Pileus Git - ~andy/linux/blob - fs/btrfs/transaction.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/sched.h>
  21 #include <linux/writeback.h>
  22 #include <linux/pagemap.h>
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26 #include "locking.h"
  27 #include "ref-cache.h"
  28
  29 static int total_trans = 0;
  30 extern struct kmem_cache *btrfs_trans_handle_cachep;
  31 extern struct kmem_cache *btrfs_transaction_cachep;
  32
  33 #define BTRFS_ROOT_TRANS_TAG 0
  34
  35 static noinline void put_transaction(struct btrfs_transaction *transaction)
  36 {
  37         WARN_ON(transaction->use_count == 0);
  38         transaction->use_count--;
  39         if (transaction->use_count == 0) {
  40                 WARN_ON(total_trans == 0);
  41                 total_trans--;
  42                 list_del_init(&transaction->list);
  43                 memset(transaction, 0, sizeof(*transaction));
  44                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  45         }
  46 }
  47
  48 static noinline int join_transaction(struct btrfs_root *root)
  49 {
  50         struct btrfs_transaction *cur_trans;
  51         cur_trans = root->fs_info->running_transaction;
  52         if (!cur_trans) {
  53                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  54                                              GFP_NOFS);
  55                 total_trans++;
  56                 BUG_ON(!cur_trans);
  57                 root->fs_info->generation++;
  58                 root->fs_info->last_alloc = 0;
  59                 root->fs_info->last_data_alloc = 0;
  60                 cur_trans->num_writers = 1;
  61                 cur_trans->num_joined = 0;
  62                 cur_trans->transid = root->fs_info->generation;
  63                 init_waitqueue_head(&cur_trans->writer_wait);
  64                 init_waitqueue_head(&cur_trans->commit_wait);
  65                 cur_trans->in_commit = 0;
  66                 cur_trans->blocked = 0;
  67                 cur_trans->use_count = 1;
  68                 cur_trans->commit_done = 0;
  69                 cur_trans->start_time = get_seconds();
  70                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  71                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  72                 extent_io_tree_init(&cur_trans->dirty_pages,
  73                                      root->fs_info->btree_inode->i_mapping,
  74                                      GFP_NOFS);
  75                 spin_lock(&root->fs_info->new_trans_lock);
  76                 root->fs_info->running_transaction = cur_trans;
  77                 spin_unlock(&root->fs_info->new_trans_lock);
  78         } else {
  79                 cur_trans->num_writers++;
  80                 cur_trans->num_joined++;
  81         }
  82
  83         return 0;
  84 }
  85
  86 static noinline int record_root_in_trans(struct btrfs_root *root)
  87 {
  88         struct btrfs_dirty_root *dirty;
  89         u64 running_trans_id = root->fs_info->running_transaction->transid;
  90         if (root->ref_cows && root->last_trans < running_trans_id) {
  91                 WARN_ON(root == root->fs_info->extent_root);
  92                 if (root->root_item.refs != 0) {
  93                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  94                                    (unsigned long)root->root_key.objectid,
  95                                    BTRFS_ROOT_TRANS_TAG);
  96
  97                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
  98                         BUG_ON(!dirty);
  99                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 100                         BUG_ON(!dirty->root);
 101                         dirty->latest_root = root;
 102                         INIT_LIST_HEAD(&dirty->list);
 103
 104                         root->commit_root = btrfs_root_node(root);
 105
 106                         memcpy(dirty->root, root, sizeof(*root));
 107                         spin_lock_init(&dirty->root->node_lock);
 108                         spin_lock_init(&dirty->root->list_lock);
 109                         mutex_init(&dirty->root->objectid_mutex);
 110                         INIT_LIST_HEAD(&dirty->root->dead_list);
 111                         dirty->root->node = root->commit_root;
 112                         dirty->root->commit_root = NULL;
 113
 114                         spin_lock(&root->list_lock);
 115                         list_add(&dirty->root->dead_list, &root->dead_list);
 116                         spin_unlock(&root->list_lock);
 117
 118                         root->dirty_root = dirty;
 119                 } else {
 120                         WARN_ON(1);
 121                 }
 122                 root->last_trans = running_trans_id;
 123         }
 124         return 0;
 125 }
 126
 127 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 128                                              int num_blocks, int join)
 129 {
 130         struct btrfs_trans_handle *h =
 131                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 132         struct btrfs_transaction *cur_trans;
 133         int ret;
 134
 135         mutex_lock(&root->fs_info->trans_mutex);
 136         cur_trans = root->fs_info->running_transaction;
 137         if (cur_trans && cur_trans->blocked && !join) {
 138                 DEFINE_WAIT(wait);
 139                 cur_trans->use_count++;
 140                 while(1) {
 141                         prepare_to_wait(&root->fs_info->transaction_wait, &wait,
 142                                         TASK_UNINTERRUPTIBLE);
 143                         if (cur_trans->blocked) {
 144                                 mutex_unlock(&root->fs_info->trans_mutex);
 145                                 schedule();
 146                                 mutex_lock(&root->fs_info->trans_mutex);
 147                                 finish_wait(&root->fs_info->transaction_wait,
 148                                             &wait);
 149                         } else {
 150                                 finish_wait(&root->fs_info->transaction_wait,
 151                                             &wait);
 152                                 break;
 153                         }
 154                 }
 155                 put_transaction(cur_trans);
 156         }
 157         ret = join_transaction(root);
 158         BUG_ON(ret);
 159
 160         record_root_in_trans(root);
 161         h->transid = root->fs_info->running_transaction->transid;
 162         h->transaction = root->fs_info->running_transaction;
 163         h->blocks_reserved = num_blocks;
 164         h->blocks_used = 0;
 165         h->block_group = NULL;
 166         h->alloc_exclude_nr = 0;
 167         h->alloc_exclude_start = 0;
 168         root->fs_info->running_transaction->use_count++;
 169         mutex_unlock(&root->fs_info->trans_mutex);
 170         return h;
 171 }
 172
 173 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 174                                                    int num_blocks)
 175 {
 176         return start_transaction(root, num_blocks, 0);
 177 }
 178 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 179                                                    int num_blocks)
 180 {
 181         return start_transaction(root, num_blocks, 1);
 182 }
 183
 184 static noinline int wait_for_commit(struct btrfs_root *root,
 185                                     struct btrfs_transaction *commit)
 186 {
 187         DEFINE_WAIT(wait);
 188         mutex_lock(&root->fs_info->trans_mutex);
 189         while(!commit->commit_done) {
 190                 prepare_to_wait(&commit->commit_wait, &wait,
 191                                 TASK_UNINTERRUPTIBLE);
 192                 if (commit->commit_done)
 193                         break;
 194                 mutex_unlock(&root->fs_info->trans_mutex);
 195                 schedule();
 196                 mutex_lock(&root->fs_info->trans_mutex);
 197         }
 198         mutex_unlock(&root->fs_info->trans_mutex);
 199         finish_wait(&commit->commit_wait, &wait);
 200         return 0;
 201 }
 202
 203 void btrfs_throttle(struct btrfs_root *root)
 204 {
 205         struct btrfs_fs_info *info = root->fs_info;
 206
 207 harder:
 208         if (atomic_read(&info->throttles)) {
 209                 DEFINE_WAIT(wait);
 210                 int thr;
 211                 int harder_count = 0;
 212                 thr = atomic_read(&info->throttle_gen);
 213
 214                 do {
 215                         prepare_to_wait(&info->transaction_throttle,
 216                                         &wait, TASK_UNINTERRUPTIBLE);
 217                         if (!atomic_read(&info->throttles)) {
 218                                 finish_wait(&info->transaction_throttle, &wait);
 219                                 break;
 220                         }
 221                         schedule();
 222                         finish_wait(&info->transaction_throttle, &wait);
 223                 } while (thr == atomic_read(&info->throttle_gen));
 224
 225                 if (harder_count < 5 &&
 226                     info->total_ref_cache_size > 5 * 1024 * 1024) {
 227                         harder_count++;
 228                         goto harder;
 229                 }
 230
 231                 if (harder_count < 10 &&
 232                     info->total_ref_cache_size > 10 * 1024 * 1024) {
 233                         harder_count++;
 234                         goto harder;
 235                 }
 236         }
 237 }
 238
 239 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 240                           struct btrfs_root *root, int throttle)
 241 {
 242         struct btrfs_transaction *cur_trans;
 243         struct btrfs_fs_info *info = root->fs_info;
 244
 245         mutex_lock(&info->trans_mutex);
 246         cur_trans = info->running_transaction;
 247         WARN_ON(cur_trans != trans->transaction);
 248         WARN_ON(cur_trans->num_writers < 1);
 249         cur_trans->num_writers--;
 250
 251         if (waitqueue_active(&cur_trans->writer_wait))
 252                 wake_up(&cur_trans->writer_wait);
 253         put_transaction(cur_trans);
 254         mutex_unlock(&info->trans_mutex);
 255         memset(trans, 0, sizeof(*trans));
 256         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 257
 258         if (throttle)
 259                 btrfs_throttle(root);
 260
 261         return 0;
 262 }
 263
 264 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 265                           struct btrfs_root *root)
 266 {
 267         return __btrfs_end_transaction(trans, root, 0);
 268 }
 269
 270 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 271                                    struct btrfs_root *root)
 272 {
 273         return __btrfs_end_transaction(trans, root, 1);
 274 }
 275
 276
 277 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 278                                      struct btrfs_root *root)
 279 {
 280         int ret;
 281         int err;
 282         int werr = 0;
 283         struct extent_io_tree *dirty_pages;
 284         struct page *page;
 285         struct inode *btree_inode = root->fs_info->btree_inode;
 286         u64 start;
 287         u64 end;
 288         unsigned long index;
 289
 290         if (!trans || !trans->transaction) {
 291                 return filemap_write_and_wait(btree_inode->i_mapping);
 292         }
 293         dirty_pages = &trans->transaction->dirty_pages;
 294         while(1) {
 295                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 296                                             EXTENT_DIRTY);
 297                 if (ret)
 298                         break;
 299                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 300                 while(start <= end) {
 301                         index = start >> PAGE_CACHE_SHIFT;
 302                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 303                         page = find_lock_page(btree_inode->i_mapping, index);
 304                         if (!page)
 305                                 continue;
 306                         if (PageWriteback(page)) {
 307                                 if (PageDirty(page))
 308                                         wait_on_page_writeback(page);
 309                                 else {
 310                                         unlock_page(page);
 311                                         page_cache_release(page);
 312                                         continue;
 313                                 }
 314                         }
 315                         err = write_one_page(page, 0);
 316                         if (err)
 317                                 werr = err;
 318                         page_cache_release(page);
 319                 }
 320         }
 321         err = filemap_fdatawait(btree_inode->i_mapping);
 322         if (err)
 323                 werr = err;
 324         return werr;
 325 }
 326
 327 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 328                                struct btrfs_root *root)
 329 {
 330         int ret;
 331         u64 old_root_bytenr;
 332         struct btrfs_root *tree_root = root->fs_info->tree_root;
 333
 334         btrfs_write_dirty_block_groups(trans, root);
 335         while(1) {
 336                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 337                 if (old_root_bytenr == root->node->start)
 338                         break;
 339                 btrfs_set_root_bytenr(&root->root_item,
 340                                        root->node->start);
 341                 btrfs_set_root_level(&root->root_item,
 342                                      btrfs_header_level(root->node));
 343                 ret = btrfs_update_root(trans, tree_root,
 344                                         &root->root_key,
 345                                         &root->root_item);
 346                 BUG_ON(ret);
 347                 btrfs_write_dirty_block_groups(trans, root);
 348         }
 349         return 0;
 350 }
 351
 352 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 353                             struct btrfs_root *root)
 354 {
 355         struct btrfs_fs_info *fs_info = root->fs_info;
 356         struct list_head *next;
 357
 358         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 359                 next = fs_info->dirty_cowonly_roots.next;
 360                 list_del_init(next);
 361                 root = list_entry(next, struct btrfs_root, dirty_list);
 362                 update_cowonly_root(trans, root);
 363         }
 364         return 0;
 365 }
 366
 367 int btrfs_add_dead_root(struct btrfs_root *root,
 368                         struct btrfs_root *latest,
 369                         struct list_head *dead_list)
 370 {
 371         struct btrfs_dirty_root *dirty;
 372
 373         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 374         if (!dirty)
 375                 return -ENOMEM;
 376         dirty->root = root;
 377         dirty->latest_root = latest;
 378         list_add(&dirty->list, dead_list);
 379         return 0;
 380 }
 381
 382 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 383                                     struct radix_tree_root *radix,
 384                                     struct list_head *list)
 385 {
 386         struct btrfs_dirty_root *dirty;
 387         struct btrfs_root *gang[8];
 388         struct btrfs_root *root;
 389         int i;
 390         int ret;
 391         int err = 0;
 392         u32 refs;
 393
 394         while(1) {
 395                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 396                                                  ARRAY_SIZE(gang),
 397                                                  BTRFS_ROOT_TRANS_TAG);
 398                 if (ret == 0)
 399                         break;
 400                 for (i = 0; i < ret; i++) {
 401                         root = gang[i];
 402                         radix_tree_tag_clear(radix,
 403                                      (unsigned long)root->root_key.objectid,
 404                                      BTRFS_ROOT_TRANS_TAG);
 405
 406                         BUG_ON(!root->ref_tree);
 407                         dirty = root->dirty_root;
 408
 409                         if (root->commit_root == root->node) {
 410                                 WARN_ON(root->node->start !=
 411                                         btrfs_root_bytenr(&root->root_item));
 412
 413                                 free_extent_buffer(root->commit_root);
 414                                 root->commit_root = NULL;
 415
 416                                 spin_lock(&root->list_lock);
 417                                 list_del_init(&dirty->root->dead_list);
 418                                 spin_unlock(&root->list_lock);
 419
 420                                 kfree(dirty->root);
 421                                 kfree(dirty);
 422
 423                                 /* make sure to update the root on disk
 424                                  * so we get any updates to the block used
 425                                  * counts
 426                                  */
 427                                 err = btrfs_update_root(trans,
 428                                                 root->fs_info->tree_root,
 429                                                 &root->root_key,
 430                                                 &root->root_item);
 431                                 continue;
 432                         }
 433
 434                         memset(&root->root_item.drop_progress, 0,
 435                                sizeof(struct btrfs_disk_key));
 436                         root->root_item.drop_level = 0;
 437                         root->commit_root = NULL;
 438                         root->root_key.offset = root->fs_info->generation;
 439                         btrfs_set_root_bytenr(&root->root_item,
 440                                               root->node->start);
 441                         btrfs_set_root_level(&root->root_item,
 442                                              btrfs_header_level(root->node));
 443                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 444                                                 &root->root_key,
 445                                                 &root->root_item);
 446                         if (err)
 447                                 break;
 448
 449                         refs = btrfs_root_refs(&dirty->root->root_item);
 450                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 451                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 452                                                 &dirty->root->root_key,
 453                                                 &dirty->root->root_item);
 454
 455                         BUG_ON(err);
 456                         if (refs == 1) {
 457                                 list_add(&dirty->list, list);
 458                         } else {
 459                                 WARN_ON(1);
 460                                 free_extent_buffer(dirty->root->node);
 461                                 kfree(dirty->root);
 462                                 kfree(dirty);
 463                         }
 464                 }
 465         }
 466         return err;
 467 }
 468
 469 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 470 {
 471         struct btrfs_fs_info *info = root->fs_info;
 472         int ret;
 473         struct btrfs_trans_handle *trans;
 474         unsigned long nr;
 475
 476         smp_mb();
 477         if (root->defrag_running)
 478                 return 0;
 479         trans = btrfs_start_transaction(root, 1);
 480         while (1) {
 481                 root->defrag_running = 1;
 482                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 483                 nr = trans->blocks_used;
 484                 btrfs_end_transaction(trans, root);
 485                 btrfs_btree_balance_dirty(info->tree_root, nr);
 486                 cond_resched();
 487
 488                 trans = btrfs_start_transaction(root, 1);
 489                 if (root->fs_info->closing || ret != -EAGAIN)
 490                         break;
 491         }
 492         root->defrag_running = 0;
 493         smp_mb();
 494         btrfs_end_transaction(trans, root);
 495         return 0;
 496 }
 497
 498 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 499                                      struct list_head *list)
 500 {
 501         struct btrfs_dirty_root *dirty;
 502         struct btrfs_trans_handle *trans;
 503         unsigned long nr;
 504         u64 num_bytes;
 505         u64 bytes_used;
 506         u64 max_useless;
 507         int ret = 0;
 508         int err;
 509
 510         while(!list_empty(list)) {
 511                 struct btrfs_root *root;
 512
 513                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 514                 list_del_init(&dirty->list);
 515
 516                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 517                 root = dirty->latest_root;
 518                 atomic_inc(&root->fs_info->throttles);
 519
 520                 mutex_lock(&root->fs_info->drop_mutex);
 521                 while(1) {
 522                         trans = btrfs_start_transaction(tree_root, 1);
 523                         ret = btrfs_drop_snapshot(trans, dirty->root);
 524                         if (ret != -EAGAIN) {
 525                                 break;
 526                         }
 527
 528                         err = btrfs_update_root(trans,
 529                                         tree_root,
 530                                         &dirty->root->root_key,
 531                                         &dirty->root->root_item);
 532                         if (err)
 533                                 ret = err;
 534                         nr = trans->blocks_used;
 535                         ret = btrfs_end_transaction(trans, tree_root);
 536                         BUG_ON(ret);
 537
 538                         mutex_unlock(&root->fs_info->drop_mutex);
 539                         btrfs_btree_balance_dirty(tree_root, nr);
 540                         cond_resched();
 541                         mutex_lock(&root->fs_info->drop_mutex);
 542                 }
 543                 BUG_ON(ret);
 544                 atomic_dec(&root->fs_info->throttles);
 545                 wake_up(&root->fs_info->transaction_throttle);
 546
 547                 mutex_lock(&root->fs_info->alloc_mutex);
 548                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 549                 bytes_used = btrfs_root_used(&root->root_item);
 550                 if (num_bytes) {
 551                         record_root_in_trans(root);
 552                         btrfs_set_root_used(&root->root_item,
 553                                             bytes_used - num_bytes);
 554                 }
 555                 mutex_unlock(&root->fs_info->alloc_mutex);
 556
 557                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 558                 if (ret) {
 559                         BUG();
 560                         break;
 561                 }
 562                 mutex_unlock(&root->fs_info->drop_mutex);
 563
 564                 spin_lock(&root->list_lock);
 565                 list_del_init(&dirty->root->dead_list);
 566                 if (!list_empty(&root->dead_list)) {
 567                         struct btrfs_root *oldest;
 568                         oldest = list_entry(root->dead_list.prev,
 569                                             struct btrfs_root, dead_list);
 570                         max_useless = oldest->root_key.offset - 1;
 571                 } else {
 572                         max_useless = root->root_key.offset - 1;
 573                 }
 574                 spin_unlock(&root->list_lock);
 575
 576                 nr = trans->blocks_used;
 577                 ret = btrfs_end_transaction(trans, tree_root);
 578                 BUG_ON(ret);
 579
 580                 ret = btrfs_remove_leaf_refs(root, max_useless);
 581                 BUG_ON(ret);
 582
 583                 free_extent_buffer(dirty->root->node);
 584                 kfree(dirty->root);
 585                 kfree(dirty);
 586
 587                 btrfs_btree_balance_dirty(tree_root, nr);
 588                 cond_resched();
 589         }
 590         return ret;
 591 }
 592
 593 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 594                                    struct btrfs_fs_info *fs_info,
 595                                    struct btrfs_pending_snapshot *pending)
 596 {
 597         struct btrfs_key key;
 598         struct btrfs_root_item *new_root_item;
 599         struct btrfs_root *tree_root = fs_info->tree_root;
 600         struct btrfs_root *root = pending->root;
 601         struct extent_buffer *tmp;
 602         struct extent_buffer *old;
 603         int ret;
 604         int namelen;
 605         u64 objectid;
 606
 607         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 608         if (!new_root_item) {
 609                 ret = -ENOMEM;
 610                 goto fail;
 611         }
 612         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 613         if (ret)
 614                 goto fail;
 615
 616         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 617
 618         key.objectid = objectid;
 619         key.offset = 1;
 620         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 621
 622         old = btrfs_lock_root_node(root);
 623         btrfs_cow_block(trans, root, old, NULL, 0, &old);
 624
 625         btrfs_copy_root(trans, root, old, &tmp, objectid);
 626         btrfs_tree_unlock(old);
 627         free_extent_buffer(old);
 628
 629         btrfs_set_root_bytenr(new_root_item, tmp->start);
 630         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 631         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 632                                 new_root_item);
 633         btrfs_tree_unlock(tmp);
 634         free_extent_buffer(tmp);
 635         if (ret)
 636                 goto fail;
 637
 638         /*
 639          * insert the directory item
 640          */
 641         key.offset = (u64)-1;
 642         namelen = strlen(pending->name);
 643         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 644                                     pending->name, namelen,
 645                                     root->fs_info->sb->s_root->d_inode->i_ino,
 646                                     &key, BTRFS_FT_DIR, 0);
 647
 648         if (ret)
 649                 goto fail;
 650
 651         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 652                              pending->name, strlen(pending->name), objectid,
 653                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 654
 655         /* Invalidate existing dcache entry for new snapshot. */
 656         btrfs_invalidate_dcache_root(root, pending->name, namelen);
 657
 658 fail:
 659         kfree(new_root_item);
 660         return ret;
 661 }
 662
 663 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 664                                              struct btrfs_fs_info *fs_info)
 665 {
 666         struct btrfs_pending_snapshot *pending;
 667         struct list_head *head = &trans->transaction->pending_snapshots;
 668         int ret;
 669
 670         while(!list_empty(head)) {
 671                 pending = list_entry(head->next,
 672                                      struct btrfs_pending_snapshot, list);
 673                 ret = create_pending_snapshot(trans, fs_info, pending);
 674                 BUG_ON(ret);
 675                 list_del(&pending->list);
 676                 kfree(pending->name);
 677                 kfree(pending);
 678         }
 679         return 0;
 680 }
 681
 682 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 683                              struct btrfs_root *root)
 684 {
 685         unsigned long joined = 0;
 686         unsigned long timeout = 1;
 687         struct btrfs_transaction *cur_trans;
 688         struct btrfs_transaction *prev_trans = NULL;
 689         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 690         struct list_head dirty_fs_roots;
 691         struct extent_io_tree *pinned_copy;
 692         DEFINE_WAIT(wait);
 693         int ret;
 694
 695         INIT_LIST_HEAD(&dirty_fs_roots);
 696
 697         mutex_lock(&root->fs_info->trans_mutex);
 698         if (trans->transaction->in_commit) {
 699                 cur_trans = trans->transaction;
 700                 trans->transaction->use_count++;
 701                 mutex_unlock(&root->fs_info->trans_mutex);
 702                 btrfs_end_transaction(trans, root);
 703
 704                 ret = wait_for_commit(root, cur_trans);
 705                 BUG_ON(ret);
 706
 707                 mutex_lock(&root->fs_info->trans_mutex);
 708                 put_transaction(cur_trans);
 709                 mutex_unlock(&root->fs_info->trans_mutex);
 710
 711                 return 0;
 712         }
 713
 714         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 715         if (!pinned_copy)
 716                 return -ENOMEM;
 717
 718         extent_io_tree_init(pinned_copy,
 719                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 720
 721         trans->transaction->in_commit = 1;
 722         trans->transaction->blocked = 1;
 723         cur_trans = trans->transaction;
 724         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 725                 prev_trans = list_entry(cur_trans->list.prev,
 726                                         struct btrfs_transaction, list);
 727                 if (!prev_trans->commit_done) {
 728                         prev_trans->use_count++;
 729                         mutex_unlock(&root->fs_info->trans_mutex);
 730
 731                         wait_for_commit(root, prev_trans);
 732
 733                         mutex_lock(&root->fs_info->trans_mutex);
 734                         put_transaction(prev_trans);
 735                 }
 736         }
 737
 738         do {
 739                 joined = cur_trans->num_joined;
 740                 WARN_ON(cur_trans != trans->transaction);
 741                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 742                                 TASK_UNINTERRUPTIBLE);
 743
 744                 if (cur_trans->num_writers > 1)
 745                         timeout = MAX_SCHEDULE_TIMEOUT;
 746                 else
 747                         timeout = 1;
 748
 749                 mutex_unlock(&root->fs_info->trans_mutex);
 750
 751                 schedule_timeout(timeout);
 752
 753                 mutex_lock(&root->fs_info->trans_mutex);
 754                 finish_wait(&cur_trans->writer_wait, &wait);
 755         } while (cur_trans->num_writers > 1 ||
 756                  (cur_trans->num_joined != joined));
 757
 758         ret = create_pending_snapshots(trans, root->fs_info);
 759         BUG_ON(ret);
 760
 761         WARN_ON(cur_trans != trans->transaction);
 762
 763         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 764                               &dirty_fs_roots);
 765         BUG_ON(ret);
 766
 767         ret = btrfs_commit_tree_roots(trans, root);
 768         BUG_ON(ret);
 769
 770         cur_trans = root->fs_info->running_transaction;
 771         spin_lock(&root->fs_info->new_trans_lock);
 772         root->fs_info->running_transaction = NULL;
 773         spin_unlock(&root->fs_info->new_trans_lock);
 774         btrfs_set_super_generation(&root->fs_info->super_copy,
 775                                    cur_trans->transid);
 776         btrfs_set_super_root(&root->fs_info->super_copy,
 777                              root->fs_info->tree_root->node->start);
 778         btrfs_set_super_root_level(&root->fs_info->super_copy,
 779                            btrfs_header_level(root->fs_info->tree_root->node));
 780
 781         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 782                                    chunk_root->node->start);
 783         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 784                                          btrfs_header_level(chunk_root->node));
 785         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 786                sizeof(root->fs_info->super_copy));
 787
 788         btrfs_copy_pinned(root, pinned_copy);
 789
 790         trans->transaction->blocked = 0;
 791         wake_up(&root->fs_info->transaction_throttle);
 792         wake_up(&root->fs_info->transaction_wait);
 793
 794         mutex_unlock(&root->fs_info->trans_mutex);
 795         ret = btrfs_write_and_wait_transaction(trans, root);
 796         BUG_ON(ret);
 797         write_ctree_super(trans, root);
 798
 799         btrfs_finish_extent_commit(trans, root, pinned_copy);
 800         mutex_lock(&root->fs_info->trans_mutex);
 801
 802         kfree(pinned_copy);
 803
 804         cur_trans->commit_done = 1;
 805         root->fs_info->last_trans_committed = cur_trans->transid;
 806         wake_up(&cur_trans->commit_wait);
 807         put_transaction(cur_trans);
 808         put_transaction(cur_trans);
 809
 810         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 811         if (root->fs_info->closing)
 812                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 813
 814         mutex_unlock(&root->fs_info->trans_mutex);
 815         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 816
 817         if (root->fs_info->closing) {
 818                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 819         }
 820         return ret;
 821 }
 822
 823 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 824 {
 825         struct list_head dirty_roots;
 826         INIT_LIST_HEAD(&dirty_roots);
 827 again:
 828         mutex_lock(&root->fs_info->trans_mutex);
 829         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 830         mutex_unlock(&root->fs_info->trans_mutex);
 831
 832         if (!list_empty(&dirty_roots)) {
 833                 drop_dirty_roots(root, &dirty_roots);
 834                 goto again;
 835         }
 836         return 0;
 837 }