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 "ctree.h"
  22 #include "disk-io.h"
  23 #include "transaction.h"
  24
  25 static int total_trans = 0;
  26 extern struct kmem_cache *btrfs_trans_handle_cachep;
  27 extern struct kmem_cache *btrfs_transaction_cachep;
  28
  29 static struct workqueue_struct *trans_wq;
  30
  31 #define BTRFS_ROOT_TRANS_TAG 0
  32 #define BTRFS_ROOT_DEFRAG_TAG 1
  33
  34 static void put_transaction(struct btrfs_transaction *transaction)
  35 {
  36         WARN_ON(transaction->use_count == 0);
  37         transaction->use_count--;
  38         if (transaction->use_count == 0) {
  39                 WARN_ON(total_trans == 0);
  40                 total_trans--;
  41                 list_del_init(&transaction->list);
  42                 memset(transaction, 0, sizeof(*transaction));
  43                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  44         }
  45 }
  46
  47 static int join_transaction(struct btrfs_root *root)
  48 {
  49         struct btrfs_transaction *cur_trans;
  50         cur_trans = root->fs_info->running_transaction;
  51         if (!cur_trans) {
  52                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  53                                              GFP_NOFS);
  54                 total_trans++;
  55                 BUG_ON(!cur_trans);
  56                 root->fs_info->generation++;
  57                 root->fs_info->running_transaction = cur_trans;
  58                 cur_trans->num_writers = 0;
  59                 cur_trans->transid = root->fs_info->generation;
  60                 init_waitqueue_head(&cur_trans->writer_wait);
  61                 init_waitqueue_head(&cur_trans->commit_wait);
  62                 cur_trans->in_commit = 0;
  63                 cur_trans->use_count = 1;
  64                 cur_trans->commit_done = 0;
  65                 cur_trans->start_time = get_seconds();
  66                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  67                 init_bit_radix(&cur_trans->dirty_pages);
  68         }
  69         cur_trans->num_writers++;
  70         return 0;
  71 }
  72
  73 static int record_root_in_trans(struct btrfs_root *root)
  74 {
  75         u64 running_trans_id = root->fs_info->running_transaction->transid;
  76         if (root->ref_cows && root->last_trans < running_trans_id) {
  77                 WARN_ON(root == root->fs_info->extent_root);
  78                 if (root->root_item.refs != 0) {
  79                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  80                                    (unsigned long)root->root_key.objectid,
  81                                    BTRFS_ROOT_TRANS_TAG);
  82                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  83                                    (unsigned long)root->root_key.objectid,
  84                                    BTRFS_ROOT_DEFRAG_TAG);
  85                         root->commit_root = root->node;
  86                         get_bh(root->node);
  87                 } else {
  88                         WARN_ON(1);
  89                 }
  90                 root->last_trans = running_trans_id;
  91         }
  92         return 0;
  93 }
  94
  95 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
  96                                                    int num_blocks)
  97 {
  98         struct btrfs_trans_handle *h =
  99                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 100         int ret;
 101
 102         mutex_lock(&root->fs_info->trans_mutex);
 103         ret = join_transaction(root);
 104         BUG_ON(ret);
 105
 106         record_root_in_trans(root);
 107         h->transid = root->fs_info->running_transaction->transid;
 108         h->transaction = root->fs_info->running_transaction;
 109         h->blocks_reserved = num_blocks;
 110         h->blocks_used = 0;
 111         h->block_group = NULL;
 112         root->fs_info->running_transaction->use_count++;
 113         mutex_unlock(&root->fs_info->trans_mutex);
 114         return h;
 115 }
 116
 117 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 118                           struct btrfs_root *root)
 119 {
 120         struct btrfs_transaction *cur_trans;
 121
 122         mutex_lock(&root->fs_info->trans_mutex);
 123         cur_trans = root->fs_info->running_transaction;
 124         WARN_ON(cur_trans != trans->transaction);
 125         WARN_ON(cur_trans->num_writers < 1);
 126         cur_trans->num_writers--;
 127         if (waitqueue_active(&cur_trans->writer_wait))
 128                 wake_up(&cur_trans->writer_wait);
 129         put_transaction(cur_trans);
 130         mutex_unlock(&root->fs_info->trans_mutex);
 131         memset(trans, 0, sizeof(*trans));
 132         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 133         return 0;
 134 }
 135
 136
 137 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 138                                      struct btrfs_root *root)
 139 {
 140         unsigned long gang[16];
 141         int ret;
 142         int i;
 143         int err;
 144         int werr = 0;
 145         struct page *page;
 146         struct radix_tree_root *dirty_pages;
 147         struct inode *btree_inode = root->fs_info->btree_inode;
 148
 149         if (!trans || !trans->transaction) {
 150                 return filemap_write_and_wait(btree_inode->i_mapping);
 151         }
 152         dirty_pages = &trans->transaction->dirty_pages;
 153         while(1) {
 154                 ret = find_first_radix_bit(dirty_pages, gang,
 155                                            0, ARRAY_SIZE(gang));
 156                 if (!ret)
 157                         break;
 158                 for (i = 0; i < ret; i++) {
 159                         /* FIXME EIO */
 160                         clear_radix_bit(dirty_pages, gang[i]);
 161                         page = find_lock_page(btree_inode->i_mapping,
 162                                               gang[i]);
 163                         if (!page)
 164                                 continue;
 165                         if (PageWriteback(page)) {
 166                                 if (PageDirty(page))
 167                                         wait_on_page_writeback(page);
 168                                 else {
 169                                         unlock_page(page);
 170                                         page_cache_release(page);
 171                                         continue;
 172                                 }
 173                         }
 174                         err = write_one_page(page, 0);
 175                         if (err)
 176                                 werr = err;
 177                         page_cache_release(page);
 178                 }
 179         }
 180         err = filemap_fdatawait(btree_inode->i_mapping);
 181         if (err)
 182                 werr = err;
 183         return werr;
 184 }
 185
 186 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 187                             struct btrfs_root *root)
 188 {
 189         int ret;
 190         u64 old_extent_block;
 191         struct btrfs_fs_info *fs_info = root->fs_info;
 192         struct btrfs_root *tree_root = fs_info->tree_root;
 193         struct btrfs_root *extent_root = fs_info->extent_root;
 194
 195         btrfs_write_dirty_block_groups(trans, extent_root);
 196         while(1) {
 197                 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
 198                 if (old_extent_block == bh_blocknr(extent_root->node))
 199                         break;
 200                 btrfs_set_root_blocknr(&extent_root->root_item,
 201                                        bh_blocknr(extent_root->node));
 202                 ret = btrfs_update_root(trans, tree_root,
 203                                         &extent_root->root_key,
 204                                         &extent_root->root_item);
 205                 BUG_ON(ret);
 206                 btrfs_write_dirty_block_groups(trans, extent_root);
 207         }
 208         return 0;
 209 }
 210
 211 static int wait_for_commit(struct btrfs_root *root,
 212                            struct btrfs_transaction *commit)
 213 {
 214         DEFINE_WAIT(wait);
 215         mutex_lock(&root->fs_info->trans_mutex);
 216         while(!commit->commit_done) {
 217                 prepare_to_wait(&commit->commit_wait, &wait,
 218                                 TASK_UNINTERRUPTIBLE);
 219                 if (commit->commit_done)
 220                         break;
 221                 mutex_unlock(&root->fs_info->trans_mutex);
 222                 schedule();
 223                 mutex_lock(&root->fs_info->trans_mutex);
 224         }
 225         mutex_unlock(&root->fs_info->trans_mutex);
 226         finish_wait(&commit->commit_wait, &wait);
 227         return 0;
 228 }
 229
 230 struct dirty_root {
 231         struct list_head list;
 232         struct btrfs_root *root;
 233 };
 234
 235 int btrfs_add_dead_root(struct btrfs_root *root, struct list_head *dead_list)
 236 {
 237         struct dirty_root *dirty;
 238
 239         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 240         if (!dirty)
 241                 return -ENOMEM;
 242         dirty->root = root;
 243         list_add(&dirty->list, dead_list);
 244         return 0;
 245 }
 246
 247 static int add_dirty_roots(struct btrfs_trans_handle *trans,
 248                            struct radix_tree_root *radix,
 249                            struct list_head *list)
 250 {
 251         struct dirty_root *dirty;
 252         struct btrfs_root *gang[8];
 253         struct btrfs_root *root;
 254         int i;
 255         int ret;
 256         int err = 0;
 257         u32 refs;
 258
 259         while(1) {
 260                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 261                                                  ARRAY_SIZE(gang),
 262                                                  BTRFS_ROOT_TRANS_TAG);
 263                 if (ret == 0)
 264                         break;
 265                 for (i = 0; i < ret; i++) {
 266                         root = gang[i];
 267                         radix_tree_tag_clear(radix,
 268                                      (unsigned long)root->root_key.objectid,
 269                                      BTRFS_ROOT_TRANS_TAG);
 270                         if (root->commit_root == root->node) {
 271                                 WARN_ON(bh_blocknr(root->node) !=
 272                                         btrfs_root_blocknr(&root->root_item));
 273                                 brelse(root->commit_root);
 274                                 root->commit_root = NULL;
 275                                 continue;
 276                         }
 277                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 278                         BUG_ON(!dirty);
 279                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 280                         BUG_ON(!dirty->root);
 281
 282                         memset(&root->root_item.drop_progress, 0,
 283                                sizeof(struct btrfs_disk_key));
 284                         root->root_item.drop_level = 0;
 285
 286                         memcpy(dirty->root, root, sizeof(*root));
 287                         dirty->root->node = root->commit_root;
 288                         root->commit_root = NULL;
 289
 290                         root->root_key.offset = root->fs_info->generation;
 291                         btrfs_set_root_blocknr(&root->root_item,
 292                                                bh_blocknr(root->node));
 293                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 294                                                 &root->root_key,
 295                                                 &root->root_item);
 296                         if (err)
 297                                 break;
 298
 299                         refs = btrfs_root_refs(&dirty->root->root_item);
 300                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 301                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 302                                                 &dirty->root->root_key,
 303                                                 &dirty->root->root_item);
 304
 305                         BUG_ON(err);
 306                         if (refs == 1) {
 307                                 list_add(&dirty->list, list);
 308                         } else {
 309                                 WARN_ON(1);
 310                                 kfree(dirty->root);
 311                                 kfree(dirty);
 312                         }
 313                 }
 314         }
 315         return err;
 316 }
 317
 318 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
 319 {
 320         struct btrfs_root *gang[1];
 321         struct btrfs_root *root;
 322         struct btrfs_root *tree_root = info->tree_root;
 323         struct btrfs_trans_handle *trans;
 324         int i;
 325         int ret;
 326         int err = 0;
 327         u64 last = 0;
 328
 329         trans = btrfs_start_transaction(tree_root, 1);
 330         while(1) {
 331                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
 332                                                  (void **)gang, last,
 333                                                  ARRAY_SIZE(gang),
 334                                                  BTRFS_ROOT_DEFRAG_TAG);
 335                 if (ret == 0)
 336                         break;
 337                 for (i = 0; i < ret; i++) {
 338                         root = gang[i];
 339                         last = root->root_key.objectid + 1;
 340                         radix_tree_tag_clear(&info->fs_roots_radix,
 341                                      (unsigned long)root->root_key.objectid,
 342                                      BTRFS_ROOT_DEFRAG_TAG);
 343                         if (root->defrag_running)
 344                                 continue;
 345
 346                         while (1) {
 347                                 mutex_lock(&root->fs_info->trans_mutex);
 348                                 record_root_in_trans(root);
 349                                 mutex_unlock(&root->fs_info->trans_mutex);
 350
 351                                 root->defrag_running = 1;
 352                                 err = btrfs_defrag_leaves(trans, root, 1);
 353                                 btrfs_end_transaction(trans, tree_root);
 354                                 mutex_unlock(&info->fs_mutex);
 355
 356                                 btrfs_btree_balance_dirty(root);
 357                                 cond_resched();
 358
 359                                 mutex_lock(&info->fs_mutex);
 360                                 trans = btrfs_start_transaction(tree_root, 1);
 361                                 if (err != -EAGAIN)
 362                                         break;
 363                         }
 364                         root->defrag_running = 0;
 365                 }
 366         }
 367         btrfs_end_transaction(trans, tree_root);
 368         return err;
 369 }
 370
 371 static int drop_dirty_roots(struct btrfs_root *tree_root,
 372                             struct list_head *list)
 373 {
 374         struct dirty_root *dirty;
 375         struct btrfs_trans_handle *trans;
 376         int ret = 0;
 377         int err;
 378
 379         while(!list_empty(list)) {
 380                 mutex_lock(&tree_root->fs_info->fs_mutex);
 381                 dirty = list_entry(list->next, struct dirty_root, list);
 382                 list_del_init(&dirty->list);
 383
 384                 while(1) {
 385                         trans = btrfs_start_transaction(tree_root, 1);
 386                         ret = btrfs_drop_snapshot(trans, dirty->root);
 387                         if (ret != -EAGAIN) {
 388                                 break;
 389                         }
 390                         err = btrfs_update_root(trans,
 391                                         tree_root,
 392                                         &dirty->root->root_key,
 393                                         &dirty->root->root_item);
 394                         if (err)
 395                                 ret = err;
 396                         ret = btrfs_end_transaction(trans, tree_root);
 397                         BUG_ON(ret);
 398                         mutex_unlock(&tree_root->fs_info->fs_mutex);
 399
 400                         btrfs_btree_balance_dirty(tree_root);
 401                         schedule();
 402
 403                         mutex_lock(&tree_root->fs_info->fs_mutex);
 404                 }
 405                 BUG_ON(ret);
 406                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 407                 if (ret)
 408                         break;
 409                 ret = btrfs_end_transaction(trans, tree_root);
 410                 BUG_ON(ret);
 411
 412                 kfree(dirty->root);
 413                 kfree(dirty);
 414                 mutex_unlock(&tree_root->fs_info->fs_mutex);
 415                 btrfs_btree_balance_dirty(tree_root);
 416                 schedule();
 417         }
 418         return ret;
 419 }
 420
 421 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 422                              struct btrfs_root *root)
 423 {
 424         int ret = 0;
 425         struct btrfs_transaction *cur_trans;
 426         struct btrfs_transaction *prev_trans = NULL;
 427         struct list_head dirty_fs_roots;
 428         struct radix_tree_root pinned_copy;
 429         DEFINE_WAIT(wait);
 430
 431         init_bit_radix(&pinned_copy);
 432         INIT_LIST_HEAD(&dirty_fs_roots);
 433
 434         mutex_lock(&root->fs_info->trans_mutex);
 435         if (trans->transaction->in_commit) {
 436                 cur_trans = trans->transaction;
 437                 trans->transaction->use_count++;
 438                 mutex_unlock(&root->fs_info->trans_mutex);
 439                 btrfs_end_transaction(trans, root);
 440
 441                 mutex_unlock(&root->fs_info->fs_mutex);
 442                 ret = wait_for_commit(root, cur_trans);
 443                 BUG_ON(ret);
 444                 put_transaction(cur_trans);
 445                 mutex_lock(&root->fs_info->fs_mutex);
 446                 return 0;
 447         }
 448         trans->transaction->in_commit = 1;
 449         cur_trans = trans->transaction;
 450         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 451                 prev_trans = list_entry(cur_trans->list.prev,
 452                                         struct btrfs_transaction, list);
 453                 if (!prev_trans->commit_done) {
 454                         prev_trans->use_count++;
 455                         mutex_unlock(&root->fs_info->fs_mutex);
 456                         mutex_unlock(&root->fs_info->trans_mutex);
 457
 458                         wait_for_commit(root, prev_trans);
 459                         put_transaction(prev_trans);
 460
 461                         mutex_lock(&root->fs_info->fs_mutex);
 462                         mutex_lock(&root->fs_info->trans_mutex);
 463                 }
 464         }
 465         while (trans->transaction->num_writers > 1) {
 466                 WARN_ON(cur_trans != trans->transaction);
 467                 prepare_to_wait(&trans->transaction->writer_wait, &wait,
 468                                 TASK_UNINTERRUPTIBLE);
 469                 if (trans->transaction->num_writers <= 1)
 470                         break;
 471                 mutex_unlock(&root->fs_info->fs_mutex);
 472                 mutex_unlock(&root->fs_info->trans_mutex);
 473                 schedule();
 474                 mutex_lock(&root->fs_info->fs_mutex);
 475                 mutex_lock(&root->fs_info->trans_mutex);
 476                 finish_wait(&trans->transaction->writer_wait, &wait);
 477         }
 478         finish_wait(&trans->transaction->writer_wait, &wait);
 479         WARN_ON(cur_trans != trans->transaction);
 480         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 481                               &dirty_fs_roots);
 482         BUG_ON(ret);
 483
 484         ret = btrfs_commit_tree_roots(trans, root);
 485         BUG_ON(ret);
 486
 487         cur_trans = root->fs_info->running_transaction;
 488         root->fs_info->running_transaction = NULL;
 489         btrfs_set_super_generation(&root->fs_info->super_copy,
 490                                    cur_trans->transid);
 491         btrfs_set_super_root(&root->fs_info->super_copy,
 492                              bh_blocknr(root->fs_info->tree_root->node));
 493         memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
 494                sizeof(root->fs_info->super_copy));
 495
 496         btrfs_copy_pinned(root, &pinned_copy);
 497
 498         mutex_unlock(&root->fs_info->trans_mutex);
 499         mutex_unlock(&root->fs_info->fs_mutex);
 500         ret = btrfs_write_and_wait_transaction(trans, root);
 501         BUG_ON(ret);
 502         write_ctree_super(trans, root);
 503         mutex_lock(&root->fs_info->fs_mutex);
 504         btrfs_finish_extent_commit(trans, root, &pinned_copy);
 505         mutex_lock(&root->fs_info->trans_mutex);
 506         cur_trans->commit_done = 1;
 507         wake_up(&cur_trans->commit_wait);
 508         put_transaction(cur_trans);
 509         put_transaction(cur_trans);
 510         if (root->fs_info->closing)
 511                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 512         else
 513                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 514         mutex_unlock(&root->fs_info->trans_mutex);
 515         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 516
 517         if (root->fs_info->closing) {
 518                 mutex_unlock(&root->fs_info->fs_mutex);
 519                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 520                 mutex_lock(&root->fs_info->fs_mutex);
 521         }
 522         return ret;
 523 }
 524
 525 void btrfs_transaction_cleaner(struct work_struct *work)
 526 {
 527         struct btrfs_fs_info *fs_info = container_of(work,
 528                                                      struct btrfs_fs_info,
 529                                                      trans_work.work);
 530
 531         struct btrfs_root *root = fs_info->tree_root;
 532         struct btrfs_transaction *cur;
 533         struct btrfs_trans_handle *trans;
 534         struct list_head dirty_roots;
 535         unsigned long now;
 536         unsigned long delay = HZ * 30;
 537         int ret;
 538
 539         INIT_LIST_HEAD(&dirty_roots);
 540         mutex_lock(&root->fs_info->fs_mutex);
 541         mutex_lock(&root->fs_info->trans_mutex);
 542         cur = root->fs_info->running_transaction;
 543         if (!cur) {
 544                 mutex_unlock(&root->fs_info->trans_mutex);
 545                 goto out;
 546         }
 547         now = get_seconds();
 548         if (now < cur->start_time || now - cur->start_time < 30) {
 549                 mutex_unlock(&root->fs_info->trans_mutex);
 550                 delay = HZ * 5;
 551                 goto out;
 552         }
 553         mutex_unlock(&root->fs_info->trans_mutex);
 554         btrfs_defrag_dirty_roots(root->fs_info);
 555         trans = btrfs_start_transaction(root, 1);
 556         ret = btrfs_commit_transaction(trans, root);
 557 out:
 558         mutex_unlock(&root->fs_info->fs_mutex);
 559
 560         mutex_lock(&root->fs_info->trans_mutex);
 561         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 562         mutex_unlock(&root->fs_info->trans_mutex);
 563
 564         if (!list_empty(&dirty_roots)) {
 565                 drop_dirty_roots(root, &dirty_roots);
 566         }
 567         btrfs_transaction_queue_work(root, delay);
 568 }
 569
 570 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
 571 {
 572         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
 573 }
 574
 575 void btrfs_transaction_flush_work(struct btrfs_root *root)
 576 {
 577         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
 578         flush_workqueue(trans_wq);
 579 }
 580
 581 void __init btrfs_init_transaction_sys(void)
 582 {
 583         trans_wq = create_workqueue("btrfs");
 584 }
 585
 586 void __exit btrfs_exit_transaction_sys(void)
 587 {
 588         destroy_workqueue(trans_wq);
 589 }
 590