Re: [PATCH] mountd/exportd: only log confirmed clients, and poll for updates

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 2047 bytes --]

On Fri, Mar 19 2021, J. Bruce Fields wrote:

> On Fri, Mar 19, 2021 at 02:38:25PM +1100, NeilBrown wrote:
>> 
>> It is possible (and common with the Linux NFS client) for the nfs server
>> to receive multiple SET_CLIENT_ID or EXCHANGE_ID requests when starting
>> a connection.  This results in some clients appearing in
>>  /proc/fs/nfsd/clients
>> which never get confirmed.  mountd currently logs these, but they aren't
>> really helpful.
>> 
>> If the kernel supports the reporting of the confirmation status of
>> clients, we can suppress the message until a client is confirmed.
>> 
>> With this patch we:
>>  - record if the client is confirmed, assuming it is if the status is
>>     not reported
>>  - don't log unconfirmed clients
>>  - check all unconfirmed clients each time any event is processed,
>>  - if there are unconfirmed clients, we request a wakeup after a
>>    exponentially decreasing time, and check again
>
> increasing not decreasing, I think.

or frequency, not time??  Thanks.

>
> Is there any better way to let userland know when the contents of a
> virtual file have changed?

i had thought of using sysfs_notify_dirent(), though the file isn't in
sysfs (or kernfs), so extra work might be needed.  And I didn't want to
hold the file descriptor open, as then I would need to worry about where
the available fds could be exhausted.

>
> Looks at inofity man page....  There's an "IN_MODIFY" event.  I think we
> could add an fsnotify_inode(inode, FS_MODIFY); at the end of
> move_to_confirmed().  (I'm not sure what's the best way to get the inode
> of the info file there.)
>
> Would that help?

Yes, that is a much better idea.
We could pass an array of dentry pointers to nfsd_client_mkdir and
thence to nfsdfs_create_files.
Then create_client could grab the required dentry and stash it next to
cl_nfsd_dentry.  Then use fsnotify_dirent() which is better than _inode
as the file name gets included in the event.

I might give that a try.

Thanks,
NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 853 bytes --]