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 #define BTRFS_ROOT_TRANS_TAG 0
  32
  33 static noinline void put_transaction(struct btrfs_transaction *transaction)
  34 {
  35         WARN_ON(transaction->use_count == 0);
  36         transaction->use_count--;
  37         if (transaction->use_count == 0) {
  38                 list_del_init(&transaction->list);
  39                 memset(transaction, 0, sizeof(*transaction));
  40                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  41         }
  42 }
  43
  44 /*
  45  * either allocate a new transaction or hop into the existing one
  46  */
  47 static noinline 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                 BUG_ON(!cur_trans);
  55                 root->fs_info->generation++;
  56                 root->fs_info->last_alloc = 0;
  57                 root->fs_info->last_data_alloc = 0;
  58                 cur_trans->num_writers = 1;
  59                 cur_trans->num_joined = 0;
  60                 cur_trans->transid = root->fs_info->generation;
  61                 init_waitqueue_head(&cur_trans->writer_wait);
  62                 init_waitqueue_head(&cur_trans->commit_wait);
  63                 cur_trans->in_commit = 0;
  64                 cur_trans->blocked = 0;
  65                 cur_trans->use_count = 1;
  66                 cur_trans->commit_done = 0;
  67                 cur_trans->start_time = get_seconds();
  68
  69                 cur_trans->delayed_refs.root.rb_node = NULL;
  70                 cur_trans->delayed_refs.num_entries = 0;
  71                 cur_trans->delayed_refs.num_heads_ready = 0;
  72                 cur_trans->delayed_refs.num_heads = 0;
  73                 cur_trans->delayed_refs.flushing = 0;
  74                 cur_trans->delayed_refs.run_delayed_start = 0;
  75                 spin_lock_init(&cur_trans->delayed_refs.lock);
  76
  77                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  78                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  79                 extent_io_tree_init(&cur_trans->dirty_pages,
  80                                      root->fs_info->btree_inode->i_mapping,
  81                                      GFP_NOFS);
  82                 spin_lock(&root->fs_info->new_trans_lock);
  83                 root->fs_info->running_transaction = cur_trans;
  84                 spin_unlock(&root->fs_info->new_trans_lock);
  85         } else {
  86                 cur_trans->num_writers++;
  87                 cur_trans->num_joined++;
  88         }
  89
  90         return 0;
  91 }
  92
  93 /*
  94  * this does all the record keeping required to make sure that a reference
  95  * counted root is properly recorded in a given transaction.  This is required
  96  * to make sure the old root from before we joined the transaction is deleted
  97  * when the transaction commits
  98  */
  99 noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
 100 {
 101         struct btrfs_dirty_root *dirty;
 102         u64 running_trans_id = root->fs_info->running_transaction->transid;
 103         if (root->ref_cows && root->last_trans < running_trans_id) {
 104                 WARN_ON(root == root->fs_info->extent_root);
 105                 if (root->root_item.refs != 0) {
 106                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
 107                                    (unsigned long)root->root_key.objectid,
 108                                    BTRFS_ROOT_TRANS_TAG);
 109
 110                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 111                         BUG_ON(!dirty);
 112                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 113                         BUG_ON(!dirty->root);
 114                         dirty->latest_root = root;
 115                         INIT_LIST_HEAD(&dirty->list);
 116
 117                         root->commit_root = btrfs_root_node(root);
 118
 119                         memcpy(dirty->root, root, sizeof(*root));
 120                         spin_lock_init(&dirty->root->node_lock);
 121                         spin_lock_init(&dirty->root->list_lock);
 122                         mutex_init(&dirty->root->objectid_mutex);
 123                         mutex_init(&dirty->root->log_mutex);
 124                         INIT_LIST_HEAD(&dirty->root->dead_list);
 125                         dirty->root->node = root->commit_root;
 126                         dirty->root->commit_root = NULL;
 127
 128                         spin_lock(&root->list_lock);
 129                         list_add(&dirty->root->dead_list, &root->dead_list);
 130                         spin_unlock(&root->list_lock);
 131
 132                         root->dirty_root = dirty;
 133                 } else {
 134                         WARN_ON(1);
 135                 }
 136                 root->last_trans = running_trans_id;
 137         }
 138         return 0;
 139 }
 140
 141 /* wait for commit against the current transaction to become unblocked
 142  * when this is done, it is safe to start a new transaction, but the current
 143  * transaction might not be fully on disk.
 144  */
 145 static void wait_current_trans(struct btrfs_root *root)
 146 {
 147         struct btrfs_transaction *cur_trans;
 148
 149         cur_trans = root->fs_info->running_transaction;
 150         if (cur_trans && cur_trans->blocked) {
 151                 DEFINE_WAIT(wait);
 152                 cur_trans->use_count++;
 153                 while (1) {
 154                         prepare_to_wait(&root->fs_info->transaction_wait, &wait,
 155                                         TASK_UNINTERRUPTIBLE);
 156                         if (cur_trans->blocked) {
 157                                 mutex_unlock(&root->fs_info->trans_mutex);
 158                                 schedule();
 159                                 mutex_lock(&root->fs_info->trans_mutex);
 160                                 finish_wait(&root->fs_info->transaction_wait,
 161                                             &wait);
 162                         } else {
 163                                 finish_wait(&root->fs_info->transaction_wait,
 164                                             &wait);
 165                                 break;
 166                         }
 167                 }
 168                 put_transaction(cur_trans);
 169         }
 170 }
 171
 172 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 173                                              int num_blocks, int wait)
 174 {
 175         struct btrfs_trans_handle *h =
 176                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 177         int ret;
 178
 179         mutex_lock(&root->fs_info->trans_mutex);
 180         if (!root->fs_info->log_root_recovering &&
 181             ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
 182                 wait_current_trans(root);
 183         ret = join_transaction(root);
 184         BUG_ON(ret);
 185
 186         btrfs_record_root_in_trans(root);
 187         h->transid = root->fs_info->running_transaction->transid;
 188         h->transaction = root->fs_info->running_transaction;
 189         h->blocks_reserved = num_blocks;
 190         h->blocks_used = 0;
 191         h->block_group = 0;
 192         h->alloc_exclude_nr = 0;
 193         h->alloc_exclude_start = 0;
 194         h->delayed_ref_updates = 0;
 195
 196         root->fs_info->running_transaction->use_count++;
 197         mutex_unlock(&root->fs_info->trans_mutex);
 198         return h;
 199 }
 200
 201 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 202                                                    int num_blocks)
 203 {
 204         return start_transaction(root, num_blocks, 1);
 205 }
 206 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
 207                                                    int num_blocks)
 208 {
 209         return start_transaction(root, num_blocks, 0);
 210 }
 211
 212 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
 213                                                          int num_blocks)
 214 {
 215         return start_transaction(r, num_blocks, 2);
 216 }
 217
 218 /* wait for a transaction commit to be fully complete */
 219 static noinline int wait_for_commit(struct btrfs_root *root,
 220                                     struct btrfs_transaction *commit)
 221 {
 222         DEFINE_WAIT(wait);
 223         mutex_lock(&root->fs_info->trans_mutex);
 224         while (!commit->commit_done) {
 225                 prepare_to_wait(&commit->commit_wait, &wait,
 226                                 TASK_UNINTERRUPTIBLE);
 227                 if (commit->commit_done)
 228                         break;
 229                 mutex_unlock(&root->fs_info->trans_mutex);
 230                 schedule();
 231                 mutex_lock(&root->fs_info->trans_mutex);
 232         }
 233         mutex_unlock(&root->fs_info->trans_mutex);
 234         finish_wait(&commit->commit_wait, &wait);
 235         return 0;
 236 }
 237
 238 /*
 239  * rate limit against the drop_snapshot code.  This helps to slow down new
 240  * operations if the drop_snapshot code isn't able to keep up.
 241  */
 242 static void throttle_on_drops(struct btrfs_root *root)
 243 {
 244         struct btrfs_fs_info *info = root->fs_info;
 245         int harder_count = 0;
 246
 247 harder:
 248         if (atomic_read(&info->throttles)) {
 249                 DEFINE_WAIT(wait);
 250                 int thr;
 251                 thr = atomic_read(&info->throttle_gen);
 252
 253                 do {
 254                         prepare_to_wait(&info->transaction_throttle,
 255                                         &wait, TASK_UNINTERRUPTIBLE);
 256                         if (!atomic_read(&info->throttles)) {
 257                                 finish_wait(&info->transaction_throttle, &wait);
 258                                 break;
 259                         }
 260                         schedule();
 261                         finish_wait(&info->transaction_throttle, &wait);
 262                 } while (thr == atomic_read(&info->throttle_gen));
 263                 harder_count++;
 264
 265                 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
 266                     harder_count < 2)
 267                         goto harder;
 268
 269                 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
 270                     harder_count < 10)
 271                         goto harder;
 272
 273                 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
 274                     harder_count < 20)
 275                         goto harder;
 276         }
 277 }
 278
 279 void btrfs_throttle(struct btrfs_root *root)
 280 {
 281         mutex_lock(&root->fs_info->trans_mutex);
 282         if (!root->fs_info->open_ioctl_trans)
 283                 wait_current_trans(root);
 284         mutex_unlock(&root->fs_info->trans_mutex);
 285         throttle_on_drops(root);
 286 }
 287
 288 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 289                           struct btrfs_root *root, int throttle)
 290 {
 291         struct btrfs_transaction *cur_trans;
 292         struct btrfs_fs_info *info = root->fs_info;
 293         int count = 0;
 294
 295         while (count < 4) {
 296                 unsigned long cur = trans->delayed_ref_updates;
 297                 trans->delayed_ref_updates = 0;
 298                 if (cur &&
 299                     trans->transaction->delayed_refs.num_heads_ready > 64) {
 300                         trans->delayed_ref_updates = 0;
 301
 302                         /*
 303                          * do a full flush if the transaction is trying
 304                          * to close
 305                          */
 306                         if (trans->transaction->delayed_refs.flushing)
 307                                 cur = 0;
 308                         btrfs_run_delayed_refs(trans, root, cur);
 309                 } else {
 310                         break;
 311                 }
 312                 count++;
 313         }
 314
 315         mutex_lock(&info->trans_mutex);
 316         cur_trans = info->running_transaction;
 317         WARN_ON(cur_trans != trans->transaction);
 318         WARN_ON(cur_trans->num_writers < 1);
 319         cur_trans->num_writers--;
 320
 321         if (waitqueue_active(&cur_trans->writer_wait))
 322                 wake_up(&cur_trans->writer_wait);
 323         put_transaction(cur_trans);
 324         mutex_unlock(&info->trans_mutex);
 325         memset(trans, 0, sizeof(*trans));
 326         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 327
 328         if (throttle)
 329                 throttle_on_drops(root);
 330
 331         return 0;
 332 }
 333
 334 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 335                           struct btrfs_root *root)
 336 {
 337         return __btrfs_end_transaction(trans, root, 0);
 338 }
 339
 340 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 341                                    struct btrfs_root *root)
 342 {
 343         return __btrfs_end_transaction(trans, root, 1);
 344 }
 345
 346 /*
 347  * when btree blocks are allocated, they have some corresponding bits set for
 348  * them in one of two extent_io trees.  This is used to make sure all of
 349  * those extents are on disk for transaction or log commit
 350  */
 351 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 352                                         struct extent_io_tree *dirty_pages)
 353 {
 354         int ret;
 355         int err = 0;
 356         int werr = 0;
 357         struct page *page;
 358         struct inode *btree_inode = root->fs_info->btree_inode;
 359         u64 start = 0;
 360         u64 end;
 361         unsigned long index;
 362
 363         while (1) {
 364                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
 365                                             EXTENT_DIRTY);
 366                 if (ret)
 367                         break;
 368                 while (start <= end) {
 369                         cond_resched();
 370
 371                         index = start >> PAGE_CACHE_SHIFT;
 372                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 373                         page = find_get_page(btree_inode->i_mapping, index);
 374                         if (!page)
 375                                 continue;
 376
 377                         btree_lock_page_hook(page);
 378                         if (!page->mapping) {
 379                                 unlock_page(page);
 380                                 page_cache_release(page);
 381                                 continue;
 382                         }
 383
 384                         if (PageWriteback(page)) {
 385                                 if (PageDirty(page))
 386                                         wait_on_page_writeback(page);
 387                                 else {
 388                                         unlock_page(page);
 389                                         page_cache_release(page);
 390                                         continue;
 391                                 }
 392                         }
 393                         err = write_one_page(page, 0);
 394                         if (err)
 395                                 werr = err;
 396                         page_cache_release(page);
 397                 }
 398         }
 399         while (1) {
 400                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 401                                             EXTENT_DIRTY);
 402                 if (ret)
 403                         break;
 404
 405                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 406                 while (start <= end) {
 407                         index = start >> PAGE_CACHE_SHIFT;
 408                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 409                         page = find_get_page(btree_inode->i_mapping, index);
 410                         if (!page)
 411                                 continue;
 412                         if (PageDirty(page)) {
 413                                 btree_lock_page_hook(page);
 414                                 wait_on_page_writeback(page);
 415                                 err = write_one_page(page, 0);
 416                                 if (err)
 417                                         werr = err;
 418                         }
 419                         wait_on_page_writeback(page);
 420                         page_cache_release(page);
 421                         cond_resched();
 422                 }
 423         }
 424         if (err)
 425                 werr = err;
 426         return werr;
 427 }
 428
 429 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 430                                      struct btrfs_root *root)
 431 {
 432         if (!trans || !trans->transaction) {
 433                 struct inode *btree_inode;
 434                 btree_inode = root->fs_info->btree_inode;
 435                 return filemap_write_and_wait(btree_inode->i_mapping);
 436         }
 437         return btrfs_write_and_wait_marked_extents(root,
 438                                            &trans->transaction->dirty_pages);
 439 }
 440
 441 /*
 442  * this is used to update the root pointer in the tree of tree roots.
 443  *
 444  * But, in the case of the extent allocation tree, updating the root
 445  * pointer may allocate blocks which may change the root of the extent
 446  * allocation tree.
 447  *
 448  * So, this loops and repeats and makes sure the cowonly root didn't
 449  * change while the root pointer was being updated in the metadata.
 450  */
 451 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 452                                struct btrfs_root *root)
 453 {
 454         int ret;
 455         u64 old_root_bytenr;
 456         struct btrfs_root *tree_root = root->fs_info->tree_root;
 457
 458         btrfs_write_dirty_block_groups(trans, root);
 459
 460         ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 461         BUG_ON(ret);
 462
 463         while (1) {
 464                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 465                 if (old_root_bytenr == root->node->start)
 466                         break;
 467                 btrfs_set_root_bytenr(&root->root_item,
 468                                        root->node->start);
 469                 btrfs_set_root_level(&root->root_item,
 470                                      btrfs_header_level(root->node));
 471                 btrfs_set_root_generation(&root->root_item, trans->transid);
 472
 473                 ret = btrfs_update_root(trans, tree_root,
 474                                         &root->root_key,
 475                                         &root->root_item);
 476                 BUG_ON(ret);
 477                 btrfs_write_dirty_block_groups(trans, root);
 478
 479                 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 480                 BUG_ON(ret);
 481         }
 482         return 0;
 483 }
 484
 485 /*
 486  * update all the cowonly tree roots on disk
 487  */
 488 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 489                             struct btrfs_root *root)
 490 {
 491         struct btrfs_fs_info *fs_info = root->fs_info;
 492         struct list_head *next;
 493         struct extent_buffer *eb;
 494         int ret;
 495
 496         ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 497         BUG_ON(ret);
 498
 499         eb = btrfs_lock_root_node(fs_info->tree_root);
 500         btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
 501         btrfs_tree_unlock(eb);
 502         free_extent_buffer(eb);
 503
 504         ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 505         BUG_ON(ret);
 506
 507         while (!list_empty(&fs_info->dirty_cowonly_roots)) {
 508                 next = fs_info->dirty_cowonly_roots.next;
 509                 list_del_init(next);
 510                 root = list_entry(next, struct btrfs_root, dirty_list);
 511
 512                 update_cowonly_root(trans, root);
 513
 514                 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
 515                 BUG_ON(ret);
 516         }
 517         return 0;
 518 }
 519
 520 /*
 521  * dead roots are old snapshots that need to be deleted.  This allocates
 522  * a dirty root struct and adds it into the list of dead roots that need to
 523  * be deleted
 524  */
 525 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
 526 {
 527         struct btrfs_dirty_root *dirty;
 528
 529         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 530         if (!dirty)
 531                 return -ENOMEM;
 532         dirty->root = root;
 533         dirty->latest_root = latest;
 534
 535         mutex_lock(&root->fs_info->trans_mutex);
 536         list_add(&dirty->list, &latest->fs_info->dead_roots);
 537         mutex_unlock(&root->fs_info->trans_mutex);
 538         return 0;
 539 }
 540
 541 /*
 542  * at transaction commit time we need to schedule the old roots for
 543  * deletion via btrfs_drop_snapshot.  This runs through all the
 544  * reference counted roots that were modified in the current
 545  * transaction and puts them into the drop list
 546  */
 547 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 548                                     struct radix_tree_root *radix,
 549                                     struct list_head *list)
 550 {
 551         struct btrfs_dirty_root *dirty;
 552         struct btrfs_root *gang[8];
 553         struct btrfs_root *root;
 554         int i;
 555         int ret;
 556         int err = 0;
 557         u32 refs;
 558
 559         while (1) {
 560                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 561                                                  ARRAY_SIZE(gang),
 562                                                  BTRFS_ROOT_TRANS_TAG);
 563                 if (ret == 0)
 564                         break;
 565                 for (i = 0; i < ret; i++) {
 566                         root = gang[i];
 567                         radix_tree_tag_clear(radix,
 568                                      (unsigned long)root->root_key.objectid,
 569                                      BTRFS_ROOT_TRANS_TAG);
 570
 571                         BUG_ON(!root->ref_tree);
 572                         dirty = root->dirty_root;
 573
 574                         btrfs_free_log(trans, root);
 575                         btrfs_free_reloc_root(trans, root);
 576
 577                         if (root->commit_root == root->node) {
 578                                 WARN_ON(root->node->start !=
 579                                         btrfs_root_bytenr(&root->root_item));
 580
 581                                 free_extent_buffer(root->commit_root);
 582                                 root->commit_root = NULL;
 583                                 root->dirty_root = NULL;
 584
 585                                 spin_lock(&root->list_lock);
 586                                 list_del_init(&dirty->root->dead_list);
 587                                 spin_unlock(&root->list_lock);
 588
 589                                 kfree(dirty->root);
 590                                 kfree(dirty);
 591
 592                                 /* make sure to update the root on disk
 593                                  * so we get any updates to the block used
 594                                  * counts
 595                                  */
 596                                 err = btrfs_update_root(trans,
 597                                                 root->fs_info->tree_root,
 598                                                 &root->root_key,
 599                                                 &root->root_item);
 600                                 continue;
 601                         }
 602
 603                         memset(&root->root_item.drop_progress, 0,
 604                                sizeof(struct btrfs_disk_key));
 605                         root->root_item.drop_level = 0;
 606                         root->commit_root = NULL;
 607                         root->dirty_root = NULL;
 608                         root->root_key.offset = root->fs_info->generation;
 609                         btrfs_set_root_bytenr(&root->root_item,
 610                                               root->node->start);
 611                         btrfs_set_root_level(&root->root_item,
 612                                              btrfs_header_level(root->node));
 613                         btrfs_set_root_generation(&root->root_item,
 614                                                   root->root_key.offset);
 615
 616                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 617                                                 &root->root_key,
 618                                                 &root->root_item);
 619                         if (err)
 620                                 break;
 621
 622                         refs = btrfs_root_refs(&dirty->root->root_item);
 623                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 624                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 625                                                 &dirty->root->root_key,
 626                                                 &dirty->root->root_item);
 627
 628                         BUG_ON(err);
 629                         if (refs == 1) {
 630                                 list_add(&dirty->list, list);
 631                         } else {
 632                                 WARN_ON(1);
 633                                 free_extent_buffer(dirty->root->node);
 634                                 kfree(dirty->root);
 635                                 kfree(dirty);
 636                         }
 637                 }
 638         }
 639         return err;
 640 }
 641
 642 /*
 643  * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 644  * otherwise every leaf in the btree is read and defragged.
 645  */
 646 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 647 {
 648         struct btrfs_fs_info *info = root->fs_info;
 649         int ret;
 650         struct btrfs_trans_handle *trans;
 651         unsigned long nr;
 652
 653         smp_mb();
 654         if (root->defrag_running)
 655                 return 0;
 656         trans = btrfs_start_transaction(root, 1);
 657         while (1) {
 658                 root->defrag_running = 1;
 659                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 660                 nr = trans->blocks_used;
 661                 btrfs_end_transaction(trans, root);
 662                 btrfs_btree_balance_dirty(info->tree_root, nr);
 663                 cond_resched();
 664
 665                 trans = btrfs_start_transaction(root, 1);
 666                 if (root->fs_info->closing || ret != -EAGAIN)
 667                         break;
 668         }
 669         root->defrag_running = 0;
 670         smp_mb();
 671         btrfs_end_transaction(trans, root);
 672         return 0;
 673 }
 674
 675 /*
 676  * when dropping snapshots, we generate a ton of delayed refs, and it makes
 677  * sense not to join the transaction while it is trying to flush the current
 678  * queue of delayed refs out.
 679  *
 680  * This is used by the drop snapshot code only
 681  */
 682 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
 683 {
 684         DEFINE_WAIT(wait);
 685
 686         mutex_lock(&info->trans_mutex);
 687         while (info->running_transaction &&
 688                info->running_transaction->delayed_refs.flushing) {
 689                 prepare_to_wait(&info->transaction_wait, &wait,
 690                                 TASK_UNINTERRUPTIBLE);
 691                 mutex_unlock(&info->trans_mutex);
 692                 schedule();
 693                 mutex_lock(&info->trans_mutex);
 694                 finish_wait(&info->transaction_wait, &wait);
 695         }
 696         mutex_unlock(&info->trans_mutex);
 697         return 0;
 698 }
 699
 700 /*
 701  * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
 702  * all of them
 703  */
 704 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 705                                      struct list_head *list)
 706 {
 707         struct btrfs_dirty_root *dirty;
 708         struct btrfs_trans_handle *trans;
 709         unsigned long nr;
 710         u64 num_bytes;
 711         u64 bytes_used;
 712         u64 max_useless;
 713         int ret = 0;
 714         int err;
 715
 716         while (!list_empty(list)) {
 717                 struct btrfs_root *root;
 718
 719                 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
 720                 list_del_init(&dirty->list);
 721
 722                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 723                 root = dirty->latest_root;
 724                 atomic_inc(&root->fs_info->throttles);
 725
 726                 while (1) {
 727                         /*
 728                          * we don't want to jump in and create a bunch of
 729                          * delayed refs if the transaction is starting to close
 730                          */
 731                         wait_transaction_pre_flush(tree_root->fs_info);
 732                         trans = btrfs_start_transaction(tree_root, 1);
 733
 734                         /*
 735                          * we've joined a transaction, make sure it isn't
 736                          * closing right now
 737                          */
 738                         if (trans->transaction->delayed_refs.flushing) {
 739                                 btrfs_end_transaction(trans, tree_root);
 740                                 continue;
 741                         }
 742
 743                         mutex_lock(&root->fs_info->drop_mutex);
 744                         ret = btrfs_drop_snapshot(trans, dirty->root);
 745                         if (ret != -EAGAIN)
 746                                 break;
 747                         mutex_unlock(&root->fs_info->drop_mutex);
 748
 749                         err = btrfs_update_root(trans,
 750                                         tree_root,
 751                                         &dirty->root->root_key,
 752                                         &dirty->root->root_item);
 753                         if (err)
 754                                 ret = err;
 755                         nr = trans->blocks_used;
 756                         ret = btrfs_end_transaction(trans, tree_root);
 757                         BUG_ON(ret);
 758
 759                         btrfs_btree_balance_dirty(tree_root, nr);
 760                         cond_resched();
 761                 }
 762                 BUG_ON(ret);
 763                 atomic_dec(&root->fs_info->throttles);
 764                 wake_up(&root->fs_info->transaction_throttle);
 765
 766                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 767                 bytes_used = btrfs_root_used(&root->root_item);
 768                 if (num_bytes) {
 769                         mutex_lock(&root->fs_info->trans_mutex);
 770                         btrfs_record_root_in_trans(root);
 771                         mutex_unlock(&root->fs_info->trans_mutex);
 772                         btrfs_set_root_used(&root->root_item,
 773                                             bytes_used - num_bytes);
 774                 }
 775
 776                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 777                 if (ret) {
 778                         BUG();
 779                         break;
 780                 }
 781                 mutex_unlock(&root->fs_info->drop_mutex);
 782
 783                 spin_lock(&root->list_lock);
 784                 list_del_init(&dirty->root->dead_list);
 785                 if (!list_empty(&root->dead_list)) {
 786                         struct btrfs_root *oldest;
 787                         oldest = list_entry(root->dead_list.prev,
 788                                             struct btrfs_root, dead_list);
 789                         max_useless = oldest->root_key.offset - 1;
 790                 } else {
 791                         max_useless = root->root_key.offset - 1;
 792                 }
 793                 spin_unlock(&root->list_lock);
 794
 795                 nr = trans->blocks_used;
 796                 ret = btrfs_end_transaction(trans, tree_root);
 797                 BUG_ON(ret);
 798
 799                 ret = btrfs_remove_leaf_refs(root, max_useless, 0);
 800                 BUG_ON(ret);
 801
 802                 free_extent_buffer(dirty->root->node);
 803                 kfree(dirty->root);
 804                 kfree(dirty);
 805
 806                 btrfs_btree_balance_dirty(tree_root, nr);
 807                 cond_resched();
 808         }
 809         return ret;
 810 }
 811
 812 /*
 813  * new snapshots need to be created at a very specific time in the
 814  * transaction commit.  This does the actual creation
 815  */
 816 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 817                                    struct btrfs_fs_info *fs_info,
 818                                    struct btrfs_pending_snapshot *pending)
 819 {
 820         struct btrfs_key key;
 821         struct btrfs_root_item *new_root_item;
 822         struct btrfs_root *tree_root = fs_info->tree_root;
 823         struct btrfs_root *root = pending->root;
 824         struct extent_buffer *tmp;
 825         struct extent_buffer *old;
 826         int ret;
 827         u64 objectid;
 828
 829         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 830         if (!new_root_item) {
 831                 ret = -ENOMEM;
 832                 goto fail;
 833         }
 834         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 835         if (ret)
 836                 goto fail;
 837
 838         btrfs_record_root_in_trans(root);
 839         btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
 840         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 841
 842         key.objectid = objectid;
 843         key.offset = trans->transid;
 844         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 845
 846         old = btrfs_lock_root_node(root);
 847         btrfs_cow_block(trans, root, old, NULL, 0, &old);
 848
 849         btrfs_copy_root(trans, root, old, &tmp, objectid);
 850         btrfs_tree_unlock(old);
 851         free_extent_buffer(old);
 852
 853         btrfs_set_root_bytenr(new_root_item, tmp->start);
 854         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 855         btrfs_set_root_generation(new_root_item, trans->transid);
 856         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 857                                 new_root_item);
 858         btrfs_tree_unlock(tmp);
 859         free_extent_buffer(tmp);
 860         if (ret)
 861                 goto fail;
 862
 863         key.offset = (u64)-1;
 864         memcpy(&pending->root_key, &key, sizeof(key));
 865 fail:
 866         kfree(new_root_item);
 867         return ret;
 868 }
 869
 870 static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
 871                                    struct btrfs_pending_snapshot *pending)
 872 {
 873         int ret;
 874         int namelen;
 875         u64 index = 0;
 876         struct btrfs_trans_handle *trans;
 877         struct inode *parent_inode;
 878         struct inode *inode;
 879         struct btrfs_root *parent_root;
 880
 881         parent_inode = pending->dentry->d_parent->d_inode;
 882         parent_root = BTRFS_I(parent_inode)->root;
 883         trans = btrfs_join_transaction(parent_root, 1);
 884
 885         /*
 886          * insert the directory item
 887          */
 888         namelen = strlen(pending->name);
 889         ret = btrfs_set_inode_index(parent_inode, &index);
 890         ret = btrfs_insert_dir_item(trans, parent_root,
 891                             pending->name, namelen,
 892                             parent_inode->i_ino,
 893                             &pending->root_key, BTRFS_FT_DIR, index);
 894
 895         if (ret)
 896                 goto fail;
 897
 898         btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
 899         ret = btrfs_update_inode(trans, parent_root, parent_inode);
 900         BUG_ON(ret);
 901
 902         /* add the backref first */
 903         ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
 904                                  pending->root_key.objectid,
 905                                  BTRFS_ROOT_BACKREF_KEY,
 906                                  parent_root->root_key.objectid,
 907                                  parent_inode->i_ino, index, pending->name,
 908                                  namelen);
 909
 910         BUG_ON(ret);
 911
 912         /* now add the forward ref */
 913         ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
 914                                  parent_root->root_key.objectid,
 915                                  BTRFS_ROOT_REF_KEY,
 916                                  pending->root_key.objectid,
 917                                  parent_inode->i_ino, index, pending->name,
 918                                  namelen);
 919
 920         inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
 921         d_instantiate(pending->dentry, inode);
 922 fail:
 923         btrfs_end_transaction(trans, fs_info->fs_root);
 924         return ret;
 925 }
 926
 927 /*
 928  * create all the snapshots we've scheduled for creation
 929  */
 930 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 931                                              struct btrfs_fs_info *fs_info)
 932 {
 933         struct btrfs_pending_snapshot *pending;
 934         struct list_head *head = &trans->transaction->pending_snapshots;
 935         int ret;
 936
 937         list_for_each_entry(pending, head, list) {
 938                 ret = create_pending_snapshot(trans, fs_info, pending);
 939                 BUG_ON(ret);
 940         }
 941         return 0;
 942 }
 943
 944 static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
 945                                              struct btrfs_fs_info *fs_info)
 946 {
 947         struct btrfs_pending_snapshot *pending;
 948         struct list_head *head = &trans->transaction->pending_snapshots;
 949         int ret;
 950
 951         while (!list_empty(head)) {
 952                 pending = list_entry(head->next,
 953                                      struct btrfs_pending_snapshot, list);
 954                 ret = finish_pending_snapshot(fs_info, pending);
 955                 BUG_ON(ret);
 956                 list_del(&pending->list);
 957                 kfree(pending->name);
 958                 kfree(pending);
 959         }
 960         return 0;
 961 }
 962
 963 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 964                              struct btrfs_root *root)
 965 {
 966         unsigned long joined = 0;
 967         unsigned long timeout = 1;
 968         struct btrfs_transaction *cur_trans;
 969         struct btrfs_transaction *prev_trans = NULL;
 970         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 971         struct list_head dirty_fs_roots;
 972         struct extent_io_tree *pinned_copy;
 973         DEFINE_WAIT(wait);
 974         int ret;
 975         int should_grow = 0;
 976         unsigned long now = get_seconds();
 977
 978         btrfs_run_ordered_operations(root, 0);
 979
 980         /* make a pass through all the delayed refs we have so far
 981          * any runnings procs may add more while we are here
 982          */
 983         ret = btrfs_run_delayed_refs(trans, root, 0);
 984         BUG_ON(ret);
 985
 986         cur_trans = trans->transaction;
 987         /*
 988          * set the flushing flag so procs in this transaction have to
 989          * start sending their work down.
 990          */
 991         cur_trans->delayed_refs.flushing = 1;
 992
 993         ret = btrfs_run_delayed_refs(trans, root, 0);
 994         BUG_ON(ret);
 995
 996         mutex_lock(&root->fs_info->trans_mutex);
 997         INIT_LIST_HEAD(&dirty_fs_roots);
 998         if (cur_trans->in_commit) {
 999                 cur_trans->use_count++;
1000                 mutex_unlock(&root->fs_info->trans_mutex);
1001                 btrfs_end_transaction(trans, root);
1002
1003                 ret = wait_for_commit(root, cur_trans);
1004                 BUG_ON(ret);
1005
1006                 mutex_lock(&root->fs_info->trans_mutex);
1007                 put_transaction(cur_trans);
1008                 mutex_unlock(&root->fs_info->trans_mutex);
1009
1010                 return 0;
1011         }
1012
1013         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
1014         if (!pinned_copy)
1015                 return -ENOMEM;
1016
1017         extent_io_tree_init(pinned_copy,
1018                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
1019
1020         trans->transaction->in_commit = 1;
1021         trans->transaction->blocked = 1;
1022         if (cur_trans->list.prev != &root->fs_info->trans_list) {
1023                 prev_trans = list_entry(cur_trans->list.prev,
1024                                         struct btrfs_transaction, list);
1025                 if (!prev_trans->commit_done) {
1026                         prev_trans->use_count++;
1027                         mutex_unlock(&root->fs_info->trans_mutex);
1028
1029                         wait_for_commit(root, prev_trans);
1030
1031                         mutex_lock(&root->fs_info->trans_mutex);
1032                         put_transaction(prev_trans);
1033                 }
1034         }
1035
1036         if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1037                 should_grow = 1;
1038
1039         do {
1040                 int snap_pending = 0;
1041                 joined = cur_trans->num_joined;
1042                 if (!list_empty(&trans->transaction->pending_snapshots))
1043                         snap_pending = 1;
1044
1045                 WARN_ON(cur_trans != trans->transaction);
1046                 prepare_to_wait(&cur_trans->writer_wait, &wait,
1047                                 TASK_UNINTERRUPTIBLE);
1048
1049                 if (cur_trans->num_writers > 1)
1050                         timeout = MAX_SCHEDULE_TIMEOUT;
1051                 else if (should_grow)
1052                         timeout = 1;
1053
1054                 mutex_unlock(&root->fs_info->trans_mutex);
1055
1056                 if (snap_pending) {
1057                         ret = btrfs_wait_ordered_extents(root, 1);
1058                         BUG_ON(ret);
1059                 }
1060
1061                 /*
1062                  * rename don't use btrfs_join_transaction, so, once we
1063                  * set the transaction to blocked above, we aren't going
1064                  * to get any new ordered operations.  We can safely run
1065                  * it here and no for sure that nothing new will be added
1066                  * to the list
1067                  */
1068                 btrfs_run_ordered_operations(root, 1);
1069
1070                 smp_mb();
1071                 if (cur_trans->num_writers > 1 || should_grow)
1072                         schedule_timeout(timeout);
1073
1074                 mutex_lock(&root->fs_info->trans_mutex);
1075                 finish_wait(&cur_trans->writer_wait, &wait);
1076         } while (cur_trans->num_writers > 1 ||
1077                  (should_grow && cur_trans->num_joined != joined));
1078
1079         ret = create_pending_snapshots(trans, root->fs_info);
1080         BUG_ON(ret);
1081
1082         ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1083         BUG_ON(ret);
1084
1085         WARN_ON(cur_trans != trans->transaction);
1086
1087         /* btrfs_commit_tree_roots is responsible for getting the
1088          * various roots consistent with each other.  Every pointer
1089          * in the tree of tree roots has to point to the most up to date
1090          * root for every subvolume and other tree.  So, we have to keep
1091          * the tree logging code from jumping in and changing any
1092          * of the trees.
1093          *
1094          * At this point in the commit, there can't be any tree-log
1095          * writers, but a little lower down we drop the trans mutex
1096          * and let new people in.  By holding the tree_log_mutex
1097          * from now until after the super is written, we avoid races
1098          * with the tree-log code.
1099          */
1100         mutex_lock(&root->fs_info->tree_log_mutex);
1101         /*
1102          * keep tree reloc code from adding new reloc trees
1103          */
1104         mutex_lock(&root->fs_info->tree_reloc_mutex);
1105
1106
1107         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
1108                               &dirty_fs_roots);
1109         BUG_ON(ret);
1110
1111         /* add_dirty_roots gets rid of all the tree log roots, it is now
1112          * safe to free the root of tree log roots
1113          */
1114         btrfs_free_log_root_tree(trans, root->fs_info);
1115
1116         ret = btrfs_commit_tree_roots(trans, root);
1117         BUG_ON(ret);
1118
1119         cur_trans = root->fs_info->running_transaction;
1120         spin_lock(&root->fs_info->new_trans_lock);
1121         root->fs_info->running_transaction = NULL;
1122         spin_unlock(&root->fs_info->new_trans_lock);
1123         btrfs_set_super_generation(&root->fs_info->super_copy,
1124                                    cur_trans->transid);
1125         btrfs_set_super_root(&root->fs_info->super_copy,
1126                              root->fs_info->tree_root->node->start);
1127         btrfs_set_super_root_level(&root->fs_info->super_copy,
1128                            btrfs_header_level(root->fs_info->tree_root->node));
1129
1130         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
1131                                    chunk_root->node->start);
1132         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
1133                                          btrfs_header_level(chunk_root->node));
1134         btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
1135                                 btrfs_header_generation(chunk_root->node));
1136
1137         if (!root->fs_info->log_root_recovering) {
1138                 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1139                 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1140         }
1141
1142         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1143                sizeof(root->fs_info->super_copy));
1144
1145         btrfs_copy_pinned(root, pinned_copy);
1146
1147         trans->transaction->blocked = 0;
1148
1149         wake_up(&root->fs_info->transaction_throttle);
1150         wake_up(&root->fs_info->transaction_wait);
1151
1152         mutex_unlock(&root->fs_info->trans_mutex);
1153         ret = btrfs_write_and_wait_transaction(trans, root);
1154         BUG_ON(ret);
1155         write_ctree_super(trans, root, 0);
1156
1157         /*
1158          * the super is written, we can safely allow the tree-loggers
1159          * to go about their business
1160          */
1161         mutex_unlock(&root->fs_info->tree_log_mutex);
1162
1163         btrfs_finish_extent_commit(trans, root, pinned_copy);
1164         kfree(pinned_copy);
1165
1166         btrfs_drop_dead_reloc_roots(root);
1167         mutex_unlock(&root->fs_info->tree_reloc_mutex);
1168
1169         /* do the directory inserts of any pending snapshot creations */
1170         finish_pending_snapshots(trans, root->fs_info);
1171
1172         mutex_lock(&root->fs_info->trans_mutex);
1173
1174         cur_trans->commit_done = 1;
1175
1176         root->fs_info->last_trans_committed = cur_trans->transid;
1177         wake_up(&cur_trans->commit_wait);
1178
1179         put_transaction(cur_trans);
1180         put_transaction(cur_trans);
1181
1182         list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
1183         if (root->fs_info->closing)
1184                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
1185
1186         mutex_unlock(&root->fs_info->trans_mutex);
1187
1188         kmem_cache_free(btrfs_trans_handle_cachep, trans);
1189
1190         if (root->fs_info->closing)
1191                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
1192         return ret;
1193 }
1194
1195 /*
1196  * interface function to delete all the snapshots we have scheduled for deletion
1197  */
1198 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1199 {
1200         struct list_head dirty_roots;
1201         INIT_LIST_HEAD(&dirty_roots);
1202 again:
1203         mutex_lock(&root->fs_info->trans_mutex);
1204         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
1205         mutex_unlock(&root->fs_info->trans_mutex);
1206
1207         if (!list_empty(&dirty_roots)) {
1208                 drop_dirty_roots(root, &dirty_roots);
1209                 goto again;
1210         }
1211         return 0;
1212 }