Re: [PATCH v2 11/35] pnfsd: get device list/info

[Benny Halevy <bhalevy@panasas.com>]

On Dec. 10, 2009, 21:04 +0200, "J. Bruce Fields" <bfields@citi.umich.edu> wrote:
> On Thu, Dec 10, 2009 at 08:53:08PM +0200, Benny Halevy wrote:
>> On Dec. 10, 2009, 19:30 +0200, " J. Bruce Fields" <bfields@citi.umich.edu> wrote:
>>> On Mon, Dec 07, 2009 at 11:32:10AM +0200, Benny Halevy wrote:
>>>> +static __be32
>>>> +nfsd4_getdevinfo(struct svc_rqst *rqstp,
>>>> +		struct nfsd4_compound_state *cstate,
>>>> +		struct nfsd4_pnfs_getdevinfo *gdp)
>>>> +{
>>>> +	struct super_block *sb;
>>>> +	struct svc_export *exp = NULL;
>>>> +	u32 fsidv = gdp->gd_devid.fsid;
>>>> +	int status;
>>>> +
>>>> +	dprintk("%s: layout_type %u dev_id %llx:%llx maxcnt %u\n",
>>>> +	       __func__, gdp->gd_layout_type, gdp->gd_devid.fsid,
>>>> +	       gdp->gd_devid.devid, gdp->gd_maxcount);
>>>> +
>>>> +	status = nfserr_inval;
>>>> +	exp = rqst_exp_find(rqstp, FSID_NUM, &fsidv);
>>> As I said before, this seems to require an fsid= option on every pnfs
>>> export.  We shouldn't need that.
>> If there a better way to get to the sb given the fsid?
>> Since we don't have a current_fh for getdeviceinfo the alternative
>> might be to map the deviceids to sb's in the generic layer.
> 
> Does the deviceid even need to persist over reboots?  If not, then just
> an index into an in-memory table might be good enough.  If it does need
> to persist, then I guess we should look at the filehandle-decoding code
> for inspiration.

No, the deviceids need not persist across reboots and we're free to put
anything we want in the major part of the deviceid, leaving the minor
for the filesystem-unique ID.

Benny

> 
> --b.