From c216775458a2ee345d9412a2770c2916acfb5d30 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Thu, 12 Nov 2009 09:34:21 +0000 Subject: Btrfs: Fix disk_i_size update corner case There are some cases file extents are inserted without involving ordered struct. In these cases, we update disk_i_size directly, without checking pending ordered extent and DELALLOC bit. This patch extends btrfs_ordered_update_i_size() to handle these cases. Signed-off-by: Yan Zheng Signed-off-by: Chris Mason --- fs/btrfs/ordered-data.c | 105 +++++++++++++++++++++++++++++++++++++----------- 1 file changed, 81 insertions(+), 24 deletions(-) (limited to 'fs/btrfs/ordered-data.c') diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index 5799bc46a30..9b16073bb87 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) /* * remove an ordered extent from the tree. No references are dropped - * but, anyone waiting on this extent is woken up. + * and you must wake_up entry->wait. You must hold the tree mutex + * while you call this function. */ -int btrfs_remove_ordered_extent(struct inode *inode, +static int __btrfs_remove_ordered_extent(struct inode *inode, struct btrfs_ordered_extent *entry) { struct btrfs_ordered_inode_tree *tree; struct rb_node *node; tree = &BTRFS_I(inode)->ordered_tree; - mutex_lock(&tree->mutex); node = &entry->rb_node; rb_erase(node, &tree->tree); tree->last = NULL; @@ -326,9 +326,26 @@ int btrfs_remove_ordered_extent(struct inode *inode, } spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); + return 0; +} + +/* + * remove an ordered extent from the tree. No references are dropped + * but any waiters are woken. + */ +int btrfs_remove_ordered_extent(struct inode *inode, + struct btrfs_ordered_extent *entry) +{ + struct btrfs_ordered_inode_tree *tree; + int ret; + + tree = &BTRFS_I(inode)->ordered_tree; + mutex_lock(&tree->mutex); + ret = __btrfs_remove_ordered_extent(inode, entry); mutex_unlock(&tree->mutex); wake_up(&entry->wait); - return 0; + + return ret; } /* @@ -589,7 +606,7 @@ out: * After an extent is done, call this to conditionally update the on disk * i_size. i_size is updated to cover any fully written part of the file. */ -int btrfs_ordered_update_i_size(struct inode *inode, +int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, struct btrfs_ordered_extent *ordered) { struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; @@ -597,18 +614,30 @@ int btrfs_ordered_update_i_size(struct inode *inode, u64 disk_i_size; u64 new_i_size; u64 i_size_test; + u64 i_size = i_size_read(inode); struct rb_node *node; + struct rb_node *prev = NULL; struct btrfs_ordered_extent *test; + int ret = 1; + + if (ordered) + offset = entry_end(ordered); mutex_lock(&tree->mutex); disk_i_size = BTRFS_I(inode)->disk_i_size; + /* truncate file */ + if (disk_i_size > i_size) { + BTRFS_I(inode)->disk_i_size = i_size; + ret = 0; + goto out; + } + /* * if the disk i_size is already at the inode->i_size, or * this ordered extent is inside the disk i_size, we're done */ - if (disk_i_size >= inode->i_size || - ordered->file_offset + ordered->len <= disk_i_size) { + if (disk_i_size == i_size || offset <= disk_i_size) { goto out; } @@ -616,8 +645,7 @@ int btrfs_ordered_update_i_size(struct inode *inode, * we can't update the disk_isize if there are delalloc bytes * between disk_i_size and this ordered extent */ - if (test_range_bit(io_tree, disk_i_size, - ordered->file_offset + ordered->len - 1, + if (test_range_bit(io_tree, disk_i_size, offset - 1, EXTENT_DELALLOC, 0, NULL)) { goto out; } @@ -626,20 +654,32 @@ int btrfs_ordered_update_i_size(struct inode *inode, * if we find an ordered extent then we can't update disk i_size * yet */ - node = &ordered->rb_node; - while (1) { - node = rb_prev(node); - if (!node) - break; + if (ordered) { + node = rb_prev(&ordered->rb_node); + } else { + prev = tree_search(tree, offset); + /* + * we insert file extents without involving ordered struct, + * so there should be no ordered struct cover this offset + */ + if (prev) { + test = rb_entry(prev, struct btrfs_ordered_extent, + rb_node); + BUG_ON(offset_in_entry(test, offset)); + } + node = prev; + } + while (node) { test = rb_entry(node, struct btrfs_ordered_extent, rb_node); if (test->file_offset + test->len <= disk_i_size) break; - if (test->file_offset >= inode->i_size) + if (test->file_offset >= i_size) break; if (test->file_offset >= disk_i_size) goto out; + node = rb_prev(node); } - new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode)); + new_i_size = min_t(u64, offset, i_size); /* * at this point, we know we can safely update i_size to at least @@ -647,7 +687,14 @@ int btrfs_ordered_update_i_size(struct inode *inode, * walk forward and see if ios from higher up in the file have * finished. */ - node = rb_next(&ordered->rb_node); + if (ordered) { + node = rb_next(&ordered->rb_node); + } else { + if (prev) + node = rb_next(prev); + else + node = rb_first(&tree->tree); + } i_size_test = 0; if (node) { /* @@ -655,10 +702,10 @@ int btrfs_ordered_update_i_size(struct inode *inode, * between our ordered extent and the next one. */ test = rb_entry(node, struct btrfs_ordered_extent, rb_node); - if (test->file_offset > entry_end(ordered)) + if (test->file_offset > offset) i_size_test = test->file_offset; } else { - i_size_test = i_size_read(inode); + i_size_test = i_size; } /* @@ -667,15 +714,25 @@ int btrfs_ordered_update_i_size(struct inode *inode, * are no delalloc bytes in this area, it is safe to update * disk_i_size to the end of the region. */ - if (i_size_test > entry_end(ordered) && - !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1, - EXTENT_DELALLOC, 0, NULL)) { - new_i_size = min_t(u64, i_size_test, i_size_read(inode)); + if (i_size_test > offset && + !test_range_bit(io_tree, offset, i_size_test - 1, + EXTENT_DELALLOC, 0, NULL)) { + new_i_size = min_t(u64, i_size_test, i_size); } BTRFS_I(inode)->disk_i_size = new_i_size; + ret = 0; out: + /* + * we need to remove the ordered extent with the tree lock held + * so that other people calling this function don't find our fully + * processed ordered entry and skip updating the i_size + */ + if (ordered) + __btrfs_remove_ordered_extent(inode, ordered); mutex_unlock(&tree->mutex); - return 0; + if (ordered) + wake_up(&ordered->wait); + return ret; } /* -- cgit v1.2.3-70-g09d2 From 24bbcf0442ee04660a5a030efdbb6d03f1c275cb Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Thu, 12 Nov 2009 09:36:34 +0000 Subject: Btrfs: Add delayed iput iput() can trigger new transactions if we are dropping the final reference, so calling it in btrfs_commit_transaction may end up deadlock. This patch adds delayed iput to avoid the issue. Signed-off-by: Yan Zheng Signed-off-by: Chris Mason --- fs/btrfs/ctree.h | 7 ++++++- fs/btrfs/disk-io.c | 4 ++++ fs/btrfs/extent-tree.c | 8 +++---- fs/btrfs/inode.c | 55 +++++++++++++++++++++++++++++++++++++++++++++++-- fs/btrfs/ordered-data.c | 10 ++++++--- fs/btrfs/ordered-data.h | 3 ++- fs/btrfs/relocation.c | 4 ++-- fs/btrfs/super.c | 4 ++-- fs/btrfs/transaction.c | 13 +++++++++--- 9 files changed, 90 insertions(+), 18 deletions(-) (limited to 'fs/btrfs/ordered-data.c') diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index a7cac2148c7..1983c889bb1 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -872,6 +872,9 @@ struct btrfs_fs_info { struct list_head dead_roots; struct list_head caching_block_groups; + spinlock_t delayed_iput_lock; + struct list_head delayed_iputs; + atomic_t nr_async_submits; atomic_t async_submit_draining; atomic_t nr_async_bios; @@ -2301,7 +2304,7 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, struct inode *inode, u64 new_size, u32 min_type); -int btrfs_start_delalloc_inodes(struct btrfs_root *root); +int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput); int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end); int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc); @@ -2341,6 +2344,8 @@ int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode); void btrfs_orphan_cleanup(struct btrfs_root *root); int btrfs_cont_expand(struct inode *inode, loff_t size); int btrfs_invalidate_inodes(struct btrfs_root *root); +void btrfs_add_delayed_iput(struct inode *inode); +void btrfs_run_delayed_iputs(struct btrfs_root *root); extern const struct dentry_operations btrfs_dentry_operations; /* ioctl.c */ diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index c1e59e33f02..009e3bd18f2 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -1476,6 +1476,7 @@ static int cleaner_kthread(void *arg) if (!(root->fs_info->sb->s_flags & MS_RDONLY) && mutex_trylock(&root->fs_info->cleaner_mutex)) { + btrfs_run_delayed_iputs(root); btrfs_clean_old_snapshots(root); mutex_unlock(&root->fs_info->cleaner_mutex); } @@ -1605,6 +1606,7 @@ struct btrfs_root *open_ctree(struct super_block *sb, INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); INIT_LIST_HEAD(&fs_info->trans_list); INIT_LIST_HEAD(&fs_info->dead_roots); + INIT_LIST_HEAD(&fs_info->delayed_iputs); INIT_LIST_HEAD(&fs_info->hashers); INIT_LIST_HEAD(&fs_info->delalloc_inodes); INIT_LIST_HEAD(&fs_info->ordered_operations); @@ -1613,6 +1615,7 @@ struct btrfs_root *open_ctree(struct super_block *sb, spin_lock_init(&fs_info->new_trans_lock); spin_lock_init(&fs_info->ref_cache_lock); spin_lock_init(&fs_info->fs_roots_radix_lock); + spin_lock_init(&fs_info->delayed_iput_lock); init_completion(&fs_info->kobj_unregister); fs_info->tree_root = tree_root; @@ -2386,6 +2389,7 @@ int btrfs_commit_super(struct btrfs_root *root) int ret; mutex_lock(&root->fs_info->cleaner_mutex); + btrfs_run_delayed_iputs(root); btrfs_clean_old_snapshots(root); mutex_unlock(&root->fs_info->cleaner_mutex); diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 4a86508ce47..fcdccfa4600 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -2880,9 +2880,9 @@ static noinline void flush_delalloc_async(struct btrfs_work *work) root = async->root; info = async->info; - btrfs_start_delalloc_inodes(root); + btrfs_start_delalloc_inodes(root, 0); wake_up(&info->flush_wait); - btrfs_wait_ordered_extents(root, 0); + btrfs_wait_ordered_extents(root, 0, 0); spin_lock(&info->lock); info->flushing = 0; @@ -2956,8 +2956,8 @@ static void flush_delalloc(struct btrfs_root *root, return; flush: - btrfs_start_delalloc_inodes(root); - btrfs_wait_ordered_extents(root, 0); + btrfs_start_delalloc_inodes(root, 0); + btrfs_wait_ordered_extents(root, 0, 0); spin_lock(&info->lock); info->flushing = 0; diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 82740a3c628..168e8c040aa 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2022,6 +2022,54 @@ zeroit: return -EIO; } +struct delayed_iput { + struct list_head list; + struct inode *inode; +}; + +void btrfs_add_delayed_iput(struct inode *inode) +{ + struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; + struct delayed_iput *delayed; + + if (atomic_add_unless(&inode->i_count, -1, 1)) + return; + + delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); + delayed->inode = inode; + + spin_lock(&fs_info->delayed_iput_lock); + list_add_tail(&delayed->list, &fs_info->delayed_iputs); + spin_unlock(&fs_info->delayed_iput_lock); +} + +void btrfs_run_delayed_iputs(struct btrfs_root *root) +{ + LIST_HEAD(list); + struct btrfs_fs_info *fs_info = root->fs_info; + struct delayed_iput *delayed; + int empty; + + spin_lock(&fs_info->delayed_iput_lock); + empty = list_empty(&fs_info->delayed_iputs); + spin_unlock(&fs_info->delayed_iput_lock); + if (empty) + return; + + down_read(&root->fs_info->cleanup_work_sem); + spin_lock(&fs_info->delayed_iput_lock); + list_splice_init(&fs_info->delayed_iputs, &list); + spin_unlock(&fs_info->delayed_iput_lock); + + while (!list_empty(&list)) { + delayed = list_entry(list.next, struct delayed_iput, list); + list_del(&delayed->list); + iput(delayed->inode); + kfree(delayed); + } + up_read(&root->fs_info->cleanup_work_sem); +} + /* * This creates an orphan entry for the given inode in case something goes * wrong in the middle of an unlink/truncate. @@ -5568,7 +5616,7 @@ out_fail: * some fairly slow code that needs optimization. This walks the list * of all the inodes with pending delalloc and forces them to disk. */ -int btrfs_start_delalloc_inodes(struct btrfs_root *root) +int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) { struct list_head *head = &root->fs_info->delalloc_inodes; struct btrfs_inode *binode; @@ -5587,7 +5635,10 @@ int btrfs_start_delalloc_inodes(struct btrfs_root *root) spin_unlock(&root->fs_info->delalloc_lock); if (inode) { filemap_flush(inode->i_mapping); - iput(inode); + if (delay_iput) + btrfs_add_delayed_iput(inode); + else + iput(inode); } cond_resched(); spin_lock(&root->fs_info->delalloc_lock); diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index 9b16073bb87..b10a49d4bc6 100644 --- a/fs/btrfs/ordered-data.c +++ b/fs/btrfs/ordered-data.c @@ -352,7 +352,8 @@ int btrfs_remove_ordered_extent(struct inode *inode, * wait for all the ordered extents in a root. This is done when balancing * space between drives. */ -int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only) +int btrfs_wait_ordered_extents(struct btrfs_root *root, + int nocow_only, int delay_iput) { struct list_head splice; struct list_head *cur; @@ -389,7 +390,10 @@ int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only) if (inode) { btrfs_start_ordered_extent(inode, ordered, 1); btrfs_put_ordered_extent(ordered); - iput(inode); + if (delay_iput) + btrfs_add_delayed_iput(inode); + else + iput(inode); } else { btrfs_put_ordered_extent(ordered); } @@ -447,7 +451,7 @@ again: btrfs_wait_ordered_range(inode, 0, (u64)-1); else filemap_flush(inode->i_mapping); - iput(inode); + btrfs_add_delayed_iput(inode); } cond_resched(); diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h index 4fa20398aec..1fe1282ef47 100644 --- a/fs/btrfs/ordered-data.h +++ b/fs/btrfs/ordered-data.h @@ -153,9 +153,10 @@ btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset); int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, struct btrfs_ordered_extent *ordered); int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, u32 *sum); -int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only); int btrfs_run_ordered_operations(struct btrfs_root *root, int wait); int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct inode *inode); +int btrfs_wait_ordered_extents(struct btrfs_root *root, + int nocow_only, int delay_iput); #endif diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c index f2aa53d2f94..a9728680eca 100644 --- a/fs/btrfs/relocation.c +++ b/fs/btrfs/relocation.c @@ -3541,8 +3541,8 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start) (unsigned long long)rc->block_group->key.objectid, (unsigned long long)rc->block_group->flags); - btrfs_start_delalloc_inodes(fs_info->tree_root); - btrfs_wait_ordered_extents(fs_info->tree_root, 0); + btrfs_start_delalloc_inodes(fs_info->tree_root, 0); + btrfs_wait_ordered_extents(fs_info->tree_root, 0, 0); while (1) { rc->extents_found = 0; diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c index 752a5463bf5..270cc96b9a4 100644 --- a/fs/btrfs/super.c +++ b/fs/btrfs/super.c @@ -405,8 +405,8 @@ int btrfs_sync_fs(struct super_block *sb, int wait) return 0; } - btrfs_start_delalloc_inodes(root); - btrfs_wait_ordered_extents(root, 0); + btrfs_start_delalloc_inodes(root, 0); + btrfs_wait_ordered_extents(root, 0, 0); trans = btrfs_start_transaction(root, 1); ret = btrfs_commit_transaction(trans, root); diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c index 728e8fe5d2c..75b31caade2 100644 --- a/fs/btrfs/transaction.c +++ b/fs/btrfs/transaction.c @@ -333,6 +333,9 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, memset(trans, 0, sizeof(*trans)); kmem_cache_free(btrfs_trans_handle_cachep, trans); + if (throttle) + btrfs_run_delayed_iputs(root); + return 0; } @@ -991,11 +994,11 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans, mutex_unlock(&root->fs_info->trans_mutex); if (flush_on_commit) { - btrfs_start_delalloc_inodes(root); - ret = btrfs_wait_ordered_extents(root, 0); + btrfs_start_delalloc_inodes(root, 1); + ret = btrfs_wait_ordered_extents(root, 0, 1); BUG_ON(ret); } else if (snap_pending) { - ret = btrfs_wait_ordered_extents(root, 1); + ret = btrfs_wait_ordered_extents(root, 0, 1); BUG_ON(ret); } @@ -1113,6 +1116,10 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans, current->journal_info = NULL; kmem_cache_free(btrfs_trans_handle_cachep, trans); + + if (current != root->fs_info->transaction_kthread) + btrfs_run_delayed_iputs(root); + return ret; } -- cgit v1.2.3-70-g09d2 From a038fab0cb873c75d6675e2bcffce8a3935bdce7 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Mon, 28 Dec 2009 05:01:58 +0000 Subject: Btrfs: align offsets for btrfs_ordered_update_i_size Some callers of btrfs_ordered_update_i_size can now pass in a NULL for the ordered extent to update against. This makes sure we properly align the offset they pass in when deciding how much to bump the on disk i_size. Signed-off-by: Chris Mason --- fs/btrfs/ordered-data.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'fs/btrfs/ordered-data.c') diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index b10a49d4bc6..5c2a9e78a94 100644 --- a/fs/btrfs/ordered-data.c +++ b/fs/btrfs/ordered-data.c @@ -626,6 +626,8 @@ int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, if (ordered) offset = entry_end(ordered); + else + offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); mutex_lock(&tree->mutex); disk_i_size = BTRFS_I(inode)->disk_i_size; -- cgit v1.2.3-70-g09d2