Re: [PATCH 1/2] nfsd: make NFSv4 lease time per net

[Stanislav Kinsbursky <skinsbursky@parallels.com>]

28.11.2012 19:15, J. Bruce Fields пишет:
> On Wed, Nov 28, 2012 at 07:12:03PM +0400, Stanislav Kinsbursky wrote:
>> 28.11.2012 19:09, J. Bruce Fields пишет:
>>> On Tue, Nov 27, 2012 at 02:11:44PM +0300, Stanislav Kinsbursky wrote:
>>>> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
>>>> index dab350d..4930981 100644
>>>> --- a/fs/nfsd/nfsctl.c
>>>> +++ b/fs/nfsd/nfsctl.c
>>>> @@ -912,7 +912,8 @@ static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size, time_
>>>>    */
>>>>   static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
>>>>   {
>>>> -	return nfsd4_write_time(file, buf, size, &nfsd4_lease);
>>>> +	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
>>>> +	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease);
>>>
>>> This is called in the context of whatever process writes to
>>> nfsv4leasetime, so should be using its network namespace, right?
>>>
>>
>> This is, actually, a interim solution to preserve existent logic.
>> I.e. I'm going to convert "nfsd" filesystem into per-net one (like rpc_pipefs). I, actually, already done it in my tree.
>> Thus proper network namespace will be taken from nfsd superblock.
>
> OK, remind me how that works?  It's mounted just once, but each network
> namespace gets a different view of the filesystem?
>

Nope. It's a single mount point, but per-namespace (network, in our case) - not globally.
Pointer to namespace will be placed on sb->s_fs_info.

> --b.
>


-- 
Best regards,
Stanislav Kinsbursky