Byzantine Routers - create/remove

From: Neil Brown (neilb@cse.unsw.edu.au)
Date: 11/17/99-06:42:21 PM Z

• Next message: Neil Brown: "SYMLINK creation??"
• Previous message: Neil Brown: "Byzantine Routers - WRITE operations"

From: Neil Brown <neilb@cse.unsw.edu.au>
Date: Thu, 18 Nov 1999 11:42:21 +1100 (EST)
Message-ID: <14387.19309.687297.999073@notabene.cse.unsw.EDU.AU>
Subject: Byzantine Routers - create/remove


While thinking about the Byzantine Router problem, I wondered about
REMOVE calls, and the possible problem that a 
   REMOVE
   OPEN/CREATE
sequence could see the REMOVE being redelivered after the create with
undesirable consequences.
Forutately, the protocol as it stands can allow the client to avoid
that.

If the client does a LOOKUP first to get the filehandle (which it
possibly has anyway), and then does:
  COMPOUND { PUTFH dir; LOOKUP name; VERIFY(filehandle==fh); PUTFH  dir; DELETE name }

Then a replay of that compound wont delete the newly created file. 
This doesn't protect against
   link a b
   unlink b
   link a b

but maybe adding the "change" attribute to the VERIFY would help.

I have suggested before that the example for VERIFY in the draft isn't
the best as it doesn't guarantee much due to the lack of atomicity.
Would this be a suitable alternate example?

NeilBrown

• Next message: Neil Brown: "SYMLINK creation??"
• Previous message: Neil Brown: "Byzantine Routers - WRITE operations"

This archive was generated by hypermail 2.1.2 : 03/04/05-01:47:56 AM Z CST