Btrfs: fix task hang under heavy compressed write
[cascardo/linux.git] / fs / btrfs / ordered-data.c
index e12441c..ac734ec 100644 (file)
@@ -484,8 +484,19 @@ void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
                                           log_list);
                list_del_init(&ordered->log_list);
                spin_unlock_irq(&log->log_extents_lock[index]);
+
+               if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
+                   !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
+                       struct inode *inode = ordered->inode;
+                       u64 start = ordered->file_offset;
+                       u64 end = ordered->file_offset + ordered->len - 1;
+
+                       WARN_ON(!inode);
+                       filemap_fdatawrite_range(inode->i_mapping, start, end);
+               }
                wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
                                                   &ordered->flags));
+
                btrfs_put_ordered_extent(ordered);
                spin_lock_irq(&log->log_extents_lock[index]);
        }
@@ -560,18 +571,6 @@ void btrfs_remove_ordered_extent(struct inode *inode,
 
        trace_btrfs_ordered_extent_remove(inode, entry);
 
-       /*
-        * we have no more ordered extents for this inode and
-        * no dirty pages.  We can safely remove it from the
-        * list of ordered extents
-        */
-       if (RB_EMPTY_ROOT(&tree->tree) &&
-           !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
-               spin_lock(&root->fs_info->ordered_root_lock);
-               list_del_init(&BTRFS_I(inode)->ordered_operations);
-               spin_unlock(&root->fs_info->ordered_root_lock);
-       }
-
        if (!root->nr_ordered_extents) {
                spin_lock(&root->fs_info->ordered_root_lock);
                BUG_ON(list_empty(&root->ordered_root));
@@ -616,6 +615,7 @@ int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
                spin_unlock(&root->ordered_extent_lock);
 
                btrfs_init_work(&ordered->flush_work,
+                               btrfs_flush_delalloc_helper,
                                btrfs_run_ordered_extent_work, NULL, NULL);
                list_add_tail(&ordered->work_list, &works);
                btrfs_queue_work(root->fs_info->flush_workers,
@@ -675,81 +675,6 @@ void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
        mutex_unlock(&fs_info->ordered_operations_mutex);
 }
 
-/*
- * this is used during transaction commit to write all the inodes
- * added to the ordered operation list.  These files must be fully on
- * disk before the transaction commits.
- *
- * we have two modes here, one is to just start the IO via filemap_flush
- * and the other is to wait for all the io.  When we wait, we have an
- * extra check to make sure the ordered operation list really is empty
- * before we return
- */
-int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans,
-                                struct btrfs_root *root, int wait)
-{
-       struct btrfs_inode *btrfs_inode;
-       struct inode *inode;
-       struct btrfs_transaction *cur_trans = trans->transaction;
-       struct list_head splice;
-       struct list_head works;
-       struct btrfs_delalloc_work *work, *next;
-       int ret = 0;
-
-       INIT_LIST_HEAD(&splice);
-       INIT_LIST_HEAD(&works);
-
-       mutex_lock(&root->fs_info->ordered_extent_flush_mutex);
-       spin_lock(&root->fs_info->ordered_root_lock);
-       list_splice_init(&cur_trans->ordered_operations, &splice);
-       while (!list_empty(&splice)) {
-               btrfs_inode = list_entry(splice.next, struct btrfs_inode,
-                                  ordered_operations);
-               inode = &btrfs_inode->vfs_inode;
-
-               list_del_init(&btrfs_inode->ordered_operations);
-
-               /*
-                * the inode may be getting freed (in sys_unlink path).
-                */
-               inode = igrab(inode);
-               if (!inode)
-                       continue;
-
-               if (!wait)
-                       list_add_tail(&BTRFS_I(inode)->ordered_operations,
-                                     &cur_trans->ordered_operations);
-               spin_unlock(&root->fs_info->ordered_root_lock);
-
-               work = btrfs_alloc_delalloc_work(inode, wait, 1);
-               if (!work) {
-                       spin_lock(&root->fs_info->ordered_root_lock);
-                       if (list_empty(&BTRFS_I(inode)->ordered_operations))
-                               list_add_tail(&btrfs_inode->ordered_operations,
-                                             &splice);
-                       list_splice_tail(&splice,
-                                        &cur_trans->ordered_operations);
-                       spin_unlock(&root->fs_info->ordered_root_lock);
-                       ret = -ENOMEM;
-                       goto out;
-               }
-               list_add_tail(&work->list, &works);
-               btrfs_queue_work(root->fs_info->flush_workers,
-                                &work->work);
-
-               cond_resched();
-               spin_lock(&root->fs_info->ordered_root_lock);
-       }
-       spin_unlock(&root->fs_info->ordered_root_lock);
-out:
-       list_for_each_entry_safe(work, next, &works, list) {
-               list_del_init(&work->list);
-               btrfs_wait_and_free_delalloc_work(work);
-       }
-       mutex_unlock(&root->fs_info->ordered_extent_flush_mutex);
-       return ret;
-}
-
 /*
  * Used to start IO or wait for a given ordered extent to finish.
  *
@@ -1109,42 +1034,6 @@ out:
        return index;
 }
 
-
-/*
- * add a given inode to the list of inodes that must be fully on
- * disk before a transaction commit finishes.
- *
- * This basically gives us the ext3 style data=ordered mode, and it is mostly
- * used to make sure renamed files are fully on disk.
- *
- * It is a noop if the inode is already fully on disk.
- *
- * If trans is not null, we'll do a friendly check for a transaction that
- * is already flushing things and force the IO down ourselves.
- */
-void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
-                                struct btrfs_root *root, struct inode *inode)
-{
-       struct btrfs_transaction *cur_trans = trans->transaction;
-       u64 last_mod;
-
-       last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
-
-       /*
-        * if this file hasn't been changed since the last transaction
-        * commit, we can safely return without doing anything
-        */
-       if (last_mod <= root->fs_info->last_trans_committed)
-               return;
-
-       spin_lock(&root->fs_info->ordered_root_lock);
-       if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
-               list_add_tail(&BTRFS_I(inode)->ordered_operations,
-                             &cur_trans->ordered_operations);
-       }
-       spin_unlock(&root->fs_info->ordered_root_lock);
-}
-
 int __init ordered_data_init(void)
 {
        btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",