Re: [PATCH v2 2/2] fuse: permit O_DIRECT flag in open()

[Anand Avati <avati@gluster.com>]

On Thu, Jul 28, 2011 at 2:35 AM, Christoph Hellwig <hch@infradead.org> wrote:
> On Wed, Jul 27, 2011 at 03:56:19AM -0700, Anand Avati wrote:
>> FUSE currently disallows O_DIRECT flag in open(). It is tricky dealing with
>> setting/unsetting of O_DIRECT flag on an open file. There are applications
>> (primarily VMs and databases) which open files with O_DIRECT flag. These
>> applications do not work on FUSE due to this limitation.
>>
>> The approach with this patch is to permit opens with O_DIRECT, but instead
>> disable setting/unsetting of the O_DIRECT flag no matter how the file was
>> opened. This limitation is for more practical than disallowing O_DIRECT
>> altogether.
>
> But it's also entirely incorrect.  Fix your userspace to have a proper
> fallback if O_DIRECT opens fail.  It's a feature only supported by a few
> filesystems, and very few operating systems.

FUSE already has a direct IO implementation (struct file_operations
fuse_direct_io_file_operation) but is currently set as a file's f_ops
only based on flags returned from the filesystem server (in the reply
of an open() call). This patch just lets the filesystem server know a
userspace application's intention to open a file with O_DIRECT and
leaves it with the decision to pick direct IO mode on the file (just
the way it already does before the patch.) There is already a
framework for direct IO operations and this patch only bridges
O_DIRECT flag with that framework. Do you still think this is entirely
incorrect?

Avati
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html