Following up with Alan request here I am sending a patch for fs/lockd/svc.c
that contains the grace period control feature through the proc filesystem,
feedback & sugestions are very welcome as is the prompt inclusion in the
linux distribution :-)
(See attached file: userl-gracep-control.patch)
Regards, Juan
On Tue, Aug 27, 2002 at 03:06:33PM -0600, Juan Gomez wrote:
> Following up with Alan request here I am sending a patch for fs/lockd/svc.c
> that contains the grace period control feature through the proc filesystem,
> feedback & sugestions are very welcome as is the prompt inclusion in the
> linux distribution :-)
What do you need this for eaxctly?
> (See attached file: userl-gracep-control.patch)
Would be much nicer if inlined and in unified diff format.
Please follow Documentation/CodingStyle and i think for that purpose sysctl
are much better than procfs
Christoph, Alan, Neil,
Attached you will find the patch with the sysctl implementation of my
previous patch to enable grace period control from user-land.
Please let me know if this looks good enough for inclusion in the kernel
distribution or whether I still need to do something else.
Note this piece is derived from net/sunrpc/sysctl.c, which by the way I
think has a problem with the READ/WRITE verifys which seem
to be swicthed which I fixed in lockd version but not there, you may want
to take a look at net/sunrpc/sysctl.c and fix that although that's a minor
thing.
(See attached file: lockd-sysctl.patch)
Regards, Juan
|---------+---------------------------->
| | Alan Cox |
| | <alan@lxorguk. |
| | ukuu.org.uk> |
| | |
| | 09/05/02 11:37 AM|
| | |
|---------+---------------------------->
>-----------------------------------------------------------------------------------------------------------------------------|
| |
| To: Juan Gomez/Almaden/IBM@IBMUS |
| cc: |
| Subject: Re: NFS lockd patch proposal for user-level control of the grace period |
| |
| |
>-----------------------------------------------------------------------------------------------------------------------------|
I was waiting for a version that used the sysctl /proc/sys interface
instead. The concept of the interface is clearly fine
On Fri, Sep 06, 2002 at 04:36:39PM -0600, Juan Gomez wrote:
> Christoph, Alan, Neil,
Any chance you could resend this as a unified diff and with a proper mailer
so I actually have a chance to read it?
Thanks in advance,
Christoph
Am Sam, 2002-09-07 um 03.36 schrieb Christoph Hellwig:
> Any chance you could resend this as a unified diff and with a proper mailer
> so I actually have a chance to read it?
It's perfectly readable here except for not being in unified form which
is a minor problem.... maybe you should go check your own mailer....
--
Servus,
Daniel
On Friday September 6, [email protected] wrote:
>
>
>
>
> Christoph, Alan, Neil,
>
> Attached you will find the patch with the sysctl implementation of my
> previous patch to enable grace period control from user-land.
> Please let me know if this looks good enough for inclusion in the kernel
> distribution or whether I still need to do something else.
> Note this piece is derived from net/sunrpc/sysctl.c, which by the way I
> think has a problem with the READ/WRITE verifys which seem
> to be swicthed which I fixed in lockd version but not there, you may want
> to take a look at net/sunrpc/sysctl.c and fix that although that's a minor
> thing.
>
> (See attached file: lockd-sysctl.patch)
>
I still haven't managed to find out exactly what you want to do with
this, and hence whether it is appropriate.
You mentioned in another Email that this was for a High Availability
setup where one server might take-over a filesystem that another
server was previously serving.
If this is the case, do you really want to change the grace period, or
do you really want to re-start the grace period.
If that is what you really want, then I think that sysctl is
un-necessary and a simple signal would do the trick.
Currently, SIGKILL will
1/ drop all locks held for clients
2/ restart the grace period.
it would probably be quite sensible (and trivial to code) for SIGHUP
(say) to restart the grace period without dropping the locks.
Would this be suitable for you?
NeilBrown
I think either way we get to the kernel would be ok with me. I picked
sysctl because some comments in the code susggested somebody already though
of supporting a sysctl gate to user land to export
grace period control, also Chrisptop suggested this was the way to go when
I previously proposed using the raw proc filesystem. I think what is nice
about sysctl is that it is extensible, If we use signals and some other
folks keep on using signals for other control purposed then they will
eventually run out and they will have to add sysctl anyway. Said that I
think either way serves the purpose I want.
Juan
|---------+---------------------------------->
| | Neil Brown |
| | <[email protected]>|
| | Sent by: |
| | linux-kernel-owner@vger|
| | .kernel.org |
| | |
| | |
| | 09/08/02 05:27 PM |
| | |
|---------+---------------------------------->
>-----------------------------------------------------------------------------------------------------------------------------|
| |
| To: Juan Gomez/Almaden/IBM@IBMUS |
| cc: Alan Cox <[email protected]>, [email protected], [email protected] |
| Subject: Re: NFS lockd patch proposal for user-level control of the grace period |
| |
| |
>-----------------------------------------------------------------------------------------------------------------------------|
On Friday September 6, [email protected] wrote:
>
>
>
>
> Christoph, Alan, Neil,
>
> Attached you will find the patch with the sysctl implementation of my
> previous patch to enable grace period control from user-land.
> Please let me know if this looks good enough for inclusion in the kernel
> distribution or whether I still need to do something else.
> Note this piece is derived from net/sunrpc/sysctl.c, which by the way I
> think has a problem with the READ/WRITE verifys which seem
> to be swicthed which I fixed in lockd version but not there, you may
want
> to take a look at net/sunrpc/sysctl.c and fix that although that's a
minor
> thing.
>
> (See attached file: lockd-sysctl.patch)
>
I still haven't managed to find out exactly what you want to do with
this, and hence whether it is appropriate.
You mentioned in another Email that this was for a High Availability
setup where one server might take-over a filesystem that another
server was previously serving.
If this is the case, do you really want to change the grace period, or
do you really want to re-start the grace period.
If that is what you really want, then I think that sysctl is
un-necessary and a simple signal would do the trick.
Currently, SIGKILL will
1/ drop all locks held for clients
2/ restart the grace period.
it would probably be quite sensible (and trivial to code) for SIGHUP
(say) to restart the grace period without dropping the locks.
Would this be suitable for you?
NeilBrown