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;
 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;
 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, 0, &start, &end,
 321                                             EXTENT_DIRTY);
 322                 if (ret)
 323                         break;
 324                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 325                 while(start <= end) {
 326                         index = start >> PAGE_CACHE_SHIFT;
 327                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 328                         page = find_lock_page(btree_inode->i_mapping, index);
 329                         if (!page)
 330                                 continue;
 331                         if (PageWriteback(page)) {
 332                                 if (PageDirty(page))
 333                                         wait_on_page_writeback(page);
 334                                 else {
 335                                         unlock_page(page);
 336                                         page_cache_release(page);
 337                                         continue;
 338                                 }
 339                         }
 340                         err = write_one_page(page, 0);
 341                         if (err)
 342                                 werr = err;
 343                         page_cache_release(page);
 344                 }
 345         }
 346         err = filemap_fdatawait(btree_inode->i_mapping);
 347         if (err)
 348                 werr = err;
 349         return werr;
 350 }
 351
 352 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 353                                struct btrfs_root *root)
 354 {
 355         int ret;
 356         u64 old_root_bytenr;
 357         struct btrfs_root *tree_root = root->fs_info->tree_root;
 358
 359         btrfs_write_dirty_block_groups(trans, root);
 360         while(1) {
 361                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 362                 if (old_root_bytenr == root->node->start)
 363                         break;
 364                 btrfs_set_root_bytenr(&root->root_item,
 365                                        root->node->start);
 366                 btrfs_set_root_level(&root->root_item,
 367                                      btrfs_header_level(root->node));
 368                 ret = btrfs_update_root(trans, tree_root,
 369                                         &root->root_key,
 370                                         &root->root_item);
 371                 BUG_ON(ret);
 372                 btrfs_write_dirty_block_groups(trans, root);
 373         }
 374         return 0;
 375 }
 376
 377 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 378                             struct btrfs_root *root)
 379 {
 380         struct btrfs_fs_info *fs_info = root->fs_info;
 381         struct list_head *next;
 382
 383         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 384                 next = fs_info->dirty_cowonly_roots.next;
 385                 list_del_init(next);
 386                 root = list_entry(next, struct btrfs_root, dirty_list);
 387                 update_cowonly_root(trans, root);
 388         }
 389         return 0;
 390 }
 391
 392 int btrfs_add_dead_root(struct btrfs_root *root,
 393                         struct btrfs_root *latest,
 394                         struct list_head *dead_list)
 395 {
 396         struct btrfs_dirty_root *dirty;
 397
 398         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 399         if (!dirty)
 400                 return -ENOMEM;
 401         dirty->root = root;
 402         dirty->latest_root = latest;
 403         list_add(&dirty->list, dead_list);
 404         return 0;
 405 }
 406
 407 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 408                                     struct radix_tree_root *radix,
 409                                     struct list_head *list)
 410 {
 411         struct btrfs_dirty_root *dirty;
 412         struct btrfs_root *gang[8];
 413         struct btrfs_root *root;
 414         int i;
 415         int ret;
 416         int err = 0;
 417         u32 refs;
 418
 419         while(1) {
 420                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 421                                                  ARRAY_SIZE(gang),
 422                                                  BTRFS_ROOT_TRANS_TAG);
 423                 if (ret == 0)
 424                         break;
 425                 for (i = 0; i < ret; i++) {
 426                         root = gang[i];
 427                         radix_tree_tag_clear(radix,
 428                                      (unsigned long)root->root_key.objectid,
 429                                      BTRFS_ROOT_TRANS_TAG);
 430
 431                         BUG_ON(!root->ref_tree);
 432                         dirty = root->dirty_root;
 433
 434                         if (root->commit_root == root->node) {
 435                                 WARN_ON(root->node->start !=
 436                                         btrfs_root_bytenr(&root->root_item));
 437
 438                                 free_extent_buffer(root->commit_root);
 439                                 root->commit_root = NULL;
 440
 441                                 spin_lock(&root->list_lock);
 442                                 list_del_init(&dirty->root->dead_list);
 443                                 spin_unlock(&root->list_lock);
 444
 445                                 kfree(dirty->root);
 446                                 kfree(dirty);
 447
 448                                 /* make sure to update the root on disk
 449                                  * so we get any updates to the block used
 450                                  * counts
 451                                  */
 452                                 err = btrfs_update_root(trans,
 453                                                 root->fs_info->tree_root,
 454                                                 &root->root_key,
 455                                                 &root->root_item);
 456                                 continue;
 457                         }
 458
 459                         memset(&root->root_item.drop_progress, 0,
 460                                sizeof(struct btrfs_disk_key));
 461                         root->root_item.drop_level = 0;
 462                         root->commit_root = NULL;
 463                         root->root_key.offset = root->fs_info->generation;
 464                         btrfs_set_root_bytenr(&root->root_item,
 465                                               root->node->start);
 466                         btrfs_set_root_level(&root->root_item,
 467                                              btrfs_header_level(root->node));
 468                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 469                                                 &root->root_key,
 470                                                 &root->root_item);
 471                         if (err)
 472                                 break;
 473
 474                         refs = btrfs_root_refs(&dirty->root->root_item);
 475                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 476                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 477                                                 &dirty->root->root_key,
 478                                                 &dirty->root->root_item);
 479
 480                         BUG_ON(err);
 481                         if (refs == 1) {
 482                                 list_add(&dirty->list, list);
 483                         } else {
 484                                 WARN_ON(1);
 485                                 free_extent_buffer(dirty->root->node);
 486                                 kfree(dirty->root);
 487                                 kfree(dirty);
 488                         }
 489                 }
 490         }
 491         return err;
 492 }
 493
 494 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 495 {
 496         struct btrfs_fs_info *info = root->fs_info;
 497         int ret;
 498         struct btrfs_trans_handle *trans;
 499         unsigned long nr;
 500
 501         smp_mb();
 502         if (root->defrag_running)
 503                 return 0;
 504         trans = btrfs_start_transaction(root, 1);
 505         while (1) {
 506                 root->defrag_running = 1;
 507                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 508                 nr = trans->blocks_used;
 509                 btrfs_end_transaction(trans, root);
 510                 btrfs_btree_balance_dirty(info->tree_root, nr);
 511                 cond_resched();
 512
 513                 trans = btrfs_start_transaction(root, 1);
 514                 if (root->fs_info->closing || ret != -EAGAIN)
 515                         break;
 516         }
 517         root->defrag_running = 0;
 518         smp_mb();
 519         btrfs_end_transaction(trans, root);
 520         return 0;
 521 }
 522
 523 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 524                                      struct list_head *list)
 525 {
 526         struct btrfs_dirty_root *dirty;
 527         struct btrfs_trans_handle *trans;
 528         unsigned long nr;
 529         u64 num_bytes;
 530         u64 bytes_used;
 531         u64 max_useless;
 532         int ret = 0;
 533         int err;
 534
 535         while(!list_empty(list)) {
 536                 struct btrfs_root *root;
 537
 538                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 539                 list_del_init(&dirty->list);
 540
 541                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 542                 root = dirty->latest_root;
 543                 atomic_inc(&root->fs_info->throttles);
 544
 545                 mutex_lock(&root->fs_info->drop_mutex);
 546                 while(1) {
 547                         trans = btrfs_start_transaction(tree_root, 1);
 548                         ret = btrfs_drop_snapshot(trans, dirty->root);
 549                         if (ret != -EAGAIN) {
 550                                 break;
 551                         }
 552
 553                         err = btrfs_update_root(trans,
 554                                         tree_root,
 555                                         &dirty->root->root_key,
 556                                         &dirty->root->root_item);
 557                         if (err)
 558                                 ret = err;
 559                         nr = trans->blocks_used;
 560                         ret = btrfs_end_transaction(trans, tree_root);
 561                         BUG_ON(ret);
 562
 563                         mutex_unlock(&root->fs_info->drop_mutex);
 564                         btrfs_btree_balance_dirty(tree_root, nr);
 565                         cond_resched();
 566                         mutex_lock(&root->fs_info->drop_mutex);
 567                 }
 568                 BUG_ON(ret);
 569                 atomic_dec(&root->fs_info->throttles);
 570                 wake_up(&root->fs_info->transaction_throttle);
 571
 572                 mutex_lock(&root->fs_info->alloc_mutex);
 573                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 574                 bytes_used = btrfs_root_used(&root->root_item);
 575                 if (num_bytes) {
 576                         record_root_in_trans(root);
 577                         btrfs_set_root_used(&root->root_item,
 578                                             bytes_used - num_bytes);
 579                 }
 580                 mutex_unlock(&root->fs_info->alloc_mutex);
 581
 582                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 583                 if (ret) {
 584                         BUG();
 585                         break;
 586                 }
 587                 mutex_unlock(&root->fs_info->drop_mutex);
 588
 589                 spin_lock(&root->list_lock);
 590                 list_del_init(&dirty->root->dead_list);
 591                 if (!list_empty(&root->dead_list)) {
 592                         struct btrfs_root *oldest;
 593                         oldest = list_entry(root->dead_list.prev,
 594                                             struct btrfs_root, dead_list);
 595                         max_useless = oldest->root_key.offset - 1;
 596                 } else {
 597                         max_useless = root->root_key.offset - 1;
 598                 }
 599                 spin_unlock(&root->list_lock);
 600
 601                 nr = trans->blocks_used;
 602                 ret = btrfs_end_transaction(trans, tree_root);
 603                 BUG_ON(ret);
 604
 605                 ret = btrfs_remove_leaf_refs(root, max_useless);
 606                 BUG_ON(ret);
 607
 608                 free_extent_buffer(dirty->root->node);
 609                 kfree(dirty->root);
 610                 kfree(dirty);
 611
 612                 btrfs_btree_balance_dirty(tree_root, nr);
 613                 cond_resched();
 614         }
 615         return ret;
 616 }
 617
 618 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 619                                    struct btrfs_fs_info *fs_info,
 620                                    struct btrfs_pending_snapshot *pending)
 621 {
 622         struct btrfs_key key;
 623         struct btrfs_root_item *new_root_item;
 624         struct btrfs_root *tree_root = fs_info->tree_root;
 625         struct btrfs_root *root = pending->root;
 626         struct extent_buffer *tmp;
 627         struct extent_buffer *old;
 628         int ret;
 629         int namelen;
 630         u64 objectid;
 631
 632         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 633         if (!new_root_item) {
 634                 ret = -ENOMEM;
 635                 goto fail;
 636         }
 637         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 638         if (ret)
 639                 goto fail;
 640
 641         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 642
 643         key.objectid = objectid;
 644         key.offset = 1;
 645         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 646
 647         old = btrfs_lock_root_node(root);
 648         btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
 649
 650         btrfs_copy_root(trans, root, old, &tmp, objectid);
 651         btrfs_tree_unlock(old);
 652         free_extent_buffer(old);
 653
 654         btrfs_set_root_bytenr(new_root_item, tmp->start);
 655         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 656         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 657                                 new_root_item);
 658         btrfs_tree_unlock(tmp);
 659         free_extent_buffer(tmp);
 660         if (ret)
 661                 goto fail;
 662
 663         /*
 664          * insert the directory item
 665          */
 666         key.offset = (u64)-1;
 667         namelen = strlen(pending->name);
 668         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 669                                     pending->name, namelen,
 670                                     root->fs_info->sb->s_root->d_inode->i_ino,
 671                                     &key, BTRFS_FT_DIR, 0);
 672
 673         if (ret)
 674                 goto fail;
 675
 676         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 677                              pending->name, strlen(pending->name), objectid,
 678                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 679
 680         /* Invalidate existing dcache entry for new snapshot. */
 681         btrfs_invalidate_dcache_root(root, pending->name, namelen);
 682
 683 fail:
 684         kfree(new_root_item);
 685         return ret;
 686 }
 687
 688 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 689                                              struct btrfs_fs_info *fs_info)
 690 {
 691         struct btrfs_pending_snapshot *pending;
 692         struct list_head *head = &trans->transaction->pending_snapshots;
 693         int ret;
 694
 695         while(!list_empty(head)) {
 696                 pending = list_entry(head->next,
 697                                      struct btrfs_pending_snapshot, list);
 698                 ret = create_pending_snapshot(trans, fs_info, pending);
 699                 BUG_ON(ret);
 700                 list_del(&pending->list);
 701                 kfree(pending->name);
 702                 kfree(pending);
 703         }
 704         return 0;
 705 }
 706
 707 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 708                              struct btrfs_root *root)
 709 {
 710         unsigned long joined = 0;
 711         unsigned long timeout = 1;
 712         struct btrfs_transaction *cur_trans;
 713         struct btrfs_transaction *prev_trans = NULL;
 714         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 715         struct list_head dirty_fs_roots;
 716         struct extent_io_tree *pinned_copy;
 717         DEFINE_WAIT(wait);
 718         int ret;
 719
 720         INIT_LIST_HEAD(&dirty_fs_roots);
 721
 722         mutex_lock(&root->fs_info->trans_mutex);
 723         if (trans->transaction->in_commit) {
 724                 cur_trans = trans->transaction;
 725                 trans->transaction->use_count++;
 726                 mutex_unlock(&root->fs_info->trans_mutex);
 727                 btrfs_end_transaction(trans, root);
 728
 729                 ret = wait_for_commit(root, cur_trans);
 730                 BUG_ON(ret);
 731
 732                 mutex_lock(&root->fs_info->trans_mutex);
 733                 put_transaction(cur_trans);
 734                 mutex_unlock(&root->fs_info->trans_mutex);
 735
 736                 return 0;
 737         }
 738
 739         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 740         if (!pinned_copy)
 741                 return -ENOMEM;
 742
 743         extent_io_tree_init(pinned_copy,
 744                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 745
 746         trans->transaction->in_commit = 1;
 747         trans->transaction->blocked = 1;
 748         cur_trans = trans->transaction;
 749         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 750                 prev_trans = list_entry(cur_trans->list.prev,
 751                                         struct btrfs_transaction, list);
 752                 if (!prev_trans->commit_done) {
 753                         prev_trans->use_count++;
 754                         mutex_unlock(&root->fs_info->trans_mutex);
 755
 756                         wait_for_commit(root, prev_trans);
 757
 758                         mutex_lock(&root->fs_info->trans_mutex);
 759                         put_transaction(prev_trans);
 760                 }
 761         }
 762
 763         do {
 764                 joined = cur_trans->num_joined;
 765                 WARN_ON(cur_trans != trans->transaction);
 766                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 767                                 TASK_UNINTERRUPTIBLE);
 768
 769                 if (cur_trans->num_writers > 1)
 770                         timeout = MAX_SCHEDULE_TIMEOUT;
 771                 else
 772                         timeout = 1;
 773
 774                 mutex_unlock(&root->fs_info->trans_mutex);
 775
 776                 schedule_timeout(timeout);
 777
 778                 mutex_lock(&root->fs_info->trans_mutex);
 779                 finish_wait(&cur_trans->writer_wait, &wait);
 780         } while (cur_trans->num_writers > 1 ||
 781                  (cur_trans->num_joined != joined));
 782
 783         ret = create_pending_snapshots(trans, root->fs_info);
 784         BUG_ON(ret);
 785
 786         WARN_ON(cur_trans != trans->transaction);
 787
 788         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 789                               &dirty_fs_roots);
 790         BUG_ON(ret);
 791
 792         ret = btrfs_commit_tree_roots(trans, root);
 793         BUG_ON(ret);
 794
 795         cur_trans = root->fs_info->running_transaction;
 796         spin_lock(&root->fs_info->new_trans_lock);
 797         root->fs_info->running_transaction = NULL;
 798         spin_unlock(&root->fs_info->new_trans_lock);
 799         btrfs_set_super_generation(&root->fs_info->super_copy,
 800                                    cur_trans->transid);
 801         btrfs_set_super_root(&root->fs_info->super_copy,
 802                              root->fs_info->tree_root->node->start);
 803         btrfs_set_super_root_level(&root->fs_info->super_copy,
 804                            btrfs_header_level(root->fs_info->tree_root->node));
 805
 806         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 807                                    chunk_root->node->start);
 808         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 809                                          btrfs_header_level(chunk_root->node));
 810         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 811                sizeof(root->fs_info->super_copy));
 812
 813         btrfs_copy_pinned(root, pinned_copy);
 814
 815         trans->transaction->blocked = 0;
 816         wake_up(&root->fs_info->transaction_throttle);
 817         wake_up(&root->fs_info->transaction_wait);
 818
 819         mutex_unlock(&root->fs_info->trans_mutex);
 820         ret = btrfs_write_and_wait_transaction(trans, root);
 821         BUG_ON(ret);
 822         write_ctree_super(trans, root);
 823
 824         btrfs_finish_extent_commit(trans, root, pinned_copy);
 825         mutex_lock(&root->fs_info->trans_mutex);
 826
 827         kfree(pinned_copy);
 828
 829         cur_trans->commit_done = 1;
 830         root->fs_info->last_trans_committed = cur_trans->transid;
 831         wake_up(&cur_trans->commit_wait);
 832         put_transaction(cur_trans);
 833         put_transaction(cur_trans);
 834
 835         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 836         if (root->fs_info->closing)
 837                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 838
 839         mutex_unlock(&root->fs_info->trans_mutex);
 840         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 841
 842         if (root->fs_info->closing) {
 843                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 844         }
 845         return ret;
 846 }
 847
 848 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 849 {
 850         struct list_head dirty_roots;
 851         INIT_LIST_HEAD(&dirty_roots);
 852 again:
 853         mutex_lock(&root->fs_info->trans_mutex);
 854         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 855         mutex_unlock(&root->fs_info->trans_mutex);
 856
 857         if (!list_empty(&dirty_roots)) {
 858                 drop_dirty_roots(root, &dirty_roots);
 859                 goto again;
 860         }
 861         return 0;
 862 }