get_best_mounts fixes.

get_best_mounts fixes.

[Chris Feist]

I'm working on a patch to get the get_best_mounts function (in 
modules/mount_nfs.c) work a little better and handle a few end cases.

What was the skiplocal variable intended to do?

It seems that if it is not 0, the only thing that get_best_mount does is 
check and see if the 'what' string is null & if it isn't return 0.  And 
the only time the get_best_mounts function is called with skiplocal set 
to 1. is when it is calling itself when all the hosts have not responded 
within 100ms.

Is it reasonable to completely remove this variable?

Re: get_best_mounts fixes.

[raven]

On Fri, 4 Feb 2005, Chris Feist wrote:

> I'm working on a patch to get the get_best_mounts function (in 
> modules/mount_nfs.c) work a little better and handle a few end cases.

Sounds great.

I really appreciate having Jeff and yourself helping out so much.
It's so much better to have people to challenge your thinking and 
contribute stuff to get the old grey matter wheels turning.

>
> What was the skiplocal variable intended to do?

Good question. I'll have a look and get back.

I don't want to escape the blame for this but ...

When I first got that patch it looked fine but was badly broken. I've 
reworked the code a good number of times now so there's bound to be 
oddities in it that I've missed.

>
> It seems that if it is not 0, the only thing that get_best_mount does is 
> check and see if the 'what' string is null & if it isn't return 0.  And the 
> only time the get_best_mounts function is called with skiplocal set to 1. is 
> when it is calling itself when all the hosts have not responded within 100ms.
>
> Is it reasonable to completely remove this variable?

I'll have a look.

Ian

Re: get_best_mounts fixes.

[raven]

On Fri, 4 Feb 2005, Chris Feist wrote:

> I'm working on a patch to get the get_best_mounts function (in 
> modules/mount_nfs.c) work a little better and handle a few end cases.

Oh, another unrelated thing.

I hope you didn't mind me jumping on the direct mounts work.

I've thought about how it could be done for so long and it was the 
original problem I wanted to fix when I first started work on autofs.

I just have to do it myself.

Anyway, sorry.

Ian

Re: get_best_mounts fixes.

[Chris Feist]

raven@themaw.net wrote:
> Oh, another unrelated thing.
> 
> I hope you didn't mind me jumping on the direct mounts work.

Sure, no problem.

> I've thought about how it could be done for so long and it was the 
> original problem I wanted to fix when I first started work on autofs.

Do you have any idea how long it may take? Let me know if there is 
anything I can do to help, we've got several users who *really* want 
this functionality.  I would also be happy to beta test it when you're 
ready.

Thanks,
Chris

Re: get_best_mounts fixes.

[Ian Kent]

On Mon, 7 Feb 2005, Chris Feist wrote:

> raven@themaw.net wrote:
> > Oh, another unrelated thing.
> > 
> > I hope you didn't mind me jumping on the direct mounts work.
> 
> Sure, no problem.
> 
> > I've thought about how it could be done for so long and it was the 
> > original problem I wanted to fix when I first started work on autofs.
> 
> Do you have any idea how long it may take? Let me know if there is 
> anything I can do to help, we've got several users who *really* want 
> this functionality.  I would also be happy to beta test it when you're 
> ready.

Thanks for the offer to help.

The implementation is pretty much something a single person needs to do.

The code is done for all but the HUP map reread. It's a bit difficult.
I can get it to mount and umount the direct mount triggers.
I can get it to mount map entries.
I'm working on expiring mounts now.
Then there's a second pass to try and make the error checking sensible.
I'll need help testing about here.
Then there's the HUP signal processing.
Then there's a third pass to attempt to address loose ends noticed during 
the development.

I don't know how long it will be as I can't really tell what holdups there 
might be if I find hard unanticipated problems. As always getting time is a 
problem. Having said that, probably not much longer. A week or a little 
more if all goes well.

There are a couple of difficulties with testing.
The code is against 4.1.4 beta1.
There's a kernel patch as well.

Ian

Re: get_best_mounts fixes.

[Ian Kent]

On Fri, 4 Feb 2005, Chris Feist wrote:

> I'm working on a patch to get the get_best_mounts function (in 
> modules/mount_nfs.c) work a little better and handle a few end cases.
> 
> What was the skiplocal variable intended to do?
> 
> It seems that if it is not 0, the only thing that get_best_mount does is 
> check and see if the 'what' string is null & if it isn't return 0.  And 
> the only time the get_best_mounts function is called with skiplocal set 
> to 1. is when it is calling itself when all the hosts have not responded 
> within 100ms.
> 
> Is it reasonable to completely remove this variable?
> 

Looking at the routine I've made a bit of a mess.
I doesn't do quite what I thought I made it do. Never mind.

I don't think that skiplocal should be removed as it should determine 
whether a check to see if the host is equivalent to localhost is done.

I'm still trying to decide, but I think that the skiplocal test in the 
early part of get_best_mount should be removed.