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 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 202                                struct btrfs_root *root)
 203 {
 204         int ret;
 205         u64 old_root_bytenr;
 206         struct btrfs_root *tree_root = root->fs_info->tree_root;
 207
 208         btrfs_write_dirty_block_groups(trans, root);
 209         while(1) {
 210                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 211                 if (old_root_bytenr == root->node->start)
 212                         break;
 213                 btrfs_set_root_bytenr(&root->root_item,
 214                                        root->node->start);
 215                 btrfs_set_root_level(&root->root_item,
 216                                      btrfs_header_level(root->node));
 217                 ret = btrfs_update_root(trans, tree_root,
 218                                         &root->root_key,
 219                                         &root->root_item);
 220                 BUG_ON(ret);
 221                 btrfs_write_dirty_block_groups(trans, root);
 222         }
 223         return 0;
 224 }
 225
 226 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 227                             struct btrfs_root *root)
 228 {
 229         struct btrfs_fs_info *fs_info = root->fs_info;
 230         struct list_head *next;
 231
 232         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 233                 next = fs_info->dirty_cowonly_roots.next;
 234                 list_del_init(next);
 235                 root = list_entry(next, struct btrfs_root, dirty_list);
 236                 update_cowonly_root(trans, root);
 237         }
 238         return 0;
 239 }
 240
 241 static noinline int wait_for_commit(struct btrfs_root *root,
 242                                     struct btrfs_transaction *commit)
 243 {
 244         DEFINE_WAIT(wait);
 245         mutex_lock(&root->fs_info->trans_mutex);
 246         while(!commit->commit_done) {
 247                 prepare_to_wait(&commit->commit_wait, &wait,
 248                                 TASK_UNINTERRUPTIBLE);
 249                 if (commit->commit_done)
 250                         break;
 251                 mutex_unlock(&root->fs_info->trans_mutex);
 252                 schedule();
 253                 mutex_lock(&root->fs_info->trans_mutex);
 254         }
 255         mutex_unlock(&root->fs_info->trans_mutex);
 256         finish_wait(&commit->commit_wait, &wait);
 257         return 0;
 258 }
 259
 260 struct dirty_root {
 261         struct list_head list;
 262         struct btrfs_root *root;
 263         struct btrfs_root *latest_root;
 264 };
 265
 266 int btrfs_add_dead_root(struct btrfs_root *root,
 267                         struct btrfs_root *latest,
 268                         struct list_head *dead_list)
 269 {
 270         struct dirty_root *dirty;
 271
 272         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 273         if (!dirty)
 274                 return -ENOMEM;
 275         dirty->root = root;
 276         dirty->latest_root = latest;
 277         list_add(&dirty->list, dead_list);
 278         return 0;
 279 }
 280
 281 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 282                                     struct radix_tree_root *radix,
 283                                     struct list_head *list)
 284 {
 285         struct dirty_root *dirty;
 286         struct btrfs_root *gang[8];
 287         struct btrfs_root *root;
 288         int i;
 289         int ret;
 290         int err = 0;
 291         u32 refs;
 292
 293         while(1) {
 294                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 295                                                  ARRAY_SIZE(gang),
 296                                                  BTRFS_ROOT_TRANS_TAG);
 297                 if (ret == 0)
 298                         break;
 299                 for (i = 0; i < ret; i++) {
 300                         root = gang[i];
 301                         radix_tree_tag_clear(radix,
 302                                      (unsigned long)root->root_key.objectid,
 303                                      BTRFS_ROOT_TRANS_TAG);
 304                         if (root->commit_root == root->node) {
 305                                 WARN_ON(root->node->start !=
 306                                         btrfs_root_bytenr(&root->root_item));
 307                                 free_extent_buffer(root->commit_root);
 308                                 root->commit_root = NULL;
 309
 310                                 /* make sure to update the root on disk
 311                                  * so we get any updates to the block used
 312                                  * counts
 313                                  */
 314                                 err = btrfs_update_root(trans,
 315                                                 root->fs_info->tree_root,
 316                                                 &root->root_key,
 317                                                 &root->root_item);
 318                                 continue;
 319                         }
 320                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 321                         BUG_ON(!dirty);
 322                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 323                         BUG_ON(!dirty->root);
 324
 325                         memset(&root->root_item.drop_progress, 0,
 326                                sizeof(struct btrfs_disk_key));
 327                         root->root_item.drop_level = 0;
 328
 329                         memcpy(dirty->root, root, sizeof(*root));
 330                         dirty->root->node = root->commit_root;
 331                         dirty->latest_root = root;
 332                         root->commit_root = NULL;
 333
 334                         root->root_key.offset = root->fs_info->generation;
 335                         btrfs_set_root_bytenr(&root->root_item,
 336                                               root->node->start);
 337                         btrfs_set_root_level(&root->root_item,
 338                                              btrfs_header_level(root->node));
 339                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 340                                                 &root->root_key,
 341                                                 &root->root_item);
 342                         if (err)
 343                                 break;
 344
 345                         refs = btrfs_root_refs(&dirty->root->root_item);
 346                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 347                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 348                                                 &dirty->root->root_key,
 349                                                 &dirty->root->root_item);
 350
 351                         BUG_ON(err);
 352                         if (refs == 1) {
 353                                 list_add(&dirty->list, list);
 354                         } else {
 355                                 WARN_ON(1);
 356                                 kfree(dirty->root);
 357                                 kfree(dirty);
 358                         }
 359                 }
 360         }
 361         return err;
 362 }
 363
 364 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 365 {
 366         struct btrfs_fs_info *info = root->fs_info;
 367         int ret;
 368         struct btrfs_trans_handle *trans;
 369         unsigned long nr;
 370
 371         if (root->defrag_running)
 372                 return 0;
 373         trans = btrfs_start_transaction(root, 1);
 374         while (1) {
 375                 root->defrag_running = 1;
 376                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 377                 nr = trans->blocks_used;
 378                 btrfs_end_transaction(trans, root);
 379                 mutex_unlock(&info->fs_mutex);
 380                 btrfs_btree_balance_dirty(info->tree_root, nr);
 381                 cond_resched();
 382
 383                 mutex_lock(&info->fs_mutex);
 384                 trans = btrfs_start_transaction(root, 1);
 385                 if (ret != -EAGAIN)
 386                         break;
 387         }
 388         root->defrag_running = 0;
 389         radix_tree_tag_clear(&info->fs_roots_radix,
 390                      (unsigned long)root->root_key.objectid,
 391                      BTRFS_ROOT_DEFRAG_TAG);
 392         btrfs_end_transaction(trans, root);
 393         return 0;
 394 }
 395
 396 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
 397 {
 398         struct btrfs_root *gang[1];
 399         struct btrfs_root *root;
 400         int i;
 401         int ret;
 402         int err = 0;
 403         u64 last = 0;
 404
 405         while(1) {
 406                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
 407                                                  (void **)gang, last,
 408                                                  ARRAY_SIZE(gang),
 409                                                  BTRFS_ROOT_DEFRAG_TAG);
 410                 if (ret == 0)
 411                         break;
 412                 for (i = 0; i < ret; i++) {
 413                         root = gang[i];
 414                         last = root->root_key.objectid + 1;
 415                         btrfs_defrag_root(root, 1);
 416                 }
 417         }
 418         btrfs_defrag_root(info->extent_root, 1);
 419         return err;
 420 }
 421
 422 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 423                                      struct list_head *list)
 424 {
 425         struct dirty_root *dirty;
 426         struct btrfs_trans_handle *trans;
 427         unsigned long nr;
 428         u64 num_bytes;
 429         u64 bytes_used;
 430         int ret = 0;
 431         int err;
 432
 433         while(!list_empty(list)) {
 434                 struct btrfs_root *root;
 435
 436                 mutex_lock(&tree_root->fs_info->fs_mutex);
 437                 dirty = list_entry(list->next, struct dirty_root, list);
 438                 list_del_init(&dirty->list);
 439
 440                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 441                 root = dirty->latest_root;
 442                 root->fs_info->throttles++;
 443
 444                 while(1) {
 445                         trans = btrfs_start_transaction(tree_root, 1);
 446                         ret = btrfs_drop_snapshot(trans, dirty->root);
 447                         if (ret != -EAGAIN) {
 448                                 break;
 449                         }
 450
 451                         err = btrfs_update_root(trans,
 452                                         tree_root,
 453                                         &dirty->root->root_key,
 454                                         &dirty->root->root_item);
 455                         if (err)
 456                                 ret = err;
 457                         nr = trans->blocks_used;
 458                         ret = btrfs_end_transaction(trans, tree_root);
 459                         BUG_ON(ret);
 460                         mutex_unlock(&tree_root->fs_info->fs_mutex);
 461                         btrfs_btree_balance_dirty(tree_root, nr);
 462                         cond_resched();
 463                         mutex_lock(&tree_root->fs_info->fs_mutex);
 464                 }
 465                 BUG_ON(ret);
 466                 root->fs_info->throttles--;
 467
 468                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 469                 bytes_used = btrfs_root_used(&root->root_item);
 470                 if (num_bytes) {
 471                         record_root_in_trans(root);
 472                         btrfs_set_root_used(&root->root_item,
 473                                             bytes_used - num_bytes);
 474                 }
 475                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 476                 if (ret) {
 477                         BUG();
 478                         break;
 479                 }
 480                 nr = trans->blocks_used;
 481                 ret = btrfs_end_transaction(trans, tree_root);
 482                 BUG_ON(ret);
 483
 484                 free_extent_buffer(dirty->root->node);
 485                 kfree(dirty->root);
 486                 kfree(dirty);
 487                 mutex_unlock(&tree_root->fs_info->fs_mutex);
 488
 489                 btrfs_btree_balance_dirty(tree_root, nr);
 490                 cond_resched();
 491         }
 492         return ret;
 493 }
 494
 495 int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
 496                                 struct btrfs_root *root)
 497 {
 498         struct btrfs_transaction *cur_trans = trans->transaction;
 499         struct inode *inode;
 500         u64 root_objectid = 0;
 501         u64 objectid = 0;
 502         int ret;
 503
 504         root->fs_info->throttles++;
 505         while(1) {
 506                 ret = btrfs_find_first_ordered_inode(
 507                                 &cur_trans->ordered_inode_tree,
 508                                 &root_objectid, &objectid, &inode);
 509                 if (!ret)
 510                         break;
 511
 512                 mutex_unlock(&root->fs_info->trans_mutex);
 513                 mutex_unlock(&root->fs_info->fs_mutex);
 514
 515                 if (S_ISREG(inode->i_mode))
 516                         filemap_fdatawrite(inode->i_mapping);
 517                 iput(inode);
 518
 519                 mutex_lock(&root->fs_info->fs_mutex);
 520                 mutex_lock(&root->fs_info->trans_mutex);
 521         }
 522         while(1) {
 523                 root_objectid = 0;
 524                 objectid = 0;
 525                 ret = btrfs_find_del_first_ordered_inode(
 526                                 &cur_trans->ordered_inode_tree,
 527                                 &root_objectid, &objectid, &inode);
 528                 if (!ret)
 529                         break;
 530                 mutex_unlock(&root->fs_info->trans_mutex);
 531                 mutex_unlock(&root->fs_info->fs_mutex);
 532
 533                 if (S_ISREG(inode->i_mode))
 534                         filemap_write_and_wait(inode->i_mapping);
 535                 atomic_dec(&inode->i_count);
 536                 iput(inode);
 537
 538                 mutex_lock(&root->fs_info->fs_mutex);
 539                 mutex_lock(&root->fs_info->trans_mutex);
 540         }
 541         root->fs_info->throttles--;
 542         return 0;
 543 }
 544
 545 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 546                                    struct btrfs_fs_info *fs_info,
 547                                    struct btrfs_pending_snapshot *pending)
 548 {
 549         struct btrfs_key key;
 550         struct btrfs_root_item *new_root_item;
 551         struct btrfs_root *tree_root = fs_info->tree_root;
 552         struct btrfs_root *root = pending->root;
 553         struct extent_buffer *tmp;
 554         int ret;
 555         u64 objectid;
 556
 557         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 558         if (!new_root_item) {
 559                 ret = -ENOMEM;
 560                 goto fail;
 561         }
 562         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 563         if (ret)
 564                 goto fail;
 565
 566         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 567
 568         key.objectid = objectid;
 569         key.offset = 1;
 570         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 571
 572         extent_buffer_get(root->node);
 573         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
 574         free_extent_buffer(tmp);
 575
 576         btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 577
 578         btrfs_set_root_bytenr(new_root_item, tmp->start);
 579         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 580         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 581                                 new_root_item);
 582         free_extent_buffer(tmp);
 583         if (ret)
 584                 goto fail;
 585
 586         /*
 587          * insert the directory item
 588          */
 589         key.offset = (u64)-1;
 590         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 591                                     pending->name, strlen(pending->name),
 592                                     root->fs_info->sb->s_root->d_inode->i_ino,
 593                                     &key, BTRFS_FT_DIR);
 594
 595         if (ret)
 596                 goto fail;
 597
 598         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 599                              pending->name, strlen(pending->name), objectid,
 600                              root->fs_info->sb->s_root->d_inode->i_ino);
 601 fail:
 602         kfree(new_root_item);
 603         return ret;
 604 }
 605
 606 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 607                                              struct btrfs_fs_info *fs_info)
 608 {
 609         struct btrfs_pending_snapshot *pending;
 610         struct list_head *head = &trans->transaction->pending_snapshots;
 611         int ret;
 612
 613         while(!list_empty(head)) {
 614                 pending = list_entry(head->next,
 615                                      struct btrfs_pending_snapshot, list);
 616                 ret = create_pending_snapshot(trans, fs_info, pending);
 617                 BUG_ON(ret);
 618                 list_del(&pending->list);
 619                 kfree(pending->name);
 620                 kfree(pending);
 621         }
 622         return 0;
 623 }
 624
 625 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 626                              struct btrfs_root *root)
 627 {
 628         unsigned long joined = 0;
 629         unsigned long timeout = 1;
 630         struct btrfs_transaction *cur_trans;
 631         struct btrfs_transaction *prev_trans = NULL;
 632         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 633         struct list_head dirty_fs_roots;
 634         struct extent_io_tree *pinned_copy;
 635         DEFINE_WAIT(wait);
 636         int ret;
 637
 638         INIT_LIST_HEAD(&dirty_fs_roots);
 639
 640         mutex_lock(&root->fs_info->trans_mutex);
 641         if (trans->transaction->in_commit) {
 642                 cur_trans = trans->transaction;
 643                 trans->transaction->use_count++;
 644                 mutex_unlock(&root->fs_info->trans_mutex);
 645                 btrfs_end_transaction(trans, root);
 646
 647                 mutex_unlock(&root->fs_info->fs_mutex);
 648                 ret = wait_for_commit(root, cur_trans);
 649                 BUG_ON(ret);
 650
 651                 mutex_lock(&root->fs_info->trans_mutex);
 652                 put_transaction(cur_trans);
 653                 mutex_unlock(&root->fs_info->trans_mutex);
 654
 655                 mutex_lock(&root->fs_info->fs_mutex);
 656                 return 0;
 657         }
 658
 659         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 660         if (!pinned_copy)
 661                 return -ENOMEM;
 662
 663         extent_io_tree_init(pinned_copy,
 664                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 665
 666         trans->transaction->in_commit = 1;
 667         cur_trans = trans->transaction;
 668         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 669                 prev_trans = list_entry(cur_trans->list.prev,
 670                                         struct btrfs_transaction, list);
 671                 if (!prev_trans->commit_done) {
 672                         prev_trans->use_count++;
 673                         mutex_unlock(&root->fs_info->fs_mutex);
 674                         mutex_unlock(&root->fs_info->trans_mutex);
 675
 676                         wait_for_commit(root, prev_trans);
 677
 678                         mutex_lock(&root->fs_info->fs_mutex);
 679                         mutex_lock(&root->fs_info->trans_mutex);
 680                         put_transaction(prev_trans);
 681                 }
 682         }
 683
 684         do {
 685                 joined = cur_trans->num_joined;
 686                 WARN_ON(cur_trans != trans->transaction);
 687                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 688                                 TASK_UNINTERRUPTIBLE);
 689
 690                 if (cur_trans->num_writers > 1)
 691                         timeout = MAX_SCHEDULE_TIMEOUT;
 692                 else
 693                         timeout = 1;
 694
 695                 mutex_unlock(&root->fs_info->fs_mutex);
 696                 mutex_unlock(&root->fs_info->trans_mutex);
 697
 698                 schedule_timeout(timeout);
 699
 700                 mutex_lock(&root->fs_info->fs_mutex);
 701                 mutex_lock(&root->fs_info->trans_mutex);
 702                 finish_wait(&cur_trans->writer_wait, &wait);
 703                 ret = btrfs_write_ordered_inodes(trans, root);
 704
 705         } while (cur_trans->num_writers > 1 ||
 706                  (cur_trans->num_joined != joined));
 707
 708         ret = create_pending_snapshots(trans, root->fs_info);
 709         BUG_ON(ret);
 710
 711         WARN_ON(cur_trans != trans->transaction);
 712
 713         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 714                               &dirty_fs_roots);
 715         BUG_ON(ret);
 716
 717         ret = btrfs_commit_tree_roots(trans, root);
 718         BUG_ON(ret);
 719
 720         cur_trans = root->fs_info->running_transaction;
 721         spin_lock(&root->fs_info->new_trans_lock);
 722         root->fs_info->running_transaction = NULL;
 723         spin_unlock(&root->fs_info->new_trans_lock);
 724         btrfs_set_super_generation(&root->fs_info->super_copy,
 725                                    cur_trans->transid);
 726         btrfs_set_super_root(&root->fs_info->super_copy,
 727                              root->fs_info->tree_root->node->start);
 728         btrfs_set_super_root_level(&root->fs_info->super_copy,
 729                            btrfs_header_level(root->fs_info->tree_root->node));
 730
 731         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 732                                    chunk_root->node->start);
 733         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 734                                          btrfs_header_level(chunk_root->node));
 735         write_extent_buffer(root->fs_info->sb_buffer,
 736                             &root->fs_info->super_copy, 0,
 737                             sizeof(root->fs_info->super_copy));
 738
 739         btrfs_copy_pinned(root, pinned_copy);
 740
 741         mutex_unlock(&root->fs_info->trans_mutex);
 742         mutex_unlock(&root->fs_info->fs_mutex);
 743         ret = btrfs_write_and_wait_transaction(trans, root);
 744         BUG_ON(ret);
 745         write_ctree_super(trans, root);
 746
 747         mutex_lock(&root->fs_info->fs_mutex);
 748         btrfs_finish_extent_commit(trans, root, pinned_copy);
 749         mutex_lock(&root->fs_info->trans_mutex);
 750
 751         kfree(pinned_copy);
 752
 753         cur_trans->commit_done = 1;
 754         root->fs_info->last_trans_committed = cur_trans->transid;
 755         wake_up(&cur_trans->commit_wait);
 756         put_transaction(cur_trans);
 757         put_transaction(cur_trans);
 758
 759         if (root->fs_info->closing)
 760                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 761         else
 762                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 763
 764         mutex_unlock(&root->fs_info->trans_mutex);
 765         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 766
 767         if (root->fs_info->closing) {
 768                 mutex_unlock(&root->fs_info->fs_mutex);
 769                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 770                 mutex_lock(&root->fs_info->fs_mutex);
 771         }
 772         return ret;
 773 }
 774
 775 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 776 {
 777         struct list_head dirty_roots;
 778         INIT_LIST_HEAD(&dirty_roots);
 779
 780         mutex_lock(&root->fs_info->trans_mutex);
 781         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 782         mutex_unlock(&root->fs_info->trans_mutex);
 783
 784         if (!list_empty(&dirty_roots)) {
 785                 drop_dirty_roots(root, &dirty_roots);
 786         }
 787         return 0;
 788 }
 789 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 790 void btrfs_transaction_cleaner(void *p)
 791 #else
 792 void btrfs_transaction_cleaner(struct work_struct *work)
 793 #endif
 794 {
 795 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 796         struct btrfs_fs_info *fs_info = p;
 797 #else
 798         struct btrfs_fs_info *fs_info = container_of(work,
 799                                                      struct btrfs_fs_info,
 800                                                      trans_work.work);
 801
 802 #endif
 803         struct btrfs_root *root = fs_info->tree_root;
 804         struct btrfs_transaction *cur;
 805         struct btrfs_trans_handle *trans;
 806         unsigned long now;
 807         unsigned long delay = HZ * 30;
 808         int ret;
 809
 810         mutex_lock(&root->fs_info->fs_mutex);
 811         mutex_lock(&root->fs_info->trans_mutex);
 812         cur = root->fs_info->running_transaction;
 813         if (!cur) {
 814                 mutex_unlock(&root->fs_info->trans_mutex);
 815                 goto out;
 816         }
 817         now = get_seconds();
 818         if (now < cur->start_time || now - cur->start_time < 30) {
 819                 mutex_unlock(&root->fs_info->trans_mutex);
 820                 delay = HZ * 5;
 821                 goto out;
 822         }
 823         mutex_unlock(&root->fs_info->trans_mutex);
 824         btrfs_defrag_dirty_roots(root->fs_info);
 825         trans = btrfs_start_transaction(root, 1);
 826         ret = btrfs_commit_transaction(trans, root);
 827 out:
 828         mutex_unlock(&root->fs_info->fs_mutex);
 829         btrfs_clean_old_snapshots(root);
 830         btrfs_transaction_queue_work(root, delay);
 831 }
 832
 833 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
 834 {
 835         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
 836 }
 837
 838 void btrfs_transaction_flush_work(struct btrfs_root *root)
 839 {
 840         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
 841         flush_workqueue(trans_wq);
 842 }
 843
 844 void __init btrfs_init_transaction_sys(void)
 845 {
 846         trans_wq = create_workqueue("btrfs");
 847 }
 848
 849 void btrfs_exit_transaction_sys(void)
 850 {
 851         destroy_workqueue(trans_wq);
 852 }
 853