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" 30581bb050SLi Zefan #include "inode-map.h" 3179154b1bSChris Mason 320f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0 330f7d52f4SChris Mason 3480b6794dSChris Mason static noinline void put_transaction(struct btrfs_transaction *transaction) 3579154b1bSChris Mason { 3613c5a93eSJosef Bacik WARN_ON(atomic_read(&transaction->use_count) == 0); 3713c5a93eSJosef Bacik if (atomic_dec_and_test(&transaction->use_count)) { 382c90e5d6SChris Mason memset(transaction, 0, sizeof(*transaction)); 392c90e5d6SChris Mason kmem_cache_free(btrfs_transaction_cachep, transaction); 4079154b1bSChris Mason } 4178fae27eSChris Mason } 4279154b1bSChris Mason 43817d52f8SJosef Bacik static noinline void switch_commit_root(struct btrfs_root *root) 44817d52f8SJosef Bacik { 45817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 46817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 47817d52f8SJosef Bacik } 48817d52f8SJosef Bacik 49d352ac68SChris Mason /* 50d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 51d352ac68SChris Mason */ 5280b6794dSChris Mason static noinline int join_transaction(struct btrfs_root *root) 5379154b1bSChris Mason { 5479154b1bSChris Mason struct btrfs_transaction *cur_trans; 5579154b1bSChris Mason cur_trans = root->fs_info->running_transaction; 5679154b1bSChris Mason if (!cur_trans) { 572c90e5d6SChris Mason cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, 582c90e5d6SChris Mason GFP_NOFS); 59db5b493aSTsutomu Itoh if (!cur_trans) 60db5b493aSTsutomu Itoh return -ENOMEM; 610f7d52f4SChris Mason root->fs_info->generation++; 6213c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 6315ee9bc7SJosef Bacik cur_trans->num_joined = 0; 640f7d52f4SChris Mason cur_trans->transid = root->fs_info->generation; 6579154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 6679154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 6779154b1bSChris Mason cur_trans->in_commit = 0; 68f9295749SChris Mason cur_trans->blocked = 0; 6913c5a93eSJosef Bacik atomic_set(&cur_trans->use_count, 1); 7079154b1bSChris Mason cur_trans->commit_done = 0; 7108607c1bSChris Mason cur_trans->start_time = get_seconds(); 7256bec294SChris Mason 736bef4d31SEric Paris cur_trans->delayed_refs.root = RB_ROOT; 7456bec294SChris Mason cur_trans->delayed_refs.num_entries = 0; 75c3e69d58SChris Mason cur_trans->delayed_refs.num_heads_ready = 0; 76c3e69d58SChris Mason cur_trans->delayed_refs.num_heads = 0; 7756bec294SChris Mason cur_trans->delayed_refs.flushing = 0; 78c3e69d58SChris Mason cur_trans->delayed_refs.run_delayed_start = 0; 7956bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 8056bec294SChris Mason 813063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 828fd17795SChris Mason list_add_tail(&cur_trans->list, &root->fs_info->trans_list); 83d1310b2eSChris Mason extent_io_tree_init(&cur_trans->dirty_pages, 845f39d397SChris Mason root->fs_info->btree_inode->i_mapping, 855f39d397SChris Mason GFP_NOFS); 8648ec2cf8SChris Mason spin_lock(&root->fs_info->new_trans_lock); 8748ec2cf8SChris Mason root->fs_info->running_transaction = cur_trans; 8848ec2cf8SChris Mason spin_unlock(&root->fs_info->new_trans_lock); 8915ee9bc7SJosef Bacik } else { 9013c5a93eSJosef Bacik atomic_inc(&cur_trans->num_writers); 9115ee9bc7SJosef Bacik cur_trans->num_joined++; 9215ee9bc7SJosef Bacik } 9315ee9bc7SJosef Bacik 9479154b1bSChris Mason return 0; 9579154b1bSChris Mason } 9679154b1bSChris Mason 97d352ac68SChris Mason /* 98d397712bSChris Mason * this does all the record keeping required to make sure that a reference 99d397712bSChris Mason * counted root is properly recorded in a given transaction. This is required 100d397712bSChris Mason * to make sure the old root from before we joined the transaction is deleted 101d397712bSChris Mason * when the transaction commits 102d352ac68SChris Mason */ 1035d4f98a2SYan Zheng static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, 1045d4f98a2SYan Zheng struct btrfs_root *root) 1056702ed49SChris Mason { 1065d4f98a2SYan Zheng if (root->ref_cows && root->last_trans < trans->transid) { 1076702ed49SChris Mason WARN_ON(root == root->fs_info->extent_root); 1085d4f98a2SYan Zheng WARN_ON(root->commit_root != root->node); 1095d4f98a2SYan Zheng 1106702ed49SChris Mason radix_tree_tag_set(&root->fs_info->fs_roots_radix, 1116702ed49SChris Mason (unsigned long)root->root_key.objectid, 1126702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 1135d4f98a2SYan Zheng root->last_trans = trans->transid; 1145d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 1156702ed49SChris Mason } 1165d4f98a2SYan Zheng return 0; 1176702ed49SChris Mason } 1185d4f98a2SYan Zheng 1195d4f98a2SYan Zheng int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 1205d4f98a2SYan Zheng struct btrfs_root *root) 1215d4f98a2SYan Zheng { 1225d4f98a2SYan Zheng if (!root->ref_cows) 1235d4f98a2SYan Zheng return 0; 1245d4f98a2SYan Zheng 1255d4f98a2SYan Zheng mutex_lock(&root->fs_info->trans_mutex); 1265d4f98a2SYan Zheng if (root->last_trans == trans->transid) { 1275d4f98a2SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 1285d4f98a2SYan Zheng return 0; 1295d4f98a2SYan Zheng } 1305d4f98a2SYan Zheng 1315d4f98a2SYan Zheng record_root_in_trans(trans, root); 1325d4f98a2SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 1336702ed49SChris Mason return 0; 1346702ed49SChris Mason } 1356702ed49SChris Mason 136d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 137d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 138d352ac68SChris Mason * transaction might not be fully on disk. 139d352ac68SChris Mason */ 14037d1aeeeSChris Mason static void wait_current_trans(struct btrfs_root *root) 14179154b1bSChris Mason { 142f9295749SChris Mason struct btrfs_transaction *cur_trans; 14379154b1bSChris Mason 144f9295749SChris Mason cur_trans = root->fs_info->running_transaction; 14537d1aeeeSChris Mason if (cur_trans && cur_trans->blocked) { 146f9295749SChris Mason DEFINE_WAIT(wait); 14713c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 148f9295749SChris Mason while (1) { 149f9295749SChris Mason prepare_to_wait(&root->fs_info->transaction_wait, &wait, 150f9295749SChris Mason TASK_UNINTERRUPTIBLE); 151471fa17dSZhao Lei if (!cur_trans->blocked) 152471fa17dSZhao Lei break; 153f9295749SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 154f9295749SChris Mason schedule(); 155f9295749SChris Mason mutex_lock(&root->fs_info->trans_mutex); 156f9295749SChris Mason } 157471fa17dSZhao Lei finish_wait(&root->fs_info->transaction_wait, &wait); 158f9295749SChris Mason put_transaction(cur_trans); 159f9295749SChris Mason } 16037d1aeeeSChris Mason } 16137d1aeeeSChris Mason 162249ac1e5SJosef Bacik enum btrfs_trans_type { 163249ac1e5SJosef Bacik TRANS_START, 164249ac1e5SJosef Bacik TRANS_JOIN, 165249ac1e5SJosef Bacik TRANS_USERSPACE, 1660af3d00bSJosef Bacik TRANS_JOIN_NOLOCK, 167249ac1e5SJosef Bacik }; 168249ac1e5SJosef Bacik 169a22285a6SYan, Zheng static int may_wait_transaction(struct btrfs_root *root, int type) 17037d1aeeeSChris Mason { 1714bef0848SChris Mason if (!root->fs_info->log_root_recovering && 172249ac1e5SJosef Bacik ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || 173249ac1e5SJosef Bacik type == TRANS_USERSPACE)) 174a22285a6SYan, Zheng return 1; 175a22285a6SYan, Zheng return 0; 176a22285a6SYan, Zheng } 177a22285a6SYan, Zheng 178a22285a6SYan, Zheng static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 179a22285a6SYan, Zheng u64 num_items, int type) 180a22285a6SYan, Zheng { 181a22285a6SYan, Zheng struct btrfs_trans_handle *h; 182a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 18306d5a589SJosef Bacik int retries = 0; 184a22285a6SYan, Zheng int ret; 185acce952bSliubo 186acce952bSliubo if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 187acce952bSliubo return ERR_PTR(-EROFS); 188a22285a6SYan, Zheng again: 189a22285a6SYan, Zheng h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); 190a22285a6SYan, Zheng if (!h) 191a22285a6SYan, Zheng return ERR_PTR(-ENOMEM); 192a22285a6SYan, Zheng 1930af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 194a22285a6SYan, Zheng mutex_lock(&root->fs_info->trans_mutex); 195a22285a6SYan, Zheng if (may_wait_transaction(root, type)) 19637d1aeeeSChris Mason wait_current_trans(root); 197a22285a6SYan, Zheng 19879154b1bSChris Mason ret = join_transaction(root); 199db5b493aSTsutomu Itoh if (ret < 0) { 2006e8df2aeSYoshinori Sano kmem_cache_free(btrfs_trans_handle_cachep, h); 201db5b493aSTsutomu Itoh if (type != TRANS_JOIN_NOLOCK) 202db5b493aSTsutomu Itoh mutex_unlock(&root->fs_info->trans_mutex); 203db5b493aSTsutomu Itoh return ERR_PTR(ret); 204db5b493aSTsutomu Itoh } 2050f7d52f4SChris Mason 206a22285a6SYan, Zheng cur_trans = root->fs_info->running_transaction; 20713c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 2080af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 209a22285a6SYan, Zheng mutex_unlock(&root->fs_info->trans_mutex); 210a22285a6SYan, Zheng 211a22285a6SYan, Zheng h->transid = cur_trans->transid; 212a22285a6SYan, Zheng h->transaction = cur_trans; 21379154b1bSChris Mason h->blocks_used = 0; 214d2fb3437SYan Zheng h->block_group = 0; 215a22285a6SYan, Zheng h->bytes_reserved = 0; 21656bec294SChris Mason h->delayed_ref_updates = 0; 217f0486c68SYan, Zheng h->block_rsv = NULL; 218b7ec40d7SChris Mason 219a22285a6SYan, Zheng smp_mb(); 220a22285a6SYan, Zheng if (cur_trans->blocked && may_wait_transaction(root, type)) { 221a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 222a22285a6SYan, Zheng goto again; 223a22285a6SYan, Zheng } 2249ed74f2dSJosef Bacik 225a22285a6SYan, Zheng if (num_items > 0) { 2268bb8ab2eSJosef Bacik ret = btrfs_trans_reserve_metadata(h, root, num_items); 22706d5a589SJosef Bacik if (ret == -EAGAIN && !retries) { 22806d5a589SJosef Bacik retries++; 229a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 230a22285a6SYan, Zheng goto again; 23106d5a589SJosef Bacik } else if (ret == -EAGAIN) { 23206d5a589SJosef Bacik /* 23306d5a589SJosef Bacik * We have already retried and got EAGAIN, so really we 23406d5a589SJosef Bacik * don't have space, so set ret to -ENOSPC. 23506d5a589SJosef Bacik */ 23606d5a589SJosef Bacik ret = -ENOSPC; 237a22285a6SYan, Zheng } 23806d5a589SJosef Bacik 239a22285a6SYan, Zheng if (ret < 0) { 240a22285a6SYan, Zheng btrfs_end_transaction(h, root); 241a22285a6SYan, Zheng return ERR_PTR(ret); 242a22285a6SYan, Zheng } 243a22285a6SYan, Zheng } 244a22285a6SYan, Zheng 2450af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 246a22285a6SYan, Zheng mutex_lock(&root->fs_info->trans_mutex); 2475d4f98a2SYan Zheng record_root_in_trans(h, root); 2480af3d00bSJosef Bacik if (type != TRANS_JOIN_NOLOCK) 24979154b1bSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 250a22285a6SYan, Zheng 251a22285a6SYan, Zheng if (!current->journal_info && type != TRANS_USERSPACE) 252a22285a6SYan, Zheng current->journal_info = h; 25379154b1bSChris Mason return h; 25479154b1bSChris Mason } 25579154b1bSChris Mason 256f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 257a22285a6SYan, Zheng int num_items) 258f9295749SChris Mason { 259a22285a6SYan, Zheng return start_transaction(root, num_items, TRANS_START); 260f9295749SChris Mason } 261f9295749SChris Mason struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, 262f9295749SChris Mason int num_blocks) 263f9295749SChris Mason { 264a22285a6SYan, Zheng return start_transaction(root, 0, TRANS_JOIN); 265f9295749SChris Mason } 266f9295749SChris Mason 2670af3d00bSJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, 2680af3d00bSJosef Bacik int num_blocks) 2690af3d00bSJosef Bacik { 2700af3d00bSJosef Bacik return start_transaction(root, 0, TRANS_JOIN_NOLOCK); 2710af3d00bSJosef Bacik } 2720af3d00bSJosef Bacik 2739ca9ee09SSage Weil struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, 2749ca9ee09SSage Weil int num_blocks) 2759ca9ee09SSage Weil { 276a22285a6SYan, Zheng return start_transaction(r, 0, TRANS_USERSPACE); 2779ca9ee09SSage Weil } 2789ca9ee09SSage Weil 279d352ac68SChris Mason /* wait for a transaction commit to be fully complete */ 28089ce8a63SChris Mason static noinline int wait_for_commit(struct btrfs_root *root, 28189ce8a63SChris Mason struct btrfs_transaction *commit) 28289ce8a63SChris Mason { 28389ce8a63SChris Mason DEFINE_WAIT(wait); 28489ce8a63SChris Mason mutex_lock(&root->fs_info->trans_mutex); 28589ce8a63SChris Mason while (!commit->commit_done) { 28689ce8a63SChris Mason prepare_to_wait(&commit->commit_wait, &wait, 28789ce8a63SChris Mason TASK_UNINTERRUPTIBLE); 28889ce8a63SChris Mason if (commit->commit_done) 28989ce8a63SChris Mason break; 29089ce8a63SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 29189ce8a63SChris Mason schedule(); 29289ce8a63SChris Mason mutex_lock(&root->fs_info->trans_mutex); 29389ce8a63SChris Mason } 29489ce8a63SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 29589ce8a63SChris Mason finish_wait(&commit->commit_wait, &wait); 29689ce8a63SChris Mason return 0; 29789ce8a63SChris Mason } 29889ce8a63SChris Mason 29946204592SSage Weil int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) 30046204592SSage Weil { 30146204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 30246204592SSage Weil int ret; 30346204592SSage Weil 30446204592SSage Weil mutex_lock(&root->fs_info->trans_mutex); 30546204592SSage Weil 30646204592SSage Weil ret = 0; 30746204592SSage Weil if (transid) { 30846204592SSage Weil if (transid <= root->fs_info->last_trans_committed) 30946204592SSage Weil goto out_unlock; 31046204592SSage Weil 31146204592SSage Weil /* find specified transaction */ 31246204592SSage Weil list_for_each_entry(t, &root->fs_info->trans_list, list) { 31346204592SSage Weil if (t->transid == transid) { 31446204592SSage Weil cur_trans = t; 31546204592SSage Weil break; 31646204592SSage Weil } 31746204592SSage Weil if (t->transid > transid) 31846204592SSage Weil break; 31946204592SSage Weil } 32046204592SSage Weil ret = -EINVAL; 32146204592SSage Weil if (!cur_trans) 32246204592SSage Weil goto out_unlock; /* bad transid */ 32346204592SSage Weil } else { 32446204592SSage Weil /* find newest transaction that is committing | committed */ 32546204592SSage Weil list_for_each_entry_reverse(t, &root->fs_info->trans_list, 32646204592SSage Weil list) { 32746204592SSage Weil if (t->in_commit) { 32846204592SSage Weil if (t->commit_done) 32946204592SSage Weil goto out_unlock; 33046204592SSage Weil cur_trans = t; 33146204592SSage Weil break; 33246204592SSage Weil } 33346204592SSage Weil } 33446204592SSage Weil if (!cur_trans) 33546204592SSage Weil goto out_unlock; /* nothing committing|committed */ 33646204592SSage Weil } 33746204592SSage Weil 33813c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 33946204592SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 34046204592SSage Weil 34146204592SSage Weil wait_for_commit(root, cur_trans); 34246204592SSage Weil 34346204592SSage Weil mutex_lock(&root->fs_info->trans_mutex); 34446204592SSage Weil put_transaction(cur_trans); 34546204592SSage Weil ret = 0; 34646204592SSage Weil out_unlock: 34746204592SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 34846204592SSage Weil return ret; 34946204592SSage Weil } 35046204592SSage Weil 3515d4f98a2SYan Zheng #if 0 352d352ac68SChris Mason /* 353d397712bSChris Mason * rate limit against the drop_snapshot code. This helps to slow down new 354d397712bSChris Mason * operations if the drop_snapshot code isn't able to keep up. 355d352ac68SChris Mason */ 35637d1aeeeSChris Mason static void throttle_on_drops(struct btrfs_root *root) 357ab78c84dSChris Mason { 358ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 3592dd3e67bSChris Mason int harder_count = 0; 360ab78c84dSChris Mason 3612dd3e67bSChris Mason harder: 362ab78c84dSChris Mason if (atomic_read(&info->throttles)) { 363ab78c84dSChris Mason DEFINE_WAIT(wait); 364ab78c84dSChris Mason int thr; 365ab78c84dSChris Mason thr = atomic_read(&info->throttle_gen); 366ab78c84dSChris Mason 367ab78c84dSChris Mason do { 368ab78c84dSChris Mason prepare_to_wait(&info->transaction_throttle, 369ab78c84dSChris Mason &wait, TASK_UNINTERRUPTIBLE); 370ab78c84dSChris Mason if (!atomic_read(&info->throttles)) { 371ab78c84dSChris Mason finish_wait(&info->transaction_throttle, &wait); 372ab78c84dSChris Mason break; 373ab78c84dSChris Mason } 374ab78c84dSChris Mason schedule(); 375ab78c84dSChris Mason finish_wait(&info->transaction_throttle, &wait); 376ab78c84dSChris Mason } while (thr == atomic_read(&info->throttle_gen)); 3772dd3e67bSChris Mason harder_count++; 3782dd3e67bSChris Mason 3792dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && 3802dd3e67bSChris Mason harder_count < 2) 3812dd3e67bSChris Mason goto harder; 3822dd3e67bSChris Mason 3832dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && 3842dd3e67bSChris Mason harder_count < 10) 3852dd3e67bSChris Mason goto harder; 3862dd3e67bSChris Mason 3872dd3e67bSChris Mason if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && 3882dd3e67bSChris Mason harder_count < 20) 3892dd3e67bSChris Mason goto harder; 390ab78c84dSChris Mason } 391ab78c84dSChris Mason } 3925d4f98a2SYan Zheng #endif 393ab78c84dSChris Mason 39437d1aeeeSChris Mason void btrfs_throttle(struct btrfs_root *root) 39537d1aeeeSChris Mason { 39637d1aeeeSChris Mason mutex_lock(&root->fs_info->trans_mutex); 3979ca9ee09SSage Weil if (!root->fs_info->open_ioctl_trans) 39837d1aeeeSChris Mason wait_current_trans(root); 39937d1aeeeSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 40037d1aeeeSChris Mason } 40137d1aeeeSChris Mason 4028929ecfaSYan, Zheng static int should_end_transaction(struct btrfs_trans_handle *trans, 4038929ecfaSYan, Zheng struct btrfs_root *root) 4048929ecfaSYan, Zheng { 4058929ecfaSYan, Zheng int ret; 4068929ecfaSYan, Zheng ret = btrfs_block_rsv_check(trans, root, 4078929ecfaSYan, Zheng &root->fs_info->global_block_rsv, 0, 5); 4088929ecfaSYan, Zheng return ret ? 1 : 0; 4098929ecfaSYan, Zheng } 4108929ecfaSYan, Zheng 4118929ecfaSYan, Zheng int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, 4128929ecfaSYan, Zheng struct btrfs_root *root) 4138929ecfaSYan, Zheng { 4148929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 4158929ecfaSYan, Zheng int updates; 4168929ecfaSYan, Zheng 4178929ecfaSYan, Zheng if (cur_trans->blocked || cur_trans->delayed_refs.flushing) 4188929ecfaSYan, Zheng return 1; 4198929ecfaSYan, Zheng 4208929ecfaSYan, Zheng updates = trans->delayed_ref_updates; 4218929ecfaSYan, Zheng trans->delayed_ref_updates = 0; 4228929ecfaSYan, Zheng if (updates) 4238929ecfaSYan, Zheng btrfs_run_delayed_refs(trans, root, updates); 4248929ecfaSYan, Zheng 4258929ecfaSYan, Zheng return should_end_transaction(trans, root); 4268929ecfaSYan, Zheng } 4278929ecfaSYan, Zheng 42889ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 4290af3d00bSJosef Bacik struct btrfs_root *root, int throttle, int lock) 43079154b1bSChris Mason { 4318929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 432ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 433c3e69d58SChris Mason int count = 0; 434d6e4a428SChris Mason 435c3e69d58SChris Mason while (count < 4) { 436c3e69d58SChris Mason unsigned long cur = trans->delayed_ref_updates; 437c3e69d58SChris Mason trans->delayed_ref_updates = 0; 438c3e69d58SChris Mason if (cur && 439c3e69d58SChris Mason trans->transaction->delayed_refs.num_heads_ready > 64) { 440c3e69d58SChris Mason trans->delayed_ref_updates = 0; 441b7ec40d7SChris Mason 442b7ec40d7SChris Mason /* 443b7ec40d7SChris Mason * do a full flush if the transaction is trying 444b7ec40d7SChris Mason * to close 445b7ec40d7SChris Mason */ 446b7ec40d7SChris Mason if (trans->transaction->delayed_refs.flushing) 447b7ec40d7SChris Mason cur = 0; 448c3e69d58SChris Mason btrfs_run_delayed_refs(trans, root, cur); 449c3e69d58SChris Mason } else { 450c3e69d58SChris Mason break; 451c3e69d58SChris Mason } 452c3e69d58SChris Mason count++; 45356bec294SChris Mason } 45456bec294SChris Mason 455a22285a6SYan, Zheng btrfs_trans_release_metadata(trans, root); 456a22285a6SYan, Zheng 4570af3d00bSJosef Bacik if (lock && !root->fs_info->open_ioctl_trans && 4588929ecfaSYan, Zheng should_end_transaction(trans, root)) 4598929ecfaSYan, Zheng trans->transaction->blocked = 1; 4608929ecfaSYan, Zheng 4610af3d00bSJosef Bacik if (lock && cur_trans->blocked && !cur_trans->in_commit) { 4628929ecfaSYan, Zheng if (throttle) 4638929ecfaSYan, Zheng return btrfs_commit_transaction(trans, root); 4648929ecfaSYan, Zheng else 4658929ecfaSYan, Zheng wake_up_process(info->transaction_kthread); 4668929ecfaSYan, Zheng } 4678929ecfaSYan, Zheng 4688929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 46913c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 47013c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 47189ce8a63SChris Mason 47299d16cbcSSage Weil smp_mb(); 47379154b1bSChris Mason if (waitqueue_active(&cur_trans->writer_wait)) 47479154b1bSChris Mason wake_up(&cur_trans->writer_wait); 47579154b1bSChris Mason put_transaction(cur_trans); 4769ed74f2dSJosef Bacik 4779ed74f2dSJosef Bacik if (current->journal_info == trans) 4789ed74f2dSJosef Bacik current->journal_info = NULL; 479d6025579SChris Mason memset(trans, 0, sizeof(*trans)); 4802c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 481ab78c84dSChris Mason 48224bbcf04SYan, Zheng if (throttle) 48324bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 48424bbcf04SYan, Zheng 48579154b1bSChris Mason return 0; 48679154b1bSChris Mason } 48779154b1bSChris Mason 48889ce8a63SChris Mason int btrfs_end_transaction(struct btrfs_trans_handle *trans, 48989ce8a63SChris Mason struct btrfs_root *root) 49089ce8a63SChris Mason { 4910af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 0, 1); 49289ce8a63SChris Mason } 49389ce8a63SChris Mason 49489ce8a63SChris Mason int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, 49589ce8a63SChris Mason struct btrfs_root *root) 49689ce8a63SChris Mason { 4970af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 1, 1); 4980af3d00bSJosef Bacik } 4990af3d00bSJosef Bacik 5000af3d00bSJosef Bacik int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, 5010af3d00bSJosef Bacik struct btrfs_root *root) 5020af3d00bSJosef Bacik { 5030af3d00bSJosef Bacik return __btrfs_end_transaction(trans, root, 0, 0); 50489ce8a63SChris Mason } 50589ce8a63SChris Mason 506d352ac68SChris Mason /* 507d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 508d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 509690587d1SChris Mason * those extents are sent to disk but does not wait on them 510d352ac68SChris Mason */ 511690587d1SChris Mason int btrfs_write_marked_extents(struct btrfs_root *root, 5128cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 51379154b1bSChris Mason { 5147c4452b9SChris Mason int ret; 515777e6bd7SChris Mason int err = 0; 5167c4452b9SChris Mason int werr = 0; 5177c4452b9SChris Mason struct page *page; 5187c4452b9SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 519777e6bd7SChris Mason u64 start = 0; 5205f39d397SChris Mason u64 end; 5215f39d397SChris Mason unsigned long index; 5227c4452b9SChris Mason 5237c4452b9SChris Mason while (1) { 524777e6bd7SChris Mason ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5258cef4e16SYan, Zheng mark); 5265f39d397SChris Mason if (ret) 5277c4452b9SChris Mason break; 5285f39d397SChris Mason while (start <= end) { 529777e6bd7SChris Mason cond_resched(); 530777e6bd7SChris Mason 5315f39d397SChris Mason index = start >> PAGE_CACHE_SHIFT; 53235ebb934SChris Mason start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 5334bef0848SChris Mason page = find_get_page(btree_inode->i_mapping, index); 5347c4452b9SChris Mason if (!page) 5357c4452b9SChris Mason continue; 5364bef0848SChris Mason 5374bef0848SChris Mason btree_lock_page_hook(page); 5384bef0848SChris Mason if (!page->mapping) { 5394bef0848SChris Mason unlock_page(page); 5404bef0848SChris Mason page_cache_release(page); 5414bef0848SChris Mason continue; 5424bef0848SChris Mason } 5434bef0848SChris Mason 5446702ed49SChris Mason if (PageWriteback(page)) { 5456702ed49SChris Mason if (PageDirty(page)) 5466702ed49SChris Mason wait_on_page_writeback(page); 5476702ed49SChris Mason else { 5486702ed49SChris Mason unlock_page(page); 5496702ed49SChris Mason page_cache_release(page); 5506702ed49SChris Mason continue; 5516702ed49SChris Mason } 5526702ed49SChris Mason } 5537c4452b9SChris Mason err = write_one_page(page, 0); 5547c4452b9SChris Mason if (err) 5557c4452b9SChris Mason werr = err; 5567c4452b9SChris Mason page_cache_release(page); 5577c4452b9SChris Mason } 5587c4452b9SChris Mason } 559690587d1SChris Mason if (err) 560690587d1SChris Mason werr = err; 561690587d1SChris Mason return werr; 562690587d1SChris Mason } 563690587d1SChris Mason 564690587d1SChris Mason /* 565690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 566690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 567690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 568690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 569690587d1SChris Mason */ 570690587d1SChris Mason int btrfs_wait_marked_extents(struct btrfs_root *root, 5718cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 572690587d1SChris Mason { 573690587d1SChris Mason int ret; 574690587d1SChris Mason int err = 0; 575690587d1SChris Mason int werr = 0; 576690587d1SChris Mason struct page *page; 577690587d1SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 578690587d1SChris Mason u64 start = 0; 579690587d1SChris Mason u64 end; 580690587d1SChris Mason unsigned long index; 581690587d1SChris Mason 582777e6bd7SChris Mason while (1) { 5838cef4e16SYan, Zheng ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5848cef4e16SYan, Zheng mark); 585777e6bd7SChris Mason if (ret) 586777e6bd7SChris Mason break; 587777e6bd7SChris Mason 5888cef4e16SYan, Zheng clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 589777e6bd7SChris Mason while (start <= end) { 590777e6bd7SChris Mason index = start >> PAGE_CACHE_SHIFT; 591777e6bd7SChris Mason start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 592777e6bd7SChris Mason page = find_get_page(btree_inode->i_mapping, index); 593777e6bd7SChris Mason if (!page) 594777e6bd7SChris Mason continue; 595777e6bd7SChris Mason if (PageDirty(page)) { 5964bef0848SChris Mason btree_lock_page_hook(page); 5974bef0848SChris Mason wait_on_page_writeback(page); 598777e6bd7SChris Mason err = write_one_page(page, 0); 599777e6bd7SChris Mason if (err) 600777e6bd7SChris Mason werr = err; 601777e6bd7SChris Mason } 602777e6bd7SChris Mason wait_on_page_writeback(page); 603777e6bd7SChris Mason page_cache_release(page); 604777e6bd7SChris Mason cond_resched(); 605777e6bd7SChris Mason } 606777e6bd7SChris Mason } 6077c4452b9SChris Mason if (err) 6087c4452b9SChris Mason werr = err; 6097c4452b9SChris Mason return werr; 61079154b1bSChris Mason } 61179154b1bSChris Mason 612690587d1SChris Mason /* 613690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 614690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 615690587d1SChris Mason * those extents are on disk for transaction or log commit 616690587d1SChris Mason */ 617690587d1SChris Mason int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 6188cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 619690587d1SChris Mason { 620690587d1SChris Mason int ret; 621690587d1SChris Mason int ret2; 622690587d1SChris Mason 6238cef4e16SYan, Zheng ret = btrfs_write_marked_extents(root, dirty_pages, mark); 6248cef4e16SYan, Zheng ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); 625690587d1SChris Mason return ret || ret2; 626690587d1SChris Mason } 627690587d1SChris Mason 628d0c803c4SChris Mason int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, 629d0c803c4SChris Mason struct btrfs_root *root) 630d0c803c4SChris Mason { 631d0c803c4SChris Mason if (!trans || !trans->transaction) { 632d0c803c4SChris Mason struct inode *btree_inode; 633d0c803c4SChris Mason btree_inode = root->fs_info->btree_inode; 634d0c803c4SChris Mason return filemap_write_and_wait(btree_inode->i_mapping); 635d0c803c4SChris Mason } 636d0c803c4SChris Mason return btrfs_write_and_wait_marked_extents(root, 6378cef4e16SYan, Zheng &trans->transaction->dirty_pages, 6388cef4e16SYan, Zheng EXTENT_DIRTY); 639d0c803c4SChris Mason } 640d0c803c4SChris Mason 641d352ac68SChris Mason /* 642d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 643d352ac68SChris Mason * 644d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 645d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 646d352ac68SChris Mason * allocation tree. 647d352ac68SChris Mason * 648d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 649d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 650d352ac68SChris Mason */ 6510b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 65279154b1bSChris Mason struct btrfs_root *root) 65379154b1bSChris Mason { 65479154b1bSChris Mason int ret; 6550b86a832SChris Mason u64 old_root_bytenr; 65686b9f2ecSYan, Zheng u64 old_root_used; 6570b86a832SChris Mason struct btrfs_root *tree_root = root->fs_info->tree_root; 65879154b1bSChris Mason 65986b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6600b86a832SChris Mason btrfs_write_dirty_block_groups(trans, root); 66156bec294SChris Mason 66279154b1bSChris Mason while (1) { 6630b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 66486b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 66586b9f2ecSYan, Zheng old_root_used == btrfs_root_used(&root->root_item)) 66679154b1bSChris Mason break; 66787ef2bb4SChris Mason 6685d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 66979154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 6700b86a832SChris Mason &root->root_key, 6710b86a832SChris Mason &root->root_item); 67279154b1bSChris Mason BUG_ON(ret); 67356bec294SChris Mason 67486b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6754a8c9a62SYan Zheng ret = btrfs_write_dirty_block_groups(trans, root); 67656bec294SChris Mason BUG_ON(ret); 6770b86a832SChris Mason } 678276e680dSYan Zheng 679276e680dSYan Zheng if (root != root->fs_info->extent_root) 680817d52f8SJosef Bacik switch_commit_root(root); 681276e680dSYan Zheng 6820b86a832SChris Mason return 0; 6830b86a832SChris Mason } 6840b86a832SChris Mason 685d352ac68SChris Mason /* 686d352ac68SChris Mason * update all the cowonly tree roots on disk 687d352ac68SChris Mason */ 6885d4f98a2SYan Zheng static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, 6890b86a832SChris Mason struct btrfs_root *root) 6900b86a832SChris Mason { 6910b86a832SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 6920b86a832SChris Mason struct list_head *next; 69384234f3aSYan Zheng struct extent_buffer *eb; 69456bec294SChris Mason int ret; 69584234f3aSYan Zheng 69656bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 69756bec294SChris Mason BUG_ON(ret); 69887ef2bb4SChris Mason 69984234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 7009fa8cfe7SChris Mason btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); 70184234f3aSYan Zheng btrfs_tree_unlock(eb); 70284234f3aSYan Zheng free_extent_buffer(eb); 7030b86a832SChris Mason 70456bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 70556bec294SChris Mason BUG_ON(ret); 70687ef2bb4SChris Mason 7070b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 7080b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 7090b86a832SChris Mason list_del_init(next); 7100b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 71187ef2bb4SChris Mason 7120b86a832SChris Mason update_cowonly_root(trans, root); 71379154b1bSChris Mason } 714276e680dSYan Zheng 715276e680dSYan Zheng down_write(&fs_info->extent_commit_sem); 716276e680dSYan Zheng switch_commit_root(fs_info->extent_root); 717276e680dSYan Zheng up_write(&fs_info->extent_commit_sem); 718276e680dSYan Zheng 71979154b1bSChris Mason return 0; 72079154b1bSChris Mason } 72179154b1bSChris Mason 722d352ac68SChris Mason /* 723d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 724d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 725d352ac68SChris Mason * be deleted 726d352ac68SChris Mason */ 7275d4f98a2SYan Zheng int btrfs_add_dead_root(struct btrfs_root *root) 7285eda7b5eSChris Mason { 729b48652c1SYan Zheng mutex_lock(&root->fs_info->trans_mutex); 7305d4f98a2SYan Zheng list_add(&root->root_list, &root->fs_info->dead_roots); 731b48652c1SYan Zheng mutex_unlock(&root->fs_info->trans_mutex); 7325eda7b5eSChris Mason return 0; 7335eda7b5eSChris Mason } 7345eda7b5eSChris Mason 735d352ac68SChris Mason /* 7365d4f98a2SYan Zheng * update all the cowonly tree roots on disk 737d352ac68SChris Mason */ 7385d4f98a2SYan Zheng static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, 7395d4f98a2SYan Zheng struct btrfs_root *root) 7400f7d52f4SChris Mason { 7410f7d52f4SChris Mason struct btrfs_root *gang[8]; 7425d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 7430f7d52f4SChris Mason int i; 7440f7d52f4SChris Mason int ret; 74554aa1f4dSChris Mason int err = 0; 74654aa1f4dSChris Mason 7470f7d52f4SChris Mason while (1) { 7485d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 7495d4f98a2SYan Zheng (void **)gang, 0, 7500f7d52f4SChris Mason ARRAY_SIZE(gang), 7510f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 7520f7d52f4SChris Mason if (ret == 0) 7530f7d52f4SChris Mason break; 7540f7d52f4SChris Mason for (i = 0; i < ret; i++) { 7550f7d52f4SChris Mason root = gang[i]; 7565d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 7572619ba1fSChris Mason (unsigned long)root->root_key.objectid, 7580f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 75931153d81SYan Zheng 760e02119d5SChris Mason btrfs_free_log(trans, root); 7615d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 762d68fc57bSYan, Zheng btrfs_orphan_commit_root(trans, root); 763e02119d5SChris Mason 76482d5902dSLi Zefan btrfs_save_ino_cache(root, trans); 76582d5902dSLi Zefan 766978d910dSYan Zheng if (root->commit_root != root->node) { 767581bb050SLi Zefan mutex_lock(&root->fs_commit_mutex); 768817d52f8SJosef Bacik switch_commit_root(root); 769581bb050SLi Zefan btrfs_unpin_free_ino(root); 770581bb050SLi Zefan mutex_unlock(&root->fs_commit_mutex); 771581bb050SLi Zefan 772978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 773978d910dSYan Zheng root->node); 774978d910dSYan Zheng } 77531153d81SYan Zheng 7765d4f98a2SYan Zheng err = btrfs_update_root(trans, fs_info->tree_root, 7770f7d52f4SChris Mason &root->root_key, 7780f7d52f4SChris Mason &root->root_item); 77954aa1f4dSChris Mason if (err) 78054aa1f4dSChris Mason break; 7810f7d52f4SChris Mason } 7829f3a7427SChris Mason } 78354aa1f4dSChris Mason return err; 7840f7d52f4SChris Mason } 7850f7d52f4SChris Mason 786d352ac68SChris Mason /* 787d352ac68SChris Mason * defrag a given btree. If cacheonly == 1, this won't read from the disk, 788d352ac68SChris Mason * otherwise every leaf in the btree is read and defragged. 789d352ac68SChris Mason */ 790e9d0b13bSChris Mason int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) 791e9d0b13bSChris Mason { 792e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 793e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 7948929ecfaSYan, Zheng int ret; 795d3c2fdcfSChris Mason unsigned long nr; 796e9d0b13bSChris Mason 7978929ecfaSYan, Zheng if (xchg(&root->defrag_running, 1)) 798e9d0b13bSChris Mason return 0; 7998929ecfaSYan, Zheng 8006b80053dSChris Mason while (1) { 8018929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 8028929ecfaSYan, Zheng if (IS_ERR(trans)) 8038929ecfaSYan, Zheng return PTR_ERR(trans); 8048929ecfaSYan, Zheng 805e9d0b13bSChris Mason ret = btrfs_defrag_leaves(trans, root, cacheonly); 8068929ecfaSYan, Zheng 807d3c2fdcfSChris Mason nr = trans->blocks_used; 808e9d0b13bSChris Mason btrfs_end_transaction(trans, root); 809d3c2fdcfSChris Mason btrfs_btree_balance_dirty(info->tree_root, nr); 810e9d0b13bSChris Mason cond_resched(); 811e9d0b13bSChris Mason 8123f157a2fSChris Mason if (root->fs_info->closing || ret != -EAGAIN) 813e9d0b13bSChris Mason break; 814e9d0b13bSChris Mason } 815e9d0b13bSChris Mason root->defrag_running = 0; 8168929ecfaSYan, Zheng return ret; 817e9d0b13bSChris Mason } 818e9d0b13bSChris Mason 8192c47e605SYan Zheng #if 0 820d352ac68SChris Mason /* 821b7ec40d7SChris Mason * when dropping snapshots, we generate a ton of delayed refs, and it makes 822b7ec40d7SChris Mason * sense not to join the transaction while it is trying to flush the current 823b7ec40d7SChris Mason * queue of delayed refs out. 824b7ec40d7SChris Mason * 825b7ec40d7SChris Mason * This is used by the drop snapshot code only 826b7ec40d7SChris Mason */ 827b7ec40d7SChris Mason static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) 828b7ec40d7SChris Mason { 829b7ec40d7SChris Mason DEFINE_WAIT(wait); 830b7ec40d7SChris Mason 831b7ec40d7SChris Mason mutex_lock(&info->trans_mutex); 832b7ec40d7SChris Mason while (info->running_transaction && 833b7ec40d7SChris Mason info->running_transaction->delayed_refs.flushing) { 834b7ec40d7SChris Mason prepare_to_wait(&info->transaction_wait, &wait, 835b7ec40d7SChris Mason TASK_UNINTERRUPTIBLE); 836b7ec40d7SChris Mason mutex_unlock(&info->trans_mutex); 83759bc5c75SChris Mason 838b7ec40d7SChris Mason schedule(); 83959bc5c75SChris Mason 840b7ec40d7SChris Mason mutex_lock(&info->trans_mutex); 841b7ec40d7SChris Mason finish_wait(&info->transaction_wait, &wait); 842b7ec40d7SChris Mason } 843b7ec40d7SChris Mason mutex_unlock(&info->trans_mutex); 844b7ec40d7SChris Mason return 0; 845b7ec40d7SChris Mason } 846b7ec40d7SChris Mason 847b7ec40d7SChris Mason /* 848d352ac68SChris Mason * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on 849d352ac68SChris Mason * all of them 850d352ac68SChris Mason */ 8515d4f98a2SYan Zheng int btrfs_drop_dead_root(struct btrfs_root *root) 8520f7d52f4SChris Mason { 8530f7d52f4SChris Mason struct btrfs_trans_handle *trans; 8545d4f98a2SYan Zheng struct btrfs_root *tree_root = root->fs_info->tree_root; 855d3c2fdcfSChris Mason unsigned long nr; 8565d4f98a2SYan Zheng int ret; 85758176a96SJosef Bacik 8589f3a7427SChris Mason while (1) { 859b7ec40d7SChris Mason /* 860b7ec40d7SChris Mason * we don't want to jump in and create a bunch of 861b7ec40d7SChris Mason * delayed refs if the transaction is starting to close 862b7ec40d7SChris Mason */ 863b7ec40d7SChris Mason wait_transaction_pre_flush(tree_root->fs_info); 8640f7d52f4SChris Mason trans = btrfs_start_transaction(tree_root, 1); 865b7ec40d7SChris Mason 866b7ec40d7SChris Mason /* 867b7ec40d7SChris Mason * we've joined a transaction, make sure it isn't 868b7ec40d7SChris Mason * closing right now 869b7ec40d7SChris Mason */ 870b7ec40d7SChris Mason if (trans->transaction->delayed_refs.flushing) { 871b7ec40d7SChris Mason btrfs_end_transaction(trans, tree_root); 872b7ec40d7SChris Mason continue; 873b7ec40d7SChris Mason } 874b7ec40d7SChris Mason 8755d4f98a2SYan Zheng ret = btrfs_drop_snapshot(trans, root); 876d397712bSChris Mason if (ret != -EAGAIN) 8779f3a7427SChris Mason break; 87858176a96SJosef Bacik 8795d4f98a2SYan Zheng ret = btrfs_update_root(trans, tree_root, 8805d4f98a2SYan Zheng &root->root_key, 8815d4f98a2SYan Zheng &root->root_item); 8825d4f98a2SYan Zheng if (ret) 88354aa1f4dSChris Mason break; 884bcc63abbSYan 885d3c2fdcfSChris Mason nr = trans->blocks_used; 8860f7d52f4SChris Mason ret = btrfs_end_transaction(trans, tree_root); 8870f7d52f4SChris Mason BUG_ON(ret); 8885eda7b5eSChris Mason 889d3c2fdcfSChris Mason btrfs_btree_balance_dirty(tree_root, nr); 8904dc11904SChris Mason cond_resched(); 8910f7d52f4SChris Mason } 8925d4f98a2SYan Zheng BUG_ON(ret); 8935d4f98a2SYan Zheng 8945d4f98a2SYan Zheng ret = btrfs_del_root(trans, tree_root, &root->root_key); 8955d4f98a2SYan Zheng BUG_ON(ret); 8965d4f98a2SYan Zheng 8975d4f98a2SYan Zheng nr = trans->blocks_used; 8985d4f98a2SYan Zheng ret = btrfs_end_transaction(trans, tree_root); 8995d4f98a2SYan Zheng BUG_ON(ret); 9005d4f98a2SYan Zheng 9015d4f98a2SYan Zheng free_extent_buffer(root->node); 9025d4f98a2SYan Zheng free_extent_buffer(root->commit_root); 9035d4f98a2SYan Zheng kfree(root); 9045d4f98a2SYan Zheng 9055d4f98a2SYan Zheng btrfs_btree_balance_dirty(tree_root, nr); 90654aa1f4dSChris Mason return ret; 9070f7d52f4SChris Mason } 9082c47e605SYan Zheng #endif 9090f7d52f4SChris Mason 910d352ac68SChris Mason /* 911d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 912d352ac68SChris Mason * transaction commit. This does the actual creation 913d352ac68SChris Mason */ 91480b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 9153063d29fSChris Mason struct btrfs_fs_info *fs_info, 9163063d29fSChris Mason struct btrfs_pending_snapshot *pending) 9173063d29fSChris Mason { 9183063d29fSChris Mason struct btrfs_key key; 91980b6794dSChris Mason struct btrfs_root_item *new_root_item; 9203063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 9213063d29fSChris Mason struct btrfs_root *root = pending->root; 9226bdb72deSSage Weil struct btrfs_root *parent_root; 9236bdb72deSSage Weil struct inode *parent_inode; 9246a912213SJosef Bacik struct dentry *parent; 925a22285a6SYan, Zheng struct dentry *dentry; 9263063d29fSChris Mason struct extent_buffer *tmp; 927925baeddSChris Mason struct extent_buffer *old; 9283063d29fSChris Mason int ret; 929d68fc57bSYan, Zheng u64 to_reserve = 0; 9306bdb72deSSage Weil u64 index = 0; 931a22285a6SYan, Zheng u64 objectid; 932b83cc969SLi Zefan u64 root_flags; 9333063d29fSChris Mason 93480b6794dSChris Mason new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); 93580b6794dSChris Mason if (!new_root_item) { 936a22285a6SYan, Zheng pending->error = -ENOMEM; 93780b6794dSChris Mason goto fail; 93880b6794dSChris Mason } 939a22285a6SYan, Zheng 940581bb050SLi Zefan ret = btrfs_find_free_objectid(tree_root, &objectid); 941a22285a6SYan, Zheng if (ret) { 942a22285a6SYan, Zheng pending->error = ret; 9433063d29fSChris Mason goto fail; 944a22285a6SYan, Zheng } 9453063d29fSChris Mason 9463fd0a558SYan, Zheng btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); 947d68fc57bSYan, Zheng btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); 948d68fc57bSYan, Zheng 949d68fc57bSYan, Zheng if (to_reserve > 0) { 950d68fc57bSYan, Zheng ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, 9518bb8ab2eSJosef Bacik to_reserve); 952d68fc57bSYan, Zheng if (ret) { 953d68fc57bSYan, Zheng pending->error = ret; 954d68fc57bSYan, Zheng goto fail; 955d68fc57bSYan, Zheng } 956d68fc57bSYan, Zheng } 957d68fc57bSYan, Zheng 9583063d29fSChris Mason key.objectid = objectid; 959a22285a6SYan, Zheng key.offset = (u64)-1; 960a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 9613063d29fSChris Mason 962a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 9636bdb72deSSage Weil 964a22285a6SYan, Zheng dentry = pending->dentry; 9656a912213SJosef Bacik parent = dget_parent(dentry); 9666a912213SJosef Bacik parent_inode = parent->d_inode; 967a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 9686bdb72deSSage Weil record_root_in_trans(trans, parent_root); 969a22285a6SYan, Zheng 9706bdb72deSSage Weil /* 9716bdb72deSSage Weil * insert the directory item 9726bdb72deSSage Weil */ 9736bdb72deSSage Weil ret = btrfs_set_inode_index(parent_inode, &index); 9746bdb72deSSage Weil BUG_ON(ret); 9756bdb72deSSage Weil ret = btrfs_insert_dir_item(trans, parent_root, 976a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len, 97733345d01SLi Zefan btrfs_ino(parent_inode), &key, 978a22285a6SYan, Zheng BTRFS_FT_DIR, index); 9796bdb72deSSage Weil BUG_ON(ret); 9806bdb72deSSage Weil 981a22285a6SYan, Zheng btrfs_i_size_write(parent_inode, parent_inode->i_size + 982a22285a6SYan, Zheng dentry->d_name.len * 2); 9836bdb72deSSage Weil ret = btrfs_update_inode(trans, parent_root, parent_inode); 9846bdb72deSSage Weil BUG_ON(ret); 9856bdb72deSSage Weil 9866bdb72deSSage Weil record_root_in_trans(trans, root); 9876bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 9886bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 98908fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 9906bdb72deSSage Weil 991b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 992b83cc969SLi Zefan if (pending->readonly) 993b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 994b83cc969SLi Zefan else 995b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 996b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 997b83cc969SLi Zefan 998925baeddSChris Mason old = btrfs_lock_root_node(root); 9999fa8cfe7SChris Mason btrfs_cow_block(trans, root, old, NULL, 0, &old); 10005d4f98a2SYan Zheng btrfs_set_lock_blocking(old); 10013063d29fSChris Mason 1002925baeddSChris Mason btrfs_copy_root(trans, root, old, &tmp, objectid); 1003925baeddSChris Mason btrfs_tree_unlock(old); 1004925baeddSChris Mason free_extent_buffer(old); 10053063d29fSChris Mason 10065d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1007a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1008a22285a6SYan, Zheng key.offset = trans->transid; 1009a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1010925baeddSChris Mason btrfs_tree_unlock(tmp); 10113063d29fSChris Mason free_extent_buffer(tmp); 10120660b5afSChris Mason BUG_ON(ret); 10130660b5afSChris Mason 1014a22285a6SYan, Zheng /* 1015a22285a6SYan, Zheng * insert root back/forward references 1016a22285a6SYan, Zheng */ 1017a22285a6SYan, Zheng ret = btrfs_add_root_ref(trans, tree_root, objectid, 1018a22285a6SYan, Zheng parent_root->root_key.objectid, 101933345d01SLi Zefan btrfs_ino(parent_inode), index, 1020a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 1021a22285a6SYan, Zheng BUG_ON(ret); 10226a912213SJosef Bacik dput(parent); 1023a22285a6SYan, Zheng 1024a22285a6SYan, Zheng key.offset = (u64)-1; 1025a22285a6SYan, Zheng pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); 1026a22285a6SYan, Zheng BUG_ON(IS_ERR(pending->snap)); 1027d68fc57bSYan, Zheng 10283fd0a558SYan, Zheng btrfs_reloc_post_snapshot(trans, pending); 1029d68fc57bSYan, Zheng btrfs_orphan_post_snapshot(trans, pending); 10303063d29fSChris Mason fail: 10316bdb72deSSage Weil kfree(new_root_item); 1032a22285a6SYan, Zheng btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); 1033a22285a6SYan, Zheng return 0; 10343063d29fSChris Mason } 10353063d29fSChris Mason 1036d352ac68SChris Mason /* 1037d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1038d352ac68SChris Mason */ 103980b6794dSChris Mason static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, 10403063d29fSChris Mason struct btrfs_fs_info *fs_info) 10413063d29fSChris Mason { 10423063d29fSChris Mason struct btrfs_pending_snapshot *pending; 10433063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 10443de4586cSChris Mason int ret; 10453de4586cSChris Mason 1046c6e30871SQinghuang Feng list_for_each_entry(pending, head, list) { 10473de4586cSChris Mason ret = create_pending_snapshot(trans, fs_info, pending); 10483de4586cSChris Mason BUG_ON(ret); 10493de4586cSChris Mason } 10503de4586cSChris Mason return 0; 10513de4586cSChris Mason } 10523de4586cSChris Mason 10535d4f98a2SYan Zheng static void update_super_roots(struct btrfs_root *root) 10545d4f98a2SYan Zheng { 10555d4f98a2SYan Zheng struct btrfs_root_item *root_item; 10565d4f98a2SYan Zheng struct btrfs_super_block *super; 10575d4f98a2SYan Zheng 10585d4f98a2SYan Zheng super = &root->fs_info->super_copy; 10595d4f98a2SYan Zheng 10605d4f98a2SYan Zheng root_item = &root->fs_info->chunk_root->root_item; 10615d4f98a2SYan Zheng super->chunk_root = root_item->bytenr; 10625d4f98a2SYan Zheng super->chunk_root_generation = root_item->generation; 10635d4f98a2SYan Zheng super->chunk_root_level = root_item->level; 10645d4f98a2SYan Zheng 10655d4f98a2SYan Zheng root_item = &root->fs_info->tree_root->root_item; 10665d4f98a2SYan Zheng super->root = root_item->bytenr; 10675d4f98a2SYan Zheng super->generation = root_item->generation; 10685d4f98a2SYan Zheng super->root_level = root_item->level; 10690af3d00bSJosef Bacik if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) 10700af3d00bSJosef Bacik super->cache_generation = root_item->generation; 10715d4f98a2SYan Zheng } 10725d4f98a2SYan Zheng 1073f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1074f36f3042SChris Mason { 1075f36f3042SChris Mason int ret = 0; 1076f36f3042SChris Mason spin_lock(&info->new_trans_lock); 1077f36f3042SChris Mason if (info->running_transaction) 1078f36f3042SChris Mason ret = info->running_transaction->in_commit; 1079f36f3042SChris Mason spin_unlock(&info->new_trans_lock); 1080f36f3042SChris Mason return ret; 1081f36f3042SChris Mason } 1082f36f3042SChris Mason 10838929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 10848929ecfaSYan, Zheng { 10858929ecfaSYan, Zheng int ret = 0; 10868929ecfaSYan, Zheng spin_lock(&info->new_trans_lock); 10878929ecfaSYan, Zheng if (info->running_transaction) 10888929ecfaSYan, Zheng ret = info->running_transaction->blocked; 10898929ecfaSYan, Zheng spin_unlock(&info->new_trans_lock); 10908929ecfaSYan, Zheng return ret; 10918929ecfaSYan, Zheng } 10928929ecfaSYan, Zheng 1093bb9c12c9SSage Weil /* 1094bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1095bb9c12c9SSage Weil * transaction joins 1096bb9c12c9SSage Weil */ 1097bb9c12c9SSage Weil static void wait_current_trans_commit_start(struct btrfs_root *root, 1098bb9c12c9SSage Weil struct btrfs_transaction *trans) 1099bb9c12c9SSage Weil { 1100bb9c12c9SSage Weil DEFINE_WAIT(wait); 1101bb9c12c9SSage Weil 1102bb9c12c9SSage Weil if (trans->in_commit) 1103bb9c12c9SSage Weil return; 1104bb9c12c9SSage Weil 1105bb9c12c9SSage Weil while (1) { 1106bb9c12c9SSage Weil prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, 1107bb9c12c9SSage Weil TASK_UNINTERRUPTIBLE); 1108bb9c12c9SSage Weil if (trans->in_commit) { 1109bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_blocked_wait, 1110bb9c12c9SSage Weil &wait); 1111bb9c12c9SSage Weil break; 1112bb9c12c9SSage Weil } 1113bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1114bb9c12c9SSage Weil schedule(); 1115bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1116bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_blocked_wait, &wait); 1117bb9c12c9SSage Weil } 1118bb9c12c9SSage Weil } 1119bb9c12c9SSage Weil 1120bb9c12c9SSage Weil /* 1121bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1122bb9c12c9SSage Weil * caller holds ref. 1123bb9c12c9SSage Weil */ 1124bb9c12c9SSage Weil static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, 1125bb9c12c9SSage Weil struct btrfs_transaction *trans) 1126bb9c12c9SSage Weil { 1127bb9c12c9SSage Weil DEFINE_WAIT(wait); 1128bb9c12c9SSage Weil 1129bb9c12c9SSage Weil if (trans->commit_done || (trans->in_commit && !trans->blocked)) 1130bb9c12c9SSage Weil return; 1131bb9c12c9SSage Weil 1132bb9c12c9SSage Weil while (1) { 1133bb9c12c9SSage Weil prepare_to_wait(&root->fs_info->transaction_wait, &wait, 1134bb9c12c9SSage Weil TASK_UNINTERRUPTIBLE); 1135bb9c12c9SSage Weil if (trans->commit_done || 1136bb9c12c9SSage Weil (trans->in_commit && !trans->blocked)) { 1137bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_wait, 1138bb9c12c9SSage Weil &wait); 1139bb9c12c9SSage Weil break; 1140bb9c12c9SSage Weil } 1141bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1142bb9c12c9SSage Weil schedule(); 1143bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1144bb9c12c9SSage Weil finish_wait(&root->fs_info->transaction_wait, 1145bb9c12c9SSage Weil &wait); 1146bb9c12c9SSage Weil } 1147bb9c12c9SSage Weil } 1148bb9c12c9SSage Weil 1149bb9c12c9SSage Weil /* 1150bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1151bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1152bb9c12c9SSage Weil */ 1153bb9c12c9SSage Weil struct btrfs_async_commit { 1154bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 1155bb9c12c9SSage Weil struct btrfs_root *root; 1156bb9c12c9SSage Weil struct delayed_work work; 1157bb9c12c9SSage Weil }; 1158bb9c12c9SSage Weil 1159bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1160bb9c12c9SSage Weil { 1161bb9c12c9SSage Weil struct btrfs_async_commit *ac = 1162bb9c12c9SSage Weil container_of(work, struct btrfs_async_commit, work.work); 1163bb9c12c9SSage Weil 1164bb9c12c9SSage Weil btrfs_commit_transaction(ac->newtrans, ac->root); 1165bb9c12c9SSage Weil kfree(ac); 1166bb9c12c9SSage Weil } 1167bb9c12c9SSage Weil 1168bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1169bb9c12c9SSage Weil struct btrfs_root *root, 1170bb9c12c9SSage Weil int wait_for_unblock) 1171bb9c12c9SSage Weil { 1172bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1173bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1174bb9c12c9SSage Weil 1175bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1176db5b493aSTsutomu Itoh if (!ac) 1177db5b493aSTsutomu Itoh return -ENOMEM; 1178bb9c12c9SSage Weil 1179bb9c12c9SSage Weil INIT_DELAYED_WORK(&ac->work, do_async_commit); 1180bb9c12c9SSage Weil ac->root = root; 1181bb9c12c9SSage Weil ac->newtrans = btrfs_join_transaction(root, 0); 11823612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 11833612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 11843612b495STsutomu Itoh kfree(ac); 11853612b495STsutomu Itoh return err; 11863612b495STsutomu Itoh } 1187bb9c12c9SSage Weil 1188bb9c12c9SSage Weil /* take transaction reference */ 1189bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1190bb9c12c9SSage Weil cur_trans = trans->transaction; 119113c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 1192bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1193bb9c12c9SSage Weil 1194bb9c12c9SSage Weil btrfs_end_transaction(trans, root); 1195bb9c12c9SSage Weil schedule_delayed_work(&ac->work, 0); 1196bb9c12c9SSage Weil 1197bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1198bb9c12c9SSage Weil mutex_lock(&root->fs_info->trans_mutex); 1199bb9c12c9SSage Weil if (wait_for_unblock) 1200bb9c12c9SSage Weil wait_current_trans_commit_start_and_unblock(root, cur_trans); 1201bb9c12c9SSage Weil else 1202bb9c12c9SSage Weil wait_current_trans_commit_start(root, cur_trans); 1203bb9c12c9SSage Weil put_transaction(cur_trans); 1204bb9c12c9SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 1205bb9c12c9SSage Weil 1206bb9c12c9SSage Weil return 0; 1207bb9c12c9SSage Weil } 1208bb9c12c9SSage Weil 1209bb9c12c9SSage Weil /* 1210bb9c12c9SSage Weil * btrfs_transaction state sequence: 1211bb9c12c9SSage Weil * in_commit = 0, blocked = 0 (initial) 1212bb9c12c9SSage Weil * in_commit = 1, blocked = 1 1213bb9c12c9SSage Weil * blocked = 0 1214bb9c12c9SSage Weil * commit_done = 1 1215bb9c12c9SSage Weil */ 121679154b1bSChris Mason int btrfs_commit_transaction(struct btrfs_trans_handle *trans, 121779154b1bSChris Mason struct btrfs_root *root) 121879154b1bSChris Mason { 121915ee9bc7SJosef Bacik unsigned long joined = 0; 122079154b1bSChris Mason struct btrfs_transaction *cur_trans; 12218fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 122279154b1bSChris Mason DEFINE_WAIT(wait); 122315ee9bc7SJosef Bacik int ret; 122489573b9cSChris Mason int should_grow = 0; 122589573b9cSChris Mason unsigned long now = get_seconds(); 1226dccae999SSage Weil int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); 122779154b1bSChris Mason 12285a3f23d5SChris Mason btrfs_run_ordered_operations(root, 0); 12295a3f23d5SChris Mason 123056bec294SChris Mason /* make a pass through all the delayed refs we have so far 123156bec294SChris Mason * any runnings procs may add more while we are here 123256bec294SChris Mason */ 123356bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 123456bec294SChris Mason BUG_ON(ret); 123556bec294SChris Mason 1236a22285a6SYan, Zheng btrfs_trans_release_metadata(trans, root); 1237a22285a6SYan, Zheng 1238b7ec40d7SChris Mason cur_trans = trans->transaction; 123956bec294SChris Mason /* 124056bec294SChris Mason * set the flushing flag so procs in this transaction have to 124156bec294SChris Mason * start sending their work down. 124256bec294SChris Mason */ 1243b7ec40d7SChris Mason cur_trans->delayed_refs.flushing = 1; 124456bec294SChris Mason 1245c3e69d58SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 124656bec294SChris Mason BUG_ON(ret); 124756bec294SChris Mason 124879154b1bSChris Mason mutex_lock(&root->fs_info->trans_mutex); 1249b7ec40d7SChris Mason if (cur_trans->in_commit) { 125013c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 1251ccd467d6SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 125279154b1bSChris Mason btrfs_end_transaction(trans, root); 1253ccd467d6SChris Mason 125479154b1bSChris Mason ret = wait_for_commit(root, cur_trans); 125579154b1bSChris Mason BUG_ON(ret); 125615ee9bc7SJosef Bacik 125715ee9bc7SJosef Bacik mutex_lock(&root->fs_info->trans_mutex); 125879154b1bSChris Mason put_transaction(cur_trans); 125915ee9bc7SJosef Bacik mutex_unlock(&root->fs_info->trans_mutex); 126015ee9bc7SJosef Bacik 126179154b1bSChris Mason return 0; 126279154b1bSChris Mason } 12634313b399SChris Mason 12642c90e5d6SChris Mason trans->transaction->in_commit = 1; 1265f9295749SChris Mason trans->transaction->blocked = 1; 1266bb9c12c9SSage Weil wake_up(&root->fs_info->transaction_blocked_wait); 1267bb9c12c9SSage Weil 1268ccd467d6SChris Mason if (cur_trans->list.prev != &root->fs_info->trans_list) { 1269ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 1270ccd467d6SChris Mason struct btrfs_transaction, list); 1271ccd467d6SChris Mason if (!prev_trans->commit_done) { 127213c5a93eSJosef Bacik atomic_inc(&prev_trans->use_count); 1273ccd467d6SChris Mason mutex_unlock(&root->fs_info->trans_mutex); 1274ccd467d6SChris Mason 1275ccd467d6SChris Mason wait_for_commit(root, prev_trans); 1276ccd467d6SChris Mason 1277ccd467d6SChris Mason mutex_lock(&root->fs_info->trans_mutex); 127815ee9bc7SJosef Bacik put_transaction(prev_trans); 1279ccd467d6SChris Mason } 1280ccd467d6SChris Mason } 128115ee9bc7SJosef Bacik 128289573b9cSChris Mason if (now < cur_trans->start_time || now - cur_trans->start_time < 1) 128389573b9cSChris Mason should_grow = 1; 128489573b9cSChris Mason 128515ee9bc7SJosef Bacik do { 12867ea394f1SYan Zheng int snap_pending = 0; 128715ee9bc7SJosef Bacik joined = cur_trans->num_joined; 12887ea394f1SYan Zheng if (!list_empty(&trans->transaction->pending_snapshots)) 12897ea394f1SYan Zheng snap_pending = 1; 12907ea394f1SYan Zheng 12912c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 129279154b1bSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 129315ee9bc7SJosef Bacik 12940bdb1db2SSage Weil if (flush_on_commit || snap_pending) { 129524bbcf04SYan, Zheng btrfs_start_delalloc_inodes(root, 1); 129624bbcf04SYan, Zheng ret = btrfs_wait_ordered_extents(root, 0, 1); 1297ebecd3d9SSage Weil BUG_ON(ret); 12987ea394f1SYan Zheng } 12997ea394f1SYan Zheng 13005a3f23d5SChris Mason /* 13015a3f23d5SChris Mason * rename don't use btrfs_join_transaction, so, once we 13025a3f23d5SChris Mason * set the transaction to blocked above, we aren't going 13035a3f23d5SChris Mason * to get any new ordered operations. We can safely run 13045a3f23d5SChris Mason * it here and no for sure that nothing new will be added 13055a3f23d5SChris Mason * to the list 13065a3f23d5SChris Mason */ 13075a3f23d5SChris Mason btrfs_run_ordered_operations(root, 1); 13085a3f23d5SChris Mason 1309ed3b3d31SChris Mason prepare_to_wait(&cur_trans->writer_wait, &wait, 1310ed3b3d31SChris Mason TASK_UNINTERRUPTIBLE); 1311ed3b3d31SChris Mason 131289573b9cSChris Mason smp_mb(); 131313c5a93eSJosef Bacik if (atomic_read(&cur_trans->num_writers) > 1) 131499d16cbcSSage Weil schedule_timeout(MAX_SCHEDULE_TIMEOUT); 131599d16cbcSSage Weil else if (should_grow) 131699d16cbcSSage Weil schedule_timeout(1); 131715ee9bc7SJosef Bacik 131879154b1bSChris Mason mutex_lock(&root->fs_info->trans_mutex); 131915ee9bc7SJosef Bacik finish_wait(&cur_trans->writer_wait, &wait); 132013c5a93eSJosef Bacik } while (atomic_read(&cur_trans->num_writers) > 1 || 132189573b9cSChris Mason (should_grow && cur_trans->num_joined != joined)); 132215ee9bc7SJosef Bacik 13233063d29fSChris Mason ret = create_pending_snapshots(trans, root->fs_info); 13243063d29fSChris Mason BUG_ON(ret); 13253063d29fSChris Mason 132656bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 132756bec294SChris Mason BUG_ON(ret); 132856bec294SChris Mason 13292c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 1330dc17ff8fSChris Mason 1331e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 1332e02119d5SChris Mason * various roots consistent with each other. Every pointer 1333e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 1334e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 1335e02119d5SChris Mason * the tree logging code from jumping in and changing any 1336e02119d5SChris Mason * of the trees. 1337e02119d5SChris Mason * 1338e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 1339e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 1340e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 1341e02119d5SChris Mason * from now until after the super is written, we avoid races 1342e02119d5SChris Mason * with the tree-log code. 1343e02119d5SChris Mason */ 1344e02119d5SChris Mason mutex_lock(&root->fs_info->tree_log_mutex); 13451a40e23bSZheng Yan 13465d4f98a2SYan Zheng ret = commit_fs_roots(trans, root); 134754aa1f4dSChris Mason BUG_ON(ret); 134854aa1f4dSChris Mason 13495d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 1350e02119d5SChris Mason * safe to free the root of tree log roots 1351e02119d5SChris Mason */ 1352e02119d5SChris Mason btrfs_free_log_root_tree(trans, root->fs_info); 1353e02119d5SChris Mason 13545d4f98a2SYan Zheng ret = commit_cowonly_roots(trans, root); 135579154b1bSChris Mason BUG_ON(ret); 135654aa1f4dSChris Mason 135711833d66SYan Zheng btrfs_prepare_extent_commit(trans, root); 135811833d66SYan Zheng 135978fae27eSChris Mason cur_trans = root->fs_info->running_transaction; 1360cee36a03SChris Mason spin_lock(&root->fs_info->new_trans_lock); 136178fae27eSChris Mason root->fs_info->running_transaction = NULL; 1362cee36a03SChris Mason spin_unlock(&root->fs_info->new_trans_lock); 13635f39d397SChris Mason 13645d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->tree_root->root_item, 13655d4f98a2SYan Zheng root->fs_info->tree_root->node); 1366817d52f8SJosef Bacik switch_commit_root(root->fs_info->tree_root); 13675d4f98a2SYan Zheng 13685d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->chunk_root->root_item, 13695d4f98a2SYan Zheng root->fs_info->chunk_root->node); 1370817d52f8SJosef Bacik switch_commit_root(root->fs_info->chunk_root); 13715d4f98a2SYan Zheng 13725d4f98a2SYan Zheng update_super_roots(root); 1373e02119d5SChris Mason 1374e02119d5SChris Mason if (!root->fs_info->log_root_recovering) { 1375e02119d5SChris Mason btrfs_set_super_log_root(&root->fs_info->super_copy, 0); 1376e02119d5SChris Mason btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); 1377e02119d5SChris Mason } 1378e02119d5SChris Mason 1379a061fc8dSChris Mason memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, 13804b52dff6SChris Mason sizeof(root->fs_info->super_copy)); 1381ccd467d6SChris Mason 1382f9295749SChris Mason trans->transaction->blocked = 0; 1383b7ec40d7SChris Mason 1384f9295749SChris Mason wake_up(&root->fs_info->transaction_wait); 1385e6dcd2dcSChris Mason 138678fae27eSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 138779154b1bSChris Mason ret = btrfs_write_and_wait_transaction(trans, root); 138879154b1bSChris Mason BUG_ON(ret); 1389a512bbf8SYan Zheng write_ctree_super(trans, root, 0); 13904313b399SChris Mason 1391e02119d5SChris Mason /* 1392e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 1393e02119d5SChris Mason * to go about their business 1394e02119d5SChris Mason */ 1395e02119d5SChris Mason mutex_unlock(&root->fs_info->tree_log_mutex); 1396e02119d5SChris Mason 139711833d66SYan Zheng btrfs_finish_extent_commit(trans, root); 13984313b399SChris Mason 13991a40e23bSZheng Yan mutex_lock(&root->fs_info->trans_mutex); 14001a40e23bSZheng Yan 14012c90e5d6SChris Mason cur_trans->commit_done = 1; 1402b7ec40d7SChris Mason 140315ee9bc7SJosef Bacik root->fs_info->last_trans_committed = cur_trans->transid; 1404817d52f8SJosef Bacik 14052c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 14063de4586cSChris Mason 140713c5a93eSJosef Bacik list_del_init(&cur_trans->list); 140879154b1bSChris Mason put_transaction(cur_trans); 140978fae27eSChris Mason put_transaction(cur_trans); 141058176a96SJosef Bacik 14111abe9b8aSliubo trace_btrfs_transaction_commit(root); 14121abe9b8aSliubo 141378fae27eSChris Mason mutex_unlock(&root->fs_info->trans_mutex); 14143de4586cSChris Mason 14159ed74f2dSJosef Bacik if (current->journal_info == trans) 14169ed74f2dSJosef Bacik current->journal_info = NULL; 14179ed74f2dSJosef Bacik 14182c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 141924bbcf04SYan, Zheng 142024bbcf04SYan, Zheng if (current != root->fs_info->transaction_kthread) 142124bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 142224bbcf04SYan, Zheng 142379154b1bSChris Mason return ret; 142479154b1bSChris Mason } 142579154b1bSChris Mason 1426d352ac68SChris Mason /* 1427d352ac68SChris Mason * interface function to delete all the snapshots we have scheduled for deletion 1428d352ac68SChris Mason */ 1429e9d0b13bSChris Mason int btrfs_clean_old_snapshots(struct btrfs_root *root) 1430e9d0b13bSChris Mason { 14315d4f98a2SYan Zheng LIST_HEAD(list); 14325d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 1433e9d0b13bSChris Mason 14345d4f98a2SYan Zheng mutex_lock(&fs_info->trans_mutex); 14355d4f98a2SYan Zheng list_splice_init(&fs_info->dead_roots, &list); 14365d4f98a2SYan Zheng mutex_unlock(&fs_info->trans_mutex); 14375d4f98a2SYan Zheng 14385d4f98a2SYan Zheng while (!list_empty(&list)) { 14395d4f98a2SYan Zheng root = list_entry(list.next, struct btrfs_root, root_list); 144076dda93cSYan, Zheng list_del(&root->root_list); 144176dda93cSYan, Zheng 144276dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 144376dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 14443fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 0); 144576dda93cSYan, Zheng else 14463fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 1); 1447e9d0b13bSChris Mason } 1448e9d0b13bSChris Mason return 0; 1449e9d0b13bSChris Mason } 1450