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 <linux/writeback.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/blkdev.h>
  24 #include "ctree.h"
  25 #include "disk-io.h"
  26 #include "transaction.h"
  27 #include "locking.h"
  28 #include "tree-log.h"
  29
  30 #define BTRFS_ROOT_TRANS_TAG 0
  31
  32 static noinline void put_transaction(struct btrfs_transaction *transaction)
  33 {
  34         WARN_ON(transaction->use_count == 0);
  35         transaction->use_count--;
  36         if (transaction->use_count == 0) {
  37                 list_del_init(&transaction->list);
  38                 memset(transaction, 0, sizeof(*transaction));
  39                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  40         }
  41 }
  42
  43 static noinline void switch_commit_root(struct btrfs_root *root)
  44 {
  45         free_extent_buffer(root->commit_root);
  46         root->commit_root = btrfs_root_node(root);
  47 }
  48
  49 /*
  50  * either allocate a new transaction or hop into the existing one
  51  */
  52 static noinline int join_transaction(struct btrfs_root *root)
  53 {
  54         struct btrfs_transaction *cur_trans;
  55         cur_trans = root->fs_info->running_transaction;
  56         if (!cur_trans) {
  57                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  58                                              GFP_NOFS);
  59                 BUG_ON(!cur_trans);
  60                 root->fs_info->generation++;
  61                 cur_trans->num_writers = 1;
  62                 cur_trans->num_joined = 0;
  63                 cur_trans->transid = root->fs_info->generation;
  64                 init_waitqueue_head(&cur_trans->writer_wait);
  65                 init_waitqueue_head(&cur_trans->commit_wait);
  66                 cur_trans->in_commit = 0;
  67                 cur_trans->blocked = 0;
  68                 cur_trans->use_count = 1;
  69                 cur_trans->commit_done = 0;
  70                 cur_trans->start_time = get_seconds();
  71
  72                 cur_trans->delayed_refs.root.rb_node = NULL;
  73                 cur_trans->delayed_refs.num_entries = 0;
  74                 cur_trans->delayed_refs.num_heads_ready = 0;
  75                 cur_trans->delayed_refs.num_heads = 0;
  76                 cur_trans->delayed_refs.flushing = 0;
  77                 cur_trans->delayed_refs.run_delayed_start = 0;
  78                 spin_lock_init(&cur_trans->delayed_refs.lock);
  79
  80                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  81                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  82                 extent_io_tree_init(&cur_trans->dirty_pages,
  83                                      root->fs_info->btree_inode->i_mapping,
  84                                      GFP_NOFS);
  85                 spin_lock(&root->fs_info->new_trans_lock);
  86                 root->fs_info->running_transaction = cur_trans;
  87                 spin_unlock(&root->fs_info->new_trans_lock);
  88         } else {
  89                 cur_trans->num_writers++;
  90                 cur_trans->num_joined++;
  91         }
  92
  93         return 0;
  94 }
  95
  96 /*
  97  * this does all the record keeping required to make sure that a reference
  98  * counted root is properly recorded in a given transaction.  This is required
  99  * to make sure the old root from before we joined the transaction is deleted
 100  * when the transaction commits
 101  */
 102 static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
 103                                          struct btrfs_root *root)
 104 {
 105         if (root->ref_cows && root->last_trans < trans->transid) {
 106                 WARN_ON(root == root->fs_info->extent_root);
 107                 WARN_ON(root->commit_root != root->node);
 108
 109                 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
 110                            (unsigned long)root->root_key.objectid,
 111                            BTRFS_ROOT_TRANS_TAG);
 112                 root->last_trans = trans->transid;
 113                 btrfs_init_reloc_root(trans, root);
 114         }
 115         return 0;
 116 }
 117
 118 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
 119                                struct btrfs_root *root)
 120 {
 121         if (!root->ref_cows)
 122                 return 0;
 123
 124         mutex_lock(&root->fs_info->trans_mutex);
 125         if (root->last_trans == trans->transid) {
 126                 mutex_unlock(&root->fs_info->trans_mutex);
 127                 return 0;
 128         }
 129
 130         record_root_in_trans(trans, root);
 131         mutex_unlock(&root->fs_info->trans_mutex);
 132         return 0;
 133 }
 134
 135 /* wait for commit against the current transaction to become unblocked
 136  * when this is done, it is safe to start a new transaction, but the current
 137  * transaction might not be fully on disk.
 138  */
 139 static void wait_current_trans(struct btrfs_root *root)
 140 {
 141         struct btrfs_transaction *cur_trans;
 142
 143         cur_trans = root->fs_info->running_transaction;
 144         if (cur_trans && cur_trans->blocked) {
 145                 DEFINE_WAIT(wait);
 146                 cur_trans->use_count++;
 147                 while (1) {
 148                         prepare_to_wait(&root->fs_info->transaction_wait, &wait,
 149                                         TASK_UNINTERRUPTIBLE);
 150                         if (cur_trans->blocked) {
 151                                 mutex_unlock(&root->fs_info->trans_mutex);
 152                                 schedule();
 153                                 mutex_lock(&root->fs_info->trans_mutex);
 154                                 finish_wait(&root->fs_info->transaction_wait,
 155                                             &wait);
 156                         } else {
 157                                 finish_wait(&root->fs_info->transaction_wait,
 158                                             &wait);
 159                                 break;
 160                         }
 161                 }
 162                 put_transaction(cur_trans);
 163         }
 164 }
 165
 166 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 167                                              int num_blocks, int wait)
 168 {
 169         struct btrfs_trans_handle *h =
 170                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 171         int ret;
 172
 173         mutex_lock(&root->fs_info->trans_mutex);
 174         if (!root->fs_info->log_root_recovering &&
 175             ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
 176                 wait_current_trans(root);
 177         ret = join_transaction(root);
 178         BUG_ON(ret);
 179
 180         h->transid = root->fs_info->running_transaction->transid;
 181         h->transaction = root->fs_info->running_transaction;
 182         h->blocks_reserved = num_blocks;
 183         h->blocks_used = 0;
 184         h->block_group = 0;
 185         h->alloc_exclude_nr = 0;
 186         h->alloc_exclude_start = 0;
 187         h->delayed_ref_updates = 0;
 188
 189         root->fs_info->running_transaction->use_count++;
 190         record_root_in_trans(h, root);
 191         mutex_unlock(&root->fs_info->trans_mutex);
 192         return h;
 193 }
 194
 195 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 196                                                    int num_blocks)
 197 {
 198         return start_transaction(root, num_blocks, 1);
 199 }
 200 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 201                                                    int num_blocks)
 202 {
 203         return start_transaction(root, num_blocks, 0);
 204 }
 205
 206 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
 207                                                          int num_blocks)
 208 {
 209         return start_transaction(r, num_blocks, 2);
 210 }
 211
 212 /* wait for a transaction commit to be fully complete */
 213 static noinline int wait_for_commit(struct btrfs_root *root,
 214                                     struct btrfs_transaction *commit)
 215 {
 216         DEFINE_WAIT(wait);
 217         mutex_lock(&root->fs_info->trans_mutex);
 218         while (!commit->commit_done) {
 219                 prepare_to_wait(&commit->commit_wait, &wait,
 220                                 TASK_UNINTERRUPTIBLE);
 221                 if (commit->commit_done)
 222                         break;
 223                 mutex_unlock(&root->fs_info->trans_mutex);
 224                 schedule();
 225                 mutex_lock(&root->fs_info->trans_mutex);
 226         }
 227         mutex_unlock(&root->fs_info->trans_mutex);
 228         finish_wait(&commit->commit_wait, &wait);
 229         return 0;
 230 }
 231
 232 #if 0
 233 /*
 234  * rate limit against the drop_snapshot code.  This helps to slow down new
 235  * operations if the drop_snapshot code isn't able to keep up.
 236  */
 237 static void throttle_on_drops(struct btrfs_root *root)
 238 {
 239         struct btrfs_fs_info *info = root->fs_info;
 240         int harder_count = 0;
 241
 242 harder:
 243         if (atomic_read(&info->throttles)) {
 244                 DEFINE_WAIT(wait);
 245                 int thr;
 246                 thr = atomic_read(&info->throttle_gen);
 247
 248                 do {
 249                         prepare_to_wait(&info->transaction_throttle,
 250                                         &wait, TASK_UNINTERRUPTIBLE);
 251                         if (!atomic_read(&info->throttles)) {
 252                                 finish_wait(&info->transaction_throttle, &wait);
 253                                 break;
 254                         }
 255                         schedule();
 256                         finish_wait(&info->transaction_throttle, &wait);
 257                 } while (thr == atomic_read(&info->throttle_gen));
 258                 harder_count++;
 259
 260                 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
 261                     harder_count < 2)
 262                         goto harder;
 263
 264                 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
 265                     harder_count < 10)
 266                         goto harder;
 267
 268                 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
 269                     harder_count < 20)
 270                         goto harder;
 271         }
 272 }
 273 #endif
 274
 275 void btrfs_throttle(struct btrfs_root *root)
 276 {
 277         mutex_lock(&root->fs_info->trans_mutex);
 278         if (!root->fs_info->open_ioctl_trans)
 279                 wait_current_trans(root);
 280         mutex_unlock(&root->fs_info->trans_mutex);
 281 }
 282
 283 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 284                           struct btrfs_root *root, int throttle)
 285 {
 286         struct btrfs_transaction *cur_trans;
 287         struct btrfs_fs_info *info = root->fs_info;
 288         int count = 0;
 289
 290         while (count < 4) {
 291                 unsigned long cur = trans->delayed_ref_updates;
 292                 trans->delayed_ref_updates = 0;
 293                 if (cur &&
 294                     trans->transaction->delayed_refs.num_heads_ready > 64) {
 295                         trans->delayed_ref_updates = 0;
 296
 297                         /*
 298                          * do a full flush if the transaction is trying
 299                          * to close
 300                          */
 301                         if (trans->transaction->delayed_refs.flushing)
 302                                 cur = 0;
 303                         btrfs_run_delayed_refs(trans, root, cur);
 304                 } else {
 305                         break;
 306                 }
 307                 count++;
 308         }
 309
 310         mutex_lock(&info->trans_mutex);
 311         cur_trans = info->running_transaction;
 312         WARN_ON(cur_trans != trans->transaction);
 313         WARN_ON(cur_trans->num_writers < 1);
 314         cur_trans->num_writers--;
 315
 316         if (waitqueue_active(&cur_trans->writer_wait))
 317                 wake_up(&cur_trans->writer_wait);
 318         put_transaction(cur_trans);
 319         mutex_unlock(&info->trans_mutex);
 320         memset(trans, 0, sizeof(*trans));
 321         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 322
 323         return 0;
 324 }
 325
 326 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 327                           struct btrfs_root *root)
 328 {
 329         return __btrfs_end_transaction(trans, root, 0);
 330 }
 331
 332 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 333                                    struct btrfs_root *root)
 334 {
 335         return __btrfs_end_transaction(trans, root, 1);
 336 }
 337
 338 /*
 339  * when btree blocks are allocated, they have some corresponding bits set for
 340  * them in one of two extent_io trees.  This is used to make sure all of
 341  * those extents are on disk for transaction or log commit
 342  */
 343 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 344                                         struct extent_io_tree *dirty_pages)
 345 {
 346         int ret;
 347         int err = 0;
 348         int werr = 0;
 349         struct page *page;
 350         struct inode *btree_inode = root->fs_info->btree_inode;
 351         u64 start = 0;
 352         u64 end;
 353         unsigned long index;
 354
 355         while (1) {
 356                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
 357                                             EXTENT_DIRTY);
 358                 if (ret)
 359                         break;
 360                 while (start <= end) {
 361                         cond_resched();
 362
 363                         index = start >> PAGE_CACHE_SHIFT;
 364                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 365                         page = find_get_page(btree_inode->i_mapping, index);
 366                         if (!page)
 367                                 continue;
 368
 369                         btree_lock_page_hook(page);
 370                         if (!page->mapping) {
 371                                 unlock_page(page);
 372                                 page_cache_release(page);
 373                                 continue;
 374                         }
 375
 376                         if (PageWriteback(page)) {
 377                                 if (PageDirty(page))
 378                                         wait_on_page_writeback(page);
 379                                 else {
 380                                         unlock_page(page);
 381                                         page_cache_release(page);
 382                                         continue;
 383                                 }
 384                         }
 385                         err = write_one_page(page, 0);
 386                         if (err)
 387                                 werr = err;
 388                         page_cache_release(page);
 389                 }
 390         }
 391         while (1) {
 392                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 393                                             EXTENT_DIRTY);
 394                 if (ret)
 395                         break;
 396
 397                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 398                 while (start <= end) {
 399                         index = start >> PAGE_CACHE_SHIFT;
 400                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 401                         page = find_get_page(btree_inode->i_mapping, index);
 402                         if (!page)
 403                                 continue;
 404                         if (PageDirty(page)) {
 405                                 btree_lock_page_hook(page);
 406                                 wait_on_page_writeback(page);
 407                                 err = write_one_page(page, 0);
 408                                 if (err)
 409                                         werr = err;
 410                         }
 411                         wait_on_page_writeback(page);
 412                         page_cache_release(page);
 413                         cond_resched();
 414                 }
 415         }
 416         if (err)
 417                 werr = err;
 418         return werr;
 419 }
 420
 421 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 422                                      struct btrfs_root *root)
 423 {
 424         if (!trans || !trans->transaction) {
 425                 struct inode *btree_inode;
 426                 btree_inode = root->fs_info->btree_inode;
 427                 return filemap_write_and_wait(btree_inode->i_mapping);
 428         }
 429         return btrfs_write_and_wait_marked_extents(root,
 430                                            &trans->transaction->dirty_pages);
 431 }
 432
 433 /*
 434  * this is used to update the root pointer in the tree of tree roots.
 435  *
 436  * But, in the case of the extent allocation tree, updating the root
 437  * pointer may allocate blocks which may change the root of the extent
 438  * allocation tree.
 439  *
 440  * So, this loops and repeats and makes sure the cowonly root didn't
 441  * change while the root pointer was being updated in the metadata.
 442  */
 443 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 444                                struct btrfs_root *root)
 445 {
 446         int ret;
 447         u64 old_root_bytenr;
 448         struct btrfs_root *tree_root = root->fs_info->tree_root;
 449
 450         btrfs_write_dirty_block_groups(trans, root);
 451
 452         while (1) {
 453                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 454                 if (old_root_bytenr == root->node->start)
 455                         break;
 456
 457                 btrfs_set_root_node(&root->root_item, root->node);
 458                 ret = btrfs_update_root(trans, tree_root,
 459                                         &root->root_key,
 460                                         &root->root_item);
 461                 BUG_ON(ret);
 462
 463                 ret = btrfs_write_dirty_block_groups(trans, root);
 464                 BUG_ON(ret);
 465         }
 466
 467         if (root != root->fs_info->extent_root)
 468                 switch_commit_root(root);
 469
 470         return 0;
 471 }
 472
 473 /*
 474  * update all the cowonly tree roots on disk
 475  */
 476 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
 477                                          struct btrfs_root *root)
 478 {
 479         struct btrfs_fs_info *fs_info = root->fs_info;
 480         struct list_head *next;
 481         struct extent_buffer *eb;
 482         int ret;
 483
 484         ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 485         BUG_ON(ret);
 486
 487         eb = btrfs_lock_root_node(fs_info->tree_root);
 488         btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
 489         btrfs_tree_unlock(eb);
 490         free_extent_buffer(eb);
 491
 492         ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 493         BUG_ON(ret);
 494
 495         while (!list_empty(&fs_info->dirty_cowonly_roots)) {
 496                 next = fs_info->dirty_cowonly_roots.next;
 497                 list_del_init(next);
 498                 root = list_entry(next, struct btrfs_root, dirty_list);
 499
 500                 update_cowonly_root(trans, root);
 501         }
 502
 503         down_write(&fs_info->extent_commit_sem);
 504         switch_commit_root(fs_info->extent_root);
 505         up_write(&fs_info->extent_commit_sem);
 506
 507         return 0;
 508 }
 509
 510 /*
 511  * dead roots are old snapshots that need to be deleted.  This allocates
 512  * a dirty root struct and adds it into the list of dead roots that need to
 513  * be deleted
 514  */
 515 int btrfs_add_dead_root(struct btrfs_root *root)
 516 {
 517         mutex_lock(&root->fs_info->trans_mutex);
 518         list_add(&root->root_list, &root->fs_info->dead_roots);
 519         mutex_unlock(&root->fs_info->trans_mutex);
 520         return 0;
 521 }
 522
 523 /*
 524  * update all the cowonly tree roots on disk
 525  */
 526 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
 527                                     struct btrfs_root *root)
 528 {
 529         struct btrfs_root *gang[8];
 530         struct btrfs_fs_info *fs_info = root->fs_info;
 531         int i;
 532         int ret;
 533         int err = 0;
 534
 535         while (1) {
 536                 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
 537                                                  (void **)gang, 0,
 538                                                  ARRAY_SIZE(gang),
 539                                                  BTRFS_ROOT_TRANS_TAG);
 540                 if (ret == 0)
 541                         break;
 542                 for (i = 0; i < ret; i++) {
 543                         root = gang[i];
 544                         radix_tree_tag_clear(&fs_info->fs_roots_radix,
 545                                         (unsigned long)root->root_key.objectid,
 546                                         BTRFS_ROOT_TRANS_TAG);
 547
 548                         btrfs_free_log(trans, root);
 549                         btrfs_update_reloc_root(trans, root);
 550
 551                         if (root->commit_root != root->node) {
 552                                 switch_commit_root(root);
 553                                 btrfs_set_root_node(&root->root_item,
 554                                                     root->node);
 555                         }
 556
 557                         err = btrfs_update_root(trans, fs_info->tree_root,
 558                                                 &root->root_key,
 559                                                 &root->root_item);
 560                         if (err)
 561                                 break;
 562                 }
 563         }
 564         return err;
 565 }
 566
 567 /*
 568  * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 569  * otherwise every leaf in the btree is read and defragged.
 570  */
 571 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 572 {
 573         struct btrfs_fs_info *info = root->fs_info;
 574         int ret;
 575         struct btrfs_trans_handle *trans;
 576         unsigned long nr;
 577
 578         smp_mb();
 579         if (root->defrag_running)
 580                 return 0;
 581         trans = btrfs_start_transaction(root, 1);
 582         while (1) {
 583                 root->defrag_running = 1;
 584                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 585                 nr = trans->blocks_used;
 586                 btrfs_end_transaction(trans, root);
 587                 btrfs_btree_balance_dirty(info->tree_root, nr);
 588                 cond_resched();
 589
 590                 trans = btrfs_start_transaction(root, 1);
 591                 if (root->fs_info->closing || ret != -EAGAIN)
 592                         break;
 593         }
 594         root->defrag_running = 0;
 595         smp_mb();
 596         btrfs_end_transaction(trans, root);
 597         return 0;
 598 }
 599
 600 #if 0
 601 /*
 602  * when dropping snapshots, we generate a ton of delayed refs, and it makes
 603  * sense not to join the transaction while it is trying to flush the current
 604  * queue of delayed refs out.
 605  *
 606  * This is used by the drop snapshot code only
 607  */
 608 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
 609 {
 610         DEFINE_WAIT(wait);
 611
 612         mutex_lock(&info->trans_mutex);
 613         while (info->running_transaction &&
 614                info->running_transaction->delayed_refs.flushing) {
 615                 prepare_to_wait(&info->transaction_wait, &wait,
 616                                 TASK_UNINTERRUPTIBLE);
 617                 mutex_unlock(&info->trans_mutex);
 618
 619                 schedule();
 620
 621                 mutex_lock(&info->trans_mutex);
 622                 finish_wait(&info->transaction_wait, &wait);
 623         }
 624         mutex_unlock(&info->trans_mutex);
 625         return 0;
 626 }
 627
 628 /*
 629  * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
 630  * all of them
 631  */
 632 int btrfs_drop_dead_root(struct btrfs_root *root)
 633 {
 634         struct btrfs_trans_handle *trans;
 635         struct btrfs_root *tree_root = root->fs_info->tree_root;
 636         unsigned long nr;
 637         int ret;
 638
 639         while (1) {
 640                 /*
 641                  * we don't want to jump in and create a bunch of
 642                  * delayed refs if the transaction is starting to close
 643                  */
 644                 wait_transaction_pre_flush(tree_root->fs_info);
 645                 trans = btrfs_start_transaction(tree_root, 1);
 646
 647                 /*
 648                  * we've joined a transaction, make sure it isn't
 649                  * closing right now
 650                  */
 651                 if (trans->transaction->delayed_refs.flushing) {
 652                         btrfs_end_transaction(trans, tree_root);
 653                         continue;
 654                 }
 655
 656                 ret = btrfs_drop_snapshot(trans, root);
 657                 if (ret != -EAGAIN)
 658                         break;
 659
 660                 ret = btrfs_update_root(trans, tree_root,
 661                                         &root->root_key,
 662                                         &root->root_item);
 663                 if (ret)
 664                         break;
 665
 666                 nr = trans->blocks_used;
 667                 ret = btrfs_end_transaction(trans, tree_root);
 668                 BUG_ON(ret);
 669
 670                 btrfs_btree_balance_dirty(tree_root, nr);
 671                 cond_resched();
 672         }
 673         BUG_ON(ret);
 674
 675         ret = btrfs_del_root(trans, tree_root, &root->root_key);
 676         BUG_ON(ret);
 677
 678         nr = trans->blocks_used;
 679         ret = btrfs_end_transaction(trans, tree_root);
 680         BUG_ON(ret);
 681
 682         free_extent_buffer(root->node);
 683         free_extent_buffer(root->commit_root);
 684         kfree(root);
 685
 686         btrfs_btree_balance_dirty(tree_root, nr);
 687         return ret;
 688 }
 689 #endif
 690
 691 /*
 692  * new snapshots need to be created at a very specific time in the
 693  * transaction commit.  This does the actual creation
 694  */
 695 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 696                                    struct btrfs_fs_info *fs_info,
 697                                    struct btrfs_pending_snapshot *pending)
 698 {
 699         struct btrfs_key key;
 700         struct btrfs_root_item *new_root_item;
 701         struct btrfs_root *tree_root = fs_info->tree_root;
 702         struct btrfs_root *root = pending->root;
 703         struct extent_buffer *tmp;
 704         struct extent_buffer *old;
 705         int ret;
 706         u64 objectid;
 707
 708         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 709         if (!new_root_item) {
 710                 ret = -ENOMEM;
 711                 goto fail;
 712         }
 713         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 714         if (ret)
 715                 goto fail;
 716
 717         record_root_in_trans(trans, root);
 718         btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
 719         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 720
 721         key.objectid = objectid;
 722         /* record when the snapshot was created in key.offset */
 723         key.offset = trans->transid;
 724         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 725
 726         old = btrfs_lock_root_node(root);
 727         btrfs_cow_block(trans, root, old, NULL, 0, &old);
 728         btrfs_set_lock_blocking(old);
 729
 730         btrfs_copy_root(trans, root, old, &tmp, objectid);
 731         btrfs_tree_unlock(old);
 732         free_extent_buffer(old);
 733
 734         btrfs_set_root_node(new_root_item, tmp);
 735         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 736                                 new_root_item);
 737         btrfs_tree_unlock(tmp);
 738         free_extent_buffer(tmp);
 739         if (ret)
 740                 goto fail;
 741
 742         key.offset = (u64)-1;
 743         memcpy(&pending->root_key, &key, sizeof(key));
 744 fail:
 745         kfree(new_root_item);
 746         return ret;
 747 }
 748
 749 static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
 750                                    struct btrfs_pending_snapshot *pending)
 751 {
 752         int ret;
 753         int namelen;
 754         u64 index = 0;
 755         struct btrfs_trans_handle *trans;
 756         struct inode *parent_inode;
 757         struct inode *inode;
 758         struct btrfs_root *parent_root;
 759
 760         parent_inode = pending->dentry->d_parent->d_inode;
 761         parent_root = BTRFS_I(parent_inode)->root;
 762         trans = btrfs_join_transaction(parent_root, 1);
 763
 764         /*
 765          * insert the directory item
 766          */
 767         namelen = strlen(pending->name);
 768         ret = btrfs_set_inode_index(parent_inode, &index);
 769         ret = btrfs_insert_dir_item(trans, parent_root,
 770                             pending->name, namelen,
 771                             parent_inode->i_ino,
 772                             &pending->root_key, BTRFS_FT_DIR, index);
 773
 774         if (ret)
 775                 goto fail;
 776
 777         btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
 778         ret = btrfs_update_inode(trans, parent_root, parent_inode);
 779         BUG_ON(ret);
 780
 781         ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
 782                                  pending->root_key.objectid,
 783                                  parent_root->root_key.objectid,
 784                                  parent_inode->i_ino, index, pending->name,
 785                                  namelen);
 786
 787         BUG_ON(ret);
 788
 789         inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
 790         d_instantiate(pending->dentry, inode);
 791 fail:
 792         btrfs_end_transaction(trans, fs_info->fs_root);
 793         return ret;
 794 }
 795
 796 /*
 797  * create all the snapshots we've scheduled for creation
 798  */
 799 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 800                                              struct btrfs_fs_info *fs_info)
 801 {
 802         struct btrfs_pending_snapshot *pending;
 803         struct list_head *head = &trans->transaction->pending_snapshots;
 804         int ret;
 805
 806         list_for_each_entry(pending, head, list) {
 807                 ret = create_pending_snapshot(trans, fs_info, pending);
 808                 BUG_ON(ret);
 809         }
 810         return 0;
 811 }
 812
 813 static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
 814                                              struct btrfs_fs_info *fs_info)
 815 {
 816         struct btrfs_pending_snapshot *pending;
 817         struct list_head *head = &trans->transaction->pending_snapshots;
 818         int ret;
 819
 820         while (!list_empty(head)) {
 821                 pending = list_entry(head->next,
 822                                      struct btrfs_pending_snapshot, list);
 823                 ret = finish_pending_snapshot(fs_info, pending);
 824                 BUG_ON(ret);
 825                 list_del(&pending->list);
 826                 kfree(pending->name);
 827                 kfree(pending);
 828         }
 829         return 0;
 830 }
 831
 832 static void update_super_roots(struct btrfs_root *root)
 833 {
 834         struct btrfs_root_item *root_item;
 835         struct btrfs_super_block *super;
 836
 837         super = &root->fs_info->super_copy;
 838
 839         root_item = &root->fs_info->chunk_root->root_item;
 840         super->chunk_root = root_item->bytenr;
 841         super->chunk_root_generation = root_item->generation;
 842         super->chunk_root_level = root_item->level;
 843
 844         root_item = &root->fs_info->tree_root->root_item;
 845         super->root = root_item->bytenr;
 846         super->generation = root_item->generation;
 847         super->root_level = root_item->level;
 848 }
 849
 850 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
 851 {
 852         int ret = 0;
 853         spin_lock(&info->new_trans_lock);
 854         if (info->running_transaction)
 855                 ret = info->running_transaction->in_commit;
 856         spin_unlock(&info->new_trans_lock);
 857         return ret;
 858 }
 859
 860 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 861                              struct btrfs_root *root)
 862 {
 863         unsigned long joined = 0;
 864         unsigned long timeout = 1;
 865         struct btrfs_transaction *cur_trans;
 866         struct btrfs_transaction *prev_trans = NULL;
 867         DEFINE_WAIT(wait);
 868         int ret;
 869         int should_grow = 0;
 870         unsigned long now = get_seconds();
 871         int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
 872
 873         btrfs_run_ordered_operations(root, 0);
 874
 875         /* make a pass through all the delayed refs we have so far
 876          * any runnings procs may add more while we are here
 877          */
 878         ret = btrfs_run_delayed_refs(trans, root, 0);
 879         BUG_ON(ret);
 880
 881         cur_trans = trans->transaction;
 882         /*
 883          * set the flushing flag so procs in this transaction have to
 884          * start sending their work down.
 885          */
 886         cur_trans->delayed_refs.flushing = 1;
 887
 888         ret = btrfs_run_delayed_refs(trans, root, 0);
 889         BUG_ON(ret);
 890
 891         mutex_lock(&root->fs_info->trans_mutex);
 892         if (cur_trans->in_commit) {
 893                 cur_trans->use_count++;
 894                 mutex_unlock(&root->fs_info->trans_mutex);
 895                 btrfs_end_transaction(trans, root);
 896
 897                 ret = wait_for_commit(root, cur_trans);
 898                 BUG_ON(ret);
 899
 900                 mutex_lock(&root->fs_info->trans_mutex);
 901                 put_transaction(cur_trans);
 902                 mutex_unlock(&root->fs_info->trans_mutex);
 903
 904                 return 0;
 905         }
 906
 907         trans->transaction->in_commit = 1;
 908         trans->transaction->blocked = 1;
 909         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 910                 prev_trans = list_entry(cur_trans->list.prev,
 911                                         struct btrfs_transaction, list);
 912                 if (!prev_trans->commit_done) {
 913                         prev_trans->use_count++;
 914                         mutex_unlock(&root->fs_info->trans_mutex);
 915
 916                         wait_for_commit(root, prev_trans);
 917
 918                         mutex_lock(&root->fs_info->trans_mutex);
 919                         put_transaction(prev_trans);
 920                 }
 921         }
 922
 923         if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
 924                 should_grow = 1;
 925
 926         do {
 927                 int snap_pending = 0;
 928                 joined = cur_trans->num_joined;
 929                 if (!list_empty(&trans->transaction->pending_snapshots))
 930                         snap_pending = 1;
 931
 932                 WARN_ON(cur_trans != trans->transaction);
 933                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 934                                 TASK_UNINTERRUPTIBLE);
 935
 936                 if (cur_trans->num_writers > 1)
 937                         timeout = MAX_SCHEDULE_TIMEOUT;
 938                 else if (should_grow)
 939                         timeout = 1;
 940
 941                 mutex_unlock(&root->fs_info->trans_mutex);
 942
 943                 if (flush_on_commit) {
 944                         btrfs_start_delalloc_inodes(root);
 945                         ret = btrfs_wait_ordered_extents(root, 0);
 946                         BUG_ON(ret);
 947                 } else if (snap_pending) {
 948                         ret = btrfs_wait_ordered_extents(root, 1);
 949                         BUG_ON(ret);
 950                 }
 951
 952                 /*
 953                  * rename don't use btrfs_join_transaction, so, once we
 954                  * set the transaction to blocked above, we aren't going
 955                  * to get any new ordered operations.  We can safely run
 956                  * it here and no for sure that nothing new will be added
 957                  * to the list
 958                  */
 959                 btrfs_run_ordered_operations(root, 1);
 960
 961                 smp_mb();
 962                 if (cur_trans->num_writers > 1 || should_grow)
 963                         schedule_timeout(timeout);
 964
 965                 mutex_lock(&root->fs_info->trans_mutex);
 966                 finish_wait(&cur_trans->writer_wait, &wait);
 967         } while (cur_trans->num_writers > 1 ||
 968                  (should_grow && cur_trans->num_joined != joined));
 969
 970         ret = create_pending_snapshots(trans, root->fs_info);
 971         BUG_ON(ret);
 972
 973         ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 974         BUG_ON(ret);
 975
 976         WARN_ON(cur_trans != trans->transaction);
 977
 978         /* btrfs_commit_tree_roots is responsible for getting the
 979          * various roots consistent with each other.  Every pointer
 980          * in the tree of tree roots has to point to the most up to date
 981          * root for every subvolume and other tree.  So, we have to keep
 982          * the tree logging code from jumping in and changing any
 983          * of the trees.
 984          *
 985          * At this point in the commit, there can't be any tree-log
 986          * writers, but a little lower down we drop the trans mutex
 987          * and let new people in.  By holding the tree_log_mutex
 988          * from now until after the super is written, we avoid races
 989          * with the tree-log code.
 990          */
 991         mutex_lock(&root->fs_info->tree_log_mutex);
 992
 993         ret = commit_fs_roots(trans, root);
 994         BUG_ON(ret);
 995
 996         /* commit_fs_roots gets rid of all the tree log roots, it is now
 997          * safe to free the root of tree log roots
 998          */
 999         btrfs_free_log_root_tree(trans, root->fs_info);
1000
1001         ret = commit_cowonly_roots(trans, root);
1002         BUG_ON(ret);
1003
1004         btrfs_prepare_extent_commit(trans, root);
1005
1006         cur_trans = root->fs_info->running_transaction;
1007         spin_lock(&root->fs_info->new_trans_lock);
1008         root->fs_info->running_transaction = NULL;
1009         spin_unlock(&root->fs_info->new_trans_lock);
1010
1011         btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1012                             root->fs_info->tree_root->node);
1013         switch_commit_root(root->fs_info->tree_root);
1014
1015         btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1016                             root->fs_info->chunk_root->node);
1017         switch_commit_root(root->fs_info->chunk_root);
1018
1019         update_super_roots(root);
1020
1021         if (!root->fs_info->log_root_recovering) {
1022                 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1023                 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1024         }
1025
1026         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1027                sizeof(root->fs_info->super_copy));
1028
1029         trans->transaction->blocked = 0;
1030
1031         wake_up(&root->fs_info->transaction_wait);
1032
1033         mutex_unlock(&root->fs_info->trans_mutex);
1034         ret = btrfs_write_and_wait_transaction(trans, root);
1035         BUG_ON(ret);
1036         write_ctree_super(trans, root, 0);
1037
1038         /*
1039          * the super is written, we can safely allow the tree-loggers
1040          * to go about their business
1041          */
1042         mutex_unlock(&root->fs_info->tree_log_mutex);
1043
1044         btrfs_finish_extent_commit(trans, root);
1045
1046         /* do the directory inserts of any pending snapshot creations */
1047         finish_pending_snapshots(trans, root->fs_info);
1048
1049         mutex_lock(&root->fs_info->trans_mutex);
1050
1051         cur_trans->commit_done = 1;
1052
1053         root->fs_info->last_trans_committed = cur_trans->transid;
1054
1055         wake_up(&cur_trans->commit_wait);
1056
1057         put_transaction(cur_trans);
1058         put_transaction(cur_trans);
1059
1060         mutex_unlock(&root->fs_info->trans_mutex);
1061
1062         kmem_cache_free(btrfs_trans_handle_cachep, trans);
1063         return ret;
1064 }
1065
1066 /*
1067  * interface function to delete all the snapshots we have scheduled for deletion
1068  */
1069 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1070 {
1071         LIST_HEAD(list);
1072         struct btrfs_fs_info *fs_info = root->fs_info;
1073
1074         mutex_lock(&fs_info->trans_mutex);
1075         list_splice_init(&fs_info->dead_roots, &list);
1076         mutex_unlock(&fs_info->trans_mutex);
1077
1078         while (!list_empty(&list)) {
1079                 root = list_entry(list.next, struct btrfs_root, root_list);
1080                 list_del(&root->root_list);
1081
1082                 if (btrfs_header_backref_rev(root->node) <
1083                     BTRFS_MIXED_BACKREF_REV)
1084                         btrfs_drop_snapshot(root, 0);
1085                 else
1086                         btrfs_drop_snapshot(root, 1);
1087         }
1088         return 0;
1089 }