Pileus Git - ~andy/linux/blob - 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                         dirty->latest_root = root;
 102                         INIT_LIST_HEAD(&dirty->list);
 103
 104                         root->commit_root = btrfs_root_node(root);
 105
 106                         memcpy(dirty->root, root, sizeof(*root));
 107                         spin_lock_init(&dirty->root->node_lock);
 108                         spin_lock_init(&dirty->root->list_lock);
 109                         mutex_init(&dirty->root->objectid_mutex);
 110                         INIT_LIST_HEAD(&dirty->root->dead_list);
 111                         dirty->root->node = root->commit_root;
 112                         dirty->root->commit_root = NULL;
 113
 114                         spin_lock(&root->list_lock);
 115                         list_add(&dirty->root->dead_list, &root->dead_list);
 116                         spin_unlock(&root->list_lock);
 117
 118                         root->dirty_root = dirty;
 119                 } else {
 120                         WARN_ON(1);
 121                 }
 122                 root->last_trans = running_trans_id;
 123         }
 124         return 0;
 125 }
 126
 127 static void wait_current_trans(struct btrfs_root *root)
 128 {
 129         struct btrfs_transaction *cur_trans;
 130
 131         cur_trans = root->fs_info->running_transaction;
 132         if (cur_trans && cur_trans->blocked) {
 133                 DEFINE_WAIT(wait);
 134                 cur_trans->use_count++;
 135                 while(1) {
 136                         prepare_to_wait(&root->fs_info->transaction_wait, &wait,
 137                                         TASK_UNINTERRUPTIBLE);
 138                         if (cur_trans->blocked) {
 139                                 mutex_unlock(&root->fs_info->trans_mutex);
 140                                 schedule();
 141                                 mutex_lock(&root->fs_info->trans_mutex);
 142                                 finish_wait(&root->fs_info->transaction_wait,
 143                                             &wait);
 144                         } else {
 145                                 finish_wait(&root->fs_info->transaction_wait,
 146                                             &wait);
 147                                 break;
 148                         }
 149                 }
 150                 put_transaction(cur_trans);
 151         }
 152 }
 153
 154 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 155                                              int num_blocks, int join)
 156 {
 157         struct btrfs_trans_handle *h =
 158                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 159         int ret;
 160
 161         mutex_lock(&root->fs_info->trans_mutex);
 162         if (!join)
 163                 wait_current_trans(root);
 164         ret = join_transaction(root);
 165         BUG_ON(ret);
 166
 167         record_root_in_trans(root);
 168         h->transid = root->fs_info->running_transaction->transid;
 169         h->transaction = root->fs_info->running_transaction;
 170         h->blocks_reserved = num_blocks;
 171         h->blocks_used = 0;
 172         h->block_group = NULL;
 173         h->alloc_exclude_nr = 0;
 174         h->alloc_exclude_start = 0;
 175         root->fs_info->running_transaction->use_count++;
 176         mutex_unlock(&root->fs_info->trans_mutex);
 177         return h;
 178 }
 179
 180 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 181                                                    int num_blocks)
 182 {
 183         return start_transaction(root, num_blocks, 0);
 184 }
 185 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 186                                                    int num_blocks)
 187 {
 188         return start_transaction(root, num_blocks, 1);
 189 }
 190
 191 static noinline int wait_for_commit(struct btrfs_root *root,
 192                                     struct btrfs_transaction *commit)
 193 {
 194         DEFINE_WAIT(wait);
 195         mutex_lock(&root->fs_info->trans_mutex);
 196         while(!commit->commit_done) {
 197                 prepare_to_wait(&commit->commit_wait, &wait,
 198                                 TASK_UNINTERRUPTIBLE);
 199                 if (commit->commit_done)
 200                         break;
 201                 mutex_unlock(&root->fs_info->trans_mutex);
 202                 schedule();
 203                 mutex_lock(&root->fs_info->trans_mutex);
 204         }
 205         mutex_unlock(&root->fs_info->trans_mutex);
 206         finish_wait(&commit->commit_wait, &wait);
 207         return 0;
 208 }
 209
 210 static void throttle_on_drops(struct btrfs_root *root)
 211 {
 212         struct btrfs_fs_info *info = root->fs_info;
 213
 214 harder:
 215         if (atomic_read(&info->throttles)) {
 216                 DEFINE_WAIT(wait);
 217                 int thr;
 218                 int harder_count = 0;
 219                 thr = atomic_read(&info->throttle_gen);
 220
 221                 do {
 222                         prepare_to_wait(&info->transaction_throttle,
 223                                         &wait, TASK_UNINTERRUPTIBLE);
 224                         if (!atomic_read(&info->throttles)) {
 225                                 finish_wait(&info->transaction_throttle, &wait);
 226                                 break;
 227                         }
 228                         schedule();
 229                         finish_wait(&info->transaction_throttle, &wait);
 230                 } while (thr == atomic_read(&info->throttle_gen));
 231
 232                 if (harder_count < 5 &&
 233                     info->total_ref_cache_size > 1 * 1024 * 1024) {
 234                         harder_count++;
 235                         goto harder;
 236                 }
 237
 238                 if (harder_count < 10 &&
 239                     info->total_ref_cache_size > 5 * 1024 * 1024) {
 240                         harder_count++;
 241                         goto harder;
 242                 }
 243         }
 244 }
 245
 246 void btrfs_throttle(struct btrfs_root *root)
 247 {
 248         mutex_lock(&root->fs_info->trans_mutex);
 249         wait_current_trans(root);
 250         mutex_unlock(&root->fs_info->trans_mutex);
 251
 252         throttle_on_drops(root);
 253 }
 254
 255 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 256                           struct btrfs_root *root, int throttle)
 257 {
 258         struct btrfs_transaction *cur_trans;
 259         struct btrfs_fs_info *info = root->fs_info;
 260
 261         mutex_lock(&info->trans_mutex);
 262         cur_trans = info->running_transaction;
 263         WARN_ON(cur_trans != trans->transaction);
 264         WARN_ON(cur_trans->num_writers < 1);
 265         cur_trans->num_writers--;
 266
 267         if (waitqueue_active(&cur_trans->writer_wait))
 268                 wake_up(&cur_trans->writer_wait);
 269         put_transaction(cur_trans);
 270         mutex_unlock(&info->trans_mutex);
 271         memset(trans, 0, sizeof(*trans));
 272         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 273
 274         if (throttle)
 275                 throttle_on_drops(root);
 276
 277         return 0;
 278 }
 279
 280 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 281                           struct btrfs_root *root)
 282 {
 283         return __btrfs_end_transaction(trans, root, 0);
 284 }
 285
 286 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 287                                    struct btrfs_root *root)
 288 {
 289         return __btrfs_end_transaction(trans, root, 1);
 290 }
 291
 292
 293 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 294                                      struct btrfs_root *root)
 295 {
 296         int ret;
 297         int err;
 298         int werr = 0;
 299         struct extent_io_tree *dirty_pages;
 300         struct page *page;
 301         struct inode *btree_inode = root->fs_info->btree_inode;
 302         u64 start;
 303         u64 end;
 304         unsigned long index;
 305
 306         if (!trans || !trans->transaction) {
 307                 return filemap_write_and_wait(btree_inode->i_mapping);
 308         }
 309         dirty_pages = &trans->transaction->dirty_pages;
 310         while(1) {
 311                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 312                                             EXTENT_DIRTY);
 313                 if (ret)
 314                         break;
 315                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 316                 while(start <= end) {
 317                         index = start >> PAGE_CACHE_SHIFT;
 318                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 319                         page = find_lock_page(btree_inode->i_mapping, index);
 320                         if (!page)
 321                                 continue;
 322                         if (PageWriteback(page)) {
 323                                 if (PageDirty(page))
 324                                         wait_on_page_writeback(page);
 325                                 else {
 326                                         unlock_page(page);
 327                                         page_cache_release(page);
 328                                         continue;
 329                                 }
 330                         }
 331                         err = write_one_page(page, 0);
 332                         if (err)
 333                                 werr = err;
 334                         page_cache_release(page);
 335                 }
 336         }
 337         err = filemap_fdatawait(btree_inode->i_mapping);
 338         if (err)
 339                 werr = err;
 340         return werr;
 341 }
 342
 343 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 344                                struct btrfs_root *root)
 345 {
 346         int ret;
 347         u64 old_root_bytenr;
 348         struct btrfs_root *tree_root = root->fs_info->tree_root;
 349
 350         btrfs_write_dirty_block_groups(trans, root);
 351         while(1) {
 352                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 353                 if (old_root_bytenr == root->node->start)
 354                         break;
 355                 btrfs_set_root_bytenr(&root->root_item,
 356                                        root->node->start);
 357                 btrfs_set_root_level(&root->root_item,
 358                                      btrfs_header_level(root->node));
 359                 ret = btrfs_update_root(trans, tree_root,
 360                                         &root->root_key,
 361                                         &root->root_item);
 362                 BUG_ON(ret);
 363                 btrfs_write_dirty_block_groups(trans, root);
 364         }
 365         return 0;
 366 }
 367
 368 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 369                             struct btrfs_root *root)
 370 {
 371         struct btrfs_fs_info *fs_info = root->fs_info;
 372         struct list_head *next;
 373
 374         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 375                 next = fs_info->dirty_cowonly_roots.next;
 376                 list_del_init(next);
 377                 root = list_entry(next, struct btrfs_root, dirty_list);
 378                 update_cowonly_root(trans, root);
 379         }
 380         return 0;
 381 }
 382
 383 int btrfs_add_dead_root(struct btrfs_root *root,
 384                         struct btrfs_root *latest,
 385                         struct list_head *dead_list)
 386 {
 387         struct btrfs_dirty_root *dirty;
 388
 389         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 390         if (!dirty)
 391                 return -ENOMEM;
 392         dirty->root = root;
 393         dirty->latest_root = latest;
 394         list_add(&dirty->list, dead_list);
 395         return 0;
 396 }
 397
 398 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 399                                     struct radix_tree_root *radix,
 400                                     struct list_head *list)
 401 {
 402         struct btrfs_dirty_root *dirty;
 403         struct btrfs_root *gang[8];
 404         struct btrfs_root *root;
 405         int i;
 406         int ret;
 407         int err = 0;
 408         u32 refs;
 409
 410         while(1) {
 411                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 412                                                  ARRAY_SIZE(gang),
 413                                                  BTRFS_ROOT_TRANS_TAG);
 414                 if (ret == 0)
 415                         break;
 416                 for (i = 0; i < ret; i++) {
 417                         root = gang[i];
 418                         radix_tree_tag_clear(radix,
 419                                      (unsigned long)root->root_key.objectid,
 420                                      BTRFS_ROOT_TRANS_TAG);
 421
 422                         BUG_ON(!root->ref_tree);
 423                         dirty = root->dirty_root;
 424
 425                         if (root->commit_root == root->node) {
 426                                 WARN_ON(root->node->start !=
 427                                         btrfs_root_bytenr(&root->root_item));
 428
 429                                 free_extent_buffer(root->commit_root);
 430                                 root->commit_root = NULL;
 431
 432                                 spin_lock(&root->list_lock);
 433                                 list_del_init(&dirty->root->dead_list);
 434                                 spin_unlock(&root->list_lock);
 435
 436                                 kfree(dirty->root);
 437                                 kfree(dirty);
 438
 439                                 /* make sure to update the root on disk
 440                                  * so we get any updates to the block used
 441                                  * counts
 442                                  */
 443                                 err = btrfs_update_root(trans,
 444                                                 root->fs_info->tree_root,
 445                                                 &root->root_key,
 446                                                 &root->root_item);
 447                                 continue;
 448                         }
 449
 450                         memset(&root->root_item.drop_progress, 0,
 451                                sizeof(struct btrfs_disk_key));
 452                         root->root_item.drop_level = 0;
 453                         root->commit_root = NULL;
 454                         root->root_key.offset = root->fs_info->generation;
 455                         btrfs_set_root_bytenr(&root->root_item,
 456                                               root->node->start);
 457                         btrfs_set_root_level(&root->root_item,
 458                                              btrfs_header_level(root->node));
 459                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 460                                                 &root->root_key,
 461                                                 &root->root_item);
 462                         if (err)
 463                                 break;
 464
 465                         refs = btrfs_root_refs(&dirty->root->root_item);
 466                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 467                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 468                                                 &dirty->root->root_key,
 469                                                 &dirty->root->root_item);
 470
 471                         BUG_ON(err);
 472                         if (refs == 1) {
 473                                 list_add(&dirty->list, list);
 474                         } else {
 475                                 WARN_ON(1);
 476                                 free_extent_buffer(dirty->root->node);
 477                                 kfree(dirty->root);
 478                                 kfree(dirty);
 479                         }
 480                 }
 481         }
 482         return err;
 483 }
 484
 485 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 486 {
 487         struct btrfs_fs_info *info = root->fs_info;
 488         int ret;
 489         struct btrfs_trans_handle *trans;
 490         unsigned long nr;
 491
 492         smp_mb();
 493         if (root->defrag_running)
 494                 return 0;
 495         trans = btrfs_start_transaction(root, 1);
 496         while (1) {
 497                 root->defrag_running = 1;
 498                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 499                 nr = trans->blocks_used;
 500                 btrfs_end_transaction(trans, root);
 501                 btrfs_btree_balance_dirty(info->tree_root, nr);
 502                 cond_resched();
 503
 504                 trans = btrfs_start_transaction(root, 1);
 505                 if (root->fs_info->closing || ret != -EAGAIN)
 506                         break;
 507         }
 508         root->defrag_running = 0;
 509         smp_mb();
 510         btrfs_end_transaction(trans, root);
 511         return 0;
 512 }
 513
 514 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 515                                      struct list_head *list)
 516 {
 517         struct btrfs_dirty_root *dirty;
 518         struct btrfs_trans_handle *trans;
 519         unsigned long nr;
 520         u64 num_bytes;
 521         u64 bytes_used;
 522         u64 max_useless;
 523         int ret = 0;
 524         int err;
 525
 526         while(!list_empty(list)) {
 527                 struct btrfs_root *root;
 528
 529                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 530                 list_del_init(&dirty->list);
 531
 532                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 533                 root = dirty->latest_root;
 534                 atomic_inc(&root->fs_info->throttles);
 535
 536                 mutex_lock(&root->fs_info->drop_mutex);
 537                 while(1) {
 538                         trans = btrfs_start_transaction(tree_root, 1);
 539                         ret = btrfs_drop_snapshot(trans, dirty->root);
 540                         if (ret != -EAGAIN) {
 541                                 break;
 542                         }
 543
 544                         err = btrfs_update_root(trans,
 545                                         tree_root,
 546                                         &dirty->root->root_key,
 547                                         &dirty->root->root_item);
 548                         if (err)
 549                                 ret = err;
 550                         nr = trans->blocks_used;
 551                         ret = btrfs_end_transaction(trans, tree_root);
 552                         BUG_ON(ret);
 553
 554                         mutex_unlock(&root->fs_info->drop_mutex);
 555                         btrfs_btree_balance_dirty(tree_root, nr);
 556                         cond_resched();
 557                         mutex_lock(&root->fs_info->drop_mutex);
 558                 }
 559                 BUG_ON(ret);
 560                 atomic_dec(&root->fs_info->throttles);
 561                 wake_up(&root->fs_info->transaction_throttle);
 562
 563                 mutex_lock(&root->fs_info->alloc_mutex);
 564                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 565                 bytes_used = btrfs_root_used(&root->root_item);
 566                 if (num_bytes) {
 567                         record_root_in_trans(root);
 568                         btrfs_set_root_used(&root->root_item,
 569                                             bytes_used - num_bytes);
 570                 }
 571                 mutex_unlock(&root->fs_info->alloc_mutex);
 572
 573                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 574                 if (ret) {
 575                         BUG();
 576                         break;
 577                 }
 578                 mutex_unlock(&root->fs_info->drop_mutex);
 579
 580                 spin_lock(&root->list_lock);
 581                 list_del_init(&dirty->root->dead_list);
 582                 if (!list_empty(&root->dead_list)) {
 583                         struct btrfs_root *oldest;
 584                         oldest = list_entry(root->dead_list.prev,
 585                                             struct btrfs_root, dead_list);
 586                         max_useless = oldest->root_key.offset - 1;
 587                 } else {
 588                         max_useless = root->root_key.offset - 1;
 589                 }
 590                 spin_unlock(&root->list_lock);
 591
 592                 nr = trans->blocks_used;
 593                 ret = btrfs_end_transaction(trans, tree_root);
 594                 BUG_ON(ret);
 595
 596                 ret = btrfs_remove_leaf_refs(root, max_useless);
 597                 BUG_ON(ret);
 598
 599                 free_extent_buffer(dirty->root->node);
 600                 kfree(dirty->root);
 601                 kfree(dirty);
 602
 603                 btrfs_btree_balance_dirty(tree_root, nr);
 604                 cond_resched();
 605         }
 606         return ret;
 607 }
 608
 609 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 610                                    struct btrfs_fs_info *fs_info,
 611                                    struct btrfs_pending_snapshot *pending)
 612 {
 613         struct btrfs_key key;
 614         struct btrfs_root_item *new_root_item;
 615         struct btrfs_root *tree_root = fs_info->tree_root;
 616         struct btrfs_root *root = pending->root;
 617         struct extent_buffer *tmp;
 618         struct extent_buffer *old;
 619         int ret;
 620         int namelen;
 621         u64 objectid;
 622
 623         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 624         if (!new_root_item) {
 625                 ret = -ENOMEM;
 626                 goto fail;
 627         }
 628         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 629         if (ret)
 630                 goto fail;
 631
 632         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 633
 634         key.objectid = objectid;
 635         key.offset = 1;
 636         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 637
 638         old = btrfs_lock_root_node(root);
 639         btrfs_cow_block(trans, root, old, NULL, 0, &old);
 640
 641         btrfs_copy_root(trans, root, old, &tmp, objectid);
 642         btrfs_tree_unlock(old);
 643         free_extent_buffer(old);
 644
 645         btrfs_set_root_bytenr(new_root_item, tmp->start);
 646         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 647         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 648                                 new_root_item);
 649         btrfs_tree_unlock(tmp);
 650         free_extent_buffer(tmp);
 651         if (ret)
 652                 goto fail;
 653
 654         /*
 655          * insert the directory item
 656          */
 657         key.offset = (u64)-1;
 658         namelen = strlen(pending->name);
 659         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 660                                     pending->name, namelen,
 661                                     root->fs_info->sb->s_root->d_inode->i_ino,
 662                                     &key, BTRFS_FT_DIR, 0);
 663
 664         if (ret)
 665                 goto fail;
 666
 667         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 668                              pending->name, strlen(pending->name), objectid,
 669                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 670
 671         /* Invalidate existing dcache entry for new snapshot. */
 672         btrfs_invalidate_dcache_root(root, pending->name, namelen);
 673
 674 fail:
 675         kfree(new_root_item);
 676         return ret;
 677 }
 678
 679 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 680                                              struct btrfs_fs_info *fs_info)
 681 {
 682         struct btrfs_pending_snapshot *pending;
 683         struct list_head *head = &trans->transaction->pending_snapshots;
 684         int ret;
 685
 686         while(!list_empty(head)) {
 687                 pending = list_entry(head->next,
 688                                      struct btrfs_pending_snapshot, list);
 689                 ret = create_pending_snapshot(trans, fs_info, pending);
 690                 BUG_ON(ret);
 691                 list_del(&pending->list);
 692                 kfree(pending->name);
 693                 kfree(pending);
 694         }
 695         return 0;
 696 }
 697
 698 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 699                              struct btrfs_root *root)
 700 {
 701         unsigned long joined = 0;
 702         unsigned long timeout = 1;
 703         struct btrfs_transaction *cur_trans;
 704         struct btrfs_transaction *prev_trans = NULL;
 705         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 706         struct list_head dirty_fs_roots;
 707         struct extent_io_tree *pinned_copy;
 708         DEFINE_WAIT(wait);
 709         int ret;
 710
 711         INIT_LIST_HEAD(&dirty_fs_roots);
 712
 713         mutex_lock(&root->fs_info->trans_mutex);
 714         if (trans->transaction->in_commit) {
 715                 cur_trans = trans->transaction;
 716                 trans->transaction->use_count++;
 717                 mutex_unlock(&root->fs_info->trans_mutex);
 718                 btrfs_end_transaction(trans, root);
 719
 720                 ret = wait_for_commit(root, cur_trans);
 721                 BUG_ON(ret);
 722
 723                 mutex_lock(&root->fs_info->trans_mutex);
 724                 put_transaction(cur_trans);
 725                 mutex_unlock(&root->fs_info->trans_mutex);
 726
 727                 return 0;
 728         }
 729
 730         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 731         if (!pinned_copy)
 732                 return -ENOMEM;
 733
 734         extent_io_tree_init(pinned_copy,
 735                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 736
 737         trans->transaction->in_commit = 1;
 738         trans->transaction->blocked = 1;
 739         cur_trans = trans->transaction;
 740         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 741                 prev_trans = list_entry(cur_trans->list.prev,
 742                                         struct btrfs_transaction, list);
 743                 if (!prev_trans->commit_done) {
 744                         prev_trans->use_count++;
 745                         mutex_unlock(&root->fs_info->trans_mutex);
 746
 747                         wait_for_commit(root, prev_trans);
 748
 749                         mutex_lock(&root->fs_info->trans_mutex);
 750                         put_transaction(prev_trans);
 751                 }
 752         }
 753
 754         do {
 755                 joined = cur_trans->num_joined;
 756                 WARN_ON(cur_trans != trans->transaction);
 757                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 758                                 TASK_UNINTERRUPTIBLE);
 759
 760                 if (cur_trans->num_writers > 1)
 761                         timeout = MAX_SCHEDULE_TIMEOUT;
 762                 else
 763                         timeout = 1;
 764
 765                 mutex_unlock(&root->fs_info->trans_mutex);
 766
 767                 schedule_timeout(timeout);
 768
 769                 mutex_lock(&root->fs_info->trans_mutex);
 770                 finish_wait(&cur_trans->writer_wait, &wait);
 771         } while (cur_trans->num_writers > 1 ||
 772                  (cur_trans->num_joined != joined));
 773
 774         ret = create_pending_snapshots(trans, root->fs_info);
 775         BUG_ON(ret);
 776
 777         WARN_ON(cur_trans != trans->transaction);
 778
 779         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 780                               &dirty_fs_roots);
 781         BUG_ON(ret);
 782
 783         ret = btrfs_commit_tree_roots(trans, root);
 784         BUG_ON(ret);
 785
 786         cur_trans = root->fs_info->running_transaction;
 787         spin_lock(&root->fs_info->new_trans_lock);
 788         root->fs_info->running_transaction = NULL;
 789         spin_unlock(&root->fs_info->new_trans_lock);
 790         btrfs_set_super_generation(&root->fs_info->super_copy,
 791                                    cur_trans->transid);
 792         btrfs_set_super_root(&root->fs_info->super_copy,
 793                              root->fs_info->tree_root->node->start);
 794         btrfs_set_super_root_level(&root->fs_info->super_copy,
 795                            btrfs_header_level(root->fs_info->tree_root->node));
 796
 797         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 798                                    chunk_root->node->start);
 799         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 800                                          btrfs_header_level(chunk_root->node));
 801         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 802                sizeof(root->fs_info->super_copy));
 803
 804         btrfs_copy_pinned(root, pinned_copy);
 805
 806         trans->transaction->blocked = 0;
 807         wake_up(&root->fs_info->transaction_throttle);
 808         wake_up(&root->fs_info->transaction_wait);
 809
 810         mutex_unlock(&root->fs_info->trans_mutex);
 811         ret = btrfs_write_and_wait_transaction(trans, root);
 812         BUG_ON(ret);
 813         write_ctree_super(trans, root);
 814
 815         btrfs_finish_extent_commit(trans, root, pinned_copy);
 816         mutex_lock(&root->fs_info->trans_mutex);
 817
 818         kfree(pinned_copy);
 819
 820         cur_trans->commit_done = 1;
 821         root->fs_info->last_trans_committed = cur_trans->transid;
 822         wake_up(&cur_trans->commit_wait);
 823         put_transaction(cur_trans);
 824         put_transaction(cur_trans);
 825
 826         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 827         if (root->fs_info->closing)
 828                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 829
 830         mutex_unlock(&root->fs_info->trans_mutex);
 831         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 832
 833         if (root->fs_info->closing) {
 834                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 835         }
 836         return ret;
 837 }
 838
 839 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 840 {
 841         struct list_head dirty_roots;
 842         INIT_LIST_HEAD(&dirty_roots);
 843 again:
 844         mutex_lock(&root->fs_info->trans_mutex);
 845         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 846         mutex_unlock(&root->fs_info->trans_mutex);
 847
 848         if (!list_empty(&dirty_roots)) {
 849                 drop_dirty_roots(root, &dirty_roots);
 850                 goto again;
 851         }
 852         return 0;
 853 }