Re: [PATCH v7 0/5] xfs: Allow pquota and gquota to be used together

[Dave Chinner <david@fromorbit.com>]

On Fri, Mar 22, 2013 at 10:00:46AM -0500, Chandra Seetharaman wrote:
> Sure. I will look at your CRC patchset and see what changes are needed
> to mine.
> 
> Meantime, you can answer my dilemma with the change in fs_quota_stat
> (see patch 4).
> 
> As per your suggestion, I changed the data structure, to include pads at
> relevant places in the data structure, which also means that the new
> modified structure is not compatible with older user space code. 
> 
> So, I created a copy of the older structure (fs_quota_stat_v1) and used
> that if the user space request has older structure. So, now the new
> kernel code is backward compatible with the old (already running) user
> space code. All is well till now....
> 
> Now, I make the changes to the user space code so as to use the new
> fs_quota_stat. Since the data structure has changed, the new user space
> code will not work with the old kernel code.

The issue that the userspace code has to handle is that old kernels
decide the version of the structure, not userspace. i.e. the
version number is an output parameter, not an input parameter. Hence
on old kernels we have to detect the version after the call rather
than assume the kernel treated the version as an input...

> This is where I have the dilemma. 
> 
> 1. Leave the user space code to work only with the newer version,
> breaking the compatibility with the older kernel (with an error
> message).
>    or
> 2. Read the superblock, and if pquotino is _not_ present conclude older
> version of fs_quota_stat would suffice for getting quota information
> from kernel (note that this will work properly with newer kernel code
> too), and use the older version of fs_quota_stat_v1 for quotactl.
>    or
> 3. Any other clever way ? :)

Because the version is only an output field for old kernels, the
userspace code should be able to handle it like this:

	struct fs_quota_stat fqs;

	ASSERT(sizeof(struct fs_quota_stat) >= sizeof(struct fs_quota_stat_v1));

	memset(&fqs, 0, sizeof(fqs));
	fqs.version = FS_QSTAT_VERSION_2;
	error = quotactl(...., &fqs);

	if (fqs.version == FS_QSTAT_VERSION) {
		struct fs_quota_stat_v1 *fqsv1 = &fqs;

		/* process as version 1 */
		.....
		return error;
	}

	/* process as version 2 */
	.....
	return error;

However, with the version field in the same place for both
structures, userspace can read the version number after the call and
determine the correct structure type for the information returned by
the kernel...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs