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