[9fans] simple questions on bind semantics

[9fans] simple questions on bind semantics

[Roman Shaposhnick]

I have a deja vu feeling that this might have been asked here before.
But since the mighty google disagrees here it goes: why does bind
have a restriction that old has be existent ? Basically why can't
I do:

   $ bind /some-file /new-name-for-some-file ?

where there's no /new-name-for-some-file before the call ?

Thanks,
Roman.

Re: [9fans] simple questions on bind semantics

[Rob Pike]

Because bind and mount are implemented by a simple substitution.
When a walk hits a file that has been bound to, the fid is dropped
and then replaced by a clone of the fid of the result of the bind.
To have a fid, one needs a file, hence the fid must exist.

There's nothing stopping a service from creating bind destinations
on demand, and I think this has even been done in a special
device(memory is vague), but fundamentally, bind needs a file to
bind to.

-rob


On Mon, 27 Sep 2004 17:46:57 -0700, Roman Shaposhnick <rvs@sun.com> wrote:
> I have a deja vu feeling that this might have been asked here before.
> But since the mighty google disagrees here it goes: why does bind
> have a restriction that old has be existent ? Basically why can't
> I do:
> 
>    $ bind /some-file /new-name-for-some-file ?
> 
> where there's no /new-name-for-some-file before the call ?
> 
> Thanks,
> Roman.
>

Re: [9fans] simple questions on bind semantics

[C H Forsyth]

mntgen(4), for instance, generates names on demand.

Re: [9fans] simple questions on bind semantics

[Roman Shaposhnick]

On Tue, Sep 28, 2004 at 02:57:42PM +0100, C H Forsyth wrote:
> mntgen(4), for instance, generates names on demand.
  
  mntgen seems new to me (I could bet it wasn't part of the 
  Plan 9 release 3, but may be I've just overlooked it)

  Anyway, is this how one accomplishes what I was trying 
  to do (bind /existing-file /non-existing-file):

   
     $ mntgen /
     $ bind /existing-file /non-existing-file

  The one thing that troubles me about this approach is that
  all of a sudden any lookup in / starts succeeding.

Thanks,
Roman.

Re: [9fans] simple questions on bind semantics

[andrey mirtchovski]

   
>   mntgen seems new to me (I could bet it wasn't part of the 
>   Plan 9 release 3, but may be I've just overlooked it)
> 

nemo wrote mntgen sometime around the initial 4e...

>   Anyway, is this how one accomplishes what I was trying 
>   to do (bind /existing-file /non-existing-file):

whenever i run into that problem i use ramfs, or i bind /tmp if
i don't care that someone may see the changes to the file.

hth: andrey

Re: [9fans] simple questions on bind semantics

[Roman Shaposhnick]

On Mon, Sep 27, 2004 at 05:51:54PM -0700, Rob Pike wrote:
> To have a fid, one needs a file, hence the fid must exist.

  That's the part I'm curious about. Why is that ? For the
  purposes of substituting you still have to maintain the
  information that file X is a 'bind' destination for file Y.
  Sure the lookup is much simpler when you really have Y and
  Z being served somewhere, but wouldn't it be quite simple
  to have a table (with a usual performance hit) that you consult
  when the real walk fails ?

  Or, to put differently, why mntgen is considered a better
  (or cleaner) solution than what I've just described.

Thanks,
Roman.

Re: [9fans] simple questions on bind semantics

[Fco. J. Ballesteros]

>>   Anyway, is this how one accomplishes what I was trying 
>>   to do (bind /existing-file /non-existing-file):

In Plan B we used to have a prefix table (now it's gone).
There you could say something like

/a/path/I/just/invented -goes-to-> a-file-server

This has the net effect that you can invent paths.
It was not that useful. I mean, you can perfectly do without
inventing files. Sometimes I missed this to patch up CDs, but
that's a different issue.

hth

Re: [9fans] simple questions on bind semantics

[Richard Miller]

> There's nothing stopping a service from creating bind destinations
> on demand, and I think this has even been done in a special
> device

That's mntgen(4).

But if you want to create only one specific bind destination instead
of having them appear on demand, you can use stub(8).

Re: [9fans] simple questions on bind semantics

[Roman Shaposhnick]

On Wed, Sep 29, 2004 at 06:06:00PM +0100, Richard Miller wrote:
> > There's nothing stopping a service from creating bind destinations
> > on demand, and I think this has even been done in a special
> > device
> 
> That's mntgen(4).
> 
> But if you want to create only one specific bind destination instead
> of having them appear on demand, you can use stub(8).

  Thanks. stub(8) looks pretty much like what I really need. However,
  it got me thinking: what if I need more stubs than one -- for every
  one of them there'll be a separate process serving one file, right ?
  Even with COW paging this sort of looks like an overkill, wouldn't 
  it be possible to serve all "stubs" with one process, using the
  third ("spec") option of mount as a distinguishing factor.

Thanks,
Roman.