btrfs: disable strict file flushes for renames and truncates

Truncates and renames are often used to replace old versions of a file
with new versions.  Applications often expect this to be an atomic
replacement, even if they haven't done anything to make sure the new
version is fully on disk.

Btrfs has strict flushing in place to make sure that renaming over an
old file with a new file will fully flush out the new file before
allowing the transaction commit with the rename to complete.

This ordering means the commit code needs to be able to lock file pages,
and there are a few paths in the filesystem where we will try to end a
transaction with the page lock held.  It's rare, but these things can
deadlock.

This patch removes the ordered flushes and switches to a best effort
filemap_flush like ext4 uses. It's not perfect, but it should fix the
deadlocks.

Signed-off-by: Chris Mason <clm@fb.com>

1 files changed, 3 insertions, 44 deletions

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 8ea7610fbaf3..73098328d040 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -7951,27 +7951,6 @@ static int btrfs_truncate(struct inode *inode)
         BUG_ON(ret);
 
         /*
-         * setattr is responsible for setting the ordered_data_close flag,
-         * but that is only tested during the last file release.  That
-         * could happen well after the next commit, leaving a great big
-         * window where new writes may get lost if someone chooses to write
-         * to this file after truncating to zero
-         *
-         * The inode doesn't have any dirty data here, and so if we commit
-         * this is a noop.  If someone immediately starts writing to the inode
-         * it is very likely we'll catch some of their writes in this
-         * transaction, and the commit will find this file on the ordered
-         * data list with good things to send down.
-         *
-         * This is a best effort solution, there is still a window where
-         * using truncate to replace the contents of the file will
-         * end up with a zero length file after a crash.
-         */
-        if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
-                                           &BTRFS_I(inode)->runtime_flags))
-                btrfs_add_ordered_operation(trans, root, inode);
-
-        /*
          * So if we truncate and then write and fsync we normally would just
          * write the extents that changed, which is a problem if we need to
          * first truncate that entire inode.  So set this flag so we write out
@@ -8118,7 +8097,6 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
         mutex_init(&ei->delalloc_mutex);
         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
         INIT_LIST_HEAD(&ei->delalloc_inodes);
-        INIT_LIST_HEAD(&ei->ordered_operations);
         RB_CLEAR_NODE(&ei->rb_node);
 
         return inode;
@@ -8158,17 +8136,6 @@ void btrfs_destroy_inode(struct inode *inode)
         if (!root)
                 goto free;
 
-        /*
-         * Make sure we're properly removed from the ordered operation
-         * lists.
-         */
-        smp_mb();
-        if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
-                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 (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
                      &BTRFS_I(inode)->runtime_flags)) {
                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
@@ -8350,12 +8317,10 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
         ret = 0;
 
         /*
-         * we're using rename to replace one file with another.
-         * and the replacement file is large.  Start IO on it now so
-         * we don't add too much work to the end of the transaction
+         * we're using rename to replace one file with another.  Start IO on it
+         * now so  we don't add too much work to the end of the transaction
          */
-        if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
-            old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
+        if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
                 filemap_flush(old_inode->i_mapping);
 
         /* close the racy window with snapshot create/destroy ioctl */
@@ -8403,12 +8368,6 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
                  */
                 btrfs_pin_log_trans(root);
         }
-        /*
-         * make sure the inode gets flushed if it is replacing
-         * something.
-         */
-        if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
-                btrfs_add_ordered_operation(trans, root, old_inode);
 
         inode_inc_iversion(old_dir);
         inode_inc_iversion(new_dir);