16cbd5570SChris Mason /* 26cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 36cbd5570SChris Mason * 46cbd5570SChris Mason * This program is free software; you can redistribute it and/or 56cbd5570SChris Mason * modify it under the terms of the GNU General Public 66cbd5570SChris Mason * License v2 as published by the Free Software Foundation. 76cbd5570SChris Mason * 86cbd5570SChris Mason * This program is distributed in the hope that it will be useful, 96cbd5570SChris Mason * but WITHOUT ANY WARRANTY; without even the implied warranty of 106cbd5570SChris Mason * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 116cbd5570SChris Mason * General Public License for more details. 126cbd5570SChris Mason * 136cbd5570SChris Mason * You should have received a copy of the GNU General Public 146cbd5570SChris Mason * License along with this program; if not, write to the 156cbd5570SChris Mason * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 166cbd5570SChris Mason * Boston, MA 021110-1307, USA. 176cbd5570SChris Mason */ 186cbd5570SChris Mason 1979154b1bSChris Mason #include <linux/fs.h> 205a0e3ad6STejun Heo #include <linux/slab.h> 2134088780SChris Mason #include <linux/sched.h> 22d3c2fdcfSChris Mason #include <linux/writeback.h> 235f39d397SChris Mason #include <linux/pagemap.h> 245f2cc086SChris Mason #include <linux/blkdev.h> 2579154b1bSChris Mason #include "ctree.h" 2679154b1bSChris Mason #include "disk-io.h" 2779154b1bSChris Mason #include "transaction.h" 28925baeddSChris Mason #include "locking.h" 29e02119d5SChris Mason #include "tree-log.h" 3079154b1bSChris Mason 310f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0 320f7d52f4SChris Mason 3380b6794dSChris Mason static noinline void put_transaction(struct btrfs_transaction *transaction) 3479154b1bSChris Mason { 352c90e5d6SChris Mason WARN_ON(transaction->use_count == 0); 3679154b1bSChris Mason transaction->use_count--; 3778fae27eSChris Mason if (transaction->use_count == 0) { 388fd17795SChris Mason list_del_init(&transaction->list); 392c90e5d6SChris Mason memset(transaction, 0, sizeof(*transaction)); 402c90e5d6SChris Mason kmem_cache_free(btrfs_transaction_cachep, transaction); 4179154b1bSChris Mason } 4278fae27eSChris Mason } 4379154b1bSChris Mason 44817d52f8SJosef Bacik static noinline void switch_commit_root(struct btrfs_root *root) 45817d52f8SJosef Bacik { 46817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 47817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 48817d52f8SJosef Bacik } 49817d52f8SJosef Bacik 50d352ac68SChris Mason /* 51d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 52d352ac68SChris Mason */ 5380b6794dSChris Mason static noinline int join_transaction(struct btrfs_root *root) 5479154b1bSChris Mason { 5579154b1bSChris Mason struct btrfs_transaction *cur_trans; 5679154b1bSChris Mason cur_trans = root->fs_info->running_transaction; 5779154b1bSChris Mason if (!cur_trans) { 582c90e5d6SChris Mason cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, 592c90e5d6SChris Mason GFP_NOFS); 60db5b493aSTsutomu Itoh if (!cur_trans) 61db5b493aSTsutomu Itoh return -ENOMEM; 620f7d52f4SChris Mason root->fs_info->generation++; 6315ee9bc7SJosef Bacik cur_trans->num_writers = 1; 6415ee9bc7SJosef Bacik cur_trans->num_joined = 0; 650f7d52f4SChris Mason cur_trans->transid = root->fs_info->generation; 6679154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 6779154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 6879154b1bSChris Mason cur_trans->in_commit = 0; 69f9295749SChris Mason cur_trans->blocked = 0; 70d5719762SChris Mason cur_trans->use_count = 1; 7179154b1bSChris Mason cur_trans->commit_done = 0; 7208607c1bSChris Mason cur_trans->start_time = get_seconds(); 7356bec294SChris Mason 746bef4d31SEric Paris cur_trans->delayed_refs.root = RB_ROOT; 7556bec294SChris Mason cur_trans->delayed_refs.num_entries = 0; 76c3e69d58SChris Mason cur_trans->delayed_refs.num_heads_ready = 0; 77c3e69d58SChris Mason cur_trans->delayed_refs.num_heads = 0; 7856bec294SChris Mason cur_trans->delayed_refs.flushing = 0; 79c3e69d58SChris Mason cur_trans->delayed_refs.run_delayed_start = 0; 8056bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 8156bec294SChris Mason 823063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 838fd17795SChris Mason list_add_tail(&cur_trans->list, &root->fs_info->trans_list); 84d1310b2eSChris Mason extent_io_tree_init(&cur_trans->dirty_pages, 855f39d397SChris Mason root->fs_info->btree_inode->i_mapping, 865f39d397SChris Mason GFP_NOFS); 8748ec2cf8SChris Mason spin_lock(&root->fs_info->new_trans_lock); 8848ec2cf8SChris Mason root->fs_info->running_transaction = cur_trans; 8948ec2cf8SChris Mason spin_unlock(&root->fs_info->new_trans_lock); 9015ee9bc7SJosef Bacik } else { 9179154b1bSChris Mason cur_trans->num_writers++; 9215ee9bc7SJosef Bacik cur_trans->num_joined++; 9315ee9bc7SJosef Bacik } 9415ee9bc7SJosef Bacik 9579154b1bSChris Mason return 0; 9679154b1bSChris Mason } 9779154b1bSChris Mason 98d352ac68SChris Mason /* 99d397712bSChris Mason * this does all the record keeping required to make sure that a reference 100d397712bSChris Mason * counted root is properly recorded in a given transaction. This is required 101d397712bSChris Mason * to make sure the old root from before we joined the transaction is deleted 102d397712bSChris Mason * when the transaction commits 103d352ac68SChris Mason */ 1045d4f98a2SYan Zheng static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, 1055d4f98a2SYan Zheng struct btrfs_root *root) 1066702ed49SChris Mason { 1075d4f98a2SYan Zheng if (root->ref_cows && root->last_trans < trans->transid) { 1086702ed49SChris Mason WARN_ON(root == root->fs_info->extent_root); 1095d4f98a2SYan Zheng WARN_ON(root->commit_root != root->node); 1105d4f98a2SYan Zheng 1116702ed49SChris Mason radix_tree_tag_set(&root->fs_info->fs_roots_radix, 1126702ed49SChris Mason (unsigned long)root->root_key.objectid, 1136702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 1145d4f98a2SYan Zheng root->last_trans = trans->transid; 1155d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 1166702ed49SChris Mason } 1175d4f98a2SYan Zheng return 0; 1186702ed49SChris Mason } 1195d4f98a2SYan Zheng 1205d4f98a2SYan Zheng int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 1215d4f98a2SYan Zheng struct btrfs_root *root) 1225d4f98a2SYan Zheng { 1235d4f98a2SYan Zheng if (!root->ref_cows) 1245d4f98a2SYan Zheng return 0; 1255d4f98a2SYan Zheng 1265d4f98a2SYan Zheng mutex_lock(&root->fs_info->trans_mutex); 1275d4f98a2SYan Zheng if (root->last_trans == trans->transid) { 1285d4f98a2SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 1295d4f98a2SYan Zheng return 0; 1305d4f98a2SYan Zheng } 1315d4f98a2SYan Zheng 1325d4f98a2SYan Zheng record_root_in_trans(trans, root); 1335d4f98a2SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 1346702ed49SChris Mason return 0; 1356702ed49SChris Mason } 1366702ed49SChris Mason 137d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 138d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 139d352ac68SChris Mason * transaction might not be fully on disk. 140d352ac68SChris Mason */ 14137d1aeeeSChris Mason static void wait_current_trans(struct btrfs_root *root) 14279154b1bSChris Mason { 143f9295749SChris Mason struct btrfs_transaction *cur_trans; 14479154b1bSChris Mason 145f9295749SChris Mason cur_trans = root->fs_info->running_transaction; 14637d1aeeeSChris Mason if (cur_trans && cur_trans->blocked) { 147f9295749SChris Mason DEFINE_WAIT(wait); 148f9295749SChris Mason cur_trans->use_count++; 149f9295749SChris Mason while (1) { 150f9295749SChris Mason prepare_to_wait(&root->fs_info->transaction_wait, &wait, 151f9295749SChris Mason TASK_UNINTERRUPTIBLE); 152471fa17dSZhao Lei if (!cur_trans->blocked) 153471fa17dSZhao Lei break; 154f9295749SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 155f9295749SChris Mason schedule(); 156f9295749SChris Mason mutex_lock(&root->fs_info->trans_mutex); 157f9295749SChris Mason } 158471fa17dSZhao Lei finish_wait(&root->fs_info->transaction_wait, &wait); 159f9295749SChris Mason put_transaction(cur_trans); 160f9295749SChris Mason } 16137d1aeeeSChris Mason } 16237d1aeeeSChris Mason 163249ac1e5SJosef Bacik enum btrfs_trans_type { 164249ac1e5SJosef Bacik TRANS_START, 165249ac1e5SJosef Bacik TRANS_JOIN, 166249ac1e5SJosef Bacik TRANS_USERSPACE, 1670af3d00bSJosef Bacik TRANS_JOIN_NOLOCK, 168249ac1e5SJosef Bacik }; 169249ac1e5SJosef Bacik 170a22285a6SYan, Zheng static int may_wait_transaction(struct btrfs_root *root, int type) 17137d1aeeeSChris Mason { 1724bef0848SChris Mason if (!root->fs_info->log_root_recovering && 173249ac1e5SJosef Bacik ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || 174249ac1e5SJosef Bacik type == TRANS_USERSPACE)) 175a22285a6SYan, Zheng return 1; 176a22285a6SYan, Zheng return 0; 177a22285a6SYan, Zheng } 178a22285a6SYan, Zheng 179a22285a6SYan, Zheng static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 180a22285a6SYan, Zheng u64 num_items, int type) 181a22285a6SYan, Zheng { 182a22285a6SYan, Zheng struct btrfs_trans_handle *h; 183a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 18406d5a589SJosef Bacik int retries = 0; 185a22285a6SYan, Zheng int ret; 186acce952bSliubo 187acce952bSliubo if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 188acce952bSliubo return ERR_PTR(-EROFS); 189a22285a6SYan, Zheng again: 190a22285a6SYan, Zheng h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); 191a22285a6SYan, Zheng if (!h) 192a22285a6SYan, Zheng return ERR_PTR(-ENOMEM); 193a22285a6SYan, Zheng 1940af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 195a22285a6SYan, Zheng mutex_lock(&root->fs_info->trans_mutex); 196a22285a6SYan, Zheng if (may_wait_transaction(root, type)) 19737d1aeeeSChris Mason wait_current_trans(root); 198a22285a6SYan, Zheng 19979154b1bSChris Mason ret = join_transaction(root); 200db5b493aSTsutomu Itoh if (ret < 0) { 2016e8df2aeSYoshinori Sano kmem_cache_free(btrfs_trans_handle_cachep, h); 202db5b493aSTsutomu Itoh if (type != TRANS_JOIN_NOLOCK) 203db5b493aSTsutomu Itoh mutex_unlock(&root->fs_info->trans_mutex); 204db5b493aSTsutomu Itoh return ERR_PTR(ret); 205db5b493aSTsutomu Itoh } 2060f7d52f4SChris Mason 207a22285a6SYan, Zheng cur_trans = root->fs_info->running_transaction; 208a22285a6SYan, Zheng cur_trans->use_count++; 2090af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 210a22285a6SYan, Zheng mutex_unlock(&root->fs_info->trans_mutex); 211a22285a6SYan, Zheng 212a22285a6SYan, Zheng h->transid = cur_trans->transid; 213a22285a6SYan, Zheng h->transaction = cur_trans; 21479154b1bSChris Mason h->blocks_used = 0; 215d2fb3437SYan Zheng h->block_group = 0; 216a22285a6SYan, Zheng h->bytes_reserved = 0; 21756bec294SChris Mason h->delayed_ref_updates = 0; 218f0486c68SYan, Zheng h->block_rsv = NULL; 219b7ec40d7SChris Mason 220a22285a6SYan, Zheng smp_mb(); 221a22285a6SYan, Zheng if (cur_trans->blocked && may_wait_transaction(root, type)) { 222a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 223a22285a6SYan, Zheng goto again; 224a22285a6SYan, Zheng } 2259ed74f2dSJosef Bacik 226a22285a6SYan, Zheng if (num_items > 0) { 2278bb8ab2eSJosef Bacik ret = btrfs_trans_reserve_metadata(h, root, num_items); 22806d5a589SJosef Bacik if (ret == -EAGAIN && !retries) { 22906d5a589SJosef Bacik retries++; 230a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 231a22285a6SYan, Zheng goto again; 23206d5a589SJosef Bacik } else if (ret == -EAGAIN) { 23306d5a589SJosef Bacik /* 23406d5a589SJosef Bacik * We have already retried and got EAGAIN, so really we 23506d5a589SJosef Bacik * don't have space, so set ret to -ENOSPC. 23606d5a589SJosef Bacik */ 23706d5a589SJosef Bacik ret = -ENOSPC; 238a22285a6SYan, Zheng } 23906d5a589SJosef Bacik 240a22285a6SYan, Zheng if (ret < 0) { 241a22285a6SYan, Zheng btrfs_end_transaction(h, root); 242a22285a6SYan, Zheng return ERR_PTR(ret); 243a22285a6SYan, Zheng } 244a22285a6SYan, Zheng } 245a22285a6SYan, Zheng 2460af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 247a22285a6SYan, Zheng mutex_lock(&root->fs_info->trans_mutex); 2485d4f98a2SYan Zheng record_root_in_trans(h, root); 2490af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 25079154b1bSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 251a22285a6SYan, Zheng 252a22285a6SYan, Zheng if (!current->journal_info && type != TRANS_USERSPACE) 253a22285a6SYan, Zheng current->journal_info = h; 25479154b1bSChris Mason return h; 25579154b1bSChris Mason } 25679154b1bSChris Mason 257f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 258a22285a6SYan, Zheng int num_items) 259f9295749SChris Mason { 260a22285a6SYan, Zheng return start_transaction(root, num_items, TRANS_START); 261f9295749SChris Mason } 262f9295749SChris Mason struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, 263f9295749SChris Mason int num_blocks) 264f9295749SChris Mason { 265a22285a6SYan, Zheng return start_transaction(root, 0, TRANS_JOIN); 266f9295749SChris Mason } 267f9295749SChris Mason 2680af3d00bSJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, 2690af3d00bSJosef Bacik int num_blocks) 2700af3d00bSJosef Bacik { 2710af3d00bSJosef Bacik return start_transaction(root, 0, TRANS_JOIN_NOLOCK); 2720af3d00bSJosef Bacik } 2730af3d00bSJosef Bacik 2749ca9ee09SSage Weil struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, 2759ca9ee09SSage Weil int num_blocks) 2769ca9ee09SSage Weil { 277a22285a6SYan, Zheng return start_transaction(r, 0, TRANS_USERSPACE); 2789ca9ee09SSage Weil } 2799ca9ee09SSage Weil 280d352ac68SChris Mason /* wait for a transaction commit to be fully complete */ 28189ce8a63SChris Mason static noinline int wait_for_commit(struct btrfs_root *root, 28289ce8a63SChris Mason struct btrfs_transaction *commit) 28389ce8a63SChris Mason { 28489ce8a63SChris Mason DEFINE_WAIT(wait); 28589ce8a63SChris Mason mutex_lock(&root->fs_info->trans_mutex); 28689ce8a63SChris Mason while (!commit->commit_done) { 28789ce8a63SChris Mason prepare_to_wait(&commit->commit_wait, &wait, 28889ce8a63SChris Mason TASK_UNINTERRUPTIBLE); 28989ce8a63SChris Mason if (commit->commit_done) 29089ce8a63SChris Mason break; 29189ce8a63SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 29289ce8a63SChris Mason schedule(); 29389ce8a63SChris Mason mutex_lock(&root->fs_info->trans_mutex); 29489ce8a63SChris Mason } 29589ce8a63SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 29689ce8a63SChris Mason finish_wait(&commit->commit_wait, &wait); 29789ce8a63SChris Mason return 0; 29889ce8a63SChris Mason } 29989ce8a63SChris Mason 30046204592SSage Weil int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) 30146204592SSage Weil { 30246204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 30346204592SSage Weil int ret; 30446204592SSage Weil 30546204592SSage Weil mutex_lock(&root->fs_info->trans_mutex); 30646204592SSage Weil 30746204592SSage Weil ret = 0; 30846204592SSage Weil if (transid) { 30946204592SSage Weil if (transid <= root->fs_info->last_trans_committed) 31046204592SSage Weil goto out_unlock; 31146204592SSage Weil 31246204592SSage Weil /* find specified transaction */ 31346204592SSage Weil list_for_each_entry(t, &root->fs_info->trans_list, list) { 31446204592SSage Weil if (t->transid == transid) { 31546204592SSage Weil cur_trans = t; 31646204592SSage Weil break; 31746204592SSage Weil } 31846204592SSage Weil if (t->transid > transid) 31946204592SSage Weil break; 32046204592SSage Weil } 32146204592SSage Weil ret = -EINVAL; 32246204592SSage Weil if (!cur_trans) 32346204592SSage Weil goto out_unlock; /* bad transid */ 32446204592SSage Weil } else { 32546204592SSage Weil /* find newest transaction that is committing | committed */ 32646204592SSage Weil list_for_each_entry_reverse(t, &root->fs_info->trans_list, 32746204592SSage Weil list) { 32846204592SSage Weil if (t->in_commit) { 32946204592SSage Weil if (t->commit_done) 33046204592SSage Weil goto out_unlock; 33146204592SSage Weil cur_trans = t; 33246204592SSage Weil break; 33346204592SSage Weil } 33446204592SSage Weil } 33546204592SSage Weil if (!cur_trans) 33646204592SSage Weil goto out_unlock; /* nothing committing|committed */ 33746204592SSage Weil } 33846204592SSage Weil 33946204592SSage Weil cur_trans->use_count++; 34046204592SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 34146204592SSage Weil 34246204592SSage Weil wait_for_commit(root, cur_trans); 34346204592SSage Weil 34446204592SSage Weil mutex_lock(&root->fs_info->trans_mutex); 34546204592SSage Weil put_transaction(cur_trans); 34646204592SSage Weil ret = 0; 34746204592SSage Weil out_unlock: 34846204592SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 34946204592SSage Weil return ret; 35046204592SSage Weil } 35146204592SSage Weil 3525d4f98a2SYan Zheng #if 0 353d352ac68SChris Mason /* 354d397712bSChris Mason * rate limit against the drop_snapshot code. This helps to slow down new 355d397712bSChris Mason * operations if the drop_snapshot code isn't able to keep up. 356d352ac68SChris Mason */ 35737d1aeeeSChris Mason static void throttle_on_drops(struct btrfs_root *root) 358ab78c84dSChris Mason { 359ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 3602dd3e67bSChris Mason int harder_count = 0; 361ab78c84dSChris Mason 3622dd3e67bSChris Mason harder: 363ab78c84dSChris Mason if (atomic_read(&info->throttles)) { 364ab78c84dSChris Mason DEFINE_WAIT(wait); 365ab78c84dSChris Mason int thr; 366ab78c84dSChris Mason thr = atomic_read(&info->throttle_gen); 367ab78c84dSChris Mason 368ab78c84dSChris Mason do { 369ab78c84dSChris Mason prepare_to_wait(&info->transaction_throttle, 370ab78c84dSChris Mason &wait, TASK_UNINTERRUPTIBLE); 371ab78c84dSChris Mason if (!atomic_read(&info->throttles)) { 372ab78c84dSChris Mason finish_wait(&info->transaction_throttle, &wait); 373ab78c84dSChris Mason break; 374ab78c84dSChris Mason } 375ab78c84dSChris Mason schedule(); 376ab78c84dSChris Mason finish_wait(&info->transaction_throttle, &wait); 377ab78c84dSChris Mason } while (thr == atomic_read(&info->throttle_gen)); 3782dd3e67bSChris Mason harder_count++; 3792dd3e67bSChris Mason 3802dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && 3812dd3e67bSChris Mason harder_count < 2) 3822dd3e67bSChris Mason goto harder; 3832dd3e67bSChris Mason 3842dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && 3852dd3e67bSChris Mason harder_count < 10) 3862dd3e67bSChris Mason goto harder; 3872dd3e67bSChris Mason 3882dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && 3892dd3e67bSChris Mason harder_count < 20) 3902dd3e67bSChris Mason goto harder; 391ab78c84dSChris Mason } 392ab78c84dSChris Mason } 3935d4f98a2SYan Zheng #endif 394ab78c84dSChris Mason 39537d1aeeeSChris Mason void btrfs_throttle(struct btrfs_root *root) 39637d1aeeeSChris Mason { 39737d1aeeeSChris Mason mutex_lock(&root->fs_info->trans_mutex); 3989ca9ee09SSage Weil if (!root->fs_info->open_ioctl_trans) 39937d1aeeeSChris Mason wait_current_trans(root); 40037d1aeeeSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 40137d1aeeeSChris Mason } 40237d1aeeeSChris Mason 4038929ecfaSYan, Zheng static int should_end_transaction(struct btrfs_trans_handle *trans, 4048929ecfaSYan, Zheng struct btrfs_root *root) 4058929ecfaSYan, Zheng { 4068929ecfaSYan, Zheng int ret; 4078929ecfaSYan, Zheng ret = btrfs_block_rsv_check(trans, root, 4088929ecfaSYan, Zheng &root->fs_info->global_block_rsv, 0, 5); 4098929ecfaSYan, Zheng return ret ? 1 : 0; 4108929ecfaSYan, Zheng } 4118929ecfaSYan, Zheng 4128929ecfaSYan, Zheng int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, 4138929ecfaSYan, Zheng struct btrfs_root *root) 4148929ecfaSYan, Zheng { 4158929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 4168929ecfaSYan, Zheng int updates; 4178929ecfaSYan, Zheng 4188929ecfaSYan, Zheng if (cur_trans->blocked || cur_trans->delayed_refs.flushing) 4198929ecfaSYan, Zheng return 1; 4208929ecfaSYan, Zheng 4218929ecfaSYan, Zheng updates = trans->delayed_ref_updates; 4228929ecfaSYan, Zheng trans->delayed_ref_updates = 0; 4238929ecfaSYan, Zheng if (updates) 4248929ecfaSYan, Zheng btrfs_run_delayed_refs(trans, root, updates); 4258929ecfaSYan, Zheng 4268929ecfaSYan, Zheng return should_end_transaction(trans, root); 4278929ecfaSYan, Zheng } 4288929ecfaSYan, Zheng 42989ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 4300af3d00bSJosef Bacik struct btrfs_root *root, int throttle, int lock) 43179154b1bSChris Mason { 4328929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 433ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 434c3e69d58SChris Mason int count = 0; 435d6e4a428SChris Mason 436c3e69d58SChris Mason while (count < 4) { 437c3e69d58SChris Mason unsigned long cur = trans->delayed_ref_updates; 438c3e69d58SChris Mason trans->delayed_ref_updates = 0; 439c3e69d58SChris Mason if (cur && 440c3e69d58SChris Mason trans->transaction->delayed_refs.num_heads_ready > 64) { 441c3e69d58SChris Mason trans->delayed_ref_updates = 0; 442b7ec40d7SChris Mason 443b7ec40d7SChris Mason /* 444b7ec40d7SChris Mason * do a full flush if the transaction is trying 445b7ec40d7SChris Mason * to close 446b7ec40d7SChris Mason */ 447b7ec40d7SChris Mason if (trans->transaction->delayed_refs.flushing) 448b7ec40d7SChris Mason cur = 0; 449c3e69d58SChris Mason btrfs_run_delayed_refs(trans, root, cur); 450c3e69d58SChris Mason } else { 451c3e69d58SChris Mason break; 452c3e69d58SChris Mason } 453c3e69d58SChris Mason count++; 45456bec294SChris Mason } 45556bec294SChris Mason 456a22285a6SYan, Zheng btrfs_trans_release_metadata(trans, root); 457a22285a6SYan, Zheng 4580af3d00bSJosef Bacik if (lock && !root->fs_info->open_ioctl_trans && 4598929ecfaSYan, Zheng should_end_transaction(trans, root)) 4608929ecfaSYan, Zheng trans->transaction->blocked = 1; 4618929ecfaSYan, Zheng 4620af3d00bSJosef Bacik if (lock && cur_trans->blocked && !cur_trans->in_commit) { 4638929ecfaSYan, Zheng if (throttle) 4648929ecfaSYan, Zheng return btrfs_commit_transaction(trans, root); 4658929ecfaSYan, Zheng else 4668929ecfaSYan, Zheng wake_up_process(info->transaction_kthread); 4678929ecfaSYan, Zheng } 4688929ecfaSYan, Zheng 4690af3d00bSJosef Bacik if (lock) 470ab78c84dSChris Mason mutex_lock(&info->trans_mutex); 4718929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 472d5719762SChris Mason WARN_ON(cur_trans->num_writers < 1); 473ccd467d6SChris Mason cur_trans->num_writers--; 47489ce8a63SChris Mason 47599d16cbcSSage Weil smp_mb(); 47679154b1bSChris Mason if (waitqueue_active(&cur_trans->writer_wait)) 47779154b1bSChris Mason wake_up(&cur_trans->writer_wait); 47879154b1bSChris Mason put_transaction(cur_trans); 4790af3d00bSJosef Bacik if (lock) 480ab78c84dSChris Mason mutex_unlock(&info->trans_mutex); 4819ed74f2dSJosef Bacik 4829ed74f2dSJosef Bacik if (current->journal_info == trans) 4839ed74f2dSJosef Bacik current->journal_info = NULL; 484d6025579SChris Mason memset(trans, 0, sizeof(*trans)); 4852c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 486ab78c84dSChris Mason 48724bbcf04SYan, Zheng if (throttle) 48824bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 48924bbcf04SYan, Zheng 49079154b1bSChris Mason return 0; 49179154b1bSChris Mason } 49279154b1bSChris Mason 49389ce8a63SChris Mason int btrfs_end_transaction(struct btrfs_trans_handle *trans, 49489ce8a63SChris Mason struct btrfs_root *root) 49589ce8a63SChris Mason { 4960af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 0, 1); 49789ce8a63SChris Mason } 49889ce8a63SChris Mason 49989ce8a63SChris Mason int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, 50089ce8a63SChris Mason struct btrfs_root *root) 50189ce8a63SChris Mason { 5020af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 1, 1); 5030af3d00bSJosef Bacik } 5040af3d00bSJosef Bacik 5050af3d00bSJosef Bacik int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, 5060af3d00bSJosef Bacik struct btrfs_root *root) 5070af3d00bSJosef Bacik { 5080af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 0, 0); 50989ce8a63SChris Mason } 51089ce8a63SChris Mason 511d352ac68SChris Mason /* 512d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 513d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 514690587d1SChris Mason * those extents are sent to disk but does not wait on them 515d352ac68SChris Mason */ 516690587d1SChris Mason int btrfs_write_marked_extents(struct btrfs_root *root, 5178cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 51879154b1bSChris Mason { 5197c4452b9SChris Mason int ret; 520777e6bd7SChris Mason int err = 0; 5217c4452b9SChris Mason int werr = 0; 5227c4452b9SChris Mason struct page *page; 5237c4452b9SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 524777e6bd7SChris Mason u64 start = 0; 5255f39d397SChris Mason u64 end; 5265f39d397SChris Mason unsigned long index; 5277c4452b9SChris Mason 5287c4452b9SChris Mason while (1) { 529777e6bd7SChris Mason ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5308cef4e16SYan, Zheng mark); 5315f39d397SChris Mason if (ret) 5327c4452b9SChris Mason break; 5335f39d397SChris Mason while (start <= end) { 534777e6bd7SChris Mason cond_resched(); 535777e6bd7SChris Mason 5365f39d397SChris Mason index = start >> PAGE_CACHE_SHIFT; 53735ebb934SChris Mason start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 5384bef0848SChris Mason page = find_get_page(btree_inode->i_mapping, index); 5397c4452b9SChris Mason if (!page) 5407c4452b9SChris Mason continue; 5414bef0848SChris Mason 5424bef0848SChris Mason btree_lock_page_hook(page); 5434bef0848SChris Mason if (!page->mapping) { 5444bef0848SChris Mason unlock_page(page); 5454bef0848SChris Mason page_cache_release(page); 5464bef0848SChris Mason continue; 5474bef0848SChris Mason } 5484bef0848SChris Mason 5496702ed49SChris Mason if (PageWriteback(page)) { 5506702ed49SChris Mason if (PageDirty(page)) 5516702ed49SChris Mason wait_on_page_writeback(page); 5526702ed49SChris Mason else { 5536702ed49SChris Mason unlock_page(page); 5546702ed49SChris Mason page_cache_release(page); 5556702ed49SChris Mason continue; 5566702ed49SChris Mason } 5576702ed49SChris Mason } 5587c4452b9SChris Mason err = write_one_page(page, 0); 5597c4452b9SChris Mason if (err) 5607c4452b9SChris Mason werr = err; 5617c4452b9SChris Mason page_cache_release(page); 5627c4452b9SChris Mason } 5637c4452b9SChris Mason } 564690587d1SChris Mason if (err) 565690587d1SChris Mason werr = err; 566690587d1SChris Mason return werr; 567690587d1SChris Mason } 568690587d1SChris Mason 569690587d1SChris Mason /* 570690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 571690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 572690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 573690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 574690587d1SChris Mason */ 575690587d1SChris Mason int btrfs_wait_marked_extents(struct btrfs_root *root, 5768cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 577690587d1SChris Mason { 578690587d1SChris Mason int ret; 579690587d1SChris Mason int err = 0; 580690587d1SChris Mason int werr = 0; 581690587d1SChris Mason struct page *page; 582690587d1SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 583690587d1SChris Mason u64 start = 0; 584690587d1SChris Mason u64 end; 585690587d1SChris Mason unsigned long index; 586690587d1SChris Mason 587777e6bd7SChris Mason while (1) { 5888cef4e16SYan, Zheng ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5898cef4e16SYan, Zheng mark); 590777e6bd7SChris Mason if (ret) 591777e6bd7SChris Mason break; 592777e6bd7SChris Mason 5938cef4e16SYan, Zheng clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 594777e6bd7SChris Mason while (start <= end) { 595777e6bd7SChris Mason index = start >> PAGE_CACHE_SHIFT; 596777e6bd7SChris Mason start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 597777e6bd7SChris Mason page = find_get_page(btree_inode->i_mapping, index); 598777e6bd7SChris Mason if (!page) 599777e6bd7SChris Mason continue; 600777e6bd7SChris Mason if (PageDirty(page)) { 6014bef0848SChris Mason btree_lock_page_hook(page); 6024bef0848SChris Mason wait_on_page_writeback(page); 603777e6bd7SChris Mason err = write_one_page(page, 0); 604777e6bd7SChris Mason if (err) 605777e6bd7SChris Mason werr = err; 606777e6bd7SChris Mason } 607777e6bd7SChris Mason wait_on_page_writeback(page); 608777e6bd7SChris Mason page_cache_release(page); 609777e6bd7SChris Mason cond_resched(); 610777e6bd7SChris Mason } 611777e6bd7SChris Mason } 6127c4452b9SChris Mason if (err) 6137c4452b9SChris Mason werr = err; 6147c4452b9SChris Mason return werr; 61579154b1bSChris Mason } 61679154b1bSChris Mason 617690587d1SChris Mason /* 618690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 619690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 620690587d1SChris Mason * those extents are on disk for transaction or log commit 621690587d1SChris Mason */ 622690587d1SChris Mason int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 6238cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 624690587d1SChris Mason { 625690587d1SChris Mason int ret; 626690587d1SChris Mason int ret2; 627690587d1SChris Mason 6288cef4e16SYan, Zheng ret = btrfs_write_marked_extents(root, dirty_pages, mark); 6298cef4e16SYan, Zheng ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); 630690587d1SChris Mason return ret || ret2; 631690587d1SChris Mason } 632690587d1SChris Mason 633d0c803c4SChris Mason int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, 634d0c803c4SChris Mason struct btrfs_root *root) 635d0c803c4SChris Mason { 636d0c803c4SChris Mason if (!trans || !trans->transaction) { 637d0c803c4SChris Mason struct inode *btree_inode; 638d0c803c4SChris Mason btree_inode = root->fs_info->btree_inode; 639d0c803c4SChris Mason return filemap_write_and_wait(btree_inode->i_mapping); 640d0c803c4SChris Mason } 641d0c803c4SChris Mason return btrfs_write_and_wait_marked_extents(root, 6428cef4e16SYan, Zheng &trans->transaction->dirty_pages, 6438cef4e16SYan, Zheng EXTENT_DIRTY); 644d0c803c4SChris Mason } 645d0c803c4SChris Mason 646d352ac68SChris Mason /* 647d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 648d352ac68SChris Mason * 649d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 650d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 651d352ac68SChris Mason * allocation tree. 652d352ac68SChris Mason * 653d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 654d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 655d352ac68SChris Mason */ 6560b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 65779154b1bSChris Mason struct btrfs_root *root) 65879154b1bSChris Mason { 65979154b1bSChris Mason int ret; 6600b86a832SChris Mason u64 old_root_bytenr; 66186b9f2ecSYan, Zheng u64 old_root_used; 6620b86a832SChris Mason struct btrfs_root *tree_root = root->fs_info->tree_root; 66379154b1bSChris Mason 66486b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6650b86a832SChris Mason btrfs_write_dirty_block_groups(trans, root); 66656bec294SChris Mason 66779154b1bSChris Mason while (1) { 6680b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 66986b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 67086b9f2ecSYan, Zheng old_root_used == btrfs_root_used(&root->root_item)) 67179154b1bSChris Mason break; 67287ef2bb4SChris Mason 6735d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 67479154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 6750b86a832SChris Mason &root->root_key, 6760b86a832SChris Mason &root->root_item); 67779154b1bSChris Mason BUG_ON(ret); 67856bec294SChris Mason 67986b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6804a8c9a62SYan Zheng ret = btrfs_write_dirty_block_groups(trans, root); 68156bec294SChris Mason BUG_ON(ret); 6820b86a832SChris Mason } 683276e680dSYan Zheng 684276e680dSYan Zheng if (root != root->fs_info->extent_root) 685817d52f8SJosef Bacik switch_commit_root(root); 686276e680dSYan Zheng 6870b86a832SChris Mason return 0; 6880b86a832SChris Mason } 6890b86a832SChris Mason 690d352ac68SChris Mason /* 691d352ac68SChris Mason * update all the cowonly tree roots on disk 692d352ac68SChris Mason */ 6935d4f98a2SYan Zheng static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, 6940b86a832SChris Mason struct btrfs_root *root) 6950b86a832SChris Mason { 6960b86a832SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 6970b86a832SChris Mason struct list_head *next; 69884234f3aSYan Zheng struct extent_buffer *eb; 69956bec294SChris Mason int ret; 70084234f3aSYan Zheng 70156bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 70256bec294SChris Mason BUG_ON(ret); 70387ef2bb4SChris Mason 70484234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 7059fa8cfe7SChris Mason btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); 70684234f3aSYan Zheng btrfs_tree_unlock(eb); 70784234f3aSYan Zheng free_extent_buffer(eb); 7080b86a832SChris Mason 70956bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 71056bec294SChris Mason BUG_ON(ret); 71187ef2bb4SChris Mason 7120b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 7130b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 7140b86a832SChris Mason list_del_init(next); 7150b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 71687ef2bb4SChris Mason 7170b86a832SChris Mason update_cowonly_root(trans, root); 71879154b1bSChris Mason } 719276e680dSYan Zheng 720276e680dSYan Zheng down_write(&fs_info->extent_commit_sem); 721276e680dSYan Zheng switch_commit_root(fs_info->extent_root); 722276e680dSYan Zheng up_write(&fs_info->extent_commit_sem); 723276e680dSYan Zheng 72479154b1bSChris Mason return 0; 72579154b1bSChris Mason } 72679154b1bSChris Mason 727d352ac68SChris Mason /* 728d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 729d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 730d352ac68SChris Mason * be deleted 731d352ac68SChris Mason */ 7325d4f98a2SYan Zheng int btrfs_add_dead_root(struct btrfs_root *root) 7335eda7b5eSChris Mason { 734b48652c1SYan Zheng mutex_lock(&root->fs_info->trans_mutex); 7355d4f98a2SYan Zheng list_add(&root->root_list, &root->fs_info->dead_roots); 736b48652c1SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 7375eda7b5eSChris Mason return 0; 7385eda7b5eSChris Mason } 7395eda7b5eSChris Mason 740d352ac68SChris Mason /* 7415d4f98a2SYan Zheng * update all the cowonly tree roots on disk 742d352ac68SChris Mason */ 7435d4f98a2SYan Zheng static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, 7445d4f98a2SYan Zheng struct btrfs_root *root) 7450f7d52f4SChris Mason { 7460f7d52f4SChris Mason struct btrfs_root *gang[8]; 7475d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 7480f7d52f4SChris Mason int i; 7490f7d52f4SChris Mason int ret; 75054aa1f4dSChris Mason int err = 0; 75154aa1f4dSChris Mason 7520f7d52f4SChris Mason while (1) { 7535d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 7545d4f98a2SYan Zheng (void **)gang, 0, 7550f7d52f4SChris Mason ARRAY_SIZE(gang), 7560f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 7570f7d52f4SChris Mason if (ret == 0) 7580f7d52f4SChris Mason break; 7590f7d52f4SChris Mason for (i = 0; i < ret; i++) { 7600f7d52f4SChris Mason root = gang[i]; 7615d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 7622619ba1fSChris Mason (unsigned long)root->root_key.objectid, 7630f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 76431153d81SYan Zheng 765e02119d5SChris Mason btrfs_free_log(trans, root); 7665d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 767d68fc57bSYan, Zheng btrfs_orphan_commit_root(trans, root); 768e02119d5SChris Mason 769978d910dSYan Zheng if (root->commit_root != root->node) { 770817d52f8SJosef Bacik switch_commit_root(root); 771978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 772978d910dSYan Zheng root->node); 773978d910dSYan Zheng } 77431153d81SYan Zheng 7755d4f98a2SYan Zheng err = btrfs_update_root(trans, fs_info->tree_root, 7760f7d52f4SChris Mason &root->root_key, 7770f7d52f4SChris Mason &root->root_item); 77854aa1f4dSChris Mason if (err) 77954aa1f4dSChris Mason break; 7800f7d52f4SChris Mason } 7819f3a7427SChris Mason } 78254aa1f4dSChris Mason return err; 7830f7d52f4SChris Mason } 7840f7d52f4SChris Mason 785d352ac68SChris Mason /* 786d352ac68SChris Mason * defrag a given btree. If cacheonly == 1, this won't read from the disk, 787d352ac68SChris Mason * otherwise every leaf in the btree is read and defragged. 788d352ac68SChris Mason */ 789e9d0b13bSChris Mason int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) 790e9d0b13bSChris Mason { 791e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 792e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 7938929ecfaSYan, Zheng int ret; 794d3c2fdcfSChris Mason unsigned long nr; 795e9d0b13bSChris Mason 7968929ecfaSYan, Zheng if (xchg(&root->defrag_running, 1)) 797e9d0b13bSChris Mason return 0; 7988929ecfaSYan, Zheng 7996b80053dSChris Mason while (1) { 8008929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 8018929ecfaSYan, Zheng if (IS_ERR(trans)) 8028929ecfaSYan, Zheng return PTR_ERR(trans); 8038929ecfaSYan, Zheng 804e9d0b13bSChris Mason ret = btrfs_defrag_leaves(trans, root, cacheonly); 8058929ecfaSYan, Zheng 806d3c2fdcfSChris Mason nr = trans->blocks_used; 807e9d0b13bSChris Mason btrfs_end_transaction(trans, root); 808d3c2fdcfSChris Mason btrfs_btree_balance_dirty(info->tree_root, nr); 809e9d0b13bSChris Mason cond_resched(); 810e9d0b13bSChris Mason 8113f157a2fSChris Mason if (root->fs_info->closing || ret != -EAGAIN) 812e9d0b13bSChris Mason break; 813e9d0b13bSChris Mason } 814e9d0b13bSChris Mason root->defrag_running = 0; 8158929ecfaSYan, Zheng return ret; 816e9d0b13bSChris Mason } 817e9d0b13bSChris Mason 8182c47e605SYan Zheng #if 0 819d352ac68SChris Mason /* 820b7ec40d7SChris Mason * when dropping snapshots, we generate a ton of delayed refs, and it makes 821b7ec40d7SChris Mason * sense not to join the transaction while it is trying to flush the current 822b7ec40d7SChris Mason * queue of delayed refs out. 823b7ec40d7SChris Mason * 824b7ec40d7SChris Mason * This is used by the drop snapshot code only 825b7ec40d7SChris Mason */ 826b7ec40d7SChris Mason static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) 827b7ec40d7SChris Mason { 828b7ec40d7SChris Mason DEFINE_WAIT(wait); 829b7ec40d7SChris Mason 830b7ec40d7SChris Mason mutex_lock(&info->trans_mutex); 831b7ec40d7SChris Mason while (info->running_transaction && 832b7ec40d7SChris Mason info->running_transaction->delayed_refs.flushing) { 833b7ec40d7SChris Mason prepare_to_wait(&info->transaction_wait, &wait, 834b7ec40d7SChris Mason TASK_UNINTERRUPTIBLE); 835b7ec40d7SChris Mason mutex_unlock(&info->trans_mutex); 83659bc5c75SChris Mason 837b7ec40d7SChris Mason schedule(); 83859bc5c75SChris Mason 839b7ec40d7SChris Mason mutex_lock(&info->trans_mutex); 840b7ec40d7SChris Mason finish_wait(&info->transaction_wait, &wait); 841b7ec40d7SChris Mason } 842b7ec40d7SChris Mason mutex_unlock(&info->trans_mutex); 843b7ec40d7SChris Mason return 0; 844b7ec40d7SChris Mason } 845b7ec40d7SChris Mason 846b7ec40d7SChris Mason /* 847d352ac68SChris Mason * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on 848d352ac68SChris Mason * all of them 849d352ac68SChris Mason */ 8505d4f98a2SYan Zheng int btrfs_drop_dead_root(struct btrfs_root *root) 8510f7d52f4SChris Mason { 8520f7d52f4SChris Mason struct btrfs_trans_handle *trans; 8535d4f98a2SYan Zheng struct btrfs_root *tree_root = root->fs_info->tree_root; 854d3c2fdcfSChris Mason unsigned long nr; 8555d4f98a2SYan Zheng int ret; 85658176a96SJosef Bacik 8579f3a7427SChris Mason while (1) { 858b7ec40d7SChris Mason /* 859b7ec40d7SChris Mason * we don't want to jump in and create a bunch of 860b7ec40d7SChris Mason * delayed refs if the transaction is starting to close 861b7ec40d7SChris Mason */ 862b7ec40d7SChris Mason wait_transaction_pre_flush(tree_root->fs_info); 8630f7d52f4SChris Mason trans = btrfs_start_transaction(tree_root, 1); 864b7ec40d7SChris Mason 865b7ec40d7SChris Mason /* 866b7ec40d7SChris Mason * we've joined a transaction, make sure it isn't 867b7ec40d7SChris Mason * closing right now 868b7ec40d7SChris Mason */ 869b7ec40d7SChris Mason if (trans->transaction->delayed_refs.flushing) { 870b7ec40d7SChris Mason btrfs_end_transaction(trans, tree_root); 871b7ec40d7SChris Mason continue; 872b7ec40d7SChris Mason } 873b7ec40d7SChris Mason 8745d4f98a2SYan Zheng ret = btrfs_drop_snapshot(trans, root); 875d397712bSChris Mason if (ret != -EAGAIN) 8769f3a7427SChris Mason break; 87758176a96SJosef Bacik 8785d4f98a2SYan Zheng ret = btrfs_update_root(trans, tree_root, 8795d4f98a2SYan Zheng &root->root_key, 8805d4f98a2SYan Zheng &root->root_item); 8815d4f98a2SYan Zheng if (ret) 88254aa1f4dSChris Mason break; 883bcc63abbSYan 884d3c2fdcfSChris Mason nr = trans->blocks_used; 8850f7d52f4SChris Mason ret = btrfs_end_transaction(trans, tree_root); 8860f7d52f4SChris Mason BUG_ON(ret); 8875eda7b5eSChris Mason 888d3c2fdcfSChris Mason btrfs_btree_balance_dirty(tree_root, nr); 8894dc11904SChris Mason cond_resched(); 8900f7d52f4SChris Mason } 8915d4f98a2SYan Zheng BUG_ON(ret); 8925d4f98a2SYan Zheng 8935d4f98a2SYan Zheng ret = btrfs_del_root(trans, tree_root, &root->root_key); 8945d4f98a2SYan Zheng BUG_ON(ret); 8955d4f98a2SYan Zheng 8965d4f98a2SYan Zheng nr = trans->blocks_used; 8975d4f98a2SYan Zheng ret = btrfs_end_transaction(trans, tree_root); 8985d4f98a2SYan Zheng BUG_ON(ret); 8995d4f98a2SYan Zheng 9005d4f98a2SYan Zheng free_extent_buffer(root->node); 9015d4f98a2SYan Zheng free_extent_buffer(root->commit_root); 9025d4f98a2SYan Zheng kfree(root); 9035d4f98a2SYan Zheng 9045d4f98a2SYan Zheng btrfs_btree_balance_dirty(tree_root, nr); 90554aa1f4dSChris Mason return ret; 9060f7d52f4SChris Mason } 9072c47e605SYan Zheng #endif 9080f7d52f4SChris Mason 909d352ac68SChris Mason /* 910d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 911d352ac68SChris Mason * transaction commit. This does the actual creation 912d352ac68SChris Mason */ 91380b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 9143063d29fSChris Mason struct btrfs_fs_info *fs_info, 9153063d29fSChris Mason struct btrfs_pending_snapshot *pending) 9163063d29fSChris Mason { 9173063d29fSChris Mason struct btrfs_key key; 91880b6794dSChris Mason struct btrfs_root_item *new_root_item; 9193063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 9203063d29fSChris Mason struct btrfs_root *root = pending->root; 9216bdb72deSSage Weil struct btrfs_root *parent_root; 9226bdb72deSSage Weil struct inode *parent_inode; 9236a912213SJosef Bacik struct dentry *parent; 924a22285a6SYan, Zheng struct dentry *dentry; 9253063d29fSChris Mason struct extent_buffer *tmp; 926925baeddSChris Mason struct extent_buffer *old; 9273063d29fSChris Mason int ret; 928d68fc57bSYan, Zheng u64 to_reserve = 0; 9296bdb72deSSage Weil u64 index = 0; 930a22285a6SYan, Zheng u64 objectid; 931b83cc969SLi Zefan u64 root_flags; 9323063d29fSChris Mason 93380b6794dSChris Mason new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); 93480b6794dSChris Mason if (!new_root_item) { 935a22285a6SYan, Zheng pending->error = -ENOMEM; 93680b6794dSChris Mason goto fail; 93780b6794dSChris Mason } 938a22285a6SYan, Zheng 9393063d29fSChris Mason ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); 940a22285a6SYan, Zheng if (ret) { 941a22285a6SYan, Zheng pending->error = ret; 9423063d29fSChris Mason goto fail; 943a22285a6SYan, Zheng } 9443063d29fSChris Mason 9453fd0a558SYan, Zheng btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); 946d68fc57bSYan, Zheng btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); 947d68fc57bSYan, Zheng 948d68fc57bSYan, Zheng if (to_reserve > 0) { 949d68fc57bSYan, Zheng ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, 9508bb8ab2eSJosef Bacik to_reserve); 951d68fc57bSYan, Zheng if (ret) { 952d68fc57bSYan, Zheng pending->error = ret; 953d68fc57bSYan, Zheng goto fail; 954d68fc57bSYan, Zheng } 955d68fc57bSYan, Zheng } 956d68fc57bSYan, Zheng 9573063d29fSChris Mason key.objectid = objectid; 958a22285a6SYan, Zheng key.offset = (u64)-1; 959a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 9603063d29fSChris Mason 961a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 9626bdb72deSSage Weil 963a22285a6SYan, Zheng dentry = pending->dentry; 9646a912213SJosef Bacik parent = dget_parent(dentry); 9656a912213SJosef Bacik parent_inode = parent->d_inode; 966a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 9676bdb72deSSage Weil record_root_in_trans(trans, parent_root); 968a22285a6SYan, Zheng 9696bdb72deSSage Weil /* 9706bdb72deSSage Weil * insert the directory item 9716bdb72deSSage Weil */ 9726bdb72deSSage Weil ret = btrfs_set_inode_index(parent_inode, &index); 9736bdb72deSSage Weil BUG_ON(ret); 9746bdb72deSSage Weil ret = btrfs_insert_dir_item(trans, parent_root, 975a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len, 976a22285a6SYan, Zheng parent_inode->i_ino, &key, 977a22285a6SYan, Zheng BTRFS_FT_DIR, index); 9786bdb72deSSage Weil BUG_ON(ret); 9796bdb72deSSage Weil 980a22285a6SYan, Zheng btrfs_i_size_write(parent_inode, parent_inode->i_size + 981a22285a6SYan, Zheng dentry->d_name.len * 2); 9826bdb72deSSage Weil ret = btrfs_update_inode(trans, parent_root, parent_inode); 9836bdb72deSSage Weil BUG_ON(ret); 9846bdb72deSSage Weil 9856bdb72deSSage Weil record_root_in_trans(trans, root); 9866bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 9876bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 98808fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 9896bdb72deSSage Weil 990b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 991b83cc969SLi Zefan if (pending->readonly) 992b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 993b83cc969SLi Zefan else 994b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 995b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 996b83cc969SLi Zefan 997925baeddSChris Mason old = btrfs_lock_root_node(root); 9989fa8cfe7SChris Mason btrfs_cow_block(trans, root, old, NULL, 0, &old); 9995d4f98a2SYan Zheng btrfs_set_lock_blocking(old); 10003063d29fSChris Mason 1001925baeddSChris Mason btrfs_copy_root(trans, root, old, &tmp, objectid); 1002925baeddSChris Mason btrfs_tree_unlock(old); 1003925baeddSChris Mason free_extent_buffer(old); 10043063d29fSChris Mason 10055d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1006a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1007a22285a6SYan, Zheng key.offset = trans->transid; 1008a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1009925baeddSChris Mason btrfs_tree_unlock(tmp); 10103063d29fSChris Mason free_extent_buffer(tmp); 10110660b5afSChris Mason BUG_ON(ret); 10120660b5afSChris Mason 1013a22285a6SYan, Zheng /* 1014a22285a6SYan, Zheng * insert root back/forward references 1015a22285a6SYan, Zheng */ 1016a22285a6SYan, Zheng ret = btrfs_add_root_ref(trans, tree_root, objectid, 1017a22285a6SYan, Zheng parent_root->root_key.objectid, 1018a22285a6SYan, Zheng parent_inode->i_ino, index, 1019a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 1020a22285a6SYan, Zheng BUG_ON(ret); 10216a912213SJosef Bacik dput(parent); 1022a22285a6SYan, Zheng 1023a22285a6SYan, Zheng key.offset = (u64)-1; 1024a22285a6SYan, Zheng pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); 1025a22285a6SYan, Zheng BUG_ON(IS_ERR(pending->snap)); 1026d68fc57bSYan, Zheng 10273fd0a558SYan, Zheng btrfs_reloc_post_snapshot(trans, pending); 1028d68fc57bSYan, Zheng btrfs_orphan_post_snapshot(trans, pending); 10293063d29fSChris Mason fail: 10306bdb72deSSage Weil kfree(new_root_item); 1031a22285a6SYan, Zheng btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); 1032a22285a6SYan, Zheng return 0; 10333063d29fSChris Mason } 10343063d29fSChris Mason 1035d352ac68SChris Mason /* 1036d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1037d352ac68SChris Mason */ 103880b6794dSChris Mason static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, 10393063d29fSChris Mason struct btrfs_fs_info *fs_info) 10403063d29fSChris Mason { 10413063d29fSChris Mason struct btrfs_pending_snapshot *pending; 10423063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 10433de4586cSChris Mason int ret; 10443de4586cSChris Mason 1045c6e30871SQinghuang Feng list_for_each_entry(pending, head, list) { 10463de4586cSChris Mason ret = create_pending_snapshot(trans, fs_info, pending); 10473de4586cSChris Mason BUG_ON(ret); 10483de4586cSChris Mason } 10493de4586cSChris Mason return 0; 10503de4586cSChris Mason } 10513de4586cSChris Mason 10525d4f98a2SYan Zheng static void update_super_roots(struct btrfs_root *root) 10535d4f98a2SYan Zheng { 10545d4f98a2SYan Zheng struct btrfs_root_item *root_item; 10555d4f98a2SYan Zheng struct btrfs_super_block *super; 10565d4f98a2SYan Zheng 10575d4f98a2SYan Zheng super = &root->fs_info->super_copy; 10585d4f98a2SYan Zheng 10595d4f98a2SYan Zheng root_item = &root->fs_info->chunk_root->root_item; 10605d4f98a2SYan Zheng super->chunk_root = root_item->bytenr; 10615d4f98a2SYan Zheng super->chunk_root_generation = root_item->generation; 10625d4f98a2SYan Zheng super->chunk_root_level = root_item->level; 10635d4f98a2SYan Zheng 10645d4f98a2SYan Zheng root_item = &root->fs_info->tree_root->root_item; 10655d4f98a2SYan Zheng super->root = root_item->bytenr; 10665d4f98a2SYan Zheng super->generation = root_item->generation; 10675d4f98a2SYan Zheng super->root_level = root_item->level; 10680af3d00bSJosef Bacik if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) 10690af3d00bSJosef Bacik super->cache_generation = root_item->generation; 10705d4f98a2SYan Zheng } 10715d4f98a2SYan Zheng 1072f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1073f36f3042SChris Mason { 1074f36f3042SChris Mason int ret = 0; 1075f36f3042SChris Mason spin_lock(&info->new_trans_lock); 1076f36f3042SChris Mason if (info->running_transaction) 1077f36f3042SChris Mason ret = info->running_transaction->in_commit; 1078f36f3042SChris Mason spin_unlock(&info->new_trans_lock); 1079f36f3042SChris Mason return ret; 1080f36f3042SChris Mason } 1081f36f3042SChris Mason 10828929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 10838929ecfaSYan, Zheng { 10848929ecfaSYan, Zheng int ret = 0; 10858929ecfaSYan, Zheng spin_lock(&info->new_trans_lock); 10868929ecfaSYan, Zheng if (info->running_transaction) 10878929ecfaSYan, Zheng ret = info->running_transaction->blocked; 10888929ecfaSYan, Zheng spin_unlock(&info->new_trans_lock); 10898929ecfaSYan, Zheng return ret; 10908929ecfaSYan, Zheng } 10918929ecfaSYan, Zheng 1092bb9c12c9SSage Weil /* 1093bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1094bb9c12c9SSage Weil * transaction joins 1095bb9c12c9SSage Weil */ 1096bb9c12c9SSage Weil static void wait_current_trans_commit_start(struct btrfs_root *root, 1097bb9c12c9SSage Weil struct btrfs_transaction *trans) 1098bb9c12c9SSage Weil { 1099bb9c12c9SSage Weil DEFINE_WAIT(wait); 1100bb9c12c9SSage Weil 1101bb9c12c9SSage Weil if (trans->in_commit) 1102bb9c12c9SSage Weil return; 1103bb9c12c9SSage Weil 1104bb9c12c9SSage Weil while (1) { 1105bb9c12c9SSage Weil prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, 1106bb9c12c9SSage Weil TASK_UNINTERRUPTIBLE); 1107bb9c12c9SSage Weil if (trans->in_commit) { 1108bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_blocked_wait, 1109bb9c12c9SSage Weil &wait); 1110bb9c12c9SSage Weil break; 1111bb9c12c9SSage Weil } 1112bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1113bb9c12c9SSage Weil schedule(); 1114bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1115bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_blocked_wait, &wait); 1116bb9c12c9SSage Weil } 1117bb9c12c9SSage Weil } 1118bb9c12c9SSage Weil 1119bb9c12c9SSage Weil /* 1120bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1121bb9c12c9SSage Weil * caller holds ref. 1122bb9c12c9SSage Weil */ 1123bb9c12c9SSage Weil static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, 1124bb9c12c9SSage Weil struct btrfs_transaction *trans) 1125bb9c12c9SSage Weil { 1126bb9c12c9SSage Weil DEFINE_WAIT(wait); 1127bb9c12c9SSage Weil 1128bb9c12c9SSage Weil if (trans->commit_done || (trans->in_commit && !trans->blocked)) 1129bb9c12c9SSage Weil return; 1130bb9c12c9SSage Weil 1131bb9c12c9SSage Weil while (1) { 1132bb9c12c9SSage Weil prepare_to_wait(&root->fs_info->transaction_wait, &wait, 1133bb9c12c9SSage Weil TASK_UNINTERRUPTIBLE); 1134bb9c12c9SSage Weil if (trans->commit_done || 1135bb9c12c9SSage Weil (trans->in_commit && !trans->blocked)) { 1136bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_wait, 1137bb9c12c9SSage Weil &wait); 1138bb9c12c9SSage Weil break; 1139bb9c12c9SSage Weil } 1140bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1141bb9c12c9SSage Weil schedule(); 1142bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1143bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_wait, 1144bb9c12c9SSage Weil &wait); 1145bb9c12c9SSage Weil } 1146bb9c12c9SSage Weil } 1147bb9c12c9SSage Weil 1148bb9c12c9SSage Weil /* 1149bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1150bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1151bb9c12c9SSage Weil */ 1152bb9c12c9SSage Weil struct btrfs_async_commit { 1153bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 1154bb9c12c9SSage Weil struct btrfs_root *root; 1155bb9c12c9SSage Weil struct delayed_work work; 1156bb9c12c9SSage Weil }; 1157bb9c12c9SSage Weil 1158bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1159bb9c12c9SSage Weil { 1160bb9c12c9SSage Weil struct btrfs_async_commit *ac = 1161bb9c12c9SSage Weil container_of(work, struct btrfs_async_commit, work.work); 1162bb9c12c9SSage Weil 1163bb9c12c9SSage Weil btrfs_commit_transaction(ac->newtrans, ac->root); 1164bb9c12c9SSage Weil kfree(ac); 1165bb9c12c9SSage Weil } 1166bb9c12c9SSage Weil 1167bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1168bb9c12c9SSage Weil struct btrfs_root *root, 1169bb9c12c9SSage Weil int wait_for_unblock) 1170bb9c12c9SSage Weil { 1171bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1172bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1173bb9c12c9SSage Weil 1174bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1175db5b493aSTsutomu Itoh if (!ac) 1176db5b493aSTsutomu Itoh return -ENOMEM; 1177bb9c12c9SSage Weil 1178bb9c12c9SSage Weil INIT_DELAYED_WORK(&ac->work, do_async_commit); 1179bb9c12c9SSage Weil ac->root = root; 1180bb9c12c9SSage Weil ac->newtrans = btrfs_join_transaction(root, 0); 11813612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 11823612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 11833612b495STsutomu Itoh kfree(ac); 11843612b495STsutomu Itoh return err; 11853612b495STsutomu Itoh } 1186bb9c12c9SSage Weil 1187bb9c12c9SSage Weil /* take transaction reference */ 1188bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1189bb9c12c9SSage Weil cur_trans = trans->transaction; 1190bb9c12c9SSage Weil cur_trans->use_count++; 1191bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1192bb9c12c9SSage Weil 1193bb9c12c9SSage Weil btrfs_end_transaction(trans, root); 1194bb9c12c9SSage Weil schedule_delayed_work(&ac->work, 0); 1195bb9c12c9SSage Weil 1196bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1197bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1198bb9c12c9SSage Weil if (wait_for_unblock) 1199bb9c12c9SSage Weil wait_current_trans_commit_start_and_unblock(root, cur_trans); 1200bb9c12c9SSage Weil else 1201bb9c12c9SSage Weil wait_current_trans_commit_start(root, cur_trans); 1202bb9c12c9SSage Weil put_transaction(cur_trans); 1203bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1204bb9c12c9SSage Weil 1205bb9c12c9SSage Weil return 0; 1206bb9c12c9SSage Weil } 1207bb9c12c9SSage Weil 1208bb9c12c9SSage Weil /* 1209bb9c12c9SSage Weil * btrfs_transaction state sequence: 1210bb9c12c9SSage Weil * in_commit = 0, blocked = 0 (initial) 1211bb9c12c9SSage Weil * in_commit = 1, blocked = 1 1212bb9c12c9SSage Weil * blocked = 0 1213bb9c12c9SSage Weil * commit_done = 1 1214bb9c12c9SSage Weil */ 121579154b1bSChris Mason int btrfs_commit_transaction(struct btrfs_trans_handle *trans, 121679154b1bSChris Mason struct btrfs_root *root) 121779154b1bSChris Mason { 121815ee9bc7SJosef Bacik unsigned long joined = 0; 121979154b1bSChris Mason struct btrfs_transaction *cur_trans; 12208fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 122179154b1bSChris Mason DEFINE_WAIT(wait); 122215ee9bc7SJosef Bacik int ret; 122389573b9cSChris Mason int should_grow = 0; 122489573b9cSChris Mason unsigned long now = get_seconds(); 1225dccae999SSage Weil int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); 122679154b1bSChris Mason 12275a3f23d5SChris Mason btrfs_run_ordered_operations(root, 0); 12285a3f23d5SChris Mason 122956bec294SChris Mason /* make a pass through all the delayed refs we have so far 123056bec294SChris Mason * any runnings procs may add more while we are here 123156bec294SChris Mason */ 123256bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 123356bec294SChris Mason BUG_ON(ret); 123456bec294SChris Mason 1235a22285a6SYan, Zheng btrfs_trans_release_metadata(trans, root); 1236a22285a6SYan, Zheng 1237b7ec40d7SChris Mason cur_trans = trans->transaction; 123856bec294SChris Mason /* 123956bec294SChris Mason * set the flushing flag so procs in this transaction have to 124056bec294SChris Mason * start sending their work down. 124156bec294SChris Mason */ 1242b7ec40d7SChris Mason cur_trans->delayed_refs.flushing = 1; 124356bec294SChris Mason 1244c3e69d58SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 124556bec294SChris Mason BUG_ON(ret); 124656bec294SChris Mason 124779154b1bSChris Mason mutex_lock(&root->fs_info->trans_mutex); 1248b7ec40d7SChris Mason if (cur_trans->in_commit) { 1249b7ec40d7SChris Mason cur_trans->use_count++; 1250ccd467d6SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 125179154b1bSChris Mason btrfs_end_transaction(trans, root); 1252ccd467d6SChris Mason 125379154b1bSChris Mason ret = wait_for_commit(root, cur_trans); 125479154b1bSChris Mason BUG_ON(ret); 125515ee9bc7SJosef Bacik 125615ee9bc7SJosef Bacik mutex_lock(&root->fs_info->trans_mutex); 125779154b1bSChris Mason put_transaction(cur_trans); 125815ee9bc7SJosef Bacik mutex_unlock(&root->fs_info->trans_mutex); 125915ee9bc7SJosef Bacik 126079154b1bSChris Mason return 0; 126179154b1bSChris Mason } 12624313b399SChris Mason 12632c90e5d6SChris Mason trans->transaction->in_commit = 1; 1264f9295749SChris Mason trans->transaction->blocked = 1; 1265bb9c12c9SSage Weil wake_up(&root->fs_info->transaction_blocked_wait); 1266bb9c12c9SSage Weil 1267ccd467d6SChris Mason if (cur_trans->list.prev != &root->fs_info->trans_list) { 1268ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 1269ccd467d6SChris Mason struct btrfs_transaction, list); 1270ccd467d6SChris Mason if (!prev_trans->commit_done) { 1271ccd467d6SChris Mason prev_trans->use_count++; 1272ccd467d6SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 1273ccd467d6SChris Mason 1274ccd467d6SChris Mason wait_for_commit(root, prev_trans); 1275ccd467d6SChris Mason 1276ccd467d6SChris Mason mutex_lock(&root->fs_info->trans_mutex); 127715ee9bc7SJosef Bacik put_transaction(prev_trans); 1278ccd467d6SChris Mason } 1279ccd467d6SChris Mason } 128015ee9bc7SJosef Bacik 128189573b9cSChris Mason if (now < cur_trans->start_time || now - cur_trans->start_time < 1) 128289573b9cSChris Mason should_grow = 1; 128389573b9cSChris Mason 128415ee9bc7SJosef Bacik do { 12857ea394f1SYan Zheng int snap_pending = 0; 128615ee9bc7SJosef Bacik joined = cur_trans->num_joined; 12877ea394f1SYan Zheng if (!list_empty(&trans->transaction->pending_snapshots)) 12887ea394f1SYan Zheng snap_pending = 1; 12897ea394f1SYan Zheng 12902c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 129179154b1bSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 129215ee9bc7SJosef Bacik 12930bdb1db2SSage Weil if (flush_on_commit || snap_pending) { 129424bbcf04SYan, Zheng btrfs_start_delalloc_inodes(root, 1); 129524bbcf04SYan, Zheng ret = btrfs_wait_ordered_extents(root, 0, 1); 1296ebecd3d9SSage Weil BUG_ON(ret); 12977ea394f1SYan Zheng } 12987ea394f1SYan Zheng 12995a3f23d5SChris Mason /* 13005a3f23d5SChris Mason * rename don't use btrfs_join_transaction, so, once we 13015a3f23d5SChris Mason * set the transaction to blocked above, we aren't going 13025a3f23d5SChris Mason * to get any new ordered operations. We can safely run 13035a3f23d5SChris Mason * it here and no for sure that nothing new will be added 13045a3f23d5SChris Mason * to the list 13055a3f23d5SChris Mason */ 13065a3f23d5SChris Mason btrfs_run_ordered_operations(root, 1); 13075a3f23d5SChris Mason 1308ed3b3d31SChris Mason prepare_to_wait(&cur_trans->writer_wait, &wait, 1309ed3b3d31SChris Mason TASK_UNINTERRUPTIBLE); 1310ed3b3d31SChris Mason 131189573b9cSChris Mason smp_mb(); 131299d16cbcSSage Weil if (cur_trans->num_writers > 1) 131399d16cbcSSage Weil schedule_timeout(MAX_SCHEDULE_TIMEOUT); 131499d16cbcSSage Weil else if (should_grow) 131599d16cbcSSage Weil schedule_timeout(1); 131615ee9bc7SJosef Bacik 131779154b1bSChris Mason mutex_lock(&root->fs_info->trans_mutex); 131815ee9bc7SJosef Bacik finish_wait(&cur_trans->writer_wait, &wait); 131915ee9bc7SJosef Bacik } while (cur_trans->num_writers > 1 || 132089573b9cSChris Mason (should_grow && cur_trans->num_joined != joined)); 132115ee9bc7SJosef Bacik 13223063d29fSChris Mason ret = create_pending_snapshots(trans, root->fs_info); 13233063d29fSChris Mason BUG_ON(ret); 13243063d29fSChris Mason 132556bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 132656bec294SChris Mason BUG_ON(ret); 132756bec294SChris Mason 13282c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 1329dc17ff8fSChris Mason 1330e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 1331e02119d5SChris Mason * various roots consistent with each other. Every pointer 1332e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 1333e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 1334e02119d5SChris Mason * the tree logging code from jumping in and changing any 1335e02119d5SChris Mason * of the trees. 1336e02119d5SChris Mason * 1337e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 1338e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 1339e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 1340e02119d5SChris Mason * from now until after the super is written, we avoid races 1341e02119d5SChris Mason * with the tree-log code. 1342e02119d5SChris Mason */ 1343e02119d5SChris Mason mutex_lock(&root->fs_info->tree_log_mutex); 13441a40e23bSZheng Yan 13455d4f98a2SYan Zheng ret = commit_fs_roots(trans, root); 134654aa1f4dSChris Mason BUG_ON(ret); 134754aa1f4dSChris Mason 13485d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 1349e02119d5SChris Mason * safe to free the root of tree log roots 1350e02119d5SChris Mason */ 1351e02119d5SChris Mason btrfs_free_log_root_tree(trans, root->fs_info); 1352e02119d5SChris Mason 13535d4f98a2SYan Zheng ret = commit_cowonly_roots(trans, root); 135479154b1bSChris Mason BUG_ON(ret); 135554aa1f4dSChris Mason 135611833d66SYan Zheng btrfs_prepare_extent_commit(trans, root); 135711833d66SYan Zheng 135878fae27eSChris Mason cur_trans = root->fs_info->running_transaction; 1359cee36a03SChris Mason spin_lock(&root->fs_info->new_trans_lock); 136078fae27eSChris Mason root->fs_info->running_transaction = NULL; 1361cee36a03SChris Mason spin_unlock(&root->fs_info->new_trans_lock); 13625f39d397SChris Mason 13635d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->tree_root->root_item, 13645d4f98a2SYan Zheng root->fs_info->tree_root->node); 1365817d52f8SJosef Bacik switch_commit_root(root->fs_info->tree_root); 13665d4f98a2SYan Zheng 13675d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->chunk_root->root_item, 13685d4f98a2SYan Zheng root->fs_info->chunk_root->node); 1369817d52f8SJosef Bacik switch_commit_root(root->fs_info->chunk_root); 13705d4f98a2SYan Zheng 13715d4f98a2SYan Zheng update_super_roots(root); 1372e02119d5SChris Mason 1373e02119d5SChris Mason if (!root->fs_info->log_root_recovering) { 1374e02119d5SChris Mason btrfs_set_super_log_root(&root->fs_info->super_copy, 0); 1375e02119d5SChris Mason btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); 1376e02119d5SChris Mason } 1377e02119d5SChris Mason 1378a061fc8dSChris Mason memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, 13794b52dff6SChris Mason sizeof(root->fs_info->super_copy)); 1380ccd467d6SChris Mason 1381f9295749SChris Mason trans->transaction->blocked = 0; 1382b7ec40d7SChris Mason 1383f9295749SChris Mason wake_up(&root->fs_info->transaction_wait); 1384e6dcd2dcSChris Mason 138578fae27eSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 138679154b1bSChris Mason ret = btrfs_write_and_wait_transaction(trans, root); 138779154b1bSChris Mason BUG_ON(ret); 1388a512bbf8SYan Zheng write_ctree_super(trans, root, 0); 13894313b399SChris Mason 1390e02119d5SChris Mason /* 1391e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 1392e02119d5SChris Mason * to go about their business 1393e02119d5SChris Mason */ 1394e02119d5SChris Mason mutex_unlock(&root->fs_info->tree_log_mutex); 1395e02119d5SChris Mason 139611833d66SYan Zheng btrfs_finish_extent_commit(trans, root); 13974313b399SChris Mason 13981a40e23bSZheng Yan mutex_lock(&root->fs_info->trans_mutex); 13991a40e23bSZheng Yan 14002c90e5d6SChris Mason cur_trans->commit_done = 1; 1401b7ec40d7SChris Mason 140215ee9bc7SJosef Bacik root->fs_info->last_trans_committed = cur_trans->transid; 1403817d52f8SJosef Bacik 14042c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 14053de4586cSChris Mason 140679154b1bSChris Mason put_transaction(cur_trans); 140778fae27eSChris Mason put_transaction(cur_trans); 140858176a96SJosef Bacik 14091abe9b8aSliubo trace_btrfs_transaction_commit(root); 14101abe9b8aSliubo 141178fae27eSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 14123de4586cSChris Mason 14139ed74f2dSJosef Bacik if (current->journal_info == trans) 14149ed74f2dSJosef Bacik current->journal_info = NULL; 14159ed74f2dSJosef Bacik 14162c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 141724bbcf04SYan, Zheng 141824bbcf04SYan, Zheng if (current != root->fs_info->transaction_kthread) 141924bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 142024bbcf04SYan, Zheng 142179154b1bSChris Mason return ret; 142279154b1bSChris Mason } 142379154b1bSChris Mason 1424d352ac68SChris Mason /* 1425d352ac68SChris Mason * interface function to delete all the snapshots we have scheduled for deletion 1426d352ac68SChris Mason */ 1427e9d0b13bSChris Mason int btrfs_clean_old_snapshots(struct btrfs_root *root) 1428e9d0b13bSChris Mason { 14295d4f98a2SYan Zheng LIST_HEAD(list); 14305d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 1431e9d0b13bSChris Mason 14325d4f98a2SYan Zheng mutex_lock(&fs_info->trans_mutex); 14335d4f98a2SYan Zheng list_splice_init(&fs_info->dead_roots, &list); 14345d4f98a2SYan Zheng mutex_unlock(&fs_info->trans_mutex); 14355d4f98a2SYan Zheng 14365d4f98a2SYan Zheng while (!list_empty(&list)) { 14375d4f98a2SYan Zheng root = list_entry(list.next, struct btrfs_root, root_list); 143876dda93cSYan, Zheng list_del(&root->root_list); 143976dda93cSYan, Zheng 144076dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 144176dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 14423fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 0); 144376dda93cSYan, Zheng else 14443fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 1); 1445e9d0b13bSChris Mason } 1446e9d0b13bSChris Mason return 0; 1447e9d0b13bSChris Mason } 1448