lkml.org 
[lkml]   [1999]   [Nov]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch] ipc updates
Andi Kleen wrote:
>
> As far as I can see it still has the resource race (resources are charged
> after sleeping without rechecking, so there is a window where the user is
> able to overcommit). Here is for reference my 2.2 patch that fixes the
> problem and Scott Maxwell's DoS.
>

I think my patch solves both problems:
* the overcommit race does not exist: the code never sleeps between
checking for free queue space and increasing q_cbytes.
I first load the message into kernel space, then acquire a spinlock and
(test for free space, add to queue, increase q_cbytes) atomically.
I know that you can slightly overcommit: each process that is sleeping
in sys_sndmsg() can overcommit by one message (ie MSGMAX), but I think
we can ignore that: it's negligable compared to the total memory for a
process (kernel stack, page tables, ...). The memory is recovered if an
interrupt aborts sys_sndmsg().

* Scott's DoS:
> +#define MSGQNUM 1024 /* <=65535 */ /* Max messages in flight / queue */

I use a simpler solution without a new limit:
sys_msgsnd():
- if(msgsz + msq->q_cbytes > msq->q_qbytes) {
- DECLARE_WAITQUEUE(wait,current);
+ if(msgsz + msq->q_cbytes > msq->q_qbytes ||
+ 1 + msq->q_qnum > msq->q_qbytes) {
+ struct msg_sender s;

I reuse the number of bytes in the queue. This allows the user to
overcommit by ~ 3%, IMHO that's acceptable.
(16384 one byte messages eat ~ 512 kB kernel memory, 2 8 Kb message +
16382 0 byte messages around 528 kB)


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:54    [W:0.079 / U:0.132 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site