diff options
Diffstat (limited to 'fs/btrfs/ordered-data.c')
| -rw-r--r-- | fs/btrfs/ordered-data.c | 115 |
1 files changed, 88 insertions, 27 deletions
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index 5799bc46a309..b10a49d4bc6a 100644 --- a/fs/btrfs/ordered-data.c +++ b/fs/btrfs/ordered-data.c | |||
| @@ -291,16 +291,16 @@ int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) | |||
| 291 | 291 | ||
| 292 | /* | 292 | /* |
| 293 | * remove an ordered extent from the tree. No references are dropped | 293 | * remove an ordered extent from the tree. No references are dropped |
| 294 | * but, anyone waiting on this extent is woken up. | 294 | * and you must wake_up entry->wait. You must hold the tree mutex |
| 295 | * while you call this function. | ||
| 295 | */ | 296 | */ |
| 296 | int btrfs_remove_ordered_extent(struct inode *inode, | 297 | static int __btrfs_remove_ordered_extent(struct inode *inode, |
| 297 | struct btrfs_ordered_extent *entry) | 298 | struct btrfs_ordered_extent *entry) |
| 298 | { | 299 | { |
| 299 | struct btrfs_ordered_inode_tree *tree; | 300 | struct btrfs_ordered_inode_tree *tree; |
| 300 | struct rb_node *node; | 301 | struct rb_node *node; |
| 301 | 302 | ||
| 302 | tree = &BTRFS_I(inode)->ordered_tree; | 303 | tree = &BTRFS_I(inode)->ordered_tree; |
| 303 | mutex_lock(&tree->mutex); | ||
| 304 | node = &entry->rb_node; | 304 | node = &entry->rb_node; |
| 305 | rb_erase(node, &tree->tree); | 305 | rb_erase(node, &tree->tree); |
| 306 | tree->last = NULL; | 306 | tree->last = NULL; |
| @@ -326,16 +326,34 @@ int btrfs_remove_ordered_extent(struct inode *inode, | |||
| 326 | } | 326 | } |
| 327 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); | 327 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); |
| 328 | 328 | ||
| 329 | return 0; | ||
| 330 | } | ||
| 331 | |||
| 332 | /* | ||
| 333 | * remove an ordered extent from the tree. No references are dropped | ||
| 334 | * but any waiters are woken. | ||
| 335 | */ | ||
| 336 | int btrfs_remove_ordered_extent(struct inode *inode, | ||
| 337 | struct btrfs_ordered_extent *entry) | ||
| 338 | { | ||
| 339 | struct btrfs_ordered_inode_tree *tree; | ||
| 340 | int ret; | ||
| 341 | |||
| 342 | tree = &BTRFS_I(inode)->ordered_tree; | ||
| 343 | mutex_lock(&tree->mutex); | ||
| 344 | ret = __btrfs_remove_ordered_extent(inode, entry); | ||
| 329 | mutex_unlock(&tree->mutex); | 345 | mutex_unlock(&tree->mutex); |
| 330 | wake_up(&entry->wait); | 346 | wake_up(&entry->wait); |
| 331 | return 0; | 347 | |
| 348 | return ret; | ||
| 332 | } | 349 | } |
| 333 | 350 | ||
| 334 | /* | 351 | /* |
| 335 | * wait for all the ordered extents in a root. This is done when balancing | 352 | * wait for all the ordered extents in a root. This is done when balancing |
| 336 | * space between drives. | 353 | * space between drives. |
| 337 | */ | 354 | */ |
| 338 | int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only) | 355 | int btrfs_wait_ordered_extents(struct btrfs_root *root, |
| 356 | int nocow_only, int delay_iput) | ||
| 339 | { | 357 | { |
| 340 | struct list_head splice; | 358 | struct list_head splice; |
| 341 | struct list_head *cur; | 359 | struct list_head *cur; |
| @@ -372,7 +390,10 @@ int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only) | |||
| 372 | if (inode) { | 390 | if (inode) { |
| 373 | btrfs_start_ordered_extent(inode, ordered, 1); | 391 | btrfs_start_ordered_extent(inode, ordered, 1); |
| 374 | btrfs_put_ordered_extent(ordered); | 392 | btrfs_put_ordered_extent(ordered); |
| 375 | iput(inode); | 393 | if (delay_iput) |
| 394 | btrfs_add_delayed_iput(inode); | ||
| 395 | else | ||
| 396 | iput(inode); | ||
| 376 | } else { | 397 | } else { |
| 377 | btrfs_put_ordered_extent(ordered); | 398 | btrfs_put_ordered_extent(ordered); |
| 378 | } | 399 | } |
| @@ -430,7 +451,7 @@ again: | |||
| 430 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | 451 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
| 431 | else | 452 | else |
| 432 | filemap_flush(inode->i_mapping); | 453 | filemap_flush(inode->i_mapping); |
| 433 | iput(inode); | 454 | btrfs_add_delayed_iput(inode); |
| 434 | } | 455 | } |
| 435 | 456 | ||
| 436 | cond_resched(); | 457 | cond_resched(); |
| @@ -589,7 +610,7 @@ out: | |||
| 589 | * After an extent is done, call this to conditionally update the on disk | 610 | * After an extent is done, call this to conditionally update the on disk |
| 590 | * i_size. i_size is updated to cover any fully written part of the file. | 611 | * i_size. i_size is updated to cover any fully written part of the file. |
| 591 | */ | 612 | */ |
| 592 | int btrfs_ordered_update_i_size(struct inode *inode, | 613 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
| 593 | struct btrfs_ordered_extent *ordered) | 614 | struct btrfs_ordered_extent *ordered) |
| 594 | { | 615 | { |
| 595 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | 616 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; |
| @@ -597,18 +618,30 @@ int btrfs_ordered_update_i_size(struct inode *inode, | |||
| 597 | u64 disk_i_size; | 618 | u64 disk_i_size; |
| 598 | u64 new_i_size; | 619 | u64 new_i_size; |
| 599 | u64 i_size_test; | 620 | u64 i_size_test; |
| 621 | u64 i_size = i_size_read(inode); | ||
| 600 | struct rb_node *node; | 622 | struct rb_node *node; |
| 623 | struct rb_node *prev = NULL; | ||
| 601 | struct btrfs_ordered_extent *test; | 624 | struct btrfs_ordered_extent *test; |
| 625 | int ret = 1; | ||
| 626 | |||
| 627 | if (ordered) | ||
| 628 | offset = entry_end(ordered); | ||
| 602 | 629 | ||
| 603 | mutex_lock(&tree->mutex); | 630 | mutex_lock(&tree->mutex); |
| 604 | disk_i_size = BTRFS_I(inode)->disk_i_size; | 631 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
| 605 | 632 | ||
| 633 | /* truncate file */ | ||
| 634 | if (disk_i_size > i_size) { | ||
| 635 | BTRFS_I(inode)->disk_i_size = i_size; | ||
| 636 | ret = 0; | ||
| 637 | goto out; | ||
| 638 | } | ||
| 639 | |||
| 606 | /* | 640 | /* |
| 607 | * if the disk i_size is already at the inode->i_size, or | 641 | * if the disk i_size is already at the inode->i_size, or |
| 608 | * this ordered extent is inside the disk i_size, we're done | 642 | * this ordered extent is inside the disk i_size, we're done |
| 609 | */ | 643 | */ |
| 610 | if (disk_i_size >= inode->i_size || | 644 | if (disk_i_size == i_size || offset <= disk_i_size) { |
| 611 | ordered->file_offset + ordered->len <= disk_i_size) { | ||
| 612 | goto out; | 645 | goto out; |
| 613 | } | 646 | } |
| 614 | 647 | ||
| @@ -616,8 +649,7 @@ int btrfs_ordered_update_i_size(struct inode *inode, | |||
| 616 | * we can't update the disk_isize if there are delalloc bytes | 649 | * we can't update the disk_isize if there are delalloc bytes |
| 617 | * between disk_i_size and this ordered extent | 650 | * between disk_i_size and this ordered extent |
| 618 | */ | 651 | */ |
| 619 | if (test_range_bit(io_tree, disk_i_size, | 652 | if (test_range_bit(io_tree, disk_i_size, offset - 1, |
| 620 | ordered->file_offset + ordered->len - 1, | ||
| 621 | EXTENT_DELALLOC, 0, NULL)) { | 653 | EXTENT_DELALLOC, 0, NULL)) { |
| 622 | goto out; | 654 | goto out; |
| 623 | } | 655 | } |
| @@ -626,20 +658,32 @@ int btrfs_ordered_update_i_size(struct inode *inode, | |||
| 626 | * if we find an ordered extent then we can't update disk i_size | 658 | * if we find an ordered extent then we can't update disk i_size |
| 627 | * yet | 659 | * yet |
| 628 | */ | 660 | */ |
| 629 | node = &ordered->rb_node; | 661 | if (ordered) { |
| 630 | while (1) { | 662 | node = rb_prev(&ordered->rb_node); |
| 631 | node = rb_prev(node); | 663 | } else { |
| 632 | if (!node) | 664 | prev = tree_search(tree, offset); |
| 633 | break; | 665 | /* |
| 666 | * we insert file extents without involving ordered struct, | ||
| 667 | * so there should be no ordered struct cover this offset | ||
| 668 | */ | ||
| 669 | if (prev) { | ||
| 670 | test = rb_entry(prev, struct btrfs_ordered_extent, | ||
| 671 | rb_node); | ||
| 672 | BUG_ON(offset_in_entry(test, offset)); | ||
| 673 | } | ||
| 674 | node = prev; | ||
| 675 | } | ||
| 676 | while (node) { | ||
| 634 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); | 677 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 635 | if (test->file_offset + test->len <= disk_i_size) | 678 | if (test->file_offset + test->len <= disk_i_size) |
| 636 | break; | 679 | break; |
| 637 | if (test->file_offset >= inode->i_size) | 680 | if (test->file_offset >= i_size) |
| 638 | break; | 681 | break; |
| 639 | if (test->file_offset >= disk_i_size) | 682 | if (test->file_offset >= disk_i_size) |
| 640 | goto out; | 683 | goto out; |
| 684 | node = rb_prev(node); | ||
| 641 | } | 685 | } |
| 642 | new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode)); | 686 | new_i_size = min_t(u64, offset, i_size); |
| 643 | 687 | ||
| 644 | /* | 688 | /* |
| 645 | * at this point, we know we can safely update i_size to at least | 689 | * at this point, we know we can safely update i_size to at least |
| @@ -647,7 +691,14 @@ int btrfs_ordered_update_i_size(struct inode *inode, | |||
| 647 | * walk forward and see if ios from higher up in the file have | 691 | * walk forward and see if ios from higher up in the file have |
| 648 | * finished. | 692 | * finished. |
| 649 | */ | 693 | */ |
| 650 | node = rb_next(&ordered->rb_node); | 694 | if (ordered) { |
| 695 | node = rb_next(&ordered->rb_node); | ||
| 696 | } else { | ||
| 697 | if (prev) | ||
| 698 | node = rb_next(prev); | ||
| 699 | else | ||
| 700 | node = rb_first(&tree->tree); | ||
| 701 | } | ||
| 651 | i_size_test = 0; | 702 | i_size_test = 0; |
| 652 | if (node) { | 703 | if (node) { |
| 653 | /* | 704 | /* |
| @@ -655,10 +706,10 @@ int btrfs_ordered_update_i_size(struct inode *inode, | |||
| 655 | * between our ordered extent and the next one. | 706 | * between our ordered extent and the next one. |
| 656 | */ | 707 | */ |
| 657 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); | 708 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 658 | if (test->file_offset > entry_end(ordered)) | 709 | if (test->file_offset > offset) |
| 659 | i_size_test = test->file_offset; | 710 | i_size_test = test->file_offset; |
| 660 | } else { | 711 | } else { |
| 661 | i_size_test = i_size_read(inode); | 712 | i_size_test = i_size; |
| 662 | } | 713 | } |
| 663 | 714 | ||
| 664 | /* | 715 | /* |
| @@ -667,15 +718,25 @@ int btrfs_ordered_update_i_size(struct inode *inode, | |||
| 667 | * are no delalloc bytes in this area, it is safe to update | 718 | * are no delalloc bytes in this area, it is safe to update |
| 668 | * disk_i_size to the end of the region. | 719 | * disk_i_size to the end of the region. |
| 669 | */ | 720 | */ |
| 670 | if (i_size_test > entry_end(ordered) && | 721 | if (i_size_test > offset && |
| 671 | !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1, | 722 | !test_range_bit(io_tree, offset, i_size_test - 1, |
| 672 | EXTENT_DELALLOC, 0, NULL)) { | 723 | EXTENT_DELALLOC, 0, NULL)) { |
| 673 | new_i_size = min_t(u64, i_size_test, i_size_read(inode)); | 724 | new_i_size = min_t(u64, i_size_test, i_size); |
| 674 | } | 725 | } |
| 675 | BTRFS_I(inode)->disk_i_size = new_i_size; | 726 | BTRFS_I(inode)->disk_i_size = new_i_size; |
| 727 | ret = 0; | ||
| 676 | out: | 728 | out: |
| 729 | /* | ||
| 730 | * we need to remove the ordered extent with the tree lock held | ||
| 731 | * so that other people calling this function don't find our fully | ||
| 732 | * processed ordered entry and skip updating the i_size | ||
| 733 | */ | ||
| 734 | if (ordered) | ||
| 735 | __btrfs_remove_ordered_extent(inode, ordered); | ||
| 677 | mutex_unlock(&tree->mutex); | 736 | mutex_unlock(&tree->mutex); |
| 678 | return 0; | 737 | if (ordered) |
| 738 | wake_up(&ordered->wait); | ||
| 739 | return ret; | ||
| 679 | } | 740 | } |
| 680 | 741 | ||
| 681 | /* | 742 | /* |
