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