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 { 3513c5a93eSJosef Bacik WARN_ON(atomic_read(&transaction->use_count) == 0); 3613c5a93eSJosef Bacik if (atomic_dec_and_test(&transaction->use_count)) { 372c90e5d6SChris Mason memset(transaction, 0, sizeof(*transaction)); 382c90e5d6SChris Mason kmem_cache_free(btrfs_transaction_cachep, transaction); 3979154b1bSChris Mason } 4078fae27eSChris Mason } 4179154b1bSChris Mason 42817d52f8SJosef Bacik static noinline void switch_commit_root(struct btrfs_root *root) 43817d52f8SJosef Bacik { 44817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 45817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 46817d52f8SJosef Bacik } 47817d52f8SJosef Bacik 48d352ac68SChris Mason /* 49d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 50d352ac68SChris Mason */ 5180b6794dSChris Mason static noinline int join_transaction(struct btrfs_root *root) 5279154b1bSChris Mason { 5379154b1bSChris Mason struct btrfs_transaction *cur_trans; 5479154b1bSChris Mason cur_trans = root->fs_info->running_transaction; 5579154b1bSChris Mason if (!cur_trans) { 562c90e5d6SChris Mason cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, 572c90e5d6SChris Mason GFP_NOFS); 58db5b493aSTsutomu Itoh if (!cur_trans) 59db5b493aSTsutomu Itoh return -ENOMEM; 600f7d52f4SChris Mason root->fs_info->generation++; 6113c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 6215ee9bc7SJosef Bacik cur_trans->num_joined = 0; 630f7d52f4SChris Mason cur_trans->transid = root->fs_info->generation; 6479154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 6579154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 6679154b1bSChris Mason cur_trans->in_commit = 0; 67f9295749SChris Mason cur_trans->blocked = 0; 6813c5a93eSJosef Bacik atomic_set(&cur_trans->use_count, 1); 6979154b1bSChris Mason cur_trans->commit_done = 0; 7008607c1bSChris Mason cur_trans->start_time = get_seconds(); 7156bec294SChris Mason 726bef4d31SEric Paris cur_trans->delayed_refs.root = RB_ROOT; 7356bec294SChris Mason cur_trans->delayed_refs.num_entries = 0; 74c3e69d58SChris Mason cur_trans->delayed_refs.num_heads_ready = 0; 75c3e69d58SChris Mason cur_trans->delayed_refs.num_heads = 0; 7656bec294SChris Mason cur_trans->delayed_refs.flushing = 0; 77c3e69d58SChris Mason cur_trans->delayed_refs.run_delayed_start = 0; 7856bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 7956bec294SChris Mason 803063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 818fd17795SChris Mason list_add_tail(&cur_trans->list, &root->fs_info->trans_list); 82d1310b2eSChris Mason extent_io_tree_init(&cur_trans->dirty_pages, 83f993c883SDavid Sterba root->fs_info->btree_inode->i_mapping); 8448ec2cf8SChris Mason spin_lock(&root->fs_info->new_trans_lock); 8548ec2cf8SChris Mason root->fs_info->running_transaction = cur_trans; 8648ec2cf8SChris Mason spin_unlock(&root->fs_info->new_trans_lock); 8715ee9bc7SJosef Bacik } else { 8813c5a93eSJosef Bacik atomic_inc(&cur_trans->num_writers); 8915ee9bc7SJosef Bacik cur_trans->num_joined++; 9015ee9bc7SJosef Bacik } 9115ee9bc7SJosef Bacik 9279154b1bSChris Mason return 0; 9379154b1bSChris Mason } 9479154b1bSChris Mason 95d352ac68SChris Mason /* 96d397712bSChris Mason * this does all the record keeping required to make sure that a reference 97d397712bSChris Mason * counted root is properly recorded in a given transaction. This is required 98d397712bSChris Mason * to make sure the old root from before we joined the transaction is deleted 99d397712bSChris Mason * when the transaction commits 100d352ac68SChris Mason */ 1015d4f98a2SYan Zheng static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, 1025d4f98a2SYan Zheng struct btrfs_root *root) 1036702ed49SChris Mason { 1045d4f98a2SYan Zheng if (root->ref_cows && root->last_trans < trans->transid) { 1056702ed49SChris Mason WARN_ON(root == root->fs_info->extent_root); 1065d4f98a2SYan Zheng WARN_ON(root->commit_root != root->node); 1075d4f98a2SYan Zheng 1086702ed49SChris Mason radix_tree_tag_set(&root->fs_info->fs_roots_radix, 1096702ed49SChris Mason (unsigned long)root->root_key.objectid, 1106702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 1115d4f98a2SYan Zheng root->last_trans = trans->transid; 1125d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 1136702ed49SChris Mason } 1145d4f98a2SYan Zheng return 0; 1156702ed49SChris Mason } 1165d4f98a2SYan Zheng 1175d4f98a2SYan Zheng int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 1185d4f98a2SYan Zheng struct btrfs_root *root) 1195d4f98a2SYan Zheng { 1205d4f98a2SYan Zheng if (!root->ref_cows) 1215d4f98a2SYan Zheng return 0; 1225d4f98a2SYan Zheng 1235d4f98a2SYan Zheng mutex_lock(&root->fs_info->trans_mutex); 1245d4f98a2SYan Zheng if (root->last_trans == trans->transid) { 1255d4f98a2SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 1265d4f98a2SYan Zheng return 0; 1275d4f98a2SYan Zheng } 1285d4f98a2SYan Zheng 1295d4f98a2SYan Zheng record_root_in_trans(trans, root); 1305d4f98a2SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 1316702ed49SChris Mason return 0; 1326702ed49SChris Mason } 1336702ed49SChris Mason 134d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 135d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 136d352ac68SChris Mason * transaction might not be fully on disk. 137d352ac68SChris Mason */ 13837d1aeeeSChris Mason static void wait_current_trans(struct btrfs_root *root) 13979154b1bSChris Mason { 140f9295749SChris Mason struct btrfs_transaction *cur_trans; 14179154b1bSChris Mason 142f9295749SChris Mason cur_trans = root->fs_info->running_transaction; 14337d1aeeeSChris Mason if (cur_trans && cur_trans->blocked) { 144f9295749SChris Mason DEFINE_WAIT(wait); 14513c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 146f9295749SChris Mason while (1) { 147f9295749SChris Mason prepare_to_wait(&root->fs_info->transaction_wait, &wait, 148f9295749SChris Mason TASK_UNINTERRUPTIBLE); 149471fa17dSZhao Lei if (!cur_trans->blocked) 150471fa17dSZhao Lei break; 151f9295749SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 152f9295749SChris Mason schedule(); 153f9295749SChris Mason mutex_lock(&root->fs_info->trans_mutex); 154f9295749SChris Mason } 155471fa17dSZhao Lei finish_wait(&root->fs_info->transaction_wait, &wait); 156f9295749SChris Mason put_transaction(cur_trans); 157f9295749SChris Mason } 15837d1aeeeSChris Mason } 15937d1aeeeSChris Mason 160249ac1e5SJosef Bacik enum btrfs_trans_type { 161249ac1e5SJosef Bacik TRANS_START, 162249ac1e5SJosef Bacik TRANS_JOIN, 163249ac1e5SJosef Bacik TRANS_USERSPACE, 1640af3d00bSJosef Bacik TRANS_JOIN_NOLOCK, 165249ac1e5SJosef Bacik }; 166249ac1e5SJosef Bacik 167a22285a6SYan, Zheng static int may_wait_transaction(struct btrfs_root *root, int type) 16837d1aeeeSChris Mason { 1694bef0848SChris Mason if (!root->fs_info->log_root_recovering && 170249ac1e5SJosef Bacik ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || 171249ac1e5SJosef Bacik type == TRANS_USERSPACE)) 172a22285a6SYan, Zheng return 1; 173a22285a6SYan, Zheng return 0; 174a22285a6SYan, Zheng } 175a22285a6SYan, Zheng 176a22285a6SYan, Zheng static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 177a22285a6SYan, Zheng u64 num_items, int type) 178a22285a6SYan, Zheng { 179a22285a6SYan, Zheng struct btrfs_trans_handle *h; 180a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 18106d5a589SJosef Bacik int retries = 0; 182a22285a6SYan, Zheng int ret; 183acce952bSliubo 184acce952bSliubo if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 185acce952bSliubo return ERR_PTR(-EROFS); 186a22285a6SYan, Zheng again: 187a22285a6SYan, Zheng h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); 188a22285a6SYan, Zheng if (!h) 189a22285a6SYan, Zheng return ERR_PTR(-ENOMEM); 190a22285a6SYan, Zheng 1910af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 192a22285a6SYan, Zheng mutex_lock(&root->fs_info->trans_mutex); 193a22285a6SYan, Zheng if (may_wait_transaction(root, type)) 19437d1aeeeSChris Mason wait_current_trans(root); 195a22285a6SYan, Zheng 19679154b1bSChris Mason ret = join_transaction(root); 197db5b493aSTsutomu Itoh if (ret < 0) { 1986e8df2aeSYoshinori Sano kmem_cache_free(btrfs_trans_handle_cachep, h); 199db5b493aSTsutomu Itoh if (type != TRANS_JOIN_NOLOCK) 200db5b493aSTsutomu Itoh mutex_unlock(&root->fs_info->trans_mutex); 201db5b493aSTsutomu Itoh return ERR_PTR(ret); 202db5b493aSTsutomu Itoh } 2030f7d52f4SChris Mason 204a22285a6SYan, Zheng cur_trans = root->fs_info->running_transaction; 20513c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 2060af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 207a22285a6SYan, Zheng mutex_unlock(&root->fs_info->trans_mutex); 208a22285a6SYan, Zheng 209a22285a6SYan, Zheng h->transid = cur_trans->transid; 210a22285a6SYan, Zheng h->transaction = cur_trans; 21179154b1bSChris Mason h->blocks_used = 0; 212d2fb3437SYan Zheng h->block_group = 0; 213a22285a6SYan, Zheng h->bytes_reserved = 0; 21456bec294SChris Mason h->delayed_ref_updates = 0; 215f0486c68SYan, Zheng h->block_rsv = NULL; 216b7ec40d7SChris Mason 217a22285a6SYan, Zheng smp_mb(); 218a22285a6SYan, Zheng if (cur_trans->blocked && may_wait_transaction(root, type)) { 219a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 220a22285a6SYan, Zheng goto again; 221a22285a6SYan, Zheng } 2229ed74f2dSJosef Bacik 223a22285a6SYan, Zheng if (num_items > 0) { 2248bb8ab2eSJosef Bacik ret = btrfs_trans_reserve_metadata(h, root, num_items); 22506d5a589SJosef Bacik if (ret == -EAGAIN && !retries) { 22606d5a589SJosef Bacik retries++; 227a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 228a22285a6SYan, Zheng goto again; 22906d5a589SJosef Bacik } else if (ret == -EAGAIN) { 23006d5a589SJosef Bacik /* 23106d5a589SJosef Bacik * We have already retried and got EAGAIN, so really we 23206d5a589SJosef Bacik * don't have space, so set ret to -ENOSPC. 23306d5a589SJosef Bacik */ 23406d5a589SJosef Bacik ret = -ENOSPC; 235a22285a6SYan, Zheng } 23606d5a589SJosef Bacik 237a22285a6SYan, Zheng if (ret < 0) { 238a22285a6SYan, Zheng btrfs_end_transaction(h, root); 239a22285a6SYan, Zheng return ERR_PTR(ret); 240a22285a6SYan, Zheng } 241a22285a6SYan, Zheng } 242a22285a6SYan, Zheng 2430af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 244a22285a6SYan, Zheng mutex_lock(&root->fs_info->trans_mutex); 2455d4f98a2SYan Zheng record_root_in_trans(h, root); 2460af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 24779154b1bSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 248a22285a6SYan, Zheng 249a22285a6SYan, Zheng if (!current->journal_info && type != TRANS_USERSPACE) 250a22285a6SYan, Zheng current->journal_info = h; 25179154b1bSChris Mason return h; 25279154b1bSChris Mason } 25379154b1bSChris Mason 254f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 255a22285a6SYan, Zheng int num_items) 256f9295749SChris Mason { 257a22285a6SYan, Zheng return start_transaction(root, num_items, TRANS_START); 258f9295749SChris Mason } 259f9295749SChris Mason struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, 260f9295749SChris Mason int num_blocks) 261f9295749SChris Mason { 262a22285a6SYan, Zheng return start_transaction(root, 0, TRANS_JOIN); 263f9295749SChris Mason } 264f9295749SChris Mason 2650af3d00bSJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, 2660af3d00bSJosef Bacik int num_blocks) 2670af3d00bSJosef Bacik { 2680af3d00bSJosef Bacik return start_transaction(root, 0, TRANS_JOIN_NOLOCK); 2690af3d00bSJosef Bacik } 2700af3d00bSJosef Bacik 2719ca9ee09SSage Weil struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, 2729ca9ee09SSage Weil int num_blocks) 2739ca9ee09SSage Weil { 274a22285a6SYan, Zheng return start_transaction(r, 0, TRANS_USERSPACE); 2759ca9ee09SSage Weil } 2769ca9ee09SSage Weil 277d352ac68SChris Mason /* wait for a transaction commit to be fully complete */ 27889ce8a63SChris Mason static noinline int wait_for_commit(struct btrfs_root *root, 27989ce8a63SChris Mason struct btrfs_transaction *commit) 28089ce8a63SChris Mason { 28189ce8a63SChris Mason DEFINE_WAIT(wait); 28289ce8a63SChris Mason mutex_lock(&root->fs_info->trans_mutex); 28389ce8a63SChris Mason while (!commit->commit_done) { 28489ce8a63SChris Mason prepare_to_wait(&commit->commit_wait, &wait, 28589ce8a63SChris Mason TASK_UNINTERRUPTIBLE); 28689ce8a63SChris Mason if (commit->commit_done) 28789ce8a63SChris Mason break; 28889ce8a63SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 28989ce8a63SChris Mason schedule(); 29089ce8a63SChris Mason mutex_lock(&root->fs_info->trans_mutex); 29189ce8a63SChris Mason } 29289ce8a63SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 29389ce8a63SChris Mason finish_wait(&commit->commit_wait, &wait); 29489ce8a63SChris Mason return 0; 29589ce8a63SChris Mason } 29689ce8a63SChris Mason 29746204592SSage Weil int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) 29846204592SSage Weil { 29946204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 30046204592SSage Weil int ret; 30146204592SSage Weil 30246204592SSage Weil mutex_lock(&root->fs_info->trans_mutex); 30346204592SSage Weil 30446204592SSage Weil ret = 0; 30546204592SSage Weil if (transid) { 30646204592SSage Weil if (transid <= root->fs_info->last_trans_committed) 30746204592SSage Weil goto out_unlock; 30846204592SSage Weil 30946204592SSage Weil /* find specified transaction */ 31046204592SSage Weil list_for_each_entry(t, &root->fs_info->trans_list, list) { 31146204592SSage Weil if (t->transid == transid) { 31246204592SSage Weil cur_trans = t; 31346204592SSage Weil break; 31446204592SSage Weil } 31546204592SSage Weil if (t->transid > transid) 31646204592SSage Weil break; 31746204592SSage Weil } 31846204592SSage Weil ret = -EINVAL; 31946204592SSage Weil if (!cur_trans) 32046204592SSage Weil goto out_unlock; /* bad transid */ 32146204592SSage Weil } else { 32246204592SSage Weil /* find newest transaction that is committing | committed */ 32346204592SSage Weil list_for_each_entry_reverse(t, &root->fs_info->trans_list, 32446204592SSage Weil list) { 32546204592SSage Weil if (t->in_commit) { 32646204592SSage Weil if (t->commit_done) 32746204592SSage Weil goto out_unlock; 32846204592SSage Weil cur_trans = t; 32946204592SSage Weil break; 33046204592SSage Weil } 33146204592SSage Weil } 33246204592SSage Weil if (!cur_trans) 33346204592SSage Weil goto out_unlock; /* nothing committing|committed */ 33446204592SSage Weil } 33546204592SSage Weil 33613c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 33746204592SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 33846204592SSage Weil 33946204592SSage Weil wait_for_commit(root, cur_trans); 34046204592SSage Weil 34146204592SSage Weil mutex_lock(&root->fs_info->trans_mutex); 34246204592SSage Weil put_transaction(cur_trans); 34346204592SSage Weil ret = 0; 34446204592SSage Weil out_unlock: 34546204592SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 34646204592SSage Weil return ret; 34746204592SSage Weil } 34846204592SSage Weil 3495d4f98a2SYan Zheng #if 0 350d352ac68SChris Mason /* 351d397712bSChris Mason * rate limit against the drop_snapshot code. This helps to slow down new 352d397712bSChris Mason * operations if the drop_snapshot code isn't able to keep up. 353d352ac68SChris Mason */ 35437d1aeeeSChris Mason static void throttle_on_drops(struct btrfs_root *root) 355ab78c84dSChris Mason { 356ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 3572dd3e67bSChris Mason int harder_count = 0; 358ab78c84dSChris Mason 3592dd3e67bSChris Mason harder: 360ab78c84dSChris Mason if (atomic_read(&info->throttles)) { 361ab78c84dSChris Mason DEFINE_WAIT(wait); 362ab78c84dSChris Mason int thr; 363ab78c84dSChris Mason thr = atomic_read(&info->throttle_gen); 364ab78c84dSChris Mason 365ab78c84dSChris Mason do { 366ab78c84dSChris Mason prepare_to_wait(&info->transaction_throttle, 367ab78c84dSChris Mason &wait, TASK_UNINTERRUPTIBLE); 368ab78c84dSChris Mason if (!atomic_read(&info->throttles)) { 369ab78c84dSChris Mason finish_wait(&info->transaction_throttle, &wait); 370ab78c84dSChris Mason break; 371ab78c84dSChris Mason } 372ab78c84dSChris Mason schedule(); 373ab78c84dSChris Mason finish_wait(&info->transaction_throttle, &wait); 374ab78c84dSChris Mason } while (thr == atomic_read(&info->throttle_gen)); 3752dd3e67bSChris Mason harder_count++; 3762dd3e67bSChris Mason 3772dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && 3782dd3e67bSChris Mason harder_count < 2) 3792dd3e67bSChris Mason goto harder; 3802dd3e67bSChris Mason 3812dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && 3822dd3e67bSChris Mason harder_count < 10) 3832dd3e67bSChris Mason goto harder; 3842dd3e67bSChris Mason 3852dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && 3862dd3e67bSChris Mason harder_count < 20) 3872dd3e67bSChris Mason goto harder; 388ab78c84dSChris Mason } 389ab78c84dSChris Mason } 3905d4f98a2SYan Zheng #endif 391ab78c84dSChris Mason 39237d1aeeeSChris Mason void btrfs_throttle(struct btrfs_root *root) 39337d1aeeeSChris Mason { 39437d1aeeeSChris Mason mutex_lock(&root->fs_info->trans_mutex); 3959ca9ee09SSage Weil if (!root->fs_info->open_ioctl_trans) 39637d1aeeeSChris Mason wait_current_trans(root); 39737d1aeeeSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 39837d1aeeeSChris Mason } 39937d1aeeeSChris Mason 4008929ecfaSYan, Zheng static int should_end_transaction(struct btrfs_trans_handle *trans, 4018929ecfaSYan, Zheng struct btrfs_root *root) 4028929ecfaSYan, Zheng { 4038929ecfaSYan, Zheng int ret; 4048929ecfaSYan, Zheng ret = btrfs_block_rsv_check(trans, root, 4058929ecfaSYan, Zheng &root->fs_info->global_block_rsv, 0, 5); 4068929ecfaSYan, Zheng return ret ? 1 : 0; 4078929ecfaSYan, Zheng } 4088929ecfaSYan, Zheng 4098929ecfaSYan, Zheng int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, 4108929ecfaSYan, Zheng struct btrfs_root *root) 4118929ecfaSYan, Zheng { 4128929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 4138929ecfaSYan, Zheng int updates; 4148929ecfaSYan, Zheng 4158929ecfaSYan, Zheng if (cur_trans->blocked || cur_trans->delayed_refs.flushing) 4168929ecfaSYan, Zheng return 1; 4178929ecfaSYan, Zheng 4188929ecfaSYan, Zheng updates = trans->delayed_ref_updates; 4198929ecfaSYan, Zheng trans->delayed_ref_updates = 0; 4208929ecfaSYan, Zheng if (updates) 4218929ecfaSYan, Zheng btrfs_run_delayed_refs(trans, root, updates); 4228929ecfaSYan, Zheng 4238929ecfaSYan, Zheng return should_end_transaction(trans, root); 4248929ecfaSYan, Zheng } 4258929ecfaSYan, Zheng 42689ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 4270af3d00bSJosef Bacik struct btrfs_root *root, int throttle, int lock) 42879154b1bSChris Mason { 4298929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 430ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 431c3e69d58SChris Mason int count = 0; 432d6e4a428SChris Mason 433c3e69d58SChris Mason while (count < 4) { 434c3e69d58SChris Mason unsigned long cur = trans->delayed_ref_updates; 435c3e69d58SChris Mason trans->delayed_ref_updates = 0; 436c3e69d58SChris Mason if (cur && 437c3e69d58SChris Mason trans->transaction->delayed_refs.num_heads_ready > 64) { 438c3e69d58SChris Mason trans->delayed_ref_updates = 0; 439b7ec40d7SChris Mason 440b7ec40d7SChris Mason /* 441b7ec40d7SChris Mason * do a full flush if the transaction is trying 442b7ec40d7SChris Mason * to close 443b7ec40d7SChris Mason */ 444b7ec40d7SChris Mason if (trans->transaction->delayed_refs.flushing) 445b7ec40d7SChris Mason cur = 0; 446c3e69d58SChris Mason btrfs_run_delayed_refs(trans, root, cur); 447c3e69d58SChris Mason } else { 448c3e69d58SChris Mason break; 449c3e69d58SChris Mason } 450c3e69d58SChris Mason count++; 45156bec294SChris Mason } 45256bec294SChris Mason 453a22285a6SYan, Zheng btrfs_trans_release_metadata(trans, root); 454a22285a6SYan, Zheng 4550af3d00bSJosef Bacik if (lock && !root->fs_info->open_ioctl_trans && 4568929ecfaSYan, Zheng should_end_transaction(trans, root)) 4578929ecfaSYan, Zheng trans->transaction->blocked = 1; 4588929ecfaSYan, Zheng 4590af3d00bSJosef Bacik if (lock && cur_trans->blocked && !cur_trans->in_commit) { 4608929ecfaSYan, Zheng if (throttle) 4618929ecfaSYan, Zheng return btrfs_commit_transaction(trans, root); 4628929ecfaSYan, Zheng else 4638929ecfaSYan, Zheng wake_up_process(info->transaction_kthread); 4648929ecfaSYan, Zheng } 4658929ecfaSYan, Zheng 4668929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 46713c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 46813c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 46989ce8a63SChris Mason 47099d16cbcSSage Weil smp_mb(); 47179154b1bSChris Mason if (waitqueue_active(&cur_trans->writer_wait)) 47279154b1bSChris Mason wake_up(&cur_trans->writer_wait); 47379154b1bSChris Mason put_transaction(cur_trans); 4749ed74f2dSJosef Bacik 4759ed74f2dSJosef Bacik if (current->journal_info == trans) 4769ed74f2dSJosef Bacik current->journal_info = NULL; 477d6025579SChris Mason memset(trans, 0, sizeof(*trans)); 4782c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 479ab78c84dSChris Mason 48024bbcf04SYan, Zheng if (throttle) 48124bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 48224bbcf04SYan, Zheng 48379154b1bSChris Mason return 0; 48479154b1bSChris Mason } 48579154b1bSChris Mason 48689ce8a63SChris Mason int btrfs_end_transaction(struct btrfs_trans_handle *trans, 48789ce8a63SChris Mason struct btrfs_root *root) 48889ce8a63SChris Mason { 4890af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 0, 1); 49089ce8a63SChris Mason } 49189ce8a63SChris Mason 49289ce8a63SChris Mason int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, 49389ce8a63SChris Mason struct btrfs_root *root) 49489ce8a63SChris Mason { 4950af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 1, 1); 4960af3d00bSJosef Bacik } 4970af3d00bSJosef Bacik 4980af3d00bSJosef Bacik int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, 4990af3d00bSJosef Bacik struct btrfs_root *root) 5000af3d00bSJosef Bacik { 5010af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 0, 0); 50289ce8a63SChris Mason } 50389ce8a63SChris Mason 504d352ac68SChris Mason /* 505d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 506d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 507690587d1SChris Mason * those extents are sent to disk but does not wait on them 508d352ac68SChris Mason */ 509690587d1SChris Mason int btrfs_write_marked_extents(struct btrfs_root *root, 5108cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 51179154b1bSChris Mason { 5127c4452b9SChris Mason int ret; 513777e6bd7SChris Mason int err = 0; 5147c4452b9SChris Mason int werr = 0; 5157c4452b9SChris Mason struct page *page; 5167c4452b9SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 517777e6bd7SChris Mason u64 start = 0; 5185f39d397SChris Mason u64 end; 5195f39d397SChris Mason unsigned long index; 5207c4452b9SChris Mason 5217c4452b9SChris Mason while (1) { 522777e6bd7SChris Mason ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5238cef4e16SYan, Zheng mark); 5245f39d397SChris Mason if (ret) 5257c4452b9SChris Mason break; 5265f39d397SChris Mason while (start <= end) { 527777e6bd7SChris Mason cond_resched(); 528777e6bd7SChris Mason 5295f39d397SChris Mason index = start >> PAGE_CACHE_SHIFT; 53035ebb934SChris Mason start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 5314bef0848SChris Mason page = find_get_page(btree_inode->i_mapping, index); 5327c4452b9SChris Mason if (!page) 5337c4452b9SChris Mason continue; 5344bef0848SChris Mason 5354bef0848SChris Mason btree_lock_page_hook(page); 5364bef0848SChris Mason if (!page->mapping) { 5374bef0848SChris Mason unlock_page(page); 5384bef0848SChris Mason page_cache_release(page); 5394bef0848SChris Mason continue; 5404bef0848SChris Mason } 5414bef0848SChris Mason 5426702ed49SChris Mason if (PageWriteback(page)) { 5436702ed49SChris Mason if (PageDirty(page)) 5446702ed49SChris Mason wait_on_page_writeback(page); 5456702ed49SChris Mason else { 5466702ed49SChris Mason unlock_page(page); 5476702ed49SChris Mason page_cache_release(page); 5486702ed49SChris Mason continue; 5496702ed49SChris Mason } 5506702ed49SChris Mason } 5517c4452b9SChris Mason err = write_one_page(page, 0); 5527c4452b9SChris Mason if (err) 5537c4452b9SChris Mason werr = err; 5547c4452b9SChris Mason page_cache_release(page); 5557c4452b9SChris Mason } 5567c4452b9SChris Mason } 557690587d1SChris Mason if (err) 558690587d1SChris Mason werr = err; 559690587d1SChris Mason return werr; 560690587d1SChris Mason } 561690587d1SChris Mason 562690587d1SChris Mason /* 563690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 564690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 565690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 566690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 567690587d1SChris Mason */ 568690587d1SChris Mason int btrfs_wait_marked_extents(struct btrfs_root *root, 5698cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 570690587d1SChris Mason { 571690587d1SChris Mason int ret; 572690587d1SChris Mason int err = 0; 573690587d1SChris Mason int werr = 0; 574690587d1SChris Mason struct page *page; 575690587d1SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 576690587d1SChris Mason u64 start = 0; 577690587d1SChris Mason u64 end; 578690587d1SChris Mason unsigned long index; 579690587d1SChris Mason 580777e6bd7SChris Mason while (1) { 5818cef4e16SYan, Zheng ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5828cef4e16SYan, Zheng mark); 583777e6bd7SChris Mason if (ret) 584777e6bd7SChris Mason break; 585777e6bd7SChris Mason 5868cef4e16SYan, Zheng clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 587777e6bd7SChris Mason while (start <= end) { 588777e6bd7SChris Mason index = start >> PAGE_CACHE_SHIFT; 589777e6bd7SChris Mason start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 590777e6bd7SChris Mason page = find_get_page(btree_inode->i_mapping, index); 591777e6bd7SChris Mason if (!page) 592777e6bd7SChris Mason continue; 593777e6bd7SChris Mason if (PageDirty(page)) { 5944bef0848SChris Mason btree_lock_page_hook(page); 5954bef0848SChris Mason wait_on_page_writeback(page); 596777e6bd7SChris Mason err = write_one_page(page, 0); 597777e6bd7SChris Mason if (err) 598777e6bd7SChris Mason werr = err; 599777e6bd7SChris Mason } 600777e6bd7SChris Mason wait_on_page_writeback(page); 601777e6bd7SChris Mason page_cache_release(page); 602777e6bd7SChris Mason cond_resched(); 603777e6bd7SChris Mason } 604777e6bd7SChris Mason } 6057c4452b9SChris Mason if (err) 6067c4452b9SChris Mason werr = err; 6077c4452b9SChris Mason return werr; 60879154b1bSChris Mason } 60979154b1bSChris Mason 610690587d1SChris Mason /* 611690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 612690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 613690587d1SChris Mason * those extents are on disk for transaction or log commit 614690587d1SChris Mason */ 615690587d1SChris Mason int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 6168cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 617690587d1SChris Mason { 618690587d1SChris Mason int ret; 619690587d1SChris Mason int ret2; 620690587d1SChris Mason 6218cef4e16SYan, Zheng ret = btrfs_write_marked_extents(root, dirty_pages, mark); 6228cef4e16SYan, Zheng ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); 623690587d1SChris Mason return ret || ret2; 624690587d1SChris Mason } 625690587d1SChris Mason 626d0c803c4SChris Mason int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, 627d0c803c4SChris Mason struct btrfs_root *root) 628d0c803c4SChris Mason { 629d0c803c4SChris Mason if (!trans || !trans->transaction) { 630d0c803c4SChris Mason struct inode *btree_inode; 631d0c803c4SChris Mason btree_inode = root->fs_info->btree_inode; 632d0c803c4SChris Mason return filemap_write_and_wait(btree_inode->i_mapping); 633d0c803c4SChris Mason } 634d0c803c4SChris Mason return btrfs_write_and_wait_marked_extents(root, 6358cef4e16SYan, Zheng &trans->transaction->dirty_pages, 6368cef4e16SYan, Zheng EXTENT_DIRTY); 637d0c803c4SChris Mason } 638d0c803c4SChris Mason 639d352ac68SChris Mason /* 640d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 641d352ac68SChris Mason * 642d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 643d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 644d352ac68SChris Mason * allocation tree. 645d352ac68SChris Mason * 646d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 647d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 648d352ac68SChris Mason */ 6490b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 65079154b1bSChris Mason struct btrfs_root *root) 65179154b1bSChris Mason { 65279154b1bSChris Mason int ret; 6530b86a832SChris Mason u64 old_root_bytenr; 65486b9f2ecSYan, Zheng u64 old_root_used; 6550b86a832SChris Mason struct btrfs_root *tree_root = root->fs_info->tree_root; 65679154b1bSChris Mason 65786b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6580b86a832SChris Mason btrfs_write_dirty_block_groups(trans, root); 65956bec294SChris Mason 66079154b1bSChris Mason while (1) { 6610b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 66286b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 66386b9f2ecSYan, Zheng old_root_used == btrfs_root_used(&root->root_item)) 66479154b1bSChris Mason break; 66587ef2bb4SChris Mason 6665d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 66779154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 6680b86a832SChris Mason &root->root_key, 6690b86a832SChris Mason &root->root_item); 67079154b1bSChris Mason BUG_ON(ret); 67156bec294SChris Mason 67286b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6734a8c9a62SYan Zheng ret = btrfs_write_dirty_block_groups(trans, root); 67456bec294SChris Mason BUG_ON(ret); 6750b86a832SChris Mason } 676276e680dSYan Zheng 677276e680dSYan Zheng if (root != root->fs_info->extent_root) 678817d52f8SJosef Bacik switch_commit_root(root); 679276e680dSYan Zheng 6800b86a832SChris Mason return 0; 6810b86a832SChris Mason } 6820b86a832SChris Mason 683d352ac68SChris Mason /* 684d352ac68SChris Mason * update all the cowonly tree roots on disk 685d352ac68SChris Mason */ 6865d4f98a2SYan Zheng static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, 6870b86a832SChris Mason struct btrfs_root *root) 6880b86a832SChris Mason { 6890b86a832SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 6900b86a832SChris Mason struct list_head *next; 69184234f3aSYan Zheng struct extent_buffer *eb; 69256bec294SChris Mason int ret; 69384234f3aSYan Zheng 69456bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 69556bec294SChris Mason BUG_ON(ret); 69687ef2bb4SChris Mason 69784234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 6989fa8cfe7SChris Mason btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); 69984234f3aSYan Zheng btrfs_tree_unlock(eb); 70084234f3aSYan Zheng free_extent_buffer(eb); 7010b86a832SChris Mason 70256bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 70356bec294SChris Mason BUG_ON(ret); 70487ef2bb4SChris Mason 7050b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 7060b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 7070b86a832SChris Mason list_del_init(next); 7080b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 70987ef2bb4SChris Mason 7100b86a832SChris Mason update_cowonly_root(trans, root); 71179154b1bSChris Mason } 712276e680dSYan Zheng 713276e680dSYan Zheng down_write(&fs_info->extent_commit_sem); 714276e680dSYan Zheng switch_commit_root(fs_info->extent_root); 715276e680dSYan Zheng up_write(&fs_info->extent_commit_sem); 716276e680dSYan Zheng 71779154b1bSChris Mason return 0; 71879154b1bSChris Mason } 71979154b1bSChris Mason 720d352ac68SChris Mason /* 721d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 722d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 723d352ac68SChris Mason * be deleted 724d352ac68SChris Mason */ 7255d4f98a2SYan Zheng int btrfs_add_dead_root(struct btrfs_root *root) 7265eda7b5eSChris Mason { 727b48652c1SYan Zheng mutex_lock(&root->fs_info->trans_mutex); 7285d4f98a2SYan Zheng list_add(&root->root_list, &root->fs_info->dead_roots); 729b48652c1SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 7305eda7b5eSChris Mason return 0; 7315eda7b5eSChris Mason } 7325eda7b5eSChris Mason 733d352ac68SChris Mason /* 7345d4f98a2SYan Zheng * update all the cowonly tree roots on disk 735d352ac68SChris Mason */ 7365d4f98a2SYan Zheng static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, 7375d4f98a2SYan Zheng struct btrfs_root *root) 7380f7d52f4SChris Mason { 7390f7d52f4SChris Mason struct btrfs_root *gang[8]; 7405d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 7410f7d52f4SChris Mason int i; 7420f7d52f4SChris Mason int ret; 74354aa1f4dSChris Mason int err = 0; 74454aa1f4dSChris Mason 7450f7d52f4SChris Mason while (1) { 7465d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 7475d4f98a2SYan Zheng (void **)gang, 0, 7480f7d52f4SChris Mason ARRAY_SIZE(gang), 7490f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 7500f7d52f4SChris Mason if (ret == 0) 7510f7d52f4SChris Mason break; 7520f7d52f4SChris Mason for (i = 0; i < ret; i++) { 7530f7d52f4SChris Mason root = gang[i]; 7545d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 7552619ba1fSChris Mason (unsigned long)root->root_key.objectid, 7560f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 75731153d81SYan Zheng 758e02119d5SChris Mason btrfs_free_log(trans, root); 7595d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 760d68fc57bSYan, Zheng btrfs_orphan_commit_root(trans, root); 761e02119d5SChris Mason 762978d910dSYan Zheng if (root->commit_root != root->node) { 763817d52f8SJosef Bacik switch_commit_root(root); 764978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 765978d910dSYan Zheng root->node); 766978d910dSYan Zheng } 76731153d81SYan Zheng 7685d4f98a2SYan Zheng err = btrfs_update_root(trans, fs_info->tree_root, 7690f7d52f4SChris Mason &root->root_key, 7700f7d52f4SChris Mason &root->root_item); 77154aa1f4dSChris Mason if (err) 77254aa1f4dSChris Mason break; 7730f7d52f4SChris Mason } 7749f3a7427SChris Mason } 77554aa1f4dSChris Mason return err; 7760f7d52f4SChris Mason } 7770f7d52f4SChris Mason 778d352ac68SChris Mason /* 779d352ac68SChris Mason * defrag a given btree. If cacheonly == 1, this won't read from the disk, 780d352ac68SChris Mason * otherwise every leaf in the btree is read and defragged. 781d352ac68SChris Mason */ 782e9d0b13bSChris Mason int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) 783e9d0b13bSChris Mason { 784e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 785e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 7868929ecfaSYan, Zheng int ret; 787d3c2fdcfSChris Mason unsigned long nr; 788e9d0b13bSChris Mason 7898929ecfaSYan, Zheng if (xchg(&root->defrag_running, 1)) 790e9d0b13bSChris Mason return 0; 7918929ecfaSYan, Zheng 7926b80053dSChris Mason while (1) { 7938929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 7948929ecfaSYan, Zheng if (IS_ERR(trans)) 7958929ecfaSYan, Zheng return PTR_ERR(trans); 7968929ecfaSYan, Zheng 797e9d0b13bSChris Mason ret = btrfs_defrag_leaves(trans, root, cacheonly); 7988929ecfaSYan, Zheng 799d3c2fdcfSChris Mason nr = trans->blocks_used; 800e9d0b13bSChris Mason btrfs_end_transaction(trans, root); 801d3c2fdcfSChris Mason btrfs_btree_balance_dirty(info->tree_root, nr); 802e9d0b13bSChris Mason cond_resched(); 803e9d0b13bSChris Mason 8043f157a2fSChris Mason if (root->fs_info->closing || ret != -EAGAIN) 805e9d0b13bSChris Mason break; 806e9d0b13bSChris Mason } 807e9d0b13bSChris Mason root->defrag_running = 0; 8088929ecfaSYan, Zheng return ret; 809e9d0b13bSChris Mason } 810e9d0b13bSChris Mason 8112c47e605SYan Zheng #if 0 812d352ac68SChris Mason /* 813b7ec40d7SChris Mason * when dropping snapshots, we generate a ton of delayed refs, and it makes 814b7ec40d7SChris Mason * sense not to join the transaction while it is trying to flush the current 815b7ec40d7SChris Mason * queue of delayed refs out. 816b7ec40d7SChris Mason * 817b7ec40d7SChris Mason * This is used by the drop snapshot code only 818b7ec40d7SChris Mason */ 819b7ec40d7SChris Mason static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) 820b7ec40d7SChris Mason { 821b7ec40d7SChris Mason DEFINE_WAIT(wait); 822b7ec40d7SChris Mason 823b7ec40d7SChris Mason mutex_lock(&info->trans_mutex); 824b7ec40d7SChris Mason while (info->running_transaction && 825b7ec40d7SChris Mason info->running_transaction->delayed_refs.flushing) { 826b7ec40d7SChris Mason prepare_to_wait(&info->transaction_wait, &wait, 827b7ec40d7SChris Mason TASK_UNINTERRUPTIBLE); 828b7ec40d7SChris Mason mutex_unlock(&info->trans_mutex); 82959bc5c75SChris Mason 830b7ec40d7SChris Mason schedule(); 83159bc5c75SChris Mason 832b7ec40d7SChris Mason mutex_lock(&info->trans_mutex); 833b7ec40d7SChris Mason finish_wait(&info->transaction_wait, &wait); 834b7ec40d7SChris Mason } 835b7ec40d7SChris Mason mutex_unlock(&info->trans_mutex); 836b7ec40d7SChris Mason return 0; 837b7ec40d7SChris Mason } 838b7ec40d7SChris Mason 839b7ec40d7SChris Mason /* 840d352ac68SChris Mason * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on 841d352ac68SChris Mason * all of them 842d352ac68SChris Mason */ 8435d4f98a2SYan Zheng int btrfs_drop_dead_root(struct btrfs_root *root) 8440f7d52f4SChris Mason { 8450f7d52f4SChris Mason struct btrfs_trans_handle *trans; 8465d4f98a2SYan Zheng struct btrfs_root *tree_root = root->fs_info->tree_root; 847d3c2fdcfSChris Mason unsigned long nr; 8485d4f98a2SYan Zheng int ret; 84958176a96SJosef Bacik 8509f3a7427SChris Mason while (1) { 851b7ec40d7SChris Mason /* 852b7ec40d7SChris Mason * we don't want to jump in and create a bunch of 853b7ec40d7SChris Mason * delayed refs if the transaction is starting to close 854b7ec40d7SChris Mason */ 855b7ec40d7SChris Mason wait_transaction_pre_flush(tree_root->fs_info); 8560f7d52f4SChris Mason trans = btrfs_start_transaction(tree_root, 1); 857b7ec40d7SChris Mason 858b7ec40d7SChris Mason /* 859b7ec40d7SChris Mason * we've joined a transaction, make sure it isn't 860b7ec40d7SChris Mason * closing right now 861b7ec40d7SChris Mason */ 862b7ec40d7SChris Mason if (trans->transaction->delayed_refs.flushing) { 863b7ec40d7SChris Mason btrfs_end_transaction(trans, tree_root); 864b7ec40d7SChris Mason continue; 865b7ec40d7SChris Mason } 866b7ec40d7SChris Mason 8675d4f98a2SYan Zheng ret = btrfs_drop_snapshot(trans, root); 868d397712bSChris Mason if (ret != -EAGAIN) 8699f3a7427SChris Mason break; 87058176a96SJosef Bacik 8715d4f98a2SYan Zheng ret = btrfs_update_root(trans, tree_root, 8725d4f98a2SYan Zheng &root->root_key, 8735d4f98a2SYan Zheng &root->root_item); 8745d4f98a2SYan Zheng if (ret) 87554aa1f4dSChris Mason break; 876bcc63abbSYan 877d3c2fdcfSChris Mason nr = trans->blocks_used; 8780f7d52f4SChris Mason ret = btrfs_end_transaction(trans, tree_root); 8790f7d52f4SChris Mason BUG_ON(ret); 8805eda7b5eSChris Mason 881d3c2fdcfSChris Mason btrfs_btree_balance_dirty(tree_root, nr); 8824dc11904SChris Mason cond_resched(); 8830f7d52f4SChris Mason } 8845d4f98a2SYan Zheng BUG_ON(ret); 8855d4f98a2SYan Zheng 8865d4f98a2SYan Zheng ret = btrfs_del_root(trans, tree_root, &root->root_key); 8875d4f98a2SYan Zheng BUG_ON(ret); 8885d4f98a2SYan Zheng 8895d4f98a2SYan Zheng nr = trans->blocks_used; 8905d4f98a2SYan Zheng ret = btrfs_end_transaction(trans, tree_root); 8915d4f98a2SYan Zheng BUG_ON(ret); 8925d4f98a2SYan Zheng 8935d4f98a2SYan Zheng free_extent_buffer(root->node); 8945d4f98a2SYan Zheng free_extent_buffer(root->commit_root); 8955d4f98a2SYan Zheng kfree(root); 8965d4f98a2SYan Zheng 8975d4f98a2SYan Zheng btrfs_btree_balance_dirty(tree_root, nr); 89854aa1f4dSChris Mason return ret; 8990f7d52f4SChris Mason } 9002c47e605SYan Zheng #endif 9010f7d52f4SChris Mason 902d352ac68SChris Mason /* 903d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 904d352ac68SChris Mason * transaction commit. This does the actual creation 905d352ac68SChris Mason */ 90680b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 9073063d29fSChris Mason struct btrfs_fs_info *fs_info, 9083063d29fSChris Mason struct btrfs_pending_snapshot *pending) 9093063d29fSChris Mason { 9103063d29fSChris Mason struct btrfs_key key; 91180b6794dSChris Mason struct btrfs_root_item *new_root_item; 9123063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 9133063d29fSChris Mason struct btrfs_root *root = pending->root; 9146bdb72deSSage Weil struct btrfs_root *parent_root; 9156bdb72deSSage Weil struct inode *parent_inode; 9166a912213SJosef Bacik struct dentry *parent; 917a22285a6SYan, Zheng struct dentry *dentry; 9183063d29fSChris Mason struct extent_buffer *tmp; 919925baeddSChris Mason struct extent_buffer *old; 9203063d29fSChris Mason int ret; 921d68fc57bSYan, Zheng u64 to_reserve = 0; 9226bdb72deSSage Weil u64 index = 0; 923a22285a6SYan, Zheng u64 objectid; 924b83cc969SLi Zefan u64 root_flags; 9253063d29fSChris Mason 92680b6794dSChris Mason new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); 92780b6794dSChris Mason if (!new_root_item) { 928a22285a6SYan, Zheng pending->error = -ENOMEM; 92980b6794dSChris Mason goto fail; 93080b6794dSChris Mason } 931a22285a6SYan, Zheng 9323063d29fSChris Mason ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); 933a22285a6SYan, Zheng if (ret) { 934a22285a6SYan, Zheng pending->error = ret; 9353063d29fSChris Mason goto fail; 936a22285a6SYan, Zheng } 9373063d29fSChris Mason 9383fd0a558SYan, Zheng btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); 939d68fc57bSYan, Zheng btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); 940d68fc57bSYan, Zheng 941d68fc57bSYan, Zheng if (to_reserve > 0) { 942d68fc57bSYan, Zheng ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, 9438bb8ab2eSJosef Bacik to_reserve); 944d68fc57bSYan, Zheng if (ret) { 945d68fc57bSYan, Zheng pending->error = ret; 946d68fc57bSYan, Zheng goto fail; 947d68fc57bSYan, Zheng } 948d68fc57bSYan, Zheng } 949d68fc57bSYan, Zheng 9503063d29fSChris Mason key.objectid = objectid; 951a22285a6SYan, Zheng key.offset = (u64)-1; 952a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 9533063d29fSChris Mason 954a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 9556bdb72deSSage Weil 956a22285a6SYan, Zheng dentry = pending->dentry; 9576a912213SJosef Bacik parent = dget_parent(dentry); 9586a912213SJosef Bacik parent_inode = parent->d_inode; 959a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 9606bdb72deSSage Weil record_root_in_trans(trans, parent_root); 961a22285a6SYan, Zheng 9626bdb72deSSage Weil /* 9636bdb72deSSage Weil * insert the directory item 9646bdb72deSSage Weil */ 9656bdb72deSSage Weil ret = btrfs_set_inode_index(parent_inode, &index); 9666bdb72deSSage Weil BUG_ON(ret); 9676bdb72deSSage Weil ret = btrfs_insert_dir_item(trans, parent_root, 968a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len, 969a22285a6SYan, Zheng parent_inode->i_ino, &key, 970a22285a6SYan, Zheng BTRFS_FT_DIR, index); 9716bdb72deSSage Weil BUG_ON(ret); 9726bdb72deSSage Weil 973a22285a6SYan, Zheng btrfs_i_size_write(parent_inode, parent_inode->i_size + 974a22285a6SYan, Zheng dentry->d_name.len * 2); 9756bdb72deSSage Weil ret = btrfs_update_inode(trans, parent_root, parent_inode); 9766bdb72deSSage Weil BUG_ON(ret); 9776bdb72deSSage Weil 9786bdb72deSSage Weil record_root_in_trans(trans, root); 9796bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 9806bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 98108fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 9826bdb72deSSage Weil 983b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 984b83cc969SLi Zefan if (pending->readonly) 985b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 986b83cc969SLi Zefan else 987b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 988b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 989b83cc969SLi Zefan 990925baeddSChris Mason old = btrfs_lock_root_node(root); 9919fa8cfe7SChris Mason btrfs_cow_block(trans, root, old, NULL, 0, &old); 9925d4f98a2SYan Zheng btrfs_set_lock_blocking(old); 9933063d29fSChris Mason 994925baeddSChris Mason btrfs_copy_root(trans, root, old, &tmp, objectid); 995925baeddSChris Mason btrfs_tree_unlock(old); 996925baeddSChris Mason free_extent_buffer(old); 9973063d29fSChris Mason 9985d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 999a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1000a22285a6SYan, Zheng key.offset = trans->transid; 1001a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1002925baeddSChris Mason btrfs_tree_unlock(tmp); 10033063d29fSChris Mason free_extent_buffer(tmp); 10040660b5afSChris Mason BUG_ON(ret); 10050660b5afSChris Mason 1006a22285a6SYan, Zheng /* 1007a22285a6SYan, Zheng * insert root back/forward references 1008a22285a6SYan, Zheng */ 1009a22285a6SYan, Zheng ret = btrfs_add_root_ref(trans, tree_root, objectid, 1010a22285a6SYan, Zheng parent_root->root_key.objectid, 1011a22285a6SYan, Zheng parent_inode->i_ino, index, 1012a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 1013a22285a6SYan, Zheng BUG_ON(ret); 10146a912213SJosef Bacik dput(parent); 1015a22285a6SYan, Zheng 1016a22285a6SYan, Zheng key.offset = (u64)-1; 1017a22285a6SYan, Zheng pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); 1018a22285a6SYan, Zheng BUG_ON(IS_ERR(pending->snap)); 1019d68fc57bSYan, Zheng 10203fd0a558SYan, Zheng btrfs_reloc_post_snapshot(trans, pending); 1021d68fc57bSYan, Zheng btrfs_orphan_post_snapshot(trans, pending); 10223063d29fSChris Mason fail: 10236bdb72deSSage Weil kfree(new_root_item); 1024a22285a6SYan, Zheng btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); 1025a22285a6SYan, Zheng return 0; 10263063d29fSChris Mason } 10273063d29fSChris Mason 1028d352ac68SChris Mason /* 1029d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1030d352ac68SChris Mason */ 103180b6794dSChris Mason static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, 10323063d29fSChris Mason struct btrfs_fs_info *fs_info) 10333063d29fSChris Mason { 10343063d29fSChris Mason struct btrfs_pending_snapshot *pending; 10353063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 10363de4586cSChris Mason int ret; 10373de4586cSChris Mason 1038c6e30871SQinghuang Feng list_for_each_entry(pending, head, list) { 10393de4586cSChris Mason ret = create_pending_snapshot(trans, fs_info, pending); 10403de4586cSChris Mason BUG_ON(ret); 10413de4586cSChris Mason } 10423de4586cSChris Mason return 0; 10433de4586cSChris Mason } 10443de4586cSChris Mason 10455d4f98a2SYan Zheng static void update_super_roots(struct btrfs_root *root) 10465d4f98a2SYan Zheng { 10475d4f98a2SYan Zheng struct btrfs_root_item *root_item; 10485d4f98a2SYan Zheng struct btrfs_super_block *super; 10495d4f98a2SYan Zheng 10505d4f98a2SYan Zheng super = &root->fs_info->super_copy; 10515d4f98a2SYan Zheng 10525d4f98a2SYan Zheng root_item = &root->fs_info->chunk_root->root_item; 10535d4f98a2SYan Zheng super->chunk_root = root_item->bytenr; 10545d4f98a2SYan Zheng super->chunk_root_generation = root_item->generation; 10555d4f98a2SYan Zheng super->chunk_root_level = root_item->level; 10565d4f98a2SYan Zheng 10575d4f98a2SYan Zheng root_item = &root->fs_info->tree_root->root_item; 10585d4f98a2SYan Zheng super->root = root_item->bytenr; 10595d4f98a2SYan Zheng super->generation = root_item->generation; 10605d4f98a2SYan Zheng super->root_level = root_item->level; 10610af3d00bSJosef Bacik if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) 10620af3d00bSJosef Bacik super->cache_generation = root_item->generation; 10635d4f98a2SYan Zheng } 10645d4f98a2SYan Zheng 1065f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1066f36f3042SChris Mason { 1067f36f3042SChris Mason int ret = 0; 1068f36f3042SChris Mason spin_lock(&info->new_trans_lock); 1069f36f3042SChris Mason if (info->running_transaction) 1070f36f3042SChris Mason ret = info->running_transaction->in_commit; 1071f36f3042SChris Mason spin_unlock(&info->new_trans_lock); 1072f36f3042SChris Mason return ret; 1073f36f3042SChris Mason } 1074f36f3042SChris Mason 10758929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 10768929ecfaSYan, Zheng { 10778929ecfaSYan, Zheng int ret = 0; 10788929ecfaSYan, Zheng spin_lock(&info->new_trans_lock); 10798929ecfaSYan, Zheng if (info->running_transaction) 10808929ecfaSYan, Zheng ret = info->running_transaction->blocked; 10818929ecfaSYan, Zheng spin_unlock(&info->new_trans_lock); 10828929ecfaSYan, Zheng return ret; 10838929ecfaSYan, Zheng } 10848929ecfaSYan, Zheng 1085bb9c12c9SSage Weil /* 1086bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1087bb9c12c9SSage Weil * transaction joins 1088bb9c12c9SSage Weil */ 1089bb9c12c9SSage Weil static void wait_current_trans_commit_start(struct btrfs_root *root, 1090bb9c12c9SSage Weil struct btrfs_transaction *trans) 1091bb9c12c9SSage Weil { 1092bb9c12c9SSage Weil DEFINE_WAIT(wait); 1093bb9c12c9SSage Weil 1094bb9c12c9SSage Weil if (trans->in_commit) 1095bb9c12c9SSage Weil return; 1096bb9c12c9SSage Weil 1097bb9c12c9SSage Weil while (1) { 1098bb9c12c9SSage Weil prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, 1099bb9c12c9SSage Weil TASK_UNINTERRUPTIBLE); 1100bb9c12c9SSage Weil if (trans->in_commit) { 1101bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_blocked_wait, 1102bb9c12c9SSage Weil &wait); 1103bb9c12c9SSage Weil break; 1104bb9c12c9SSage Weil } 1105bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1106bb9c12c9SSage Weil schedule(); 1107bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1108bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_blocked_wait, &wait); 1109bb9c12c9SSage Weil } 1110bb9c12c9SSage Weil } 1111bb9c12c9SSage Weil 1112bb9c12c9SSage Weil /* 1113bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1114bb9c12c9SSage Weil * caller holds ref. 1115bb9c12c9SSage Weil */ 1116bb9c12c9SSage Weil static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, 1117bb9c12c9SSage Weil struct btrfs_transaction *trans) 1118bb9c12c9SSage Weil { 1119bb9c12c9SSage Weil DEFINE_WAIT(wait); 1120bb9c12c9SSage Weil 1121bb9c12c9SSage Weil if (trans->commit_done || (trans->in_commit && !trans->blocked)) 1122bb9c12c9SSage Weil return; 1123bb9c12c9SSage Weil 1124bb9c12c9SSage Weil while (1) { 1125bb9c12c9SSage Weil prepare_to_wait(&root->fs_info->transaction_wait, &wait, 1126bb9c12c9SSage Weil TASK_UNINTERRUPTIBLE); 1127bb9c12c9SSage Weil if (trans->commit_done || 1128bb9c12c9SSage Weil (trans->in_commit && !trans->blocked)) { 1129bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_wait, 1130bb9c12c9SSage Weil &wait); 1131bb9c12c9SSage Weil break; 1132bb9c12c9SSage Weil } 1133bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1134bb9c12c9SSage Weil schedule(); 1135bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1136bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_wait, 1137bb9c12c9SSage Weil &wait); 1138bb9c12c9SSage Weil } 1139bb9c12c9SSage Weil } 1140bb9c12c9SSage Weil 1141bb9c12c9SSage Weil /* 1142bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1143bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1144bb9c12c9SSage Weil */ 1145bb9c12c9SSage Weil struct btrfs_async_commit { 1146bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 1147bb9c12c9SSage Weil struct btrfs_root *root; 1148bb9c12c9SSage Weil struct delayed_work work; 1149bb9c12c9SSage Weil }; 1150bb9c12c9SSage Weil 1151bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1152bb9c12c9SSage Weil { 1153bb9c12c9SSage Weil struct btrfs_async_commit *ac = 1154bb9c12c9SSage Weil container_of(work, struct btrfs_async_commit, work.work); 1155bb9c12c9SSage Weil 1156bb9c12c9SSage Weil btrfs_commit_transaction(ac->newtrans, ac->root); 1157bb9c12c9SSage Weil kfree(ac); 1158bb9c12c9SSage Weil } 1159bb9c12c9SSage Weil 1160bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1161bb9c12c9SSage Weil struct btrfs_root *root, 1162bb9c12c9SSage Weil int wait_for_unblock) 1163bb9c12c9SSage Weil { 1164bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1165bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1166bb9c12c9SSage Weil 1167bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1168db5b493aSTsutomu Itoh if (!ac) 1169db5b493aSTsutomu Itoh return -ENOMEM; 1170bb9c12c9SSage Weil 1171bb9c12c9SSage Weil INIT_DELAYED_WORK(&ac->work, do_async_commit); 1172bb9c12c9SSage Weil ac->root = root; 1173bb9c12c9SSage Weil ac->newtrans = btrfs_join_transaction(root, 0); 11743612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 11753612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 11763612b495STsutomu Itoh kfree(ac); 11773612b495STsutomu Itoh return err; 11783612b495STsutomu Itoh } 1179bb9c12c9SSage Weil 1180bb9c12c9SSage Weil /* take transaction reference */ 1181bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1182bb9c12c9SSage Weil cur_trans = trans->transaction; 118313c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 1184bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1185bb9c12c9SSage Weil 1186bb9c12c9SSage Weil btrfs_end_transaction(trans, root); 1187bb9c12c9SSage Weil schedule_delayed_work(&ac->work, 0); 1188bb9c12c9SSage Weil 1189bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1190bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1191bb9c12c9SSage Weil if (wait_for_unblock) 1192bb9c12c9SSage Weil wait_current_trans_commit_start_and_unblock(root, cur_trans); 1193bb9c12c9SSage Weil else 1194bb9c12c9SSage Weil wait_current_trans_commit_start(root, cur_trans); 1195bb9c12c9SSage Weil put_transaction(cur_trans); 1196bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1197bb9c12c9SSage Weil 1198bb9c12c9SSage Weil return 0; 1199bb9c12c9SSage Weil } 1200bb9c12c9SSage Weil 1201bb9c12c9SSage Weil /* 1202bb9c12c9SSage Weil * btrfs_transaction state sequence: 1203bb9c12c9SSage Weil * in_commit = 0, blocked = 0 (initial) 1204bb9c12c9SSage Weil * in_commit = 1, blocked = 1 1205bb9c12c9SSage Weil * blocked = 0 1206bb9c12c9SSage Weil * commit_done = 1 1207bb9c12c9SSage Weil */ 120879154b1bSChris Mason int btrfs_commit_transaction(struct btrfs_trans_handle *trans, 120979154b1bSChris Mason struct btrfs_root *root) 121079154b1bSChris Mason { 121115ee9bc7SJosef Bacik unsigned long joined = 0; 121279154b1bSChris Mason struct btrfs_transaction *cur_trans; 12138fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 121479154b1bSChris Mason DEFINE_WAIT(wait); 121515ee9bc7SJosef Bacik int ret; 121689573b9cSChris Mason int should_grow = 0; 121789573b9cSChris Mason unsigned long now = get_seconds(); 1218dccae999SSage Weil int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); 121979154b1bSChris Mason 12205a3f23d5SChris Mason btrfs_run_ordered_operations(root, 0); 12215a3f23d5SChris Mason 122256bec294SChris Mason /* make a pass through all the delayed refs we have so far 122356bec294SChris Mason * any runnings procs may add more while we are here 122456bec294SChris Mason */ 122556bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 122656bec294SChris Mason BUG_ON(ret); 122756bec294SChris Mason 1228a22285a6SYan, Zheng btrfs_trans_release_metadata(trans, root); 1229a22285a6SYan, Zheng 1230b7ec40d7SChris Mason cur_trans = trans->transaction; 123156bec294SChris Mason /* 123256bec294SChris Mason * set the flushing flag so procs in this transaction have to 123356bec294SChris Mason * start sending their work down. 123456bec294SChris Mason */ 1235b7ec40d7SChris Mason cur_trans->delayed_refs.flushing = 1; 123656bec294SChris Mason 1237c3e69d58SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 123856bec294SChris Mason BUG_ON(ret); 123956bec294SChris Mason 124079154b1bSChris Mason mutex_lock(&root->fs_info->trans_mutex); 1241b7ec40d7SChris Mason if (cur_trans->in_commit) { 124213c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 1243ccd467d6SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 124479154b1bSChris Mason btrfs_end_transaction(trans, root); 1245ccd467d6SChris Mason 124679154b1bSChris Mason ret = wait_for_commit(root, cur_trans); 124779154b1bSChris Mason BUG_ON(ret); 124815ee9bc7SJosef Bacik 124915ee9bc7SJosef Bacik mutex_lock(&root->fs_info->trans_mutex); 125079154b1bSChris Mason put_transaction(cur_trans); 125115ee9bc7SJosef Bacik mutex_unlock(&root->fs_info->trans_mutex); 125215ee9bc7SJosef Bacik 125379154b1bSChris Mason return 0; 125479154b1bSChris Mason } 12554313b399SChris Mason 12562c90e5d6SChris Mason trans->transaction->in_commit = 1; 1257f9295749SChris Mason trans->transaction->blocked = 1; 1258bb9c12c9SSage Weil wake_up(&root->fs_info->transaction_blocked_wait); 1259bb9c12c9SSage Weil 1260ccd467d6SChris Mason if (cur_trans->list.prev != &root->fs_info->trans_list) { 1261ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 1262ccd467d6SChris Mason struct btrfs_transaction, list); 1263ccd467d6SChris Mason if (!prev_trans->commit_done) { 126413c5a93eSJosef Bacik atomic_inc(&prev_trans->use_count); 1265ccd467d6SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 1266ccd467d6SChris Mason 1267ccd467d6SChris Mason wait_for_commit(root, prev_trans); 1268ccd467d6SChris Mason 1269ccd467d6SChris Mason mutex_lock(&root->fs_info->trans_mutex); 127015ee9bc7SJosef Bacik put_transaction(prev_trans); 1271ccd467d6SChris Mason } 1272ccd467d6SChris Mason } 127315ee9bc7SJosef Bacik 127489573b9cSChris Mason if (now < cur_trans->start_time || now - cur_trans->start_time < 1) 127589573b9cSChris Mason should_grow = 1; 127689573b9cSChris Mason 127715ee9bc7SJosef Bacik do { 12787ea394f1SYan Zheng int snap_pending = 0; 127915ee9bc7SJosef Bacik joined = cur_trans->num_joined; 12807ea394f1SYan Zheng if (!list_empty(&trans->transaction->pending_snapshots)) 12817ea394f1SYan Zheng snap_pending = 1; 12827ea394f1SYan Zheng 12832c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 128479154b1bSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 128515ee9bc7SJosef Bacik 12860bdb1db2SSage Weil if (flush_on_commit || snap_pending) { 128724bbcf04SYan, Zheng btrfs_start_delalloc_inodes(root, 1); 128824bbcf04SYan, Zheng ret = btrfs_wait_ordered_extents(root, 0, 1); 1289ebecd3d9SSage Weil BUG_ON(ret); 12907ea394f1SYan Zheng } 12917ea394f1SYan Zheng 12925a3f23d5SChris Mason /* 12935a3f23d5SChris Mason * rename don't use btrfs_join_transaction, so, once we 12945a3f23d5SChris Mason * set the transaction to blocked above, we aren't going 12955a3f23d5SChris Mason * to get any new ordered operations. We can safely run 12965a3f23d5SChris Mason * it here and no for sure that nothing new will be added 12975a3f23d5SChris Mason * to the list 12985a3f23d5SChris Mason */ 12995a3f23d5SChris Mason btrfs_run_ordered_operations(root, 1); 13005a3f23d5SChris Mason 1301ed3b3d31SChris Mason prepare_to_wait(&cur_trans->writer_wait, &wait, 1302ed3b3d31SChris Mason TASK_UNINTERRUPTIBLE); 1303ed3b3d31SChris Mason 130489573b9cSChris Mason smp_mb(); 130513c5a93eSJosef Bacik if (atomic_read(&cur_trans->num_writers) > 1) 130699d16cbcSSage Weil schedule_timeout(MAX_SCHEDULE_TIMEOUT); 130799d16cbcSSage Weil else if (should_grow) 130899d16cbcSSage Weil schedule_timeout(1); 130915ee9bc7SJosef Bacik 131079154b1bSChris Mason mutex_lock(&root->fs_info->trans_mutex); 131115ee9bc7SJosef Bacik finish_wait(&cur_trans->writer_wait, &wait); 131213c5a93eSJosef Bacik } while (atomic_read(&cur_trans->num_writers) > 1 || 131389573b9cSChris Mason (should_grow && cur_trans->num_joined != joined)); 131415ee9bc7SJosef Bacik 13153063d29fSChris Mason ret = create_pending_snapshots(trans, root->fs_info); 13163063d29fSChris Mason BUG_ON(ret); 13173063d29fSChris Mason 131856bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 131956bec294SChris Mason BUG_ON(ret); 132056bec294SChris Mason 13212c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 1322dc17ff8fSChris Mason 1323e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 1324e02119d5SChris Mason * various roots consistent with each other. Every pointer 1325e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 1326e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 1327e02119d5SChris Mason * the tree logging code from jumping in and changing any 1328e02119d5SChris Mason * of the trees. 1329e02119d5SChris Mason * 1330e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 1331e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 1332e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 1333e02119d5SChris Mason * from now until after the super is written, we avoid races 1334e02119d5SChris Mason * with the tree-log code. 1335e02119d5SChris Mason */ 1336e02119d5SChris Mason mutex_lock(&root->fs_info->tree_log_mutex); 13371a40e23bSZheng Yan 13385d4f98a2SYan Zheng ret = commit_fs_roots(trans, root); 133954aa1f4dSChris Mason BUG_ON(ret); 134054aa1f4dSChris Mason 13415d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 1342e02119d5SChris Mason * safe to free the root of tree log roots 1343e02119d5SChris Mason */ 1344e02119d5SChris Mason btrfs_free_log_root_tree(trans, root->fs_info); 1345e02119d5SChris Mason 13465d4f98a2SYan Zheng ret = commit_cowonly_roots(trans, root); 134779154b1bSChris Mason BUG_ON(ret); 134854aa1f4dSChris Mason 134911833d66SYan Zheng btrfs_prepare_extent_commit(trans, root); 135011833d66SYan Zheng 135178fae27eSChris Mason cur_trans = root->fs_info->running_transaction; 1352cee36a03SChris Mason spin_lock(&root->fs_info->new_trans_lock); 135378fae27eSChris Mason root->fs_info->running_transaction = NULL; 1354cee36a03SChris Mason spin_unlock(&root->fs_info->new_trans_lock); 13555f39d397SChris Mason 13565d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->tree_root->root_item, 13575d4f98a2SYan Zheng root->fs_info->tree_root->node); 1358817d52f8SJosef Bacik switch_commit_root(root->fs_info->tree_root); 13595d4f98a2SYan Zheng 13605d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->chunk_root->root_item, 13615d4f98a2SYan Zheng root->fs_info->chunk_root->node); 1362817d52f8SJosef Bacik switch_commit_root(root->fs_info->chunk_root); 13635d4f98a2SYan Zheng 13645d4f98a2SYan Zheng update_super_roots(root); 1365e02119d5SChris Mason 1366e02119d5SChris Mason if (!root->fs_info->log_root_recovering) { 1367e02119d5SChris Mason btrfs_set_super_log_root(&root->fs_info->super_copy, 0); 1368e02119d5SChris Mason btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); 1369e02119d5SChris Mason } 1370e02119d5SChris Mason 1371a061fc8dSChris Mason memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, 13724b52dff6SChris Mason sizeof(root->fs_info->super_copy)); 1373ccd467d6SChris Mason 1374f9295749SChris Mason trans->transaction->blocked = 0; 1375b7ec40d7SChris Mason 1376f9295749SChris Mason wake_up(&root->fs_info->transaction_wait); 1377e6dcd2dcSChris Mason 137878fae27eSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 137979154b1bSChris Mason ret = btrfs_write_and_wait_transaction(trans, root); 138079154b1bSChris Mason BUG_ON(ret); 1381a512bbf8SYan Zheng write_ctree_super(trans, root, 0); 13824313b399SChris Mason 1383e02119d5SChris Mason /* 1384e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 1385e02119d5SChris Mason * to go about their business 1386e02119d5SChris Mason */ 1387e02119d5SChris Mason mutex_unlock(&root->fs_info->tree_log_mutex); 1388e02119d5SChris Mason 138911833d66SYan Zheng btrfs_finish_extent_commit(trans, root); 13904313b399SChris Mason 13911a40e23bSZheng Yan mutex_lock(&root->fs_info->trans_mutex); 13921a40e23bSZheng Yan 13932c90e5d6SChris Mason cur_trans->commit_done = 1; 1394b7ec40d7SChris Mason 139515ee9bc7SJosef Bacik root->fs_info->last_trans_committed = cur_trans->transid; 1396817d52f8SJosef Bacik 13972c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 13983de4586cSChris Mason 139913c5a93eSJosef Bacik list_del_init(&cur_trans->list); 140079154b1bSChris Mason put_transaction(cur_trans); 140178fae27eSChris Mason put_transaction(cur_trans); 140258176a96SJosef Bacik 14031abe9b8aSliubo trace_btrfs_transaction_commit(root); 14041abe9b8aSliubo 140578fae27eSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 14063de4586cSChris Mason 14079ed74f2dSJosef Bacik if (current->journal_info == trans) 14089ed74f2dSJosef Bacik current->journal_info = NULL; 14099ed74f2dSJosef Bacik 14102c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 141124bbcf04SYan, Zheng 141224bbcf04SYan, Zheng if (current != root->fs_info->transaction_kthread) 141324bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 141424bbcf04SYan, Zheng 141579154b1bSChris Mason return ret; 141679154b1bSChris Mason } 141779154b1bSChris Mason 1418d352ac68SChris Mason /* 1419d352ac68SChris Mason * interface function to delete all the snapshots we have scheduled for deletion 1420d352ac68SChris Mason */ 1421e9d0b13bSChris Mason int btrfs_clean_old_snapshots(struct btrfs_root *root) 1422e9d0b13bSChris Mason { 14235d4f98a2SYan Zheng LIST_HEAD(list); 14245d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 1425e9d0b13bSChris Mason 14265d4f98a2SYan Zheng mutex_lock(&fs_info->trans_mutex); 14275d4f98a2SYan Zheng list_splice_init(&fs_info->dead_roots, &list); 14285d4f98a2SYan Zheng mutex_unlock(&fs_info->trans_mutex); 14295d4f98a2SYan Zheng 14305d4f98a2SYan Zheng while (!list_empty(&list)) { 14315d4f98a2SYan Zheng root = list_entry(list.next, struct btrfs_root, root_list); 143276dda93cSYan, Zheng list_del(&root->root_list); 143376dda93cSYan, Zheng 143476dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 143576dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 14363fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 0); 143776dda93cSYan, Zheng else 14383fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 1); 1439e9d0b13bSChris Mason } 1440e9d0b13bSChris Mason return 0; 1441e9d0b13bSChris Mason } 1442