Btrfs: New data=ordered implementation
The old data=ordered code would force commit to wait until
all the data extents from the transaction were fully on disk. This
introduced large latencies into the commit and stalled new writers
in the transaction for a long time.
The new code changes the way data allocations and extents work:
* When delayed allocation is filled, data extents are reserved, and
the extent bit EXTENT_ORDERED is set on the entire range of the extent.
A struct btrfs_ordered_extent is allocated an inserted into a per-inode
rbtree to track the pending extents.
* As each page is written EXTENT_ORDERED is cleared on the bytes corresponding
to that page.
* When all of the bytes corresponding to a single struct btrfs_ordered_extent
are written, The previously reserved extent is inserted into the FS
btree and into the extent allocation trees. The checksums for the file
data are also updated.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index a8a3cb0..86a5acc 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -67,7 +67,6 @@
cur_trans->start_time = get_seconds();
INIT_LIST_HEAD(&cur_trans->pending_snapshots);
list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
- btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
extent_io_tree_init(&cur_trans->dirty_pages,
root->fs_info->btree_inode->i_mapping,
GFP_NOFS);
@@ -158,10 +157,12 @@
wake_up(&cur_trans->writer_wait);
if (cur_trans->in_commit && throttle) {
- int ret;
+ DEFINE_WAIT(wait);
mutex_unlock(&root->fs_info->trans_mutex);
- ret = wait_for_commit(root, cur_trans);
- BUG_ON(ret);
+ prepare_to_wait(&root->fs_info->transaction_throttle, &wait,
+ TASK_UNINTERRUPTIBLE);
+ schedule();
+ finish_wait(&root->fs_info->transaction_throttle, &wait);
mutex_lock(&root->fs_info->trans_mutex);
}
@@ -486,58 +487,6 @@
return ret;
}
-int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
- struct btrfs_root *root)
-{
- struct btrfs_transaction *cur_trans = trans->transaction;
- struct inode *inode;
- u64 root_objectid = 0;
- u64 objectid = 0;
- int ret;
-
- atomic_inc(&root->fs_info->throttles);
- while(1) {
- ret = btrfs_find_first_ordered_inode(
- &cur_trans->ordered_inode_tree,
- &root_objectid, &objectid, &inode);
- if (!ret)
- break;
-
- mutex_unlock(&root->fs_info->trans_mutex);
-
- if (S_ISREG(inode->i_mode)) {
- atomic_inc(&BTRFS_I(inode)->ordered_writeback);
- filemap_fdatawrite(inode->i_mapping);
- atomic_dec(&BTRFS_I(inode)->ordered_writeback);
- }
- iput(inode);
-
- mutex_lock(&root->fs_info->trans_mutex);
- }
- while(1) {
- root_objectid = 0;
- objectid = 0;
- ret = btrfs_find_del_first_ordered_inode(
- &cur_trans->ordered_inode_tree,
- &root_objectid, &objectid, &inode);
- if (!ret)
- break;
- mutex_unlock(&root->fs_info->trans_mutex);
-
- if (S_ISREG(inode->i_mode)) {
- atomic_inc(&BTRFS_I(inode)->ordered_writeback);
- filemap_write_and_wait(inode->i_mapping);
- atomic_dec(&BTRFS_I(inode)->ordered_writeback);
- }
- atomic_dec(&inode->i_count);
- iput(inode);
-
- mutex_lock(&root->fs_info->trans_mutex);
- }
- atomic_dec(&root->fs_info->throttles);
- return 0;
-}
-
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
struct btrfs_fs_info *fs_info,
struct btrfs_pending_snapshot *pending)
@@ -666,6 +615,7 @@
extent_io_tree_init(pinned_copy,
root->fs_info->btree_inode->i_mapping, GFP_NOFS);
+printk("commit trans %Lu\n", trans->transid);
trans->transaction->in_commit = 1;
cur_trans = trans->transaction;
if (cur_trans->list.prev != &root->fs_info->trans_list) {
@@ -699,8 +649,6 @@
mutex_lock(&root->fs_info->trans_mutex);
finish_wait(&cur_trans->writer_wait, &wait);
- ret = btrfs_write_ordered_inodes(trans, root);
-
} while (cur_trans->num_writers > 1 ||
(cur_trans->num_joined != joined));
@@ -736,6 +684,8 @@
btrfs_copy_pinned(root, pinned_copy);
+ wake_up(&root->fs_info->transaction_throttle);
+
mutex_unlock(&root->fs_info->trans_mutex);
ret = btrfs_write_and_wait_transaction(trans, root);
BUG_ON(ret);
@@ -758,6 +708,7 @@
list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
mutex_unlock(&root->fs_info->trans_mutex);
+printk("done commit trans %Lu\n", trans->transid);
kmem_cache_free(btrfs_trans_handle_cachep, trans);
if (root->fs_info->closing) {