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