On Fri, Jan 31, 2003 at 08:18:50PM -0500, Wietse Venema wrote:
> Either you have solved the problem of assigning globally unique
> identifiers to objects. This seems unlikely to me.
>
> Or, more likely, you allow multiple uses of the same identifier,
> in which case identifier collisions are by definition possible.

If we could start over: in case of a successful write of tmp/uniq, do
not remove it, but have the MUA remove it after it removed new/uniq or
cur/uniq:info.

Since the above behavior is not compatible with the present one,
maintain the `uniq' identifier in a separate directory, col/ (for
collision), under Maildir. Both MDA and MUA will recognize whether to
use the old or new protocol by checking for the existence of the new
directory, col.

More precisely

0) Instead of three, create four subdirectories in a maildir: col,
   cur, new, tmp.

1) MDA creates tmp/uniq (and writes the message to it).

2) MDA links tmp/uniq to col/uniq;
   if cannot,

   unlink tmp/uniq
   sleep and start at 1) with a new `uniq'

3) MDA links tmp/uniq to new/uniq;
   if cannot

   unlink col/uniq
   unlink tmp/uniq
   sleep and start at 1) with a new `uniq'

   if successful

   unlink tmp/uniq

MDA is done; now the MUA will do either

A) unlink new/uniq
   unlink col/uniq

   at which point `uniq' can be safely reused by the MDA
or

A') rename new/uniq to cur/uniq:info

and later

B) unlink cur/uniq:info
   unlink col/uniq

   at which point `uniq' can be safely reused by the MDA.

>
> I use time as the label to distinguish between multiple instances
> of the same VnIn.hostname string.
>

I am not familiar with future plans, but would not the above be
problematic with NFS?

>
> > > I don't worry about blackhats resetting the clock on time servers
> > > or spoofing NTP server replies.
> >
> > Why not?
>
> Because it's a non-problem.

Meaning, it is not possible to lose mail as a result of it?

Mate

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]