git.oblomov.eu Git - linux-2.6/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 static void wait_current_trans(struct btrfs_root *root)
 128 {
 129         struct btrfs_transaction *cur_trans;
 130
 131         cur_trans = root->fs_info->running_transaction;
 132         if (cur_trans && cur_trans->blocked) {
 133                 DEFINE_WAIT(wait);
 134                 cur_trans->use_count++;
 135                 while(1) {
 136                         prepare_to_wait(&root->fs_info->transaction_wait, &wait,
 137                                         TASK_UNINTERRUPTIBLE);
 138                         if (cur_trans->blocked) {
 139                                 mutex_unlock(&root->fs_info->trans_mutex);
 140                                 schedule();
 141                                 mutex_lock(&root->fs_info->trans_mutex);
 142                                 finish_wait(&root->fs_info->transaction_wait,
 143                                             &wait);
 144                         } else {
 145                                 finish_wait(&root->fs_info->transaction_wait,
 146                                             &wait);
 147                                 break;
 148                         }
 149                 }
 150                 put_transaction(cur_trans);
 151         }
 152 }
 153
 154 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 155                                              int num_blocks, int wait)
 156 {
 157         struct btrfs_trans_handle *h =
 158                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 159         int ret;
 160
 161         mutex_lock(&root->fs_info->trans_mutex);
 162         if ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)
 163                 wait_current_trans(root);
 164         ret = join_transaction(root);
 165         BUG_ON(ret);
 166
 167         record_root_in_trans(root);
 168         h->transid = root->fs_info->running_transaction->transid;
 169         h->transaction = root->fs_info->running_transaction;
 170         h->blocks_reserved = num_blocks;
 171         h->blocks_used = 0;
 172         h->block_group = NULL;
 173         h->alloc_exclude_nr = 0;
 174         h->alloc_exclude_start = 0;
 175         root->fs_info->running_transaction->use_count++;
 176         mutex_unlock(&root->fs_info->trans_mutex);
 177         return h;
 178 }
 179
 180 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 181                                                    int num_blocks)
 182 {
 183         return start_transaction(root, num_blocks, 1);
 184 }
 185 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 186                                                    int num_blocks)
 187 {
 188         return start_transaction(root, num_blocks, 0);
 189 }
 190
 191 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
 192                                                          int num_blocks)
 193 {
 194         return start_transaction(r, num_blocks, 2);
 195 }
 196
 197
 198 static noinline int wait_for_commit(struct btrfs_root *root,
 199                                     struct btrfs_transaction *commit)
 200 {
 201         DEFINE_WAIT(wait);
 202         mutex_lock(&root->fs_info->trans_mutex);
 203         while(!commit->commit_done) {
 204                 prepare_to_wait(&commit->commit_wait, &wait,
 205                                 TASK_UNINTERRUPTIBLE);
 206                 if (commit->commit_done)
 207                         break;
 208                 mutex_unlock(&root->fs_info->trans_mutex);
 209                 schedule();
 210                 mutex_lock(&root->fs_info->trans_mutex);
 211         }
 212         mutex_unlock(&root->fs_info->trans_mutex);
 213         finish_wait(&commit->commit_wait, &wait);
 214         return 0;
 215 }
 216
 217 static void throttle_on_drops(struct btrfs_root *root)
 218 {
 219         struct btrfs_fs_info *info = root->fs_info;
 220         int harder_count = 0;
 221
 222 harder:
 223         if (atomic_read(&info->throttles)) {
 224                 DEFINE_WAIT(wait);
 225                 int thr;
 226                 thr = atomic_read(&info->throttle_gen);
 227
 228                 do {
 229                         prepare_to_wait(&info->transaction_throttle,
 230                                         &wait, TASK_UNINTERRUPTIBLE);
 231                         if (!atomic_read(&info->throttles)) {
 232                                 finish_wait(&info->transaction_throttle, &wait);
 233                                 break;
 234                         }
 235                         schedule();
 236                         finish_wait(&info->transaction_throttle, &wait);
 237                 } while (thr == atomic_read(&info->throttle_gen));
 238                 harder_count++;
 239
 240                 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
 241                     harder_count < 2)
 242                         goto harder;
 243
 244                 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
 245                     harder_count < 10)
 246                         goto harder;
 247
 248                 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
 249                     harder_count < 20)
 250                         goto harder;
 251         }
 252 }
 253
 254 void btrfs_throttle(struct btrfs_root *root)
 255 {
 256         mutex_lock(&root->fs_info->trans_mutex);
 257         if (!root->fs_info->open_ioctl_trans)
 258                 wait_current_trans(root);
 259         mutex_unlock(&root->fs_info->trans_mutex);
 260
 261         throttle_on_drops(root);
 262 }
 263
 264 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 265                           struct btrfs_root *root, int throttle)
 266 {
 267         struct btrfs_transaction *cur_trans;
 268         struct btrfs_fs_info *info = root->fs_info;
 269
 270         mutex_lock(&info->trans_mutex);
 271         cur_trans = info->running_transaction;
 272         WARN_ON(cur_trans != trans->transaction);
 273         WARN_ON(cur_trans->num_writers < 1);
 274         cur_trans->num_writers--;
 275
 276         if (waitqueue_active(&cur_trans->writer_wait))
 277                 wake_up(&cur_trans->writer_wait);
 278         put_transaction(cur_trans);
 279         mutex_unlock(&info->trans_mutex);
 280         memset(trans, 0, sizeof(*trans));
 281         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 282
 283         if (throttle)
 284                 throttle_on_drops(root);
 285
 286         return 0;
 287 }
 288
 289 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 290                           struct btrfs_root *root)
 291 {
 292         return __btrfs_end_transaction(trans, root, 0);
 293 }
 294
 295 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 296                                    struct btrfs_root *root)
 297 {
 298         return __btrfs_end_transaction(trans, root, 1);
 299 }
 300
 301
 302 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 303                                      struct btrfs_root *root)
 304 {
 305         int ret;
 306         int err = 0;
 307         int werr = 0;
 308         struct extent_io_tree *dirty_pages;
 309         struct page *page;
 310         struct inode *btree_inode = root->fs_info->btree_inode;
 311         u64 start = 0;
 312         u64 end;
 313         unsigned long index;
 314
 315         if (!trans || !trans->transaction) {
 316                 return filemap_write_and_wait(btree_inode->i_mapping);
 317         }
 318         dirty_pages = &trans->transaction->dirty_pages;
 319         while(1) {
 320                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
 321                                             EXTENT_DIRTY);
 322                 if (ret)
 323                         break;
 324                 while(start <= end) {
 325                         cond_resched();
 326
 327                         index = start >> PAGE_CACHE_SHIFT;
 328                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 329                         page = find_lock_page(btree_inode->i_mapping, index);
 330                         if (!page)
 331                                 continue;
 332                         if (PageWriteback(page)) {
 333                                 if (PageDirty(page))
 334                                         wait_on_page_writeback(page);
 335                                 else {
 336                                         unlock_page(page);
 337                                         page_cache_release(page);
 338                                         continue;
 339                                 }
 340                         }
 341                         err = write_one_page(page, 0);
 342                         if (err)
 343                                 werr = err;
 344                         page_cache_release(page);
 345                 }
 346         }
 347         while(1) {
 348                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 349                                             EXTENT_DIRTY);
 350                 if (ret)
 351                         break;
 352
 353                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 354                 while(start <= end) {
 355                         index = start >> PAGE_CACHE_SHIFT;
 356                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 357                         page = find_get_page(btree_inode->i_mapping, index);
 358                         if (!page)
 359                                 continue;
 360                         if (PageDirty(page)) {
 361                                 lock_page(page);
 362                                 err = write_one_page(page, 0);
 363                                 if (err)
 364                                         werr = err;
 365                         }
 366                         wait_on_page_writeback(page);
 367                         page_cache_release(page);
 368                         cond_resched();
 369                 }
 370         }
 371         if (err)
 372                 werr = err;
 373         return werr;
 374 }
 375
 376 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 377                                struct btrfs_root *root)
 378 {
 379         int ret;
 380         u64 old_root_bytenr;
 381         struct btrfs_root *tree_root = root->fs_info->tree_root;
 382
 383         btrfs_write_dirty_block_groups(trans, root);
 384         while(1) {
 385                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 386                 if (old_root_bytenr == root->node->start)
 387                         break;
 388                 btrfs_set_root_bytenr(&root->root_item,
 389                                        root->node->start);
 390                 btrfs_set_root_level(&root->root_item,
 391                                      btrfs_header_level(root->node));
 392                 ret = btrfs_update_root(trans, tree_root,
 393                                         &root->root_key,
 394                                         &root->root_item);
 395                 BUG_ON(ret);
 396                 btrfs_write_dirty_block_groups(trans, root);
 397         }
 398         return 0;
 399 }
 400
 401 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 402                             struct btrfs_root *root)
 403 {
 404         struct btrfs_fs_info *fs_info = root->fs_info;
 405         struct list_head *next;
 406
 407         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 408                 next = fs_info->dirty_cowonly_roots.next;
 409                 list_del_init(next);
 410                 root = list_entry(next, struct btrfs_root, dirty_list);
 411                 update_cowonly_root(trans, root);
 412         }
 413         return 0;
 414 }
 415
 416 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
 417 {
 418         struct btrfs_dirty_root *dirty;
 419
 420         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 421         if (!dirty)
 422                 return -ENOMEM;
 423         dirty->root = root;
 424         dirty->latest_root = latest;
 425
 426         mutex_lock(&root->fs_info->trans_mutex);
 427         list_add(&dirty->list, &latest->fs_info->dead_roots);
 428         mutex_unlock(&root->fs_info->trans_mutex);
 429         return 0;
 430 }
 431
 432 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 433                                     struct radix_tree_root *radix,
 434                                     struct list_head *list)
 435 {
 436         struct btrfs_dirty_root *dirty;
 437         struct btrfs_root *gang[8];
 438         struct btrfs_root *root;
 439         int i;
 440         int ret;
 441         int err = 0;
 442         u32 refs;
 443
 444         while(1) {
 445                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 446                                                  ARRAY_SIZE(gang),
 447                                                  BTRFS_ROOT_TRANS_TAG);
 448                 if (ret == 0)
 449                         break;
 450                 for (i = 0; i < ret; i++) {
 451                         root = gang[i];
 452                         radix_tree_tag_clear(radix,
 453                                      (unsigned long)root->root_key.objectid,
 454                                      BTRFS_ROOT_TRANS_TAG);
 455
 456                         BUG_ON(!root->ref_tree);
 457                         dirty = root->dirty_root;
 458
 459                         if (root->commit_root == root->node) {
 460                                 WARN_ON(root->node->start !=
 461                                         btrfs_root_bytenr(&root->root_item));
 462
 463                                 free_extent_buffer(root->commit_root);
 464                                 root->commit_root = NULL;
 465                                 root->dirty_root = NULL;
 466
 467                                 spin_lock(&root->list_lock);
 468                                 list_del_init(&dirty->root->dead_list);
 469                                 spin_unlock(&root->list_lock);
 470
 471                                 kfree(dirty->root);
 472                                 kfree(dirty);
 473
 474                                 /* make sure to update the root on disk
 475                                  * so we get any updates to the block used
 476                                  * counts
 477                                  */
 478                                 err = btrfs_update_root(trans,
 479                                                 root->fs_info->tree_root,
 480                                                 &root->root_key,
 481                                                 &root->root_item);
 482                                 continue;
 483                         }
 484
 485                         memset(&root->root_item.drop_progress, 0,
 486                                sizeof(struct btrfs_disk_key));
 487                         root->root_item.drop_level = 0;
 488                         root->commit_root = NULL;
 489                         root->dirty_root = NULL;
 490                         root->root_key.offset = root->fs_info->generation;
 491                         btrfs_set_root_bytenr(&root->root_item,
 492                                               root->node->start);
 493                         btrfs_set_root_level(&root->root_item,
 494                                              btrfs_header_level(root->node));
 495                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 496                                                 &root->root_key,
 497                                                 &root->root_item);
 498                         if (err)
 499                                 break;
 500
 501                         refs = btrfs_root_refs(&dirty->root->root_item);
 502                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 503                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 504                                                 &dirty->root->root_key,
 505                                                 &dirty->root->root_item);
 506
 507                         BUG_ON(err);
 508                         if (refs == 1) {
 509                                 list_add(&dirty->list, list);
 510                         } else {
 511                                 WARN_ON(1);
 512                                 free_extent_buffer(dirty->root->node);
 513                                 kfree(dirty->root);
 514                                 kfree(dirty);
 515                         }
 516                 }
 517         }
 518         return err;
 519 }
 520
 521 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 522 {
 523         struct btrfs_fs_info *info = root->fs_info;
 524         int ret;
 525         struct btrfs_trans_handle *trans;
 526         unsigned long nr;
 527
 528         smp_mb();
 529         if (root->defrag_running)
 530                 return 0;
 531         trans = btrfs_start_transaction(root, 1);
 532         while (1) {
 533                 root->defrag_running = 1;
 534                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 535                 nr = trans->blocks_used;
 536                 btrfs_end_transaction(trans, root);
 537                 btrfs_btree_balance_dirty(info->tree_root, nr);
 538                 cond_resched();
 539
 540                 trans = btrfs_start_transaction(root, 1);
 541                 if (root->fs_info->closing || ret != -EAGAIN)
 542                         break;
 543         }
 544         root->defrag_running = 0;
 545         smp_mb();
 546         btrfs_end_transaction(trans, root);
 547         return 0;
 548 }
 549
 550 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 551                                      struct list_head *list)
 552 {
 553         struct btrfs_dirty_root *dirty;
 554         struct btrfs_trans_handle *trans;
 555         unsigned long nr;
 556         u64 num_bytes;
 557         u64 bytes_used;
 558         u64 max_useless;
 559         int ret = 0;
 560         int err;
 561
 562         while(!list_empty(list)) {
 563                 struct btrfs_root *root;
 564
 565                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 566                 list_del_init(&dirty->list);
 567
 568                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 569                 root = dirty->latest_root;
 570                 atomic_inc(&root->fs_info->throttles);
 571
 572                 mutex_lock(&root->fs_info->drop_mutex);
 573                 while(1) {
 574                         trans = btrfs_start_transaction(tree_root, 1);
 575                         ret = btrfs_drop_snapshot(trans, dirty->root);
 576                         if (ret != -EAGAIN) {
 577                                 break;
 578                         }
 579
 580                         err = btrfs_update_root(trans,
 581                                         tree_root,
 582                                         &dirty->root->root_key,
 583                                         &dirty->root->root_item);
 584                         if (err)
 585                                 ret = err;
 586                         nr = trans->blocks_used;
 587                         ret = btrfs_end_transaction(trans, tree_root);
 588                         BUG_ON(ret);
 589
 590                         mutex_unlock(&root->fs_info->drop_mutex);
 591                         btrfs_btree_balance_dirty(tree_root, nr);
 592                         cond_resched();
 593                         mutex_lock(&root->fs_info->drop_mutex);
 594                 }
 595                 BUG_ON(ret);
 596                 atomic_dec(&root->fs_info->throttles);
 597                 wake_up(&root->fs_info->transaction_throttle);
 598
 599                 mutex_lock(&root->fs_info->alloc_mutex);
 600                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 601                 bytes_used = btrfs_root_used(&root->root_item);
 602                 if (num_bytes) {
 603                         record_root_in_trans(root);
 604                         btrfs_set_root_used(&root->root_item,
 605                                             bytes_used - num_bytes);
 606                 }
 607                 mutex_unlock(&root->fs_info->alloc_mutex);
 608
 609                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 610                 if (ret) {
 611                         BUG();
 612                         break;
 613                 }
 614                 mutex_unlock(&root->fs_info->drop_mutex);
 615
 616                 spin_lock(&root->list_lock);
 617                 list_del_init(&dirty->root->dead_list);
 618                 if (!list_empty(&root->dead_list)) {
 619                         struct btrfs_root *oldest;
 620                         oldest = list_entry(root->dead_list.prev,
 621                                             struct btrfs_root, dead_list);
 622                         max_useless = oldest->root_key.offset - 1;
 623                 } else {
 624                         max_useless = root->root_key.offset - 1;
 625                 }
 626                 spin_unlock(&root->list_lock);
 627
 628                 nr = trans->blocks_used;
 629                 ret = btrfs_end_transaction(trans, tree_root);
 630                 BUG_ON(ret);
 631
 632                 ret = btrfs_remove_leaf_refs(root, max_useless);
 633                 BUG_ON(ret);
 634
 635                 free_extent_buffer(dirty->root->node);
 636                 kfree(dirty->root);
 637                 kfree(dirty);
 638
 639                 btrfs_btree_balance_dirty(tree_root, nr);
 640                 cond_resched();
 641         }
 642         return ret;
 643 }
 644
 645 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 646                                    struct btrfs_fs_info *fs_info,
 647                                    struct btrfs_pending_snapshot *pending)
 648 {
 649         struct btrfs_key key;
 650         struct btrfs_root_item *new_root_item;
 651         struct btrfs_root *tree_root = fs_info->tree_root;
 652         struct btrfs_root *root = pending->root;
 653         struct extent_buffer *tmp;
 654         struct extent_buffer *old;
 655         int ret;
 656         int namelen;
 657         u64 objectid;
 658
 659         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 660         if (!new_root_item) {
 661                 ret = -ENOMEM;
 662                 goto fail;
 663         }
 664         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 665         if (ret)
 666                 goto fail;
 667
 668         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 669
 670         key.objectid = objectid;
 671         key.offset = 1;
 672         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 673
 674         old = btrfs_lock_root_node(root);
 675         btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
 676
 677         btrfs_copy_root(trans, root, old, &tmp, objectid);
 678         btrfs_tree_unlock(old);
 679         free_extent_buffer(old);
 680
 681         btrfs_set_root_bytenr(new_root_item, tmp->start);
 682         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 683         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 684                                 new_root_item);
 685         btrfs_tree_unlock(tmp);
 686         free_extent_buffer(tmp);
 687         if (ret)
 688                 goto fail;
 689
 690         /*
 691          * insert the directory item
 692          */
 693         key.offset = (u64)-1;
 694         namelen = strlen(pending->name);
 695         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 696                                     pending->name, namelen,
 697                                     root->fs_info->sb->s_root->d_inode->i_ino,
 698                                     &key, BTRFS_FT_DIR, 0);
 699
 700         if (ret)
 701                 goto fail;
 702
 703         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 704                              pending->name, strlen(pending->name), objectid,
 705                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 706
 707         /* Invalidate existing dcache entry for new snapshot. */
 708         btrfs_invalidate_dcache_root(root, pending->name, namelen);
 709
 710 fail:
 711         kfree(new_root_item);
 712         return ret;
 713 }
 714
 715 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 716                                              struct btrfs_fs_info *fs_info)
 717 {
 718         struct btrfs_pending_snapshot *pending;
 719         struct list_head *head = &trans->transaction->pending_snapshots;
 720         int ret;
 721
 722         while(!list_empty(head)) {
 723                 pending = list_entry(head->next,
 724                                      struct btrfs_pending_snapshot, list);
 725                 ret = create_pending_snapshot(trans, fs_info, pending);
 726                 BUG_ON(ret);
 727                 list_del(&pending->list);
 728                 kfree(pending->name);
 729                 kfree(pending);
 730         }
 731         return 0;
 732 }
 733
 734 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 735                              struct btrfs_root *root)
 736 {
 737         unsigned long joined = 0;
 738         unsigned long timeout = 1;
 739         struct btrfs_transaction *cur_trans;
 740         struct btrfs_transaction *prev_trans = NULL;
 741         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 742         struct list_head dirty_fs_roots;
 743         struct extent_io_tree *pinned_copy;
 744         DEFINE_WAIT(wait);
 745         int ret;
 746
 747         INIT_LIST_HEAD(&dirty_fs_roots);
 748
 749         mutex_lock(&root->fs_info->trans_mutex);
 750         if (trans->transaction->in_commit) {
 751                 cur_trans = trans->transaction;
 752                 trans->transaction->use_count++;
 753                 mutex_unlock(&root->fs_info->trans_mutex);
 754                 btrfs_end_transaction(trans, root);
 755
 756                 ret = wait_for_commit(root, cur_trans);
 757                 BUG_ON(ret);
 758
 759                 mutex_lock(&root->fs_info->trans_mutex);
 760                 put_transaction(cur_trans);
 761                 mutex_unlock(&root->fs_info->trans_mutex);
 762
 763                 return 0;
 764         }
 765
 766         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 767         if (!pinned_copy)
 768                 return -ENOMEM;
 769
 770         extent_io_tree_init(pinned_copy,
 771                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 772
 773         trans->transaction->in_commit = 1;
 774         trans->transaction->blocked = 1;
 775         cur_trans = trans->transaction;
 776         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 777                 prev_trans = list_entry(cur_trans->list.prev,
 778                                         struct btrfs_transaction, list);
 779                 if (!prev_trans->commit_done) {
 780                         prev_trans->use_count++;
 781                         mutex_unlock(&root->fs_info->trans_mutex);
 782
 783                         wait_for_commit(root, prev_trans);
 784
 785                         mutex_lock(&root->fs_info->trans_mutex);
 786                         put_transaction(prev_trans);
 787                 }
 788         }
 789
 790         do {
 791                 int snap_pending = 0;
 792                 joined = cur_trans->num_joined;
 793                 if (!list_empty(&trans->transaction->pending_snapshots))
 794                         snap_pending = 1;
 795
 796                 WARN_ON(cur_trans != trans->transaction);
 797                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 798                                 TASK_UNINTERRUPTIBLE);
 799
 800                 if (cur_trans->num_writers > 1)
 801                         timeout = MAX_SCHEDULE_TIMEOUT;
 802                 else
 803                         timeout = 1;
 804
 805                 mutex_unlock(&root->fs_info->trans_mutex);
 806
 807                 if (snap_pending) {
 808                         ret = btrfs_wait_ordered_extents(root, 1);
 809                         BUG_ON(ret);
 810                 }
 811
 812                 schedule_timeout(timeout);
 813
 814                 mutex_lock(&root->fs_info->trans_mutex);
 815                 finish_wait(&cur_trans->writer_wait, &wait);
 816         } while (cur_trans->num_writers > 1 ||
 817                  (cur_trans->num_joined != joined));
 818
 819         ret = create_pending_snapshots(trans, root->fs_info);
 820         BUG_ON(ret);
 821
 822         WARN_ON(cur_trans != trans->transaction);
 823
 824         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 825                               &dirty_fs_roots);
 826         BUG_ON(ret);
 827
 828         ret = btrfs_commit_tree_roots(trans, root);
 829         BUG_ON(ret);
 830
 831         cur_trans = root->fs_info->running_transaction;
 832         spin_lock(&root->fs_info->new_trans_lock);
 833         root->fs_info->running_transaction = NULL;
 834         spin_unlock(&root->fs_info->new_trans_lock);
 835         btrfs_set_super_generation(&root->fs_info->super_copy,
 836                                    cur_trans->transid);
 837         btrfs_set_super_root(&root->fs_info->super_copy,
 838                              root->fs_info->tree_root->node->start);
 839         btrfs_set_super_root_level(&root->fs_info->super_copy,
 840                            btrfs_header_level(root->fs_info->tree_root->node));
 841
 842         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 843                                    chunk_root->node->start);
 844         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 845                                          btrfs_header_level(chunk_root->node));
 846         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 847                sizeof(root->fs_info->super_copy));
 848
 849         btrfs_copy_pinned(root, pinned_copy);
 850
 851         trans->transaction->blocked = 0;
 852         wake_up(&root->fs_info->transaction_throttle);
 853         wake_up(&root->fs_info->transaction_wait);
 854
 855         mutex_unlock(&root->fs_info->trans_mutex);
 856         ret = btrfs_write_and_wait_transaction(trans, root);
 857         BUG_ON(ret);
 858         write_ctree_super(trans, root);
 859
 860         btrfs_finish_extent_commit(trans, root, pinned_copy);
 861         mutex_lock(&root->fs_info->trans_mutex);
 862
 863         kfree(pinned_copy);
 864
 865         cur_trans->commit_done = 1;
 866         root->fs_info->last_trans_committed = cur_trans->transid;
 867         wake_up(&cur_trans->commit_wait);
 868         put_transaction(cur_trans);
 869         put_transaction(cur_trans);
 870
 871         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 872         if (root->fs_info->closing)
 873                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 874
 875         mutex_unlock(&root->fs_info->trans_mutex);
 876         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 877
 878         if (root->fs_info->closing) {
 879                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 880         }
 881         return ret;
 882 }
 883
 884 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 885 {
 886         struct list_head dirty_roots;
 887         INIT_LIST_HEAD(&dirty_roots);
 888 again:
 889         mutex_lock(&root->fs_info->trans_mutex);
 890         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 891         mutex_unlock(&root->fs_info->trans_mutex);
 892
 893         if (!list_empty(&dirty_roots)) {
 894                 drop_dirty_roots(root, &dirty_roots);
 895                 goto again;
 896         }
 897         return 0;
 898 }