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 #include "tree-log.h"
  29
  30 static int total_trans = 0;
  31 extern struct kmem_cache *btrfs_trans_handle_cachep;
  32 extern struct kmem_cache *btrfs_transaction_cachep;
  33
  34 #define BTRFS_ROOT_TRANS_TAG 0
  35
  36 static noinline void put_transaction(struct btrfs_transaction *transaction)
  37 {
  38         WARN_ON(transaction->use_count == 0);
  39         transaction->use_count--;
  40         if (transaction->use_count == 0) {
  41                 WARN_ON(total_trans == 0);
  42                 total_trans--;
  43                 list_del_init(&transaction->list);
  44                 memset(transaction, 0, sizeof(*transaction));
  45                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  46         }
  47 }
  48
  49 static noinline int join_transaction(struct btrfs_root *root)
  50 {
  51         struct btrfs_transaction *cur_trans;
  52         cur_trans = root->fs_info->running_transaction;
  53         if (!cur_trans) {
  54                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  55                                              GFP_NOFS);
  56                 total_trans++;
  57                 BUG_ON(!cur_trans);
  58                 root->fs_info->generation++;
  59                 root->fs_info->last_alloc = 0;
  60                 root->fs_info->last_data_alloc = 0;
  61                 root->fs_info->last_log_alloc = 0;
  62                 cur_trans->num_writers = 1;
  63                 cur_trans->num_joined = 0;
  64                 cur_trans->transid = root->fs_info->generation;
  65                 init_waitqueue_head(&cur_trans->writer_wait);
  66                 init_waitqueue_head(&cur_trans->commit_wait);
  67                 cur_trans->in_commit = 0;
  68                 cur_trans->blocked = 0;
  69                 cur_trans->use_count = 1;
  70                 cur_trans->commit_done = 0;
  71                 cur_trans->start_time = get_seconds();
  72                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  73                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  74                 extent_io_tree_init(&cur_trans->dirty_pages,
  75                                      root->fs_info->btree_inode->i_mapping,
  76                                      GFP_NOFS);
  77                 spin_lock(&root->fs_info->new_trans_lock);
  78                 root->fs_info->running_transaction = cur_trans;
  79                 spin_unlock(&root->fs_info->new_trans_lock);
  80         } else {
  81                 cur_trans->num_writers++;
  82                 cur_trans->num_joined++;
  83         }
  84
  85         return 0;
  86 }
  87
  88 noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
  89 {
  90         struct btrfs_dirty_root *dirty;
  91         u64 running_trans_id = root->fs_info->running_transaction->transid;
  92         if (root->ref_cows && root->last_trans < running_trans_id) {
  93                 WARN_ON(root == root->fs_info->extent_root);
  94                 if (root->root_item.refs != 0) {
  95                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  96                                    (unsigned long)root->root_key.objectid,
  97                                    BTRFS_ROOT_TRANS_TAG);
  98
  99                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 100                         BUG_ON(!dirty);
 101                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 102                         BUG_ON(!dirty->root);
 103                         dirty->latest_root = root;
 104                         INIT_LIST_HEAD(&dirty->list);
 105
 106                         root->commit_root = btrfs_root_node(root);
 107
 108                         memcpy(dirty->root, root, sizeof(*root));
 109                         spin_lock_init(&dirty->root->node_lock);
 110                         spin_lock_init(&dirty->root->list_lock);
 111                         mutex_init(&dirty->root->objectid_mutex);
 112                         INIT_LIST_HEAD(&dirty->root->dead_list);
 113                         dirty->root->node = root->commit_root;
 114                         dirty->root->commit_root = NULL;
 115
 116                         spin_lock(&root->list_lock);
 117                         list_add(&dirty->root->dead_list, &root->dead_list);
 118                         spin_unlock(&root->list_lock);
 119
 120                         root->dirty_root = dirty;
 121                 } else {
 122                         WARN_ON(1);
 123                 }
 124                 root->last_trans = running_trans_id;
 125         }
 126         return 0;
 127 }
 128
 129 static void wait_current_trans(struct btrfs_root *root)
 130 {
 131         struct btrfs_transaction *cur_trans;
 132
 133         cur_trans = root->fs_info->running_transaction;
 134         if (cur_trans && cur_trans->blocked) {
 135                 DEFINE_WAIT(wait);
 136                 cur_trans->use_count++;
 137                 while(1) {
 138                         prepare_to_wait(&root->fs_info->transaction_wait, &wait,
 139                                         TASK_UNINTERRUPTIBLE);
 140                         if (cur_trans->blocked) {
 141                                 mutex_unlock(&root->fs_info->trans_mutex);
 142                                 schedule();
 143                                 mutex_lock(&root->fs_info->trans_mutex);
 144                                 finish_wait(&root->fs_info->transaction_wait,
 145                                             &wait);
 146                         } else {
 147                                 finish_wait(&root->fs_info->transaction_wait,
 148                                             &wait);
 149                                 break;
 150                         }
 151                 }
 152                 put_transaction(cur_trans);
 153         }
 154 }
 155
 156 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 157                                              int num_blocks, int wait)
 158 {
 159         struct btrfs_trans_handle *h =
 160                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 161         int ret;
 162
 163         mutex_lock(&root->fs_info->trans_mutex);
 164         if (!root->fs_info->log_root_recovering &&
 165             ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
 166                 wait_current_trans(root);
 167         ret = join_transaction(root);
 168         BUG_ON(ret);
 169
 170         btrfs_record_root_in_trans(root);
 171         h->transid = root->fs_info->running_transaction->transid;
 172         h->transaction = root->fs_info->running_transaction;
 173         h->blocks_reserved = num_blocks;
 174         h->blocks_used = 0;
 175         h->block_group = NULL;
 176         h->alloc_exclude_nr = 0;
 177         h->alloc_exclude_start = 0;
 178         root->fs_info->running_transaction->use_count++;
 179         mutex_unlock(&root->fs_info->trans_mutex);
 180         return h;
 181 }
 182
 183 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 184                                                    int num_blocks)
 185 {
 186         return start_transaction(root, num_blocks, 1);
 187 }
 188 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 189                                                    int num_blocks)
 190 {
 191         return start_transaction(root, num_blocks, 0);
 192 }
 193
 194 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
 195                                                          int num_blocks)
 196 {
 197         return start_transaction(r, num_blocks, 2);
 198 }
 199
 200
 201 static noinline int wait_for_commit(struct btrfs_root *root,
 202                                     struct btrfs_transaction *commit)
 203 {
 204         DEFINE_WAIT(wait);
 205         mutex_lock(&root->fs_info->trans_mutex);
 206         while(!commit->commit_done) {
 207                 prepare_to_wait(&commit->commit_wait, &wait,
 208                                 TASK_UNINTERRUPTIBLE);
 209                 if (commit->commit_done)
 210                         break;
 211                 mutex_unlock(&root->fs_info->trans_mutex);
 212                 schedule();
 213                 mutex_lock(&root->fs_info->trans_mutex);
 214         }
 215         mutex_unlock(&root->fs_info->trans_mutex);
 216         finish_wait(&commit->commit_wait, &wait);
 217         return 0;
 218 }
 219
 220 static void throttle_on_drops(struct btrfs_root *root)
 221 {
 222         struct btrfs_fs_info *info = root->fs_info;
 223         int harder_count = 0;
 224
 225 harder:
 226         if (atomic_read(&info->throttles)) {
 227                 DEFINE_WAIT(wait);
 228                 int thr;
 229                 thr = atomic_read(&info->throttle_gen);
 230
 231                 do {
 232                         prepare_to_wait(&info->transaction_throttle,
 233                                         &wait, TASK_UNINTERRUPTIBLE);
 234                         if (!atomic_read(&info->throttles)) {
 235                                 finish_wait(&info->transaction_throttle, &wait);
 236                                 break;
 237                         }
 238                         schedule();
 239                         finish_wait(&info->transaction_throttle, &wait);
 240                 } while (thr == atomic_read(&info->throttle_gen));
 241                 harder_count++;
 242
 243                 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
 244                     harder_count < 2)
 245                         goto harder;
 246
 247                 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
 248                     harder_count < 10)
 249                         goto harder;
 250
 251                 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
 252                     harder_count < 20)
 253                         goto harder;
 254         }
 255 }
 256
 257 void btrfs_throttle(struct btrfs_root *root)
 258 {
 259         mutex_lock(&root->fs_info->trans_mutex);
 260         if (!root->fs_info->open_ioctl_trans)
 261                 wait_current_trans(root);
 262         mutex_unlock(&root->fs_info->trans_mutex);
 263
 264         throttle_on_drops(root);
 265 }
 266
 267 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 268                           struct btrfs_root *root, int throttle)
 269 {
 270         struct btrfs_transaction *cur_trans;
 271         struct btrfs_fs_info *info = root->fs_info;
 272
 273         mutex_lock(&info->trans_mutex);
 274         cur_trans = info->running_transaction;
 275         WARN_ON(cur_trans != trans->transaction);
 276         WARN_ON(cur_trans->num_writers < 1);
 277         cur_trans->num_writers--;
 278
 279         if (waitqueue_active(&cur_trans->writer_wait))
 280                 wake_up(&cur_trans->writer_wait);
 281         put_transaction(cur_trans);
 282         mutex_unlock(&info->trans_mutex);
 283         memset(trans, 0, sizeof(*trans));
 284         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 285
 286         if (throttle)
 287                 throttle_on_drops(root);
 288
 289         return 0;
 290 }
 291
 292 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 293                           struct btrfs_root *root)
 294 {
 295         return __btrfs_end_transaction(trans, root, 0);
 296 }
 297
 298 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 299                                    struct btrfs_root *root)
 300 {
 301         return __btrfs_end_transaction(trans, root, 1);
 302 }
 303
 304
 305 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 306                                      struct btrfs_root *root)
 307 {
 308         int ret;
 309         int err = 0;
 310         int werr = 0;
 311         struct extent_io_tree *dirty_pages;
 312         struct page *page;
 313         struct inode *btree_inode = root->fs_info->btree_inode;
 314         u64 start = 0;
 315         u64 end;
 316         unsigned long index;
 317
 318         if (!trans || !trans->transaction) {
 319                 return filemap_write_and_wait(btree_inode->i_mapping);
 320         }
 321         dirty_pages = &trans->transaction->dirty_pages;
 322         while(1) {
 323                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
 324                                             EXTENT_DIRTY);
 325                 if (ret)
 326                         break;
 327                 while(start <= end) {
 328                         cond_resched();
 329
 330                         index = start >> PAGE_CACHE_SHIFT;
 331                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 332                         page = find_get_page(btree_inode->i_mapping, index);
 333                         if (!page)
 334                                 continue;
 335
 336                         btree_lock_page_hook(page);
 337                         if (!page->mapping) {
 338                                 unlock_page(page);
 339                                 page_cache_release(page);
 340                                 continue;
 341                         }
 342
 343                         if (PageWriteback(page)) {
 344                                 if (PageDirty(page))
 345                                         wait_on_page_writeback(page);
 346                                 else {
 347                                         unlock_page(page);
 348                                         page_cache_release(page);
 349                                         continue;
 350                                 }
 351                         }
 352                         err = write_one_page(page, 0);
 353                         if (err)
 354                                 werr = err;
 355                         page_cache_release(page);
 356                 }
 357         }
 358         while(1) {
 359                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 360                                             EXTENT_DIRTY);
 361                 if (ret)
 362                         break;
 363
 364                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 365                 while(start <= end) {
 366                         index = start >> PAGE_CACHE_SHIFT;
 367                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 368                         page = find_get_page(btree_inode->i_mapping, index);
 369                         if (!page)
 370                                 continue;
 371                         if (PageDirty(page)) {
 372                                 btree_lock_page_hook(page);
 373                                 wait_on_page_writeback(page);
 374                                 err = write_one_page(page, 0);
 375                                 if (err)
 376                                         werr = err;
 377                         }
 378                         wait_on_page_writeback(page);
 379                         page_cache_release(page);
 380                         cond_resched();
 381                 }
 382         }
 383         if (err)
 384                 werr = err;
 385         return werr;
 386 }
 387
 388 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 389                                struct btrfs_root *root)
 390 {
 391         int ret;
 392         u64 old_root_bytenr;
 393         struct btrfs_root *tree_root = root->fs_info->tree_root;
 394
 395         btrfs_write_dirty_block_groups(trans, root);
 396         while(1) {
 397                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 398                 if (old_root_bytenr == root->node->start)
 399                         break;
 400                 btrfs_set_root_bytenr(&root->root_item,
 401                                        root->node->start);
 402                 btrfs_set_root_level(&root->root_item,
 403                                      btrfs_header_level(root->node));
 404                 ret = btrfs_update_root(trans, tree_root,
 405                                         &root->root_key,
 406                                         &root->root_item);
 407                 BUG_ON(ret);
 408                 btrfs_write_dirty_block_groups(trans, root);
 409         }
 410         return 0;
 411 }
 412
 413 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 414                             struct btrfs_root *root)
 415 {
 416         struct btrfs_fs_info *fs_info = root->fs_info;
 417         struct list_head *next;
 418
 419         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 420                 next = fs_info->dirty_cowonly_roots.next;
 421                 list_del_init(next);
 422                 root = list_entry(next, struct btrfs_root, dirty_list);
 423                 update_cowonly_root(trans, root);
 424         }
 425         return 0;
 426 }
 427
 428 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
 429 {
 430         struct btrfs_dirty_root *dirty;
 431
 432         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 433         if (!dirty)
 434                 return -ENOMEM;
 435         dirty->root = root;
 436         dirty->latest_root = latest;
 437
 438         mutex_lock(&root->fs_info->trans_mutex);
 439         list_add(&dirty->list, &latest->fs_info->dead_roots);
 440         mutex_unlock(&root->fs_info->trans_mutex);
 441         return 0;
 442 }
 443
 444 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 445                                     struct radix_tree_root *radix,
 446                                     struct list_head *list)
 447 {
 448         struct btrfs_dirty_root *dirty;
 449         struct btrfs_root *gang[8];
 450         struct btrfs_root *root;
 451         int i;
 452         int ret;
 453         int err = 0;
 454         u32 refs;
 455
 456         while(1) {
 457                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 458                                                  ARRAY_SIZE(gang),
 459                                                  BTRFS_ROOT_TRANS_TAG);
 460                 if (ret == 0)
 461                         break;
 462                 for (i = 0; i < ret; i++) {
 463                         root = gang[i];
 464                         radix_tree_tag_clear(radix,
 465                                      (unsigned long)root->root_key.objectid,
 466                                      BTRFS_ROOT_TRANS_TAG);
 467
 468                         BUG_ON(!root->ref_tree);
 469                         dirty = root->dirty_root;
 470
 471                         btrfs_free_log(trans, root);
 472
 473                         if (root->commit_root == root->node) {
 474                                 WARN_ON(root->node->start !=
 475                                         btrfs_root_bytenr(&root->root_item));
 476
 477                                 free_extent_buffer(root->commit_root);
 478                                 root->commit_root = NULL;
 479                                 root->dirty_root = NULL;
 480
 481                                 spin_lock(&root->list_lock);
 482                                 list_del_init(&dirty->root->dead_list);
 483                                 spin_unlock(&root->list_lock);
 484
 485                                 kfree(dirty->root);
 486                                 kfree(dirty);
 487
 488                                 /* make sure to update the root on disk
 489                                  * so we get any updates to the block used
 490                                  * counts
 491                                  */
 492                                 err = btrfs_update_root(trans,
 493                                                 root->fs_info->tree_root,
 494                                                 &root->root_key,
 495                                                 &root->root_item);
 496                                 continue;
 497                         }
 498
 499                         memset(&root->root_item.drop_progress, 0,
 500                                sizeof(struct btrfs_disk_key));
 501                         root->root_item.drop_level = 0;
 502                         root->commit_root = NULL;
 503                         root->dirty_root = NULL;
 504                         root->root_key.offset = root->fs_info->generation;
 505                         btrfs_set_root_bytenr(&root->root_item,
 506                                               root->node->start);
 507                         btrfs_set_root_level(&root->root_item,
 508                                              btrfs_header_level(root->node));
 509                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 510                                                 &root->root_key,
 511                                                 &root->root_item);
 512                         if (err)
 513                                 break;
 514
 515                         refs = btrfs_root_refs(&dirty->root->root_item);
 516                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 517                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 518                                                 &dirty->root->root_key,
 519                                                 &dirty->root->root_item);
 520
 521                         BUG_ON(err);
 522                         if (refs == 1) {
 523                                 list_add(&dirty->list, list);
 524                         } else {
 525                                 WARN_ON(1);
 526                                 free_extent_buffer(dirty->root->node);
 527                                 kfree(dirty->root);
 528                                 kfree(dirty);
 529                         }
 530                 }
 531         }
 532         return err;
 533 }
 534
 535 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 536 {
 537         struct btrfs_fs_info *info = root->fs_info;
 538         int ret;
 539         struct btrfs_trans_handle *trans;
 540         unsigned long nr;
 541
 542         smp_mb();
 543         if (root->defrag_running)
 544                 return 0;
 545         trans = btrfs_start_transaction(root, 1);
 546         while (1) {
 547                 root->defrag_running = 1;
 548                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 549                 nr = trans->blocks_used;
 550                 btrfs_end_transaction(trans, root);
 551                 btrfs_btree_balance_dirty(info->tree_root, nr);
 552                 cond_resched();
 553
 554                 trans = btrfs_start_transaction(root, 1);
 555                 if (root->fs_info->closing || ret != -EAGAIN)
 556                         break;
 557         }
 558         root->defrag_running = 0;
 559         smp_mb();
 560         btrfs_end_transaction(trans, root);
 561         return 0;
 562 }
 563
 564 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 565                                      struct list_head *list)
 566 {
 567         struct btrfs_dirty_root *dirty;
 568         struct btrfs_trans_handle *trans;
 569         unsigned long nr;
 570         u64 num_bytes;
 571         u64 bytes_used;
 572         u64 max_useless;
 573         int ret = 0;
 574         int err;
 575
 576         while(!list_empty(list)) {
 577                 struct btrfs_root *root;
 578
 579                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 580                 list_del_init(&dirty->list);
 581
 582                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 583                 root = dirty->latest_root;
 584                 atomic_inc(&root->fs_info->throttles);
 585
 586                 mutex_lock(&root->fs_info->drop_mutex);
 587                 while(1) {
 588                         trans = btrfs_start_transaction(tree_root, 1);
 589                         ret = btrfs_drop_snapshot(trans, dirty->root);
 590                         if (ret != -EAGAIN) {
 591                                 break;
 592                         }
 593
 594                         err = btrfs_update_root(trans,
 595                                         tree_root,
 596                                         &dirty->root->root_key,
 597                                         &dirty->root->root_item);
 598                         if (err)
 599                                 ret = err;
 600                         nr = trans->blocks_used;
 601                         ret = btrfs_end_transaction(trans, tree_root);
 602                         BUG_ON(ret);
 603
 604                         mutex_unlock(&root->fs_info->drop_mutex);
 605                         btrfs_btree_balance_dirty(tree_root, nr);
 606                         cond_resched();
 607                         mutex_lock(&root->fs_info->drop_mutex);
 608                 }
 609                 BUG_ON(ret);
 610                 atomic_dec(&root->fs_info->throttles);
 611                 wake_up(&root->fs_info->transaction_throttle);
 612
 613                 mutex_lock(&root->fs_info->alloc_mutex);
 614                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 615                 bytes_used = btrfs_root_used(&root->root_item);
 616                 if (num_bytes) {
 617                         btrfs_record_root_in_trans(root);
 618                         btrfs_set_root_used(&root->root_item,
 619                                             bytes_used - num_bytes);
 620                 }
 621                 mutex_unlock(&root->fs_info->alloc_mutex);
 622
 623                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 624                 if (ret) {
 625                         BUG();
 626                         break;
 627                 }
 628                 mutex_unlock(&root->fs_info->drop_mutex);
 629
 630                 spin_lock(&root->list_lock);
 631                 list_del_init(&dirty->root->dead_list);
 632                 if (!list_empty(&root->dead_list)) {
 633                         struct btrfs_root *oldest;
 634                         oldest = list_entry(root->dead_list.prev,
 635                                             struct btrfs_root, dead_list);
 636                         max_useless = oldest->root_key.offset - 1;
 637                 } else {
 638                         max_useless = root->root_key.offset - 1;
 639                 }
 640                 spin_unlock(&root->list_lock);
 641
 642                 nr = trans->blocks_used;
 643                 ret = btrfs_end_transaction(trans, tree_root);
 644                 BUG_ON(ret);
 645
 646                 ret = btrfs_remove_leaf_refs(root, max_useless);
 647                 BUG_ON(ret);
 648
 649                 free_extent_buffer(dirty->root->node);
 650                 kfree(dirty->root);
 651                 kfree(dirty);
 652
 653                 btrfs_btree_balance_dirty(tree_root, nr);
 654                 cond_resched();
 655         }
 656         return ret;
 657 }
 658
 659 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 660                                    struct btrfs_fs_info *fs_info,
 661                                    struct btrfs_pending_snapshot *pending)
 662 {
 663         struct btrfs_key key;
 664         struct btrfs_root_item *new_root_item;
 665         struct btrfs_root *tree_root = fs_info->tree_root;
 666         struct btrfs_root *root = pending->root;
 667         struct extent_buffer *tmp;
 668         struct extent_buffer *old;
 669         int ret;
 670         int namelen;
 671         u64 objectid;
 672
 673         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 674         if (!new_root_item) {
 675                 ret = -ENOMEM;
 676                 goto fail;
 677         }
 678         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 679         if (ret)
 680                 goto fail;
 681
 682         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 683
 684         key.objectid = objectid;
 685         key.offset = 1;
 686         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 687
 688         old = btrfs_lock_root_node(root);
 689         btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
 690
 691         btrfs_copy_root(trans, root, old, &tmp, objectid);
 692         btrfs_tree_unlock(old);
 693         free_extent_buffer(old);
 694
 695         btrfs_set_root_bytenr(new_root_item, tmp->start);
 696         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 697         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 698                                 new_root_item);
 699         btrfs_tree_unlock(tmp);
 700         free_extent_buffer(tmp);
 701         if (ret)
 702                 goto fail;
 703
 704         /*
 705          * insert the directory item
 706          */
 707         key.offset = (u64)-1;
 708         namelen = strlen(pending->name);
 709         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 710                                     pending->name, namelen,
 711                                     root->fs_info->sb->s_root->d_inode->i_ino,
 712                                     &key, BTRFS_FT_DIR, 0);
 713
 714         if (ret)
 715                 goto fail;
 716
 717         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 718                              pending->name, strlen(pending->name), objectid,
 719                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 720
 721         /* Invalidate existing dcache entry for new snapshot. */
 722         btrfs_invalidate_dcache_root(root, pending->name, namelen);
 723
 724 fail:
 725         kfree(new_root_item);
 726         return ret;
 727 }
 728
 729 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 730                                              struct btrfs_fs_info *fs_info)
 731 {
 732         struct btrfs_pending_snapshot *pending;
 733         struct list_head *head = &trans->transaction->pending_snapshots;
 734         int ret;
 735
 736         while(!list_empty(head)) {
 737                 pending = list_entry(head->next,
 738                                      struct btrfs_pending_snapshot, list);
 739                 ret = create_pending_snapshot(trans, fs_info, pending);
 740                 BUG_ON(ret);
 741                 list_del(&pending->list);
 742                 kfree(pending->name);
 743                 kfree(pending);
 744         }
 745         return 0;
 746 }
 747
 748 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 749                              struct btrfs_root *root)
 750 {
 751         unsigned long joined = 0;
 752         unsigned long timeout = 1;
 753         struct btrfs_transaction *cur_trans;
 754         struct btrfs_transaction *prev_trans = NULL;
 755         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 756         struct list_head dirty_fs_roots;
 757         struct extent_io_tree *pinned_copy;
 758         DEFINE_WAIT(wait);
 759         int ret;
 760
 761         INIT_LIST_HEAD(&dirty_fs_roots);
 762         mutex_lock(&root->fs_info->trans_mutex);
 763         if (trans->transaction->in_commit) {
 764                 cur_trans = trans->transaction;
 765                 trans->transaction->use_count++;
 766                 mutex_unlock(&root->fs_info->trans_mutex);
 767                 btrfs_end_transaction(trans, root);
 768
 769                 ret = wait_for_commit(root, cur_trans);
 770                 BUG_ON(ret);
 771
 772                 mutex_lock(&root->fs_info->trans_mutex);
 773                 put_transaction(cur_trans);
 774                 mutex_unlock(&root->fs_info->trans_mutex);
 775
 776                 return 0;
 777         }
 778
 779         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 780         if (!pinned_copy)
 781                 return -ENOMEM;
 782
 783         extent_io_tree_init(pinned_copy,
 784                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 785
 786         trans->transaction->in_commit = 1;
 787         trans->transaction->blocked = 1;
 788         cur_trans = trans->transaction;
 789         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 790                 prev_trans = list_entry(cur_trans->list.prev,
 791                                         struct btrfs_transaction, list);
 792                 if (!prev_trans->commit_done) {
 793                         prev_trans->use_count++;
 794                         mutex_unlock(&root->fs_info->trans_mutex);
 795
 796                         wait_for_commit(root, prev_trans);
 797
 798                         mutex_lock(&root->fs_info->trans_mutex);
 799                         put_transaction(prev_trans);
 800                 }
 801         }
 802
 803         do {
 804                 int snap_pending = 0;
 805                 joined = cur_trans->num_joined;
 806                 if (!list_empty(&trans->transaction->pending_snapshots))
 807                         snap_pending = 1;
 808
 809                 WARN_ON(cur_trans != trans->transaction);
 810                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 811                                 TASK_UNINTERRUPTIBLE);
 812
 813                 if (cur_trans->num_writers > 1)
 814                         timeout = MAX_SCHEDULE_TIMEOUT;
 815                 else
 816                         timeout = 1;
 817
 818                 mutex_unlock(&root->fs_info->trans_mutex);
 819
 820                 if (snap_pending) {
 821                         ret = btrfs_wait_ordered_extents(root, 1);
 822                         BUG_ON(ret);
 823                 }
 824
 825                 schedule_timeout(timeout);
 826
 827                 mutex_lock(&root->fs_info->trans_mutex);
 828                 finish_wait(&cur_trans->writer_wait, &wait);
 829         } while (cur_trans->num_writers > 1 ||
 830                  (cur_trans->num_joined != joined));
 831
 832         ret = create_pending_snapshots(trans, root->fs_info);
 833         BUG_ON(ret);
 834
 835         WARN_ON(cur_trans != trans->transaction);
 836
 837         /* btrfs_commit_tree_roots is responsible for getting the
 838          * various roots consistent with each other.  Every pointer
 839          * in the tree of tree roots has to point to the most up to date
 840          * root for every subvolume and other tree.  So, we have to keep
 841          * the tree logging code from jumping in and changing any
 842          * of the trees.
 843          *
 844          * At this point in the commit, there can't be any tree-log
 845          * writers, but a little lower down we drop the trans mutex
 846          * and let new people in.  By holding the tree_log_mutex
 847          * from now until after the super is written, we avoid races
 848          * with the tree-log code.
 849          */
 850         mutex_lock(&root->fs_info->tree_log_mutex);
 851
 852         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 853                               &dirty_fs_roots);
 854         BUG_ON(ret);
 855
 856         /* add_dirty_roots gets rid of all the tree log roots, it is now
 857          * safe to free the root of tree log roots
 858          */
 859         btrfs_free_log_root_tree(trans, root->fs_info);
 860
 861         ret = btrfs_commit_tree_roots(trans, root);
 862         BUG_ON(ret);
 863
 864         cur_trans = root->fs_info->running_transaction;
 865         spin_lock(&root->fs_info->new_trans_lock);
 866         root->fs_info->running_transaction = NULL;
 867         spin_unlock(&root->fs_info->new_trans_lock);
 868         btrfs_set_super_generation(&root->fs_info->super_copy,
 869                                    cur_trans->transid);
 870         btrfs_set_super_root(&root->fs_info->super_copy,
 871                              root->fs_info->tree_root->node->start);
 872         btrfs_set_super_root_level(&root->fs_info->super_copy,
 873                            btrfs_header_level(root->fs_info->tree_root->node));
 874
 875         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 876                                    chunk_root->node->start);
 877         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 878                                          btrfs_header_level(chunk_root->node));
 879
 880         if (!root->fs_info->log_root_recovering) {
 881                 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
 882                 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
 883         }
 884
 885         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 886                sizeof(root->fs_info->super_copy));
 887
 888         btrfs_copy_pinned(root, pinned_copy);
 889
 890         trans->transaction->blocked = 0;
 891         wake_up(&root->fs_info->transaction_throttle);
 892         wake_up(&root->fs_info->transaction_wait);
 893
 894         mutex_unlock(&root->fs_info->trans_mutex);
 895         ret = btrfs_write_and_wait_transaction(trans, root);
 896         BUG_ON(ret);
 897         write_ctree_super(trans, root);
 898
 899         /*
 900          * the super is written, we can safely allow the tree-loggers
 901          * to go about their business
 902          */
 903         mutex_unlock(&root->fs_info->tree_log_mutex);
 904
 905         btrfs_finish_extent_commit(trans, root, pinned_copy);
 906         mutex_lock(&root->fs_info->trans_mutex);
 907
 908         kfree(pinned_copy);
 909
 910         cur_trans->commit_done = 1;
 911         root->fs_info->last_trans_committed = cur_trans->transid;
 912         wake_up(&cur_trans->commit_wait);
 913         put_transaction(cur_trans);
 914         put_transaction(cur_trans);
 915
 916         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 917         if (root->fs_info->closing)
 918                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 919
 920         mutex_unlock(&root->fs_info->trans_mutex);
 921         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 922
 923         if (root->fs_info->closing) {
 924                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 925         }
 926         return ret;
 927 }
 928
 929 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 930 {
 931         struct list_head dirty_roots;
 932         INIT_LIST_HEAD(&dirty_roots);
 933 again:
 934         mutex_lock(&root->fs_info->trans_mutex);
 935         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 936         mutex_unlock(&root->fs_info->trans_mutex);
 937
 938         if (!list_empty(&dirty_roots)) {
 939                 drop_dirty_roots(root, &dirty_roots);
 940                 goto again;
 941         }
 942         return 0;
 943 }