fs/btrfs/transaction.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/sched.h>
  21 #include <linux/writeback.h>
  22 #include <linux/pagemap.h>
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26
  27 static int total_trans = 0;
  28 extern struct kmem_cache *btrfs_trans_handle_cachep;
  29 extern struct kmem_cache *btrfs_transaction_cachep;
  30
  31 static struct workqueue_struct *trans_wq;
  32
  33 #define BTRFS_ROOT_TRANS_TAG 0
  34 #define BTRFS_ROOT_DEFRAG_TAG 1
  35
  36 static noinline void put_transaction(struct btrfs_transaction *transaction)
  37 {
  38         WARN_ON(transaction->use_count == 0);
  39         transaction->use_count--;
  40         if (transaction->use_count == 0) {
  41                 WARN_ON(total_trans == 0);
  42                 total_trans--;
  43                 list_del_init(&transaction->list);
  44                 memset(transaction, 0, sizeof(*transaction));
  45                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  46         }
  47 }
  48
  49 static noinline int join_transaction(struct btrfs_root *root)
  50 {
  51         struct btrfs_transaction *cur_trans;
  52         cur_trans = root->fs_info->running_transaction;
  53         if (!cur_trans) {
  54                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  55                                              GFP_NOFS);
  56                 total_trans++;
  57                 BUG_ON(!cur_trans);
  58                 root->fs_info->generation++;
  59                 root->fs_info->running_transaction = cur_trans;
  60                 root->fs_info->last_alloc = 0;
  61                 root->fs_info->last_data_alloc = 0;
  62                 cur_trans->num_writers = 1;
  63                 cur_trans->num_joined = 0;
  64                 cur_trans->transid = root->fs_info->generation;
  65                 init_waitqueue_head(&cur_trans->writer_wait);
  66                 init_waitqueue_head(&cur_trans->commit_wait);
  67                 cur_trans->in_commit = 0;
  68                 cur_trans->use_count = 1;
  69                 cur_trans->commit_done = 0;
  70                 cur_trans->start_time = get_seconds();
  71                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  72                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  73                 btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
  74                 extent_io_tree_init(&cur_trans->dirty_pages,
  75                                      root->fs_info->btree_inode->i_mapping,
  76                                      GFP_NOFS);
  77         } else {
  78                 cur_trans->num_writers++;
  79                 cur_trans->num_joined++;
  80         }
  81
  82         return 0;
  83 }
  84
  85 static noinline int record_root_in_trans(struct btrfs_root *root)
  86 {
  87         u64 running_trans_id = root->fs_info->running_transaction->transid;
  88         if (root->ref_cows && root->last_trans < running_trans_id) {
  89                 WARN_ON(root == root->fs_info->extent_root);
  90                 if (root->root_item.refs != 0) {
  91                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  92                                    (unsigned long)root->root_key.objectid,
  93                                    BTRFS_ROOT_TRANS_TAG);
  94                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  95                                    (unsigned long)root->root_key.objectid,
  96                                    BTRFS_ROOT_DEFRAG_TAG);
  97                         root->commit_root = root->node;
  98                         extent_buffer_get(root->node);
  99                 } else {
 100                         WARN_ON(1);
 101                 }
 102                 root->last_trans = running_trans_id;
 103         }
 104         return 0;
 105 }
 106
 107 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 108                                                    int num_blocks)
 109 {
 110         struct btrfs_trans_handle *h =
 111                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 112         int ret;
 113
 114         mutex_lock(&root->fs_info->trans_mutex);
 115         ret = join_transaction(root);
 116         BUG_ON(ret);
 117
 118         record_root_in_trans(root);
 119         h->transid = root->fs_info->running_transaction->transid;
 120         h->transaction = root->fs_info->running_transaction;
 121         h->blocks_reserved = num_blocks;
 122         h->blocks_used = 0;
 123         h->block_group = NULL;
 124         h->alloc_exclude_nr = 0;
 125         h->alloc_exclude_start = 0;
 126         root->fs_info->running_transaction->use_count++;
 127         mutex_unlock(&root->fs_info->trans_mutex);
 128         return h;
 129 }
 130
 131 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 132                           struct btrfs_root *root)
 133 {
 134         struct btrfs_transaction *cur_trans;
 135
 136         mutex_lock(&root->fs_info->trans_mutex);
 137         cur_trans = root->fs_info->running_transaction;
 138         WARN_ON(cur_trans != trans->transaction);
 139         WARN_ON(cur_trans->num_writers < 1);
 140         cur_trans->num_writers--;
 141         if (waitqueue_active(&cur_trans->writer_wait))
 142                 wake_up(&cur_trans->writer_wait);
 143         put_transaction(cur_trans);
 144         mutex_unlock(&root->fs_info->trans_mutex);
 145         memset(trans, 0, sizeof(*trans));
 146         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 147         return 0;
 148 }
 149
 150
 151 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 152                                      struct btrfs_root *root)
 153 {
 154         int ret;
 155         int err;
 156         int werr = 0;
 157         struct extent_io_tree *dirty_pages;
 158         struct page *page;
 159         struct inode *btree_inode = root->fs_info->btree_inode;
 160         u64 start;
 161         u64 end;
 162         unsigned long index;
 163
 164         if (!trans || !trans->transaction) {
 165                 return filemap_write_and_wait(btree_inode->i_mapping);
 166         }
 167         dirty_pages = &trans->transaction->dirty_pages;
 168         while(1) {
 169                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 170                                             EXTENT_DIRTY);
 171                 if (ret)
 172                         break;
 173                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 174                 while(start <= end) {
 175                         index = start >> PAGE_CACHE_SHIFT;
 176                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 177                         page = find_lock_page(btree_inode->i_mapping, index);
 178                         if (!page)
 179                                 continue;
 180                         if (PageWriteback(page)) {
 181                                 if (PageDirty(page))
 182                                         wait_on_page_writeback(page);
 183                                 else {
 184                                         unlock_page(page);
 185                                         page_cache_release(page);
 186                                         continue;
 187                                 }
 188                         }
 189                         err = write_one_page(page, 0);
 190                         if (err)
 191                                 werr = err;
 192                         page_cache_release(page);
 193                 }
 194         }
 195         err = filemap_fdatawait(btree_inode->i_mapping);
 196         if (err)
 197                 werr = err;
 198         return werr;
 199 }
 200
 201 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 202                             struct btrfs_root *root)
 203 {
 204         int ret;
 205         u64 old_extent_block;
 206         struct btrfs_fs_info *fs_info = root->fs_info;
 207         struct btrfs_root *tree_root = fs_info->tree_root;
 208         struct btrfs_root *extent_root = fs_info->extent_root;
 209
 210         btrfs_write_dirty_block_groups(trans, extent_root);
 211         while(1) {
 212                 old_extent_block = btrfs_root_bytenr(&extent_root->root_item);
 213                 if (old_extent_block == extent_root->node->start)
 214                         break;
 215                 btrfs_set_root_bytenr(&extent_root->root_item,
 216                                       extent_root->node->start);
 217                 btrfs_set_root_level(&extent_root->root_item,
 218                                      btrfs_header_level(extent_root->node));
 219                 ret = btrfs_update_root(trans, tree_root,
 220                                         &extent_root->root_key,
 221                                         &extent_root->root_item);
 222                 BUG_ON(ret);
 223                 btrfs_write_dirty_block_groups(trans, extent_root);
 224         }
 225         return 0;
 226 }
 227
 228 static noinline int wait_for_commit(struct btrfs_root *root,
 229                                     struct btrfs_transaction *commit)
 230 {
 231         DEFINE_WAIT(wait);
 232         mutex_lock(&root->fs_info->trans_mutex);
 233         while(!commit->commit_done) {
 234                 prepare_to_wait(&commit->commit_wait, &wait,
 235                                 TASK_UNINTERRUPTIBLE);
 236                 if (commit->commit_done)
 237                         break;
 238                 mutex_unlock(&root->fs_info->trans_mutex);
 239                 schedule();
 240                 mutex_lock(&root->fs_info->trans_mutex);
 241         }
 242         mutex_unlock(&root->fs_info->trans_mutex);
 243         finish_wait(&commit->commit_wait, &wait);
 244         return 0;
 245 }
 246
 247 struct dirty_root {
 248         struct list_head list;
 249         struct btrfs_root *root;
 250         struct btrfs_root *latest_root;
 251 };
 252
 253 int btrfs_add_dead_root(struct btrfs_root *root,
 254                         struct btrfs_root *latest,
 255                         struct list_head *dead_list)
 256 {
 257         struct dirty_root *dirty;
 258
 259         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 260         if (!dirty)
 261                 return -ENOMEM;
 262         dirty->root = root;
 263         dirty->latest_root = latest;
 264         list_add(&dirty->list, dead_list);
 265         return 0;
 266 }
 267
 268 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 269                                     struct radix_tree_root *radix,
 270                                     struct list_head *list)
 271 {
 272         struct dirty_root *dirty;
 273         struct btrfs_root *gang[8];
 274         struct btrfs_root *root;
 275         int i;
 276         int ret;
 277         int err = 0;
 278         u32 refs;
 279
 280         while(1) {
 281                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 282                                                  ARRAY_SIZE(gang),
 283                                                  BTRFS_ROOT_TRANS_TAG);
 284                 if (ret == 0)
 285                         break;
 286                 for (i = 0; i < ret; i++) {
 287                         root = gang[i];
 288                         radix_tree_tag_clear(radix,
 289                                      (unsigned long)root->root_key.objectid,
 290                                      BTRFS_ROOT_TRANS_TAG);
 291                         if (root->commit_root == root->node) {
 292                                 WARN_ON(root->node->start !=
 293                                         btrfs_root_bytenr(&root->root_item));
 294                                 free_extent_buffer(root->commit_root);
 295                                 root->commit_root = NULL;
 296
 297                                 /* make sure to update the root on disk
 298                                  * so we get any updates to the block used
 299                                  * counts
 300                                  */
 301                                 err = btrfs_update_root(trans,
 302                                                 root->fs_info->tree_root,
 303                                                 &root->root_key,
 304                                                 &root->root_item);
 305                                 continue;
 306                         }
 307                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 308                         BUG_ON(!dirty);
 309                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 310                         BUG_ON(!dirty->root);
 311
 312                         memset(&root->root_item.drop_progress, 0,
 313                                sizeof(struct btrfs_disk_key));
 314                         root->root_item.drop_level = 0;
 315
 316                         memcpy(dirty->root, root, sizeof(*root));
 317                         dirty->root->node = root->commit_root;
 318                         dirty->latest_root = root;
 319                         root->commit_root = NULL;
 320
 321                         root->root_key.offset = root->fs_info->generation;
 322                         btrfs_set_root_bytenr(&root->root_item,
 323                                               root->node->start);
 324                         btrfs_set_root_level(&root->root_item,
 325                                              btrfs_header_level(root->node));
 326                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 327                                                 &root->root_key,
 328                                                 &root->root_item);
 329                         if (err)
 330                                 break;
 331
 332                         refs = btrfs_root_refs(&dirty->root->root_item);
 333                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 334                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 335                                                 &dirty->root->root_key,
 336                                                 &dirty->root->root_item);
 337
 338                         BUG_ON(err);
 339                         if (refs == 1) {
 340                                 list_add(&dirty->list, list);
 341                         } else {
 342                                 WARN_ON(1);
 343                                 kfree(dirty->root);
 344                                 kfree(dirty);
 345                         }
 346                 }
 347         }
 348         return err;
 349 }
 350
 351 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 352 {
 353         struct btrfs_fs_info *info = root->fs_info;
 354         int ret;
 355         struct btrfs_trans_handle *trans;
 356         unsigned long nr;
 357
 358         if (root->defrag_running)
 359                 return 0;
 360         trans = btrfs_start_transaction(root, 1);
 361         while (1) {
 362                 root->defrag_running = 1;
 363                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 364                 nr = trans->blocks_used;
 365                 btrfs_end_transaction(trans, root);
 366                 mutex_unlock(&info->fs_mutex);
 367                 btrfs_btree_balance_dirty(info->tree_root, nr);
 368                 cond_resched();
 369
 370                 mutex_lock(&info->fs_mutex);
 371                 trans = btrfs_start_transaction(root, 1);
 372                 if (ret != -EAGAIN)
 373                         break;
 374         }
 375         root->defrag_running = 0;
 376         radix_tree_tag_clear(&info->fs_roots_radix,
 377                      (unsigned long)root->root_key.objectid,
 378                      BTRFS_ROOT_DEFRAG_TAG);
 379         btrfs_end_transaction(trans, root);
 380         return 0;
 381 }
 382
 383 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
 384 {
 385         struct btrfs_root *gang[1];
 386         struct btrfs_root *root;
 387         int i;
 388         int ret;
 389         int err = 0;
 390         u64 last = 0;
 391
 392         while(1) {
 393                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
 394                                                  (void **)gang, last,
 395                                                  ARRAY_SIZE(gang),
 396                                                  BTRFS_ROOT_DEFRAG_TAG);
 397                 if (ret == 0)
 398                         break;
 399                 for (i = 0; i < ret; i++) {
 400                         root = gang[i];
 401                         last = root->root_key.objectid + 1;
 402                         btrfs_defrag_root(root, 1);
 403                 }
 404         }
 405         btrfs_defrag_root(info->extent_root, 1);
 406         return err;
 407 }
 408
 409 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 410                                      struct list_head *list)
 411 {
 412         struct dirty_root *dirty;
 413         struct btrfs_trans_handle *trans;
 414         unsigned long nr;
 415         u64 num_bytes;
 416         u64 bytes_used;
 417         int ret = 0;
 418         int err;
 419
 420         while(!list_empty(list)) {
 421                 struct btrfs_root *root;
 422
 423                 mutex_lock(&tree_root->fs_info->fs_mutex);
 424                 dirty = list_entry(list->next, struct dirty_root, list);
 425                 list_del_init(&dirty->list);
 426
 427                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 428                 root = dirty->latest_root;
 429                 root->fs_info->throttles++;
 430
 431                 while(1) {
 432                         trans = btrfs_start_transaction(tree_root, 1);
 433                         ret = btrfs_drop_snapshot(trans, dirty->root);
 434                         if (ret != -EAGAIN) {
 435                                 break;
 436                         }
 437
 438                         err = btrfs_update_root(trans,
 439                                         tree_root,
 440                                         &dirty->root->root_key,
 441                                         &dirty->root->root_item);
 442                         if (err)
 443                                 ret = err;
 444                         nr = trans->blocks_used;
 445                         ret = btrfs_end_transaction(trans, tree_root);
 446                         BUG_ON(ret);
 447                         mutex_unlock(&tree_root->fs_info->fs_mutex);
 448                         btrfs_btree_balance_dirty(tree_root, nr);
 449                         cond_resched();
 450                         mutex_lock(&tree_root->fs_info->fs_mutex);
 451                 }
 452                 BUG_ON(ret);
 453                 root->fs_info->throttles--;
 454
 455                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 456                 bytes_used = btrfs_root_used(&root->root_item);
 457                 if (num_bytes) {
 458                         record_root_in_trans(root);
 459                         btrfs_set_root_used(&root->root_item,
 460                                             bytes_used - num_bytes);
 461                 }
 462                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 463                 if (ret) {
 464                         BUG();
 465                         break;
 466                 }
 467                 nr = trans->blocks_used;
 468                 ret = btrfs_end_transaction(trans, tree_root);
 469                 BUG_ON(ret);
 470
 471                 free_extent_buffer(dirty->root->node);
 472                 kfree(dirty->root);
 473                 kfree(dirty);
 474                 mutex_unlock(&tree_root->fs_info->fs_mutex);
 475
 476                 btrfs_btree_balance_dirty(tree_root, nr);
 477                 cond_resched();
 478         }
 479         return ret;
 480 }
 481
 482 int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
 483                                 struct btrfs_root *root)
 484 {
 485         struct btrfs_transaction *cur_trans = trans->transaction;
 486         struct inode *inode;
 487         u64 root_objectid = 0;
 488         u64 objectid = 0;
 489         int ret;
 490
 491         root->fs_info->throttles++;
 492         while(1) {
 493                 ret = btrfs_find_first_ordered_inode(
 494                                 &cur_trans->ordered_inode_tree,
 495                                 &root_objectid, &objectid, &inode);
 496                 if (!ret)
 497                         break;
 498
 499                 mutex_unlock(&root->fs_info->trans_mutex);
 500                 mutex_unlock(&root->fs_info->fs_mutex);
 501
 502                 if (S_ISREG(inode->i_mode))
 503                         filemap_fdatawrite(inode->i_mapping);
 504                 iput(inode);
 505
 506                 mutex_lock(&root->fs_info->fs_mutex);
 507                 mutex_lock(&root->fs_info->trans_mutex);
 508         }
 509         while(1) {
 510                 root_objectid = 0;
 511                 objectid = 0;
 512                 ret = btrfs_find_del_first_ordered_inode(
 513                                 &cur_trans->ordered_inode_tree,
 514                                 &root_objectid, &objectid, &inode);
 515                 if (!ret)
 516                         break;
 517                 mutex_unlock(&root->fs_info->trans_mutex);
 518                 mutex_unlock(&root->fs_info->fs_mutex);
 519
 520                 if (S_ISREG(inode->i_mode))
 521                         filemap_write_and_wait(inode->i_mapping);
 522                 atomic_dec(&inode->i_count);
 523                 iput(inode);
 524
 525                 mutex_lock(&root->fs_info->fs_mutex);
 526                 mutex_lock(&root->fs_info->trans_mutex);
 527         }
 528         root->fs_info->throttles--;
 529         return 0;
 530 }
 531
 532 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 533                                    struct btrfs_fs_info *fs_info,
 534                                    struct btrfs_pending_snapshot *pending)
 535 {
 536         struct btrfs_key key;
 537         struct btrfs_root_item *new_root_item;
 538         struct btrfs_root *tree_root = fs_info->tree_root;
 539         struct btrfs_root *root = pending->root;
 540         struct extent_buffer *tmp;
 541         int ret;
 542         u64 objectid;
 543
 544         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 545         if (!new_root_item) {
 546                 ret = -ENOMEM;
 547                 goto fail;
 548         }
 549         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 550         if (ret)
 551                 goto fail;
 552
 553         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 554
 555         key.objectid = objectid;
 556         key.offset = 1;
 557         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 558
 559         extent_buffer_get(root->node);
 560         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
 561         free_extent_buffer(tmp);
 562
 563         btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 564
 565         btrfs_set_root_bytenr(new_root_item, tmp->start);
 566         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 567         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 568                                 new_root_item);
 569         free_extent_buffer(tmp);
 570         if (ret)
 571                 goto fail;
 572
 573         /*
 574          * insert the directory item
 575          */
 576         key.offset = (u64)-1;
 577         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 578                                     pending->name, strlen(pending->name),
 579                                     root->fs_info->sb->s_root->d_inode->i_ino,
 580                                     &key, BTRFS_FT_DIR);
 581
 582         if (ret)
 583                 goto fail;
 584
 585         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 586                              pending->name, strlen(pending->name), objectid,
 587                              root->fs_info->sb->s_root->d_inode->i_ino);
 588 fail:
 589         kfree(new_root_item);
 590         return ret;
 591 }
 592
 593 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 594                                              struct btrfs_fs_info *fs_info)
 595 {
 596         struct btrfs_pending_snapshot *pending;
 597         struct list_head *head = &trans->transaction->pending_snapshots;
 598         int ret;
 599
 600         while(!list_empty(head)) {
 601                 pending = list_entry(head->next,
 602                                      struct btrfs_pending_snapshot, list);
 603                 ret = create_pending_snapshot(trans, fs_info, pending);
 604                 BUG_ON(ret);
 605                 list_del(&pending->list);
 606                 kfree(pending->name);
 607                 kfree(pending);
 608         }
 609         return 0;
 610 }
 611
 612 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 613                              struct btrfs_root *root)
 614 {
 615         unsigned long joined = 0;
 616         unsigned long timeout = 1;
 617         struct btrfs_transaction *cur_trans;
 618         struct btrfs_transaction *prev_trans = NULL;
 619         struct list_head dirty_fs_roots;
 620         struct extent_io_tree *pinned_copy;
 621         DEFINE_WAIT(wait);
 622         int ret;
 623
 624         INIT_LIST_HEAD(&dirty_fs_roots);
 625
 626         mutex_lock(&root->fs_info->trans_mutex);
 627         if (trans->transaction->in_commit) {
 628                 cur_trans = trans->transaction;
 629                 trans->transaction->use_count++;
 630                 mutex_unlock(&root->fs_info->trans_mutex);
 631                 btrfs_end_transaction(trans, root);
 632
 633                 mutex_unlock(&root->fs_info->fs_mutex);
 634                 ret = wait_for_commit(root, cur_trans);
 635                 BUG_ON(ret);
 636
 637                 mutex_lock(&root->fs_info->trans_mutex);
 638                 put_transaction(cur_trans);
 639                 mutex_unlock(&root->fs_info->trans_mutex);
 640
 641                 mutex_lock(&root->fs_info->fs_mutex);
 642                 return 0;
 643         }
 644
 645         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 646         if (!pinned_copy)
 647                 return -ENOMEM;
 648
 649         extent_io_tree_init(pinned_copy,
 650                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 651
 652         trans->transaction->in_commit = 1;
 653         cur_trans = trans->transaction;
 654         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 655                 prev_trans = list_entry(cur_trans->list.prev,
 656                                         struct btrfs_transaction, list);
 657                 if (!prev_trans->commit_done) {
 658                         prev_trans->use_count++;
 659                         mutex_unlock(&root->fs_info->fs_mutex);
 660                         mutex_unlock(&root->fs_info->trans_mutex);
 661
 662                         wait_for_commit(root, prev_trans);
 663
 664                         mutex_lock(&root->fs_info->fs_mutex);
 665                         mutex_lock(&root->fs_info->trans_mutex);
 666                         put_transaction(prev_trans);
 667                 }
 668         }
 669
 670         do {
 671                 joined = cur_trans->num_joined;
 672                 WARN_ON(cur_trans != trans->transaction);
 673                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 674                                 TASK_UNINTERRUPTIBLE);
 675
 676                 if (cur_trans->num_writers > 1)
 677                         timeout = MAX_SCHEDULE_TIMEOUT;
 678                 else
 679                         timeout = 1;
 680
 681                 mutex_unlock(&root->fs_info->fs_mutex);
 682                 mutex_unlock(&root->fs_info->trans_mutex);
 683
 684                 schedule_timeout(timeout);
 685
 686                 mutex_lock(&root->fs_info->fs_mutex);
 687                 mutex_lock(&root->fs_info->trans_mutex);
 688                 finish_wait(&cur_trans->writer_wait, &wait);
 689                 ret = btrfs_write_ordered_inodes(trans, root);
 690
 691         } while (cur_trans->num_writers > 1 ||
 692                  (cur_trans->num_joined != joined));
 693
 694         ret = create_pending_snapshots(trans, root->fs_info);
 695         BUG_ON(ret);
 696
 697         WARN_ON(cur_trans != trans->transaction);
 698
 699         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 700                               &dirty_fs_roots);
 701         BUG_ON(ret);
 702
 703         ret = btrfs_commit_tree_roots(trans, root);
 704         BUG_ON(ret);
 705
 706         cur_trans = root->fs_info->running_transaction;
 707         spin_lock(&root->fs_info->new_trans_lock);
 708         root->fs_info->running_transaction = NULL;
 709         spin_unlock(&root->fs_info->new_trans_lock);
 710         btrfs_set_super_generation(&root->fs_info->super_copy,
 711                                    cur_trans->transid);
 712         btrfs_set_super_root(&root->fs_info->super_copy,
 713                              root->fs_info->tree_root->node->start);
 714         btrfs_set_super_root_level(&root->fs_info->super_copy,
 715                            btrfs_header_level(root->fs_info->tree_root->node));
 716
 717         write_extent_buffer(root->fs_info->sb_buffer,
 718                             &root->fs_info->super_copy, 0,
 719                             sizeof(root->fs_info->super_copy));
 720
 721         btrfs_copy_pinned(root, pinned_copy);
 722
 723         mutex_unlock(&root->fs_info->trans_mutex);
 724         mutex_unlock(&root->fs_info->fs_mutex);
 725         ret = btrfs_write_and_wait_transaction(trans, root);
 726         BUG_ON(ret);
 727         write_ctree_super(trans, root);
 728
 729         mutex_lock(&root->fs_info->fs_mutex);
 730         btrfs_finish_extent_commit(trans, root, pinned_copy);
 731         mutex_lock(&root->fs_info->trans_mutex);
 732
 733         kfree(pinned_copy);
 734
 735         cur_trans->commit_done = 1;
 736         root->fs_info->last_trans_committed = cur_trans->transid;
 737         wake_up(&cur_trans->commit_wait);
 738         put_transaction(cur_trans);
 739         put_transaction(cur_trans);
 740
 741         if (root->fs_info->closing)
 742                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 743         else
 744                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 745
 746         mutex_unlock(&root->fs_info->trans_mutex);
 747         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 748
 749         if (root->fs_info->closing) {
 750                 mutex_unlock(&root->fs_info->fs_mutex);
 751                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 752                 mutex_lock(&root->fs_info->fs_mutex);
 753         }
 754         return ret;
 755 }
 756
 757 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 758 {
 759         struct list_head dirty_roots;
 760         INIT_LIST_HEAD(&dirty_roots);
 761
 762         mutex_lock(&root->fs_info->trans_mutex);
 763         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 764         mutex_unlock(&root->fs_info->trans_mutex);
 765
 766         if (!list_empty(&dirty_roots)) {
 767                 drop_dirty_roots(root, &dirty_roots);
 768         }
 769         return 0;
 770 }
 771 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 772 void btrfs_transaction_cleaner(void *p)
 773 #else
 774 void btrfs_transaction_cleaner(struct work_struct *work)
 775 #endif
 776 {
 777 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 778         struct btrfs_fs_info *fs_info = p;
 779 #else
 780         struct btrfs_fs_info *fs_info = container_of(work,
 781                                                      struct btrfs_fs_info,
 782                                                      trans_work.work);
 783
 784 #endif
 785         struct btrfs_root *root = fs_info->tree_root;
 786         struct btrfs_transaction *cur;
 787         struct btrfs_trans_handle *trans;
 788         unsigned long now;
 789         unsigned long delay = HZ * 30;
 790         int ret;
 791
 792         mutex_lock(&root->fs_info->fs_mutex);
 793         mutex_lock(&root->fs_info->trans_mutex);
 794         cur = root->fs_info->running_transaction;
 795         if (!cur) {
 796                 mutex_unlock(&root->fs_info->trans_mutex);
 797                 goto out;
 798         }
 799         now = get_seconds();
 800         if (now < cur->start_time || now - cur->start_time < 30) {
 801                 mutex_unlock(&root->fs_info->trans_mutex);
 802                 delay = HZ * 5;
 803                 goto out;
 804         }
 805         mutex_unlock(&root->fs_info->trans_mutex);
 806         btrfs_defrag_dirty_roots(root->fs_info);
 807         trans = btrfs_start_transaction(root, 1);
 808         ret = btrfs_commit_transaction(trans, root);
 809 out:
 810         mutex_unlock(&root->fs_info->fs_mutex);
 811         btrfs_clean_old_snapshots(root);
 812         btrfs_transaction_queue_work(root, delay);
 813 }
 814
 815 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
 816 {
 817         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
 818 }
 819
 820 void btrfs_transaction_flush_work(struct btrfs_root *root)
 821 {
 822         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
 823         flush_workqueue(trans_wq);
 824 }
 825
 826 void __init btrfs_init_transaction_sys(void)
 827 {
 828         trans_wq = create_workqueue("btrfs");
 829 }
 830
 831 void btrfs_exit_transaction_sys(void)
 832 {
 833         destroy_workqueue(trans_wq);
 834 }
 835