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" 34*7572dec8SJosef Bacik #include "ioctl.h" 3579154b1bSChris Mason 36956504a3SJosef Bacik static struct kmem_cache *btrfs_trans_handle_cachep; 37956504a3SJosef Bacik 38fc7cbcd4SDavid Sterba #define BTRFS_ROOT_TRANS_TAG 0 390f7d52f4SChris Mason 4061c047b5SQu Wenruo /* 4161c047b5SQu Wenruo * Transaction states and transitions 4261c047b5SQu Wenruo * 4361c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 4461c047b5SQu Wenruo * | 4561c047b5SQu Wenruo * | To next stage: 4661c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 4761c047b5SQu Wenruo * V 4861c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 4961c047b5SQu Wenruo * | 5061c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 5161c047b5SQu Wenruo * | start_transaction() variants. 5261c047b5SQu Wenruo * | 5361c047b5SQu Wenruo * | To next stage: 5461c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 5561c047b5SQu Wenruo * | transaction N 5661c047b5SQu Wenruo * V 5761c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 5861c047b5SQu Wenruo * | 5961c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 6061c047b5SQu Wenruo * | is one 6161c047b5SQu Wenruo * | 6261c047b5SQu Wenruo * | Then one of the following happes: 6361c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 6461c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 6561c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 6661c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 6761c047b5SQu Wenruo * | 6861c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 6961c047b5SQu Wenruo * | to this running transaction. 7061c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 7161c047b5SQu Wenruo * | transaction N+1. 7261c047b5SQu Wenruo * | 7361c047b5SQu Wenruo * | To next stage: 7461c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 7561c047b5SQu Wenruo * | haven been released. 7661c047b5SQu Wenruo * V 7761c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 7861c047b5SQu Wenruo * | 7961c047b5SQu Wenruo * | The heavy lifting transaction work is started. 8061c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 8161c047b5SQu Wenruo * | snapshots, running qgroups. 8261c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 8361c047b5SQu Wenruo * | trees. 8461c047b5SQu Wenruo * | 8561c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 8661c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 8761c047b5SQu Wenruo * | 8861c047b5SQu Wenruo * | To next stage: 8961c047b5SQu Wenruo * | Until all supporting trees are updated. 9061c047b5SQu Wenruo * V 9161c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 9261c047b5SQu Wenruo * | Transaction N+1 9361c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 9461c047b5SQu Wenruo * | need to write them back to disk and update | 9561c047b5SQu Wenruo * | super blocks. | 9661c047b5SQu Wenruo * | | 9761c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 9861c047b5SQu Wenruo * | start. | 9961c047b5SQu Wenruo * | All new start_transaction() calls will be | 10061c047b5SQu Wenruo * | attached to transid N+1. | 10161c047b5SQu Wenruo * | | 10261c047b5SQu Wenruo * | To next stage: | 10361c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 10461c047b5SQu Wenruo * | written to block devices | 10561c047b5SQu Wenruo * V | 10661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 10761c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 10861c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 10961c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 11061c047b5SQu Wenruo */ 111e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1124a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 113bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 114bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1154a9d8bdeSMiao Xie __TRANS_ATTACH | 116a6d155d2SFilipe Manana __TRANS_JOIN | 117a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 118bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1194a9d8bdeSMiao Xie __TRANS_ATTACH | 1204a9d8bdeSMiao Xie __TRANS_JOIN | 121a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 122a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 123d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 124d0c2f4faSFilipe Manana __TRANS_ATTACH | 125d0c2f4faSFilipe Manana __TRANS_JOIN | 126d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 127d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 128bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1294a9d8bdeSMiao Xie __TRANS_ATTACH | 1304a9d8bdeSMiao Xie __TRANS_JOIN | 131a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 132a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1334a9d8bdeSMiao Xie }; 1344a9d8bdeSMiao Xie 135724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 13679154b1bSChris Mason { 1379b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1389b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 139a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1405c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1415c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 14281f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 14381f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1441262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 145ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 146ab8d0fc4SJeff Mahoney "pending csums is %llu", 1471262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1487785a663SFilipe Manana /* 1497785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1507785a663SFilipe Manana * because the transaction was aborted and a commit did not 1517785a663SFilipe Manana * happen (things failed before writing the new superblock 1527785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1537785a663SFilipe Manana * discard the physical locations of the block groups. 1547785a663SFilipe Manana */ 1557785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 15632da5386SDavid Sterba struct btrfs_block_group *cache; 1577785a663SFilipe Manana 1587785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 15932da5386SDavid Sterba struct btrfs_block_group, 1607785a663SFilipe Manana bg_list); 1617785a663SFilipe Manana list_del_init(&cache->bg_list); 1626b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1637785a663SFilipe Manana btrfs_put_block_group(cache); 1647785a663SFilipe Manana } 165bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1664b5faeacSDavid Sterba kfree(transaction); 16779154b1bSChris Mason } 16878fae27eSChris Mason } 16979154b1bSChris Mason 170889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 171817d52f8SJosef Bacik { 172889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 17316916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1749e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 1759e351cc8SJosef Bacik 176dfba78dcSFilipe Manana /* 177dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 178dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 179dfba78dcSFilipe Manana */ 180dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 181dfba78dcSFilipe Manana 1829e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 183d96b3424SFilipe Manana 184d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 185d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 186d96b3424SFilipe Manana 187889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1889e351cc8SJosef Bacik dirty_list) { 1899e351cc8SJosef Bacik list_del_init(&root->dirty_list); 190817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 191817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 19241e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 193370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1949e351cc8SJosef Bacik } 1952b9dbef2SJosef Bacik 1962b9dbef2SJosef Bacik /* We can free old roots now. */ 197889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 198889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 199889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 2002b9dbef2SJosef Bacik struct btrfs_root, root_list); 2012b9dbef2SJosef Bacik list_del_init(&root->root_list); 202889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 203889bfa39SJosef Bacik btrfs_free_log(trans, root); 2042b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 205889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 2062b9dbef2SJosef Bacik } 207889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 20827d56e62SJosef Bacik 2099e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 210817d52f8SJosef Bacik } 211817d52f8SJosef Bacik 2120860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2130860adfdSMiao Xie unsigned int type) 2140860adfdSMiao Xie { 2150860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2160860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2170860adfdSMiao Xie } 2180860adfdSMiao Xie 2190860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2200860adfdSMiao Xie unsigned int type) 2210860adfdSMiao Xie { 2220860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2230860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2240860adfdSMiao Xie } 2250860adfdSMiao Xie 2260860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2270860adfdSMiao Xie unsigned int type) 2280860adfdSMiao Xie { 2290860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2300860adfdSMiao Xie } 2310860adfdSMiao Xie 2320860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2330860adfdSMiao Xie { 2340860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 235178260b2SMiao Xie } 236178260b2SMiao Xie 237d352ac68SChris Mason /* 23879bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 23979bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 24079bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 24179bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 24279bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 243fb6dea26SJosef Bacik */ 244fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 245fb6dea26SJosef Bacik { 246fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 247fb6dea26SJosef Bacik 248fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 249fb6dea26SJosef Bacik return; 250fb6dea26SJosef Bacik 251fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 25263f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 253fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 254fb6dea26SJosef Bacik } 255fb6dea26SJosef Bacik 256fb6dea26SJosef Bacik /* 257d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 258d352ac68SChris Mason */ 2592ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2602ff7e61eSJeff Mahoney unsigned int type) 26179154b1bSChris Mason { 26279154b1bSChris Mason struct btrfs_transaction *cur_trans; 263a4abeea4SJosef Bacik 26419ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 265d43317dcSChris Mason loop: 26649b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 26784961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 26819ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 26949b25e05SJeff Mahoney return -EROFS; 27049b25e05SJeff Mahoney } 27149b25e05SJeff Mahoney 27219ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 273a4abeea4SJosef Bacik if (cur_trans) { 274bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 27519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 27649b25e05SJeff Mahoney return cur_trans->aborted; 277871383beSDavid Sterba } 2784a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 279178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 280178260b2SMiao Xie return -EBUSY; 281178260b2SMiao Xie } 2829b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 283a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2840860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 28519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 286e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 2875a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 288a4abeea4SJosef Bacik return 0; 289a4abeea4SJosef Bacik } 29019ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 291a4abeea4SJosef Bacik 292354aa0fbSMiao Xie /* 293354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 294354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 295354aa0fbSMiao Xie */ 296354aa0fbSMiao Xie if (type == TRANS_ATTACH) 297354aa0fbSMiao Xie return -ENOENT; 298354aa0fbSMiao Xie 2994a9d8bdeSMiao Xie /* 3004a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3014a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3024a9d8bdeSMiao Xie */ 3034a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3044a9d8bdeSMiao Xie 3054b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 306db5b493aSTsutomu Itoh if (!cur_trans) 307db5b493aSTsutomu Itoh return -ENOMEM; 308d43317dcSChris Mason 309e1489b4fSIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 3105a9ba670SIoannis Angelakopoulos btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 311e1489b4fSIoannis Angelakopoulos 31219ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 31319ae4e81SJan Schmidt if (fs_info->running_transaction) { 314d43317dcSChris Mason /* 315d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3164a9d8bdeSMiao Xie * to redo the checks above 317d43317dcSChris Mason */ 3185a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 319e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3204b5faeacSDavid Sterba kfree(cur_trans); 321d43317dcSChris Mason goto loop; 32284961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 323e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3245a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 325e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 3264b5faeacSDavid Sterba kfree(cur_trans); 3277b8b92afSJosef Bacik return -EROFS; 328a4abeea4SJosef Bacik } 329d43317dcSChris Mason 330ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 33148778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 33248778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 33313c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3340860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 33579154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 33679154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3374a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 338a4abeea4SJosef Bacik /* 339a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 340a4abeea4SJosef Bacik * commit the transaction. 341a4abeea4SJosef Bacik */ 3429b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3433204d33cSJosef Bacik cur_trans->flags = 0; 344afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 34556bec294SChris Mason 346a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 347a099d0fdSAlexandru Moise 3485c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3493368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 350d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 35120b297d6SJan Schmidt 35220b297d6SJan Schmidt /* 35320b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 35420b297d6SJan Schmidt * the log must never go across transaction boundaries. 35520b297d6SJan Schmidt */ 35620b297d6SJan Schmidt smp_mb(); 35731b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3585d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 35931b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3605d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 361fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 36220b297d6SJan Schmidt 36356bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 36456bec294SChris Mason 3653063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 366bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3679e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 368ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3691bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3702b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3711bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 372ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 373e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3742b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 375d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 376d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 37719ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 378c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 379efb0645bSJosef Bacik IO_TREE_TRANS_DIRTY_PAGES, NULL); 380fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 381fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 38219ae4e81SJan Schmidt fs_info->generation++; 38319ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 38419ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 38549b25e05SJeff Mahoney cur_trans->aborted = 0; 38619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 38715ee9bc7SJosef Bacik 38879154b1bSChris Mason return 0; 38979154b1bSChris Mason } 39079154b1bSChris Mason 391d352ac68SChris Mason /* 39292a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 39392a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 39492a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 39592a7cc42SQu Wenruo * transaction commits. 396d352ac68SChris Mason */ 3977585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 3986426c7adSQu Wenruo struct btrfs_root *root, 3996426c7adSQu Wenruo int force) 4006702ed49SChris Mason { 4010b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 40203a7e111SJosef Bacik int ret = 0; 4030b246afaSJeff Mahoney 40492a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4056426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4064d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4075d4f98a2SYan Zheng 4087585717fSChris Mason /* 40927cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4107585717fSChris Mason * we have the reloc mutex held now, so there 4117585717fSChris Mason * is only one writer in this function 4127585717fSChris Mason */ 41327cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4147585717fSChris Mason 41527cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4167585717fSChris Mason * they find our root->last_trans update 4177585717fSChris Mason */ 4187585717fSChris Mason smp_wmb(); 4197585717fSChris Mason 420fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4216426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 422fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 423a4abeea4SJosef Bacik return 0; 424a4abeea4SJosef Bacik } 425fc7cbcd4SDavid Sterba radix_tree_tag_set(&fs_info->fs_roots_radix, 4266702ed49SChris Mason (unsigned long)root->root_key.objectid, 4276702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 428fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4297585717fSChris Mason root->last_trans = trans->transid; 4307585717fSChris Mason 4317585717fSChris Mason /* this is pretty tricky. We don't want to 4327585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4337585717fSChris Mason * unless we're really doing the first setup for this root in 4347585717fSChris Mason * this transaction. 4357585717fSChris Mason * 4367585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4377585717fSChris Mason * if we want to take the expensive mutex. 4387585717fSChris Mason * 4397585717fSChris Mason * But, we have to set root->last_trans before we 4407585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4417585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 44227cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4437585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4447585717fSChris Mason * 4457585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4467585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4477585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4487585717fSChris Mason * done before we pop in the zero below 4497585717fSChris Mason */ 45003a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 451c7548af6SChris Mason smp_mb__before_atomic(); 45227cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4536702ed49SChris Mason } 45403a7e111SJosef Bacik return ret; 4556702ed49SChris Mason } 4565d4f98a2SYan Zheng 4577585717fSChris Mason 4582b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4592b9dbef2SJosef Bacik struct btrfs_root *root) 4602b9dbef2SJosef Bacik { 4610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4622b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4632b9dbef2SJosef Bacik 4642b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4652b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4662b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4672b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4682b9dbef2SJosef Bacik 4692b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 470fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 471fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 4722b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4732b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 474fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4752b9dbef2SJosef Bacik } 4762b9dbef2SJosef Bacik 4777585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4787585717fSChris Mason struct btrfs_root *root) 4797585717fSChris Mason { 4800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4811409e6ccSJosef Bacik int ret; 4820b246afaSJeff Mahoney 48392a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4847585717fSChris Mason return 0; 4857585717fSChris Mason 4867585717fSChris Mason /* 48727cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4887585717fSChris Mason * and barriers 4897585717fSChris Mason */ 4907585717fSChris Mason smp_rmb(); 4917585717fSChris Mason if (root->last_trans == trans->transid && 49227cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4937585717fSChris Mason return 0; 4947585717fSChris Mason 4950b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4961409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 4970b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 4987585717fSChris Mason 4991409e6ccSJosef Bacik return ret; 5007585717fSChris Mason } 5017585717fSChris Mason 5024a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5034a9d8bdeSMiao Xie { 5043296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 505501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 506bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5074a9d8bdeSMiao Xie } 5084a9d8bdeSMiao Xie 509d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 510d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 511d352ac68SChris Mason * transaction might not be fully on disk. 512d352ac68SChris Mason */ 5132ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 51479154b1bSChris Mason { 515f9295749SChris Mason struct btrfs_transaction *cur_trans; 51679154b1bSChris Mason 5170b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5180b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5194a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5209b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5210b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52272d63ed6SLi Zefan 5233e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 5240b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 525501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 526bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 527724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 528a4abeea4SJosef Bacik } else { 5290b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 530f9295749SChris Mason } 53137d1aeeeSChris Mason } 53237d1aeeeSChris Mason 5332ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 53437d1aeeeSChris Mason { 5350b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 536a4abeea4SJosef Bacik return 0; 537a4abeea4SJosef Bacik 53892e2f7e3SNikolay Borisov if (type == TRANS_START) 539a4abeea4SJosef Bacik return 1; 540a4abeea4SJosef Bacik 541a22285a6SYan, Zheng return 0; 542a22285a6SYan, Zheng } 543a22285a6SYan, Zheng 54420dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 54520dd2cbfSMiao Xie { 5460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5470b246afaSJeff Mahoney 5480b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 54992a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 55020dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 55120dd2cbfSMiao Xie root->reloc_root) 55220dd2cbfSMiao Xie return false; 55320dd2cbfSMiao Xie 55420dd2cbfSMiao Xie return true; 55520dd2cbfSMiao Xie } 55620dd2cbfSMiao Xie 55708e007d2SMiao Xie static struct btrfs_trans_handle * 5585aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 559003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 560003d7c59SJeff Mahoney bool enforce_qgroups) 561a22285a6SYan, Zheng { 5620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 563ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 564a22285a6SYan, Zheng struct btrfs_trans_handle *h; 565a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 566b5009945SJosef Bacik u64 num_bytes = 0; 567c5567237SArne Jansen u64 qgroup_reserved = 0; 56820dd2cbfSMiao Xie bool reloc_reserved = false; 5699c343784SJosef Bacik bool do_chunk_alloc = false; 57020dd2cbfSMiao Xie int ret; 571acce952bSliubo 57284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 573acce952bSliubo return ERR_PTR(-EROFS); 5742a1eb461SJosef Bacik 57546c4e71eSFilipe Manana if (current->journal_info) { 5760860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5772a1eb461SJosef Bacik h = current->journal_info; 578b50fff81SDavid Sterba refcount_inc(&h->use_count); 579b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5802a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5812a1eb461SJosef Bacik h->block_rsv = NULL; 5822a1eb461SJosef Bacik goto got_it; 5832a1eb461SJosef Bacik } 584b5009945SJosef Bacik 585b5009945SJosef Bacik /* 586b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 587b5009945SJosef Bacik * the appropriate flushing if need be. 588b5009945SJosef Bacik */ 589003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 590ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 591ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 592ba2c4d4eSJosef Bacik 5930b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 594733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 595003d7c59SJeff Mahoney enforce_qgroups); 596c5567237SArne Jansen if (ret) 597c5567237SArne Jansen return ERR_PTR(ret); 598c5567237SArne Jansen 599ba2c4d4eSJosef Bacik /* 600ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 601ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 602ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 603ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 604ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 605ba2c4d4eSJosef Bacik */ 6062bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6077f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 608748f553cSDavid Sterba btrfs_block_rsv_full(delayed_refs_rsv) == 0) { 609ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 610ba2c4d4eSJosef Bacik num_bytes <<= 1; 611ba2c4d4eSJosef Bacik } 612ba2c4d4eSJosef Bacik 61320dd2cbfSMiao Xie /* 61420dd2cbfSMiao Xie * Do the reservation for the relocation root creation 61520dd2cbfSMiao Xie */ 616ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6170b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 61820dd2cbfSMiao Xie reloc_reserved = true; 61920dd2cbfSMiao Xie } 62020dd2cbfSMiao Xie 6219270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 622ba2c4d4eSJosef Bacik if (ret) 623ba2c4d4eSJosef Bacik goto reserve_fail; 624ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 625ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 626ba2c4d4eSJosef Bacik delayed_refs_bytes); 627ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 628ba2c4d4eSJosef Bacik } 6299c343784SJosef Bacik 6309c343784SJosef Bacik if (rsv->space_info->force_alloc) 6319c343784SJosef Bacik do_chunk_alloc = true; 632ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 633748f553cSDavid Sterba !btrfs_block_rsv_full(delayed_refs_rsv)) { 634ba2c4d4eSJosef Bacik /* 635ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 636ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 637ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 638ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 639ba2c4d4eSJosef Bacik * here. 640ba2c4d4eSJosef Bacik */ 641ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 642b5009945SJosef Bacik if (ret) 643843fcf35SMiao Xie goto reserve_fail; 644b5009945SJosef Bacik } 645a22285a6SYan, Zheng again: 646f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 647843fcf35SMiao Xie if (!h) { 648843fcf35SMiao Xie ret = -ENOMEM; 649843fcf35SMiao Xie goto alloc_fail; 650843fcf35SMiao Xie } 651a22285a6SYan, Zheng 65298114659SJosef Bacik /* 65398114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 65498114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 65598114659SJosef Bacik * because we're already holding a ref. We need this because we could 65698114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 65798114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 658354aa0fbSMiao Xie * 659354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 660354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 66198114659SJosef Bacik */ 6620860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6630b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 664b2b5ef5cSJan Kara 6652ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6662ff7e61eSJeff Mahoney wait_current_trans(fs_info); 667a22285a6SYan, Zheng 668a4abeea4SJosef Bacik do { 6692ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 670178260b2SMiao Xie if (ret == -EBUSY) { 6712ff7e61eSJeff Mahoney wait_current_trans(fs_info); 672a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 673a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 674178260b2SMiao Xie ret = -ENOENT; 675178260b2SMiao Xie } 676a4abeea4SJosef Bacik } while (ret == -EBUSY); 677a4abeea4SJosef Bacik 678a43f7f82SLiu Bo if (ret < 0) 679843fcf35SMiao Xie goto join_fail; 6800f7d52f4SChris Mason 6810b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 682a22285a6SYan, Zheng 683a22285a6SYan, Zheng h->transid = cur_trans->transid; 684a22285a6SYan, Zheng h->transaction = cur_trans; 685b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 68664b63580SJeff Mahoney h->fs_info = root->fs_info; 6877174109cSQu Wenruo 688a698d075SMiao Xie h->type = type; 689ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 690b7ec40d7SChris Mason 691a22285a6SYan, Zheng smp_mb(); 6923296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 6932ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 694abdd2e80SFilipe Manana current->journal_info = h; 6953a45bb20SJeff Mahoney btrfs_commit_transaction(h); 696a22285a6SYan, Zheng goto again; 697a22285a6SYan, Zheng } 6989ed74f2dSJosef Bacik 699b5009945SJosef Bacik if (num_bytes) { 7000b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7012bcc0328SLiu Bo h->transid, num_bytes, 1); 7020b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 703b5009945SJosef Bacik h->bytes_reserved = num_bytes; 70420dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 705a22285a6SYan, Zheng } 706a22285a6SYan, Zheng 7072a1eb461SJosef Bacik got_it: 708bcf3a3e7SNikolay Borisov if (!current->journal_info) 709a22285a6SYan, Zheng current->journal_info = h; 710fcc99734SQu Wenruo 711fcc99734SQu Wenruo /* 7129c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7139c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7149c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7159c343784SJosef Bacik * value here. 7169c343784SJosef Bacik */ 7179c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7189c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7199c343784SJosef Bacik 7209c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7219c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7229c343784SJosef Bacik } 7239c343784SJosef Bacik 7249c343784SJosef Bacik /* 725fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 726fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 727fcc99734SQu Wenruo * transaction. 728fcc99734SQu Wenruo * 729fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 730fcc99734SQu Wenruo * or we can deadlock. 731fcc99734SQu Wenruo */ 73268075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 73368075ea8SJosef Bacik if (ret) { 73468075ea8SJosef Bacik /* 73568075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 73668075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 73768075ea8SJosef Bacik * not just freed. 73868075ea8SJosef Bacik */ 73968075ea8SJosef Bacik btrfs_end_transaction(h); 74068075ea8SJosef Bacik return ERR_PTR(ret); 74168075ea8SJosef Bacik } 742fcc99734SQu Wenruo 74379154b1bSChris Mason return h; 744843fcf35SMiao Xie 745843fcf35SMiao Xie join_fail: 7460860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7470b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 748843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 749843fcf35SMiao Xie alloc_fail: 750843fcf35SMiao Xie if (num_bytes) 7512ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 75263f018beSNikolay Borisov num_bytes, NULL); 753843fcf35SMiao Xie reserve_fail: 754733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 755843fcf35SMiao Xie return ERR_PTR(ret); 75679154b1bSChris Mason } 75779154b1bSChris Mason 758f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7595aed1dd8SAlexandru Moise unsigned int num_items) 760f9295749SChris Mason { 76108e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 762003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 763f9295749SChris Mason } 764003d7c59SJeff Mahoney 7658eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7668eab77ffSFilipe Manana struct btrfs_root *root, 7677f9fe614SJosef Bacik unsigned int num_items) 7688eab77ffSFilipe Manana { 7697f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7707f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7718eab77ffSFilipe Manana } 7728407aa46SMiao Xie 7737a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 774f9295749SChris Mason { 775003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 776003d7c59SJeff Mahoney true); 777f9295749SChris Mason } 778f9295749SChris Mason 7798d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7800af3d00bSJosef Bacik { 781575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 782003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7830af3d00bSJosef Bacik } 7840af3d00bSJosef Bacik 785d4edf39bSMiao Xie /* 786a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 787a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 788a6d155d2SFilipe Manana */ 789a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 790a6d155d2SFilipe Manana { 791a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 792a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 793a6d155d2SFilipe Manana } 794a6d155d2SFilipe Manana 795a6d155d2SFilipe Manana /* 796d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 797d4edf39bSMiao Xie * 798d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 799d4edf39bSMiao Xie * don't want to start a new one. 800d4edf39bSMiao Xie * 801d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 802d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 803d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 804d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 805d4edf39bSMiao Xie * invoke 806d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 807d4edf39bSMiao Xie */ 808354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 80960376ce4SJosef Bacik { 810575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 811003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 81260376ce4SJosef Bacik } 81360376ce4SJosef Bacik 814d4edf39bSMiao Xie /* 81590b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 816d4edf39bSMiao Xie * 81752042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 818d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 819d4edf39bSMiao Xie * complete. 820d4edf39bSMiao Xie */ 821d4edf39bSMiao Xie struct btrfs_trans_handle * 822d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 823d4edf39bSMiao Xie { 824d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 825d4edf39bSMiao Xie 826575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 827003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8288d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8292ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 830d4edf39bSMiao Xie 831d4edf39bSMiao Xie return trans; 832d4edf39bSMiao Xie } 833d4edf39bSMiao Xie 834d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 835d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 836d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 83789ce8a63SChris Mason { 8385fd76bf3SOmar Sandoval struct btrfs_fs_info *fs_info = commit->fs_info; 8395fd76bf3SOmar Sandoval u64 transid = commit->transid; 8405fd76bf3SOmar Sandoval bool put = false; 8415fd76bf3SOmar Sandoval 8423e738c53SIoannis Angelakopoulos /* 8433e738c53SIoannis Angelakopoulos * At the moment this function is called with min_state either being 8443e738c53SIoannis Angelakopoulos * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 8453e738c53SIoannis Angelakopoulos */ 8463e738c53SIoannis Angelakopoulos if (min_state == TRANS_STATE_COMPLETED) 8473e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 8483e738c53SIoannis Angelakopoulos else 8493e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 8503e738c53SIoannis Angelakopoulos 8515fd76bf3SOmar Sandoval while (1) { 852d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 8535fd76bf3SOmar Sandoval if (put) 8545fd76bf3SOmar Sandoval btrfs_put_transaction(commit); 8555fd76bf3SOmar Sandoval 8565fd76bf3SOmar Sandoval if (min_state < TRANS_STATE_COMPLETED) 8575fd76bf3SOmar Sandoval break; 8585fd76bf3SOmar Sandoval 8595fd76bf3SOmar Sandoval /* 8605fd76bf3SOmar Sandoval * A transaction isn't really completed until all of the 8615fd76bf3SOmar Sandoval * previous transactions are completed, but with fsync we can 8625fd76bf3SOmar Sandoval * end up with SUPER_COMMITTED transactions before a COMPLETED 8635fd76bf3SOmar Sandoval * transaction. Wait for those. 8645fd76bf3SOmar Sandoval */ 8655fd76bf3SOmar Sandoval 8665fd76bf3SOmar Sandoval spin_lock(&fs_info->trans_lock); 8675fd76bf3SOmar Sandoval commit = list_first_entry_or_null(&fs_info->trans_list, 8685fd76bf3SOmar Sandoval struct btrfs_transaction, 8695fd76bf3SOmar Sandoval list); 8705fd76bf3SOmar Sandoval if (!commit || commit->transid > transid) { 8715fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8725fd76bf3SOmar Sandoval break; 8735fd76bf3SOmar Sandoval } 8745fd76bf3SOmar Sandoval refcount_inc(&commit->use_count); 8755fd76bf3SOmar Sandoval put = true; 8765fd76bf3SOmar Sandoval spin_unlock(&fs_info->trans_lock); 8775fd76bf3SOmar Sandoval } 87889ce8a63SChris Mason } 87989ce8a63SChris Mason 8802ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 88146204592SSage Weil { 88246204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8838cd2807fSMiao Xie int ret = 0; 88446204592SSage Weil 88546204592SSage Weil if (transid) { 8860b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 887a4abeea4SJosef Bacik goto out; 88846204592SSage Weil 88946204592SSage Weil /* find specified transaction */ 8900b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8910b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 89246204592SSage Weil if (t->transid == transid) { 89346204592SSage Weil cur_trans = t; 8949b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8958cd2807fSMiao Xie ret = 0; 89646204592SSage Weil break; 89746204592SSage Weil } 8988cd2807fSMiao Xie if (t->transid > transid) { 8998cd2807fSMiao Xie ret = 0; 90046204592SSage Weil break; 90146204592SSage Weil } 9028cd2807fSMiao Xie } 9030b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 90442383020SSage Weil 90542383020SSage Weil /* 90642383020SSage Weil * The specified transaction doesn't exist, or we 90742383020SSage Weil * raced with btrfs_commit_transaction 90842383020SSage Weil */ 90942383020SSage Weil if (!cur_trans) { 9100b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 91142383020SSage Weil ret = -EINVAL; 9128cd2807fSMiao Xie goto out; 91342383020SSage Weil } 91446204592SSage Weil } else { 91546204592SSage Weil /* find newest transaction that is committing | committed */ 9160b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 9170b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 91846204592SSage Weil list) { 9194a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9204a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9213473f3c0SJosef Bacik break; 92246204592SSage Weil cur_trans = t; 9239b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 92446204592SSage Weil break; 92546204592SSage Weil } 92646204592SSage Weil } 9270b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 92846204592SSage Weil if (!cur_trans) 929a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 93046204592SSage Weil } 93146204592SSage Weil 932d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 933724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 934a4abeea4SJosef Bacik out: 93546204592SSage Weil return ret; 93646204592SSage Weil } 93746204592SSage Weil 9382ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 93937d1aeeeSChris Mason { 9402ff7e61eSJeff Mahoney wait_current_trans(fs_info); 94137d1aeeeSChris Mason } 94237d1aeeeSChris Mason 9438a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9448929ecfaSYan, Zheng { 9452ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9460b246afaSJeff Mahoney 94764403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9488a8f4deaSNikolay Borisov return true; 94936ba022aSJosef Bacik 9502ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9518929ecfaSYan, Zheng } 9528929ecfaSYan, Zheng 953a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9548929ecfaSYan, Zheng { 9558929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9568929ecfaSYan, Zheng 9573296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 958e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 959a2633b6aSNikolay Borisov return true; 9608929ecfaSYan, Zheng 9612ff7e61eSJeff Mahoney return should_end_transaction(trans); 9628929ecfaSYan, Zheng } 9638929ecfaSYan, Zheng 964dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 965dc60c525SNikolay Borisov 9660e34693fSNikolay Borisov { 967dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 968dc60c525SNikolay Borisov 9690e34693fSNikolay Borisov if (!trans->block_rsv) { 9700e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9710e34693fSNikolay Borisov return; 9720e34693fSNikolay Borisov } 9730e34693fSNikolay Borisov 9740e34693fSNikolay Borisov if (!trans->bytes_reserved) 9750e34693fSNikolay Borisov return; 9760e34693fSNikolay Borisov 9770e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9780e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9790e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9800e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 98163f018beSNikolay Borisov trans->bytes_reserved, NULL); 9820e34693fSNikolay Borisov trans->bytes_reserved = 0; 9830e34693fSNikolay Borisov } 9840e34693fSNikolay Borisov 98589ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9863a45bb20SJeff Mahoney int throttle) 98779154b1bSChris Mason { 9883a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9898929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9904edc2ca3SDave Jones int err = 0; 991d6e4a428SChris Mason 992b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 993b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9942a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9952a1eb461SJosef Bacik return 0; 9962a1eb461SJosef Bacik } 9972a1eb461SJosef Bacik 998dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9994c13d758SJosef Bacik trans->block_rsv = NULL; 1000c5567237SArne Jansen 10016c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 1002ea658badSJosef Bacik 10034fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 10044fbcdf66SFilipe Manana 10050860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 10060b246afaSJeff Mahoney sb_end_intwrite(info->sb); 10076df7881aSJosef Bacik 10088929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 100913c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 101013c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 10110860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 101289ce8a63SChris Mason 1013093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 1014e1489b4fSIoannis Angelakopoulos 10155a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1016e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(info, btrfs_trans_num_writers); 1017e1489b4fSIoannis Angelakopoulos 1018724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 10199ed74f2dSJosef Bacik 10209ed74f2dSJosef Bacik if (current->journal_info == trans) 10219ed74f2dSJosef Bacik current->journal_info = NULL; 1022ab78c84dSChris Mason 102324bbcf04SYan, Zheng if (throttle) 10242ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 102524bbcf04SYan, Zheng 102684961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10274e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1028fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1029fbabd4a3SJosef Bacik err = trans->aborted; 1030fbabd4a3SJosef Bacik else 1031fbabd4a3SJosef Bacik err = -EROFS; 10324e121c06SJosef Bacik } 103349b25e05SJeff Mahoney 10344edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10354edc2ca3SDave Jones return err; 103679154b1bSChris Mason } 103779154b1bSChris Mason 10383a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 103989ce8a63SChris Mason { 10403a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 104189ce8a63SChris Mason } 104289ce8a63SChris Mason 10433a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 104489ce8a63SChris Mason { 10453a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 104616cdcec7SMiao Xie } 104716cdcec7SMiao Xie 1048d352ac68SChris Mason /* 1049d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1050d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1051690587d1SChris Mason * those extents are sent to disk but does not wait on them 1052d352ac68SChris Mason */ 10532ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10548cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 105579154b1bSChris Mason { 1056777e6bd7SChris Mason int err = 0; 10577c4452b9SChris Mason int werr = 0; 10580b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1059e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1060777e6bd7SChris Mason u64 start = 0; 10615f39d397SChris Mason u64 end; 10627c4452b9SChris Mason 10636300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10641728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1065e6138876SJosef Bacik mark, &cached_state)) { 1066663dfbb0SFilipe Manana bool wait_writeback = false; 1067663dfbb0SFilipe Manana 1068663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1069663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1070210aa277SDavid Sterba mark, &cached_state); 1071663dfbb0SFilipe Manana /* 1072663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1073663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1074663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1075663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1076bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1077bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1078bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1079bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1080bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1081663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 108241e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1083663dfbb0SFilipe Manana */ 1084663dfbb0SFilipe Manana if (err == -ENOMEM) { 1085663dfbb0SFilipe Manana err = 0; 1086663dfbb0SFilipe Manana wait_writeback = true; 1087663dfbb0SFilipe Manana } 1088663dfbb0SFilipe Manana if (!err) 10891728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10907c4452b9SChris Mason if (err) 10917c4452b9SChris Mason werr = err; 1092663dfbb0SFilipe Manana else if (wait_writeback) 1093663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1094e38e2ed7SFilipe Manana free_extent_state(cached_state); 1095663dfbb0SFilipe Manana cached_state = NULL; 10961728366eSJosef Bacik cond_resched(); 10971728366eSJosef Bacik start = end + 1; 10987c4452b9SChris Mason } 10996300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1100690587d1SChris Mason return werr; 1101690587d1SChris Mason } 1102690587d1SChris Mason 1103690587d1SChris Mason /* 1104690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1105690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1106690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1107690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1108690587d1SChris Mason */ 1109bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1110bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1111690587d1SChris Mason { 1112690587d1SChris Mason int err = 0; 1113690587d1SChris Mason int werr = 0; 11140b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1115e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1116690587d1SChris Mason u64 start = 0; 1117690587d1SChris Mason u64 end; 1118690587d1SChris Mason 11191728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1120e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1121663dfbb0SFilipe Manana /* 1122663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1123663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1124663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1125663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1126663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 112741e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1128663dfbb0SFilipe Manana */ 1129663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1130bd015294SJosef Bacik EXTENT_NEED_WAIT, &cached_state); 1131663dfbb0SFilipe Manana if (err == -ENOMEM) 1132663dfbb0SFilipe Manana err = 0; 1133663dfbb0SFilipe Manana if (!err) 11341728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1135777e6bd7SChris Mason if (err) 1136777e6bd7SChris Mason werr = err; 1137e38e2ed7SFilipe Manana free_extent_state(cached_state); 1138e38e2ed7SFilipe Manana cached_state = NULL; 1139777e6bd7SChris Mason cond_resched(); 11401728366eSJosef Bacik start = end + 1; 1141777e6bd7SChris Mason } 11427c4452b9SChris Mason if (err) 11437c4452b9SChris Mason werr = err; 1144bf89d38fSJeff Mahoney return werr; 1145bf89d38fSJeff Mahoney } 1146656f30dbSFilipe Manana 1147b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1148bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1149bf89d38fSJeff Mahoney { 1150bf89d38fSJeff Mahoney bool errors = false; 1151bf89d38fSJeff Mahoney int err; 1152bf89d38fSJeff Mahoney 1153bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1154bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1155bf89d38fSJeff Mahoney errors = true; 1156bf89d38fSJeff Mahoney 1157bf89d38fSJeff Mahoney if (errors && !err) 1158bf89d38fSJeff Mahoney err = -EIO; 1159bf89d38fSJeff Mahoney return err; 1160bf89d38fSJeff Mahoney } 1161bf89d38fSJeff Mahoney 1162bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1163bf89d38fSJeff Mahoney { 1164bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1165bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1166bf89d38fSJeff Mahoney bool errors = false; 1167bf89d38fSJeff Mahoney int err; 1168bf89d38fSJeff Mahoney 1169bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1170bf89d38fSJeff Mahoney 1171bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1172656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11730b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1174656f30dbSFilipe Manana errors = true; 1175656f30dbSFilipe Manana 1176656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11770b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1178656f30dbSFilipe Manana errors = true; 1179656f30dbSFilipe Manana 1180bf89d38fSJeff Mahoney if (errors && !err) 1181bf89d38fSJeff Mahoney err = -EIO; 1182bf89d38fSJeff Mahoney return err; 118379154b1bSChris Mason } 118479154b1bSChris Mason 1185690587d1SChris Mason /* 1186c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1187c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1188c9b577c0SNikolay Borisov * log commit. 1189c9b577c0SNikolay Borisov * 1190c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1191690587d1SChris Mason */ 119270458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1193d0c803c4SChris Mason { 1194663dfbb0SFilipe Manana int ret; 1195c9b577c0SNikolay Borisov int ret2; 1196c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 119770458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1198c9b577c0SNikolay Borisov struct blk_plug plug; 1199663dfbb0SFilipe Manana 1200c9b577c0SNikolay Borisov blk_start_plug(&plug); 1201c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1202c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1203c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1204c9b577c0SNikolay Borisov 120541e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1206663dfbb0SFilipe Manana 1207c9b577c0SNikolay Borisov if (ret) 1208663dfbb0SFilipe Manana return ret; 1209c9b577c0SNikolay Borisov else if (ret2) 1210c9b577c0SNikolay Borisov return ret2; 1211c9b577c0SNikolay Borisov else 1212c9b577c0SNikolay Borisov return 0; 1213d0c803c4SChris Mason } 1214d0c803c4SChris Mason 1215d352ac68SChris Mason /* 1216d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1217d352ac68SChris Mason * 1218d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1219d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1220d352ac68SChris Mason * allocation tree. 1221d352ac68SChris Mason * 1222d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1223d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1224d352ac68SChris Mason */ 12250b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 122679154b1bSChris Mason struct btrfs_root *root) 122779154b1bSChris Mason { 122879154b1bSChris Mason int ret; 12290b86a832SChris Mason u64 old_root_bytenr; 123086b9f2ecSYan, Zheng u64 old_root_used; 12310b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12320b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 123379154b1bSChris Mason 123486b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 123556bec294SChris Mason 123679154b1bSChris Mason while (1) { 12370b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 123886b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1239ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 124079154b1bSChris Mason break; 124187ef2bb4SChris Mason 12425d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 124379154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12440b86a832SChris Mason &root->root_key, 12450b86a832SChris Mason &root->root_item); 124649b25e05SJeff Mahoney if (ret) 124749b25e05SJeff Mahoney return ret; 124856bec294SChris Mason 124986b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1250e7070be1SJosef Bacik } 1251276e680dSYan Zheng 12520b86a832SChris Mason return 0; 12530b86a832SChris Mason } 12540b86a832SChris Mason 1255d352ac68SChris Mason /* 1256d352ac68SChris Mason * update all the cowonly tree roots on disk 125749b25e05SJeff Mahoney * 125849b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 125949b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 126049b25e05SJeff Mahoney * to clean up the delayed refs. 1261d352ac68SChris Mason */ 12629386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12630b86a832SChris Mason { 12649386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1265ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12661bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12670b86a832SChris Mason struct list_head *next; 126884234f3aSYan Zheng struct extent_buffer *eb; 126956bec294SChris Mason int ret; 127084234f3aSYan Zheng 1271dfba78dcSFilipe Manana /* 1272dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1273dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1274dfba78dcSFilipe Manana */ 1275dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1276dfba78dcSFilipe Manana 127784234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 127849b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12799631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 128084234f3aSYan Zheng btrfs_tree_unlock(eb); 128184234f3aSYan Zheng free_extent_buffer(eb); 12820b86a832SChris Mason 128349b25e05SJeff Mahoney if (ret) 128449b25e05SJeff Mahoney return ret; 128549b25e05SJeff Mahoney 1286196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1287c16ce190SJosef Bacik if (ret) 1288c16ce190SJosef Bacik return ret; 12892b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1290c16ce190SJosef Bacik if (ret) 1291c16ce190SJosef Bacik return ret; 1292280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1293c16ce190SJosef Bacik if (ret) 1294c16ce190SJosef Bacik return ret; 1295546adb0dSJan Schmidt 1296bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1297dcdf7f6dSJosef Bacik if (ret) 1298dcdf7f6dSJosef Bacik return ret; 1299dcdf7f6dSJosef Bacik 1300ea526d18SJosef Bacik again: 13010b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 13022ff7e61eSJeff Mahoney struct btrfs_root *root; 13030b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 13040b86a832SChris Mason list_del_init(next); 13050b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1306e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 130787ef2bb4SChris Mason 13089e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13099e351cc8SJosef Bacik &trans->transaction->switch_commits); 131049b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 131149b25e05SJeff Mahoney if (ret) 131249b25e05SJeff Mahoney return ret; 1313488bc2a2SJosef Bacik } 1314488bc2a2SJosef Bacik 1315488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1316c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1317ea526d18SJosef Bacik if (ret) 1318ea526d18SJosef Bacik return ret; 1319276e680dSYan Zheng 13201bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 13215742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1322ea526d18SJosef Bacik if (ret) 1323ea526d18SJosef Bacik return ret; 1324488bc2a2SJosef Bacik 1325488bc2a2SJosef Bacik /* 1326488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1327488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1328488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1329488bc2a2SJosef Bacik * everything gets run. 1330488bc2a2SJosef Bacik */ 1331c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1332ea526d18SJosef Bacik if (ret) 1333ea526d18SJosef Bacik return ret; 1334ea526d18SJosef Bacik } 1335ea526d18SJosef Bacik 1336ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1337ea526d18SJosef Bacik goto again; 1338ea526d18SJosef Bacik 13399f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13409f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13419f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13428dabb742SStefan Behrens 134379154b1bSChris Mason return 0; 134479154b1bSChris Mason } 134579154b1bSChris Mason 1346d352ac68SChris Mason /* 1347b4be6aefSJosef Bacik * If we had a pending drop we need to see if there are any others left in our 1348b4be6aefSJosef Bacik * dead roots list, and if not clear our bit and wake any waiters. 1349b4be6aefSJosef Bacik */ 1350b4be6aefSJosef Bacik void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1351b4be6aefSJosef Bacik { 1352b4be6aefSJosef Bacik /* 1353b4be6aefSJosef Bacik * We put the drop in progress roots at the front of the list, so if the 1354b4be6aefSJosef Bacik * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1355b4be6aefSJosef Bacik * up. 1356b4be6aefSJosef Bacik */ 1357b4be6aefSJosef Bacik spin_lock(&fs_info->trans_lock); 1358b4be6aefSJosef Bacik if (!list_empty(&fs_info->dead_roots)) { 1359b4be6aefSJosef Bacik struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1360b4be6aefSJosef Bacik struct btrfs_root, 1361b4be6aefSJosef Bacik root_list); 1362b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1363b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1364b4be6aefSJosef Bacik return; 1365b4be6aefSJosef Bacik } 1366b4be6aefSJosef Bacik } 1367b4be6aefSJosef Bacik spin_unlock(&fs_info->trans_lock); 1368b4be6aefSJosef Bacik 1369b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 1370b4be6aefSJosef Bacik } 1371b4be6aefSJosef Bacik 1372b4be6aefSJosef Bacik /* 1373d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1374d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1375d352ac68SChris Mason * be deleted 1376d352ac68SChris Mason */ 1377cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13785eda7b5eSChris Mason { 13790b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13800b246afaSJeff Mahoney 13810b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1382dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1383dc9492c1SJosef Bacik btrfs_grab_root(root); 1384b4be6aefSJosef Bacik 1385b4be6aefSJosef Bacik /* We want to process the partially complete drops first. */ 1386b4be6aefSJosef Bacik if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1387b4be6aefSJosef Bacik list_add(&root->root_list, &fs_info->dead_roots); 1388b4be6aefSJosef Bacik else 13890b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1390dc9492c1SJosef Bacik } 13910b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13925eda7b5eSChris Mason } 13935eda7b5eSChris Mason 1394d352ac68SChris Mason /* 1395dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1396dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1397d352ac68SChris Mason */ 13987e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13990f7d52f4SChris Mason { 14007e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1401fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 1402fc7cbcd4SDavid Sterba int i; 1403fc7cbcd4SDavid Sterba int ret; 140454aa1f4dSChris Mason 1405dfba78dcSFilipe Manana /* 1406dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1407dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1408dfba78dcSFilipe Manana */ 1409dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1410dfba78dcSFilipe Manana 1411fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1412fc7cbcd4SDavid Sterba while (1) { 1413fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1414fc7cbcd4SDavid Sterba (void **)gang, 0, 1415fc7cbcd4SDavid Sterba ARRAY_SIZE(gang), 1416fc7cbcd4SDavid Sterba BTRFS_ROOT_TRANS_TAG); 1417fc7cbcd4SDavid Sterba if (ret == 0) 1418fc7cbcd4SDavid Sterba break; 1419fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 1420fc7cbcd4SDavid Sterba struct btrfs_root *root = gang[i]; 1421fc7cbcd4SDavid Sterba int ret2; 14224f4317c1SJosef Bacik 1423dfba78dcSFilipe Manana /* 1424dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1425dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1426dfba78dcSFilipe Manana */ 1427dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1428dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1429dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1430dfba78dcSFilipe Manana 1431fc7cbcd4SDavid Sterba radix_tree_tag_clear(&fs_info->fs_roots_radix, 14322619ba1fSChris Mason (unsigned long)root->root_key.objectid, 14330f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1434fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 143531153d81SYan Zheng 1436e02119d5SChris Mason btrfs_free_log(trans, root); 1437fc7cbcd4SDavid Sterba ret2 = btrfs_update_reloc_root(trans, root); 1438fc7cbcd4SDavid Sterba if (ret2) 1439fc7cbcd4SDavid Sterba return ret2; 1440e02119d5SChris Mason 1441fc7cbcd4SDavid Sterba /* see comments in should_cow_block() */ 144227cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1443c7548af6SChris Mason smp_mb__after_atomic(); 1444f1ebcc74SLiu Bo 1445978d910dSYan Zheng if (root->commit_root != root->node) { 14469e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 14479e351cc8SJosef Bacik &trans->transaction->switch_commits); 1448fc7cbcd4SDavid Sterba btrfs_set_root_node(&root->root_item, 1449fc7cbcd4SDavid Sterba root->node); 1450978d910dSYan Zheng } 145131153d81SYan Zheng 1452fc7cbcd4SDavid Sterba ret2 = btrfs_update_root(trans, fs_info->tree_root, 1453fc7cbcd4SDavid Sterba &root->root_key, 1454fc7cbcd4SDavid Sterba &root->root_item); 1455fc7cbcd4SDavid Sterba if (ret2) 1456fc7cbcd4SDavid Sterba return ret2; 1457fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1458733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14590f7d52f4SChris Mason } 1460fc7cbcd4SDavid Sterba } 1461fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 14624f4317c1SJosef Bacik return 0; 14630f7d52f4SChris Mason } 14640f7d52f4SChris Mason 1465d352ac68SChris Mason /* 1466de78b51aSEric Sandeen * defrag a given btree. 1467de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1468d352ac68SChris Mason */ 1469de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1470e9d0b13bSChris Mason { 1471e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1472e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14738929ecfaSYan, Zheng int ret; 1474e9d0b13bSChris Mason 147527cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1476e9d0b13bSChris Mason return 0; 14778929ecfaSYan, Zheng 14786b80053dSChris Mason while (1) { 14798929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14806819703fSDavid Sterba if (IS_ERR(trans)) { 14816819703fSDavid Sterba ret = PTR_ERR(trans); 14826819703fSDavid Sterba break; 14836819703fSDavid Sterba } 14848929ecfaSYan, Zheng 1485de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14868929ecfaSYan, Zheng 14873a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14882ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1489e9d0b13bSChris Mason cond_resched(); 1490e9d0b13bSChris Mason 1491ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1492e9d0b13bSChris Mason break; 1493210549ebSDavid Sterba 1494ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1495ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1496210549ebSDavid Sterba ret = -EAGAIN; 1497210549ebSDavid Sterba break; 1498210549ebSDavid Sterba } 1499e9d0b13bSChris Mason } 150027cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 15018929ecfaSYan, Zheng return ret; 1502e9d0b13bSChris Mason } 1503e9d0b13bSChris Mason 1504d352ac68SChris Mason /* 15056426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 15066426c7adSQu Wenruo * 15076426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 15086426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 15096426c7adSQu Wenruo * qgroup works. 15106426c7adSQu Wenruo */ 15116426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 15126426c7adSQu Wenruo struct btrfs_root *src, 15136426c7adSQu Wenruo struct btrfs_root *parent, 15146426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 15156426c7adSQu Wenruo u64 dst_objectid) 15166426c7adSQu Wenruo { 15176426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 15186426c7adSQu Wenruo int ret; 15196426c7adSQu Wenruo 15206426c7adSQu Wenruo /* 15216426c7adSQu Wenruo * Save some performance in the case that qgroups are not 15226426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 15236426c7adSQu Wenruo * kick in anyway. 15246426c7adSQu Wenruo */ 15259ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 15266426c7adSQu Wenruo return 0; 15276426c7adSQu Wenruo 15286426c7adSQu Wenruo /* 152952042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 15304d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 15314d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 15324d31778aSQu Wenruo * item. 15334d31778aSQu Wenruo */ 15341c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 15351c442d22SJosef Bacik if (ret) 15361c442d22SJosef Bacik return ret; 15374d31778aSQu Wenruo 15384d31778aSQu Wenruo /* 15392a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 15402a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 15412a4d84c1SJosef Bacik * 15422a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 15432a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 15442a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 15452a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 15462a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 15472a4d84c1SJosef Bacik * qgroup counters could end up wrong. 15482a4d84c1SJosef Bacik */ 15492a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 15502a4d84c1SJosef Bacik if (ret) { 15512a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 155244365827SNaohiro Aota return ret; 15532a4d84c1SJosef Bacik } 15542a4d84c1SJosef Bacik 15557e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15566426c7adSQu Wenruo if (ret) 15576426c7adSQu Wenruo goto out; 1558460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15596426c7adSQu Wenruo if (ret < 0) 15606426c7adSQu Wenruo goto out; 15616426c7adSQu Wenruo 15626426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1563a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15646426c7adSQu Wenruo inherit); 15656426c7adSQu Wenruo if (ret < 0) 15666426c7adSQu Wenruo goto out; 15676426c7adSQu Wenruo 15686426c7adSQu Wenruo /* 15696426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15706426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15716426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15726426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15736426c7adSQu Wenruo * 2) write all btree blocks onto disk 15746426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15756426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15766426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15776426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15786426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15796426c7adSQu Wenruo */ 15809386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15816426c7adSQu Wenruo if (ret) 15826426c7adSQu Wenruo goto out; 1583889bfa39SJosef Bacik switch_commit_roots(trans); 158470458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15856426c7adSQu Wenruo if (ret) 1586f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15876426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15886426c7adSQu Wenruo 15896426c7adSQu Wenruo out: 15906426c7adSQu Wenruo /* 15916426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15926426c7adSQu Wenruo * last_trans == cur_transid. 15936426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15946426c7adSQu Wenruo * insert_dir_item() 15956426c7adSQu Wenruo */ 15966426c7adSQu Wenruo if (!ret) 15971c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15986426c7adSQu Wenruo return ret; 15996426c7adSQu Wenruo } 16006426c7adSQu Wenruo 16016426c7adSQu Wenruo /* 1602d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1603aec8030aSMiao Xie * transaction commit. This does the actual creation. 1604aec8030aSMiao Xie * 1605aec8030aSMiao Xie * Note: 1606aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1607aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1608aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1609d352ac68SChris Mason */ 161080b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 16113063d29fSChris Mason struct btrfs_pending_snapshot *pending) 16123063d29fSChris Mason { 161308d50ca3SNikolay Borisov 161408d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 16153063d29fSChris Mason struct btrfs_key key; 161680b6794dSChris Mason struct btrfs_root_item *new_root_item; 16173063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 16183063d29fSChris Mason struct btrfs_root *root = pending->root; 16196bdb72deSSage Weil struct btrfs_root *parent_root; 162098c9942aSLiu Bo struct btrfs_block_rsv *rsv; 1621ab3c5c18SSweet Tea Dorminy struct inode *parent_inode = pending->dir; 162242874b3dSMiao Xie struct btrfs_path *path; 162342874b3dSMiao Xie struct btrfs_dir_item *dir_item; 16243063d29fSChris Mason struct extent_buffer *tmp; 1625925baeddSChris Mason struct extent_buffer *old; 162695582b00SDeepa Dinamani struct timespec64 cur_time; 1627aec8030aSMiao Xie int ret = 0; 1628d68fc57bSYan, Zheng u64 to_reserve = 0; 16296bdb72deSSage Weil u64 index = 0; 1630a22285a6SYan, Zheng u64 objectid; 1631b83cc969SLi Zefan u64 root_flags; 1632ab3c5c18SSweet Tea Dorminy unsigned int nofs_flags; 1633ab3c5c18SSweet Tea Dorminy struct fscrypt_name fname; 16343063d29fSChris Mason 16358546b570SDavid Sterba ASSERT(pending->path); 16368546b570SDavid Sterba path = pending->path; 163742874b3dSMiao Xie 1638b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1639b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1640a22285a6SYan, Zheng 1641ab3c5c18SSweet Tea Dorminy /* 1642ab3c5c18SSweet Tea Dorminy * We're inside a transaction and must make sure that any potential 1643ab3c5c18SSweet Tea Dorminy * allocations with GFP_KERNEL in fscrypt won't recurse back to 1644ab3c5c18SSweet Tea Dorminy * filesystem. 1645ab3c5c18SSweet Tea Dorminy */ 1646ab3c5c18SSweet Tea Dorminy nofs_flags = memalloc_nofs_save(); 1647ab3c5c18SSweet Tea Dorminy pending->error = fscrypt_setup_filename(parent_inode, 1648ab3c5c18SSweet Tea Dorminy &pending->dentry->d_name, 0, 1649ab3c5c18SSweet Tea Dorminy &fname); 1650ab3c5c18SSweet Tea Dorminy memalloc_nofs_restore(nofs_flags); 1651ab3c5c18SSweet Tea Dorminy if (pending->error) 1652ab3c5c18SSweet Tea Dorminy goto free_pending; 1653ab3c5c18SSweet Tea Dorminy 1654543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1655aec8030aSMiao Xie if (pending->error) 1656ab3c5c18SSweet Tea Dorminy goto free_fname; 16573063d29fSChris Mason 1658d6726335SQu Wenruo /* 1659d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1660d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1661d6726335SQu Wenruo */ 1662d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1663d6726335SQu Wenruo 1664147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1665d68fc57bSYan, Zheng 1666d68fc57bSYan, Zheng if (to_reserve > 0) { 16679270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1668aec8030aSMiao Xie &pending->block_rsv, 166908e007d2SMiao Xie to_reserve, 167008e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1671aec8030aSMiao Xie if (pending->error) 1672d6726335SQu Wenruo goto clear_skip_qgroup; 1673d68fc57bSYan, Zheng } 1674d68fc57bSYan, Zheng 16753063d29fSChris Mason key.objectid = objectid; 1676a22285a6SYan, Zheng key.offset = (u64)-1; 1677a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16783063d29fSChris Mason 16796fa9700eSMiao Xie rsv = trans->block_rsv; 1680a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16812382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16820b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 168388d3a5aaSJosef Bacik trans->transid, 168488d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1685a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1686f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1687f0118cb6SJosef Bacik if (ret) 1688f0118cb6SJosef Bacik goto fail; 1689c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 169004b285f3SDeepa Dinamani 16916bdb72deSSage Weil /* 16926bdb72deSSage Weil * insert the directory item 16936bdb72deSSage Weil */ 1694877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 169549b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 169642874b3dSMiao Xie 169742874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 169842874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16994a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 17006db75318SSweet Tea Dorminy &fname.disk_name, 0); 170142874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1702fe66a05aSChris Mason pending->error = -EEXIST; 1703aec8030aSMiao Xie goto dir_item_existed; 170442874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 170542874b3dSMiao Xie ret = PTR_ERR(dir_item); 170666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17078732d44fSMiao Xie goto fail; 170879787eaaSJeff Mahoney } 170942874b3dSMiao Xie btrfs_release_path(path); 17106bdb72deSSage Weil 1711e999376fSChris Mason /* 1712e999376fSChris Mason * pull in the delayed directory update 1713e999376fSChris Mason * and the delayed inode item 1714e999376fSChris Mason * otherwise we corrupt the FS during 1715e999376fSChris Mason * snapshot 1716e999376fSChris Mason */ 1717e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 17188732d44fSMiao Xie if (ret) { /* Transaction aborted */ 171966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17208732d44fSMiao Xie goto fail; 17218732d44fSMiao Xie } 1722e999376fSChris Mason 1723f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1724f0118cb6SJosef Bacik if (ret) { 1725f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1726f0118cb6SJosef Bacik goto fail; 1727f0118cb6SJosef Bacik } 17286bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 17296bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 173008fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 17316bdb72deSSage Weil 1732b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1733b83cc969SLi Zefan if (pending->readonly) 1734b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1735b83cc969SLi Zefan else 1736b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1737b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1738b83cc969SLi Zefan 17398ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 17408ea05e3aSAlexander Block trans->transid); 1741807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 17428ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 17438ea05e3aSAlexander Block BTRFS_UUID_SIZE); 174470023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 174570023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 174670023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 17478ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 17488ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 17498ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 17508ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 175170023da2SStefan Behrens } 17523cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 17533cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 175470023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 17558ea05e3aSAlexander Block 1756925baeddSChris Mason old = btrfs_lock_root_node(root); 17579631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 17589631e4ccSJosef Bacik BTRFS_NESTING_COW); 175979787eaaSJeff Mahoney if (ret) { 176079787eaaSJeff Mahoney btrfs_tree_unlock(old); 176179787eaaSJeff Mahoney free_extent_buffer(old); 176266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17638732d44fSMiao Xie goto fail; 176479787eaaSJeff Mahoney } 176549b25e05SJeff Mahoney 176649b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 176779787eaaSJeff Mahoney /* clean up in any case */ 1768925baeddSChris Mason btrfs_tree_unlock(old); 1769925baeddSChris Mason free_extent_buffer(old); 17708732d44fSMiao Xie if (ret) { 177166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17728732d44fSMiao Xie goto fail; 17738732d44fSMiao Xie } 1774f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 177527cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1776f1ebcc74SLiu Bo smp_wmb(); 1777f1ebcc74SLiu Bo 17785d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1779a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1780a22285a6SYan, Zheng key.offset = trans->transid; 1781a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1782925baeddSChris Mason btrfs_tree_unlock(tmp); 17833063d29fSChris Mason free_extent_buffer(tmp); 17848732d44fSMiao Xie if (ret) { 178566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17868732d44fSMiao Xie goto fail; 17878732d44fSMiao Xie } 17880660b5afSChris Mason 1789a22285a6SYan, Zheng /* 1790a22285a6SYan, Zheng * insert root back/forward references 1791a22285a6SYan, Zheng */ 17926025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1793a22285a6SYan, Zheng parent_root->root_key.objectid, 17944a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 17956db75318SSweet Tea Dorminy &fname.disk_name); 17968732d44fSMiao Xie if (ret) { 179766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17988732d44fSMiao Xie goto fail; 17998732d44fSMiao Xie } 1800a22285a6SYan, Zheng 1801a22285a6SYan, Zheng key.offset = (u64)-1; 18022dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 180379787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 180479787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 18052d892ccdSFilipe Manana pending->snap = NULL; 180666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18078732d44fSMiao Xie goto fail; 180879787eaaSJeff Mahoney } 1809d68fc57bSYan, Zheng 181049b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 18118732d44fSMiao Xie if (ret) { 181266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18138732d44fSMiao Xie goto fail; 18148732d44fSMiao Xie } 1815361048f5SMiao Xie 18166426c7adSQu Wenruo /* 18176426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 18186426c7adSQu Wenruo * snapshot hack to do fast snapshot. 18196426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 18206426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 18216426c7adSQu Wenruo */ 18226426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 18236426c7adSQu Wenruo pending->inherit, objectid); 18246426c7adSQu Wenruo if (ret < 0) 18256426c7adSQu Wenruo goto fail; 18266426c7adSQu Wenruo 18276db75318SSweet Tea Dorminy ret = btrfs_insert_dir_item(trans, &fname.disk_name, 18286db75318SSweet Tea Dorminy BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, 18296db75318SSweet Tea Dorminy index); 183042874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 18319c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 18328732d44fSMiao Xie if (ret) { 183366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 18348732d44fSMiao Xie goto fail; 18358732d44fSMiao Xie } 183642874b3dSMiao Xie 18376ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 18386db75318SSweet Tea Dorminy fname.disk_name.len * 2); 1839c1867eb3SDavid Sterba parent_inode->i_mtime = current_time(parent_inode); 1840c1867eb3SDavid Sterba parent_inode->i_ctime = parent_inode->i_mtime; 1841729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1842dd5f9615SStefan Behrens if (ret) { 184366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1844dd5f9615SStefan Behrens goto fail; 1845dd5f9615SStefan Behrens } 1846807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1847807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1848cdb345a8SLu Fengqi objectid); 1849dd5f9615SStefan Behrens if (ret) { 185066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1851dd5f9615SStefan Behrens goto fail; 1852dd5f9615SStefan Behrens } 1853dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1854cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1855dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1856dd5f9615SStefan Behrens objectid); 1857dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 185866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1859dd5f9615SStefan Behrens goto fail; 1860dd5f9615SStefan Behrens } 1861dd5f9615SStefan Behrens } 1862d6726335SQu Wenruo 18633063d29fSChris Mason fail: 1864aec8030aSMiao Xie pending->error = ret; 1865aec8030aSMiao Xie dir_item_existed: 186698c9942aSLiu Bo trans->block_rsv = rsv; 18672382c5ccSLiu Bo trans->bytes_reserved = 0; 1868d6726335SQu Wenruo clear_skip_qgroup: 1869d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 1870ab3c5c18SSweet Tea Dorminy free_fname: 1871ab3c5c18SSweet Tea Dorminy fscrypt_free_filename(&fname); 1872ab3c5c18SSweet Tea Dorminy free_pending: 18736fa9700eSMiao Xie kfree(new_root_item); 1874b0c0ea63SDavid Sterba pending->root_item = NULL; 187542874b3dSMiao Xie btrfs_free_path(path); 18768546b570SDavid Sterba pending->path = NULL; 18778546b570SDavid Sterba 187849b25e05SJeff Mahoney return ret; 18793063d29fSChris Mason } 18803063d29fSChris Mason 1881d352ac68SChris Mason /* 1882d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1883d352ac68SChris Mason */ 188408d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18853063d29fSChris Mason { 1886aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18873063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1888aec8030aSMiao Xie int ret = 0; 18893de4586cSChris Mason 1890aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1891aec8030aSMiao Xie list_del(&pending->list); 189208d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1893aec8030aSMiao Xie if (ret) 1894aec8030aSMiao Xie break; 1895aec8030aSMiao Xie } 1896aec8030aSMiao Xie return ret; 18973de4586cSChris Mason } 18983de4586cSChris Mason 18992ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 19005d4f98a2SYan Zheng { 19015d4f98a2SYan Zheng struct btrfs_root_item *root_item; 19025d4f98a2SYan Zheng struct btrfs_super_block *super; 19035d4f98a2SYan Zheng 19040b246afaSJeff Mahoney super = fs_info->super_copy; 19055d4f98a2SYan Zheng 19060b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1907093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1908093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1909093e037cSDavid Sterba super->chunk_root_level = root_item->level; 19105d4f98a2SYan Zheng 19110b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1912093e037cSDavid Sterba super->root = root_item->bytenr; 1913093e037cSDavid Sterba super->generation = root_item->generation; 1914093e037cSDavid Sterba super->root_level = root_item->level; 19150b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1916093e037cSDavid Sterba super->cache_generation = root_item->generation; 191794846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 191894846229SBoris Burkov super->cache_generation = 0; 19190b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1920093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 19215d4f98a2SYan Zheng } 19225d4f98a2SYan Zheng 1923f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1924f36f3042SChris Mason { 19254a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1926f36f3042SChris Mason int ret = 0; 19274a9d8bdeSMiao Xie 1928a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19294a9d8bdeSMiao Xie trans = info->running_transaction; 19304a9d8bdeSMiao Xie if (trans) 19314a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1932a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1933f36f3042SChris Mason return ret; 1934f36f3042SChris Mason } 1935f36f3042SChris Mason 19368929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 19378929ecfaSYan, Zheng { 19384a9d8bdeSMiao Xie struct btrfs_transaction *trans; 19398929ecfaSYan, Zheng int ret = 0; 19404a9d8bdeSMiao Xie 1941a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 19424a9d8bdeSMiao Xie trans = info->running_transaction; 19434a9d8bdeSMiao Xie if (trans) 19444a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1945a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 19468929ecfaSYan, Zheng return ret; 19478929ecfaSYan, Zheng } 19488929ecfaSYan, Zheng 1949fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1950bb9c12c9SSage Weil { 19513a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1952bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1953bb9c12c9SSage Weil 1954fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1955fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1956fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1957bb9c12c9SSage Weil 1958bb9c12c9SSage Weil /* take transaction reference */ 1959bb9c12c9SSage Weil cur_trans = trans->transaction; 19609b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1961bb9c12c9SSage Weil 19623a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19636fc4e354SSage Weil 19646fc4e354SSage Weil /* 1965ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1966ae5d29d4SDavid Sterba * subsequent transaction joins 1967ae5d29d4SDavid Sterba */ 19683e738c53SIoannis Angelakopoulos btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1969ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1970ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1971ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1972724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1973bb9c12c9SSage Weil } 1974bb9c12c9SSage Weil 197597cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 197649b25e05SJeff Mahoney { 197797cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 197849b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 197949b25e05SJeff Mahoney 1980b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 198149b25e05SJeff Mahoney 198266642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19837b8b92afSJosef Bacik 19840b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 198566b6135bSLiu Bo 198625d8c284SMiao Xie /* 198725d8c284SMiao Xie * If the transaction is removed from the list, it means this 198825d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 198925d8c284SMiao Xie * to call the cleanup function. 199025d8c284SMiao Xie */ 199125d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 199266b6135bSLiu Bo 19930b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19944a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19950b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1996e1489b4fSIoannis Angelakopoulos 1997e1489b4fSIoannis Angelakopoulos /* 1998e1489b4fSIoannis Angelakopoulos * The thread has already released the lockdep map as reader 1999e1489b4fSIoannis Angelakopoulos * already in btrfs_commit_transaction(). 2000e1489b4fSIoannis Angelakopoulos */ 2001e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2002f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 2003f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 2004f094ac32SLiu Bo 20050b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 2006d7096fc3SJosef Bacik } 2007061dde82SFilipe Manana 2008061dde82SFilipe Manana /* 2009061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 2010061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 2011061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 2012061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 2013061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 2014061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 2015061dde82SFilipe Manana */ 2016061dde82SFilipe Manana list_del_init(&cur_trans->list); 2017061dde82SFilipe Manana 20180b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 201949b25e05SJeff Mahoney 20202ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 202149b25e05SJeff Mahoney 20220b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20230b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 20240b246afaSJeff Mahoney fs_info->running_transaction = NULL; 20250b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20264a9d8bdeSMiao Xie 2027e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 20280b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2029724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2030724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 203149b25e05SJeff Mahoney 20322e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 203349b25e05SJeff Mahoney 203449b25e05SJeff Mahoney if (current->journal_info == trans) 203549b25e05SJeff Mahoney current->journal_info = NULL; 20360b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 203749b25e05SJeff Mahoney 203849b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 203949b25e05SJeff Mahoney } 204049b25e05SJeff Mahoney 2041c7cc64a9SDavid Sterba /* 2042c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2043c7cc64a9SDavid Sterba * transaction and remove them from the list 2044c7cc64a9SDavid Sterba */ 2045c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2046c7cc64a9SDavid Sterba { 2047c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 204832da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2049c7cc64a9SDavid Sterba 2050c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2051c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2052c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2053c7cc64a9SDavid Sterba } 2054c7cc64a9SDavid Sterba } 2055c7cc64a9SDavid Sterba 205688090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 205782436617SMiao Xie { 2058ce8ea7ccSJosef Bacik /* 2059a0f0cf83SFilipe Manana * We use try_to_writeback_inodes_sb() here because if we used 2060ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2061ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2062ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2063ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2064ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2065ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2066a0f0cf83SFilipe Manana * 2067a0f0cf83SFilipe Manana * Note that try_to_writeback_inodes_sb() will only trigger writeback 2068a0f0cf83SFilipe Manana * if it can read lock sb->s_umount. It will always be able to lock it, 2069a0f0cf83SFilipe Manana * except when the filesystem is being unmounted or being frozen, but in 2070a0f0cf83SFilipe Manana * those cases sync_filesystem() is called, which results in calling 2071a0f0cf83SFilipe Manana * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2072a0f0cf83SFilipe Manana * Note that we don't call writeback_inodes_sb() directly, because it 2073a0f0cf83SFilipe Manana * will emit a warning if sb->s_umount is not locked. 2074ce8ea7ccSJosef Bacik */ 207588090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2076a0f0cf83SFilipe Manana try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 207782436617SMiao Xie return 0; 207882436617SMiao Xie } 207982436617SMiao Xie 208088090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 208182436617SMiao Xie { 208288090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20836374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 208482436617SMiao Xie } 208582436617SMiao Xie 208628b21c55SFilipe Manana /* 208728b21c55SFilipe Manana * Add a pending snapshot associated with the given transaction handle to the 208828b21c55SFilipe Manana * respective handle. This must be called after the transaction commit started 208928b21c55SFilipe Manana * and while holding fs_info->trans_lock. 209028b21c55SFilipe Manana * This serves to guarantee a caller of btrfs_commit_transaction() that it can 209128b21c55SFilipe Manana * safely free the pending snapshot pointer in case btrfs_commit_transaction() 209228b21c55SFilipe Manana * returns an error. 209328b21c55SFilipe Manana */ 209428b21c55SFilipe Manana static void add_pending_snapshot(struct btrfs_trans_handle *trans) 209528b21c55SFilipe Manana { 209628b21c55SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 209728b21c55SFilipe Manana 209828b21c55SFilipe Manana if (!trans->pending_snapshot) 209928b21c55SFilipe Manana return; 210028b21c55SFilipe Manana 210128b21c55SFilipe Manana lockdep_assert_held(&trans->fs_info->trans_lock); 210228b21c55SFilipe Manana ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 210328b21c55SFilipe Manana 210428b21c55SFilipe Manana list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 210528b21c55SFilipe Manana } 210628b21c55SFilipe Manana 2107e55958c8SIoannis Angelakopoulos static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2108e55958c8SIoannis Angelakopoulos { 2109e55958c8SIoannis Angelakopoulos fs_info->commit_stats.commit_count++; 2110e55958c8SIoannis Angelakopoulos fs_info->commit_stats.last_commit_dur = interval; 2111e55958c8SIoannis Angelakopoulos fs_info->commit_stats.max_commit_dur = 2112e55958c8SIoannis Angelakopoulos max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2113e55958c8SIoannis Angelakopoulos fs_info->commit_stats.total_commit_dur += interval; 2114e55958c8SIoannis Angelakopoulos } 2115e55958c8SIoannis Angelakopoulos 21163a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 211779154b1bSChris Mason { 21183a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 211949b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 21208fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 212125287e0aSMiao Xie int ret; 2122e55958c8SIoannis Angelakopoulos ktime_t start_time; 2123e55958c8SIoannis Angelakopoulos ktime_t interval; 212479154b1bSChris Mason 212535b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 21263e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 212735b814f3SNikolay Borisov 2128c52cc7b7SJosef Bacik clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); 2129c52cc7b7SJosef Bacik 21308d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2131bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 213225287e0aSMiao Xie ret = cur_trans->aborted; 21333e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 213425287e0aSMiao Xie } 213549b25e05SJeff Mahoney 2136f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2137f45c752bSJosef Bacik trans->block_rsv = NULL; 2138f45c752bSJosef Bacik 2139e19eb11fSJosef Bacik /* 2140e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2141e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2142e19eb11fSJosef Bacik */ 2143e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2144e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2145e19eb11fSJosef Bacik /* 2146e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2147e19eb11fSJosef Bacik * Any running threads may add more while we are here. 214856bec294SChris Mason */ 2149c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 21503e738c53SIoannis Angelakopoulos if (ret) 21513e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2152e19eb11fSJosef Bacik } 215356bec294SChris Mason 21546c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2155ea658badSJosef Bacik 21563204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 21571bbc621eSChris Mason int run_it = 0; 21581bbc621eSChris Mason 21591bbc621eSChris Mason /* this mutex is also taken before trying to set 21601bbc621eSChris Mason * block groups readonly. We need to make sure 21611bbc621eSChris Mason * that nobody has set a block group readonly 21621bbc621eSChris Mason * after a extents from that block group have been 21631bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21641bbc621eSChris Mason * will wait for the transaction to commit if it 21653204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21661bbc621eSChris Mason * 21673204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21683204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21691bbc621eSChris Mason * hurt to have more than one go through, but there's no 21701bbc621eSChris Mason * real advantage to it either. 21711bbc621eSChris Mason */ 21720b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21733204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21743204d33cSJosef Bacik &cur_trans->flags)) 21751bbc621eSChris Mason run_it = 1; 21760b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21771bbc621eSChris Mason 2178f9cacae3SNikolay Borisov if (run_it) { 217921217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21803e738c53SIoannis Angelakopoulos if (ret) 21813e738c53SIoannis Angelakopoulos goto lockdep_trans_commit_start_release; 2182f9cacae3SNikolay Borisov } 2183f9cacae3SNikolay Borisov } 21841bbc621eSChris Mason 21850b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21864a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2187d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2188d0c2f4faSFilipe Manana 218928b21c55SFilipe Manana add_pending_snapshot(trans); 219028b21c55SFilipe Manana 21910b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21929b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2193ccd467d6SChris Mason 2194d0c2f4faSFilipe Manana if (trans->in_fsync) 2195d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 21963e738c53SIoannis Angelakopoulos 21973e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, 21983e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 2199d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2200d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 220115ee9bc7SJosef Bacik 2202bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2203b4924a0fSLiu Bo ret = cur_trans->aborted; 2204b4924a0fSLiu Bo 2205724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 220615ee9bc7SJosef Bacik 220749b25e05SJeff Mahoney return ret; 220879154b1bSChris Mason } 22094313b399SChris Mason 22104a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 22110b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 22123e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2213bb9c12c9SSage Weil 22140b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2215d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2216d0c2f4faSFilipe Manana 2217d0c2f4faSFilipe Manana if (trans->in_fsync) 2218d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2219d0c2f4faSFilipe Manana 2220ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2221ccd467d6SChris Mason struct btrfs_transaction, list); 2222d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 22239b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 22240b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2225ccd467d6SChris Mason 2226d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2227d0c2f4faSFilipe Manana 2228bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2229ccd467d6SChris Mason 2230724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 22311f9b8c8fSFilipe Manana if (ret) 2232e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2233a4abeea4SJosef Bacik } else { 22340b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2235ccd467d6SChris Mason } 2236a4abeea4SJosef Bacik } else { 22370b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2238cb2d3dadSFilipe Manana /* 2239cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2240cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2241cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2242cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2243cb2d3dadSFilipe Manana */ 224484961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2245cb2d3dadSFilipe Manana ret = -EROFS; 2246e1489b4fSIoannis Angelakopoulos goto lockdep_release; 2247cb2d3dadSFilipe Manana } 2248ccd467d6SChris Mason } 224915ee9bc7SJosef Bacik 2250e55958c8SIoannis Angelakopoulos /* 2251e55958c8SIoannis Angelakopoulos * Get the time spent on the work done by the commit thread and not 2252e55958c8SIoannis Angelakopoulos * the time spent waiting on a previous commit 2253e55958c8SIoannis Angelakopoulos */ 2254e55958c8SIoannis Angelakopoulos start_time = ktime_get_ns(); 2255e55958c8SIoannis Angelakopoulos 22560860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 22570860adfdSMiao Xie 225888090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 225982436617SMiao Xie if (ret) 2260e1489b4fSIoannis Angelakopoulos goto lockdep_release; 226182436617SMiao Xie 2262e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 226349b25e05SJeff Mahoney if (ret) 2264e1489b4fSIoannis Angelakopoulos goto lockdep_release; 226516cdcec7SMiao Xie 22665a9ba670SIoannis Angelakopoulos /* 22675a9ba670SIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 22685a9ba670SIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 22695a9ba670SIoannis Angelakopoulos * lockdep map as a writer. 22705a9ba670SIoannis Angelakopoulos */ 22715a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 22725a9ba670SIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2273581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2274581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2275ed3b3d31SChris Mason 2276581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2277e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2278e1489b4fSIoannis Angelakopoulos if (ret) { 2279e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2280ca469637SMiao Xie goto cleanup_transaction; 2281e1489b4fSIoannis Angelakopoulos } 2282ca469637SMiao Xie 228388090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2284cb7ab021SWang Shilong 228548778179SFilipe Manana /* 228648778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 228748778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 228848778179SFilipe Manana * extents complete we lose logged data after a power failure. 228948778179SFilipe Manana */ 22908b53779eSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 229148778179SFilipe Manana wait_event(cur_trans->pending_wait, 229248778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 229348778179SFilipe Manana 22942ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2295ed0ca140SJosef Bacik /* 2296ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2297ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22984a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2299ed0ca140SJosef Bacik */ 23000b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 230128b21c55SFilipe Manana add_pending_snapshot(trans); 23024a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 23030b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2304e1489b4fSIoannis Angelakopoulos 2305e1489b4fSIoannis Angelakopoulos /* 2306e1489b4fSIoannis Angelakopoulos * The thread has started/joined the transaction thus it holds the 2307e1489b4fSIoannis Angelakopoulos * lockdep map as a reader. It has to release it before acquiring the 2308e1489b4fSIoannis Angelakopoulos * lockdep map as a writer. 2309e1489b4fSIoannis Angelakopoulos */ 2310e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2311e1489b4fSIoannis Angelakopoulos btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2312ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2313ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 231415ee9bc7SJosef Bacik 2315fdfbf020SJosef Bacik /* 23163e738c53SIoannis Angelakopoulos * Make lockdep happy by acquiring the state locks after 23173e738c53SIoannis Angelakopoulos * btrfs_trans_num_writers is released. If we acquired the state locks 23183e738c53SIoannis Angelakopoulos * before releasing the btrfs_trans_num_writers lock then lockdep would 23193e738c53SIoannis Angelakopoulos * complain because we did not follow the reverse order unlocking rule. 23203e738c53SIoannis Angelakopoulos */ 23213e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 23223e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 23233e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23243e738c53SIoannis Angelakopoulos 23253e738c53SIoannis Angelakopoulos /* 2326fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2327fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2328fdfbf020SJosef Bacik * kthread could do the work. 2329fdfbf020SJosef Bacik */ 2330fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2331fdfbf020SJosef Bacik 2332bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23332cba30f1SMiao Xie ret = cur_trans->aborted; 23343e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 23356cf7f77eSWang Shilong goto scrub_continue; 23362cba30f1SMiao Xie } 23377585717fSChris Mason /* 23387585717fSChris Mason * the reloc mutex makes sure that we stop 23397585717fSChris Mason * the balancing code from coming in and moving 23407585717fSChris Mason * extents around in the middle of the commit 23417585717fSChris Mason */ 23420b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23437585717fSChris Mason 234442874b3dSMiao Xie /* 234542874b3dSMiao Xie * We needn't worry about the delayed items because we will 234642874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 234742874b3dSMiao Xie * core function of the snapshot creation. 234842874b3dSMiao Xie */ 234908d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 235056e9f6eaSDavid Sterba if (ret) 235156e9f6eaSDavid Sterba goto unlock_reloc; 23523063d29fSChris Mason 235342874b3dSMiao Xie /* 235442874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 235542874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 235642874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 235742874b3dSMiao Xie * them. 235842874b3dSMiao Xie * 235942874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 236042874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 236142874b3dSMiao Xie * the nodes and leaves. 236242874b3dSMiao Xie */ 2363e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 236456e9f6eaSDavid Sterba if (ret) 236556e9f6eaSDavid Sterba goto unlock_reloc; 236616cdcec7SMiao Xie 2367c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 236856e9f6eaSDavid Sterba if (ret) 236956e9f6eaSDavid Sterba goto unlock_reloc; 237056bec294SChris Mason 2371e999376fSChris Mason /* 2372e999376fSChris Mason * make sure none of the code above managed to slip in a 2373e999376fSChris Mason * delayed item 2374e999376fSChris Mason */ 2375ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2376e999376fSChris Mason 23772c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2378dc17ff8fSChris Mason 23797e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 238056e9f6eaSDavid Sterba if (ret) 2381dfba78dcSFilipe Manana goto unlock_reloc; 238254aa1f4dSChris Mason 23835d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2384e02119d5SChris Mason * safe to free the root of tree log roots 2385e02119d5SChris Mason */ 23860b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2387e02119d5SChris Mason 23880ed4792aSQu Wenruo /* 23890ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 23900ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 23910ed4792aSQu Wenruo */ 2392460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 239356e9f6eaSDavid Sterba if (ret < 0) 2394dfba78dcSFilipe Manana goto unlock_reloc; 23950ed4792aSQu Wenruo 23969386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 239756e9f6eaSDavid Sterba if (ret) 2398dfba78dcSFilipe Manana goto unlock_reloc; 239954aa1f4dSChris Mason 24002cba30f1SMiao Xie /* 24012cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 24022cba30f1SMiao Xie * update ->aborted, check it. 24032cba30f1SMiao Xie */ 2404bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 24052cba30f1SMiao Xie ret = cur_trans->aborted; 2406dfba78dcSFilipe Manana goto unlock_reloc; 24072cba30f1SMiao Xie } 24082cba30f1SMiao Xie 24090b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 24105f39d397SChris Mason 24110b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 24120b246afaSJeff Mahoney fs_info->tree_root->node); 24130b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 24149e351cc8SJosef Bacik &cur_trans->switch_commits); 24155d4f98a2SYan Zheng 24160b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 24170b246afaSJeff Mahoney fs_info->chunk_root->node); 24180b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 24199e351cc8SJosef Bacik &cur_trans->switch_commits); 24209e351cc8SJosef Bacik 2421f7238e50SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2422f7238e50SJosef Bacik btrfs_set_root_node(&fs_info->block_group_root->root_item, 2423f7238e50SJosef Bacik fs_info->block_group_root->node); 2424f7238e50SJosef Bacik list_add_tail(&fs_info->block_group_root->dirty_list, 2425f7238e50SJosef Bacik &cur_trans->switch_commits); 2426f7238e50SJosef Bacik } 2427f7238e50SJosef Bacik 2428889bfa39SJosef Bacik switch_commit_roots(trans); 24295d4f98a2SYan Zheng 2430ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 24311bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 24322ff7e61eSJeff Mahoney update_super_roots(fs_info); 2433e02119d5SChris Mason 24340b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 24350b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 24360b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 24370b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2438ccd467d6SChris Mason 2439bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2440935e5cc9SMiao Xie 24410b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 24420b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2443656f30dbSFilipe Manana 24444fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24454fbcdf66SFilipe Manana 2446dfba78dcSFilipe Manana /* 2447dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2448dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2449dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2450dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2451dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2452dfba78dcSFilipe Manana * writing its superblock. 2453dfba78dcSFilipe Manana */ 2454dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2455dfba78dcSFilipe Manana 24560b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 24574a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 24580b246afaSJeff Mahoney fs_info->running_transaction = NULL; 24590b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 24600b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2461b7ec40d7SChris Mason 24620b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 24633e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2464e6dcd2dcSChris Mason 246570458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 246649b25e05SJeff Mahoney if (ret) { 24670b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 246808748810SDavid Sterba "Error while writing out transaction"); 24690b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 24706cf7f77eSWang Shilong goto scrub_continue; 247149b25e05SJeff Mahoney } 247249b25e05SJeff Mahoney 2473d3575156SNaohiro Aota /* 2474d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2475d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2476d3575156SNaohiro Aota * buffers. 2477d3575156SNaohiro Aota */ 2478d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2479d3575156SNaohiro Aota 2480eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2481e02119d5SChris Mason /* 2482e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2483e02119d5SChris Mason * to go about their business 2484e02119d5SChris Mason */ 24850b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2486c1f32b7cSAnand Jain if (ret) 2487c1f32b7cSAnand Jain goto scrub_continue; 2488e02119d5SChris Mason 2489d0c2f4faSFilipe Manana /* 2490d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2491d0c2f4faSFilipe Manana * which can change it. 2492d0c2f4faSFilipe Manana */ 2493d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2494d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 24953e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2496d0c2f4faSFilipe Manana 24975ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 24984313b399SChris Mason 24993204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 25000b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 250113212b54SZhao Lei 25020b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 25034a9d8bdeSMiao Xie /* 25044a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 25054a9d8bdeSMiao Xie * which can change it. 25064a9d8bdeSMiao Xie */ 25074a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 25082c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 25093e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25103de4586cSChris Mason 25110b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 251213c5a93eSJosef Bacik list_del_init(&cur_trans->list); 25130b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2514a4abeea4SJosef Bacik 2515724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2516724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 251758176a96SJosef Bacik 25180860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 25190b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2520b2b5ef5cSJan Kara 25212e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 25221abe9b8aSliubo 2523e55958c8SIoannis Angelakopoulos interval = ktime_get_ns() - start_time; 2524e55958c8SIoannis Angelakopoulos 25252ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2526a2de733cSArne Jansen 25279ed74f2dSJosef Bacik if (current->journal_info == trans) 25289ed74f2dSJosef Bacik current->journal_info = NULL; 25299ed74f2dSJosef Bacik 25302c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 253124bbcf04SYan, Zheng 2532e55958c8SIoannis Angelakopoulos update_commit_stats(fs_info, interval); 2533e55958c8SIoannis Angelakopoulos 253479154b1bSChris Mason return ret; 253549b25e05SJeff Mahoney 253656e9f6eaSDavid Sterba unlock_reloc: 253756e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 25383e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 25396cf7f77eSWang Shilong scrub_continue: 25403e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 25413e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 25422ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 254349b25e05SJeff Mahoney cleanup_transaction: 2544dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2545c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 25464fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 25470e721106SJosef Bacik trans->block_rsv = NULL; 25480b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 254949b25e05SJeff Mahoney if (current->journal_info == trans) 255049b25e05SJeff Mahoney current->journal_info = NULL; 255197cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 255249b25e05SJeff Mahoney 255349b25e05SJeff Mahoney return ret; 2554e1489b4fSIoannis Angelakopoulos 2555e1489b4fSIoannis Angelakopoulos lockdep_release: 25565a9ba670SIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2557e1489b4fSIoannis Angelakopoulos btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2558e1489b4fSIoannis Angelakopoulos goto cleanup_transaction; 25593e738c53SIoannis Angelakopoulos 25603e738c53SIoannis Angelakopoulos lockdep_trans_commit_start_release: 25613e738c53SIoannis Angelakopoulos btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 25623e738c53SIoannis Angelakopoulos btrfs_end_transaction(trans); 25633e738c53SIoannis Angelakopoulos return ret; 256479154b1bSChris Mason } 256579154b1bSChris Mason 2566d352ac68SChris Mason /* 25679d1a2a3aSDavid Sterba * return < 0 if error 25689d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 25699d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 25709d1a2a3aSDavid Sterba * 25719d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 25729d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 25739d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 25749d1a2a3aSDavid Sterba * few seconds later. 2575d352ac68SChris Mason */ 257633c44184SJosef Bacik int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2577e9d0b13bSChris Mason { 257833c44184SJosef Bacik struct btrfs_root *root; 25799d1a2a3aSDavid Sterba int ret; 2580e9d0b13bSChris Mason 2581a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 25829d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 25839d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 25849d1a2a3aSDavid Sterba return 0; 25859d1a2a3aSDavid Sterba } 25869d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 25879d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2588cfad392bSJosef Bacik list_del_init(&root->root_list); 2589a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 25905d4f98a2SYan Zheng 25914fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 259276dda93cSYan, Zheng 259316cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 259416cdcec7SMiao Xie 259576dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 259676dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 25970078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 259876dda93cSYan, Zheng else 25990078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 260032471dc2SDavid Sterba 2601dc9492c1SJosef Bacik btrfs_put_root(root); 26026596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2603e9d0b13bSChris Mason } 2604572d9ab7SDavid Sterba 2605956504a3SJosef Bacik int __init btrfs_transaction_init(void) 2606956504a3SJosef Bacik { 2607956504a3SJosef Bacik btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2608956504a3SJosef Bacik sizeof(struct btrfs_trans_handle), 0, 2609956504a3SJosef Bacik SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2610956504a3SJosef Bacik if (!btrfs_trans_handle_cachep) 2611956504a3SJosef Bacik return -ENOMEM; 2612956504a3SJosef Bacik return 0; 2613956504a3SJosef Bacik } 2614956504a3SJosef Bacik 2615956504a3SJosef Bacik void __cold btrfs_transaction_exit(void) 2616956504a3SJosef Bacik { 2617956504a3SJosef Bacik kmem_cache_destroy(btrfs_trans_handle_cachep); 2618956504a3SJosef Bacik } 2619