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 "ref-cache.h"
  29 #include "tree-log.h"
  30
  31 extern struct kmem_cache *btrfs_trans_handle_cachep;
  32 extern struct kmem_cache *btrfs_transaction_cachep;
  33
  34 #define BTRFS_ROOT_TRANS_TAG 0
  35
  36 static noinline void put_transaction(struct btrfs_transaction *transaction)
  37 {
  38         WARN_ON(transaction->use_count == 0);
  39         transaction->use_count--;
  40         if (transaction->use_count == 0) {
  41                 list_del_init(&transaction->list);
  42                 memset(transaction, 0, sizeof(*transaction));
  43                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  44         }
  45 }
  46
  47 /*
  48  * either allocate a new transaction or hop into the existing one
  49  */
  50 static noinline int join_transaction(struct btrfs_root *root)
  51 {
  52         struct btrfs_transaction *cur_trans;
  53         cur_trans = root->fs_info->running_transaction;
  54         if (!cur_trans) {
  55                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  56                                              GFP_NOFS);
  57                 BUG_ON(!cur_trans);
  58                 root->fs_info->generation++;
  59                 root->fs_info->last_alloc = 0;
  60                 root->fs_info->last_data_alloc = 0;
  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                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  72                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  73                 extent_io_tree_init(&cur_trans->dirty_pages,
  74                                      root->fs_info->btree_inode->i_mapping,
  75                                      GFP_NOFS);
  76                 spin_lock(&root->fs_info->new_trans_lock);
  77                 root->fs_info->running_transaction = cur_trans;
  78                 spin_unlock(&root->fs_info->new_trans_lock);
  79         } else {
  80                 cur_trans->num_writers++;
  81                 cur_trans->num_joined++;
  82         }
  83
  84         return 0;
  85 }
  86
  87 /*
  88  * this does all the record keeping required to make sure that a
  89  * reference counted root is properly recorded in a given transaction.
  90  * This is required to make sure the old root from before we joined the transaction
  91  * is deleted when the transaction commits
  92  */
  93 noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
  94 {
  95         struct btrfs_dirty_root *dirty;
  96         u64 running_trans_id = root->fs_info->running_transaction->transid;
  97         if (root->ref_cows && root->last_trans < running_trans_id) {
  98                 WARN_ON(root == root->fs_info->extent_root);
  99                 if (root->root_item.refs != 0) {
 100                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
 101                                    (unsigned long)root->root_key.objectid,
 102                                    BTRFS_ROOT_TRANS_TAG);
 103
 104                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 105                         BUG_ON(!dirty);
 106                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 107                         BUG_ON(!dirty->root);
 108                         dirty->latest_root = root;
 109                         INIT_LIST_HEAD(&dirty->list);
 110
 111                         root->commit_root = btrfs_root_node(root);
 112
 113                         memcpy(dirty->root, root, sizeof(*root));
 114                         spin_lock_init(&dirty->root->node_lock);
 115                         spin_lock_init(&dirty->root->list_lock);
 116                         mutex_init(&dirty->root->objectid_mutex);
 117                         mutex_init(&dirty->root->log_mutex);
 118                         INIT_LIST_HEAD(&dirty->root->dead_list);
 119                         dirty->root->node = root->commit_root;
 120                         dirty->root->commit_root = NULL;
 121
 122                         spin_lock(&root->list_lock);
 123                         list_add(&dirty->root->dead_list, &root->dead_list);
 124                         spin_unlock(&root->list_lock);
 125
 126                         root->dirty_root = dirty;
 127                 } else {
 128                         WARN_ON(1);
 129                 }
 130                 root->last_trans = running_trans_id;
 131         }
 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         btrfs_record_root_in_trans(root);
 181         h->transid = root->fs_info->running_transaction->transid;
 182         h->transaction = root->fs_info->running_transaction;
 183         h->blocks_reserved = num_blocks;
 184         h->blocks_used = 0;
 185         h->block_group = NULL;
 186         h->alloc_exclude_nr = 0;
 187         h->alloc_exclude_start = 0;
 188         root->fs_info->running_transaction->use_count++;
 189         mutex_unlock(&root->fs_info->trans_mutex);
 190         return h;
 191 }
 192
 193 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 194                                                    int num_blocks)
 195 {
 196         return start_transaction(root, num_blocks, 1);
 197 }
 198 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 199                                                    int num_blocks)
 200 {
 201         return start_transaction(root, num_blocks, 0);
 202 }
 203
 204 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
 205                                                          int num_blocks)
 206 {
 207         return start_transaction(r, num_blocks, 2);
 208 }
 209
 210 /* wait for a transaction commit to be fully complete */
 211 static noinline int wait_for_commit(struct btrfs_root *root,
 212                                     struct btrfs_transaction *commit)
 213 {
 214         DEFINE_WAIT(wait);
 215         mutex_lock(&root->fs_info->trans_mutex);
 216         while(!commit->commit_done) {
 217                 prepare_to_wait(&commit->commit_wait, &wait,
 218                                 TASK_UNINTERRUPTIBLE);
 219                 if (commit->commit_done)
 220                         break;
 221                 mutex_unlock(&root->fs_info->trans_mutex);
 222                 schedule();
 223                 mutex_lock(&root->fs_info->trans_mutex);
 224         }
 225         mutex_unlock(&root->fs_info->trans_mutex);
 226         finish_wait(&commit->commit_wait, &wait);
 227         return 0;
 228 }
 229
 230 /*
 231  * rate limit against the drop_snapshot code.  This helps to slow down new operations
 232  * if the drop_snapshot code isn't able to keep up.
 233  */
 234 static void throttle_on_drops(struct btrfs_root *root)
 235 {
 236         struct btrfs_fs_info *info = root->fs_info;
 237         int harder_count = 0;
 238
 239 harder:
 240         if (atomic_read(&info->throttles)) {
 241                 DEFINE_WAIT(wait);
 242                 int thr;
 243                 thr = atomic_read(&info->throttle_gen);
 244
 245                 do {
 246                         prepare_to_wait(&info->transaction_throttle,
 247                                         &wait, TASK_UNINTERRUPTIBLE);
 248                         if (!atomic_read(&info->throttles)) {
 249                                 finish_wait(&info->transaction_throttle, &wait);
 250                                 break;
 251                         }
 252                         schedule();
 253                         finish_wait(&info->transaction_throttle, &wait);
 254                 } while (thr == atomic_read(&info->throttle_gen));
 255                 harder_count++;
 256
 257                 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
 258                     harder_count < 2)
 259                         goto harder;
 260
 261                 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
 262                     harder_count < 10)
 263                         goto harder;
 264
 265                 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
 266                     harder_count < 20)
 267                         goto harder;
 268         }
 269 }
 270
 271 void btrfs_throttle(struct btrfs_root *root)
 272 {
 273         mutex_lock(&root->fs_info->trans_mutex);
 274         if (!root->fs_info->open_ioctl_trans)
 275                 wait_current_trans(root);
 276         mutex_unlock(&root->fs_info->trans_mutex);
 277
 278         throttle_on_drops(root);
 279 }
 280
 281 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 282                           struct btrfs_root *root, int throttle)
 283 {
 284         struct btrfs_transaction *cur_trans;
 285         struct btrfs_fs_info *info = root->fs_info;
 286
 287         mutex_lock(&info->trans_mutex);
 288         cur_trans = info->running_transaction;
 289         WARN_ON(cur_trans != trans->transaction);
 290         WARN_ON(cur_trans->num_writers < 1);
 291         cur_trans->num_writers--;
 292
 293         if (waitqueue_active(&cur_trans->writer_wait))
 294                 wake_up(&cur_trans->writer_wait);
 295         put_transaction(cur_trans);
 296         mutex_unlock(&info->trans_mutex);
 297         memset(trans, 0, sizeof(*trans));
 298         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 299
 300         if (throttle)
 301                 throttle_on_drops(root);
 302
 303         return 0;
 304 }
 305
 306 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 307                           struct btrfs_root *root)
 308 {
 309         return __btrfs_end_transaction(trans, root, 0);
 310 }
 311
 312 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 313                                    struct btrfs_root *root)
 314 {
 315         return __btrfs_end_transaction(trans, root, 1);
 316 }
 317
 318 /*
 319  * when btree blocks are allocated, they have some corresponding bits set for
 320  * them in one of two extent_io trees.  This is used to make sure all of
 321  * those extents are on disk for transaction or log commit
 322  */
 323 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 324                                         struct extent_io_tree *dirty_pages)
 325 {
 326         int ret;
 327         int err = 0;
 328         int werr = 0;
 329         struct page *page;
 330         struct inode *btree_inode = root->fs_info->btree_inode;
 331         u64 start = 0;
 332         u64 end;
 333         unsigned long index;
 334
 335         while(1) {
 336                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
 337                                             EXTENT_DIRTY);
 338                 if (ret)
 339                         break;
 340                 while(start <= end) {
 341                         cond_resched();
 342
 343                         index = start >> PAGE_CACHE_SHIFT;
 344                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 345                         page = find_get_page(btree_inode->i_mapping, index);
 346                         if (!page)
 347                                 continue;
 348
 349                         btree_lock_page_hook(page);
 350                         if (!page->mapping) {
 351                                 unlock_page(page);
 352                                 page_cache_release(page);
 353                                 continue;
 354                         }
 355
 356                         if (PageWriteback(page)) {
 357                                 if (PageDirty(page))
 358                                         wait_on_page_writeback(page);
 359                                 else {
 360                                         unlock_page(page);
 361                                         page_cache_release(page);
 362                                         continue;
 363                                 }
 364                         }
 365                         err = write_one_page(page, 0);
 366                         if (err)
 367                                 werr = err;
 368                         page_cache_release(page);
 369                 }
 370         }
 371         while(1) {
 372                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 373                                             EXTENT_DIRTY);
 374                 if (ret)
 375                         break;
 376
 377                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 378                 while(start <= end) {
 379                         index = start >> PAGE_CACHE_SHIFT;
 380                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 381                         page = find_get_page(btree_inode->i_mapping, index);
 382                         if (!page)
 383                                 continue;
 384                         if (PageDirty(page)) {
 385                                 btree_lock_page_hook(page);
 386                                 wait_on_page_writeback(page);
 387                                 err = write_one_page(page, 0);
 388                                 if (err)
 389                                         werr = err;
 390                         }
 391                         wait_on_page_writeback(page);
 392                         page_cache_release(page);
 393                         cond_resched();
 394                 }
 395         }
 396         if (err)
 397                 werr = err;
 398         return werr;
 399 }
 400
 401 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 402                                      struct btrfs_root *root)
 403 {
 404         if (!trans || !trans->transaction) {
 405                 struct inode *btree_inode;
 406                 btree_inode = root->fs_info->btree_inode;
 407                 return filemap_write_and_wait(btree_inode->i_mapping);
 408         }
 409         return btrfs_write_and_wait_marked_extents(root,
 410                                            &trans->transaction->dirty_pages);
 411 }
 412
 413 /*
 414  * this is used to update the root pointer in the tree of tree roots.
 415  *
 416  * But, in the case of the extent allocation tree, updating the root
 417  * pointer may allocate blocks which may change the root of the extent
 418  * allocation tree.
 419  *
 420  * So, this loops and repeats and makes sure the cowonly root didn't
 421  * change while the root pointer was being updated in the metadata.
 422  */
 423 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 424                                struct btrfs_root *root)
 425 {
 426         int ret;
 427         u64 old_root_bytenr;
 428         struct btrfs_root *tree_root = root->fs_info->tree_root;
 429
 430         btrfs_extent_post_op(trans, root);
 431         btrfs_write_dirty_block_groups(trans, root);
 432         btrfs_extent_post_op(trans, root);
 433
 434         while(1) {
 435                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 436                 if (old_root_bytenr == root->node->start)
 437                         break;
 438                 btrfs_set_root_bytenr(&root->root_item,
 439                                        root->node->start);
 440                 btrfs_set_root_level(&root->root_item,
 441                                      btrfs_header_level(root->node));
 442                 btrfs_set_root_generation(&root->root_item, trans->transid);
 443
 444                 btrfs_extent_post_op(trans, root);
 445
 446                 ret = btrfs_update_root(trans, tree_root,
 447                                         &root->root_key,
 448                                         &root->root_item);
 449                 BUG_ON(ret);
 450                 btrfs_write_dirty_block_groups(trans, root);
 451                 btrfs_extent_post_op(trans, root);
 452         }
 453         return 0;
 454 }
 455
 456 /*
 457  * update all the cowonly tree roots on disk
 458  */
 459 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 460                             struct btrfs_root *root)
 461 {
 462         struct btrfs_fs_info *fs_info = root->fs_info;
 463         struct list_head *next;
 464         struct extent_buffer *eb;
 465
 466         btrfs_extent_post_op(trans, fs_info->tree_root);
 467
 468         eb = btrfs_lock_root_node(fs_info->tree_root);
 469         btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0);
 470         btrfs_tree_unlock(eb);
 471         free_extent_buffer(eb);
 472
 473         btrfs_extent_post_op(trans, fs_info->tree_root);
 474
 475         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 476                 next = fs_info->dirty_cowonly_roots.next;
 477                 list_del_init(next);
 478                 root = list_entry(next, struct btrfs_root, dirty_list);
 479
 480                 update_cowonly_root(trans, root);
 481         }
 482         return 0;
 483 }
 484
 485 /*
 486  * dead roots are old snapshots that need to be deleted.  This allocates
 487  * a dirty root struct and adds it into the list of dead roots that need to
 488  * be deleted
 489  */
 490 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
 491 {
 492         struct btrfs_dirty_root *dirty;
 493
 494         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 495         if (!dirty)
 496                 return -ENOMEM;
 497         dirty->root = root;
 498         dirty->latest_root = latest;
 499
 500         mutex_lock(&root->fs_info->trans_mutex);
 501         list_add(&dirty->list, &latest->fs_info->dead_roots);
 502         mutex_unlock(&root->fs_info->trans_mutex);
 503         return 0;
 504 }
 505
 506 /*
 507  * at transaction commit time we need to schedule the old roots for
 508  * deletion via btrfs_drop_snapshot.  This runs through all the
 509  * reference counted roots that were modified in the current
 510  * transaction and puts them into the drop list
 511  */
 512 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 513                                     struct radix_tree_root *radix,
 514                                     struct list_head *list)
 515 {
 516         struct btrfs_dirty_root *dirty;
 517         struct btrfs_root *gang[8];
 518         struct btrfs_root *root;
 519         int i;
 520         int ret;
 521         int err = 0;
 522         u32 refs;
 523
 524         while(1) {
 525                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 526                                                  ARRAY_SIZE(gang),
 527                                                  BTRFS_ROOT_TRANS_TAG);
 528                 if (ret == 0)
 529                         break;
 530                 for (i = 0; i < ret; i++) {
 531                         root = gang[i];
 532                         radix_tree_tag_clear(radix,
 533                                      (unsigned long)root->root_key.objectid,
 534                                      BTRFS_ROOT_TRANS_TAG);
 535
 536                         BUG_ON(!root->ref_tree);
 537                         dirty = root->dirty_root;
 538
 539                         btrfs_free_log(trans, root);
 540                         btrfs_free_reloc_root(trans, root);
 541
 542                         if (root->commit_root == root->node) {
 543                                 WARN_ON(root->node->start !=
 544                                         btrfs_root_bytenr(&root->root_item));
 545
 546                                 free_extent_buffer(root->commit_root);
 547                                 root->commit_root = NULL;
 548                                 root->dirty_root = NULL;
 549
 550                                 spin_lock(&root->list_lock);
 551                                 list_del_init(&dirty->root->dead_list);
 552                                 spin_unlock(&root->list_lock);
 553
 554                                 kfree(dirty->root);
 555                                 kfree(dirty);
 556
 557                                 /* make sure to update the root on disk
 558                                  * so we get any updates to the block used
 559                                  * counts
 560                                  */
 561                                 err = btrfs_update_root(trans,
 562                                                 root->fs_info->tree_root,
 563                                                 &root->root_key,
 564                                                 &root->root_item);
 565                                 continue;
 566                         }
 567
 568                         memset(&root->root_item.drop_progress, 0,
 569                                sizeof(struct btrfs_disk_key));
 570                         root->root_item.drop_level = 0;
 571                         root->commit_root = NULL;
 572                         root->dirty_root = NULL;
 573                         root->root_key.offset = root->fs_info->generation;
 574                         btrfs_set_root_bytenr(&root->root_item,
 575                                               root->node->start);
 576                         btrfs_set_root_level(&root->root_item,
 577                                              btrfs_header_level(root->node));
 578                         btrfs_set_root_generation(&root->root_item,
 579                                                   root->root_key.offset);
 580
 581                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 582                                                 &root->root_key,
 583                                                 &root->root_item);
 584                         if (err)
 585                                 break;
 586
 587                         refs = btrfs_root_refs(&dirty->root->root_item);
 588                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 589                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 590                                                 &dirty->root->root_key,
 591                                                 &dirty->root->root_item);
 592
 593                         BUG_ON(err);
 594                         if (refs == 1) {
 595                                 list_add(&dirty->list, list);
 596                         } else {
 597                                 WARN_ON(1);
 598                                 free_extent_buffer(dirty->root->node);
 599                                 kfree(dirty->root);
 600                                 kfree(dirty);
 601                         }
 602                 }
 603         }
 604         return err;
 605 }
 606
 607 /*
 608  * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 609  * otherwise every leaf in the btree is read and defragged.
 610  */
 611 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 612 {
 613         struct btrfs_fs_info *info = root->fs_info;
 614         int ret;
 615         struct btrfs_trans_handle *trans;
 616         unsigned long nr;
 617
 618         smp_mb();
 619         if (root->defrag_running)
 620                 return 0;
 621         trans = btrfs_start_transaction(root, 1);
 622         while (1) {
 623                 root->defrag_running = 1;
 624                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 625                 nr = trans->blocks_used;
 626                 btrfs_end_transaction(trans, root);
 627                 btrfs_btree_balance_dirty(info->tree_root, nr);
 628                 cond_resched();
 629
 630                 trans = btrfs_start_transaction(root, 1);
 631                 if (root->fs_info->closing || ret != -EAGAIN)
 632                         break;
 633         }
 634         root->defrag_running = 0;
 635         smp_mb();
 636         btrfs_end_transaction(trans, root);
 637         return 0;
 638 }
 639
 640 /*
 641  * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
 642  * all of them
 643  */
 644 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 645                                      struct list_head *list)
 646 {
 647         struct btrfs_dirty_root *dirty;
 648         struct btrfs_trans_handle *trans;
 649         unsigned long nr;
 650         u64 num_bytes;
 651         u64 bytes_used;
 652         u64 max_useless;
 653         int ret = 0;
 654         int err;
 655
 656         while(!list_empty(list)) {
 657                 struct btrfs_root *root;
 658
 659                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 660                 list_del_init(&dirty->list);
 661
 662                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 663                 root = dirty->latest_root;
 664                 atomic_inc(&root->fs_info->throttles);
 665
 666                 while(1) {
 667                         trans = btrfs_start_transaction(tree_root, 1);
 668                         mutex_lock(&root->fs_info->drop_mutex);
 669                         ret = btrfs_drop_snapshot(trans, dirty->root);
 670                         if (ret != -EAGAIN) {
 671                                 break;
 672                         }
 673                         mutex_unlock(&root->fs_info->drop_mutex);
 674
 675                         err = btrfs_update_root(trans,
 676                                         tree_root,
 677                                         &dirty->root->root_key,
 678                                         &dirty->root->root_item);
 679                         if (err)
 680                                 ret = err;
 681                         nr = trans->blocks_used;
 682                         ret = btrfs_end_transaction(trans, tree_root);
 683                         BUG_ON(ret);
 684
 685                         btrfs_btree_balance_dirty(tree_root, nr);
 686                         cond_resched();
 687                 }
 688                 BUG_ON(ret);
 689                 atomic_dec(&root->fs_info->throttles);
 690                 wake_up(&root->fs_info->transaction_throttle);
 691
 692                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 693                 bytes_used = btrfs_root_used(&root->root_item);
 694                 if (num_bytes) {
 695                         btrfs_record_root_in_trans(root);
 696                         btrfs_set_root_used(&root->root_item,
 697                                             bytes_used - num_bytes);
 698                 }
 699
 700                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 701                 if (ret) {
 702                         BUG();
 703                         break;
 704                 }
 705                 mutex_unlock(&root->fs_info->drop_mutex);
 706
 707                 spin_lock(&root->list_lock);
 708                 list_del_init(&dirty->root->dead_list);
 709                 if (!list_empty(&root->dead_list)) {
 710                         struct btrfs_root *oldest;
 711                         oldest = list_entry(root->dead_list.prev,
 712                                             struct btrfs_root, dead_list);
 713                         max_useless = oldest->root_key.offset - 1;
 714                 } else {
 715                         max_useless = root->root_key.offset - 1;
 716                 }
 717                 spin_unlock(&root->list_lock);
 718
 719                 nr = trans->blocks_used;
 720                 ret = btrfs_end_transaction(trans, tree_root);
 721                 BUG_ON(ret);
 722
 723                 ret = btrfs_remove_leaf_refs(root, max_useless, 0);
 724                 BUG_ON(ret);
 725
 726                 free_extent_buffer(dirty->root->node);
 727                 kfree(dirty->root);
 728                 kfree(dirty);
 729
 730                 btrfs_btree_balance_dirty(tree_root, nr);
 731                 cond_resched();
 732         }
 733         return ret;
 734 }
 735
 736 /*
 737  * new snapshots need to be created at a very specific time in the
 738  * transaction commit.  This does the actual creation
 739  */
 740 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 741                                    struct btrfs_fs_info *fs_info,
 742                                    struct btrfs_pending_snapshot *pending)
 743 {
 744         struct btrfs_key key;
 745         struct btrfs_root_item *new_root_item;
 746         struct btrfs_root *tree_root = fs_info->tree_root;
 747         struct btrfs_root *root = pending->root;
 748         struct extent_buffer *tmp;
 749         struct extent_buffer *old;
 750         int ret;
 751         u64 objectid;
 752
 753         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 754         if (!new_root_item) {
 755                 ret = -ENOMEM;
 756                 goto fail;
 757         }
 758         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 759         if (ret)
 760                 goto fail;
 761
 762         btrfs_record_root_in_trans(root);
 763         btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
 764         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 765
 766         key.objectid = objectid;
 767         key.offset = trans->transid;
 768         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 769
 770         old = btrfs_lock_root_node(root);
 771         btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
 772
 773         btrfs_copy_root(trans, root, old, &tmp, objectid);
 774         btrfs_tree_unlock(old);
 775         free_extent_buffer(old);
 776
 777         btrfs_set_root_bytenr(new_root_item, tmp->start);
 778         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 779         btrfs_set_root_generation(new_root_item, trans->transid);
 780         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 781                                 new_root_item);
 782         btrfs_tree_unlock(tmp);
 783         free_extent_buffer(tmp);
 784         if (ret)
 785                 goto fail;
 786
 787         key.offset = (u64)-1;
 788         memcpy(&pending->root_key, &key, sizeof(key));
 789 fail:
 790         kfree(new_root_item);
 791         return ret;
 792 }
 793
 794 static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
 795                                    struct btrfs_pending_snapshot *pending)
 796 {
 797         int ret;
 798         int namelen;
 799         u64 index = 0;
 800         struct btrfs_trans_handle *trans;
 801         struct inode *parent_inode;
 802         struct inode *inode;
 803         struct btrfs_root *parent_root;
 804
 805         parent_inode = pending->dentry->d_parent->d_inode;
 806         parent_root = BTRFS_I(parent_inode)->root;
 807         trans = btrfs_start_transaction(parent_root, 1);
 808
 809         /*
 810          * insert the directory item
 811          */
 812         namelen = strlen(pending->name);
 813         ret = btrfs_set_inode_index(parent_inode, &index);
 814         ret = btrfs_insert_dir_item(trans, parent_root,
 815                             pending->name, namelen,
 816                             parent_inode->i_ino,
 817                             &pending->root_key, BTRFS_FT_DIR, index);
 818
 819         if (ret)
 820                 goto fail;
 821
 822         /* add the backref first */
 823         ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
 824                                  pending->root_key.objectid,
 825                                  BTRFS_ROOT_BACKREF_KEY,
 826                                  parent_root->root_key.objectid,
 827                                  parent_inode->i_ino, index, pending->name,
 828                                  namelen);
 829
 830         BUG_ON(ret);
 831
 832         /* now add the forward ref */
 833         ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
 834                                  parent_root->root_key.objectid,
 835                                  BTRFS_ROOT_REF_KEY,
 836                                  pending->root_key.objectid,
 837                                  parent_inode->i_ino, index, pending->name,
 838                                  namelen);
 839
 840         inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
 841         d_instantiate(pending->dentry, inode);
 842 fail:
 843         btrfs_end_transaction(trans, fs_info->fs_root);
 844         return ret;
 845 }
 846
 847 /*
 848  * create all the snapshots we've scheduled for creation
 849  */
 850 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 851                                              struct btrfs_fs_info *fs_info)
 852 {
 853         struct btrfs_pending_snapshot *pending;
 854         struct list_head *head = &trans->transaction->pending_snapshots;
 855         struct list_head *cur;
 856         int ret;
 857
 858         list_for_each(cur, head) {
 859                 pending = list_entry(cur, struct btrfs_pending_snapshot, list);
 860                 ret = create_pending_snapshot(trans, fs_info, pending);
 861                 BUG_ON(ret);
 862         }
 863         return 0;
 864 }
 865
 866 static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
 867                                              struct btrfs_fs_info *fs_info)
 868 {
 869         struct btrfs_pending_snapshot *pending;
 870         struct list_head *head = &trans->transaction->pending_snapshots;
 871         int ret;
 872
 873         while(!list_empty(head)) {
 874                 pending = list_entry(head->next,
 875                                      struct btrfs_pending_snapshot, list);
 876                 ret = finish_pending_snapshot(fs_info, pending);
 877                 BUG_ON(ret);
 878                 list_del(&pending->list);
 879                 kfree(pending->name);
 880                 kfree(pending);
 881         }
 882         return 0;
 883 }
 884
 885 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 886                              struct btrfs_root *root)
 887 {
 888         unsigned long joined = 0;
 889         unsigned long timeout = 1;
 890         struct btrfs_transaction *cur_trans;
 891         struct btrfs_transaction *prev_trans = NULL;
 892         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 893         struct list_head dirty_fs_roots;
 894         struct extent_io_tree *pinned_copy;
 895         DEFINE_WAIT(wait);
 896         int ret;
 897
 898         INIT_LIST_HEAD(&dirty_fs_roots);
 899         mutex_lock(&root->fs_info->trans_mutex);
 900         if (trans->transaction->in_commit) {
 901                 cur_trans = trans->transaction;
 902                 trans->transaction->use_count++;
 903                 mutex_unlock(&root->fs_info->trans_mutex);
 904                 btrfs_end_transaction(trans, root);
 905
 906                 ret = wait_for_commit(root, cur_trans);
 907                 BUG_ON(ret);
 908
 909                 mutex_lock(&root->fs_info->trans_mutex);
 910                 put_transaction(cur_trans);
 911                 mutex_unlock(&root->fs_info->trans_mutex);
 912
 913                 return 0;
 914         }
 915
 916         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 917         if (!pinned_copy)
 918                 return -ENOMEM;
 919
 920         extent_io_tree_init(pinned_copy,
 921                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 922
 923         trans->transaction->in_commit = 1;
 924         trans->transaction->blocked = 1;
 925         cur_trans = trans->transaction;
 926         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 927                 prev_trans = list_entry(cur_trans->list.prev,
 928                                         struct btrfs_transaction, list);
 929                 if (!prev_trans->commit_done) {
 930                         prev_trans->use_count++;
 931                         mutex_unlock(&root->fs_info->trans_mutex);
 932
 933                         wait_for_commit(root, prev_trans);
 934
 935                         mutex_lock(&root->fs_info->trans_mutex);
 936                         put_transaction(prev_trans);
 937                 }
 938         }
 939
 940         do {
 941                 int snap_pending = 0;
 942                 joined = cur_trans->num_joined;
 943                 if (!list_empty(&trans->transaction->pending_snapshots))
 944                         snap_pending = 1;
 945
 946                 WARN_ON(cur_trans != trans->transaction);
 947                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 948                                 TASK_UNINTERRUPTIBLE);
 949
 950                 if (cur_trans->num_writers > 1)
 951                         timeout = MAX_SCHEDULE_TIMEOUT;
 952                 else
 953                         timeout = 1;
 954
 955                 mutex_unlock(&root->fs_info->trans_mutex);
 956
 957                 if (snap_pending) {
 958                         ret = btrfs_wait_ordered_extents(root, 1);
 959                         BUG_ON(ret);
 960                 }
 961
 962                 schedule_timeout(timeout);
 963
 964                 mutex_lock(&root->fs_info->trans_mutex);
 965                 finish_wait(&cur_trans->writer_wait, &wait);
 966         } while (cur_trans->num_writers > 1 ||
 967                  (cur_trans->num_joined != joined));
 968
 969         ret = create_pending_snapshots(trans, root->fs_info);
 970         BUG_ON(ret);
 971
 972         WARN_ON(cur_trans != trans->transaction);
 973
 974         /* btrfs_commit_tree_roots is responsible for getting the
 975          * various roots consistent with each other.  Every pointer
 976          * in the tree of tree roots has to point to the most up to date
 977          * root for every subvolume and other tree.  So, we have to keep
 978          * the tree logging code from jumping in and changing any
 979          * of the trees.
 980          *
 981          * At this point in the commit, there can't be any tree-log
 982          * writers, but a little lower down we drop the trans mutex
 983          * and let new people in.  By holding the tree_log_mutex
 984          * from now until after the super is written, we avoid races
 985          * with the tree-log code.
 986          */
 987         mutex_lock(&root->fs_info->tree_log_mutex);
 988         /*
 989          * keep tree reloc code from adding new reloc trees
 990          */
 991         mutex_lock(&root->fs_info->tree_reloc_mutex);
 992
 993
 994         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 995                               &dirty_fs_roots);
 996         BUG_ON(ret);
 997
 998         /* add_dirty_roots gets rid of all the tree log roots, it is now
 999          * safe to free the root of tree log roots
1000          */
1001         btrfs_free_log_root_tree(trans, root->fs_info);
1002
1003         ret = btrfs_commit_tree_roots(trans, root);
1004         BUG_ON(ret);
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         btrfs_set_super_generation(&root->fs_info->super_copy,
1011                                    cur_trans->transid);
1012         btrfs_set_super_root(&root->fs_info->super_copy,
1013                              root->fs_info->tree_root->node->start);
1014         btrfs_set_super_root_level(&root->fs_info->super_copy,
1015                            btrfs_header_level(root->fs_info->tree_root->node));
1016
1017         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
1018                                    chunk_root->node->start);
1019         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
1020                                          btrfs_header_level(chunk_root->node));
1021         btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
1022                                 btrfs_header_generation(chunk_root->node));
1023
1024         if (!root->fs_info->log_root_recovering) {
1025                 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1026                 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1027         }
1028
1029         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1030                sizeof(root->fs_info->super_copy));
1031
1032         btrfs_copy_pinned(root, pinned_copy);
1033
1034         trans->transaction->blocked = 0;
1035         wake_up(&root->fs_info->transaction_throttle);
1036         wake_up(&root->fs_info->transaction_wait);
1037
1038         mutex_unlock(&root->fs_info->trans_mutex);
1039         ret = btrfs_write_and_wait_transaction(trans, root);
1040         BUG_ON(ret);
1041         write_ctree_super(trans, root, 0);
1042
1043         /*
1044          * the super is written, we can safely allow the tree-loggers
1045          * to go about their business
1046          */
1047         mutex_unlock(&root->fs_info->tree_log_mutex);
1048
1049         btrfs_finish_extent_commit(trans, root, pinned_copy);
1050         kfree(pinned_copy);
1051
1052         btrfs_drop_dead_reloc_roots(root);
1053         mutex_unlock(&root->fs_info->tree_reloc_mutex);
1054
1055         /* do the directory inserts of any pending snapshot creations */
1056         finish_pending_snapshots(trans, root->fs_info);
1057
1058         mutex_lock(&root->fs_info->trans_mutex);
1059
1060         cur_trans->commit_done = 1;
1061         root->fs_info->last_trans_committed = cur_trans->transid;
1062         wake_up(&cur_trans->commit_wait);
1063
1064         put_transaction(cur_trans);
1065         put_transaction(cur_trans);
1066
1067         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
1068         if (root->fs_info->closing)
1069                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
1070
1071         mutex_unlock(&root->fs_info->trans_mutex);
1072
1073         kmem_cache_free(btrfs_trans_handle_cachep, trans);
1074
1075         if (root->fs_info->closing) {
1076                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
1077         }
1078         return ret;
1079 }
1080
1081 /*
1082  * interface function to delete all the snapshots we have scheduled for deletion
1083  */
1084 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1085 {
1086         struct list_head dirty_roots;
1087         INIT_LIST_HEAD(&dirty_roots);
1088 again:
1089         mutex_lock(&root->fs_info->trans_mutex);
1090         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
1091         mutex_unlock(&root->fs_info->trans_mutex);
1092
1093         if (!list_empty(&dirty_roots)) {
1094                 drop_dirty_roots(root, &dirty_roots);
1095                 goto again;
1096         }
1097         return 0;
1098 }