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 #include "locking.h"
  27 #include "ref-cache.h"
  28
  29 static int total_trans = 0;
  30 extern struct kmem_cache *btrfs_trans_handle_cachep;
  31 extern struct kmem_cache *btrfs_transaction_cachep;
  32
  33 #define BTRFS_ROOT_TRANS_TAG 0
  34
  35 static noinline void put_transaction(struct btrfs_transaction *transaction)
  36 {
  37         WARN_ON(transaction->use_count == 0);
  38         transaction->use_count--;
  39         if (transaction->use_count == 0) {
  40                 WARN_ON(total_trans == 0);
  41                 total_trans--;
  42                 list_del_init(&transaction->list);
  43                 memset(transaction, 0, sizeof(*transaction));
  44                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  45         }
  46 }
  47
  48 static noinline int join_transaction(struct btrfs_root *root)
  49 {
  50         struct btrfs_transaction *cur_trans;
  51         cur_trans = root->fs_info->running_transaction;
  52         if (!cur_trans) {
  53                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  54                                              GFP_NOFS);
  55                 total_trans++;
  56                 BUG_ON(!cur_trans);
  57                 root->fs_info->generation++;
  58                 root->fs_info->last_alloc = 0;
  59                 root->fs_info->last_data_alloc = 0;
  60                 cur_trans->num_writers = 1;
  61                 cur_trans->num_joined = 0;
  62                 cur_trans->transid = root->fs_info->generation;
  63                 init_waitqueue_head(&cur_trans->writer_wait);
  64                 init_waitqueue_head(&cur_trans->commit_wait);
  65                 cur_trans->in_commit = 0;
  66                 cur_trans->blocked = 0;
  67                 cur_trans->use_count = 1;
  68                 cur_trans->commit_done = 0;
  69                 cur_trans->start_time = get_seconds();
  70                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  71                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  72                 extent_io_tree_init(&cur_trans->dirty_pages,
  73                                      root->fs_info->btree_inode->i_mapping,
  74                                      GFP_NOFS);
  75                 spin_lock(&root->fs_info->new_trans_lock);
  76                 root->fs_info->running_transaction = cur_trans;
  77                 spin_unlock(&root->fs_info->new_trans_lock);
  78         } else {
  79                 cur_trans->num_writers++;
  80                 cur_trans->num_joined++;
  81         }
  82
  83         return 0;
  84 }
  85
  86 static noinline int record_root_in_trans(struct btrfs_root *root)
  87 {
  88         struct btrfs_dirty_root *dirty;
  89         u64 running_trans_id = root->fs_info->running_transaction->transid;
  90         if (root->ref_cows && root->last_trans < running_trans_id) {
  91                 WARN_ON(root == root->fs_info->extent_root);
  92                 if (root->root_item.refs != 0) {
  93                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  94                                    (unsigned long)root->root_key.objectid,
  95                                    BTRFS_ROOT_TRANS_TAG);
  96
  97                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
  98                         BUG_ON(!dirty);
  99                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 100                         BUG_ON(!dirty->root);
 101
 102                         dirty->latest_root = root;
 103                         INIT_LIST_HEAD(&dirty->list);
 104
 105                         root->commit_root = btrfs_root_node(root);
 106                         root->dirty_root = dirty;
 107
 108                         memcpy(dirty->root, root, sizeof(*root));
 109                         dirty->root->ref_tree = &root->ref_tree_struct;
 110
 111                         spin_lock_init(&dirty->root->node_lock);
 112                         mutex_init(&dirty->root->objectid_mutex);
 113                         dirty->root->node = root->commit_root;
 114                         dirty->root->commit_root = NULL;
 115                 } else {
 116                         WARN_ON(1);
 117                 }
 118                 root->last_trans = running_trans_id;
 119         }
 120         return 0;
 121 }
 122
 123 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 124                                              int num_blocks, int join)
 125 {
 126         struct btrfs_trans_handle *h =
 127                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 128         struct btrfs_transaction *cur_trans;
 129         int ret;
 130
 131         mutex_lock(&root->fs_info->trans_mutex);
 132         cur_trans = root->fs_info->running_transaction;
 133         if (cur_trans && cur_trans->blocked && !join) {
 134                 DEFINE_WAIT(wait);
 135                 cur_trans->use_count++;
 136                 while(1) {
 137                         prepare_to_wait(&root->fs_info->transaction_wait, &wait,
 138                                         TASK_UNINTERRUPTIBLE);
 139                         if (cur_trans->blocked) {
 140                                 mutex_unlock(&root->fs_info->trans_mutex);
 141                                 schedule();
 142                                 mutex_lock(&root->fs_info->trans_mutex);
 143                                 finish_wait(&root->fs_info->transaction_wait,
 144                                             &wait);
 145                         } else {
 146                                 finish_wait(&root->fs_info->transaction_wait,
 147                                             &wait);
 148                                 break;
 149                         }
 150                 }
 151                 put_transaction(cur_trans);
 152         }
 153         ret = join_transaction(root);
 154         BUG_ON(ret);
 155
 156         record_root_in_trans(root);
 157         h->transid = root->fs_info->running_transaction->transid;
 158         h->transaction = root->fs_info->running_transaction;
 159         h->blocks_reserved = num_blocks;
 160         h->blocks_used = 0;
 161         h->block_group = NULL;
 162         h->alloc_exclude_nr = 0;
 163         h->alloc_exclude_start = 0;
 164         root->fs_info->running_transaction->use_count++;
 165         mutex_unlock(&root->fs_info->trans_mutex);
 166         return h;
 167 }
 168
 169 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 170                                                    int num_blocks)
 171 {
 172         return start_transaction(root, num_blocks, 0);
 173 }
 174 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 175                                                    int num_blocks)
 176 {
 177         return start_transaction(root, num_blocks, 1);
 178 }
 179
 180 static noinline int wait_for_commit(struct btrfs_root *root,
 181                                     struct btrfs_transaction *commit)
 182 {
 183         DEFINE_WAIT(wait);
 184         mutex_lock(&root->fs_info->trans_mutex);
 185         while(!commit->commit_done) {
 186                 prepare_to_wait(&commit->commit_wait, &wait,
 187                                 TASK_UNINTERRUPTIBLE);
 188                 if (commit->commit_done)
 189                         break;
 190                 mutex_unlock(&root->fs_info->trans_mutex);
 191                 schedule();
 192                 mutex_lock(&root->fs_info->trans_mutex);
 193         }
 194         mutex_unlock(&root->fs_info->trans_mutex);
 195         finish_wait(&commit->commit_wait, &wait);
 196         return 0;
 197 }
 198
 199 void btrfs_throttle(struct btrfs_root *root)
 200 {
 201         struct btrfs_fs_info *info = root->fs_info;
 202
 203 harder:
 204         if (atomic_read(&info->throttles)) {
 205                 DEFINE_WAIT(wait);
 206                 int thr;
 207                 int harder_count = 0;
 208                 thr = atomic_read(&info->throttle_gen);
 209
 210                 do {
 211                         prepare_to_wait(&info->transaction_throttle,
 212                                         &wait, TASK_UNINTERRUPTIBLE);
 213                         if (!atomic_read(&info->throttles)) {
 214                                 finish_wait(&info->transaction_throttle, &wait);
 215                                 break;
 216                         }
 217                         schedule();
 218                         finish_wait(&info->transaction_throttle, &wait);
 219                 } while (thr == atomic_read(&info->throttle_gen));
 220
 221                 if (harder_count < 5 &&
 222                     info->total_ref_cache_size > 5 * 1024 * 1024) {
 223                         harder_count++;
 224                         goto harder;
 225                 }
 226
 227                 if (harder_count < 10 &&
 228                     info->total_ref_cache_size > 10 * 1024 * 1024) {
 229                         harder_count++;
 230                         goto harder;
 231                 }
 232         }
 233 }
 234
 235 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 236                           struct btrfs_root *root, int throttle)
 237 {
 238         struct btrfs_transaction *cur_trans;
 239         struct btrfs_fs_info *info = root->fs_info;
 240
 241         mutex_lock(&info->trans_mutex);
 242         cur_trans = info->running_transaction;
 243         WARN_ON(cur_trans != trans->transaction);
 244         WARN_ON(cur_trans->num_writers < 1);
 245         cur_trans->num_writers--;
 246
 247         if (waitqueue_active(&cur_trans->writer_wait))
 248                 wake_up(&cur_trans->writer_wait);
 249         put_transaction(cur_trans);
 250         mutex_unlock(&info->trans_mutex);
 251         memset(trans, 0, sizeof(*trans));
 252         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 253
 254         if (throttle)
 255                 btrfs_throttle(root);
 256
 257         return 0;
 258 }
 259
 260 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 261                           struct btrfs_root *root)
 262 {
 263         return __btrfs_end_transaction(trans, root, 0);
 264 }
 265
 266 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 267                                    struct btrfs_root *root)
 268 {
 269         return __btrfs_end_transaction(trans, root, 1);
 270 }
 271
 272
 273 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 274                                      struct btrfs_root *root)
 275 {
 276         int ret;
 277         int err;
 278         int werr = 0;
 279         struct extent_io_tree *dirty_pages;
 280         struct page *page;
 281         struct inode *btree_inode = root->fs_info->btree_inode;
 282         u64 start;
 283         u64 end;
 284         unsigned long index;
 285
 286         if (!trans || !trans->transaction) {
 287                 return filemap_write_and_wait(btree_inode->i_mapping);
 288         }
 289         dirty_pages = &trans->transaction->dirty_pages;
 290         while(1) {
 291                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 292                                             EXTENT_DIRTY);
 293                 if (ret)
 294                         break;
 295                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 296                 while(start <= end) {
 297                         index = start >> PAGE_CACHE_SHIFT;
 298                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 299                         page = find_lock_page(btree_inode->i_mapping, index);
 300                         if (!page)
 301                                 continue;
 302                         if (PageWriteback(page)) {
 303                                 if (PageDirty(page))
 304                                         wait_on_page_writeback(page);
 305                                 else {
 306                                         unlock_page(page);
 307                                         page_cache_release(page);
 308                                         continue;
 309                                 }
 310                         }
 311                         err = write_one_page(page, 0);
 312                         if (err)
 313                                 werr = err;
 314                         page_cache_release(page);
 315                 }
 316         }
 317         err = filemap_fdatawait(btree_inode->i_mapping);
 318         if (err)
 319                 werr = err;
 320         return werr;
 321 }
 322
 323 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 324                                struct btrfs_root *root)
 325 {
 326         int ret;
 327         u64 old_root_bytenr;
 328         struct btrfs_root *tree_root = root->fs_info->tree_root;
 329
 330         btrfs_write_dirty_block_groups(trans, root);
 331         while(1) {
 332                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 333                 if (old_root_bytenr == root->node->start)
 334                         break;
 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                 ret = btrfs_update_root(trans, tree_root,
 340                                         &root->root_key,
 341                                         &root->root_item);
 342                 BUG_ON(ret);
 343                 btrfs_write_dirty_block_groups(trans, root);
 344         }
 345         return 0;
 346 }
 347
 348 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 349                             struct btrfs_root *root)
 350 {
 351         struct btrfs_fs_info *fs_info = root->fs_info;
 352         struct list_head *next;
 353
 354         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 355                 next = fs_info->dirty_cowonly_roots.next;
 356                 list_del_init(next);
 357                 root = list_entry(next, struct btrfs_root, dirty_list);
 358                 update_cowonly_root(trans, root);
 359                 if (root->fs_info->closing)
 360                         btrfs_remove_leaf_refs(root);
 361         }
 362         return 0;
 363 }
 364
 365 int btrfs_add_dead_root(struct btrfs_root *root,
 366                         struct btrfs_root *latest,
 367                         struct list_head *dead_list)
 368 {
 369         struct btrfs_dirty_root *dirty;
 370
 371         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 372         if (!dirty)
 373                 return -ENOMEM;
 374         dirty->root = root;
 375         dirty->latest_root = latest;
 376         list_add(&dirty->list, dead_list);
 377         return 0;
 378 }
 379
 380 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 381                                     struct radix_tree_root *radix,
 382                                     struct list_head *list)
 383 {
 384         struct btrfs_dirty_root *dirty;
 385         struct btrfs_root *gang[8];
 386         struct btrfs_root *root;
 387         int i;
 388         int ret;
 389         int err = 0;
 390         u32 refs;
 391
 392         while(1) {
 393                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 394                                                  ARRAY_SIZE(gang),
 395                                                  BTRFS_ROOT_TRANS_TAG);
 396                 if (ret == 0)
 397                         break;
 398                 for (i = 0; i < ret; i++) {
 399                         root = gang[i];
 400                         radix_tree_tag_clear(radix,
 401                                      (unsigned long)root->root_key.objectid,
 402                                      BTRFS_ROOT_TRANS_TAG);
 403
 404                         BUG_ON(!root->ref_tree);
 405                         dirty = root->dirty_root;
 406
 407                         if (root->commit_root == root->node) {
 408                                 WARN_ON(root->node->start !=
 409                                         btrfs_root_bytenr(&root->root_item));
 410
 411                                 free_extent_buffer(root->commit_root);
 412                                 root->commit_root = NULL;
 413
 414                                 kfree(dirty->root);
 415                                 kfree(dirty);
 416
 417                                 /* make sure to update the root on disk
 418                                  * so we get any updates to the block used
 419                                  * counts
 420                                  */
 421                                 err = btrfs_update_root(trans,
 422                                                 root->fs_info->tree_root,
 423                                                 &root->root_key,
 424                                                 &root->root_item);
 425                                 continue;
 426                         }
 427
 428                         memset(&root->root_item.drop_progress, 0,
 429                                sizeof(struct btrfs_disk_key));
 430                         root->root_item.drop_level = 0;
 431                         root->commit_root = NULL;
 432                         root->root_key.offset = root->fs_info->generation;
 433                         btrfs_set_root_bytenr(&root->root_item,
 434                                               root->node->start);
 435                         btrfs_set_root_level(&root->root_item,
 436                                              btrfs_header_level(root->node));
 437                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 438                                                 &root->root_key,
 439                                                 &root->root_item);
 440                         if (err)
 441                                 break;
 442
 443                         refs = btrfs_root_refs(&dirty->root->root_item);
 444                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 445                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 446                                                 &dirty->root->root_key,
 447                                                 &dirty->root->root_item);
 448
 449                         BUG_ON(err);
 450                         if (refs == 1) {
 451                                 list_add(&dirty->list, list);
 452                         } else {
 453                                 WARN_ON(1);
 454                                 free_extent_buffer(dirty->root->node);
 455                                 kfree(dirty->root);
 456                                 kfree(dirty);
 457                         }
 458                 }
 459         }
 460         return err;
 461 }
 462
 463 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 464 {
 465         struct btrfs_fs_info *info = root->fs_info;
 466         int ret;
 467         struct btrfs_trans_handle *trans;
 468         unsigned long nr;
 469
 470         smp_mb();
 471         if (root->defrag_running)
 472                 return 0;
 473         trans = btrfs_start_transaction(root, 1);
 474         while (1) {
 475                 root->defrag_running = 1;
 476                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 477                 nr = trans->blocks_used;
 478                 btrfs_end_transaction(trans, root);
 479                 btrfs_btree_balance_dirty(info->tree_root, nr);
 480                 cond_resched();
 481
 482                 trans = btrfs_start_transaction(root, 1);
 483                 if (root->fs_info->closing || ret != -EAGAIN)
 484                         break;
 485         }
 486         root->defrag_running = 0;
 487         smp_mb();
 488         btrfs_end_transaction(trans, root);
 489         return 0;
 490 }
 491
 492 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 493                                      struct list_head *list)
 494 {
 495         struct btrfs_dirty_root *dirty;
 496         struct btrfs_trans_handle *trans;
 497         unsigned long nr;
 498         u64 num_bytes;
 499         u64 bytes_used;
 500         int ret = 0;
 501         int err;
 502
 503         while(!list_empty(list)) {
 504                 struct btrfs_root *root;
 505
 506                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 507                 list_del_init(&dirty->list);
 508
 509                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 510                 root = dirty->latest_root;
 511                 atomic_inc(&root->fs_info->throttles);
 512
 513                 mutex_lock(&root->fs_info->drop_mutex);
 514                 while(1) {
 515                         trans = btrfs_start_transaction(tree_root, 1);
 516                         ret = btrfs_drop_snapshot(trans, dirty->root);
 517                         if (ret != -EAGAIN) {
 518                                 break;
 519                         }
 520
 521                         err = btrfs_update_root(trans,
 522                                         tree_root,
 523                                         &dirty->root->root_key,
 524                                         &dirty->root->root_item);
 525                         if (err)
 526                                 ret = err;
 527                         nr = trans->blocks_used;
 528                         ret = btrfs_end_transaction(trans, tree_root);
 529                         BUG_ON(ret);
 530
 531                         mutex_unlock(&root->fs_info->drop_mutex);
 532                         btrfs_btree_balance_dirty(tree_root, nr);
 533                         cond_resched();
 534                         mutex_lock(&root->fs_info->drop_mutex);
 535                 }
 536                 BUG_ON(ret);
 537                 atomic_dec(&root->fs_info->throttles);
 538                 wake_up(&root->fs_info->transaction_throttle);
 539
 540                 mutex_lock(&root->fs_info->alloc_mutex);
 541                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 542                 bytes_used = btrfs_root_used(&root->root_item);
 543                 if (num_bytes) {
 544                         record_root_in_trans(root);
 545                         btrfs_set_root_used(&root->root_item,
 546                                             bytes_used - num_bytes);
 547                 }
 548                 mutex_unlock(&root->fs_info->alloc_mutex);
 549
 550                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 551                 if (ret) {
 552                         BUG();
 553                         break;
 554                 }
 555                 mutex_unlock(&root->fs_info->drop_mutex);
 556
 557                 nr = trans->blocks_used;
 558                 ret = btrfs_end_transaction(trans, tree_root);
 559                 BUG_ON(ret);
 560
 561                 free_extent_buffer(dirty->root->node);
 562                 kfree(dirty->root);
 563                 kfree(dirty);
 564
 565                 btrfs_btree_balance_dirty(tree_root, nr);
 566                 cond_resched();
 567         }
 568         return ret;
 569 }
 570
 571 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 572                                    struct btrfs_fs_info *fs_info,
 573                                    struct btrfs_pending_snapshot *pending)
 574 {
 575         struct btrfs_key key;
 576         struct btrfs_root_item *new_root_item;
 577         struct btrfs_root *tree_root = fs_info->tree_root;
 578         struct btrfs_root *root = pending->root;
 579         struct extent_buffer *tmp;
 580         struct extent_buffer *old;
 581         int ret;
 582         int namelen;
 583         u64 objectid;
 584
 585         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 586         if (!new_root_item) {
 587                 ret = -ENOMEM;
 588                 goto fail;
 589         }
 590         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 591         if (ret)
 592                 goto fail;
 593
 594         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 595
 596         key.objectid = objectid;
 597         key.offset = 1;
 598         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 599
 600         old = btrfs_lock_root_node(root);
 601         btrfs_cow_block(trans, root, old, NULL, 0, &old);
 602
 603         btrfs_copy_root(trans, root, old, &tmp, objectid);
 604         btrfs_tree_unlock(old);
 605         free_extent_buffer(old);
 606
 607         btrfs_set_root_bytenr(new_root_item, tmp->start);
 608         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 609         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 610                                 new_root_item);
 611         btrfs_tree_unlock(tmp);
 612         free_extent_buffer(tmp);
 613         if (ret)
 614                 goto fail;
 615
 616         /*
 617          * insert the directory item
 618          */
 619         key.offset = (u64)-1;
 620         namelen = strlen(pending->name);
 621         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 622                                     pending->name, namelen,
 623                                     root->fs_info->sb->s_root->d_inode->i_ino,
 624                                     &key, BTRFS_FT_DIR, 0);
 625
 626         if (ret)
 627                 goto fail;
 628
 629         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 630                              pending->name, strlen(pending->name), objectid,
 631                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 632
 633         /* Invalidate existing dcache entry for new snapshot. */
 634         btrfs_invalidate_dcache_root(root, pending->name, namelen);
 635
 636 fail:
 637         kfree(new_root_item);
 638         return ret;
 639 }
 640
 641 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 642                                              struct btrfs_fs_info *fs_info)
 643 {
 644         struct btrfs_pending_snapshot *pending;
 645         struct list_head *head = &trans->transaction->pending_snapshots;
 646         int ret;
 647
 648         while(!list_empty(head)) {
 649                 pending = list_entry(head->next,
 650                                      struct btrfs_pending_snapshot, list);
 651                 ret = create_pending_snapshot(trans, fs_info, pending);
 652                 BUG_ON(ret);
 653                 list_del(&pending->list);
 654                 kfree(pending->name);
 655                 kfree(pending);
 656         }
 657         return 0;
 658 }
 659
 660 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 661                              struct btrfs_root *root)
 662 {
 663         unsigned long joined = 0;
 664         unsigned long timeout = 1;
 665         struct btrfs_transaction *cur_trans;
 666         struct btrfs_transaction *prev_trans = NULL;
 667         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 668         struct list_head dirty_fs_roots;
 669         struct extent_io_tree *pinned_copy;
 670         DEFINE_WAIT(wait);
 671         int ret;
 672
 673         INIT_LIST_HEAD(&dirty_fs_roots);
 674
 675         mutex_lock(&root->fs_info->trans_mutex);
 676         if (trans->transaction->in_commit) {
 677                 cur_trans = trans->transaction;
 678                 trans->transaction->use_count++;
 679                 mutex_unlock(&root->fs_info->trans_mutex);
 680                 btrfs_end_transaction(trans, root);
 681
 682                 ret = wait_for_commit(root, cur_trans);
 683                 BUG_ON(ret);
 684
 685                 mutex_lock(&root->fs_info->trans_mutex);
 686                 put_transaction(cur_trans);
 687                 mutex_unlock(&root->fs_info->trans_mutex);
 688
 689                 return 0;
 690         }
 691
 692         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 693         if (!pinned_copy)
 694                 return -ENOMEM;
 695
 696         extent_io_tree_init(pinned_copy,
 697                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 698
 699         trans->transaction->in_commit = 1;
 700         trans->transaction->blocked = 1;
 701         cur_trans = trans->transaction;
 702         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 703                 prev_trans = list_entry(cur_trans->list.prev,
 704                                         struct btrfs_transaction, list);
 705                 if (!prev_trans->commit_done) {
 706                         prev_trans->use_count++;
 707                         mutex_unlock(&root->fs_info->trans_mutex);
 708
 709                         wait_for_commit(root, prev_trans);
 710
 711                         mutex_lock(&root->fs_info->trans_mutex);
 712                         put_transaction(prev_trans);
 713                 }
 714         }
 715
 716         do {
 717                 joined = cur_trans->num_joined;
 718                 WARN_ON(cur_trans != trans->transaction);
 719                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 720                                 TASK_UNINTERRUPTIBLE);
 721
 722                 if (cur_trans->num_writers > 1)
 723                         timeout = MAX_SCHEDULE_TIMEOUT;
 724                 else
 725                         timeout = 1;
 726
 727                 mutex_unlock(&root->fs_info->trans_mutex);
 728
 729                 schedule_timeout(timeout);
 730
 731                 mutex_lock(&root->fs_info->trans_mutex);
 732                 finish_wait(&cur_trans->writer_wait, &wait);
 733         } while (cur_trans->num_writers > 1 ||
 734                  (cur_trans->num_joined != joined));
 735
 736         ret = create_pending_snapshots(trans, root->fs_info);
 737         BUG_ON(ret);
 738
 739         WARN_ON(cur_trans != trans->transaction);
 740
 741         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 742                               &dirty_fs_roots);
 743         BUG_ON(ret);
 744
 745         ret = btrfs_commit_tree_roots(trans, root);
 746         BUG_ON(ret);
 747
 748         cur_trans = root->fs_info->running_transaction;
 749         spin_lock(&root->fs_info->new_trans_lock);
 750         root->fs_info->running_transaction = NULL;
 751         spin_unlock(&root->fs_info->new_trans_lock);
 752         btrfs_set_super_generation(&root->fs_info->super_copy,
 753                                    cur_trans->transid);
 754         btrfs_set_super_root(&root->fs_info->super_copy,
 755                              root->fs_info->tree_root->node->start);
 756         btrfs_set_super_root_level(&root->fs_info->super_copy,
 757                            btrfs_header_level(root->fs_info->tree_root->node));
 758
 759         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 760                                    chunk_root->node->start);
 761         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 762                                          btrfs_header_level(chunk_root->node));
 763         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 764                sizeof(root->fs_info->super_copy));
 765
 766         btrfs_copy_pinned(root, pinned_copy);
 767
 768         trans->transaction->blocked = 0;
 769         wake_up(&root->fs_info->transaction_throttle);
 770         wake_up(&root->fs_info->transaction_wait);
 771
 772         mutex_unlock(&root->fs_info->trans_mutex);
 773         ret = btrfs_write_and_wait_transaction(trans, root);
 774         BUG_ON(ret);
 775         write_ctree_super(trans, root);
 776
 777         btrfs_finish_extent_commit(trans, root, pinned_copy);
 778         mutex_lock(&root->fs_info->trans_mutex);
 779
 780         kfree(pinned_copy);
 781
 782         cur_trans->commit_done = 1;
 783         root->fs_info->last_trans_committed = cur_trans->transid;
 784         wake_up(&cur_trans->commit_wait);
 785         put_transaction(cur_trans);
 786         put_transaction(cur_trans);
 787
 788         if (root->fs_info->closing)
 789                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 790         else
 791                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 792
 793         mutex_unlock(&root->fs_info->trans_mutex);
 794         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 795
 796         if (root->fs_info->closing) {
 797                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 798         }
 799         return ret;
 800 }
 801
 802 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 803 {
 804         struct list_head dirty_roots;
 805         INIT_LIST_HEAD(&dirty_roots);
 806 again:
 807         mutex_lock(&root->fs_info->trans_mutex);
 808         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 809         mutex_unlock(&root->fs_info->trans_mutex);
 810
 811         if (!list_empty(&dirty_roots)) {
 812                 drop_dirty_roots(root, &dirty_roots);
 813                 goto again;
 814         }
 815         return 0;
 816 }
 817