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 "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
  33 static void put_transaction(struct btrfs_transaction *transaction)
  34 {
  35         WARN_ON(transaction->use_count == 0);
  36         transaction->use_count--;
  37         if (transaction->use_count == 0) {
  38                 WARN_ON(total_trans == 0);
  39                 total_trans--;
  40                 list_del_init(&transaction->list);
  41                 memset(transaction, 0, sizeof(*transaction));
  42                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  43         }
  44 }
  45
  46 static int join_transaction(struct btrfs_root *root)
  47 {
  48         struct btrfs_transaction *cur_trans;
  49         cur_trans = root->fs_info->running_transaction;
  50         if (!cur_trans) {
  51                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  52                                              GFP_NOFS);
  53                 total_trans++;
  54                 BUG_ON(!cur_trans);
  55                 root->fs_info->generation++;
  56                 root->fs_info->running_transaction = cur_trans;
  57                 cur_trans->num_writers = 0;
  58                 cur_trans->transid = root->fs_info->generation;
  59                 init_waitqueue_head(&cur_trans->writer_wait);
  60                 init_waitqueue_head(&cur_trans->commit_wait);
  61                 cur_trans->in_commit = 0;
  62                 cur_trans->use_count = 1;
  63                 cur_trans->commit_done = 0;
  64                 cur_trans->start_time = get_seconds();
  65                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  66                 init_bit_radix(&cur_trans->dirty_pages);
  67         }
  68         cur_trans->num_writers++;
  69         return 0;
  70 }
  71
  72 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
  73                                                    int num_blocks)
  74 {
  75         struct btrfs_trans_handle *h =
  76                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
  77         int ret;
  78         u64 running_trans_id;
  79
  80         mutex_lock(&root->fs_info->trans_mutex);
  81         ret = join_transaction(root);
  82         BUG_ON(ret);
  83         running_trans_id = root->fs_info->running_transaction->transid;
  84
  85         if (root != root->fs_info->tree_root && root->last_trans <
  86             running_trans_id) {
  87                 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  88                                    (unsigned long)root->root_key.objectid,
  89                                    BTRFS_ROOT_TRANS_TAG);
  90                 root->commit_root = root->node;
  91                 get_bh(root->node);
  92         }
  93         root->last_trans = running_trans_id;
  94         h->transid = running_trans_id;
  95         h->transaction = root->fs_info->running_transaction;
  96         h->blocks_reserved = num_blocks;
  97         h->blocks_used = 0;
  98         h->block_group = NULL;
  99         root->fs_info->running_transaction->use_count++;
 100         mutex_unlock(&root->fs_info->trans_mutex);
 101         return h;
 102 }
 103
 104 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 105                           struct btrfs_root *root)
 106 {
 107         struct btrfs_transaction *cur_trans;
 108
 109         mutex_lock(&root->fs_info->trans_mutex);
 110         cur_trans = root->fs_info->running_transaction;
 111         WARN_ON(cur_trans->num_writers < 1);
 112         if (waitqueue_active(&cur_trans->writer_wait))
 113                 wake_up(&cur_trans->writer_wait);
 114         cur_trans->num_writers--;
 115         put_transaction(cur_trans);
 116         mutex_unlock(&root->fs_info->trans_mutex);
 117         memset(trans, 0, sizeof(*trans));
 118         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 119         return 0;
 120 }
 121
 122
 123 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 124                                      struct btrfs_root *root)
 125 {
 126         unsigned long gang[16];
 127         int ret;
 128         int i;
 129         int err;
 130         int werr = 0;
 131         struct page *page;
 132         struct radix_tree_root *dirty_pages;
 133         struct inode *btree_inode = root->fs_info->btree_inode;
 134
 135         if (!trans || !trans->transaction) {
 136                 return filemap_write_and_wait(btree_inode->i_mapping);
 137         }
 138         dirty_pages = &trans->transaction->dirty_pages;
 139         while(1) {
 140                 ret = find_first_radix_bit(dirty_pages, gang,
 141                                            0, ARRAY_SIZE(gang));
 142                 if (!ret)
 143                         break;
 144                 for (i = 0; i < ret; i++) {
 145                         /* FIXME EIO */
 146                         clear_radix_bit(dirty_pages, gang[i]);
 147                         page = find_lock_page(btree_inode->i_mapping,
 148                                               gang[i]);
 149                         if (!page)
 150                                 continue;
 151                         err = write_one_page(page, 0);
 152                         if (err)
 153                                 werr = err;
 154                         page_cache_release(page);
 155                 }
 156         }
 157         err = filemap_fdatawait(btree_inode->i_mapping);
 158         if (err)
 159                 werr = err;
 160         return werr;
 161 }
 162
 163 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 164                             struct btrfs_root *root)
 165 {
 166         int ret;
 167         u64 old_extent_block;
 168         struct btrfs_fs_info *fs_info = root->fs_info;
 169         struct btrfs_root *tree_root = fs_info->tree_root;
 170         struct btrfs_root *extent_root = fs_info->extent_root;
 171
 172         btrfs_write_dirty_block_groups(trans, extent_root);
 173         while(1) {
 174                 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
 175                 if (old_extent_block == bh_blocknr(extent_root->node))
 176                         break;
 177                 btrfs_set_root_blocknr(&extent_root->root_item,
 178                                        bh_blocknr(extent_root->node));
 179                 ret = btrfs_update_root(trans, tree_root,
 180                                         &extent_root->root_key,
 181                                         &extent_root->root_item);
 182                 BUG_ON(ret);
 183                 btrfs_write_dirty_block_groups(trans, extent_root);
 184         }
 185         return 0;
 186 }
 187
 188 static int wait_for_commit(struct btrfs_root *root,
 189                            struct btrfs_transaction *commit)
 190 {
 191         DEFINE_WAIT(wait);
 192         while(!commit->commit_done) {
 193                 prepare_to_wait(&commit->commit_wait, &wait,
 194                                 TASK_UNINTERRUPTIBLE);
 195                 if (commit->commit_done)
 196                         break;
 197                 mutex_unlock(&root->fs_info->trans_mutex);
 198                 schedule();
 199                 mutex_lock(&root->fs_info->trans_mutex);
 200         }
 201         finish_wait(&commit->commit_wait, &wait);
 202         return 0;
 203 }
 204
 205 struct dirty_root {
 206         struct list_head list;
 207         struct btrfs_key snap_key;
 208         struct buffer_head *commit_root;
 209         struct btrfs_root *root;
 210 };
 211
 212 static int add_dirty_roots(struct btrfs_trans_handle *trans,
 213                            struct radix_tree_root *radix,
 214                            struct list_head *list)
 215 {
 216         struct dirty_root *dirty;
 217         struct btrfs_root *gang[8];
 218         struct btrfs_root *root;
 219         int i;
 220         int ret;
 221         int err;
 222         while(1) {
 223                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 224                                                  ARRAY_SIZE(gang),
 225                                                  BTRFS_ROOT_TRANS_TAG);
 226                 if (ret == 0)
 227                         break;
 228                 for (i = 0; i < ret; i++) {
 229                         root = gang[i];
 230                         radix_tree_tag_clear(radix,
 231                                      (unsigned long)root->root_key.objectid,
 232                                      BTRFS_ROOT_TRANS_TAG);
 233                         if (root->commit_root == root->node) {
 234                                 WARN_ON(bh_blocknr(root->node) !=
 235                                         btrfs_root_blocknr(&root->root_item));
 236                                 brelse(root->commit_root);
 237                                 root->commit_root = NULL;
 238                                 continue;
 239                         }
 240                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 241                         BUG_ON(!dirty);
 242                         memcpy(&dirty->snap_key, &root->root_key,
 243                                sizeof(root->root_key));
 244                         dirty->commit_root = root->commit_root;
 245                         root->commit_root = NULL;
 246                         dirty->root = root;
 247                         root->root_key.offset = root->fs_info->generation;
 248                         btrfs_set_root_blocknr(&root->root_item,
 249                                                bh_blocknr(root->node));
 250                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 251                                                 &root->root_key,
 252                                                 &root->root_item);
 253                         BUG_ON(err);
 254                         list_add(&dirty->list, list);
 255                 }
 256         }
 257         return 0;
 258 }
 259
 260 static int drop_dirty_roots(struct btrfs_root *tree_root,
 261                             struct list_head *list)
 262 {
 263         struct dirty_root *dirty;
 264         struct btrfs_trans_handle *trans;
 265         int ret;
 266         while(!list_empty(list)) {
 267                 mutex_lock(&tree_root->fs_info->fs_mutex);
 268                 dirty = list_entry(list->next, struct dirty_root, list);
 269                 list_del_init(&dirty->list);
 270                 trans = btrfs_start_transaction(tree_root, 1);
 271                 ret = btrfs_drop_snapshot(trans, dirty->root,
 272                                           dirty->commit_root);
 273                 BUG_ON(ret);
 274
 275                 ret = btrfs_del_root(trans, tree_root, &dirty->snap_key);
 276                 BUG_ON(ret);
 277                 ret = btrfs_end_transaction(trans, tree_root);
 278                 BUG_ON(ret);
 279                 kfree(dirty);
 280                 mutex_unlock(&tree_root->fs_info->fs_mutex);
 281         }
 282         return 0;
 283 }
 284
 285 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 286                              struct btrfs_root *root)
 287 {
 288         int ret = 0;
 289         struct btrfs_transaction *cur_trans;
 290         struct btrfs_transaction *prev_trans = NULL;
 291         struct list_head dirty_fs_roots;
 292         DEFINE_WAIT(wait);
 293
 294         INIT_LIST_HEAD(&dirty_fs_roots);
 295
 296         mutex_lock(&root->fs_info->trans_mutex);
 297         if (trans->transaction->in_commit) {
 298                 cur_trans = trans->transaction;
 299                 trans->transaction->use_count++;
 300                 btrfs_end_transaction(trans, root);
 301                 ret = wait_for_commit(root, cur_trans);
 302                 BUG_ON(ret);
 303                 put_transaction(cur_trans);
 304                 mutex_unlock(&root->fs_info->trans_mutex);
 305                 return 0;
 306         }
 307         cur_trans = trans->transaction;
 308         trans->transaction->in_commit = 1;
 309         while (trans->transaction->num_writers > 1) {
 310                 WARN_ON(cur_trans != trans->transaction);
 311                 prepare_to_wait(&trans->transaction->writer_wait, &wait,
 312                                 TASK_UNINTERRUPTIBLE);
 313                 if (trans->transaction->num_writers <= 1)
 314                         break;
 315                 mutex_unlock(&root->fs_info->trans_mutex);
 316                 schedule();
 317                 mutex_lock(&root->fs_info->trans_mutex);
 318                 finish_wait(&trans->transaction->writer_wait, &wait);
 319         }
 320         finish_wait(&trans->transaction->writer_wait, &wait);
 321         WARN_ON(cur_trans != trans->transaction);
 322         add_dirty_roots(trans, &root->fs_info->fs_roots_radix, &dirty_fs_roots);
 323         ret = btrfs_commit_tree_roots(trans, root);
 324         BUG_ON(ret);
 325         cur_trans = root->fs_info->running_transaction;
 326         root->fs_info->running_transaction = NULL;
 327         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 328                 prev_trans = list_entry(cur_trans->list.prev,
 329                                         struct btrfs_transaction, list);
 330                 if (prev_trans->commit_done)
 331                         prev_trans = NULL;
 332                 else
 333                         prev_trans->use_count++;
 334         }
 335         mutex_unlock(&root->fs_info->trans_mutex);
 336         mutex_unlock(&root->fs_info->fs_mutex);
 337         ret = btrfs_write_and_wait_transaction(trans, root);
 338         if (prev_trans) {
 339                 mutex_lock(&root->fs_info->trans_mutex);
 340                 wait_for_commit(root, prev_trans);
 341                 put_transaction(prev_trans);
 342                 mutex_unlock(&root->fs_info->trans_mutex);
 343         }
 344         btrfs_set_super_generation(root->fs_info->disk_super,
 345                                    cur_trans->transid);
 346         BUG_ON(ret);
 347         write_ctree_super(trans, root);
 348
 349         mutex_lock(&root->fs_info->fs_mutex);
 350         btrfs_finish_extent_commit(trans, root);
 351         mutex_lock(&root->fs_info->trans_mutex);
 352         cur_trans->commit_done = 1;
 353         wake_up(&cur_trans->commit_wait);
 354         put_transaction(cur_trans);
 355         put_transaction(cur_trans);
 356         if (root->fs_info->closing)
 357                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 358         else
 359                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 360         mutex_unlock(&root->fs_info->trans_mutex);
 361         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 362
 363         if (root->fs_info->closing) {
 364                 mutex_unlock(&root->fs_info->fs_mutex);
 365                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 366                 mutex_lock(&root->fs_info->fs_mutex);
 367         }
 368         return ret;
 369 }
 370
 371 void btrfs_transaction_cleaner(struct work_struct *work)
 372 {
 373         struct btrfs_fs_info *fs_info = container_of(work,
 374                                                      struct btrfs_fs_info,
 375                                                      trans_work.work);
 376
 377         struct btrfs_root *root = fs_info->tree_root;
 378         struct btrfs_transaction *cur;
 379         struct btrfs_trans_handle *trans;
 380         struct list_head dirty_roots;
 381         unsigned long now;
 382         unsigned long delay = HZ * 30;
 383         int ret;
 384
 385         INIT_LIST_HEAD(&dirty_roots);
 386         mutex_lock(&root->fs_info->fs_mutex);
 387         mutex_lock(&root->fs_info->trans_mutex);
 388         cur = root->fs_info->running_transaction;
 389         if (!cur) {
 390                 mutex_unlock(&root->fs_info->trans_mutex);
 391                 goto out;
 392         }
 393         now = get_seconds();
 394         if (now < cur->start_time || now - cur->start_time < 30) {
 395                 mutex_unlock(&root->fs_info->trans_mutex);
 396                 delay = HZ * 5;
 397                 goto out;
 398         }
 399         mutex_unlock(&root->fs_info->trans_mutex);
 400         trans = btrfs_start_transaction(root, 1);
 401         ret = btrfs_commit_transaction(trans, root);
 402 out:
 403         mutex_unlock(&root->fs_info->fs_mutex);
 404
 405         mutex_lock(&root->fs_info->trans_mutex);
 406         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 407         mutex_unlock(&root->fs_info->trans_mutex);
 408
 409         if (!list_empty(&dirty_roots)) {
 410                 drop_dirty_roots(root, &dirty_roots);
 411         }
 412         btrfs_transaction_queue_work(root, delay);
 413 }
 414
 415 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
 416 {
 417         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
 418 }
 419
 420 void btrfs_transaction_flush_work(struct btrfs_root *root)
 421 {
 422         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
 423         flush_workqueue(trans_wq);
 424 }
 425
 426 void __init btrfs_init_transaction_sys(void)
 427 {
 428         trans_wq = create_workqueue("btrfs");
 429 }
 430
 431 void __exit btrfs_exit_transaction_sys(void)
 432 {
 433         destroy_workqueue(trans_wq);
 434 }
 435