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/slab.h> 21#include <linux/sched.h> 22#include <linux/writeback.h> 23#include <linux/pagemap.h> 24#include <linux/blkdev.h> 25#include "ctree.h" 26#include "disk-io.h" 27#include "transaction.h" 28#include "locking.h" 29#include "tree-log.h" 30 31#define BTRFS_ROOT_TRANS_TAG 0 32 33static 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 44static noinline void switch_commit_root(struct btrfs_root *root) 45{ 46 free_extent_buffer(root->commit_root); 47 root->commit_root = btrfs_root_node(root); 48} 49 50/* 51 * either allocate a new transaction or hop into the existing one 52 */ 53static noinline int join_transaction(struct btrfs_root *root) 54{ 55 struct btrfs_transaction *cur_trans; 56 cur_trans = root->fs_info->running_transaction; 57 if (!cur_trans) { 58 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, 59 GFP_NOFS); 60 BUG_ON(!cur_trans); 61 root->fs_info->generation++; 62 cur_trans->num_writers = 1; 63 cur_trans->num_joined = 0; 64 cur_trans->transid = root->fs_info->generation; 65 init_waitqueue_head(&cur_trans->writer_wait); 66 init_waitqueue_head(&cur_trans->commit_wait); 67 cur_trans->in_commit = 0; 68 cur_trans->blocked = 0; 69 cur_trans->use_count = 1; 70 cur_trans->commit_done = 0; 71 cur_trans->start_time = get_seconds(); 72 73 cur_trans->delayed_refs.root = RB_ROOT; 74 cur_trans->delayed_refs.num_entries = 0; 75 cur_trans->delayed_refs.num_heads_ready = 0; 76 cur_trans->delayed_refs.num_heads = 0; 77 cur_trans->delayed_refs.flushing = 0; 78 cur_trans->delayed_refs.run_delayed_start = 0; 79 spin_lock_init(&cur_trans->delayed_refs.lock); 80 81 INIT_LIST_HEAD(&cur_trans->pending_snapshots); 82 list_add_tail(&cur_trans->list, &root->fs_info->trans_list); 83 extent_io_tree_init(&cur_trans->dirty_pages, 84 root->fs_info->btree_inode->i_mapping, 85 GFP_NOFS); 86 spin_lock(&root->fs_info->new_trans_lock); 87 root->fs_info->running_transaction = cur_trans; 88 spin_unlock(&root->fs_info->new_trans_lock); 89 } else { 90 cur_trans->num_writers++; 91 cur_trans->num_joined++; 92 } 93 94 return 0; 95} 96 97/* 98 * this does all the record keeping required to make sure that a reference 99 * counted root is properly recorded in a given transaction. This is required 100 * to make sure the old root from before we joined the transaction is deleted 101 * when the transaction commits 102 */ 103static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, 104 struct btrfs_root *root) 105{ 106 if (root->ref_cows && root->last_trans < trans->transid) { 107 WARN_ON(root == root->fs_info->extent_root); 108 WARN_ON(root->commit_root != root->node); 109 110 radix_tree_tag_set(&root->fs_info->fs_roots_radix, 111 (unsigned long)root->root_key.objectid, 112 BTRFS_ROOT_TRANS_TAG); 113 root->last_trans = trans->transid; 114 btrfs_init_reloc_root(trans, root); 115 } 116 return 0; 117} 118 119int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 120 struct btrfs_root *root) 121{ 122 if (!root->ref_cows) 123 return 0; 124 125 mutex_lock(&root->fs_info->trans_mutex); 126 if (root->last_trans == trans->transid) { 127 mutex_unlock(&root->fs_info->trans_mutex); 128 return 0; 129 } 130 131 record_root_in_trans(trans, root); 132 mutex_unlock(&root->fs_info->trans_mutex); 133 return 0; 134} 135 136/* wait for commit against the current transaction to become unblocked 137 * when this is done, it is safe to start a new transaction, but the current 138 * transaction might not be fully on disk. 139 */ 140static void wait_current_trans(struct btrfs_root *root) 141{ 142 struct btrfs_transaction *cur_trans; 143 144 cur_trans = root->fs_info->running_transaction; 145 if (cur_trans && cur_trans->blocked) { 146 DEFINE_WAIT(wait); 147 cur_trans->use_count++; 148 while (1) { 149 prepare_to_wait(&root->fs_info->transaction_wait, &wait, 150 TASK_UNINTERRUPTIBLE); 151 if (!cur_trans->blocked) 152 break; 153 mutex_unlock(&root->fs_info->trans_mutex); 154 schedule(); 155 mutex_lock(&root->fs_info->trans_mutex); 156 } 157 finish_wait(&root->fs_info->transaction_wait, &wait); 158 put_transaction(cur_trans); 159 } 160} 161 162enum btrfs_trans_type { 163 TRANS_START, 164 TRANS_JOIN, 165 TRANS_USERSPACE, 166}; 167 168static int may_wait_transaction(struct btrfs_root *root, int type) 169{ 170 if (!root->fs_info->log_root_recovering && 171 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || 172 type == TRANS_USERSPACE)) 173 return 1; 174 return 0; 175} 176 177static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 178 u64 num_items, int type) 179{ 180 struct btrfs_trans_handle *h; 181 struct btrfs_transaction *cur_trans; 182 int retries = 0; 183 int ret; 184again: 185 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); 186 if (!h) 187 return ERR_PTR(-ENOMEM); 188 189 mutex_lock(&root->fs_info->trans_mutex); 190 if (may_wait_transaction(root, type)) 191 wait_current_trans(root); 192 193 ret = join_transaction(root); 194 BUG_ON(ret); 195 196 cur_trans = root->fs_info->running_transaction; 197 cur_trans->use_count++; 198 mutex_unlock(&root->fs_info->trans_mutex); 199 200 h->transid = cur_trans->transid; 201 h->transaction = cur_trans; 202 h->blocks_used = 0; 203 h->block_group = 0; 204 h->bytes_reserved = 0; 205 h->delayed_ref_updates = 0; 206 h->block_rsv = NULL; 207 208 smp_mb(); 209 if (cur_trans->blocked && may_wait_transaction(root, type)) { 210 btrfs_commit_transaction(h, root); 211 goto again; 212 } 213 214 if (num_items > 0) { 215 ret = btrfs_trans_reserve_metadata(h, root, num_items, 216 &retries); 217 if (ret == -EAGAIN) { 218 btrfs_commit_transaction(h, root); 219 goto again; 220 } 221 if (ret < 0) { 222 btrfs_end_transaction(h, root); 223 return ERR_PTR(ret); 224 } 225 } 226 227 mutex_lock(&root->fs_info->trans_mutex); 228 record_root_in_trans(h, root); 229 mutex_unlock(&root->fs_info->trans_mutex); 230 231 if (!current->journal_info && type != TRANS_USERSPACE) 232 current->journal_info = h; 233 return h; 234} 235 236struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 237 int num_items) 238{ 239 return start_transaction(root, num_items, TRANS_START); 240} 241struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, 242 int num_blocks) 243{ 244 return start_transaction(root, 0, TRANS_JOIN); 245} 246 247struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, 248 int num_blocks) 249{ 250 return start_transaction(r, 0, TRANS_USERSPACE); 251} 252 253/* wait for a transaction commit to be fully complete */ 254static noinline int wait_for_commit(struct btrfs_root *root, 255 struct btrfs_transaction *commit) 256{ 257 DEFINE_WAIT(wait); 258 mutex_lock(&root->fs_info->trans_mutex); 259 while (!commit->commit_done) { 260 prepare_to_wait(&commit->commit_wait, &wait, 261 TASK_UNINTERRUPTIBLE); 262 if (commit->commit_done) 263 break; 264 mutex_unlock(&root->fs_info->trans_mutex); 265 schedule(); 266 mutex_lock(&root->fs_info->trans_mutex); 267 } 268 mutex_unlock(&root->fs_info->trans_mutex); 269 finish_wait(&commit->commit_wait, &wait); 270 return 0; 271} 272 273 274void btrfs_throttle(struct btrfs_root *root) 275{ 276 mutex_lock(&root->fs_info->trans_mutex); 277 if (!root->fs_info->open_ioctl_trans) 278 wait_current_trans(root); 279 mutex_unlock(&root->fs_info->trans_mutex); 280} 281 282static int should_end_transaction(struct btrfs_trans_handle *trans, 283 struct btrfs_root *root) 284{ 285 int ret; 286 ret = btrfs_block_rsv_check(trans, root, 287 &root->fs_info->global_block_rsv, 0, 5); 288 return ret ? 1 : 0; 289} 290 291int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, 292 struct btrfs_root *root) 293{ 294 struct btrfs_transaction *cur_trans = trans->transaction; 295 int updates; 296 297 if (cur_trans->blocked || cur_trans->delayed_refs.flushing) 298 return 1; 299 300 updates = trans->delayed_ref_updates; 301 trans->delayed_ref_updates = 0; 302 if (updates) 303 btrfs_run_delayed_refs(trans, root, updates); 304 305 return should_end_transaction(trans, root); 306} 307 308static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 309 struct btrfs_root *root, int throttle) 310{ 311 struct btrfs_transaction *cur_trans = trans->transaction; 312 struct btrfs_fs_info *info = root->fs_info; 313 int count = 0; 314 315 while (count < 4) { 316 unsigned long cur = trans->delayed_ref_updates; 317 trans->delayed_ref_updates = 0; 318 if (cur && 319 trans->transaction->delayed_refs.num_heads_ready > 64) { 320 trans->delayed_ref_updates = 0; 321 322 /* 323 * do a full flush if the transaction is trying 324 * to close 325 */ 326 if (trans->transaction->delayed_refs.flushing) 327 cur = 0; 328 btrfs_run_delayed_refs(trans, root, cur); 329 } else { 330 break; 331 } 332 count++; 333 } 334 335 btrfs_trans_release_metadata(trans, root); 336 337 if (!root->fs_info->open_ioctl_trans && 338 should_end_transaction(trans, root)) 339 trans->transaction->blocked = 1; 340 341 if (cur_trans->blocked && !cur_trans->in_commit) { 342 if (throttle) 343 return btrfs_commit_transaction(trans, root); 344 else 345 wake_up_process(info->transaction_kthread); 346 } 347 348 mutex_lock(&info->trans_mutex); 349 WARN_ON(cur_trans != info->running_transaction); 350 WARN_ON(cur_trans->num_writers < 1); 351 cur_trans->num_writers--; 352 353 if (waitqueue_active(&cur_trans->writer_wait)) 354 wake_up(&cur_trans->writer_wait); 355 put_transaction(cur_trans); 356 mutex_unlock(&info->trans_mutex); 357 358 if (current->journal_info == trans) 359 current->journal_info = NULL; 360 memset(trans, 0, sizeof(*trans)); 361 kmem_cache_free(btrfs_trans_handle_cachep, trans); 362 363 if (throttle) 364 btrfs_run_delayed_iputs(root); 365 366 return 0; 367} 368 369int btrfs_end_transaction(struct btrfs_trans_handle *trans, 370 struct btrfs_root *root) 371{ 372 return __btrfs_end_transaction(trans, root, 0); 373} 374 375int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, 376 struct btrfs_root *root) 377{ 378 return __btrfs_end_transaction(trans, root, 1); 379} 380 381/* 382 * when btree blocks are allocated, they have some corresponding bits set for 383 * them in one of two extent_io trees. This is used to make sure all of 384 * those extents are sent to disk but does not wait on them 385 */ 386int btrfs_write_marked_extents(struct btrfs_root *root, 387 struct extent_io_tree *dirty_pages, int mark) 388{ 389 int ret; 390 int err = 0; 391 int werr = 0; 392 struct page *page; 393 struct inode *btree_inode = root->fs_info->btree_inode; 394 u64 start = 0; 395 u64 end; 396 unsigned long index; 397 398 while (1) { 399 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 400 mark); 401 if (ret) 402 break; 403 while (start <= end) { 404 cond_resched(); 405 406 index = start >> PAGE_CACHE_SHIFT; 407 start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 408 page = find_get_page(btree_inode->i_mapping, index); 409 if (!page) 410 continue; 411 412 btree_lock_page_hook(page); 413 if (!page->mapping) { 414 unlock_page(page); 415 page_cache_release(page); 416 continue; 417 } 418 419 if (PageWriteback(page)) { 420 if (PageDirty(page)) 421 wait_on_page_writeback(page); 422 else { 423 unlock_page(page); 424 page_cache_release(page); 425 continue; 426 } 427 } 428 err = write_one_page(page, 0); 429 if (err) 430 werr = err; 431 page_cache_release(page); 432 } 433 } 434 if (err) 435 werr = err; 436 return werr; 437} 438 439/* 440 * when btree blocks are allocated, they have some corresponding bits set for 441 * them in one of two extent_io trees. This is used to make sure all of 442 * those extents are on disk for transaction or log commit. We wait 443 * on all the pages and clear them from the dirty pages state tree 444 */ 445int btrfs_wait_marked_extents(struct btrfs_root *root, 446 struct extent_io_tree *dirty_pages, int mark) 447{ 448 int ret; 449 int err = 0; 450 int werr = 0; 451 struct page *page; 452 struct inode *btree_inode = root->fs_info->btree_inode; 453 u64 start = 0; 454 u64 end; 455 unsigned long index; 456 457 while (1) { 458 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 459 mark); 460 if (ret) 461 break; 462 463 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 464 while (start <= end) { 465 index = start >> PAGE_CACHE_SHIFT; 466 start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 467 page = find_get_page(btree_inode->i_mapping, index); 468 if (!page) 469 continue; 470 if (PageDirty(page)) { 471 btree_lock_page_hook(page); 472 wait_on_page_writeback(page); 473 err = write_one_page(page, 0); 474 if (err) 475 werr = err; 476 } 477 wait_on_page_writeback(page); 478 page_cache_release(page); 479 cond_resched(); 480 } 481 } 482 if (err) 483 werr = err; 484 return werr; 485} 486 487/* 488 * when btree blocks are allocated, they have some corresponding bits set for 489 * them in one of two extent_io trees. This is used to make sure all of 490 * those extents are on disk for transaction or log commit 491 */ 492int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 493 struct extent_io_tree *dirty_pages, int mark) 494{ 495 int ret; 496 int ret2; 497 498 ret = btrfs_write_marked_extents(root, dirty_pages, mark); 499 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); 500 return ret || ret2; 501} 502 503int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, 504 struct btrfs_root *root) 505{ 506 if (!trans || !trans->transaction) { 507 struct inode *btree_inode; 508 btree_inode = root->fs_info->btree_inode; 509 return filemap_write_and_wait(btree_inode->i_mapping); 510 } 511 return btrfs_write_and_wait_marked_extents(root, 512 &trans->transaction->dirty_pages, 513 EXTENT_DIRTY); 514} 515 516/* 517 * this is used to update the root pointer in the tree of tree roots. 518 * 519 * But, in the case of the extent allocation tree, updating the root 520 * pointer may allocate blocks which may change the root of the extent 521 * allocation tree. 522 * 523 * So, this loops and repeats and makes sure the cowonly root didn't 524 * change while the root pointer was being updated in the metadata. 525 */ 526static int update_cowonly_root(struct btrfs_trans_handle *trans, 527 struct btrfs_root *root) 528{ 529 int ret; 530 u64 old_root_bytenr; 531 u64 old_root_used; 532 struct btrfs_root *tree_root = root->fs_info->tree_root; 533 534 old_root_used = btrfs_root_used(&root->root_item); 535 btrfs_write_dirty_block_groups(trans, root); 536 537 while (1) { 538 old_root_bytenr = btrfs_root_bytenr(&root->root_item); 539 if (old_root_bytenr == root->node->start && 540 old_root_used == btrfs_root_used(&root->root_item)) 541 break; 542 543 btrfs_set_root_node(&root->root_item, root->node); 544 ret = btrfs_update_root(trans, tree_root, 545 &root->root_key, 546 &root->root_item); 547 BUG_ON(ret); 548 549 old_root_used = btrfs_root_used(&root->root_item); 550 ret = btrfs_write_dirty_block_groups(trans, root); 551 BUG_ON(ret); 552 } 553 554 if (root != root->fs_info->extent_root) 555 switch_commit_root(root); 556 557 return 0; 558} 559 560/* 561 * update all the cowonly tree roots on disk 562 */ 563static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, 564 struct btrfs_root *root) 565{ 566 struct btrfs_fs_info *fs_info = root->fs_info; 567 struct list_head *next; 568 struct extent_buffer *eb; 569 int ret; 570 571 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 572 BUG_ON(ret); 573 574 eb = btrfs_lock_root_node(fs_info->tree_root); 575 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); 576 btrfs_tree_unlock(eb); 577 free_extent_buffer(eb); 578 579 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 580 BUG_ON(ret); 581 582 while (!list_empty(&fs_info->dirty_cowonly_roots)) { 583 next = fs_info->dirty_cowonly_roots.next; 584 list_del_init(next); 585 root = list_entry(next, struct btrfs_root, dirty_list); 586 587 update_cowonly_root(trans, root); 588 } 589 590 down_write(&fs_info->extent_commit_sem); 591 switch_commit_root(fs_info->extent_root); 592 up_write(&fs_info->extent_commit_sem); 593 594 return 0; 595} 596 597/* 598 * dead roots are old snapshots that need to be deleted. This allocates 599 * a dirty root struct and adds it into the list of dead roots that need to 600 * be deleted 601 */ 602int btrfs_add_dead_root(struct btrfs_root *root) 603{ 604 mutex_lock(&root->fs_info->trans_mutex); 605 list_add(&root->root_list, &root->fs_info->dead_roots); 606 mutex_unlock(&root->fs_info->trans_mutex); 607 return 0; 608} 609 610/* 611 * update all the cowonly tree roots on disk 612 */ 613static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, 614 struct btrfs_root *root) 615{ 616 struct btrfs_root *gang[8]; 617 struct btrfs_fs_info *fs_info = root->fs_info; 618 int i; 619 int ret; 620 int err = 0; 621 622 while (1) { 623 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 624 (void **)gang, 0, 625 ARRAY_SIZE(gang), 626 BTRFS_ROOT_TRANS_TAG); 627 if (ret == 0) 628 break; 629 for (i = 0; i < ret; i++) { 630 root = gang[i]; 631 radix_tree_tag_clear(&fs_info->fs_roots_radix, 632 (unsigned long)root->root_key.objectid, 633 BTRFS_ROOT_TRANS_TAG); 634 635 btrfs_free_log(trans, root); 636 btrfs_update_reloc_root(trans, root); 637 btrfs_orphan_commit_root(trans, root); 638 639 if (root->commit_root != root->node) { 640 switch_commit_root(root); 641 btrfs_set_root_node(&root->root_item, 642 root->node); 643 } 644 645 err = btrfs_update_root(trans, fs_info->tree_root, 646 &root->root_key, 647 &root->root_item); 648 if (err) 649 break; 650 } 651 } 652 return err; 653} 654 655/* 656 * defrag a given btree. If cacheonly == 1, this won't read from the disk, 657 * otherwise every leaf in the btree is read and defragged. 658 */ 659int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) 660{ 661 struct btrfs_fs_info *info = root->fs_info; 662 struct btrfs_trans_handle *trans; 663 int ret; 664 unsigned long nr; 665 666 if (xchg(&root->defrag_running, 1)) 667 return 0; 668 669 while (1) { 670 trans = btrfs_start_transaction(root, 0); 671 if (IS_ERR(trans)) 672 return PTR_ERR(trans); 673 674 ret = btrfs_defrag_leaves(trans, root, cacheonly); 675 676 nr = trans->blocks_used; 677 btrfs_end_transaction(trans, root); 678 btrfs_btree_balance_dirty(info->tree_root, nr); 679 cond_resched(); 680 681 if (root->fs_info->closing || ret != -EAGAIN) 682 break; 683 } 684 root->defrag_running = 0; 685 return ret; 686} 687 688 689/* 690 * new snapshots need to be created at a very specific time in the 691 * transaction commit. This does the actual creation 692 */ 693static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 694 struct btrfs_fs_info *fs_info, 695 struct btrfs_pending_snapshot *pending) 696{ 697 struct btrfs_key key; 698 struct btrfs_root_item *new_root_item; 699 struct btrfs_root *tree_root = fs_info->tree_root; 700 struct btrfs_root *root = pending->root; 701 struct btrfs_root *parent_root; 702 struct inode *parent_inode; 703 struct dentry *dentry; 704 struct extent_buffer *tmp; 705 struct extent_buffer *old; 706 int ret; 707 int retries = 0; 708 u64 to_reserve = 0; 709 u64 index = 0; 710 u64 objectid; 711 712 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); 713 if (!new_root_item) { 714 pending->error = -ENOMEM; 715 goto fail; 716 } 717 718 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); 719 if (ret) { 720 pending->error = ret; 721 goto fail; 722 } 723 724 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); 725 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); 726 727 if (to_reserve > 0) { 728 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, 729 to_reserve, &retries); 730 if (ret) { 731 pending->error = ret; 732 goto fail; 733 } 734 } 735 736 key.objectid = objectid; 737 key.offset = (u64)-1; 738 key.type = BTRFS_ROOT_ITEM_KEY; 739 740 trans->block_rsv = &pending->block_rsv; 741 742 dentry = pending->dentry; 743 parent_inode = dentry->d_parent->d_inode; 744 parent_root = BTRFS_I(parent_inode)->root; 745 record_root_in_trans(trans, parent_root); 746 747 /* 748 * insert the directory item 749 */ 750 ret = btrfs_set_inode_index(parent_inode, &index); 751 BUG_ON(ret); 752 ret = btrfs_insert_dir_item(trans, parent_root, 753 dentry->d_name.name, dentry->d_name.len, 754 parent_inode->i_ino, &key, 755 BTRFS_FT_DIR, index); 756 BUG_ON(ret); 757 758 btrfs_i_size_write(parent_inode, parent_inode->i_size + 759 dentry->d_name.len * 2); 760 ret = btrfs_update_inode(trans, parent_root, parent_inode); 761 BUG_ON(ret); 762 763 record_root_in_trans(trans, root); 764 btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 765 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 766 767 old = btrfs_lock_root_node(root); 768 btrfs_cow_block(trans, root, old, NULL, 0, &old); 769 btrfs_set_lock_blocking(old); 770 771 btrfs_copy_root(trans, root, old, &tmp, objectid); 772 btrfs_tree_unlock(old); 773 free_extent_buffer(old); 774 775 btrfs_set_root_node(new_root_item, tmp); 776 /* record when the snapshot was created in key.offset */ 777 key.offset = trans->transid; 778 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 779 btrfs_tree_unlock(tmp); 780 free_extent_buffer(tmp); 781 BUG_ON(ret); 782 783 /* 784 * insert root back/forward references 785 */ 786 ret = btrfs_add_root_ref(trans, tree_root, objectid, 787 parent_root->root_key.objectid, 788 parent_inode->i_ino, index, 789 dentry->d_name.name, dentry->d_name.len); 790 BUG_ON(ret); 791 792 key.offset = (u64)-1; 793 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); 794 BUG_ON(IS_ERR(pending->snap)); 795 796 btrfs_reloc_post_snapshot(trans, pending); 797 btrfs_orphan_post_snapshot(trans, pending); 798fail: 799 kfree(new_root_item); 800 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); 801 return 0; 802} 803 804/* 805 * create all the snapshots we've scheduled for creation 806 */ 807static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, 808 struct btrfs_fs_info *fs_info) 809{ 810 struct btrfs_pending_snapshot *pending; 811 struct list_head *head = &trans->transaction->pending_snapshots; 812 int ret; 813 814 list_for_each_entry(pending, head, list) { 815 ret = create_pending_snapshot(trans, fs_info, pending); 816 BUG_ON(ret); 817 } 818 return 0; 819} 820 821static void update_super_roots(struct btrfs_root *root) 822{ 823 struct btrfs_root_item *root_item; 824 struct btrfs_super_block *super; 825 826 super = &root->fs_info->super_copy; 827 828 root_item = &root->fs_info->chunk_root->root_item; 829 super->chunk_root = root_item->bytenr; 830 super->chunk_root_generation = root_item->generation; 831 super->chunk_root_level = root_item->level; 832 833 root_item = &root->fs_info->tree_root->root_item; 834 super->root = root_item->bytenr; 835 super->generation = root_item->generation; 836 super->root_level = root_item->level; 837} 838 839int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 840{ 841 int ret = 0; 842 spin_lock(&info->new_trans_lock); 843 if (info->running_transaction) 844 ret = info->running_transaction->in_commit; 845 spin_unlock(&info->new_trans_lock); 846 return ret; 847} 848 849int btrfs_transaction_blocked(struct btrfs_fs_info *info) 850{ 851 int ret = 0; 852 spin_lock(&info->new_trans_lock); 853 if (info->running_transaction) 854 ret = info->running_transaction->blocked; 855 spin_unlock(&info->new_trans_lock); 856 return ret; 857} 858 859int btrfs_commit_transaction(struct btrfs_trans_handle *trans, 860 struct btrfs_root *root) 861{ 862 unsigned long joined = 0; 863 unsigned long timeout = 1; 864 struct btrfs_transaction *cur_trans; 865 struct btrfs_transaction *prev_trans = NULL; 866 DEFINE_WAIT(wait); 867 int ret; 868 int should_grow = 0; 869 unsigned long now = get_seconds(); 870 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); 871 872 btrfs_run_ordered_operations(root, 0); 873 874 /* make a pass through all the delayed refs we have so far 875 * any runnings procs may add more while we are here 876 */ 877 ret = btrfs_run_delayed_refs(trans, root, 0); 878 BUG_ON(ret); 879 880 btrfs_trans_release_metadata(trans, root); 881 882 cur_trans = trans->transaction; 883 /* 884 * set the flushing flag so procs in this transaction have to 885 * start sending their work down. 886 */ 887 cur_trans->delayed_refs.flushing = 1; 888 889 ret = btrfs_run_delayed_refs(trans, root, 0); 890 BUG_ON(ret); 891 892 mutex_lock(&root->fs_info->trans_mutex); 893 if (cur_trans->in_commit) { 894 cur_trans->use_count++; 895 mutex_unlock(&root->fs_info->trans_mutex); 896 btrfs_end_transaction(trans, root); 897 898 ret = wait_for_commit(root, cur_trans); 899 BUG_ON(ret); 900 901 mutex_lock(&root->fs_info->trans_mutex); 902 put_transaction(cur_trans); 903 mutex_unlock(&root->fs_info->trans_mutex); 904 905 return 0; 906 } 907 908 trans->transaction->in_commit = 1; 909 trans->transaction->blocked = 1; 910 if (cur_trans->list.prev != &root->fs_info->trans_list) { 911 prev_trans = list_entry(cur_trans->list.prev, 912 struct btrfs_transaction, list); 913 if (!prev_trans->commit_done) { 914 prev_trans->use_count++; 915 mutex_unlock(&root->fs_info->trans_mutex); 916 917 wait_for_commit(root, prev_trans); 918 919 mutex_lock(&root->fs_info->trans_mutex); 920 put_transaction(prev_trans); 921 } 922 } 923 924 if (now < cur_trans->start_time || now - cur_trans->start_time < 1) 925 should_grow = 1; 926 927 do { 928 int snap_pending = 0; 929 joined = cur_trans->num_joined; 930 if (!list_empty(&trans->transaction->pending_snapshots)) 931 snap_pending = 1; 932 933 WARN_ON(cur_trans != trans->transaction); 934 if (cur_trans->num_writers > 1) 935 timeout = MAX_SCHEDULE_TIMEOUT; 936 else if (should_grow) 937 timeout = 1; 938 939 mutex_unlock(&root->fs_info->trans_mutex); 940 941 if (flush_on_commit || snap_pending) { 942 btrfs_start_delalloc_inodes(root, 1); 943 ret = btrfs_wait_ordered_extents(root, 0, 1); 944 BUG_ON(ret); 945 } 946 947 /* 948 * rename don't use btrfs_join_transaction, so, once we 949 * set the transaction to blocked above, we aren't going 950 * to get any new ordered operations. We can safely run 951 * it here and no for sure that nothing new will be added 952 * to the list 953 */ 954 btrfs_run_ordered_operations(root, 1); 955 956 prepare_to_wait(&cur_trans->writer_wait, &wait, 957 TASK_UNINTERRUPTIBLE); 958 959 smp_mb(); 960 if (cur_trans->num_writers > 1 || should_grow) 961 schedule_timeout(timeout); 962 963 mutex_lock(&root->fs_info->trans_mutex); 964 finish_wait(&cur_trans->writer_wait, &wait); 965 } while (cur_trans->num_writers > 1 || 966 (should_grow && cur_trans->num_joined != joined)); 967 968 ret = create_pending_snapshots(trans, root->fs_info); 969 BUG_ON(ret); 970 971 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 972 BUG_ON(ret); 973 974 WARN_ON(cur_trans != trans->transaction); 975 976 /* btrfs_commit_tree_roots is responsible for getting the 977 * various roots consistent with each other. Every pointer 978 * in the tree of tree roots has to point to the most up to date 979 * root for every subvolume and other tree. So, we have to keep 980 * the tree logging code from jumping in and changing any 981 * of the trees. 982 * 983 * At this point in the commit, there can't be any tree-log 984 * writers, but a little lower down we drop the trans mutex 985 * and let new people in. By holding the tree_log_mutex 986 * from now until after the super is written, we avoid races 987 * with the tree-log code. 988 */ 989 mutex_lock(&root->fs_info->tree_log_mutex); 990 991 ret = commit_fs_roots(trans, root); 992 BUG_ON(ret); 993 994 /* commit_fs_roots gets rid of all the tree log roots, it is now 995 * safe to free the root of tree log roots 996 */ 997 btrfs_free_log_root_tree(trans, root->fs_info); 998 999 ret = commit_cowonly_roots(trans, root); 1000 BUG_ON(ret); 1001 1002 btrfs_prepare_extent_commit(trans, root); 1003 1004 cur_trans = root->fs_info->running_transaction; 1005 spin_lock(&root->fs_info->new_trans_lock); 1006 root->fs_info->running_transaction = NULL; 1007 spin_unlock(&root->fs_info->new_trans_lock); 1008 1009 btrfs_set_root_node(&root->fs_info->tree_root->root_item, 1010 root->fs_info->tree_root->node); 1011 switch_commit_root(root->fs_info->tree_root); 1012 1013 btrfs_set_root_node(&root->fs_info->chunk_root->root_item, 1014 root->fs_info->chunk_root->node); 1015 switch_commit_root(root->fs_info->chunk_root); 1016 1017 update_super_roots(root); 1018 1019 if (!root->fs_info->log_root_recovering) { 1020 btrfs_set_super_log_root(&root->fs_info->super_copy, 0); 1021 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); 1022 } 1023 1024 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, 1025 sizeof(root->fs_info->super_copy)); 1026 1027 trans->transaction->blocked = 0; 1028 1029 wake_up(&root->fs_info->transaction_wait); 1030 1031 mutex_unlock(&root->fs_info->trans_mutex); 1032 ret = btrfs_write_and_wait_transaction(trans, root); 1033 BUG_ON(ret); 1034 write_ctree_super(trans, root, 0); 1035 1036 /* 1037 * the super is written, we can safely allow the tree-loggers 1038 * to go about their business 1039 */ 1040 mutex_unlock(&root->fs_info->tree_log_mutex); 1041 1042 btrfs_finish_extent_commit(trans, root); 1043 1044 mutex_lock(&root->fs_info->trans_mutex); 1045 1046 cur_trans->commit_done = 1; 1047 1048 root->fs_info->last_trans_committed = cur_trans->transid; 1049 1050 wake_up(&cur_trans->commit_wait); 1051 1052 put_transaction(cur_trans); 1053 put_transaction(cur_trans); 1054 1055 mutex_unlock(&root->fs_info->trans_mutex); 1056 1057 if (current->journal_info == trans) 1058 current->journal_info = NULL; 1059 1060 kmem_cache_free(btrfs_trans_handle_cachep, trans); 1061 1062 if (current != root->fs_info->transaction_kthread) 1063 btrfs_run_delayed_iputs(root); 1064 1065 return ret; 1066} 1067 1068/* 1069 * interface function to delete all the snapshots we have scheduled for deletion 1070 */ 1071int btrfs_clean_old_snapshots(struct btrfs_root *root) 1072{ 1073 LIST_HEAD(list); 1074 struct btrfs_fs_info *fs_info = root->fs_info; 1075 1076 mutex_lock(&fs_info->trans_mutex); 1077 list_splice_init(&fs_info->dead_roots, &list); 1078 mutex_unlock(&fs_info->trans_mutex); 1079 1080 while (!list_empty(&list)) { 1081 root = list_entry(list.next, struct btrfs_root, root_list); 1082 list_del(&root->root_list); 1083 1084 if (btrfs_header_backref_rev(root->node) < 1085 BTRFS_MIXED_BACKREF_REV) 1086 btrfs_drop_snapshot(root, NULL, 0); 1087 else 1088 btrfs_drop_snapshot(root, NULL, 1); 1089 } 1090 return 0; 1091} 1092