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