NFS: Remove racy size manipulations in O_DIRECT

On success, the RPC callbacks will ensure that we make the appropriate calls
to nfs_writeback_update_inode()

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
This commit is contained in:
Trond Myklebust 2016-06-22 14:38:06 -04:00
parent a5314a7492
commit 2f3c7d87a3

View File

@ -376,15 +376,6 @@ static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write)
{
struct inode *inode = dreq->inode;
if (dreq->iocb && write) {
loff_t pos = dreq->iocb->ki_pos + dreq->count;
spin_lock(&inode->i_lock);
if (i_size_read(inode) < pos)
i_size_write(inode, pos);
spin_unlock(&inode->i_lock);
}
if (write)
nfs_zap_mapping(inode, inode->i_mapping);
@ -1058,14 +1049,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter)
if (!result) {
result = nfs_direct_wait(dreq);
if (result > 0) {
struct inode *inode = mapping->host;
iocb->ki_pos = pos + result;
spin_lock(&inode->i_lock);
if (i_size_read(inode) < iocb->ki_pos)
i_size_write(inode, iocb->ki_pos);
spin_unlock(&inode->i_lock);
/* XXX: should check the generic_write_sync retval */
generic_write_sync(iocb, result);
}