040b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 22051bbc621eSChris Mason 2206f9cacae3SNikolay Borisov if (run_it) { 220721217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 22083e738c53SIoannis Angelakopoulos if (ret) 22093e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2210f9cacae3SNikolay Borisov } 2211f9cacae3SNikolay Borisov } 22121bbc621eSChris Mason 22130b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 22144a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2215d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2216d0c2f4faSFilipe Manana 221728b21c55SFilipe Manana add_pending_snapshot(trans); 221828b21c55SFilipe Manana 22190b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 22209b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2221ccd467d6SChris Mason 2222d0c2f4faSFilipe Manana if (trans->in_fsync) 2223d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 22243e738c53SIoannis Angelakopoulos 22253e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 22263e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2227d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2228d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 222915ee9bc7SJosef Bacik 2230bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2231b4924a0fSLiu Bo ret = cur_trans->aborted; 2232b4924a0fSLiu Bo 2233724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 223415ee9bc7SJosef Bacik 223549b25e05SJeff Mahoney return ret; 223679154b1bSChris Mason } 22374313b399SChris Mason 22384a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 22390b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 22403e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2241bb9c12c9SSage Weil 22420b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2243d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2244d0c2f4faSFilipe Manana 2245d0c2f4faSFilipe Manana if (trans->in_fsync) 2246d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2247d0c2f4faSFilipe Manana 2248ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2249ccd467d6SChris Mason struct btrfs_transaction, list); 2250d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 22519b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 22520b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2253ccd467d6SChris Mason 2254d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2255d0c2f4faSFilipe Manana 2256bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2257ccd467d6SChris Mason 2258724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22591f9b8c8fSFilipe Manana if (ret) 2260e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2261a4abeea4SJosef Bacik } else { 22620b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2263ccd467d6SChris Mason } 2264a4abeea4SJosef Bacik } else { 22650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2266cb2d3dadSFilipe Manana /* 2267cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2268cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2269cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2270cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2271cb2d3dadSFilipe Manana */ 227284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2273cb2d3dadSFilipe Manana ret = -EROFS; 2274e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2275cb2d3dadSFilipe Manana } 2276ccd467d6SChris Mason } 227715ee9bc7SJosef Bacik 2278e55958c8SIoannis Angelakopoulos /* 2279e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2280e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2281e55958c8SIoannis Angelakopoulos */ 2282e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2283e55958c8SIoannis Angelakopoulos 22840860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22850860adfdSMiao Xie 228688090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 228782436617SMiao Xie if (ret) 2288e1489b4fSIoannis Angelakopoulos goto lockdep_release; 228982436617SMiao Xie 2290e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 229149b25e05SJeff Mahoney if (ret) 2292e1489b4fSIoannis Angelakopoulos goto lockdep_release; 229316cdcec7SMiao Xie 22945a9ba670SIoannis Angelakopoulos /* 22955a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22965a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22975a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22985a9ba670SIoannis Angelakopoulos */ 22995a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 23005a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2301581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2302581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2303ed3b3d31SChris Mason 2304581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2305e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2306e1489b4fSIoannis Angelakopoulos if (ret) { 2307e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2308ca469637SMiao Xie goto cleanup_transaction; 2309e1489b4fSIoannis Angelakopoulos } 2310ca469637SMiao Xie 231188090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2312cb7ab021SWang Shilong 231348778179SFilipe Manana /* 231448778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 231548778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 231648778179SFilipe Manana * extents complete we lose logged data after a power failure. 231748778179SFilipe Manana */ 23188b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 231948778179SFilipe Manana wait_event(cur_trans->pending_wait, 232048778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 232148778179SFilipe Manana 23222ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2323ed0ca140SJosef Bacik /* 2324ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2325ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 23264a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2327ed0ca140SJosef Bacik */ 23280b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 232928b21c55SFilipe Manana add_pending_snapshot(trans); 23304a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 23310b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2332e1489b4fSIoannis Angelakopoulos 2333e1489b4fSIoannis Angelakopoulos /* 2334e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2335e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2336e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2337e1489b4fSIoannis Angelakopoulos */ 2338e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2339e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2340ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2341ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 234215ee9bc7SJosef Bacik 2343fdfbf020SJosef Bacik /* 23443e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 23453e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 23463e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 23473e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 23483e738c53SIoannis Angelakopoulos */ 23493e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 23503e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 23513e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23523e738c53SIoannis Angelakopoulos 23533e738c53SIoannis Angelakopoulos /* 2354fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2355fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2356fdfbf020SJosef Bacik * kthread could do the work. 2357fdfbf020SJosef Bacik */ 2358fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2359fdfbf020SJosef Bacik 2360bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23612cba30f1SMiao Xie ret = cur_trans->aborted; 23623e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23636cf7f77eSWang Shilong goto scrub_continue; 23642cba30f1SMiao Xie } 23657585717fSChris Mason /* 23667585717fSChris Mason * the reloc mutex makes sure that we stop 23677585717fSChris Mason * the balancing code from coming in and moving 23687585717fSChris Mason * extents around in the middle of the commit 23697585717fSChris Mason */ 23700b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23717585717fSChris Mason 237242874b3dSMiao Xie /* 237342874b3dSMiao Xie * We needn't worry about the delayed items because we will 237442874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 237542874b3dSMiao Xie * core function of the snapshot creation. 237642874b3dSMiao Xie */ 237708d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 237856e9f6eaSDavid Sterba if (ret) 237956e9f6eaSDavid Sterba goto unlock_reloc; 23803063d29fSChris Mason 238142874b3dSMiao Xie /* 238242874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 238342874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 238442874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 238542874b3dSMiao Xie * them. 238642874b3dSMiao Xie * 238742874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 238842874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 238942874b3dSMiao Xie * the nodes and leaves. 239042874b3dSMiao Xie */ 2391e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 239256e9f6eaSDavid Sterba if (ret) 239356e9f6eaSDavid Sterba goto unlock_reloc; 239416cdcec7SMiao Xie 2395c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 239656e9f6eaSDavid Sterba if (ret) 239756e9f6eaSDavid Sterba goto unlock_reloc; 239856bec294SChris Mason 2399e999376fSChris Mason /* 2400e999376fSChris Mason * make sure none of the code above managed to slip in a 2401e999376fSChris Mason * delayed item 2402e999376fSChris Mason */ 2403ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2404e999376fSChris Mason 24052c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2406dc17ff8fSChris Mason 24077e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 240856e9f6eaSDavid Sterba if (ret) 2409dfba78dcSFilipe Manana goto unlock_reloc; 241054aa1f4dSChris Mason 24115d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2412e02119d5SChris Mason * safe to free the root of tree log roots 2413e02119d5SChris Mason */ 24140b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2415e02119d5SChris Mason 24160ed4792aSQu Wenruo /* 24170ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 24180ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 24190ed4792aSQu Wenruo */ 2420460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 242156e9f6eaSDavid Sterba if (ret < 0) 2422dfba78dcSFilipe Manana goto unlock_reloc; 24230ed4792aSQu Wenruo 24249386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 242556e9f6eaSDavid Sterba if (ret) 2426dfba78dcSFilipe Manana goto unlock_reloc; 242754aa1f4dSChris Mason 24282cba30f1SMiao Xie /* 24292cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 24302cba30f1SMiao Xie * update ->aborted, check it. 24312cba30f1SMiao Xie */ 2432bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 24332cba30f1SMiao Xie ret = cur_trans->aborted; 2434dfba78dcSFilipe Manana goto unlock_reloc; 24352cba30f1SMiao Xie } 24362cba30f1SMiao Xie 24370b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 24385f39d397SChris Mason 24390b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 24400b246afaSJeff Mahoney fs_info->tree_root->node); 24410b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 24429e351cc8SJosef Bacik &cur_trans->switch_commits); 24435d4f98a2SYan Zheng 24440b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 24450b246afaSJeff Mahoney fs_info->chunk_root->node); 24460b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 24479e351cc8SJosef Bacik &cur_trans->switch_commits); 24489e351cc8SJosef Bacik 2449f7238e50SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2450f7238e50SJosef Bacik btrfs_set_root_node(&fs_info->block_group_root->root_item, 2451f7238e50SJosef Bacik fs_info->block_group_root->node); 2452f7238e50SJosef Bacik list_add_tail(&fs_info->block_group_root->dirty_list, 2453f7238e50SJosef Bacik &cur_trans->switch_commits); 2454f7238e50SJosef Bacik } 2455f7238e50SJosef Bacik 2456889bfa39SJosef Bacik switch_commit_roots(trans); 24575d4f98a2SYan Zheng 2458ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 24591bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 24602ff7e61eSJeff Mahoney update_super_roots(fs_info); 2461e02119d5SChris Mason 24620b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 24630b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 24640b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 24650b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2466ccd467d6SChris Mason 2467bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2468935e5cc9SMiao Xie 24690b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 24700b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2471656f30dbSFilipe Manana 24724fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24734fbcdf66SFilipe Manana 2474dfba78dcSFilipe Manana /* 2475dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2476dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2477dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2478dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2479dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2480dfba78dcSFilipe Manana * writing its superblock. 2481dfba78dcSFilipe Manana */ 2482dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2483dfba78dcSFilipe Manana 24840b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 24854a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 24860b246afaSJeff Mahoney fs_info->running_transaction = NULL; 24870b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 24880b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2489b7ec40d7SChris Mason 24900b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 24913e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2492e6dcd2dcSChris Mason 2493b7625f46SQu Wenruo /* If we have features changed, wake up the cleaner to update sysfs. */ 2494b7625f46SQu Wenruo if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) && 2495b7625f46SQu Wenruo fs_info->cleaner_kthread) 2496b7625f46SQu Wenruo wake_up_process(fs_info->cleaner_kthread); 2497b7625f46SQu Wenruo 249870458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 249949b25e05SJeff Mahoney if (ret) { 25000b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 250108748810SDavid Sterba "Error while writing out transaction"); 25020b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 25036cf7f77eSWang Shilong goto scrub_continue; 250449b25e05SJeff Mahoney } 250549b25e05SJeff Mahoney 2506eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2507e02119d5SChris Mason /* 2508e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2509e02119d5SChris Mason * to go about their business 2510e02119d5SChris Mason */ 25110b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2512c1f32b7cSAnand Jain if (ret) 2513c1f32b7cSAnand Jain goto scrub_continue; 2514e02119d5SChris Mason 2515d0c2f4faSFilipe Manana /* 2516d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2517d0c2f4faSFilipe Manana * which can change it. 2518d0c2f4faSFilipe Manana */ 2519d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2520d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 25213e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2522d0c2f4faSFilipe Manana 25235ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 25244313b399SChris Mason 25253204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 25260b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 252713212b54SZhao Lei 25280b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 25294a9d8bdeSMiao Xie /* 25304a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 25314a9d8bdeSMiao Xie * which can change it. 25324a9d8bdeSMiao Xie */ 25334a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 25342c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 25353e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25363de4586cSChris Mason 25370b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 253813c5a93eSJosef Bacik list_del_init(&cur_trans->list); 25390b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2540a4abeea4SJosef Bacik 2541724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2542724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 254358176a96SJosef Bacik 25440860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 25450b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2546b2b5ef5cSJan Kara 25472e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 25481abe9b8aSliubo 2549e55958c8SIoannis Angelakopoulos interval = ktime_get_ns() - start_time; 2550e55958c8SIoannis Angelakopoulos 25512ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2552a2de733cSArne Jansen 25539ed74f2dSJosef Bacik if (current->journal_info == trans) 25549ed74f2dSJosef Bacik current->journal_info = NULL; 25559ed74f2dSJosef Bacik 25562c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 255724bbcf04SYan, Zheng 2558e55958c8SIoannis Angelakopoulos update_commit_stats(fs_info, interval); 2559e55958c8SIoannis Angelakopoulos 256079154b1bSChris Mason return ret; 256149b25e05SJeff Mahoney 256256e9f6eaSDavid Sterba unlock_reloc: 256356e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 25643e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 25656cf7f77eSWang Shilong scrub_continue: 25663e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 25673e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25682ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 256949b25e05SJeff Mahoney cleanup_transaction: 2570dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2571c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 25724fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 25730e721106SJosef Bacik trans->block_rsv = NULL; 25740b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 257549b25e05SJeff Mahoney if (current->journal_info == trans) 257649b25e05SJeff Mahoney current->journal_info = NULL; 257797cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 257849b25e05SJeff Mahoney 257949b25e05SJeff Mahoney return ret; 2580e1489b4fSIoannis Angelakopoulos 2581e1489b4fSIoannis Angelakopoulos lockdep_release: 25825a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2583e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2584e1489b4fSIoannis Angelakopoulos goto cleanup_transaction; 25853e738c53SIoannis Angelakopoulos 25863e738c53SIoannis Angelakopoulos lockdep_trans_commit_start_release: 25873e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 25883e738c53SIoannis Angelakopoulos btrfs_end_transaction(trans); 25893e738c53SIoannis Angelakopoulos return ret; 259079154b1bSChris Mason } 259179154b1bSChris Mason 2592d352ac68SChris Mason /* 25939d1a2a3aSDavid Sterba * return < 0 if error 25949d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 25959d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 25969d1a2a3aSDavid Sterba * 25979d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 25989d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 25999d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 26009d1a2a3aSDavid Sterba * few seconds later. 2601d352ac68SChris Mason */ 260233c44184SJosef Bacik int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2603e9d0b13bSChris Mason { 260433c44184SJosef Bacik struct btrfs_root *root; 26059d1a2a3aSDavid Sterba int ret; 2606e9d0b13bSChris Mason 2607a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 26089d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 26099d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 26109d1a2a3aSDavid Sterba return 0; 26119d1a2a3aSDavid Sterba } 26129d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 26139d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2614cfad392bSJosef Bacik list_del_init(&root->root_list); 2615a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 26165d4f98a2SYan Zheng 26174fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 261876dda93cSYan, Zheng 261916cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 262016cdcec7SMiao Xie 262176dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 262276dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 26230078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 262476dda93cSYan, Zheng else 26250078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 262632471dc2SDavid Sterba 2627dc9492c1SJosef Bacik btrfs_put_root(root); 26286596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2629e9d0b13bSChris Mason } 2630572d9ab7SDavid Sterba 2631fccf0c84SJosef Bacik /* 2632fccf0c84SJosef Bacik * We only mark the transaction aborted and then set the file system read-only. 2633fccf0c84SJosef Bacik * This will prevent new transactions from starting or trying to join this 2634fccf0c84SJosef Bacik * one. 2635fccf0c84SJosef Bacik * 2636fccf0c84SJosef Bacik * This means that error recovery at the call site is limited to freeing 2637fccf0c84SJosef Bacik * any local memory allocations and passing the error code up without 2638fccf0c84SJosef Bacik * further cleanup. The transaction should complete as it normally would 2639fccf0c84SJosef Bacik * in the call path but will return -EIO. 2640fccf0c84SJosef Bacik * 2641fccf0c84SJosef Bacik * We'll complete the cleanup in btrfs_end_transaction and 2642fccf0c84SJosef Bacik * btrfs_commit_transaction. 2643fccf0c84SJosef Bacik */ 2644fccf0c84SJosef Bacik void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, 2645fccf0c84SJosef Bacik const char *function, 2646fccf0c84SJosef Bacik unsigned int line, int errno, bool first_hit) 2647fccf0c84SJosef Bacik { 2648fccf0c84SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 2649fccf0c84SJosef Bacik 2650fccf0c84SJosef Bacik WRITE_ONCE(trans->aborted, errno); 2651fccf0c84SJosef Bacik WRITE_ONCE(trans->transaction->aborted, errno); 2652fccf0c84SJosef Bacik if (first_hit && errno == -ENOSPC) 2653fccf0c84SJosef Bacik btrfs_dump_space_info_for_trans_abort(fs_info); 2654fccf0c84SJosef Bacik /* Wake up anybody who may be waiting on this transaction */ 2655fccf0c84SJosef Bacik wake_up(&fs_info->transaction_wait); 2656fccf0c84SJosef Bacik wake_up(&fs_info->transaction_blocked_wait); 2657fccf0c84SJosef Bacik __btrfs_handle_fs_error(fs_info, function, line, errno, NULL); 2658fccf0c84SJosef Bacik } 2659fccf0c84SJosef Bacik 2660956504a3SJosef Bacik int __init btrfs_transaction_init(void) 2661956504a3SJosef Bacik { 2662956504a3SJosef Bacik btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2663956504a3SJosef Bacik sizeof(struct btrfs_trans_handle), 0, 2664956504a3SJosef Bacik SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2665956504a3SJosef Bacik if (!btrfs_trans_handle_cachep) 2666956504a3SJosef Bacik return -ENOMEM; 2667956504a3SJosef Bacik return 0; 2668956504a3SJosef Bacik } 2669956504a3SJosef Bacik 2670956504a3SJosef Bacik void __cold btrfs_transaction_exit(void) 2671956504a3SJosef Bacik { 2672956504a3SJosef Bacik kmem_cache_destroy(btrfs_trans_handle_cachep); 2673956504a3SJosef Bacik } 2674