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