Re: [PATCH] close race condition in shared memory mapping/unmapping

[Neil Horman <nhorman@redhat.com>]

Chris Wright wrote:
> * Neil Horman (nhorman@redhat.com) wrote:
> 
>>Hey all-
>>	Found this the other day poking through the ipc code.  There appears to 
>>be a race condition in the counter that records how many processes are 
>>accessing a given shared memory segment.  In most places the shm_nattch 
>>variable is protected by the shm_ids.sem semaphore, but there are a few 
>>openings which appear to be able to allow a corruption of this variable 
>>when run on SMP systems.  I've attached a patch to 2.6.9-rc2 for review. 
>>  The locking may be a little over-aggressive (I was following examples 
>>from other points in this file), but I figure better safe than sorry :).
> 
> 
> Are you sure you've got this right?  I thought that the shmid_kernel
> struct protects shm_nattch with a local (per structure) lock which is
> embedded in kern_ipc_perm.  Did you find shm_nattch changes w/out
> shm_lock/shm_unlock around it?  I believe shm_ids.sem is protecting the
> id allocation, not per object data such as shm_nattch.
> 
> thanks,
> -chris
You're right, its not correct.  I'm sorry.  I'm looking into a locking 
bug in 2.4, which does its ipc locking for shared memory very 
differently.  Since the shared memory code looks simmilar in 2.6, I was 
making the assumption that the change applied upstram as well, but I 
don't think thats the case after looking more carefully.  My bad.
Neil

-- 
/***************************************************
  *Neil Horman
  *Software Engineer
  *Red Hat, Inc.
  *nhorman@redhat.com
  *gpg keyid: 1024D / 0x92A74FA1
  *http://pgp.mit.edu
  ***************************************************/