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 wait)
 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 ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)
 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, 1);
 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, 0);
 189 }
 190
 191 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
 192                                                          int num_blocks)
 193 {
 194         return start_transaction(r, num_blocks, 2);
 195 }
 196
 197
 198 static noinline int wait_for_commit(struct btrfs_root *root,
 199                                     struct btrfs_transaction *commit)
 200 {
 201         DEFINE_WAIT(wait);
 202         mutex_lock(&root->fs_info->trans_mutex);
 203         while(!commit->commit_done) {
 204                 prepare_to_wait(&commit->commit_wait, &wait,
 205                                 TASK_UNINTERRUPTIBLE);
 206                 if (commit->commit_done)
 207                         break;
 208                 mutex_unlock(&root->fs_info->trans_mutex);
 209                 schedule();
 210                 mutex_lock(&root->fs_info->trans_mutex);
 211         }
 212         mutex_unlock(&root->fs_info->trans_mutex);
 213         finish_wait(&commit->commit_wait, &wait);
 214         return 0;
 215 }
 216
 217 static void throttle_on_drops(struct btrfs_root *root)
 218 {
 219         struct btrfs_fs_info *info = root->fs_info;
 220         int harder_count = 0;
 221
 222 harder:
 223         if (atomic_read(&info->throttles)) {
 224                 DEFINE_WAIT(wait);
 225                 int thr;
 226                 thr = atomic_read(&info->throttle_gen);
 227
 228                 do {
 229                         prepare_to_wait(&info->transaction_throttle,
 230                                         &wait, TASK_UNINTERRUPTIBLE);
 231                         if (!atomic_read(&info->throttles)) {
 232                                 finish_wait(&info->transaction_throttle, &wait);
 233                                 break;
 234                         }
 235                         schedule();
 236                         finish_wait(&info->transaction_throttle, &wait);
 237                 } while (thr == atomic_read(&info->throttle_gen));
 238                 harder_count++;
 239
 240                 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
 241                     harder_count < 2)
 242                         goto harder;
 243
 244                 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
 245                     harder_count < 10)
 246                         goto harder;
 247
 248                 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
 249                     harder_count < 20)
 250                         goto harder;
 251         }
 252 }
 253
 254 void btrfs_throttle(struct btrfs_root *root)
 255 {
 256         mutex_lock(&root->fs_info->trans_mutex);
 257         if (!root->fs_info->open_ioctl_trans)
 258                 wait_current_trans(root);
 259         mutex_unlock(&root->fs_info->trans_mutex);
 260
 261         throttle_on_drops(root);
 262 }
 263
 264 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 265                           struct btrfs_root *root, int throttle)
 266 {
 267         struct btrfs_transaction *cur_trans;
 268         struct btrfs_fs_info *info = root->fs_info;
 269
 270         mutex_lock(&info->trans_mutex);
 271         cur_trans = info->running_transaction;
 272         WARN_ON(cur_trans != trans->transaction);
 273         WARN_ON(cur_trans->num_writers < 1);
 274         cur_trans->num_writers--;
 275
 276         if (waitqueue_active(&cur_trans->writer_wait))
 277                 wake_up(&cur_trans->writer_wait);
 278         put_transaction(cur_trans);
 279         mutex_unlock(&info->trans_mutex);
 280         memset(trans, 0, sizeof(*trans));
 281         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 282
 283         if (throttle)
 284                 throttle_on_drops(root);
 285
 286         return 0;
 287 }
 288
 289 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 290                           struct btrfs_root *root)
 291 {
 292         return __btrfs_end_transaction(trans, root, 0);
 293 }
 294
 295 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 296                                    struct btrfs_root *root)
 297 {
 298         return __btrfs_end_transaction(trans, root, 1);
 299 }
 300
 301
 302 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 303                                      struct btrfs_root *root)
 304 {
 305         int ret;
 306         int err = 0;
 307         int werr = 0;
 308         struct extent_io_tree *dirty_pages;
 309         struct page *page;
 310         struct inode *btree_inode = root->fs_info->btree_inode;
 311         u64 start = 0;
 312         u64 end;
 313         unsigned long index;
 314
 315         if (!trans || !trans->transaction) {
 316                 return filemap_write_and_wait(btree_inode->i_mapping);
 317         }
 318         dirty_pages = &trans->transaction->dirty_pages;
 319         while(1) {
 320                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
 321                                             EXTENT_DIRTY);
 322                 if (ret)
 323                         break;
 324                 while(start <= end) {
 325                         if (btrfs_congested_async(root->fs_info, 0))
 326                                 congestion_wait(WRITE, HZ/10);
 327                         cond_resched();
 328
 329                         index = start >> PAGE_CACHE_SHIFT;
 330                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 331                         page = find_lock_page(btree_inode->i_mapping, index);
 332                         if (!page)
 333                                 continue;
 334                         if (PageWriteback(page)) {
 335                                 if (PageDirty(page))
 336                                         wait_on_page_writeback(page);
 337                                 else {
 338                                         unlock_page(page);
 339                                         page_cache_release(page);
 340                                         continue;
 341                                 }
 342                         }
 343                         err = write_one_page(page, 0);
 344                         if (err)
 345                                 werr = err;
 346                         page_cache_release(page);
 347                 }
 348         }
 349         while(1) {
 350                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 351                                             EXTENT_DIRTY);
 352                 if (ret)
 353                         break;
 354
 355                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 356                 while(start <= end) {
 357                         index = start >> PAGE_CACHE_SHIFT;
 358                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 359                         page = find_get_page(btree_inode->i_mapping, index);
 360                         if (!page)
 361                                 continue;
 362                         if (PageDirty(page)) {
 363                                 lock_page(page);
 364                                 err = write_one_page(page, 0);
 365                                 if (err)
 366                                         werr = err;
 367                         }
 368                         wait_on_page_writeback(page);
 369                         page_cache_release(page);
 370                         cond_resched();
 371                 }
 372         }
 373         if (err)
 374                 werr = err;
 375         return werr;
 376 }
 377
 378 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 379                                struct btrfs_root *root)
 380 {
 381         int ret;
 382         u64 old_root_bytenr;
 383         struct btrfs_root *tree_root = root->fs_info->tree_root;
 384
 385         btrfs_write_dirty_block_groups(trans, root);
 386         while(1) {
 387                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 388                 if (old_root_bytenr == root->node->start)
 389                         break;
 390                 btrfs_set_root_bytenr(&root->root_item,
 391                                        root->node->start);
 392                 btrfs_set_root_level(&root->root_item,
 393                                      btrfs_header_level(root->node));
 394                 ret = btrfs_update_root(trans, tree_root,
 395                                         &root->root_key,
 396                                         &root->root_item);
 397                 BUG_ON(ret);
 398                 btrfs_write_dirty_block_groups(trans, root);
 399         }
 400         return 0;
 401 }
 402
 403 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 404                             struct btrfs_root *root)
 405 {
 406         struct btrfs_fs_info *fs_info = root->fs_info;
 407         struct list_head *next;
 408
 409         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 410                 next = fs_info->dirty_cowonly_roots.next;
 411                 list_del_init(next);
 412                 root = list_entry(next, struct btrfs_root, dirty_list);
 413                 update_cowonly_root(trans, root);
 414         }
 415         return 0;
 416 }
 417
 418 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
 419 {
 420         struct btrfs_dirty_root *dirty;
 421
 422         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 423         if (!dirty)
 424                 return -ENOMEM;
 425         dirty->root = root;
 426         dirty->latest_root = latest;
 427
 428         mutex_lock(&root->fs_info->trans_mutex);
 429         list_add(&dirty->list, &latest->fs_info->dead_roots);
 430         mutex_unlock(&root->fs_info->trans_mutex);
 431         return 0;
 432 }
 433
 434 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 435                                     struct radix_tree_root *radix,
 436                                     struct list_head *list)
 437 {
 438         struct btrfs_dirty_root *dirty;
 439         struct btrfs_root *gang[8];
 440         struct btrfs_root *root;
 441         int i;
 442         int ret;
 443         int err = 0;
 444         u32 refs;
 445
 446         while(1) {
 447                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 448                                                  ARRAY_SIZE(gang),
 449                                                  BTRFS_ROOT_TRANS_TAG);
 450                 if (ret == 0)
 451                         break;
 452                 for (i = 0; i < ret; i++) {
 453                         root = gang[i];
 454                         radix_tree_tag_clear(radix,
 455                                      (unsigned long)root->root_key.objectid,
 456                                      BTRFS_ROOT_TRANS_TAG);
 457
 458                         BUG_ON(!root->ref_tree);
 459                         dirty = root->dirty_root;
 460
 461                         if (root->commit_root == root->node) {
 462                                 WARN_ON(root->node->start !=
 463                                         btrfs_root_bytenr(&root->root_item));
 464
 465                                 free_extent_buffer(root->commit_root);
 466                                 root->commit_root = NULL;
 467                                 root->dirty_root = NULL;
 468
 469                                 spin_lock(&root->list_lock);
 470                                 list_del_init(&dirty->root->dead_list);
 471                                 spin_unlock(&root->list_lock);
 472
 473                                 kfree(dirty->root);
 474                                 kfree(dirty);
 475
 476                                 /* make sure to update the root on disk
 477                                  * so we get any updates to the block used
 478                                  * counts
 479                                  */
 480                                 err = btrfs_update_root(trans,
 481                                                 root->fs_info->tree_root,
 482                                                 &root->root_key,
 483                                                 &root->root_item);
 484                                 continue;
 485                         }
 486
 487                         memset(&root->root_item.drop_progress, 0,
 488                                sizeof(struct btrfs_disk_key));
 489                         root->root_item.drop_level = 0;
 490                         root->commit_root = NULL;
 491                         root->dirty_root = NULL;
 492                         root->root_key.offset = root->fs_info->generation;
 493                         btrfs_set_root_bytenr(&root->root_item,
 494                                               root->node->start);
 495                         btrfs_set_root_level(&root->root_item,
 496                                              btrfs_header_level(root->node));
 497                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 498                                                 &root->root_key,
 499                                                 &root->root_item);
 500                         if (err)
 501                                 break;
 502
 503                         refs = btrfs_root_refs(&dirty->root->root_item);
 504                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 505                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 506                                                 &dirty->root->root_key,
 507                                                 &dirty->root->root_item);
 508
 509                         BUG_ON(err);
 510                         if (refs == 1) {
 511                                 list_add(&dirty->list, list);
 512                         } else {
 513                                 WARN_ON(1);
 514                                 free_extent_buffer(dirty->root->node);
 515                                 kfree(dirty->root);
 516                                 kfree(dirty);
 517                         }
 518                 }
 519         }
 520         return err;
 521 }
 522
 523 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 524 {
 525         struct btrfs_fs_info *info = root->fs_info;
 526         int ret;
 527         struct btrfs_trans_handle *trans;
 528         unsigned long nr;
 529
 530         smp_mb();
 531         if (root->defrag_running)
 532                 return 0;
 533         trans = btrfs_start_transaction(root, 1);
 534         while (1) {
 535                 root->defrag_running = 1;
 536                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 537                 nr = trans->blocks_used;
 538                 btrfs_end_transaction(trans, root);
 539                 btrfs_btree_balance_dirty(info->tree_root, nr);
 540                 cond_resched();
 541
 542                 trans = btrfs_start_transaction(root, 1);
 543                 if (root->fs_info->closing || ret != -EAGAIN)
 544                         break;
 545         }
 546         root->defrag_running = 0;
 547         smp_mb();
 548         btrfs_end_transaction(trans, root);
 549         return 0;
 550 }
 551
 552 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 553                                      struct list_head *list)
 554 {
 555         struct btrfs_dirty_root *dirty;
 556         struct btrfs_trans_handle *trans;
 557         unsigned long nr;
 558         u64 num_bytes;
 559         u64 bytes_used;
 560         u64 max_useless;
 561         int ret = 0;
 562         int err;
 563
 564         while(!list_empty(list)) {
 565                 struct btrfs_root *root;
 566
 567                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 568                 list_del_init(&dirty->list);
 569
 570                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 571                 root = dirty->latest_root;
 572                 atomic_inc(&root->fs_info->throttles);
 573
 574                 mutex_lock(&root->fs_info->drop_mutex);
 575                 while(1) {
 576                         trans = btrfs_start_transaction(tree_root, 1);
 577                         ret = btrfs_drop_snapshot(trans, dirty->root);
 578                         if (ret != -EAGAIN) {
 579                                 break;
 580                         }
 581
 582                         err = btrfs_update_root(trans,
 583                                         tree_root,
 584                                         &dirty->root->root_key,
 585                                         &dirty->root->root_item);
 586                         if (err)
 587                                 ret = err;
 588                         nr = trans->blocks_used;
 589                         ret = btrfs_end_transaction(trans, tree_root);
 590                         BUG_ON(ret);
 591
 592                         mutex_unlock(&root->fs_info->drop_mutex);
 593                         btrfs_btree_balance_dirty(tree_root, nr);
 594                         cond_resched();
 595                         mutex_lock(&root->fs_info->drop_mutex);
 596                 }
 597                 BUG_ON(ret);
 598                 atomic_dec(&root->fs_info->throttles);
 599                 wake_up(&root->fs_info->transaction_throttle);
 600
 601                 mutex_lock(&root->fs_info->alloc_mutex);
 602                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 603                 bytes_used = btrfs_root_used(&root->root_item);
 604                 if (num_bytes) {
 605                         record_root_in_trans(root);
 606                         btrfs_set_root_used(&root->root_item,
 607                                             bytes_used - num_bytes);
 608                 }
 609                 mutex_unlock(&root->fs_info->alloc_mutex);
 610
 611                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 612                 if (ret) {
 613                         BUG();
 614                         break;
 615                 }
 616                 mutex_unlock(&root->fs_info->drop_mutex);
 617
 618                 spin_lock(&root->list_lock);
 619                 list_del_init(&dirty->root->dead_list);
 620                 if (!list_empty(&root->dead_list)) {
 621                         struct btrfs_root *oldest;
 622                         oldest = list_entry(root->dead_list.prev,
 623                                             struct btrfs_root, dead_list);
 624                         max_useless = oldest->root_key.offset - 1;
 625                 } else {
 626                         max_useless = root->root_key.offset - 1;
 627                 }
 628                 spin_unlock(&root->list_lock);
 629
 630                 nr = trans->blocks_used;
 631                 ret = btrfs_end_transaction(trans, tree_root);
 632                 BUG_ON(ret);
 633
 634                 ret = btrfs_remove_leaf_refs(root, max_useless);
 635                 BUG_ON(ret);
 636
 637                 free_extent_buffer(dirty->root->node);
 638                 kfree(dirty->root);
 639                 kfree(dirty);
 640
 641                 btrfs_btree_balance_dirty(tree_root, nr);
 642                 cond_resched();
 643         }
 644         return ret;
 645 }
 646
 647 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 648                                    struct btrfs_fs_info *fs_info,
 649                                    struct btrfs_pending_snapshot *pending)
 650 {
 651         struct btrfs_key key;
 652         struct btrfs_root_item *new_root_item;
 653         struct btrfs_root *tree_root = fs_info->tree_root;
 654         struct btrfs_root *root = pending->root;
 655         struct extent_buffer *tmp;
 656         struct extent_buffer *old;
 657         int ret;
 658         int namelen;
 659         u64 objectid;
 660
 661         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 662         if (!new_root_item) {
 663                 ret = -ENOMEM;
 664                 goto fail;
 665         }
 666         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 667         if (ret)
 668                 goto fail;
 669
 670         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 671
 672         key.objectid = objectid;
 673         key.offset = 1;
 674         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 675
 676         old = btrfs_lock_root_node(root);
 677         btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
 678
 679         btrfs_copy_root(trans, root, old, &tmp, objectid);
 680         btrfs_tree_unlock(old);
 681         free_extent_buffer(old);
 682
 683         btrfs_set_root_bytenr(new_root_item, tmp->start);
 684         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 685         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 686                                 new_root_item);
 687         btrfs_tree_unlock(tmp);
 688         free_extent_buffer(tmp);
 689         if (ret)
 690                 goto fail;
 691
 692         /*
 693          * insert the directory item
 694          */
 695         key.offset = (u64)-1;
 696         namelen = strlen(pending->name);
 697         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 698                                     pending->name, namelen,
 699                                     root->fs_info->sb->s_root->d_inode->i_ino,
 700                                     &key, BTRFS_FT_DIR, 0);
 701
 702         if (ret)
 703                 goto fail;
 704
 705         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 706                              pending->name, strlen(pending->name), objectid,
 707                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 708
 709         /* Invalidate existing dcache entry for new snapshot. */
 710         btrfs_invalidate_dcache_root(root, pending->name, namelen);
 711
 712 fail:
 713         kfree(new_root_item);
 714         return ret;
 715 }
 716
 717 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 718                                              struct btrfs_fs_info *fs_info)
 719 {
 720         struct btrfs_pending_snapshot *pending;
 721         struct list_head *head = &trans->transaction->pending_snapshots;
 722         int ret;
 723
 724         while(!list_empty(head)) {
 725                 pending = list_entry(head->next,
 726                                      struct btrfs_pending_snapshot, list);
 727                 ret = create_pending_snapshot(trans, fs_info, pending);
 728                 BUG_ON(ret);
 729                 list_del(&pending->list);
 730                 kfree(pending->name);
 731                 kfree(pending);
 732         }
 733         return 0;
 734 }
 735
 736 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 737                              struct btrfs_root *root)
 738 {
 739         unsigned long joined = 0;
 740         unsigned long timeout = 1;
 741         struct btrfs_transaction *cur_trans;
 742         struct btrfs_transaction *prev_trans = NULL;
 743         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 744         struct list_head dirty_fs_roots;
 745         struct extent_io_tree *pinned_copy;
 746         DEFINE_WAIT(wait);
 747         int ret;
 748
 749         INIT_LIST_HEAD(&dirty_fs_roots);
 750
 751         mutex_lock(&root->fs_info->trans_mutex);
 752         if (trans->transaction->in_commit) {
 753                 cur_trans = trans->transaction;
 754                 trans->transaction->use_count++;
 755                 mutex_unlock(&root->fs_info->trans_mutex);
 756                 btrfs_end_transaction(trans, root);
 757
 758                 ret = wait_for_commit(root, cur_trans);
 759                 BUG_ON(ret);
 760
 761                 mutex_lock(&root->fs_info->trans_mutex);
 762                 put_transaction(cur_trans);
 763                 mutex_unlock(&root->fs_info->trans_mutex);
 764
 765                 return 0;
 766         }
 767
 768         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 769         if (!pinned_copy)
 770                 return -ENOMEM;
 771
 772         extent_io_tree_init(pinned_copy,
 773                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 774
 775         trans->transaction->in_commit = 1;
 776         trans->transaction->blocked = 1;
 777         cur_trans = trans->transaction;
 778         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 779                 prev_trans = list_entry(cur_trans->list.prev,
 780                                         struct btrfs_transaction, list);
 781                 if (!prev_trans->commit_done) {
 782                         prev_trans->use_count++;
 783                         mutex_unlock(&root->fs_info->trans_mutex);
 784
 785                         wait_for_commit(root, prev_trans);
 786
 787                         mutex_lock(&root->fs_info->trans_mutex);
 788                         put_transaction(prev_trans);
 789                 }
 790         }
 791
 792         do {
 793                 int snap_pending = 0;
 794                 joined = cur_trans->num_joined;
 795                 if (!list_empty(&trans->transaction->pending_snapshots))
 796                         snap_pending = 1;
 797
 798                 WARN_ON(cur_trans != trans->transaction);
 799                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 800                                 TASK_UNINTERRUPTIBLE);
 801
 802                 if (cur_trans->num_writers > 1)
 803                         timeout = MAX_SCHEDULE_TIMEOUT;
 804                 else
 805                         timeout = 1;
 806
 807                 mutex_unlock(&root->fs_info->trans_mutex);
 808
 809                 if (snap_pending) {
 810                         ret = btrfs_wait_ordered_extents(root, 1);
 811                         BUG_ON(ret);
 812                 }
 813
 814                 schedule_timeout(timeout);
 815
 816                 mutex_lock(&root->fs_info->trans_mutex);
 817                 finish_wait(&cur_trans->writer_wait, &wait);
 818         } while (cur_trans->num_writers > 1 ||
 819                  (cur_trans->num_joined != joined));
 820
 821         ret = create_pending_snapshots(trans, root->fs_info);
 822         BUG_ON(ret);
 823
 824         WARN_ON(cur_trans != trans->transaction);
 825
 826         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 827                               &dirty_fs_roots);
 828         BUG_ON(ret);
 829
 830         ret = btrfs_commit_tree_roots(trans, root);
 831         BUG_ON(ret);
 832
 833         cur_trans = root->fs_info->running_transaction;
 834         spin_lock(&root->fs_info->new_trans_lock);
 835         root->fs_info->running_transaction = NULL;
 836         spin_unlock(&root->fs_info->new_trans_lock);
 837         btrfs_set_super_generation(&root->fs_info->super_copy,
 838                                    cur_trans->transid);
 839         btrfs_set_super_root(&root->fs_info->super_copy,
 840                              root->fs_info->tree_root->node->start);
 841         btrfs_set_super_root_level(&root->fs_info->super_copy,
 842                            btrfs_header_level(root->fs_info->tree_root->node));
 843
 844         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 845                                    chunk_root->node->start);
 846         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 847                                          btrfs_header_level(chunk_root->node));
 848         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 849                sizeof(root->fs_info->super_copy));
 850
 851         btrfs_copy_pinned(root, pinned_copy);
 852
 853         trans->transaction->blocked = 0;
 854         wake_up(&root->fs_info->transaction_throttle);
 855         wake_up(&root->fs_info->transaction_wait);
 856
 857         mutex_unlock(&root->fs_info->trans_mutex);
 858         ret = btrfs_write_and_wait_transaction(trans, root);
 859         BUG_ON(ret);
 860         write_ctree_super(trans, root);
 861
 862         btrfs_finish_extent_commit(trans, root, pinned_copy);
 863         mutex_lock(&root->fs_info->trans_mutex);
 864
 865         kfree(pinned_copy);
 866
 867         cur_trans->commit_done = 1;
 868         root->fs_info->last_trans_committed = cur_trans->transid;
 869         wake_up(&cur_trans->commit_wait);
 870         put_transaction(cur_trans);
 871         put_transaction(cur_trans);
 872
 873         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 874         if (root->fs_info->closing)
 875                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 876
 877         mutex_unlock(&root->fs_info->trans_mutex);
 878         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 879
 880         if (root->fs_info->closing) {
 881                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 882         }
 883         return ret;
 884 }
 885
 886 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 887 {
 888         struct list_head dirty_roots;
 889         INIT_LIST_HEAD(&dirty_roots);
 890 again:
 891         mutex_lock(&root->fs_info->trans_mutex);
 892         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 893         mutex_unlock(&root->fs_info->trans_mutex);
 894
 895         if (!list_empty(&dirty_roots)) {
 896                 drop_dirty_roots(root, &dirty_roots);
 897                 goto again;
 898         }
 899         return 0;
 900 }