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" 19581bb050SLi Zefan #include "inode-map.h" 20733f4fbbSStefan Behrens #include "volumes.h" 218dabb742SStefan Behrens #include "dev-replace.h" 22fcebe456SJosef Bacik #include "qgroup.h" 23aac0023cSJosef Bacik #include "block-group.h" 249c343784SJosef Bacik #include "space-info.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), 111bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1124a9d8bdeSMiao Xie __TRANS_ATTACH | 1134a9d8bdeSMiao Xie __TRANS_JOIN | 114a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 115a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1164a9d8bdeSMiao Xie }; 1174a9d8bdeSMiao Xie 118724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 11979154b1bSChris Mason { 1209b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1219b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 122a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1235c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1245c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 12581f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 12681f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1271262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 128ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 129ab8d0fc4SJeff Mahoney "pending csums is %llu", 1301262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1317785a663SFilipe Manana /* 1327785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1337785a663SFilipe Manana * because the transaction was aborted and a commit did not 1347785a663SFilipe Manana * happen (things failed before writing the new superblock 1357785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1367785a663SFilipe Manana * discard the physical locations of the block groups. 1377785a663SFilipe Manana */ 1387785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 13932da5386SDavid Sterba struct btrfs_block_group *cache; 1407785a663SFilipe Manana 1417785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 14232da5386SDavid Sterba struct btrfs_block_group, 1437785a663SFilipe Manana bg_list); 1447785a663SFilipe Manana list_del_init(&cache->bg_list); 1456b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1467785a663SFilipe Manana btrfs_put_block_group(cache); 1477785a663SFilipe Manana } 148bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1494b5faeacSDavid Sterba kfree(transaction); 15079154b1bSChris Mason } 15178fae27eSChris Mason } 15279154b1bSChris Mason 153889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 154817d52f8SJosef Bacik { 155889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 15616916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1579e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 15827d56e62SJosef Bacik struct btrfs_caching_control *caching_ctl, *next; 1599e351cc8SJosef Bacik 1609e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 161889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1629e351cc8SJosef Bacik dirty_list) { 1639e351cc8SJosef Bacik list_del_init(&root->dirty_list); 164817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 165817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 1664fd786e6SMisono Tomohiro if (is_fstree(root->root_key.objectid)) 1679e351cc8SJosef Bacik btrfs_unpin_free_ino(root); 16841e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 169370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1709e351cc8SJosef Bacik } 1712b9dbef2SJosef Bacik 1722b9dbef2SJosef Bacik /* We can free old roots now. */ 173889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 174889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 175889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1762b9dbef2SJosef Bacik struct btrfs_root, root_list); 1772b9dbef2SJosef Bacik list_del_init(&root->root_list); 178889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 179889bfa39SJosef Bacik btrfs_free_log(trans, root); 1802b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 181889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1822b9dbef2SJosef Bacik } 183889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 18427d56e62SJosef Bacik 18527d56e62SJosef Bacik /* 18627d56e62SJosef Bacik * We have to update the last_byte_to_unpin under the commit_root_sem, 18727d56e62SJosef Bacik * at the same time we swap out the commit roots. 18827d56e62SJosef Bacik * 18927d56e62SJosef Bacik * This is because we must have a real view of the last spot the caching 19027d56e62SJosef Bacik * kthreads were while caching. Consider the following views of the 19127d56e62SJosef Bacik * extent tree for a block group 19227d56e62SJosef Bacik * 19327d56e62SJosef Bacik * commit root 19427d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 19527d56e62SJosef Bacik * |\\\\| |\\\\|\\\\| |\\\\|\\\\| 19627d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 19727d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 19827d56e62SJosef Bacik * 19927d56e62SJosef Bacik * new commit root 20027d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20127d56e62SJosef Bacik * | | | |\\\\| | |\\\\| 20227d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20327d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 20427d56e62SJosef Bacik * 20527d56e62SJosef Bacik * If the cache_ctl->progress was at 3, then we are only allowed to 20627d56e62SJosef Bacik * unpin [0,1) and [2,3], because the caching thread has already 20727d56e62SJosef Bacik * processed those extents. We are not allowed to unpin [5,6), because 20827d56e62SJosef Bacik * the caching thread will re-start it's search from 3, and thus find 20927d56e62SJosef Bacik * the hole from [4,6) to add to the free space cache. 21027d56e62SJosef Bacik */ 211*bbb86a37SJosef Bacik spin_lock(&fs_info->block_group_cache_lock); 21227d56e62SJosef Bacik list_for_each_entry_safe(caching_ctl, next, 21327d56e62SJosef Bacik &fs_info->caching_block_groups, list) { 21427d56e62SJosef Bacik struct btrfs_block_group *cache = caching_ctl->block_group; 21527d56e62SJosef Bacik 21627d56e62SJosef Bacik if (btrfs_block_group_done(cache)) { 21727d56e62SJosef Bacik cache->last_byte_to_unpin = (u64)-1; 21827d56e62SJosef Bacik list_del_init(&caching_ctl->list); 21927d56e62SJosef Bacik btrfs_put_caching_control(caching_ctl); 22027d56e62SJosef Bacik } else { 22127d56e62SJosef Bacik cache->last_byte_to_unpin = caching_ctl->progress; 22227d56e62SJosef Bacik } 22327d56e62SJosef Bacik } 224*bbb86a37SJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 2259e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 226817d52f8SJosef Bacik } 227817d52f8SJosef Bacik 2280860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2290860adfdSMiao Xie unsigned int type) 2300860adfdSMiao Xie { 2310860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2320860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2330860adfdSMiao Xie } 2340860adfdSMiao Xie 2350860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2360860adfdSMiao Xie unsigned int type) 2370860adfdSMiao Xie { 2380860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2390860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2400860adfdSMiao Xie } 2410860adfdSMiao Xie 2420860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2430860adfdSMiao Xie unsigned int type) 2440860adfdSMiao Xie { 2450860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2460860adfdSMiao Xie } 2470860adfdSMiao Xie 2480860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2490860adfdSMiao Xie { 2500860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 251178260b2SMiao Xie } 252178260b2SMiao Xie 253d352ac68SChris Mason /* 254fb6dea26SJosef Bacik * To be called after all the new block groups attached to the transaction 255fb6dea26SJosef Bacik * handle have been created (btrfs_create_pending_block_groups()). 256fb6dea26SJosef Bacik */ 257fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 258fb6dea26SJosef Bacik { 259fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 260fb6dea26SJosef Bacik 261fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 262fb6dea26SJosef Bacik return; 263fb6dea26SJosef Bacik 264fb6dea26SJosef Bacik WARN_ON_ONCE(!list_empty(&trans->new_bgs)); 265fb6dea26SJosef Bacik 266fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 26763f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 268fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 269fb6dea26SJosef Bacik } 270fb6dea26SJosef Bacik 271fb6dea26SJosef Bacik /* 272d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 273d352ac68SChris Mason */ 2742ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2752ff7e61eSJeff Mahoney unsigned int type) 27679154b1bSChris Mason { 27779154b1bSChris Mason struct btrfs_transaction *cur_trans; 278a4abeea4SJosef Bacik 27919ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 280d43317dcSChris Mason loop: 28149b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 28287533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 28319ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 28449b25e05SJeff Mahoney return -EROFS; 28549b25e05SJeff Mahoney } 28649b25e05SJeff Mahoney 28719ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 288a4abeea4SJosef Bacik if (cur_trans) { 289bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 29019ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 29149b25e05SJeff Mahoney return cur_trans->aborted; 292871383beSDavid Sterba } 2934a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 294178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 295178260b2SMiao Xie return -EBUSY; 296178260b2SMiao Xie } 2979b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 298a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 2990860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 30019ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 301a4abeea4SJosef Bacik return 0; 302a4abeea4SJosef Bacik } 30319ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 304a4abeea4SJosef Bacik 305354aa0fbSMiao Xie /* 306354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 307354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 308354aa0fbSMiao Xie */ 309354aa0fbSMiao Xie if (type == TRANS_ATTACH) 310354aa0fbSMiao Xie return -ENOENT; 311354aa0fbSMiao Xie 3124a9d8bdeSMiao Xie /* 3134a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3144a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3154a9d8bdeSMiao Xie */ 3164a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3174a9d8bdeSMiao Xie 3184b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 319db5b493aSTsutomu Itoh if (!cur_trans) 320db5b493aSTsutomu Itoh return -ENOMEM; 321d43317dcSChris Mason 32219ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 32319ae4e81SJan Schmidt if (fs_info->running_transaction) { 324d43317dcSChris Mason /* 325d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3264a9d8bdeSMiao Xie * to redo the checks above 327d43317dcSChris Mason */ 3284b5faeacSDavid Sterba kfree(cur_trans); 329d43317dcSChris Mason goto loop; 33087533c47SMiao Xie } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 331e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3324b5faeacSDavid Sterba kfree(cur_trans); 3337b8b92afSJosef Bacik return -EROFS; 334a4abeea4SJosef Bacik } 335d43317dcSChris Mason 336ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 33748778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 33848778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 33913c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3400860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 34179154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 34279154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3434a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 344a4abeea4SJosef Bacik /* 345a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 346a4abeea4SJosef Bacik * commit the transaction. 347a4abeea4SJosef Bacik */ 3489b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3493204d33cSJosef Bacik cur_trans->flags = 0; 350afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 35156bec294SChris Mason 352a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 353a099d0fdSAlexandru Moise 3545c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3553368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 356d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 35720b297d6SJan Schmidt 35820b297d6SJan Schmidt /* 35920b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 36020b297d6SJan Schmidt * the log must never go across transaction boundaries. 36120b297d6SJan Schmidt */ 36220b297d6SJan Schmidt smp_mb(); 36331b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3645d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 36531b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3665d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 367fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 36820b297d6SJan Schmidt 36956bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 37056bec294SChris Mason 3713063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 372bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3739e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 374ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3751bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3762b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3771bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 378ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 379e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3802b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 38119ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 382c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 38343eb5f29SQu Wenruo IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); 384fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 385fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 38619ae4e81SJan Schmidt fs_info->generation++; 38719ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 38819ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 38949b25e05SJeff Mahoney cur_trans->aborted = 0; 39019ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 39115ee9bc7SJosef Bacik 39279154b1bSChris Mason return 0; 39379154b1bSChris Mason } 39479154b1bSChris Mason 395d352ac68SChris Mason /* 39692a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 39792a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 39892a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 39992a7cc42SQu Wenruo * transaction commits. 400d352ac68SChris Mason */ 4017585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 4026426c7adSQu Wenruo struct btrfs_root *root, 4036426c7adSQu Wenruo int force) 4046702ed49SChris Mason { 4050b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4060b246afaSJeff Mahoney 40792a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4086426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4090b246afaSJeff Mahoney WARN_ON(root == fs_info->extent_root); 4104d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4115d4f98a2SYan Zheng 4127585717fSChris Mason /* 41327cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4147585717fSChris Mason * we have the reloc mutex held now, so there 4157585717fSChris Mason * is only one writer in this function 4167585717fSChris Mason */ 41727cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4187585717fSChris Mason 41927cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4207585717fSChris Mason * they find our root->last_trans update 4217585717fSChris Mason */ 4227585717fSChris Mason smp_wmb(); 4237585717fSChris Mason 4240b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4256426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 4260b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 427a4abeea4SJosef Bacik return 0; 428a4abeea4SJosef Bacik } 4290b246afaSJeff Mahoney radix_tree_tag_set(&fs_info->fs_roots_radix, 4306702ed49SChris Mason (unsigned long)root->root_key.objectid, 4316702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 4320b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4337585717fSChris Mason root->last_trans = trans->transid; 4347585717fSChris Mason 4357585717fSChris Mason /* this is pretty tricky. We don't want to 4367585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4377585717fSChris Mason * unless we're really doing the first setup for this root in 4387585717fSChris Mason * this transaction. 4397585717fSChris Mason * 4407585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4417585717fSChris Mason * if we want to take the expensive mutex. 4427585717fSChris Mason * 4437585717fSChris Mason * But, we have to set root->last_trans before we 4447585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4457585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 44627cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4477585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4487585717fSChris Mason * 4497585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4507585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4517585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4527585717fSChris Mason * done before we pop in the zero below 4537585717fSChris Mason */ 4545d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 455c7548af6SChris Mason smp_mb__before_atomic(); 45627cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4576702ed49SChris Mason } 4585d4f98a2SYan Zheng return 0; 4596702ed49SChris Mason } 4605d4f98a2SYan Zheng 4617585717fSChris Mason 4622b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4632b9dbef2SJosef Bacik struct btrfs_root *root) 4642b9dbef2SJosef Bacik { 4650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4662b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4672b9dbef2SJosef Bacik 4682b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4692b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4702b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4712b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4722b9dbef2SJosef Bacik 4732b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 4740b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4750b246afaSJeff Mahoney radix_tree_tag_clear(&fs_info->fs_roots_radix, 4762b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4772b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 4780b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4792b9dbef2SJosef Bacik } 4802b9dbef2SJosef Bacik 4817585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4827585717fSChris Mason struct btrfs_root *root) 4837585717fSChris Mason { 4840b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4850b246afaSJeff Mahoney 48692a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4877585717fSChris Mason return 0; 4887585717fSChris Mason 4897585717fSChris Mason /* 49027cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4917585717fSChris Mason * and barriers 4927585717fSChris Mason */ 4937585717fSChris Mason smp_rmb(); 4947585717fSChris Mason if (root->last_trans == trans->transid && 49527cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4967585717fSChris Mason return 0; 4977585717fSChris Mason 4980b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 4996426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 5000b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 5017585717fSChris Mason 5027585717fSChris Mason return 0; 5037585717fSChris Mason } 5047585717fSChris Mason 5054a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5064a9d8bdeSMiao Xie { 5073296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 508501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 509bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5104a9d8bdeSMiao Xie } 5114a9d8bdeSMiao Xie 512d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 513d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 514d352ac68SChris Mason * transaction might not be fully on disk. 515d352ac68SChris Mason */ 5162ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 51779154b1bSChris Mason { 518f9295749SChris Mason struct btrfs_transaction *cur_trans; 51979154b1bSChris Mason 5200b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5210b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5224a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5239b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5240b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52572d63ed6SLi Zefan 5260b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 527501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 528bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 529724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 530a4abeea4SJosef Bacik } else { 5310b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 532f9295749SChris Mason } 53337d1aeeeSChris Mason } 53437d1aeeeSChris Mason 5352ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 53637d1aeeeSChris Mason { 5370b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 538a4abeea4SJosef Bacik return 0; 539a4abeea4SJosef Bacik 54092e2f7e3SNikolay Borisov if (type == TRANS_START) 541a4abeea4SJosef Bacik return 1; 542a4abeea4SJosef Bacik 543a22285a6SYan, Zheng return 0; 544a22285a6SYan, Zheng } 545a22285a6SYan, Zheng 54620dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 54720dd2cbfSMiao Xie { 5480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5490b246afaSJeff Mahoney 5500b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 55192a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 55220dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 55320dd2cbfSMiao Xie root->reloc_root) 55420dd2cbfSMiao Xie return false; 55520dd2cbfSMiao Xie 55620dd2cbfSMiao Xie return true; 55720dd2cbfSMiao Xie } 55820dd2cbfSMiao Xie 55908e007d2SMiao Xie static struct btrfs_trans_handle * 5605aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 561003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 562003d7c59SJeff Mahoney bool enforce_qgroups) 563a22285a6SYan, Zheng { 5640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 565ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 566a22285a6SYan, Zheng struct btrfs_trans_handle *h; 567a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 568b5009945SJosef Bacik u64 num_bytes = 0; 569c5567237SArne Jansen u64 qgroup_reserved = 0; 57020dd2cbfSMiao Xie bool reloc_reserved = false; 5719c343784SJosef Bacik bool do_chunk_alloc = false; 57220dd2cbfSMiao Xie int ret; 573acce952bSliubo 57446c4e71eSFilipe Manana /* Send isn't supposed to start transactions. */ 5752755a0deSDavid Sterba ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); 57646c4e71eSFilipe Manana 5770b246afaSJeff Mahoney if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 578acce952bSliubo return ERR_PTR(-EROFS); 5792a1eb461SJosef Bacik 58046c4e71eSFilipe Manana if (current->journal_info) { 5810860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5822a1eb461SJosef Bacik h = current->journal_info; 583b50fff81SDavid Sterba refcount_inc(&h->use_count); 584b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5852a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5862a1eb461SJosef Bacik h->block_rsv = NULL; 5872a1eb461SJosef Bacik goto got_it; 5882a1eb461SJosef Bacik } 589b5009945SJosef Bacik 590b5009945SJosef Bacik /* 591b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 592b5009945SJosef Bacik * the appropriate flushing if need be. 593b5009945SJosef Bacik */ 594003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 595ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 596ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 597ba2c4d4eSJosef Bacik 5980b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 599733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 600003d7c59SJeff Mahoney enforce_qgroups); 601c5567237SArne Jansen if (ret) 602c5567237SArne Jansen return ERR_PTR(ret); 603c5567237SArne Jansen 604ba2c4d4eSJosef Bacik /* 605ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 606ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 607ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 608ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 609ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 610ba2c4d4eSJosef Bacik */ 6112bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6127f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 6137f9fe614SJosef Bacik delayed_refs_rsv->full == 0) { 614ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 615ba2c4d4eSJosef Bacik num_bytes <<= 1; 616ba2c4d4eSJosef Bacik } 617ba2c4d4eSJosef Bacik 61820dd2cbfSMiao Xie /* 61920dd2cbfSMiao Xie * Do the reservation for the relocation root creation 62020dd2cbfSMiao Xie */ 621ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6220b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 62320dd2cbfSMiao Xie reloc_reserved = true; 62420dd2cbfSMiao Xie } 62520dd2cbfSMiao Xie 626ba2c4d4eSJosef Bacik ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); 627ba2c4d4eSJosef Bacik if (ret) 628ba2c4d4eSJosef Bacik goto reserve_fail; 629ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 630ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 631ba2c4d4eSJosef Bacik delayed_refs_bytes); 632ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 633ba2c4d4eSJosef Bacik } 6349c343784SJosef Bacik 6359c343784SJosef Bacik if (rsv->space_info->force_alloc) 6369c343784SJosef Bacik do_chunk_alloc = true; 637ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 638ba2c4d4eSJosef Bacik !delayed_refs_rsv->full) { 639ba2c4d4eSJosef Bacik /* 640ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 641ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 642ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 643ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 644ba2c4d4eSJosef Bacik * here. 645ba2c4d4eSJosef Bacik */ 646ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 647b5009945SJosef Bacik if (ret) 648843fcf35SMiao Xie goto reserve_fail; 649b5009945SJosef Bacik } 650a22285a6SYan, Zheng again: 651f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 652843fcf35SMiao Xie if (!h) { 653843fcf35SMiao Xie ret = -ENOMEM; 654843fcf35SMiao Xie goto alloc_fail; 655843fcf35SMiao Xie } 656a22285a6SYan, Zheng 65798114659SJosef Bacik /* 65898114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 65998114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 66098114659SJosef Bacik * because we're already holding a ref. We need this because we could 66198114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 66298114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 663354aa0fbSMiao Xie * 664354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 665354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 66698114659SJosef Bacik */ 6670860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6680b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 669b2b5ef5cSJan Kara 6702ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6712ff7e61eSJeff Mahoney wait_current_trans(fs_info); 672a22285a6SYan, Zheng 673a4abeea4SJosef Bacik do { 6742ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 675178260b2SMiao Xie if (ret == -EBUSY) { 6762ff7e61eSJeff Mahoney wait_current_trans(fs_info); 677a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 678a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 679178260b2SMiao Xie ret = -ENOENT; 680178260b2SMiao Xie } 681a4abeea4SJosef Bacik } while (ret == -EBUSY); 682a4abeea4SJosef Bacik 683a43f7f82SLiu Bo if (ret < 0) 684843fcf35SMiao Xie goto join_fail; 6850f7d52f4SChris Mason 6860b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 687a22285a6SYan, Zheng 688a22285a6SYan, Zheng h->transid = cur_trans->transid; 689a22285a6SYan, Zheng h->transaction = cur_trans; 690d13603efSArne Jansen h->root = root; 691b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 69264b63580SJeff Mahoney h->fs_info = root->fs_info; 6937174109cSQu Wenruo 694a698d075SMiao Xie h->type = type; 695d9a0540aSFilipe Manana h->can_flush_pending_bgs = true; 696ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 697b7ec40d7SChris Mason 698a22285a6SYan, Zheng smp_mb(); 6993296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 7002ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 701abdd2e80SFilipe Manana current->journal_info = h; 7023a45bb20SJeff Mahoney btrfs_commit_transaction(h); 703a22285a6SYan, Zheng goto again; 704a22285a6SYan, Zheng } 7059ed74f2dSJosef Bacik 706b5009945SJosef Bacik if (num_bytes) { 7070b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7082bcc0328SLiu Bo h->transid, num_bytes, 1); 7090b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 710b5009945SJosef Bacik h->bytes_reserved = num_bytes; 71120dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 712a22285a6SYan, Zheng } 713a22285a6SYan, Zheng 7142a1eb461SJosef Bacik got_it: 715bcf3a3e7SNikolay Borisov if (!current->journal_info) 716a22285a6SYan, Zheng current->journal_info = h; 717fcc99734SQu Wenruo 718fcc99734SQu Wenruo /* 7199c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7209c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7219c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7229c343784SJosef Bacik * value here. 7239c343784SJosef Bacik */ 7249c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7259c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7269c343784SJosef Bacik 7279c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7289c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7299c343784SJosef Bacik } 7309c343784SJosef Bacik 7319c343784SJosef Bacik /* 732fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 733fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 734fcc99734SQu Wenruo * transaction. 735fcc99734SQu Wenruo * 736fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 737fcc99734SQu Wenruo * or we can deadlock. 738fcc99734SQu Wenruo */ 739fcc99734SQu Wenruo btrfs_record_root_in_trans(h, root); 740fcc99734SQu Wenruo 74179154b1bSChris Mason return h; 742843fcf35SMiao Xie 743843fcf35SMiao Xie join_fail: 7440860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7450b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 746843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 747843fcf35SMiao Xie alloc_fail: 748843fcf35SMiao Xie if (num_bytes) 7492ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 75063f018beSNikolay Borisov num_bytes, NULL); 751843fcf35SMiao Xie reserve_fail: 752733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 753843fcf35SMiao Xie return ERR_PTR(ret); 75479154b1bSChris Mason } 75579154b1bSChris Mason 756f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7575aed1dd8SAlexandru Moise unsigned int num_items) 758f9295749SChris Mason { 75908e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 760003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 761f9295749SChris Mason } 762003d7c59SJeff Mahoney 7638eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7648eab77ffSFilipe Manana struct btrfs_root *root, 7657f9fe614SJosef Bacik unsigned int num_items) 7668eab77ffSFilipe Manana { 7677f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7687f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7698eab77ffSFilipe Manana } 7708407aa46SMiao Xie 7717a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 772f9295749SChris Mason { 773003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 774003d7c59SJeff Mahoney true); 775f9295749SChris Mason } 776f9295749SChris Mason 7778d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7780af3d00bSJosef Bacik { 779575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 780003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7810af3d00bSJosef Bacik } 7820af3d00bSJosef Bacik 783d4edf39bSMiao Xie /* 784a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 785a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 786a6d155d2SFilipe Manana */ 787a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 788a6d155d2SFilipe Manana { 789a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 790a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 791a6d155d2SFilipe Manana } 792a6d155d2SFilipe Manana 793a6d155d2SFilipe Manana /* 794d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 795d4edf39bSMiao Xie * 796d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 797d4edf39bSMiao Xie * don't want to start a new one. 798d4edf39bSMiao Xie * 799d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 800d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 801d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 802d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 803d4edf39bSMiao Xie * invoke 804d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 805d4edf39bSMiao Xie */ 806354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 80760376ce4SJosef Bacik { 808575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 809003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 81060376ce4SJosef Bacik } 81160376ce4SJosef Bacik 812d4edf39bSMiao Xie /* 81390b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 814d4edf39bSMiao Xie * 81552042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 816d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 817d4edf39bSMiao Xie * complete. 818d4edf39bSMiao Xie */ 819d4edf39bSMiao Xie struct btrfs_trans_handle * 820d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 821d4edf39bSMiao Xie { 822d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 823d4edf39bSMiao Xie 824575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 825003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8268d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8272ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 828d4edf39bSMiao Xie 829d4edf39bSMiao Xie return trans; 830d4edf39bSMiao Xie } 831d4edf39bSMiao Xie 832d352ac68SChris Mason /* wait for a transaction commit to be fully complete */ 8332ff7e61eSJeff Mahoney static noinline void wait_for_commit(struct btrfs_transaction *commit) 83489ce8a63SChris Mason { 8354a9d8bdeSMiao Xie wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); 83689ce8a63SChris Mason } 83789ce8a63SChris Mason 8382ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 83946204592SSage Weil { 84046204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8418cd2807fSMiao Xie int ret = 0; 84246204592SSage Weil 84346204592SSage Weil if (transid) { 8440b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 845a4abeea4SJosef Bacik goto out; 84646204592SSage Weil 84746204592SSage Weil /* find specified transaction */ 8480b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8490b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 85046204592SSage Weil if (t->transid == transid) { 85146204592SSage Weil cur_trans = t; 8529b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8538cd2807fSMiao Xie ret = 0; 85446204592SSage Weil break; 85546204592SSage Weil } 8568cd2807fSMiao Xie if (t->transid > transid) { 8578cd2807fSMiao Xie ret = 0; 85846204592SSage Weil break; 85946204592SSage Weil } 8608cd2807fSMiao Xie } 8610b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 86242383020SSage Weil 86342383020SSage Weil /* 86442383020SSage Weil * The specified transaction doesn't exist, or we 86542383020SSage Weil * raced with btrfs_commit_transaction 86642383020SSage Weil */ 86742383020SSage Weil if (!cur_trans) { 8680b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 86942383020SSage Weil ret = -EINVAL; 8708cd2807fSMiao Xie goto out; 87142383020SSage Weil } 87246204592SSage Weil } else { 87346204592SSage Weil /* find newest transaction that is committing | committed */ 8740b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8750b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 87646204592SSage Weil list) { 8774a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 8784a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 8793473f3c0SJosef Bacik break; 88046204592SSage Weil cur_trans = t; 8819b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 88246204592SSage Weil break; 88346204592SSage Weil } 88446204592SSage Weil } 8850b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 88646204592SSage Weil if (!cur_trans) 887a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 88846204592SSage Weil } 88946204592SSage Weil 8902ff7e61eSJeff Mahoney wait_for_commit(cur_trans); 891724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 892a4abeea4SJosef Bacik out: 89346204592SSage Weil return ret; 89446204592SSage Weil } 89546204592SSage Weil 8962ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 89737d1aeeeSChris Mason { 8982ff7e61eSJeff Mahoney wait_current_trans(fs_info); 89937d1aeeeSChris Mason } 90037d1aeeeSChris Mason 9012ff7e61eSJeff Mahoney static int should_end_transaction(struct btrfs_trans_handle *trans) 9028929ecfaSYan, Zheng { 9032ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9040b246afaSJeff Mahoney 90564403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9061be41b78SJosef Bacik return 1; 90736ba022aSJosef Bacik 9082ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9098929ecfaSYan, Zheng } 9108929ecfaSYan, Zheng 9113a45bb20SJeff Mahoney int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9128929ecfaSYan, Zheng { 9138929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9148929ecfaSYan, Zheng 915a4abeea4SJosef Bacik smp_mb(); 9163296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 9174a9d8bdeSMiao Xie cur_trans->delayed_refs.flushing) 9188929ecfaSYan, Zheng return 1; 9198929ecfaSYan, Zheng 9202ff7e61eSJeff Mahoney return should_end_transaction(trans); 9218929ecfaSYan, Zheng } 9228929ecfaSYan, Zheng 923dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 924dc60c525SNikolay Borisov 9250e34693fSNikolay Borisov { 926dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 927dc60c525SNikolay Borisov 9280e34693fSNikolay Borisov if (!trans->block_rsv) { 9290e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9300e34693fSNikolay Borisov return; 9310e34693fSNikolay Borisov } 9320e34693fSNikolay Borisov 9330e34693fSNikolay Borisov if (!trans->bytes_reserved) 9340e34693fSNikolay Borisov return; 9350e34693fSNikolay Borisov 9360e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9370e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9380e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9390e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 94063f018beSNikolay Borisov trans->bytes_reserved, NULL); 9410e34693fSNikolay Borisov trans->bytes_reserved = 0; 9420e34693fSNikolay Borisov } 9430e34693fSNikolay Borisov 94489ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9453a45bb20SJeff Mahoney int throttle) 94679154b1bSChris Mason { 9473a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9488929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9494edc2ca3SDave Jones int err = 0; 950d6e4a428SChris Mason 951b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 952b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9532a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9542a1eb461SJosef Bacik return 0; 9552a1eb461SJosef Bacik } 9562a1eb461SJosef Bacik 957dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9584c13d758SJosef Bacik trans->block_rsv = NULL; 959c5567237SArne Jansen 9606c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 961ea658badSJosef Bacik 9624fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9634fbcdf66SFilipe Manana 9640860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9650b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9666df7881aSJosef Bacik 9678929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 96813c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 96913c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 9700860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 97189ce8a63SChris Mason 972093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 973724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 9749ed74f2dSJosef Bacik 9759ed74f2dSJosef Bacik if (current->journal_info == trans) 9769ed74f2dSJosef Bacik current->journal_info = NULL; 977ab78c84dSChris Mason 97824bbcf04SYan, Zheng if (throttle) 9792ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 98024bbcf04SYan, Zheng 981bf31f87fSDavid Sterba if (TRANS_ABORTED(trans) || 9820b246afaSJeff Mahoney test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { 9834e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 984fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 985fbabd4a3SJosef Bacik err = trans->aborted; 986fbabd4a3SJosef Bacik else 987fbabd4a3SJosef Bacik err = -EROFS; 9884e121c06SJosef Bacik } 98949b25e05SJeff Mahoney 9904edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 9914edc2ca3SDave Jones return err; 99279154b1bSChris Mason } 99379154b1bSChris Mason 9943a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 99589ce8a63SChris Mason { 9963a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 99789ce8a63SChris Mason } 99889ce8a63SChris Mason 9993a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 100089ce8a63SChris Mason { 10013a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 100216cdcec7SMiao Xie } 100316cdcec7SMiao Xie 1004d352ac68SChris Mason /* 1005d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1006d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1007690587d1SChris Mason * those extents are sent to disk but does not wait on them 1008d352ac68SChris Mason */ 10092ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10108cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 101179154b1bSChris Mason { 1012777e6bd7SChris Mason int err = 0; 10137c4452b9SChris Mason int werr = 0; 10140b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1015e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1016777e6bd7SChris Mason u64 start = 0; 10175f39d397SChris Mason u64 end; 10187c4452b9SChris Mason 10196300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10201728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1021e6138876SJosef Bacik mark, &cached_state)) { 1022663dfbb0SFilipe Manana bool wait_writeback = false; 1023663dfbb0SFilipe Manana 1024663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1025663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1026210aa277SDavid Sterba mark, &cached_state); 1027663dfbb0SFilipe Manana /* 1028663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1029663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1030663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1031663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1032bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1033bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1034bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1035bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1036bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1037663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 103841e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1039663dfbb0SFilipe Manana */ 1040663dfbb0SFilipe Manana if (err == -ENOMEM) { 1041663dfbb0SFilipe Manana err = 0; 1042663dfbb0SFilipe Manana wait_writeback = true; 1043663dfbb0SFilipe Manana } 1044663dfbb0SFilipe Manana if (!err) 10451728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10467c4452b9SChris Mason if (err) 10477c4452b9SChris Mason werr = err; 1048663dfbb0SFilipe Manana else if (wait_writeback) 1049663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1050e38e2ed7SFilipe Manana free_extent_state(cached_state); 1051663dfbb0SFilipe Manana cached_state = NULL; 10521728366eSJosef Bacik cond_resched(); 10531728366eSJosef Bacik start = end + 1; 10547c4452b9SChris Mason } 10556300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1056690587d1SChris Mason return werr; 1057690587d1SChris Mason } 1058690587d1SChris Mason 1059690587d1SChris Mason /* 1060690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1061690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1062690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1063690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1064690587d1SChris Mason */ 1065bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1066bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1067690587d1SChris Mason { 1068690587d1SChris Mason int err = 0; 1069690587d1SChris Mason int werr = 0; 10700b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1071e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1072690587d1SChris Mason u64 start = 0; 1073690587d1SChris Mason u64 end; 1074690587d1SChris Mason 10751728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1076e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1077663dfbb0SFilipe Manana /* 1078663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1079663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1080663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1081663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1082663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 108341e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1084663dfbb0SFilipe Manana */ 1085663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1086ae0f1625SDavid Sterba EXTENT_NEED_WAIT, 0, 0, &cached_state); 1087663dfbb0SFilipe Manana if (err == -ENOMEM) 1088663dfbb0SFilipe Manana err = 0; 1089663dfbb0SFilipe Manana if (!err) 10901728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1091777e6bd7SChris Mason if (err) 1092777e6bd7SChris Mason werr = err; 1093e38e2ed7SFilipe Manana free_extent_state(cached_state); 1094e38e2ed7SFilipe Manana cached_state = NULL; 1095777e6bd7SChris Mason cond_resched(); 10961728366eSJosef Bacik start = end + 1; 1097777e6bd7SChris Mason } 10987c4452b9SChris Mason if (err) 10997c4452b9SChris Mason werr = err; 1100bf89d38fSJeff Mahoney return werr; 1101bf89d38fSJeff Mahoney } 1102656f30dbSFilipe Manana 1103b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1104bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1105bf89d38fSJeff Mahoney { 1106bf89d38fSJeff Mahoney bool errors = false; 1107bf89d38fSJeff Mahoney int err; 1108bf89d38fSJeff Mahoney 1109bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1110bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1111bf89d38fSJeff Mahoney errors = true; 1112bf89d38fSJeff Mahoney 1113bf89d38fSJeff Mahoney if (errors && !err) 1114bf89d38fSJeff Mahoney err = -EIO; 1115bf89d38fSJeff Mahoney return err; 1116bf89d38fSJeff Mahoney } 1117bf89d38fSJeff Mahoney 1118bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1119bf89d38fSJeff Mahoney { 1120bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1121bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1122bf89d38fSJeff Mahoney bool errors = false; 1123bf89d38fSJeff Mahoney int err; 1124bf89d38fSJeff Mahoney 1125bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1126bf89d38fSJeff Mahoney 1127bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1128656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11290b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1130656f30dbSFilipe Manana errors = true; 1131656f30dbSFilipe Manana 1132656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11330b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1134656f30dbSFilipe Manana errors = true; 1135656f30dbSFilipe Manana 1136bf89d38fSJeff Mahoney if (errors && !err) 1137bf89d38fSJeff Mahoney err = -EIO; 1138bf89d38fSJeff Mahoney return err; 113979154b1bSChris Mason } 114079154b1bSChris Mason 1141690587d1SChris Mason /* 1142c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1143c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1144c9b577c0SNikolay Borisov * log commit. 1145c9b577c0SNikolay Borisov * 1146c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1147690587d1SChris Mason */ 114870458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1149d0c803c4SChris Mason { 1150663dfbb0SFilipe Manana int ret; 1151c9b577c0SNikolay Borisov int ret2; 1152c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 115370458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1154c9b577c0SNikolay Borisov struct blk_plug plug; 1155663dfbb0SFilipe Manana 1156c9b577c0SNikolay Borisov blk_start_plug(&plug); 1157c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1158c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1159c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1160c9b577c0SNikolay Borisov 116141e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1162663dfbb0SFilipe Manana 1163c9b577c0SNikolay Borisov if (ret) 1164663dfbb0SFilipe Manana return ret; 1165c9b577c0SNikolay Borisov else if (ret2) 1166c9b577c0SNikolay Borisov return ret2; 1167c9b577c0SNikolay Borisov else 1168c9b577c0SNikolay Borisov return 0; 1169d0c803c4SChris Mason } 1170d0c803c4SChris Mason 1171d352ac68SChris Mason /* 1172d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1173d352ac68SChris Mason * 1174d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1175d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1176d352ac68SChris Mason * allocation tree. 1177d352ac68SChris Mason * 1178d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1179d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1180d352ac68SChris Mason */ 11810b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 118279154b1bSChris Mason struct btrfs_root *root) 118379154b1bSChris Mason { 118479154b1bSChris Mason int ret; 11850b86a832SChris Mason u64 old_root_bytenr; 118686b9f2ecSYan, Zheng u64 old_root_used; 11870b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 11880b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 118979154b1bSChris Mason 119086b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 119156bec294SChris Mason 119279154b1bSChris Mason while (1) { 11930b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 119486b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1195ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 119679154b1bSChris Mason break; 119787ef2bb4SChris Mason 11985d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 119979154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12000b86a832SChris Mason &root->root_key, 12010b86a832SChris Mason &root->root_item); 120249b25e05SJeff Mahoney if (ret) 120349b25e05SJeff Mahoney return ret; 120456bec294SChris Mason 120586b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1206e7070be1SJosef Bacik } 1207276e680dSYan Zheng 12080b86a832SChris Mason return 0; 12090b86a832SChris Mason } 12100b86a832SChris Mason 1211d352ac68SChris Mason /* 1212d352ac68SChris Mason * update all the cowonly tree roots on disk 121349b25e05SJeff Mahoney * 121449b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 121549b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 121649b25e05SJeff Mahoney * to clean up the delayed refs. 1217d352ac68SChris Mason */ 12189386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12190b86a832SChris Mason { 12209386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1221ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12221bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12230b86a832SChris Mason struct list_head *next; 122484234f3aSYan Zheng struct extent_buffer *eb; 122556bec294SChris Mason int ret; 122684234f3aSYan Zheng 122784234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 122849b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12299631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 123084234f3aSYan Zheng btrfs_tree_unlock(eb); 123184234f3aSYan Zheng free_extent_buffer(eb); 12320b86a832SChris Mason 123349b25e05SJeff Mahoney if (ret) 123449b25e05SJeff Mahoney return ret; 123549b25e05SJeff Mahoney 1236c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 123749b25e05SJeff Mahoney if (ret) 123849b25e05SJeff Mahoney return ret; 123987ef2bb4SChris Mason 1240196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1241c16ce190SJosef Bacik if (ret) 1242c16ce190SJosef Bacik return ret; 12432b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1244c16ce190SJosef Bacik if (ret) 1245c16ce190SJosef Bacik return ret; 1246280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1247c16ce190SJosef Bacik if (ret) 1248c16ce190SJosef Bacik return ret; 1249546adb0dSJan Schmidt 1250bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1251dcdf7f6dSJosef Bacik if (ret) 1252dcdf7f6dSJosef Bacik return ret; 1253dcdf7f6dSJosef Bacik 1254546adb0dSJan Schmidt /* run_qgroups might have added some more refs */ 1255c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1256c16ce190SJosef Bacik if (ret) 1257c16ce190SJosef Bacik return ret; 1258ea526d18SJosef Bacik again: 12590b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12602ff7e61eSJeff Mahoney struct btrfs_root *root; 12610b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12620b86a832SChris Mason list_del_init(next); 12630b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1264e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 126587ef2bb4SChris Mason 12669e351cc8SJosef Bacik if (root != fs_info->extent_root) 12679e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 12689e351cc8SJosef Bacik &trans->transaction->switch_commits); 126949b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 127049b25e05SJeff Mahoney if (ret) 127149b25e05SJeff Mahoney return ret; 1272c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1273ea526d18SJosef Bacik if (ret) 1274ea526d18SJosef Bacik return ret; 127579154b1bSChris Mason } 1276276e680dSYan Zheng 12771bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 12785742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1279ea526d18SJosef Bacik if (ret) 1280ea526d18SJosef Bacik return ret; 1281c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1282ea526d18SJosef Bacik if (ret) 1283ea526d18SJosef Bacik return ret; 1284ea526d18SJosef Bacik } 1285ea526d18SJosef Bacik 1286ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1287ea526d18SJosef Bacik goto again; 1288ea526d18SJosef Bacik 12899e351cc8SJosef Bacik list_add_tail(&fs_info->extent_root->dirty_list, 12909e351cc8SJosef Bacik &trans->transaction->switch_commits); 12919f6cbcbbSDavid Sterba 12929f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 12939f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 12949f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 12958dabb742SStefan Behrens 129679154b1bSChris Mason return 0; 129779154b1bSChris Mason } 129879154b1bSChris Mason 1299d352ac68SChris Mason /* 1300d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1301d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1302d352ac68SChris Mason * be deleted 1303d352ac68SChris Mason */ 1304cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13055eda7b5eSChris Mason { 13060b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13070b246afaSJeff Mahoney 13080b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1309dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1310dc9492c1SJosef Bacik btrfs_grab_root(root); 13110b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1312dc9492c1SJosef Bacik } 13130b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13145eda7b5eSChris Mason } 13155eda7b5eSChris Mason 1316d352ac68SChris Mason /* 13175d4f98a2SYan Zheng * update all the cowonly tree roots on disk 1318d352ac68SChris Mason */ 13197e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13200f7d52f4SChris Mason { 13217e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 13220f7d52f4SChris Mason struct btrfs_root *gang[8]; 13230f7d52f4SChris Mason int i; 13240f7d52f4SChris Mason int ret; 132554aa1f4dSChris Mason int err = 0; 132654aa1f4dSChris Mason 1327a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 13280f7d52f4SChris Mason while (1) { 13295d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 13305d4f98a2SYan Zheng (void **)gang, 0, 13310f7d52f4SChris Mason ARRAY_SIZE(gang), 13320f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 13330f7d52f4SChris Mason if (ret == 0) 13340f7d52f4SChris Mason break; 13350f7d52f4SChris Mason for (i = 0; i < ret; i++) { 13365b4aacefSJeff Mahoney struct btrfs_root *root = gang[i]; 13375d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 13382619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13390f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1340a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 134131153d81SYan Zheng 1342e02119d5SChris Mason btrfs_free_log(trans, root); 13435d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 1344e02119d5SChris Mason 134582d5902dSLi Zefan btrfs_save_ino_cache(root, trans); 134682d5902dSLi Zefan 1347f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 134827cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1349c7548af6SChris Mason smp_mb__after_atomic(); 1350f1ebcc74SLiu Bo 1351978d910dSYan Zheng if (root->commit_root != root->node) { 13529e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13539e351cc8SJosef Bacik &trans->transaction->switch_commits); 1354978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 1355978d910dSYan Zheng root->node); 1356978d910dSYan Zheng } 135731153d81SYan Zheng 13585d4f98a2SYan Zheng err = btrfs_update_root(trans, fs_info->tree_root, 13590f7d52f4SChris Mason &root->root_key, 13600f7d52f4SChris Mason &root->root_item); 1361a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 136254aa1f4dSChris Mason if (err) 136354aa1f4dSChris Mason break; 1364733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 13650f7d52f4SChris Mason } 13669f3a7427SChris Mason } 1367a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 136854aa1f4dSChris Mason return err; 13690f7d52f4SChris Mason } 13700f7d52f4SChris Mason 1371d352ac68SChris Mason /* 1372de78b51aSEric Sandeen * defrag a given btree. 1373de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1374d352ac68SChris Mason */ 1375de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1376e9d0b13bSChris Mason { 1377e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1378e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 13798929ecfaSYan, Zheng int ret; 1380e9d0b13bSChris Mason 138127cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1382e9d0b13bSChris Mason return 0; 13838929ecfaSYan, Zheng 13846b80053dSChris Mason while (1) { 13858929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 13868929ecfaSYan, Zheng if (IS_ERR(trans)) 13878929ecfaSYan, Zheng return PTR_ERR(trans); 13888929ecfaSYan, Zheng 1389de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 13908929ecfaSYan, Zheng 13913a45bb20SJeff Mahoney btrfs_end_transaction(trans); 13922ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1393e9d0b13bSChris Mason cond_resched(); 1394e9d0b13bSChris Mason 1395ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1396e9d0b13bSChris Mason break; 1397210549ebSDavid Sterba 1398ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1399ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1400210549ebSDavid Sterba ret = -EAGAIN; 1401210549ebSDavid Sterba break; 1402210549ebSDavid Sterba } 1403e9d0b13bSChris Mason } 140427cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14058929ecfaSYan, Zheng return ret; 1406e9d0b13bSChris Mason } 1407e9d0b13bSChris Mason 1408d352ac68SChris Mason /* 14096426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14106426c7adSQu Wenruo * 14116426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14126426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 14136426c7adSQu Wenruo * qgroup works. 14146426c7adSQu Wenruo */ 14156426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 14166426c7adSQu Wenruo struct btrfs_root *src, 14176426c7adSQu Wenruo struct btrfs_root *parent, 14186426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 14196426c7adSQu Wenruo u64 dst_objectid) 14206426c7adSQu Wenruo { 14216426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 14226426c7adSQu Wenruo int ret; 14236426c7adSQu Wenruo 14246426c7adSQu Wenruo /* 14256426c7adSQu Wenruo * Save some performance in the case that qgroups are not 14266426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 14276426c7adSQu Wenruo * kick in anyway. 14286426c7adSQu Wenruo */ 14299ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 14306426c7adSQu Wenruo return 0; 14316426c7adSQu Wenruo 14326426c7adSQu Wenruo /* 143352042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 14344d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 14354d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 14364d31778aSQu Wenruo * item. 14374d31778aSQu Wenruo */ 14384d31778aSQu Wenruo record_root_in_trans(trans, src, 1); 14394d31778aSQu Wenruo 14404d31778aSQu Wenruo /* 14416426c7adSQu Wenruo * We are going to commit transaction, see btrfs_commit_transaction() 14426426c7adSQu Wenruo * comment for reason locking tree_log_mutex 14436426c7adSQu Wenruo */ 14446426c7adSQu Wenruo mutex_lock(&fs_info->tree_log_mutex); 14456426c7adSQu Wenruo 14467e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 14476426c7adSQu Wenruo if (ret) 14486426c7adSQu Wenruo goto out; 1449460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 14506426c7adSQu Wenruo if (ret < 0) 14516426c7adSQu Wenruo goto out; 14526426c7adSQu Wenruo 14536426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1454a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 14556426c7adSQu Wenruo inherit); 14566426c7adSQu Wenruo if (ret < 0) 14576426c7adSQu Wenruo goto out; 14586426c7adSQu Wenruo 14596426c7adSQu Wenruo /* 14606426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 14616426c7adSQu Wenruo * 1) commit all subvolume and extent tree 14626426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 14636426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 14646426c7adSQu Wenruo * 2) write all btree blocks onto disk 14656426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 14666426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 14676426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 14686426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 14696426c7adSQu Wenruo * And we don't write super to avoid half committed status. 14706426c7adSQu Wenruo */ 14719386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 14726426c7adSQu Wenruo if (ret) 14736426c7adSQu Wenruo goto out; 1474889bfa39SJosef Bacik switch_commit_roots(trans); 147570458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 14766426c7adSQu Wenruo if (ret) 1477f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 14786426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 14796426c7adSQu Wenruo 14806426c7adSQu Wenruo out: 14816426c7adSQu Wenruo mutex_unlock(&fs_info->tree_log_mutex); 14826426c7adSQu Wenruo 14836426c7adSQu Wenruo /* 14846426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 14856426c7adSQu Wenruo * last_trans == cur_transid. 14866426c7adSQu Wenruo * Or it won't be committed again onto disk after later 14876426c7adSQu Wenruo * insert_dir_item() 14886426c7adSQu Wenruo */ 14896426c7adSQu Wenruo if (!ret) 14906426c7adSQu Wenruo record_root_in_trans(trans, parent, 1); 14916426c7adSQu Wenruo return ret; 14926426c7adSQu Wenruo } 14936426c7adSQu Wenruo 14946426c7adSQu Wenruo /* 1495d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1496aec8030aSMiao Xie * transaction commit. This does the actual creation. 1497aec8030aSMiao Xie * 1498aec8030aSMiao Xie * Note: 1499aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1500aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1501aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1502d352ac68SChris Mason */ 150380b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 15043063d29fSChris Mason struct btrfs_pending_snapshot *pending) 15053063d29fSChris Mason { 150608d50ca3SNikolay Borisov 150708d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 15083063d29fSChris Mason struct btrfs_key key; 150980b6794dSChris Mason struct btrfs_root_item *new_root_item; 15103063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 15113063d29fSChris Mason struct btrfs_root *root = pending->root; 15126bdb72deSSage Weil struct btrfs_root *parent_root; 151398c9942aSLiu Bo struct btrfs_block_rsv *rsv; 15146bdb72deSSage Weil struct inode *parent_inode; 151542874b3dSMiao Xie struct btrfs_path *path; 151642874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1517a22285a6SYan, Zheng struct dentry *dentry; 15183063d29fSChris Mason struct extent_buffer *tmp; 1519925baeddSChris Mason struct extent_buffer *old; 152095582b00SDeepa Dinamani struct timespec64 cur_time; 1521aec8030aSMiao Xie int ret = 0; 1522d68fc57bSYan, Zheng u64 to_reserve = 0; 15236bdb72deSSage Weil u64 index = 0; 1524a22285a6SYan, Zheng u64 objectid; 1525b83cc969SLi Zefan u64 root_flags; 15263063d29fSChris Mason 15278546b570SDavid Sterba ASSERT(pending->path); 15288546b570SDavid Sterba path = pending->path; 152942874b3dSMiao Xie 1530b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1531b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1532a22285a6SYan, Zheng 1533aec8030aSMiao Xie pending->error = btrfs_find_free_objectid(tree_root, &objectid); 1534aec8030aSMiao Xie if (pending->error) 15356fa9700eSMiao Xie goto no_free_objectid; 15363063d29fSChris Mason 1537d6726335SQu Wenruo /* 1538d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1539d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1540d6726335SQu Wenruo */ 1541d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1542d6726335SQu Wenruo 1543147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1544d68fc57bSYan, Zheng 1545d68fc57bSYan, Zheng if (to_reserve > 0) { 1546aec8030aSMiao Xie pending->error = btrfs_block_rsv_add(root, 1547aec8030aSMiao Xie &pending->block_rsv, 154808e007d2SMiao Xie to_reserve, 154908e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1550aec8030aSMiao Xie if (pending->error) 1551d6726335SQu Wenruo goto clear_skip_qgroup; 1552d68fc57bSYan, Zheng } 1553d68fc57bSYan, Zheng 15543063d29fSChris Mason key.objectid = objectid; 1555a22285a6SYan, Zheng key.offset = (u64)-1; 1556a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 15573063d29fSChris Mason 15586fa9700eSMiao Xie rsv = trans->block_rsv; 1559a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 15602382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 15610b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 156288d3a5aaSJosef Bacik trans->transid, 156388d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1564a22285a6SYan, Zheng dentry = pending->dentry; 1565e9662f70SMiao Xie parent_inode = pending->dir; 1566a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 15676426c7adSQu Wenruo record_root_in_trans(trans, parent_root, 0); 1568a22285a6SYan, Zheng 1569c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 157004b285f3SDeepa Dinamani 15716bdb72deSSage Weil /* 15726bdb72deSSage Weil * insert the directory item 15736bdb72deSSage Weil */ 1574877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 157549b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 157642874b3dSMiao Xie 157742874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 157842874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 15794a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 158042874b3dSMiao Xie dentry->d_name.name, 158142874b3dSMiao Xie dentry->d_name.len, 0); 158242874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1583fe66a05aSChris Mason pending->error = -EEXIST; 1584aec8030aSMiao Xie goto dir_item_existed; 158542874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 158642874b3dSMiao Xie ret = PTR_ERR(dir_item); 158766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 15888732d44fSMiao Xie goto fail; 158979787eaaSJeff Mahoney } 159042874b3dSMiao Xie btrfs_release_path(path); 15916bdb72deSSage Weil 1592e999376fSChris Mason /* 1593e999376fSChris Mason * pull in the delayed directory update 1594e999376fSChris Mason * and the delayed inode item 1595e999376fSChris Mason * otherwise we corrupt the FS during 1596e999376fSChris Mason * snapshot 1597e999376fSChris Mason */ 1598e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 15998732d44fSMiao Xie if (ret) { /* Transaction aborted */ 160066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16018732d44fSMiao Xie goto fail; 16028732d44fSMiao Xie } 1603e999376fSChris Mason 16046426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 16056bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 16066bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 160708fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 16086bdb72deSSage Weil 1609b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1610b83cc969SLi Zefan if (pending->readonly) 1611b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1612b83cc969SLi Zefan else 1613b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1614b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1615b83cc969SLi Zefan 16168ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 16178ea05e3aSAlexander Block trans->transid); 1618807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 16198ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 16208ea05e3aSAlexander Block BTRFS_UUID_SIZE); 162170023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 162270023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 162370023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 16248ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 16258ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 16268ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 16278ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 162870023da2SStefan Behrens } 16293cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 16303cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 163170023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 16328ea05e3aSAlexander Block 1633925baeddSChris Mason old = btrfs_lock_root_node(root); 16349631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 16359631e4ccSJosef Bacik BTRFS_NESTING_COW); 163679787eaaSJeff Mahoney if (ret) { 163779787eaaSJeff Mahoney btrfs_tree_unlock(old); 163879787eaaSJeff Mahoney free_extent_buffer(old); 163966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16408732d44fSMiao Xie goto fail; 164179787eaaSJeff Mahoney } 164249b25e05SJeff Mahoney 164349b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 164479787eaaSJeff Mahoney /* clean up in any case */ 1645925baeddSChris Mason btrfs_tree_unlock(old); 1646925baeddSChris Mason free_extent_buffer(old); 16478732d44fSMiao Xie if (ret) { 164866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16498732d44fSMiao Xie goto fail; 16508732d44fSMiao Xie } 1651f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 165227cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1653f1ebcc74SLiu Bo smp_wmb(); 1654f1ebcc74SLiu Bo 16555d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1656a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1657a22285a6SYan, Zheng key.offset = trans->transid; 1658a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1659925baeddSChris Mason btrfs_tree_unlock(tmp); 16603063d29fSChris Mason free_extent_buffer(tmp); 16618732d44fSMiao Xie if (ret) { 166266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16638732d44fSMiao Xie goto fail; 16648732d44fSMiao Xie } 16650660b5afSChris Mason 1666a22285a6SYan, Zheng /* 1667a22285a6SYan, Zheng * insert root back/forward references 1668a22285a6SYan, Zheng */ 16696025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1670a22285a6SYan, Zheng parent_root->root_key.objectid, 16714a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1672a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 16738732d44fSMiao Xie if (ret) { 167466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16758732d44fSMiao Xie goto fail; 16768732d44fSMiao Xie } 1677a22285a6SYan, Zheng 1678a22285a6SYan, Zheng key.offset = (u64)-1; 16792dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 168079787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 168179787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 16822d892ccdSFilipe Manana pending->snap = NULL; 168366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16848732d44fSMiao Xie goto fail; 168579787eaaSJeff Mahoney } 1686d68fc57bSYan, Zheng 168749b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 16888732d44fSMiao Xie if (ret) { 168966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16908732d44fSMiao Xie goto fail; 16918732d44fSMiao Xie } 1692361048f5SMiao Xie 1693c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 16948732d44fSMiao Xie if (ret) { 169566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16968732d44fSMiao Xie goto fail; 16978732d44fSMiao Xie } 169842874b3dSMiao Xie 16996426c7adSQu Wenruo /* 17006426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17016426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17026426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 17036426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 17046426c7adSQu Wenruo */ 17056426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 17066426c7adSQu Wenruo pending->inherit, objectid); 17076426c7adSQu Wenruo if (ret < 0) 17086426c7adSQu Wenruo goto fail; 17096426c7adSQu Wenruo 1710684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1711684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1712684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 171342874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 17149c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 17158732d44fSMiao Xie if (ret) { 171666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17178732d44fSMiao Xie goto fail; 17188732d44fSMiao Xie } 171942874b3dSMiao Xie 17206ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 172142874b3dSMiao Xie dentry->d_name.len * 2); 172204b285f3SDeepa Dinamani parent_inode->i_mtime = parent_inode->i_ctime = 1723c2050a45SDeepa Dinamani current_time(parent_inode); 1724be6aef60SJosef Bacik ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); 1725dd5f9615SStefan Behrens if (ret) { 172666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1727dd5f9615SStefan Behrens goto fail; 1728dd5f9615SStefan Behrens } 1729807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1730807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1731cdb345a8SLu Fengqi objectid); 1732dd5f9615SStefan Behrens if (ret) { 173366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1734dd5f9615SStefan Behrens goto fail; 1735dd5f9615SStefan Behrens } 1736dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1737cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1738dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1739dd5f9615SStefan Behrens objectid); 1740dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 174166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1742dd5f9615SStefan Behrens goto fail; 1743dd5f9615SStefan Behrens } 1744dd5f9615SStefan Behrens } 1745d6726335SQu Wenruo 1746c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1747d6726335SQu Wenruo if (ret) { 174866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1749d6726335SQu Wenruo goto fail; 1750d6726335SQu Wenruo } 1751d6726335SQu Wenruo 17523063d29fSChris Mason fail: 1753aec8030aSMiao Xie pending->error = ret; 1754aec8030aSMiao Xie dir_item_existed: 175598c9942aSLiu Bo trans->block_rsv = rsv; 17562382c5ccSLiu Bo trans->bytes_reserved = 0; 1757d6726335SQu Wenruo clear_skip_qgroup: 1758d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 17596fa9700eSMiao Xie no_free_objectid: 17606fa9700eSMiao Xie kfree(new_root_item); 1761b0c0ea63SDavid Sterba pending->root_item = NULL; 176242874b3dSMiao Xie btrfs_free_path(path); 17638546b570SDavid Sterba pending->path = NULL; 17648546b570SDavid Sterba 176549b25e05SJeff Mahoney return ret; 17663063d29fSChris Mason } 17673063d29fSChris Mason 1768d352ac68SChris Mason /* 1769d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1770d352ac68SChris Mason */ 177108d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 17723063d29fSChris Mason { 1773aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 17743063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1775aec8030aSMiao Xie int ret = 0; 17763de4586cSChris Mason 1777aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1778aec8030aSMiao Xie list_del(&pending->list); 177908d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1780aec8030aSMiao Xie if (ret) 1781aec8030aSMiao Xie break; 1782aec8030aSMiao Xie } 1783aec8030aSMiao Xie return ret; 17843de4586cSChris Mason } 17853de4586cSChris Mason 17862ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 17875d4f98a2SYan Zheng { 17885d4f98a2SYan Zheng struct btrfs_root_item *root_item; 17895d4f98a2SYan Zheng struct btrfs_super_block *super; 17905d4f98a2SYan Zheng 17910b246afaSJeff Mahoney super = fs_info->super_copy; 17925d4f98a2SYan Zheng 17930b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1794093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1795093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1796093e037cSDavid Sterba super->chunk_root_level = root_item->level; 17975d4f98a2SYan Zheng 17980b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1799093e037cSDavid Sterba super->root = root_item->bytenr; 1800093e037cSDavid Sterba super->generation = root_item->generation; 1801093e037cSDavid Sterba super->root_level = root_item->level; 18020b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1803093e037cSDavid Sterba super->cache_generation = root_item->generation; 18040b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1805093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18065d4f98a2SYan Zheng } 18075d4f98a2SYan Zheng 1808f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1809f36f3042SChris Mason { 18104a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1811f36f3042SChris Mason int ret = 0; 18124a9d8bdeSMiao Xie 1813a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18144a9d8bdeSMiao Xie trans = info->running_transaction; 18154a9d8bdeSMiao Xie if (trans) 18164a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1817a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1818f36f3042SChris Mason return ret; 1819f36f3042SChris Mason } 1820f36f3042SChris Mason 18218929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 18228929ecfaSYan, Zheng { 18234a9d8bdeSMiao Xie struct btrfs_transaction *trans; 18248929ecfaSYan, Zheng int ret = 0; 18254a9d8bdeSMiao Xie 1826a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18274a9d8bdeSMiao Xie trans = info->running_transaction; 18284a9d8bdeSMiao Xie if (trans) 18294a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1830a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 18318929ecfaSYan, Zheng return ret; 18328929ecfaSYan, Zheng } 18338929ecfaSYan, Zheng 1834bb9c12c9SSage Weil /* 1835bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1836bb9c12c9SSage Weil * transaction joins 1837bb9c12c9SSage Weil */ 18382ff7e61eSJeff Mahoney static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, 1839bb9c12c9SSage Weil struct btrfs_transaction *trans) 1840bb9c12c9SSage Weil { 18412ff7e61eSJeff Mahoney wait_event(fs_info->transaction_blocked_wait, 1842bf31f87fSDavid Sterba trans->state >= TRANS_STATE_COMMIT_START || 1843bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1844bb9c12c9SSage Weil } 1845bb9c12c9SSage Weil 1846bb9c12c9SSage Weil /* 1847bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1848bb9c12c9SSage Weil * caller holds ref. 1849bb9c12c9SSage Weil */ 18502ff7e61eSJeff Mahoney static void wait_current_trans_commit_start_and_unblock( 18512ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 1852bb9c12c9SSage Weil struct btrfs_transaction *trans) 1853bb9c12c9SSage Weil { 18542ff7e61eSJeff Mahoney wait_event(fs_info->transaction_wait, 1855bf31f87fSDavid Sterba trans->state >= TRANS_STATE_UNBLOCKED || 1856bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1857bb9c12c9SSage Weil } 1858bb9c12c9SSage Weil 1859bb9c12c9SSage Weil /* 1860bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1861bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1862bb9c12c9SSage Weil */ 1863bb9c12c9SSage Weil struct btrfs_async_commit { 1864bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 18657892b5afSMiao Xie struct work_struct work; 1866bb9c12c9SSage Weil }; 1867bb9c12c9SSage Weil 1868bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1869bb9c12c9SSage Weil { 1870bb9c12c9SSage Weil struct btrfs_async_commit *ac = 18717892b5afSMiao Xie container_of(work, struct btrfs_async_commit, work); 1872bb9c12c9SSage Weil 18736fc4e354SSage Weil /* 18746fc4e354SSage Weil * We've got freeze protection passed with the transaction. 18756fc4e354SSage Weil * Tell lockdep about it. 18766fc4e354SSage Weil */ 1877b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 18783a45bb20SJeff Mahoney __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); 18796fc4e354SSage Weil 1880e209db7aSSage Weil current->journal_info = ac->newtrans; 1881e209db7aSSage Weil 18823a45bb20SJeff Mahoney btrfs_commit_transaction(ac->newtrans); 1883bb9c12c9SSage Weil kfree(ac); 1884bb9c12c9SSage Weil } 1885bb9c12c9SSage Weil 1886bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1887bb9c12c9SSage Weil int wait_for_unblock) 1888bb9c12c9SSage Weil { 18893a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1890bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1891bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1892bb9c12c9SSage Weil 1893bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1894db5b493aSTsutomu Itoh if (!ac) 1895db5b493aSTsutomu Itoh return -ENOMEM; 1896bb9c12c9SSage Weil 18977892b5afSMiao Xie INIT_WORK(&ac->work, do_async_commit); 18983a45bb20SJeff Mahoney ac->newtrans = btrfs_join_transaction(trans->root); 18993612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 19003612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 19013612b495STsutomu Itoh kfree(ac); 19023612b495STsutomu Itoh return err; 19033612b495STsutomu Itoh } 1904bb9c12c9SSage Weil 1905bb9c12c9SSage Weil /* take transaction reference */ 1906bb9c12c9SSage Weil cur_trans = trans->transaction; 19079b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1908bb9c12c9SSage Weil 19093a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19106fc4e354SSage Weil 19116fc4e354SSage Weil /* 19126fc4e354SSage Weil * Tell lockdep we've released the freeze rwsem, since the 19136fc4e354SSage Weil * async commit thread will be the one to unlock it. 19146fc4e354SSage Weil */ 1915b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 19160b246afaSJeff Mahoney __sb_writers_release(fs_info->sb, SB_FREEZE_FS); 19176fc4e354SSage Weil 19187892b5afSMiao Xie schedule_work(&ac->work); 1919bb9c12c9SSage Weil 1920bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1921bb9c12c9SSage Weil if (wait_for_unblock) 19222ff7e61eSJeff Mahoney wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); 1923bb9c12c9SSage Weil else 19242ff7e61eSJeff Mahoney wait_current_trans_commit_start(fs_info, cur_trans); 1925bb9c12c9SSage Weil 192638e88054SSage Weil if (current->journal_info == trans) 192738e88054SSage Weil current->journal_info = NULL; 192838e88054SSage Weil 1929724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1930bb9c12c9SSage Weil return 0; 1931bb9c12c9SSage Weil } 1932bb9c12c9SSage Weil 193349b25e05SJeff Mahoney 193497cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 193549b25e05SJeff Mahoney { 193697cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 193749b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 193849b25e05SJeff Mahoney 1939b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 194049b25e05SJeff Mahoney 194166642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19427b8b92afSJosef Bacik 19430b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 194466b6135bSLiu Bo 194525d8c284SMiao Xie /* 194625d8c284SMiao Xie * If the transaction is removed from the list, it means this 194725d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 194825d8c284SMiao Xie * to call the cleanup function. 194925d8c284SMiao Xie */ 195025d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 195166b6135bSLiu Bo 195249b25e05SJeff Mahoney list_del_init(&cur_trans->list); 19530b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19544a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19550b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1956f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1957f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1958f094ac32SLiu Bo 19590b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1960d7096fc3SJosef Bacik } 19610b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 196249b25e05SJeff Mahoney 19632ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 196449b25e05SJeff Mahoney 19650b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 19660b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 19670b246afaSJeff Mahoney fs_info->running_transaction = NULL; 19680b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 19694a9d8bdeSMiao Xie 1970e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 19710b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 1972724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1973724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 197449b25e05SJeff Mahoney 197597cb39bbSNikolay Borisov trace_btrfs_transaction_commit(trans->root); 197649b25e05SJeff Mahoney 197749b25e05SJeff Mahoney if (current->journal_info == trans) 197849b25e05SJeff Mahoney current->journal_info = NULL; 19790b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 198049b25e05SJeff Mahoney 198149b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 198249b25e05SJeff Mahoney } 198349b25e05SJeff Mahoney 1984c7cc64a9SDavid Sterba /* 1985c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 1986c7cc64a9SDavid Sterba * transaction and remove them from the list 1987c7cc64a9SDavid Sterba */ 1988c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 1989c7cc64a9SDavid Sterba { 1990c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 199132da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 1992c7cc64a9SDavid Sterba 1993c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 1994c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 1995c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 1996c7cc64a9SDavid Sterba } 1997c7cc64a9SDavid Sterba } 1998c7cc64a9SDavid Sterba 199988090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 200082436617SMiao Xie { 2001ce8ea7ccSJosef Bacik /* 2002ce8ea7ccSJosef Bacik * We use writeback_inodes_sb here because if we used 2003ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2004ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2005ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2006ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2007ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2008ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2009ce8ea7ccSJosef Bacik */ 201088090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2011ce8ea7ccSJosef Bacik writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 201282436617SMiao Xie return 0; 201382436617SMiao Xie } 201482436617SMiao Xie 201588090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 201682436617SMiao Xie { 201788090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20186374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 201982436617SMiao Xie } 202082436617SMiao Xie 20213a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 202279154b1bSChris Mason { 20233a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 202449b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20258fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 202625287e0aSMiao Xie int ret; 202779154b1bSChris Mason 202835b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 202935b814f3SNikolay Borisov 2030d62b23c9SJosef Bacik /* 2031d62b23c9SJosef Bacik * Some places just start a transaction to commit it. We need to make 2032d62b23c9SJosef Bacik * sure that if this commit fails that the abort code actually marks the 2033d62b23c9SJosef Bacik * transaction as failed, so set trans->dirty to make the abort code do 2034d62b23c9SJosef Bacik * the right thing. 2035d62b23c9SJosef Bacik */ 2036d62b23c9SJosef Bacik trans->dirty = true; 2037d62b23c9SJosef Bacik 20388d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2039bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 204025287e0aSMiao Xie ret = cur_trans->aborted; 20413a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2042e4a2bcacSJosef Bacik return ret; 204325287e0aSMiao Xie } 204449b25e05SJeff Mahoney 2045f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2046f45c752bSJosef Bacik trans->block_rsv = NULL; 2047f45c752bSJosef Bacik 204856bec294SChris Mason /* make a pass through all the delayed refs we have so far 204956bec294SChris Mason * any runnings procs may add more while we are here 205056bec294SChris Mason */ 2051c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 2052e4a2bcacSJosef Bacik if (ret) { 20533a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2054e4a2bcacSJosef Bacik return ret; 2055e4a2bcacSJosef Bacik } 205656bec294SChris Mason 2057b7ec40d7SChris Mason cur_trans = trans->transaction; 205849b25e05SJeff Mahoney 205956bec294SChris Mason /* 206056bec294SChris Mason * set the flushing flag so procs in this transaction have to 206156bec294SChris Mason * start sending their work down. 206256bec294SChris Mason */ 2063b7ec40d7SChris Mason cur_trans->delayed_refs.flushing = 1; 20641be41b78SJosef Bacik smp_wmb(); 206556bec294SChris Mason 20666c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2067ea658badSJosef Bacik 2068c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 2069e4a2bcacSJosef Bacik if (ret) { 20703a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2071e4a2bcacSJosef Bacik return ret; 2072e4a2bcacSJosef Bacik } 207356bec294SChris Mason 20743204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 20751bbc621eSChris Mason int run_it = 0; 20761bbc621eSChris Mason 20771bbc621eSChris Mason /* this mutex is also taken before trying to set 20781bbc621eSChris Mason * block groups readonly. We need to make sure 20791bbc621eSChris Mason * that nobody has set a block group readonly 20801bbc621eSChris Mason * after a extents from that block group have been 20811bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 20821bbc621eSChris Mason * will wait for the transaction to commit if it 20833204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 20841bbc621eSChris Mason * 20853204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 20863204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 20871bbc621eSChris Mason * hurt to have more than one go through, but there's no 20881bbc621eSChris Mason * real advantage to it either. 20891bbc621eSChris Mason */ 20900b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 20913204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 20923204d33cSJosef Bacik &cur_trans->flags)) 20931bbc621eSChris Mason run_it = 1; 20940b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 20951bbc621eSChris Mason 2096f9cacae3SNikolay Borisov if (run_it) { 209721217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 20981bbc621eSChris Mason if (ret) { 20993a45bb20SJeff Mahoney btrfs_end_transaction(trans); 21001bbc621eSChris Mason return ret; 21011bbc621eSChris Mason } 2102f9cacae3SNikolay Borisov } 2103f9cacae3SNikolay Borisov } 21041bbc621eSChris Mason 21050b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21064a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 21070b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21089b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 21093a45bb20SJeff Mahoney ret = btrfs_end_transaction(trans); 2110ccd467d6SChris Mason 21112ff7e61eSJeff Mahoney wait_for_commit(cur_trans); 211215ee9bc7SJosef Bacik 2113bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2114b4924a0fSLiu Bo ret = cur_trans->aborted; 2115b4924a0fSLiu Bo 2116724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 211715ee9bc7SJosef Bacik 211849b25e05SJeff Mahoney return ret; 211979154b1bSChris Mason } 21204313b399SChris Mason 21214a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21220b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 2123bb9c12c9SSage Weil 21240b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2125ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2126ccd467d6SChris Mason struct btrfs_transaction, list); 21274a9d8bdeSMiao Xie if (prev_trans->state != TRANS_STATE_COMPLETED) { 21289b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 21290b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2130ccd467d6SChris Mason 21312ff7e61eSJeff Mahoney wait_for_commit(prev_trans); 2132bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2133ccd467d6SChris Mason 2134724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 21351f9b8c8fSFilipe Manana if (ret) 21361f9b8c8fSFilipe Manana goto cleanup_transaction; 2137a4abeea4SJosef Bacik } else { 21380b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2139ccd467d6SChris Mason } 2140a4abeea4SJosef Bacik } else { 21410b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2142cb2d3dadSFilipe Manana /* 2143cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2144cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2145cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2146cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2147cb2d3dadSFilipe Manana */ 2148cb2d3dadSFilipe Manana if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { 2149cb2d3dadSFilipe Manana ret = -EROFS; 2150cb2d3dadSFilipe Manana goto cleanup_transaction; 2151cb2d3dadSFilipe Manana } 2152ccd467d6SChris Mason } 215315ee9bc7SJosef Bacik 21540860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 21550860adfdSMiao Xie 215688090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 215782436617SMiao Xie if (ret) 215882436617SMiao Xie goto cleanup_transaction; 215982436617SMiao Xie 2160e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 216149b25e05SJeff Mahoney if (ret) 216249b25e05SJeff Mahoney goto cleanup_transaction; 216316cdcec7SMiao Xie 2164581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2165581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2166ed3b3d31SChris Mason 2167581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2168e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2169ca469637SMiao Xie if (ret) 2170ca469637SMiao Xie goto cleanup_transaction; 2171ca469637SMiao Xie 217288090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2173cb7ab021SWang Shilong 217448778179SFilipe Manana /* 217548778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 217648778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 217748778179SFilipe Manana * extents complete we lose logged data after a power failure. 217848778179SFilipe Manana */ 217948778179SFilipe Manana wait_event(cur_trans->pending_wait, 218048778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 218148778179SFilipe Manana 21822ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2183ed0ca140SJosef Bacik /* 2184ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2185ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 21864a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2187ed0ca140SJosef Bacik */ 21880b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21894a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 21900b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2191ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2192ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 219315ee9bc7SJosef Bacik 2194bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 21952cba30f1SMiao Xie ret = cur_trans->aborted; 21966cf7f77eSWang Shilong goto scrub_continue; 21972cba30f1SMiao Xie } 21987585717fSChris Mason /* 21997585717fSChris Mason * the reloc mutex makes sure that we stop 22007585717fSChris Mason * the balancing code from coming in and moving 22017585717fSChris Mason * extents around in the middle of the commit 22027585717fSChris Mason */ 22030b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 22047585717fSChris Mason 220542874b3dSMiao Xie /* 220642874b3dSMiao Xie * We needn't worry about the delayed items because we will 220742874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 220842874b3dSMiao Xie * core function of the snapshot creation. 220942874b3dSMiao Xie */ 221008d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 221156e9f6eaSDavid Sterba if (ret) 221256e9f6eaSDavid Sterba goto unlock_reloc; 22133063d29fSChris Mason 221442874b3dSMiao Xie /* 221542874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 221642874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 221742874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 221842874b3dSMiao Xie * them. 221942874b3dSMiao Xie * 222042874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 222142874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 222242874b3dSMiao Xie * the nodes and leaves. 222342874b3dSMiao Xie */ 2224e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 222556e9f6eaSDavid Sterba if (ret) 222656e9f6eaSDavid Sterba goto unlock_reloc; 222716cdcec7SMiao Xie 2228c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 222956e9f6eaSDavid Sterba if (ret) 223056e9f6eaSDavid Sterba goto unlock_reloc; 223156bec294SChris Mason 2232e999376fSChris Mason /* 2233e999376fSChris Mason * make sure none of the code above managed to slip in a 2234e999376fSChris Mason * delayed item 2235e999376fSChris Mason */ 2236ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2237e999376fSChris Mason 22382c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2239dc17ff8fSChris Mason 2240e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 2241e02119d5SChris Mason * various roots consistent with each other. Every pointer 2242e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 2243e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 2244e02119d5SChris Mason * the tree logging code from jumping in and changing any 2245e02119d5SChris Mason * of the trees. 2246e02119d5SChris Mason * 2247e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 2248e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 2249e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 2250e02119d5SChris Mason * from now until after the super is written, we avoid races 2251e02119d5SChris Mason * with the tree-log code. 2252e02119d5SChris Mason */ 22530b246afaSJeff Mahoney mutex_lock(&fs_info->tree_log_mutex); 22541a40e23bSZheng Yan 22557e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 225656e9f6eaSDavid Sterba if (ret) 225756e9f6eaSDavid Sterba goto unlock_tree_log; 225854aa1f4dSChris Mason 22593818aea2SQu Wenruo /* 22607e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 22617e1876acSDavid Sterba * before the next transaction. 22623818aea2SQu Wenruo */ 22630b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 22643818aea2SQu Wenruo 22655d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2266e02119d5SChris Mason * safe to free the root of tree log roots 2267e02119d5SChris Mason */ 22680b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2269e02119d5SChris Mason 22700ed4792aSQu Wenruo /* 227182bafb38SQu Wenruo * commit_fs_roots() can call btrfs_save_ino_cache(), which generates 227282bafb38SQu Wenruo * new delayed refs. Must handle them or qgroup can be wrong. 227382bafb38SQu Wenruo */ 2274c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 227556e9f6eaSDavid Sterba if (ret) 227656e9f6eaSDavid Sterba goto unlock_tree_log; 227782bafb38SQu Wenruo 227882bafb38SQu Wenruo /* 22790ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 22800ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 22810ed4792aSQu Wenruo */ 2282460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 228356e9f6eaSDavid Sterba if (ret < 0) 228456e9f6eaSDavid Sterba goto unlock_tree_log; 22850ed4792aSQu Wenruo 22869386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 228756e9f6eaSDavid Sterba if (ret) 228856e9f6eaSDavid Sterba goto unlock_tree_log; 228954aa1f4dSChris Mason 22902cba30f1SMiao Xie /* 22912cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 22922cba30f1SMiao Xie * update ->aborted, check it. 22932cba30f1SMiao Xie */ 2294bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22952cba30f1SMiao Xie ret = cur_trans->aborted; 229656e9f6eaSDavid Sterba goto unlock_tree_log; 22972cba30f1SMiao Xie } 22982cba30f1SMiao Xie 22990b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 23005f39d397SChris Mason 23010b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 23020b246afaSJeff Mahoney fs_info->tree_root->node); 23030b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 23049e351cc8SJosef Bacik &cur_trans->switch_commits); 23055d4f98a2SYan Zheng 23060b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 23070b246afaSJeff Mahoney fs_info->chunk_root->node); 23080b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 23099e351cc8SJosef Bacik &cur_trans->switch_commits); 23109e351cc8SJosef Bacik 2311889bfa39SJosef Bacik switch_commit_roots(trans); 23125d4f98a2SYan Zheng 2313ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 23141bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 23152ff7e61eSJeff Mahoney update_super_roots(fs_info); 2316e02119d5SChris Mason 23170b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 23180b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 23190b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 23200b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2321ccd467d6SChris Mason 2322bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2323935e5cc9SMiao Xie 23240b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 23250b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2326656f30dbSFilipe Manana 23274fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 23284fbcdf66SFilipe Manana 23290b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 23304a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 23310b246afaSJeff Mahoney fs_info->running_transaction = NULL; 23320b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 23330b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2334b7ec40d7SChris Mason 23350b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 2336e6dcd2dcSChris Mason 233770458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 233849b25e05SJeff Mahoney if (ret) { 23390b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 234008748810SDavid Sterba "Error while writing out transaction"); 234156e9f6eaSDavid Sterba /* 234256e9f6eaSDavid Sterba * reloc_mutex has been unlocked, tree_log_mutex is still held 234356e9f6eaSDavid Sterba * but we can't jump to unlock_tree_log causing double unlock 234456e9f6eaSDavid Sterba */ 23450b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 23466cf7f77eSWang Shilong goto scrub_continue; 234749b25e05SJeff Mahoney } 234849b25e05SJeff Mahoney 2349eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2350e02119d5SChris Mason /* 2351e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2352e02119d5SChris Mason * to go about their business 2353e02119d5SChris Mason */ 23540b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2355c1f32b7cSAnand Jain if (ret) 2356c1f32b7cSAnand Jain goto scrub_continue; 2357e02119d5SChris Mason 23585ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 23594313b399SChris Mason 23603204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 23610b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 236213212b54SZhao Lei 23630b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 23644a9d8bdeSMiao Xie /* 23654a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 23664a9d8bdeSMiao Xie * which can change it. 23674a9d8bdeSMiao Xie */ 23684a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 23692c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 23703de4586cSChris Mason 23710b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 237213c5a93eSJosef Bacik list_del_init(&cur_trans->list); 23730b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2374a4abeea4SJosef Bacik 2375724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2376724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 237758176a96SJosef Bacik 23780860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 23790b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2380b2b5ef5cSJan Kara 23813a45bb20SJeff Mahoney trace_btrfs_transaction_commit(trans->root); 23821abe9b8aSliubo 23832ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2384a2de733cSArne Jansen 23859ed74f2dSJosef Bacik if (current->journal_info == trans) 23869ed74f2dSJosef Bacik current->journal_info = NULL; 23879ed74f2dSJosef Bacik 23882c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 238924bbcf04SYan, Zheng 239079154b1bSChris Mason return ret; 239149b25e05SJeff Mahoney 239256e9f6eaSDavid Sterba unlock_tree_log: 239356e9f6eaSDavid Sterba mutex_unlock(&fs_info->tree_log_mutex); 239456e9f6eaSDavid Sterba unlock_reloc: 239556e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 23966cf7f77eSWang Shilong scrub_continue: 23972ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 239849b25e05SJeff Mahoney cleanup_transaction: 2399dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2400c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 24014fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24020e721106SJosef Bacik trans->block_rsv = NULL; 24030b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 240449b25e05SJeff Mahoney if (current->journal_info == trans) 240549b25e05SJeff Mahoney current->journal_info = NULL; 240697cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 240749b25e05SJeff Mahoney 240849b25e05SJeff Mahoney return ret; 240979154b1bSChris Mason } 241079154b1bSChris Mason 2411d352ac68SChris Mason /* 24129d1a2a3aSDavid Sterba * return < 0 if error 24139d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 24149d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 24159d1a2a3aSDavid Sterba * 24169d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 24179d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 24189d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 24199d1a2a3aSDavid Sterba * few seconds later. 2420d352ac68SChris Mason */ 24219d1a2a3aSDavid Sterba int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) 2422e9d0b13bSChris Mason { 24239d1a2a3aSDavid Sterba int ret; 24245d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 2425e9d0b13bSChris Mason 2426a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 24279d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 24289d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 24299d1a2a3aSDavid Sterba return 0; 24309d1a2a3aSDavid Sterba } 24319d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 24329d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2433cfad392bSJosef Bacik list_del_init(&root->root_list); 2434a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 24355d4f98a2SYan Zheng 24364fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 243776dda93cSYan, Zheng 243816cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 24390e996e7fSJosef Bacik if (root->ino_cache_inode) { 24400e996e7fSJosef Bacik iput(root->ino_cache_inode); 24410e996e7fSJosef Bacik root->ino_cache_inode = NULL; 24420e996e7fSJosef Bacik } 244316cdcec7SMiao Xie 244476dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 244576dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 24460078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 244776dda93cSYan, Zheng else 24480078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 244932471dc2SDavid Sterba 2450dc9492c1SJosef Bacik btrfs_put_root(root); 24516596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2452e9d0b13bSChris Mason } 2453572d9ab7SDavid Sterba 2454572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2455572d9ab7SDavid Sterba { 2456572d9ab7SDavid Sterba unsigned long prev; 2457572d9ab7SDavid Sterba unsigned long bit; 2458572d9ab7SDavid Sterba 24596c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2460572d9ab7SDavid Sterba if (!prev) 2461572d9ab7SDavid Sterba return; 2462572d9ab7SDavid Sterba 24637e1876acSDavid Sterba bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; 24647e1876acSDavid Sterba if (prev & bit) 24657e1876acSDavid Sterba btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); 24667e1876acSDavid Sterba prev &= ~bit; 24677e1876acSDavid Sterba 24687e1876acSDavid Sterba bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; 24697e1876acSDavid Sterba if (prev & bit) 24707e1876acSDavid Sterba btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); 24717e1876acSDavid Sterba prev &= ~bit; 24727e1876acSDavid Sterba 2473d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2474d51033d0SDavid Sterba if (prev & bit) 2475d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2476d51033d0SDavid Sterba prev &= ~bit; 2477d51033d0SDavid Sterba 2478572d9ab7SDavid Sterba if (prev) 2479572d9ab7SDavid Sterba btrfs_warn(fs_info, 2480572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2481572d9ab7SDavid Sterba } 2482