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> 9d3c2fdcfSChris Mason #include <linux/writeback.h> 105f39d397SChris Mason #include <linux/pagemap.h> 115f2cc086SChris Mason #include <linux/blkdev.h> 128ea05e3aSAlexander Block #include <linux/uuid.h> 13602cbe91SDavid Sterba #include "misc.h" 1479154b1bSChris Mason #include "ctree.h" 1579154b1bSChris Mason #include "disk-io.h" 1679154b1bSChris Mason #include "transaction.h" 17925baeddSChris Mason #include "locking.h" 18e02119d5SChris Mason #include "tree-log.h" 19733f4fbbSStefan Behrens #include "volumes.h" 208dabb742SStefan Behrens #include "dev-replace.h" 21fcebe456SJosef Bacik #include "qgroup.h" 22aac0023cSJosef Bacik #include "block-group.h" 239c343784SJosef Bacik #include "space-info.h" 24d3575156SNaohiro Aota #include "zoned.h" 2579154b1bSChris Mason 260f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0 270f7d52f4SChris Mason 2861c047b5SQu Wenruo /* 2961c047b5SQu Wenruo * Transaction states and transitions 3061c047b5SQu Wenruo * 3161c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 3261c047b5SQu Wenruo * | 3361c047b5SQu Wenruo * | To next stage: 3461c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 3561c047b5SQu Wenruo * V 3661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 3761c047b5SQu Wenruo * | 3861c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 3961c047b5SQu Wenruo * | start_transaction() variants. 4061c047b5SQu Wenruo * | 4161c047b5SQu Wenruo * | To next stage: 4261c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 4361c047b5SQu Wenruo * | transaction N 4461c047b5SQu Wenruo * V 4561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 4661c047b5SQu Wenruo * | 4761c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 4861c047b5SQu Wenruo * | is one 4961c047b5SQu Wenruo * | 5061c047b5SQu Wenruo * | Then one of the following happes: 5161c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 5261c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 5361c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 5461c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 5561c047b5SQu Wenruo * | 5661c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 5761c047b5SQu Wenruo * | to this running transaction. 5861c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 5961c047b5SQu Wenruo * | transaction N+1. 6061c047b5SQu Wenruo * | 6161c047b5SQu Wenruo * | To next stage: 6261c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 6361c047b5SQu Wenruo * | haven been released. 6461c047b5SQu Wenruo * V 6561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 6661c047b5SQu Wenruo * | 6761c047b5SQu Wenruo * | The heavy lifting transaction work is started. 6861c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 6961c047b5SQu Wenruo * | snapshots, running qgroups. 7061c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 7161c047b5SQu Wenruo * | trees. 7261c047b5SQu Wenruo * | 7361c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 7461c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 7561c047b5SQu Wenruo * | 7661c047b5SQu Wenruo * | To next stage: 7761c047b5SQu Wenruo * | Until all supporting trees are updated. 7861c047b5SQu Wenruo * V 7961c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 8061c047b5SQu Wenruo * | Transaction N+1 8161c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 8261c047b5SQu Wenruo * | need to write them back to disk and update | 8361c047b5SQu Wenruo * | super blocks. | 8461c047b5SQu Wenruo * | | 8561c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 8661c047b5SQu Wenruo * | start. | 8761c047b5SQu Wenruo * | All new start_transaction() calls will be | 8861c047b5SQu Wenruo * | attached to transid N+1. | 8961c047b5SQu Wenruo * | | 9061c047b5SQu Wenruo * | To next stage: | 9161c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 9261c047b5SQu Wenruo * | written to block devices | 9361c047b5SQu Wenruo * V | 9461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 9561c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 9661c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 9761c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 9861c047b5SQu Wenruo */ 99e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1004a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 101bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 102bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1034a9d8bdeSMiao Xie __TRANS_ATTACH | 104a6d155d2SFilipe Manana __TRANS_JOIN | 105a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 106bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1074a9d8bdeSMiao Xie __TRANS_ATTACH | 1084a9d8bdeSMiao Xie __TRANS_JOIN | 109a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 110a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 111d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 112d0c2f4faSFilipe Manana __TRANS_ATTACH | 113d0c2f4faSFilipe Manana __TRANS_JOIN | 114d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 115d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 116bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1174a9d8bdeSMiao Xie __TRANS_ATTACH | 1184a9d8bdeSMiao Xie __TRANS_JOIN | 119a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 120a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1214a9d8bdeSMiao Xie }; 1224a9d8bdeSMiao Xie 123724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 12479154b1bSChris Mason { 1259b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1269b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 127a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1285c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1295c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 13081f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 13181f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1321262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 133ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 134ab8d0fc4SJeff Mahoney "pending csums is %llu", 1351262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1367785a663SFilipe Manana /* 1377785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1387785a663SFilipe Manana * because the transaction was aborted and a commit did not 1397785a663SFilipe Manana * happen (things failed before writing the new superblock 1407785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1417785a663SFilipe Manana * discard the physical locations of the block groups. 1427785a663SFilipe Manana */ 1437785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 14432da5386SDavid Sterba struct btrfs_block_group *cache; 1457785a663SFilipe Manana 1467785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 14732da5386SDavid Sterba struct btrfs_block_group, 1487785a663SFilipe Manana bg_list); 1497785a663SFilipe Manana list_del_init(&cache->bg_list); 1506b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1517785a663SFilipe Manana btrfs_put_block_group(cache); 1527785a663SFilipe Manana } 153bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1544b5faeacSDavid Sterba kfree(transaction); 15579154b1bSChris Mason } 15678fae27eSChris Mason } 15779154b1bSChris Mason 158889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 159817d52f8SJosef Bacik { 160889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 16116916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1629e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 16327d56e62SJosef Bacik struct btrfs_caching_control *caching_ctl, *next; 1649e351cc8SJosef Bacik 1659e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 166889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1679e351cc8SJosef Bacik dirty_list) { 1689e351cc8SJosef Bacik list_del_init(&root->dirty_list); 169817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 170817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 17141e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 172370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1739e351cc8SJosef Bacik } 1742b9dbef2SJosef Bacik 1752b9dbef2SJosef Bacik /* We can free old roots now. */ 176889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 177889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 178889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1792b9dbef2SJosef Bacik struct btrfs_root, root_list); 1802b9dbef2SJosef Bacik list_del_init(&root->root_list); 181889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 182889bfa39SJosef Bacik btrfs_free_log(trans, root); 1832b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 184889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1852b9dbef2SJosef Bacik } 186889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 18727d56e62SJosef Bacik 18827d56e62SJosef Bacik /* 18927d56e62SJosef Bacik * We have to update the last_byte_to_unpin under the commit_root_sem, 19027d56e62SJosef Bacik * at the same time we swap out the commit roots. 19127d56e62SJosef Bacik * 19227d56e62SJosef Bacik * This is because we must have a real view of the last spot the caching 19327d56e62SJosef Bacik * kthreads were while caching. Consider the following views of the 19427d56e62SJosef Bacik * extent tree for a block group 19527d56e62SJosef Bacik * 19627d56e62SJosef Bacik * commit root 19727d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 19827d56e62SJosef Bacik * |\\\\| |\\\\|\\\\| |\\\\|\\\\| 19927d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20027d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 20127d56e62SJosef Bacik * 20227d56e62SJosef Bacik * new commit root 20327d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20427d56e62SJosef Bacik * | | | |\\\\| | |\\\\| 20527d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20627d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 20727d56e62SJosef Bacik * 20827d56e62SJosef Bacik * If the cache_ctl->progress was at 3, then we are only allowed to 20927d56e62SJosef Bacik * unpin [0,1) and [2,3], because the caching thread has already 21027d56e62SJosef Bacik * processed those extents. We are not allowed to unpin [5,6), because 21127d56e62SJosef Bacik * the caching thread will re-start it's search from 3, and thus find 21227d56e62SJosef Bacik * the hole from [4,6) to add to the free space cache. 21327d56e62SJosef Bacik */ 214bbb86a37SJosef Bacik spin_lock(&fs_info->block_group_cache_lock); 21527d56e62SJosef Bacik list_for_each_entry_safe(caching_ctl, next, 21627d56e62SJosef Bacik &fs_info->caching_block_groups, list) { 21727d56e62SJosef Bacik struct btrfs_block_group *cache = caching_ctl->block_group; 21827d56e62SJosef Bacik 21927d56e62SJosef Bacik if (btrfs_block_group_done(cache)) { 22027d56e62SJosef Bacik cache->last_byte_to_unpin = (u64)-1; 22127d56e62SJosef Bacik list_del_init(&caching_ctl->list); 22227d56e62SJosef Bacik btrfs_put_caching_control(caching_ctl); 22327d56e62SJosef Bacik } else { 22427d56e62SJosef Bacik cache->last_byte_to_unpin = caching_ctl->progress; 22527d56e62SJosef Bacik } 22627d56e62SJosef Bacik } 227bbb86a37SJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 2289e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 229817d52f8SJosef Bacik } 230817d52f8SJosef Bacik 2310860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2320860adfdSMiao Xie unsigned int type) 2330860adfdSMiao Xie { 2340860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2350860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2360860adfdSMiao Xie } 2370860adfdSMiao Xie 2380860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2390860adfdSMiao Xie unsigned int type) 2400860adfdSMiao Xie { 2410860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2420860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2430860adfdSMiao Xie } 2440860adfdSMiao Xie 2450860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2460860adfdSMiao Xie unsigned int type) 2470860adfdSMiao Xie { 2480860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2490860adfdSMiao Xie } 2500860adfdSMiao Xie 2510860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2520860adfdSMiao Xie { 2530860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 254178260b2SMiao Xie } 255178260b2SMiao Xie 256d352ac68SChris Mason /* 257fb6dea26SJosef Bacik * To be called after all the new block groups attached to the transaction 258fb6dea26SJosef Bacik * handle have been created (btrfs_create_pending_block_groups()). 259fb6dea26SJosef Bacik */ 260fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 261fb6dea26SJosef Bacik { 262fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 263eafa4fd0SFilipe Manana struct btrfs_transaction *cur_trans = trans->transaction; 264fb6dea26SJosef Bacik 265fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 266fb6dea26SJosef Bacik return; 267fb6dea26SJosef Bacik 268fb6dea26SJosef Bacik WARN_ON_ONCE(!list_empty(&trans->new_bgs)); 269fb6dea26SJosef Bacik 270fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 27163f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 272eafa4fd0SFilipe Manana atomic64_sub(trans->chunk_bytes_reserved, &cur_trans->chunk_bytes_reserved); 273eafa4fd0SFilipe Manana cond_wake_up(&cur_trans->chunk_reserve_wait); 274fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 275fb6dea26SJosef Bacik } 276fb6dea26SJosef Bacik 277fb6dea26SJosef Bacik /* 278d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 279d352ac68SChris Mason */ 2802ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2812ff7e61eSJeff Mahoney unsigned int type) 28279154b1bSChris Mason { 28379154b1bSChris Mason struct btrfs_transaction *cur_trans; 284a4abeea4SJosef Bacik 28519ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 286d43317dcSChris Mason loop: 28749b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 28887533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 28919ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 29049b25e05SJeff Mahoney return -EROFS; 29149b25e05SJeff Mahoney } 29249b25e05SJeff Mahoney 29319ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 294a4abeea4SJosef Bacik if (cur_trans) { 295bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 29619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 29749b25e05SJeff Mahoney return cur_trans->aborted; 298871383beSDavid Sterba } 2994a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 300178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 301178260b2SMiao Xie return -EBUSY; 302178260b2SMiao Xie } 3039b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 304a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 3050860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 30619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 307a4abeea4SJosef Bacik return 0; 308a4abeea4SJosef Bacik } 30919ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 310a4abeea4SJosef Bacik 311354aa0fbSMiao Xie /* 312354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 313354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 314354aa0fbSMiao Xie */ 315354aa0fbSMiao Xie if (type == TRANS_ATTACH) 316354aa0fbSMiao Xie return -ENOENT; 317354aa0fbSMiao Xie 3184a9d8bdeSMiao Xie /* 3194a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3204a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3214a9d8bdeSMiao Xie */ 3224a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3234a9d8bdeSMiao Xie 3244b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 325db5b493aSTsutomu Itoh if (!cur_trans) 326db5b493aSTsutomu Itoh return -ENOMEM; 327d43317dcSChris Mason 32819ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 32919ae4e81SJan Schmidt if (fs_info->running_transaction) { 330d43317dcSChris Mason /* 331d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3324a9d8bdeSMiao Xie * to redo the checks above 333d43317dcSChris Mason */ 3344b5faeacSDavid Sterba kfree(cur_trans); 335d43317dcSChris Mason goto loop; 33687533c47SMiao Xie } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 337e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3384b5faeacSDavid Sterba kfree(cur_trans); 3397b8b92afSJosef Bacik return -EROFS; 340a4abeea4SJosef Bacik } 341d43317dcSChris Mason 342ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 34348778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 34448778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 34513c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3460860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 34779154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 34879154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3494a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 350a4abeea4SJosef Bacik /* 351a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 352a4abeea4SJosef Bacik * commit the transaction. 353a4abeea4SJosef Bacik */ 3549b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3553204d33cSJosef Bacik cur_trans->flags = 0; 356afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 35756bec294SChris Mason 358a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 359a099d0fdSAlexandru Moise 3605c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3613368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 362d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 36320b297d6SJan Schmidt 36420b297d6SJan Schmidt /* 36520b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 36620b297d6SJan Schmidt * the log must never go across transaction boundaries. 36720b297d6SJan Schmidt */ 36820b297d6SJan Schmidt smp_mb(); 36931b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3705d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 37131b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3725d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 373fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 37420b297d6SJan Schmidt 37556bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 37656bec294SChris Mason 3773063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 378bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3799e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 380ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3811bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3822b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3831bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 384ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 385e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3862b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 387d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 388d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 389eafa4fd0SFilipe Manana atomic64_set(&cur_trans->chunk_bytes_reserved, 0); 390eafa4fd0SFilipe Manana init_waitqueue_head(&cur_trans->chunk_reserve_wait); 39119ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 392c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 39343eb5f29SQu Wenruo IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); 394fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 395fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 39619ae4e81SJan Schmidt fs_info->generation++; 39719ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 39819ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 39949b25e05SJeff Mahoney cur_trans->aborted = 0; 40019ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 40115ee9bc7SJosef Bacik 40279154b1bSChris Mason return 0; 40379154b1bSChris Mason } 40479154b1bSChris Mason 405d352ac68SChris Mason /* 40692a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 40792a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 40892a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 40992a7cc42SQu Wenruo * transaction commits. 410d352ac68SChris Mason */ 4117585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 4126426c7adSQu Wenruo struct btrfs_root *root, 4136426c7adSQu Wenruo int force) 4146702ed49SChris Mason { 4150b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4160b246afaSJeff Mahoney 41792a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4186426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4190b246afaSJeff Mahoney WARN_ON(root == fs_info->extent_root); 4204d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4215d4f98a2SYan Zheng 4227585717fSChris Mason /* 42327cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4247585717fSChris Mason * we have the reloc mutex held now, so there 4257585717fSChris Mason * is only one writer in this function 4267585717fSChris Mason */ 42727cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4287585717fSChris Mason 42927cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4307585717fSChris Mason * they find our root->last_trans update 4317585717fSChris Mason */ 4327585717fSChris Mason smp_wmb(); 4337585717fSChris Mason 4340b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4356426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 4360b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 437a4abeea4SJosef Bacik return 0; 438a4abeea4SJosef Bacik } 4390b246afaSJeff Mahoney radix_tree_tag_set(&fs_info->fs_roots_radix, 4406702ed49SChris Mason (unsigned long)root->root_key.objectid, 4416702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 4420b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4437585717fSChris Mason root->last_trans = trans->transid; 4447585717fSChris Mason 4457585717fSChris Mason /* this is pretty tricky. We don't want to 4467585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4477585717fSChris Mason * unless we're really doing the first setup for this root in 4487585717fSChris Mason * this transaction. 4497585717fSChris Mason * 4507585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4517585717fSChris Mason * if we want to take the expensive mutex. 4527585717fSChris Mason * 4537585717fSChris Mason * But, we have to set root->last_trans before we 4547585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4557585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 45627cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4577585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4587585717fSChris Mason * 4597585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4607585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4617585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4627585717fSChris Mason * done before we pop in the zero below 4637585717fSChris Mason */ 4645d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 465c7548af6SChris Mason smp_mb__before_atomic(); 46627cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4676702ed49SChris Mason } 4685d4f98a2SYan Zheng return 0; 4696702ed49SChris Mason } 4705d4f98a2SYan Zheng 4717585717fSChris Mason 4722b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4732b9dbef2SJosef Bacik struct btrfs_root *root) 4742b9dbef2SJosef Bacik { 4750b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4762b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4772b9dbef2SJosef Bacik 4782b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4792b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4802b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4812b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4822b9dbef2SJosef Bacik 4832b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 4840b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4850b246afaSJeff Mahoney radix_tree_tag_clear(&fs_info->fs_roots_radix, 4862b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4872b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 4880b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4892b9dbef2SJosef Bacik } 4902b9dbef2SJosef Bacik 4917585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4927585717fSChris Mason struct btrfs_root *root) 4937585717fSChris Mason { 4940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4950b246afaSJeff Mahoney 49692a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4977585717fSChris Mason return 0; 4987585717fSChris Mason 4997585717fSChris Mason /* 50027cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 5017585717fSChris Mason * and barriers 5027585717fSChris Mason */ 5037585717fSChris Mason smp_rmb(); 5047585717fSChris Mason if (root->last_trans == trans->transid && 50527cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 5067585717fSChris Mason return 0; 5077585717fSChris Mason 5080b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 5096426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 5100b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 5117585717fSChris Mason 5127585717fSChris Mason return 0; 5137585717fSChris Mason } 5147585717fSChris Mason 5154a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5164a9d8bdeSMiao Xie { 5173296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 518501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 519bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5204a9d8bdeSMiao Xie } 5214a9d8bdeSMiao Xie 522d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 523d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 524d352ac68SChris Mason * transaction might not be fully on disk. 525d352ac68SChris Mason */ 5262ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 52779154b1bSChris Mason { 528f9295749SChris Mason struct btrfs_transaction *cur_trans; 52979154b1bSChris Mason 5300b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5310b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5324a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5339b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5340b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 53572d63ed6SLi Zefan 5360b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 537501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 538bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 539724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 540a4abeea4SJosef Bacik } else { 5410b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 542f9295749SChris Mason } 54337d1aeeeSChris Mason } 54437d1aeeeSChris Mason 5452ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 54637d1aeeeSChris Mason { 5470b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 548a4abeea4SJosef Bacik return 0; 549a4abeea4SJosef Bacik 55092e2f7e3SNikolay Borisov if (type == TRANS_START) 551a4abeea4SJosef Bacik return 1; 552a4abeea4SJosef Bacik 553a22285a6SYan, Zheng return 0; 554a22285a6SYan, Zheng } 555a22285a6SYan, Zheng 55620dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 55720dd2cbfSMiao Xie { 5580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5590b246afaSJeff Mahoney 5600b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 56192a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 56220dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 56320dd2cbfSMiao Xie root->reloc_root) 56420dd2cbfSMiao Xie return false; 56520dd2cbfSMiao Xie 56620dd2cbfSMiao Xie return true; 56720dd2cbfSMiao Xie } 56820dd2cbfSMiao Xie 56908e007d2SMiao Xie static struct btrfs_trans_handle * 5705aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 571003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 572003d7c59SJeff Mahoney bool enforce_qgroups) 573a22285a6SYan, Zheng { 5740b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 575ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 576a22285a6SYan, Zheng struct btrfs_trans_handle *h; 577a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 578b5009945SJosef Bacik u64 num_bytes = 0; 579c5567237SArne Jansen u64 qgroup_reserved = 0; 58020dd2cbfSMiao Xie bool reloc_reserved = false; 5819c343784SJosef Bacik bool do_chunk_alloc = false; 58220dd2cbfSMiao Xie int ret; 583acce952bSliubo 58446c4e71eSFilipe Manana /* Send isn't supposed to start transactions. */ 5852755a0deSDavid Sterba ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); 58646c4e71eSFilipe Manana 5870b246afaSJeff Mahoney if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 588acce952bSliubo return ERR_PTR(-EROFS); 5892a1eb461SJosef Bacik 59046c4e71eSFilipe Manana if (current->journal_info) { 5910860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5922a1eb461SJosef Bacik h = current->journal_info; 593b50fff81SDavid Sterba refcount_inc(&h->use_count); 594b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5952a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5962a1eb461SJosef Bacik h->block_rsv = NULL; 5972a1eb461SJosef Bacik goto got_it; 5982a1eb461SJosef Bacik } 599b5009945SJosef Bacik 600b5009945SJosef Bacik /* 601b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 602b5009945SJosef Bacik * the appropriate flushing if need be. 603b5009945SJosef Bacik */ 604003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 605ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 606ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 607ba2c4d4eSJosef Bacik 6080b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 609733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 610003d7c59SJeff Mahoney enforce_qgroups); 611c5567237SArne Jansen if (ret) 612c5567237SArne Jansen return ERR_PTR(ret); 613c5567237SArne Jansen 614ba2c4d4eSJosef Bacik /* 615ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 616ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 617ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 618ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 619ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 620ba2c4d4eSJosef Bacik */ 6212bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6227f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 6237f9fe614SJosef Bacik delayed_refs_rsv->full == 0) { 624ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 625ba2c4d4eSJosef Bacik num_bytes <<= 1; 626ba2c4d4eSJosef Bacik } 627ba2c4d4eSJosef Bacik 62820dd2cbfSMiao Xie /* 62920dd2cbfSMiao Xie * Do the reservation for the relocation root creation 63020dd2cbfSMiao Xie */ 631ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6320b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 63320dd2cbfSMiao Xie reloc_reserved = true; 63420dd2cbfSMiao Xie } 63520dd2cbfSMiao Xie 636ba2c4d4eSJosef Bacik ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); 637ba2c4d4eSJosef Bacik if (ret) 638ba2c4d4eSJosef Bacik goto reserve_fail; 639ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 640ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 641ba2c4d4eSJosef Bacik delayed_refs_bytes); 642ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 643ba2c4d4eSJosef Bacik } 6449c343784SJosef Bacik 6459c343784SJosef Bacik if (rsv->space_info->force_alloc) 6469c343784SJosef Bacik do_chunk_alloc = true; 647ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 648ba2c4d4eSJosef Bacik !delayed_refs_rsv->full) { 649ba2c4d4eSJosef Bacik /* 650ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 651ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 652ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 653ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 654ba2c4d4eSJosef Bacik * here. 655ba2c4d4eSJosef Bacik */ 656ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 657b5009945SJosef Bacik if (ret) 658843fcf35SMiao Xie goto reserve_fail; 659b5009945SJosef Bacik } 660a22285a6SYan, Zheng again: 661f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 662843fcf35SMiao Xie if (!h) { 663843fcf35SMiao Xie ret = -ENOMEM; 664843fcf35SMiao Xie goto alloc_fail; 665843fcf35SMiao Xie } 666a22285a6SYan, Zheng 66798114659SJosef Bacik /* 66898114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 66998114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 67098114659SJosef Bacik * because we're already holding a ref. We need this because we could 67198114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 67298114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 673354aa0fbSMiao Xie * 674354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 675354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 67698114659SJosef Bacik */ 6770860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6780b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 679b2b5ef5cSJan Kara 6802ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6812ff7e61eSJeff Mahoney wait_current_trans(fs_info); 682a22285a6SYan, Zheng 683a4abeea4SJosef Bacik do { 6842ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 685178260b2SMiao Xie if (ret == -EBUSY) { 6862ff7e61eSJeff Mahoney wait_current_trans(fs_info); 687a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 688a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 689178260b2SMiao Xie ret = -ENOENT; 690178260b2SMiao Xie } 691a4abeea4SJosef Bacik } while (ret == -EBUSY); 692a4abeea4SJosef Bacik 693a43f7f82SLiu Bo if (ret < 0) 694843fcf35SMiao Xie goto join_fail; 6950f7d52f4SChris Mason 6960b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 697a22285a6SYan, Zheng 698a22285a6SYan, Zheng h->transid = cur_trans->transid; 699a22285a6SYan, Zheng h->transaction = cur_trans; 700d13603efSArne Jansen h->root = root; 701b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 70264b63580SJeff Mahoney h->fs_info = root->fs_info; 7037174109cSQu Wenruo 704a698d075SMiao Xie h->type = type; 705d9a0540aSFilipe Manana h->can_flush_pending_bgs = true; 706ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 707b7ec40d7SChris Mason 708a22285a6SYan, Zheng smp_mb(); 7093296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 7102ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 711abdd2e80SFilipe Manana current->journal_info = h; 7123a45bb20SJeff Mahoney btrfs_commit_transaction(h); 713a22285a6SYan, Zheng goto again; 714a22285a6SYan, Zheng } 7159ed74f2dSJosef Bacik 716b5009945SJosef Bacik if (num_bytes) { 7170b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7182bcc0328SLiu Bo h->transid, num_bytes, 1); 7190b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 720b5009945SJosef Bacik h->bytes_reserved = num_bytes; 72120dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 722a22285a6SYan, Zheng } 723a22285a6SYan, Zheng 7242a1eb461SJosef Bacik got_it: 725bcf3a3e7SNikolay Borisov if (!current->journal_info) 726a22285a6SYan, Zheng current->journal_info = h; 727fcc99734SQu Wenruo 728fcc99734SQu Wenruo /* 7299c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7309c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7319c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7329c343784SJosef Bacik * value here. 7339c343784SJosef Bacik */ 7349c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7359c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7369c343784SJosef Bacik 7379c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7389c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7399c343784SJosef Bacik } 7409c343784SJosef Bacik 7419c343784SJosef Bacik /* 742fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 743fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 744fcc99734SQu Wenruo * transaction. 745fcc99734SQu Wenruo * 746fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 747fcc99734SQu Wenruo * or we can deadlock. 748fcc99734SQu Wenruo */ 74968075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 75068075ea8SJosef Bacik if (ret) { 75168075ea8SJosef Bacik /* 75268075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 75368075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 75468075ea8SJosef Bacik * not just freed. 75568075ea8SJosef Bacik */ 75668075ea8SJosef Bacik btrfs_end_transaction(h); 75768075ea8SJosef Bacik return ERR_PTR(ret); 75868075ea8SJosef Bacik } 759fcc99734SQu Wenruo 76079154b1bSChris Mason return h; 761843fcf35SMiao Xie 762843fcf35SMiao Xie join_fail: 7630860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7640b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 765843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 766843fcf35SMiao Xie alloc_fail: 767843fcf35SMiao Xie if (num_bytes) 7682ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 76963f018beSNikolay Borisov num_bytes, NULL); 770843fcf35SMiao Xie reserve_fail: 771733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 772843fcf35SMiao Xie return ERR_PTR(ret); 77379154b1bSChris Mason } 77479154b1bSChris Mason 775f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7765aed1dd8SAlexandru Moise unsigned int num_items) 777f9295749SChris Mason { 77808e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 779003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 780f9295749SChris Mason } 781003d7c59SJeff Mahoney 7828eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7838eab77ffSFilipe Manana struct btrfs_root *root, 7847f9fe614SJosef Bacik unsigned int num_items) 7858eab77ffSFilipe Manana { 7867f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7877f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7888eab77ffSFilipe Manana } 7898407aa46SMiao Xie 7907a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 791f9295749SChris Mason { 792003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 793003d7c59SJeff Mahoney true); 794f9295749SChris Mason } 795f9295749SChris Mason 7968d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7970af3d00bSJosef Bacik { 798575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 799003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8000af3d00bSJosef Bacik } 8010af3d00bSJosef Bacik 802d4edf39bSMiao Xie /* 803a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 804a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 805a6d155d2SFilipe Manana */ 806a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 807a6d155d2SFilipe Manana { 808a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 809a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 810a6d155d2SFilipe Manana } 811a6d155d2SFilipe Manana 812a6d155d2SFilipe Manana /* 813d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 814d4edf39bSMiao Xie * 815d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 816d4edf39bSMiao Xie * don't want to start a new one. 817d4edf39bSMiao Xie * 818d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 819d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 820d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 821d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 822d4edf39bSMiao Xie * invoke 823d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 824d4edf39bSMiao Xie */ 825354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 82660376ce4SJosef Bacik { 827575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 828003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 82960376ce4SJosef Bacik } 83060376ce4SJosef Bacik 831d4edf39bSMiao Xie /* 83290b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 833d4edf39bSMiao Xie * 83452042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 835d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 836d4edf39bSMiao Xie * complete. 837d4edf39bSMiao Xie */ 838d4edf39bSMiao Xie struct btrfs_trans_handle * 839d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 840d4edf39bSMiao Xie { 841d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 842d4edf39bSMiao Xie 843575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 844003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8458d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8462ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 847d4edf39bSMiao Xie 848d4edf39bSMiao Xie return trans; 849d4edf39bSMiao Xie } 850d4edf39bSMiao Xie 851d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 852d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 853d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 85489ce8a63SChris Mason { 855d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 85689ce8a63SChris Mason } 85789ce8a63SChris Mason 8582ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 85946204592SSage Weil { 86046204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8618cd2807fSMiao Xie int ret = 0; 86246204592SSage Weil 86346204592SSage Weil if (transid) { 8640b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 865a4abeea4SJosef Bacik goto out; 86646204592SSage Weil 86746204592SSage Weil /* find specified transaction */ 8680b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8690b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 87046204592SSage Weil if (t->transid == transid) { 87146204592SSage Weil cur_trans = t; 8729b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8738cd2807fSMiao Xie ret = 0; 87446204592SSage Weil break; 87546204592SSage Weil } 8768cd2807fSMiao Xie if (t->transid > transid) { 8778cd2807fSMiao Xie ret = 0; 87846204592SSage Weil break; 87946204592SSage Weil } 8808cd2807fSMiao Xie } 8810b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 88242383020SSage Weil 88342383020SSage Weil /* 88442383020SSage Weil * The specified transaction doesn't exist, or we 88542383020SSage Weil * raced with btrfs_commit_transaction 88642383020SSage Weil */ 88742383020SSage Weil if (!cur_trans) { 8880b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 88942383020SSage Weil ret = -EINVAL; 8908cd2807fSMiao Xie goto out; 89142383020SSage Weil } 89246204592SSage Weil } else { 89346204592SSage Weil /* find newest transaction that is committing | committed */ 8940b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8950b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 89646204592SSage Weil list) { 8974a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 8984a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 8993473f3c0SJosef Bacik break; 90046204592SSage Weil cur_trans = t; 9019b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 90246204592SSage Weil break; 90346204592SSage Weil } 90446204592SSage Weil } 9050b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 90646204592SSage Weil if (!cur_trans) 907a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 90846204592SSage Weil } 90946204592SSage Weil 910d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 911724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 912a4abeea4SJosef Bacik out: 91346204592SSage Weil return ret; 91446204592SSage Weil } 91546204592SSage Weil 9162ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 91737d1aeeeSChris Mason { 9182ff7e61eSJeff Mahoney wait_current_trans(fs_info); 91937d1aeeeSChris Mason } 92037d1aeeeSChris Mason 9218a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9228929ecfaSYan, Zheng { 9232ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9240b246afaSJeff Mahoney 92564403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9268a8f4deaSNikolay Borisov return true; 92736ba022aSJosef Bacik 9282ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9298929ecfaSYan, Zheng } 9308929ecfaSYan, Zheng 931a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9328929ecfaSYan, Zheng { 9338929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9348929ecfaSYan, Zheng 9353296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 936e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 937a2633b6aSNikolay Borisov return true; 9388929ecfaSYan, Zheng 9392ff7e61eSJeff Mahoney return should_end_transaction(trans); 9408929ecfaSYan, Zheng } 9418929ecfaSYan, Zheng 942dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 943dc60c525SNikolay Borisov 9440e34693fSNikolay Borisov { 945dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 946dc60c525SNikolay Borisov 9470e34693fSNikolay Borisov if (!trans->block_rsv) { 9480e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9490e34693fSNikolay Borisov return; 9500e34693fSNikolay Borisov } 9510e34693fSNikolay Borisov 9520e34693fSNikolay Borisov if (!trans->bytes_reserved) 9530e34693fSNikolay Borisov return; 9540e34693fSNikolay Borisov 9550e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9560e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9570e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9580e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 95963f018beSNikolay Borisov trans->bytes_reserved, NULL); 9600e34693fSNikolay Borisov trans->bytes_reserved = 0; 9610e34693fSNikolay Borisov } 9620e34693fSNikolay Borisov 96389ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9643a45bb20SJeff Mahoney int throttle) 96579154b1bSChris Mason { 9663a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9678929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9684edc2ca3SDave Jones int err = 0; 969d6e4a428SChris Mason 970b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 971b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9722a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9732a1eb461SJosef Bacik return 0; 9742a1eb461SJosef Bacik } 9752a1eb461SJosef Bacik 976dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9774c13d758SJosef Bacik trans->block_rsv = NULL; 978c5567237SArne Jansen 9796c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 980ea658badSJosef Bacik 9814fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9824fbcdf66SFilipe Manana 9830860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9840b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9856df7881aSJosef Bacik 9868929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 98713c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 98813c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 9890860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 99089ce8a63SChris Mason 991093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 992724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 9939ed74f2dSJosef Bacik 9949ed74f2dSJosef Bacik if (current->journal_info == trans) 9959ed74f2dSJosef Bacik current->journal_info = NULL; 996ab78c84dSChris Mason 99724bbcf04SYan, Zheng if (throttle) 9982ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 99924bbcf04SYan, Zheng 1000bf31f87fSDavid Sterba if (TRANS_ABORTED(trans) || 10010b246afaSJeff Mahoney test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { 10024e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1003fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1004fbabd4a3SJosef Bacik err = trans->aborted; 1005fbabd4a3SJosef Bacik else 1006fbabd4a3SJosef Bacik err = -EROFS; 10074e121c06SJosef Bacik } 100849b25e05SJeff Mahoney 10094edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10104edc2ca3SDave Jones return err; 101179154b1bSChris Mason } 101279154b1bSChris Mason 10133a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 101489ce8a63SChris Mason { 10153a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 101689ce8a63SChris Mason } 101789ce8a63SChris Mason 10183a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 101989ce8a63SChris Mason { 10203a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 102116cdcec7SMiao Xie } 102216cdcec7SMiao Xie 1023d352ac68SChris Mason /* 1024d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1025d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1026690587d1SChris Mason * those extents are sent to disk but does not wait on them 1027d352ac68SChris Mason */ 10282ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10298cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 103079154b1bSChris Mason { 1031777e6bd7SChris Mason int err = 0; 10327c4452b9SChris Mason int werr = 0; 10330b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1034e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1035777e6bd7SChris Mason u64 start = 0; 10365f39d397SChris Mason u64 end; 10377c4452b9SChris Mason 10386300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10391728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1040e6138876SJosef Bacik mark, &cached_state)) { 1041663dfbb0SFilipe Manana bool wait_writeback = false; 1042663dfbb0SFilipe Manana 1043663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1044663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1045210aa277SDavid Sterba mark, &cached_state); 1046663dfbb0SFilipe Manana /* 1047663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1048663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1049663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1050663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1051bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1052bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1053bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1054bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1055bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1056663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 105741e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1058663dfbb0SFilipe Manana */ 1059663dfbb0SFilipe Manana if (err == -ENOMEM) { 1060663dfbb0SFilipe Manana err = 0; 1061663dfbb0SFilipe Manana wait_writeback = true; 1062663dfbb0SFilipe Manana } 1063663dfbb0SFilipe Manana if (!err) 10641728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10657c4452b9SChris Mason if (err) 10667c4452b9SChris Mason werr = err; 1067663dfbb0SFilipe Manana else if (wait_writeback) 1068663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1069e38e2ed7SFilipe Manana free_extent_state(cached_state); 1070663dfbb0SFilipe Manana cached_state = NULL; 10711728366eSJosef Bacik cond_resched(); 10721728366eSJosef Bacik start = end + 1; 10737c4452b9SChris Mason } 10746300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1075690587d1SChris Mason return werr; 1076690587d1SChris Mason } 1077690587d1SChris Mason 1078690587d1SChris Mason /* 1079690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1080690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1081690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1082690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1083690587d1SChris Mason */ 1084bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1085bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1086690587d1SChris Mason { 1087690587d1SChris Mason int err = 0; 1088690587d1SChris Mason int werr = 0; 10890b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1090e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1091690587d1SChris Mason u64 start = 0; 1092690587d1SChris Mason u64 end; 1093690587d1SChris Mason 10941728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1095e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1096663dfbb0SFilipe Manana /* 1097663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1098663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1099663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1100663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1101663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 110241e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1103663dfbb0SFilipe Manana */ 1104663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1105ae0f1625SDavid Sterba EXTENT_NEED_WAIT, 0, 0, &cached_state); 1106663dfbb0SFilipe Manana if (err == -ENOMEM) 1107663dfbb0SFilipe Manana err = 0; 1108663dfbb0SFilipe Manana if (!err) 11091728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1110777e6bd7SChris Mason if (err) 1111777e6bd7SChris Mason werr = err; 1112e38e2ed7SFilipe Manana free_extent_state(cached_state); 1113e38e2ed7SFilipe Manana cached_state = NULL; 1114777e6bd7SChris Mason cond_resched(); 11151728366eSJosef Bacik start = end + 1; 1116777e6bd7SChris Mason } 11177c4452b9SChris Mason if (err) 11187c4452b9SChris Mason werr = err; 1119bf89d38fSJeff Mahoney return werr; 1120bf89d38fSJeff Mahoney } 1121656f30dbSFilipe Manana 1122b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1123bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1124bf89d38fSJeff Mahoney { 1125bf89d38fSJeff Mahoney bool errors = false; 1126bf89d38fSJeff Mahoney int err; 1127bf89d38fSJeff Mahoney 1128bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1129bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1130bf89d38fSJeff Mahoney errors = true; 1131bf89d38fSJeff Mahoney 1132bf89d38fSJeff Mahoney if (errors && !err) 1133bf89d38fSJeff Mahoney err = -EIO; 1134bf89d38fSJeff Mahoney return err; 1135bf89d38fSJeff Mahoney } 1136bf89d38fSJeff Mahoney 1137bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1138bf89d38fSJeff Mahoney { 1139bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1140bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1141bf89d38fSJeff Mahoney bool errors = false; 1142bf89d38fSJeff Mahoney int err; 1143bf89d38fSJeff Mahoney 1144bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1145bf89d38fSJeff Mahoney 1146bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1147656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11480b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1149656f30dbSFilipe Manana errors = true; 1150656f30dbSFilipe Manana 1151656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11520b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1153656f30dbSFilipe Manana errors = true; 1154656f30dbSFilipe Manana 1155bf89d38fSJeff Mahoney if (errors && !err) 1156bf89d38fSJeff Mahoney err = -EIO; 1157bf89d38fSJeff Mahoney return err; 115879154b1bSChris Mason } 115979154b1bSChris Mason 1160690587d1SChris Mason /* 1161c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1162c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1163c9b577c0SNikolay Borisov * log commit. 1164c9b577c0SNikolay Borisov * 1165c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1166690587d1SChris Mason */ 116770458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1168d0c803c4SChris Mason { 1169663dfbb0SFilipe Manana int ret; 1170c9b577c0SNikolay Borisov int ret2; 1171c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 117270458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1173c9b577c0SNikolay Borisov struct blk_plug plug; 1174663dfbb0SFilipe Manana 1175c9b577c0SNikolay Borisov blk_start_plug(&plug); 1176c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1177c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1178c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1179c9b577c0SNikolay Borisov 118041e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1181663dfbb0SFilipe Manana 1182c9b577c0SNikolay Borisov if (ret) 1183663dfbb0SFilipe Manana return ret; 1184c9b577c0SNikolay Borisov else if (ret2) 1185c9b577c0SNikolay Borisov return ret2; 1186c9b577c0SNikolay Borisov else 1187c9b577c0SNikolay Borisov return 0; 1188d0c803c4SChris Mason } 1189d0c803c4SChris Mason 1190d352ac68SChris Mason /* 1191d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1192d352ac68SChris Mason * 1193d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1194d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1195d352ac68SChris Mason * allocation tree. 1196d352ac68SChris Mason * 1197d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1198d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1199d352ac68SChris Mason */ 12000b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 120179154b1bSChris Mason struct btrfs_root *root) 120279154b1bSChris Mason { 120379154b1bSChris Mason int ret; 12040b86a832SChris Mason u64 old_root_bytenr; 120586b9f2ecSYan, Zheng u64 old_root_used; 12060b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12070b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 120879154b1bSChris Mason 120986b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 121056bec294SChris Mason 121179154b1bSChris Mason while (1) { 12120b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 121386b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1214ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 121579154b1bSChris Mason break; 121687ef2bb4SChris Mason 12175d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 121879154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12190b86a832SChris Mason &root->root_key, 12200b86a832SChris Mason &root->root_item); 122149b25e05SJeff Mahoney if (ret) 122249b25e05SJeff Mahoney return ret; 122356bec294SChris Mason 122486b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1225e7070be1SJosef Bacik } 1226276e680dSYan Zheng 12270b86a832SChris Mason return 0; 12280b86a832SChris Mason } 12290b86a832SChris Mason 1230d352ac68SChris Mason /* 1231d352ac68SChris Mason * update all the cowonly tree roots on disk 123249b25e05SJeff Mahoney * 123349b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 123449b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 123549b25e05SJeff Mahoney * to clean up the delayed refs. 1236d352ac68SChris Mason */ 12379386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12380b86a832SChris Mason { 12399386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1240ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12411bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12420b86a832SChris Mason struct list_head *next; 124384234f3aSYan Zheng struct extent_buffer *eb; 124456bec294SChris Mason int ret; 124584234f3aSYan Zheng 124684234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 124749b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12489631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 124984234f3aSYan Zheng btrfs_tree_unlock(eb); 125084234f3aSYan Zheng free_extent_buffer(eb); 12510b86a832SChris Mason 125249b25e05SJeff Mahoney if (ret) 125349b25e05SJeff Mahoney return ret; 125449b25e05SJeff Mahoney 1255196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1256c16ce190SJosef Bacik if (ret) 1257c16ce190SJosef Bacik return ret; 12582b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1259c16ce190SJosef Bacik if (ret) 1260c16ce190SJosef Bacik return ret; 1261280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1262c16ce190SJosef Bacik if (ret) 1263c16ce190SJosef Bacik return ret; 1264546adb0dSJan Schmidt 1265bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1266dcdf7f6dSJosef Bacik if (ret) 1267dcdf7f6dSJosef Bacik return ret; 1268dcdf7f6dSJosef Bacik 1269ea526d18SJosef Bacik again: 12700b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12712ff7e61eSJeff Mahoney struct btrfs_root *root; 12720b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12730b86a832SChris Mason list_del_init(next); 12740b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1275e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 127687ef2bb4SChris Mason 12779e351cc8SJosef Bacik if (root != fs_info->extent_root) 12789e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 12799e351cc8SJosef Bacik &trans->transaction->switch_commits); 128049b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 128149b25e05SJeff Mahoney if (ret) 128249b25e05SJeff Mahoney return ret; 1283488bc2a2SJosef Bacik } 1284488bc2a2SJosef Bacik 1285488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1286c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1287ea526d18SJosef Bacik if (ret) 1288ea526d18SJosef Bacik return ret; 1289276e680dSYan Zheng 12901bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 12915742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1292ea526d18SJosef Bacik if (ret) 1293ea526d18SJosef Bacik return ret; 1294488bc2a2SJosef Bacik 1295488bc2a2SJosef Bacik /* 1296488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1297488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1298488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1299488bc2a2SJosef Bacik * everything gets run. 1300488bc2a2SJosef Bacik */ 1301c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1302ea526d18SJosef Bacik if (ret) 1303ea526d18SJosef Bacik return ret; 1304ea526d18SJosef Bacik } 1305ea526d18SJosef Bacik 1306ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1307ea526d18SJosef Bacik goto again; 1308ea526d18SJosef Bacik 13099e351cc8SJosef Bacik list_add_tail(&fs_info->extent_root->dirty_list, 13109e351cc8SJosef Bacik &trans->transaction->switch_commits); 13119f6cbcbbSDavid Sterba 13129f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13139f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13149f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13158dabb742SStefan Behrens 131679154b1bSChris Mason return 0; 131779154b1bSChris Mason } 131879154b1bSChris Mason 1319d352ac68SChris Mason /* 1320d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1321d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1322d352ac68SChris Mason * be deleted 1323d352ac68SChris Mason */ 1324cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13255eda7b5eSChris Mason { 13260b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13270b246afaSJeff Mahoney 13280b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1329dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1330dc9492c1SJosef Bacik btrfs_grab_root(root); 13310b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1332dc9492c1SJosef Bacik } 13330b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13345eda7b5eSChris Mason } 13355eda7b5eSChris Mason 1336d352ac68SChris Mason /* 13375d4f98a2SYan Zheng * update all the cowonly tree roots on disk 1338d352ac68SChris Mason */ 13397e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13400f7d52f4SChris Mason { 13417e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 13420f7d52f4SChris Mason struct btrfs_root *gang[8]; 13430f7d52f4SChris Mason int i; 13440f7d52f4SChris Mason int ret; 134554aa1f4dSChris Mason 1346a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 13470f7d52f4SChris Mason while (1) { 13485d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 13495d4f98a2SYan Zheng (void **)gang, 0, 13500f7d52f4SChris Mason ARRAY_SIZE(gang), 13510f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 13520f7d52f4SChris Mason if (ret == 0) 13530f7d52f4SChris Mason break; 13540f7d52f4SChris Mason for (i = 0; i < ret; i++) { 13555b4aacefSJeff Mahoney struct btrfs_root *root = gang[i]; 13564f4317c1SJosef Bacik int ret2; 13574f4317c1SJosef Bacik 13585d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 13592619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13600f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1361a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 136231153d81SYan Zheng 1363e02119d5SChris Mason btrfs_free_log(trans, root); 13645d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 1365e02119d5SChris Mason 1366f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 136727cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1368c7548af6SChris Mason smp_mb__after_atomic(); 1369f1ebcc74SLiu Bo 1370978d910dSYan Zheng if (root->commit_root != root->node) { 13719e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13729e351cc8SJosef Bacik &trans->transaction->switch_commits); 1373978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 1374978d910dSYan Zheng root->node); 1375978d910dSYan Zheng } 137631153d81SYan Zheng 13774f4317c1SJosef Bacik ret2 = btrfs_update_root(trans, fs_info->tree_root, 13780f7d52f4SChris Mason &root->root_key, 13790f7d52f4SChris Mason &root->root_item); 13804f4317c1SJosef Bacik if (ret2) 13814f4317c1SJosef Bacik return ret2; 1382a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1383733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 13840f7d52f4SChris Mason } 13859f3a7427SChris Mason } 1386a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 13874f4317c1SJosef Bacik return 0; 13880f7d52f4SChris Mason } 13890f7d52f4SChris Mason 1390d352ac68SChris Mason /* 1391de78b51aSEric Sandeen * defrag a given btree. 1392de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1393d352ac68SChris Mason */ 1394de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1395e9d0b13bSChris Mason { 1396e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1397e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 13988929ecfaSYan, Zheng int ret; 1399e9d0b13bSChris Mason 140027cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1401e9d0b13bSChris Mason return 0; 14028929ecfaSYan, Zheng 14036b80053dSChris Mason while (1) { 14048929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14058929ecfaSYan, Zheng if (IS_ERR(trans)) 14068929ecfaSYan, Zheng return PTR_ERR(trans); 14078929ecfaSYan, Zheng 1408de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14098929ecfaSYan, Zheng 14103a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14112ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1412e9d0b13bSChris Mason cond_resched(); 1413e9d0b13bSChris Mason 1414ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1415e9d0b13bSChris Mason break; 1416210549ebSDavid Sterba 1417ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1418ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1419210549ebSDavid Sterba ret = -EAGAIN; 1420210549ebSDavid Sterba break; 1421210549ebSDavid Sterba } 1422e9d0b13bSChris Mason } 142327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14248929ecfaSYan, Zheng return ret; 1425e9d0b13bSChris Mason } 1426e9d0b13bSChris Mason 1427d352ac68SChris Mason /* 14286426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14296426c7adSQu Wenruo * 14306426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14316426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 14326426c7adSQu Wenruo * qgroup works. 14336426c7adSQu Wenruo */ 14346426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 14356426c7adSQu Wenruo struct btrfs_root *src, 14366426c7adSQu Wenruo struct btrfs_root *parent, 14376426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 14386426c7adSQu Wenruo u64 dst_objectid) 14396426c7adSQu Wenruo { 14406426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 14416426c7adSQu Wenruo int ret; 14426426c7adSQu Wenruo 14436426c7adSQu Wenruo /* 14446426c7adSQu Wenruo * Save some performance in the case that qgroups are not 14456426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 14466426c7adSQu Wenruo * kick in anyway. 14476426c7adSQu Wenruo */ 14489ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 14496426c7adSQu Wenruo return 0; 14506426c7adSQu Wenruo 14516426c7adSQu Wenruo /* 145252042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 14534d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 14544d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 14554d31778aSQu Wenruo * item. 14564d31778aSQu Wenruo */ 1457*1c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 1458*1c442d22SJosef Bacik if (ret) 1459*1c442d22SJosef Bacik return ret; 14604d31778aSQu Wenruo 14614d31778aSQu Wenruo /* 14622a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 14632a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 14642a4d84c1SJosef Bacik * 14652a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 14662a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 14672a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 14682a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 14692a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 14702a4d84c1SJosef Bacik * qgroup counters could end up wrong. 14712a4d84c1SJosef Bacik */ 14722a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 14732a4d84c1SJosef Bacik if (ret) { 14742a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 14752a4d84c1SJosef Bacik goto out; 14762a4d84c1SJosef Bacik } 14772a4d84c1SJosef Bacik 14782a4d84c1SJosef Bacik /* 14796426c7adSQu Wenruo * We are going to commit transaction, see btrfs_commit_transaction() 14806426c7adSQu Wenruo * comment for reason locking tree_log_mutex 14816426c7adSQu Wenruo */ 14826426c7adSQu Wenruo mutex_lock(&fs_info->tree_log_mutex); 14836426c7adSQu Wenruo 14847e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 14856426c7adSQu Wenruo if (ret) 14866426c7adSQu Wenruo goto out; 1487460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 14886426c7adSQu Wenruo if (ret < 0) 14896426c7adSQu Wenruo goto out; 14906426c7adSQu Wenruo 14916426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1492a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 14936426c7adSQu Wenruo inherit); 14946426c7adSQu Wenruo if (ret < 0) 14956426c7adSQu Wenruo goto out; 14966426c7adSQu Wenruo 14976426c7adSQu Wenruo /* 14986426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 14996426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15006426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15016426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15026426c7adSQu Wenruo * 2) write all btree blocks onto disk 15036426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15046426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15056426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15066426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15076426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15086426c7adSQu Wenruo */ 15099386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15106426c7adSQu Wenruo if (ret) 15116426c7adSQu Wenruo goto out; 1512889bfa39SJosef Bacik switch_commit_roots(trans); 151370458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15146426c7adSQu Wenruo if (ret) 1515f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15166426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15176426c7adSQu Wenruo 15186426c7adSQu Wenruo out: 15196426c7adSQu Wenruo mutex_unlock(&fs_info->tree_log_mutex); 15206426c7adSQu Wenruo 15216426c7adSQu Wenruo /* 15226426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15236426c7adSQu Wenruo * last_trans == cur_transid. 15246426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15256426c7adSQu Wenruo * insert_dir_item() 15266426c7adSQu Wenruo */ 15276426c7adSQu Wenruo if (!ret) 1528*1c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15296426c7adSQu Wenruo return ret; 15306426c7adSQu Wenruo } 15316426c7adSQu Wenruo 15326426c7adSQu Wenruo /* 1533d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1534aec8030aSMiao Xie * transaction commit. This does the actual creation. 1535aec8030aSMiao Xie * 1536aec8030aSMiao Xie * Note: 1537aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1538aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1539aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1540d352ac68SChris Mason */ 154180b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 15423063d29fSChris Mason struct btrfs_pending_snapshot *pending) 15433063d29fSChris Mason { 154408d50ca3SNikolay Borisov 154508d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 15463063d29fSChris Mason struct btrfs_key key; 154780b6794dSChris Mason struct btrfs_root_item *new_root_item; 15483063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 15493063d29fSChris Mason struct btrfs_root *root = pending->root; 15506bdb72deSSage Weil struct btrfs_root *parent_root; 155198c9942aSLiu Bo struct btrfs_block_rsv *rsv; 15526bdb72deSSage Weil struct inode *parent_inode; 155342874b3dSMiao Xie struct btrfs_path *path; 155442874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1555a22285a6SYan, Zheng struct dentry *dentry; 15563063d29fSChris Mason struct extent_buffer *tmp; 1557925baeddSChris Mason struct extent_buffer *old; 155895582b00SDeepa Dinamani struct timespec64 cur_time; 1559aec8030aSMiao Xie int ret = 0; 1560d68fc57bSYan, Zheng u64 to_reserve = 0; 15616bdb72deSSage Weil u64 index = 0; 1562a22285a6SYan, Zheng u64 objectid; 1563b83cc969SLi Zefan u64 root_flags; 15643063d29fSChris Mason 15658546b570SDavid Sterba ASSERT(pending->path); 15668546b570SDavid Sterba path = pending->path; 156742874b3dSMiao Xie 1568b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1569b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1570a22285a6SYan, Zheng 1571543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1572aec8030aSMiao Xie if (pending->error) 15736fa9700eSMiao Xie goto no_free_objectid; 15743063d29fSChris Mason 1575d6726335SQu Wenruo /* 1576d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1577d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1578d6726335SQu Wenruo */ 1579d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1580d6726335SQu Wenruo 1581147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1582d68fc57bSYan, Zheng 1583d68fc57bSYan, Zheng if (to_reserve > 0) { 1584aec8030aSMiao Xie pending->error = btrfs_block_rsv_add(root, 1585aec8030aSMiao Xie &pending->block_rsv, 158608e007d2SMiao Xie to_reserve, 158708e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1588aec8030aSMiao Xie if (pending->error) 1589d6726335SQu Wenruo goto clear_skip_qgroup; 1590d68fc57bSYan, Zheng } 1591d68fc57bSYan, Zheng 15923063d29fSChris Mason key.objectid = objectid; 1593a22285a6SYan, Zheng key.offset = (u64)-1; 1594a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 15953063d29fSChris Mason 15966fa9700eSMiao Xie rsv = trans->block_rsv; 1597a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 15982382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 15990b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 160088d3a5aaSJosef Bacik trans->transid, 160188d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1602a22285a6SYan, Zheng dentry = pending->dentry; 1603e9662f70SMiao Xie parent_inode = pending->dir; 1604a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 16056426c7adSQu Wenruo record_root_in_trans(trans, parent_root, 0); 1606a22285a6SYan, Zheng 1607c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 160804b285f3SDeepa Dinamani 16096bdb72deSSage Weil /* 16106bdb72deSSage Weil * insert the directory item 16116bdb72deSSage Weil */ 1612877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 161349b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 161442874b3dSMiao Xie 161542874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 161642874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16174a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 161842874b3dSMiao Xie dentry->d_name.name, 161942874b3dSMiao Xie dentry->d_name.len, 0); 162042874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1621fe66a05aSChris Mason pending->error = -EEXIST; 1622aec8030aSMiao Xie goto dir_item_existed; 162342874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 162442874b3dSMiao Xie ret = PTR_ERR(dir_item); 162566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16268732d44fSMiao Xie goto fail; 162779787eaaSJeff Mahoney } 162842874b3dSMiao Xie btrfs_release_path(path); 16296bdb72deSSage Weil 1630e999376fSChris Mason /* 1631e999376fSChris Mason * pull in the delayed directory update 1632e999376fSChris Mason * and the delayed inode item 1633e999376fSChris Mason * otherwise we corrupt the FS during 1634e999376fSChris Mason * snapshot 1635e999376fSChris Mason */ 1636e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 16378732d44fSMiao Xie if (ret) { /* Transaction aborted */ 163866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16398732d44fSMiao Xie goto fail; 16408732d44fSMiao Xie } 1641e999376fSChris Mason 16426426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 16436bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 16446bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 164508fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 16466bdb72deSSage Weil 1647b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1648b83cc969SLi Zefan if (pending->readonly) 1649b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1650b83cc969SLi Zefan else 1651b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1652b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1653b83cc969SLi Zefan 16548ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 16558ea05e3aSAlexander Block trans->transid); 1656807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 16578ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 16588ea05e3aSAlexander Block BTRFS_UUID_SIZE); 165970023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 166070023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 166170023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 16628ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 16638ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 16648ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 16658ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 166670023da2SStefan Behrens } 16673cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 16683cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 166970023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 16708ea05e3aSAlexander Block 1671925baeddSChris Mason old = btrfs_lock_root_node(root); 16729631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 16739631e4ccSJosef Bacik BTRFS_NESTING_COW); 167479787eaaSJeff Mahoney if (ret) { 167579787eaaSJeff Mahoney btrfs_tree_unlock(old); 167679787eaaSJeff Mahoney free_extent_buffer(old); 167766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16788732d44fSMiao Xie goto fail; 167979787eaaSJeff Mahoney } 168049b25e05SJeff Mahoney 168149b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 168279787eaaSJeff Mahoney /* clean up in any case */ 1683925baeddSChris Mason btrfs_tree_unlock(old); 1684925baeddSChris Mason free_extent_buffer(old); 16858732d44fSMiao Xie if (ret) { 168666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16878732d44fSMiao Xie goto fail; 16888732d44fSMiao Xie } 1689f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 169027cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1691f1ebcc74SLiu Bo smp_wmb(); 1692f1ebcc74SLiu Bo 16935d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1694a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1695a22285a6SYan, Zheng key.offset = trans->transid; 1696a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1697925baeddSChris Mason btrfs_tree_unlock(tmp); 16983063d29fSChris Mason free_extent_buffer(tmp); 16998732d44fSMiao Xie if (ret) { 170066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17018732d44fSMiao Xie goto fail; 17028732d44fSMiao Xie } 17030660b5afSChris Mason 1704a22285a6SYan, Zheng /* 1705a22285a6SYan, Zheng * insert root back/forward references 1706a22285a6SYan, Zheng */ 17076025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1708a22285a6SYan, Zheng parent_root->root_key.objectid, 17094a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1710a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 17118732d44fSMiao Xie if (ret) { 171266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17138732d44fSMiao Xie goto fail; 17148732d44fSMiao Xie } 1715a22285a6SYan, Zheng 1716a22285a6SYan, Zheng key.offset = (u64)-1; 17172dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 171879787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 171979787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17202d892ccdSFilipe Manana pending->snap = NULL; 172166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17228732d44fSMiao Xie goto fail; 172379787eaaSJeff Mahoney } 1724d68fc57bSYan, Zheng 172549b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17268732d44fSMiao Xie if (ret) { 172766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17288732d44fSMiao Xie goto fail; 17298732d44fSMiao Xie } 1730361048f5SMiao Xie 17316426c7adSQu Wenruo /* 17326426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17336426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17346426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 17356426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 17366426c7adSQu Wenruo */ 17376426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 17386426c7adSQu Wenruo pending->inherit, objectid); 17396426c7adSQu Wenruo if (ret < 0) 17406426c7adSQu Wenruo goto fail; 17416426c7adSQu Wenruo 1742684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1743684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1744684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 174542874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 17469c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 17478732d44fSMiao Xie if (ret) { 174866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17498732d44fSMiao Xie goto fail; 17508732d44fSMiao Xie } 175142874b3dSMiao Xie 17526ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 175342874b3dSMiao Xie dentry->d_name.len * 2); 175404b285f3SDeepa Dinamani parent_inode->i_mtime = parent_inode->i_ctime = 1755c2050a45SDeepa Dinamani current_time(parent_inode); 1756729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1757dd5f9615SStefan Behrens if (ret) { 175866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1759dd5f9615SStefan Behrens goto fail; 1760dd5f9615SStefan Behrens } 1761807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1762807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1763cdb345a8SLu Fengqi objectid); 1764dd5f9615SStefan Behrens if (ret) { 176566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1766dd5f9615SStefan Behrens goto fail; 1767dd5f9615SStefan Behrens } 1768dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1769cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1770dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1771dd5f9615SStefan Behrens objectid); 1772dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 177366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1774dd5f9615SStefan Behrens goto fail; 1775dd5f9615SStefan Behrens } 1776dd5f9615SStefan Behrens } 1777d6726335SQu Wenruo 17783063d29fSChris Mason fail: 1779aec8030aSMiao Xie pending->error = ret; 1780aec8030aSMiao Xie dir_item_existed: 178198c9942aSLiu Bo trans->block_rsv = rsv; 17822382c5ccSLiu Bo trans->bytes_reserved = 0; 1783d6726335SQu Wenruo clear_skip_qgroup: 1784d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 17856fa9700eSMiao Xie no_free_objectid: 17866fa9700eSMiao Xie kfree(new_root_item); 1787b0c0ea63SDavid Sterba pending->root_item = NULL; 178842874b3dSMiao Xie btrfs_free_path(path); 17898546b570SDavid Sterba pending->path = NULL; 17908546b570SDavid Sterba 179149b25e05SJeff Mahoney return ret; 17923063d29fSChris Mason } 17933063d29fSChris Mason 1794d352ac68SChris Mason /* 1795d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1796d352ac68SChris Mason */ 179708d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 17983063d29fSChris Mason { 1799aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18003063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1801aec8030aSMiao Xie int ret = 0; 18023de4586cSChris Mason 1803aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1804aec8030aSMiao Xie list_del(&pending->list); 180508d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1806aec8030aSMiao Xie if (ret) 1807aec8030aSMiao Xie break; 1808aec8030aSMiao Xie } 1809aec8030aSMiao Xie return ret; 18103de4586cSChris Mason } 18113de4586cSChris Mason 18122ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18135d4f98a2SYan Zheng { 18145d4f98a2SYan Zheng struct btrfs_root_item *root_item; 18155d4f98a2SYan Zheng struct btrfs_super_block *super; 18165d4f98a2SYan Zheng 18170b246afaSJeff Mahoney super = fs_info->super_copy; 18185d4f98a2SYan Zheng 18190b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1820093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1821093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1822093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18235d4f98a2SYan Zheng 18240b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1825093e037cSDavid Sterba super->root = root_item->bytenr; 1826093e037cSDavid Sterba super->generation = root_item->generation; 1827093e037cSDavid Sterba super->root_level = root_item->level; 18280b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1829093e037cSDavid Sterba super->cache_generation = root_item->generation; 183094846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 183194846229SBoris Burkov super->cache_generation = 0; 18320b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1833093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18345d4f98a2SYan Zheng } 18355d4f98a2SYan Zheng 1836f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1837f36f3042SChris Mason { 18384a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1839f36f3042SChris Mason int ret = 0; 18404a9d8bdeSMiao Xie 1841a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18424a9d8bdeSMiao Xie trans = info->running_transaction; 18434a9d8bdeSMiao Xie if (trans) 18444a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1845a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1846f36f3042SChris Mason return ret; 1847f36f3042SChris Mason } 1848f36f3042SChris Mason 18498929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 18508929ecfaSYan, Zheng { 18514a9d8bdeSMiao Xie struct btrfs_transaction *trans; 18528929ecfaSYan, Zheng int ret = 0; 18534a9d8bdeSMiao Xie 1854a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18554a9d8bdeSMiao Xie trans = info->running_transaction; 18564a9d8bdeSMiao Xie if (trans) 18574a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1858a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 18598929ecfaSYan, Zheng return ret; 18608929ecfaSYan, Zheng } 18618929ecfaSYan, Zheng 1862bb9c12c9SSage Weil /* 1863bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1864bb9c12c9SSage Weil * transaction joins 1865bb9c12c9SSage Weil */ 18662ff7e61eSJeff Mahoney static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, 1867bb9c12c9SSage Weil struct btrfs_transaction *trans) 1868bb9c12c9SSage Weil { 18692ff7e61eSJeff Mahoney wait_event(fs_info->transaction_blocked_wait, 1870bf31f87fSDavid Sterba trans->state >= TRANS_STATE_COMMIT_START || 1871bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1872bb9c12c9SSage Weil } 1873bb9c12c9SSage Weil 1874bb9c12c9SSage Weil /* 1875bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1876bb9c12c9SSage Weil * caller holds ref. 1877bb9c12c9SSage Weil */ 18782ff7e61eSJeff Mahoney static void wait_current_trans_commit_start_and_unblock( 18792ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 1880bb9c12c9SSage Weil struct btrfs_transaction *trans) 1881bb9c12c9SSage Weil { 18822ff7e61eSJeff Mahoney wait_event(fs_info->transaction_wait, 1883bf31f87fSDavid Sterba trans->state >= TRANS_STATE_UNBLOCKED || 1884bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1885bb9c12c9SSage Weil } 1886bb9c12c9SSage Weil 1887bb9c12c9SSage Weil /* 1888bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1889bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1890bb9c12c9SSage Weil */ 1891bb9c12c9SSage Weil struct btrfs_async_commit { 1892bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 18937892b5afSMiao Xie struct work_struct work; 1894bb9c12c9SSage Weil }; 1895bb9c12c9SSage Weil 1896bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1897bb9c12c9SSage Weil { 1898bb9c12c9SSage Weil struct btrfs_async_commit *ac = 18997892b5afSMiao Xie container_of(work, struct btrfs_async_commit, work); 1900bb9c12c9SSage Weil 19016fc4e354SSage Weil /* 19026fc4e354SSage Weil * We've got freeze protection passed with the transaction. 19036fc4e354SSage Weil * Tell lockdep about it. 19046fc4e354SSage Weil */ 1905b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 19063a45bb20SJeff Mahoney __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); 19076fc4e354SSage Weil 1908e209db7aSSage Weil current->journal_info = ac->newtrans; 1909e209db7aSSage Weil 19103a45bb20SJeff Mahoney btrfs_commit_transaction(ac->newtrans); 1911bb9c12c9SSage Weil kfree(ac); 1912bb9c12c9SSage Weil } 1913bb9c12c9SSage Weil 1914bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1915bb9c12c9SSage Weil int wait_for_unblock) 1916bb9c12c9SSage Weil { 19173a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1918bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1919bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1920bb9c12c9SSage Weil 1921bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1922db5b493aSTsutomu Itoh if (!ac) 1923db5b493aSTsutomu Itoh return -ENOMEM; 1924bb9c12c9SSage Weil 19257892b5afSMiao Xie INIT_WORK(&ac->work, do_async_commit); 19263a45bb20SJeff Mahoney ac->newtrans = btrfs_join_transaction(trans->root); 19273612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 19283612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 19293612b495STsutomu Itoh kfree(ac); 19303612b495STsutomu Itoh return err; 19313612b495STsutomu Itoh } 1932bb9c12c9SSage Weil 1933bb9c12c9SSage Weil /* take transaction reference */ 1934bb9c12c9SSage Weil cur_trans = trans->transaction; 19359b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1936bb9c12c9SSage Weil 19373a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19386fc4e354SSage Weil 19396fc4e354SSage Weil /* 19406fc4e354SSage Weil * Tell lockdep we've released the freeze rwsem, since the 19416fc4e354SSage Weil * async commit thread will be the one to unlock it. 19426fc4e354SSage Weil */ 1943b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 19440b246afaSJeff Mahoney __sb_writers_release(fs_info->sb, SB_FREEZE_FS); 19456fc4e354SSage Weil 19467892b5afSMiao Xie schedule_work(&ac->work); 1947bb9c12c9SSage Weil 1948bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1949bb9c12c9SSage Weil if (wait_for_unblock) 19502ff7e61eSJeff Mahoney wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); 1951bb9c12c9SSage Weil else 19522ff7e61eSJeff Mahoney wait_current_trans_commit_start(fs_info, cur_trans); 1953bb9c12c9SSage Weil 195438e88054SSage Weil if (current->journal_info == trans) 195538e88054SSage Weil current->journal_info = NULL; 195638e88054SSage Weil 1957724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1958bb9c12c9SSage Weil return 0; 1959bb9c12c9SSage Weil } 1960bb9c12c9SSage Weil 196149b25e05SJeff Mahoney 196297cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 196349b25e05SJeff Mahoney { 196497cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 196549b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 196649b25e05SJeff Mahoney 1967b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 196849b25e05SJeff Mahoney 196966642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19707b8b92afSJosef Bacik 19710b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 197266b6135bSLiu Bo 197325d8c284SMiao Xie /* 197425d8c284SMiao Xie * If the transaction is removed from the list, it means this 197525d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 197625d8c284SMiao Xie * to call the cleanup function. 197725d8c284SMiao Xie */ 197825d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 197966b6135bSLiu Bo 198049b25e05SJeff Mahoney list_del_init(&cur_trans->list); 19810b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19824a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19830b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1984f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1985f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1986f094ac32SLiu Bo 19870b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1988d7096fc3SJosef Bacik } 19890b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 199049b25e05SJeff Mahoney 19912ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 199249b25e05SJeff Mahoney 19930b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 19940b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 19950b246afaSJeff Mahoney fs_info->running_transaction = NULL; 19960b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 19974a9d8bdeSMiao Xie 1998e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 19990b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2000724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2001724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 200249b25e05SJeff Mahoney 200397cb39bbSNikolay Borisov trace_btrfs_transaction_commit(trans->root); 200449b25e05SJeff Mahoney 200549b25e05SJeff Mahoney if (current->journal_info == trans) 200649b25e05SJeff Mahoney current->journal_info = NULL; 20070b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 200849b25e05SJeff Mahoney 200949b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 201049b25e05SJeff Mahoney } 201149b25e05SJeff Mahoney 2012c7cc64a9SDavid Sterba /* 2013c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2014c7cc64a9SDavid Sterba * transaction and remove them from the list 2015c7cc64a9SDavid Sterba */ 2016c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2017c7cc64a9SDavid Sterba { 2018c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 201932da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2020c7cc64a9SDavid Sterba 2021c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2022c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2023c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2024c7cc64a9SDavid Sterba } 2025c7cc64a9SDavid Sterba } 2026c7cc64a9SDavid Sterba 202788090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 202882436617SMiao Xie { 2029ce8ea7ccSJosef Bacik /* 2030ce8ea7ccSJosef Bacik * We use writeback_inodes_sb here because if we used 2031ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2032ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2033ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2034ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2035ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2036ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2037ce8ea7ccSJosef Bacik */ 203888090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2039ce8ea7ccSJosef Bacik writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 204082436617SMiao Xie return 0; 204182436617SMiao Xie } 204282436617SMiao Xie 204388090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 204482436617SMiao Xie { 204588090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20466374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 204782436617SMiao Xie } 204882436617SMiao Xie 20493a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 205079154b1bSChris Mason { 20513a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 205249b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20538fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 205425287e0aSMiao Xie int ret; 205579154b1bSChris Mason 205635b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 205735b814f3SNikolay Borisov 2058d62b23c9SJosef Bacik /* 2059d62b23c9SJosef Bacik * Some places just start a transaction to commit it. We need to make 2060d62b23c9SJosef Bacik * sure that if this commit fails that the abort code actually marks the 2061d62b23c9SJosef Bacik * transaction as failed, so set trans->dirty to make the abort code do 2062d62b23c9SJosef Bacik * the right thing. 2063d62b23c9SJosef Bacik */ 2064d62b23c9SJosef Bacik trans->dirty = true; 2065d62b23c9SJosef Bacik 20668d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2067bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 206825287e0aSMiao Xie ret = cur_trans->aborted; 20693a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2070e4a2bcacSJosef Bacik return ret; 207125287e0aSMiao Xie } 207249b25e05SJeff Mahoney 2073f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2074f45c752bSJosef Bacik trans->block_rsv = NULL; 2075f45c752bSJosef Bacik 2076e19eb11fSJosef Bacik /* 2077e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2078e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2079e19eb11fSJosef Bacik */ 2080e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2081e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2082e19eb11fSJosef Bacik /* 2083e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2084e19eb11fSJosef Bacik * Any running threads may add more while we are here. 208556bec294SChris Mason */ 2086c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 2087e4a2bcacSJosef Bacik if (ret) { 20883a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2089e4a2bcacSJosef Bacik return ret; 2090e4a2bcacSJosef Bacik } 2091e19eb11fSJosef Bacik } 209256bec294SChris Mason 20936c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2094ea658badSJosef Bacik 20953204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 20961bbc621eSChris Mason int run_it = 0; 20971bbc621eSChris Mason 20981bbc621eSChris Mason /* this mutex is also taken before trying to set 20991bbc621eSChris Mason * block groups readonly. We need to make sure 21001bbc621eSChris Mason * that nobody has set a block group readonly 21011bbc621eSChris Mason * after a extents from that block group have been 21021bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 21031bbc621eSChris Mason * will wait for the transaction to commit if it 21043204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 21051bbc621eSChris Mason * 21063204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 21073204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 21081bbc621eSChris Mason * hurt to have more than one go through, but there's no 21091bbc621eSChris Mason * real advantage to it either. 21101bbc621eSChris Mason */ 21110b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 21123204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 21133204d33cSJosef Bacik &cur_trans->flags)) 21141bbc621eSChris Mason run_it = 1; 21150b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21161bbc621eSChris Mason 2117f9cacae3SNikolay Borisov if (run_it) { 211821217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21191bbc621eSChris Mason if (ret) { 21203a45bb20SJeff Mahoney btrfs_end_transaction(trans); 21211bbc621eSChris Mason return ret; 21221bbc621eSChris Mason } 2123f9cacae3SNikolay Borisov } 2124f9cacae3SNikolay Borisov } 21251bbc621eSChris Mason 21260b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21274a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2128d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2129d0c2f4faSFilipe Manana 21300b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21319b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2132ccd467d6SChris Mason 2133d0c2f4faSFilipe Manana if (trans->in_fsync) 2134d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2135d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2136d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 213715ee9bc7SJosef Bacik 2138bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2139b4924a0fSLiu Bo ret = cur_trans->aborted; 2140b4924a0fSLiu Bo 2141724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 214215ee9bc7SJosef Bacik 214349b25e05SJeff Mahoney return ret; 214479154b1bSChris Mason } 21454313b399SChris Mason 21464a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21470b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 2148bb9c12c9SSage Weil 21490b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2150d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2151d0c2f4faSFilipe Manana 2152d0c2f4faSFilipe Manana if (trans->in_fsync) 2153d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2154d0c2f4faSFilipe Manana 2155ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2156ccd467d6SChris Mason struct btrfs_transaction, list); 2157d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 21589b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 21590b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2160ccd467d6SChris Mason 2161d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2162d0c2f4faSFilipe Manana 2163bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2164ccd467d6SChris Mason 2165724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 21661f9b8c8fSFilipe Manana if (ret) 21671f9b8c8fSFilipe Manana goto cleanup_transaction; 2168a4abeea4SJosef Bacik } else { 21690b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2170ccd467d6SChris Mason } 2171a4abeea4SJosef Bacik } else { 21720b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2173cb2d3dadSFilipe Manana /* 2174cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2175cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2176cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2177cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2178cb2d3dadSFilipe Manana */ 2179cb2d3dadSFilipe Manana if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { 2180cb2d3dadSFilipe Manana ret = -EROFS; 2181cb2d3dadSFilipe Manana goto cleanup_transaction; 2182cb2d3dadSFilipe Manana } 2183ccd467d6SChris Mason } 218415ee9bc7SJosef Bacik 21850860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 21860860adfdSMiao Xie 218788090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 218882436617SMiao Xie if (ret) 218982436617SMiao Xie goto cleanup_transaction; 219082436617SMiao Xie 2191e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 219249b25e05SJeff Mahoney if (ret) 219349b25e05SJeff Mahoney goto cleanup_transaction; 219416cdcec7SMiao Xie 2195581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2196581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2197ed3b3d31SChris Mason 2198581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2199e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2200ca469637SMiao Xie if (ret) 2201ca469637SMiao Xie goto cleanup_transaction; 2202ca469637SMiao Xie 220388090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2204cb7ab021SWang Shilong 220548778179SFilipe Manana /* 220648778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 220748778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 220848778179SFilipe Manana * extents complete we lose logged data after a power failure. 220948778179SFilipe Manana */ 221048778179SFilipe Manana wait_event(cur_trans->pending_wait, 221148778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 221248778179SFilipe Manana 22132ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2214ed0ca140SJosef Bacik /* 2215ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2216ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22174a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2218ed0ca140SJosef Bacik */ 22190b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 22204a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 22210b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2222ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2223ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 222415ee9bc7SJosef Bacik 2225bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22262cba30f1SMiao Xie ret = cur_trans->aborted; 22276cf7f77eSWang Shilong goto scrub_continue; 22282cba30f1SMiao Xie } 22297585717fSChris Mason /* 22307585717fSChris Mason * the reloc mutex makes sure that we stop 22317585717fSChris Mason * the balancing code from coming in and moving 22327585717fSChris Mason * extents around in the middle of the commit 22337585717fSChris Mason */ 22340b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 22357585717fSChris Mason 223642874b3dSMiao Xie /* 223742874b3dSMiao Xie * We needn't worry about the delayed items because we will 223842874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 223942874b3dSMiao Xie * core function of the snapshot creation. 224042874b3dSMiao Xie */ 224108d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 224256e9f6eaSDavid Sterba if (ret) 224356e9f6eaSDavid Sterba goto unlock_reloc; 22443063d29fSChris Mason 224542874b3dSMiao Xie /* 224642874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 224742874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 224842874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 224942874b3dSMiao Xie * them. 225042874b3dSMiao Xie * 225142874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 225242874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 225342874b3dSMiao Xie * the nodes and leaves. 225442874b3dSMiao Xie */ 2255e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 225656e9f6eaSDavid Sterba if (ret) 225756e9f6eaSDavid Sterba goto unlock_reloc; 225816cdcec7SMiao Xie 2259c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 226056e9f6eaSDavid Sterba if (ret) 226156e9f6eaSDavid Sterba goto unlock_reloc; 226256bec294SChris Mason 2263e999376fSChris Mason /* 2264e999376fSChris Mason * make sure none of the code above managed to slip in a 2265e999376fSChris Mason * delayed item 2266e999376fSChris Mason */ 2267ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2268e999376fSChris Mason 22692c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2270dc17ff8fSChris Mason 2271e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 2272e02119d5SChris Mason * various roots consistent with each other. Every pointer 2273e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 2274e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 2275e02119d5SChris Mason * the tree logging code from jumping in and changing any 2276e02119d5SChris Mason * of the trees. 2277e02119d5SChris Mason * 2278e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 2279e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 2280e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 2281e02119d5SChris Mason * from now until after the super is written, we avoid races 2282e02119d5SChris Mason * with the tree-log code. 2283e02119d5SChris Mason */ 22840b246afaSJeff Mahoney mutex_lock(&fs_info->tree_log_mutex); 22851a40e23bSZheng Yan 22867e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 228756e9f6eaSDavid Sterba if (ret) 228856e9f6eaSDavid Sterba goto unlock_tree_log; 228954aa1f4dSChris Mason 22903818aea2SQu Wenruo /* 22917e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 22927e1876acSDavid Sterba * before the next transaction. 22933818aea2SQu Wenruo */ 22940b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 22953818aea2SQu Wenruo 22965d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2297e02119d5SChris Mason * safe to free the root of tree log roots 2298e02119d5SChris Mason */ 22990b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2300e02119d5SChris Mason 23010ed4792aSQu Wenruo /* 23020ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 23030ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 23040ed4792aSQu Wenruo */ 2305460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 230656e9f6eaSDavid Sterba if (ret < 0) 230756e9f6eaSDavid Sterba goto unlock_tree_log; 23080ed4792aSQu Wenruo 23099386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 231056e9f6eaSDavid Sterba if (ret) 231156e9f6eaSDavid Sterba goto unlock_tree_log; 231254aa1f4dSChris Mason 23132cba30f1SMiao Xie /* 23142cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 23152cba30f1SMiao Xie * update ->aborted, check it. 23162cba30f1SMiao Xie */ 2317bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23182cba30f1SMiao Xie ret = cur_trans->aborted; 231956e9f6eaSDavid Sterba goto unlock_tree_log; 23202cba30f1SMiao Xie } 23212cba30f1SMiao Xie 23220b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 23235f39d397SChris Mason 23240b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 23250b246afaSJeff Mahoney fs_info->tree_root->node); 23260b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 23279e351cc8SJosef Bacik &cur_trans->switch_commits); 23285d4f98a2SYan Zheng 23290b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 23300b246afaSJeff Mahoney fs_info->chunk_root->node); 23310b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 23329e351cc8SJosef Bacik &cur_trans->switch_commits); 23339e351cc8SJosef Bacik 2334889bfa39SJosef Bacik switch_commit_roots(trans); 23355d4f98a2SYan Zheng 2336ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 23371bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 23382ff7e61eSJeff Mahoney update_super_roots(fs_info); 2339e02119d5SChris Mason 23400b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 23410b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 23420b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 23430b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2344ccd467d6SChris Mason 2345bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2346935e5cc9SMiao Xie 23470b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 23480b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2349656f30dbSFilipe Manana 23504fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 23514fbcdf66SFilipe Manana 23520b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 23534a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 23540b246afaSJeff Mahoney fs_info->running_transaction = NULL; 23550b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 23560b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2357b7ec40d7SChris Mason 23580b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 2359e6dcd2dcSChris Mason 236070458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 236149b25e05SJeff Mahoney if (ret) { 23620b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 236308748810SDavid Sterba "Error while writing out transaction"); 236456e9f6eaSDavid Sterba /* 236556e9f6eaSDavid Sterba * reloc_mutex has been unlocked, tree_log_mutex is still held 236656e9f6eaSDavid Sterba * but we can't jump to unlock_tree_log causing double unlock 236756e9f6eaSDavid Sterba */ 23680b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 23696cf7f77eSWang Shilong goto scrub_continue; 237049b25e05SJeff Mahoney } 237149b25e05SJeff Mahoney 2372d3575156SNaohiro Aota /* 2373d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2374d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2375d3575156SNaohiro Aota * buffers. 2376d3575156SNaohiro Aota */ 2377d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2378d3575156SNaohiro Aota 2379eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2380e02119d5SChris Mason /* 2381e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2382e02119d5SChris Mason * to go about their business 2383e02119d5SChris Mason */ 23840b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2385c1f32b7cSAnand Jain if (ret) 2386c1f32b7cSAnand Jain goto scrub_continue; 2387e02119d5SChris Mason 2388d0c2f4faSFilipe Manana /* 2389d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2390d0c2f4faSFilipe Manana * which can change it. 2391d0c2f4faSFilipe Manana */ 2392d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2393d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 2394d0c2f4faSFilipe Manana 23955ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 23964313b399SChris Mason 23973204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 23980b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 239913212b54SZhao Lei 24000b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 24014a9d8bdeSMiao Xie /* 24024a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 24034a9d8bdeSMiao Xie * which can change it. 24044a9d8bdeSMiao Xie */ 24054a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 24062c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 24073de4586cSChris Mason 24080b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 240913c5a93eSJosef Bacik list_del_init(&cur_trans->list); 24100b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2411a4abeea4SJosef Bacik 2412724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2413724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 241458176a96SJosef Bacik 24150860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 24160b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2417b2b5ef5cSJan Kara 24183a45bb20SJeff Mahoney trace_btrfs_transaction_commit(trans->root); 24191abe9b8aSliubo 24202ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2421a2de733cSArne Jansen 24229ed74f2dSJosef Bacik if (current->journal_info == trans) 24239ed74f2dSJosef Bacik current->journal_info = NULL; 24249ed74f2dSJosef Bacik 24252c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 242624bbcf04SYan, Zheng 242779154b1bSChris Mason return ret; 242849b25e05SJeff Mahoney 242956e9f6eaSDavid Sterba unlock_tree_log: 243056e9f6eaSDavid Sterba mutex_unlock(&fs_info->tree_log_mutex); 243156e9f6eaSDavid Sterba unlock_reloc: 243256e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 24336cf7f77eSWang Shilong scrub_continue: 24342ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 243549b25e05SJeff Mahoney cleanup_transaction: 2436dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2437c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 24384fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24390e721106SJosef Bacik trans->block_rsv = NULL; 24400b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 244149b25e05SJeff Mahoney if (current->journal_info == trans) 244249b25e05SJeff Mahoney current->journal_info = NULL; 244397cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 244449b25e05SJeff Mahoney 244549b25e05SJeff Mahoney return ret; 244679154b1bSChris Mason } 244779154b1bSChris Mason 2448d352ac68SChris Mason /* 24499d1a2a3aSDavid Sterba * return < 0 if error 24509d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 24519d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 24529d1a2a3aSDavid Sterba * 24539d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 24549d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 24559d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 24569d1a2a3aSDavid Sterba * few seconds later. 2457d352ac68SChris Mason */ 24589d1a2a3aSDavid Sterba int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) 2459e9d0b13bSChris Mason { 24609d1a2a3aSDavid Sterba int ret; 24615d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 2462e9d0b13bSChris Mason 2463a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 24649d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 24659d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 24669d1a2a3aSDavid Sterba return 0; 24679d1a2a3aSDavid Sterba } 24689d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 24699d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2470cfad392bSJosef Bacik list_del_init(&root->root_list); 2471a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 24725d4f98a2SYan Zheng 24734fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 247476dda93cSYan, Zheng 247516cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 247616cdcec7SMiao Xie 247776dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 247876dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 24790078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 248076dda93cSYan, Zheng else 24810078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 248232471dc2SDavid Sterba 2483dc9492c1SJosef Bacik btrfs_put_root(root); 24846596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2485e9d0b13bSChris Mason } 2486572d9ab7SDavid Sterba 2487572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2488572d9ab7SDavid Sterba { 2489572d9ab7SDavid Sterba unsigned long prev; 2490572d9ab7SDavid Sterba unsigned long bit; 2491572d9ab7SDavid Sterba 24926c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2493572d9ab7SDavid Sterba if (!prev) 2494572d9ab7SDavid Sterba return; 2495572d9ab7SDavid Sterba 2496d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2497d51033d0SDavid Sterba if (prev & bit) 2498d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2499d51033d0SDavid Sterba prev &= ~bit; 2500d51033d0SDavid Sterba 2501572d9ab7SDavid Sterba if (prev) 2502572d9ab7SDavid Sterba btrfs_warn(fs_info, 2503572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2504572d9ab7SDavid Sterba } 2505