Patchwork AF_PACKET: Fix PACKET_MMAP on 64-bit kernel/32-bit user systems

login
register
mail settings
Submitter Chase Douglas
Date Aug. 10, 2009, 5:03 p.m.
Message ID <20090810130359.p2hixk2eoow8kwg8@imap.linux.ibm.com>
Download mbox | patch
Permalink /patch/31090/
State Rejected
Delegated to: David Miller
Headers show

Comments

Chase Douglas - Aug. 10, 2009, 5:03 p.m.
PACKET_MMAP sets up a ring buffer at a memory location so that kernel  
and user space can read from it without extra overhead. However, the  
data is written with a header, struct tpacket_hdr, which begins with  
an unsigned long field, tp_status. This means in a split 64-bit/32-bit  
environment the kernel- and user-space reads from the mmap'd region  
will be different, since in kernel-space the field is 64-bit while in  
user-space the field is 32-bit.

The following patch makes the size of the structure constant by  
changing the tp_status field to 32-bits. It does not appear to need  
64-bits, since only 4-bits are actually used throughout the AF_PACKET  
code.

Signed-off-by: Chase Douglas <cndougla@linux.vnet.ibm.com>

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Patrick McHardy - Aug. 10, 2009, 5:17 p.m.
cndougla@linux.vnet.ibm.com wrote:
> PACKET_MMAP sets up a ring buffer at a memory location so that kernel
> and user space can read from it without extra overhead. However, the
> data is written with a header, struct tpacket_hdr, which begins with an
> unsigned long field, tp_status. This means in a split 64-bit/32-bit
> environment the kernel- and user-space reads from the mmap'd region will
> be different, since in kernel-space the field is 64-bit while in
> user-space the field is 32-bit.
> 
> The following patch makes the size of the structure constant by changing
> the tp_status field to 32-bits. It does not appear to need 64-bits,
> since only 4-bits are actually used throughout the AF_PACKET code.
> 
> Signed-off-by: Chase Douglas <cndougla@linux.vnet.ibm.com>
> --- a/include/linux/if_packet.h   2009-08-10 16:53:11.000000000 +0000
> +++ b/include/linux/if_packet.h   2009-08-10 16:53:28.000000000 +0000
> @@ -65,7 +65,7 @@ struct tpacket_auxdata
> 
>  struct tpacket_hdr
>  {
> -       unsigned long   tp_status;
> +       unsigned int    tp_status;

This changes the layout on 64 bit and thus breaks compatibility.
The tpacket2_hdr doesn't suffer from this problem.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Chase Douglas - Aug. 10, 2009, 6:52 p.m.
On Aug 10, 2009, at 1:17 PM, Patrick McHardy wrote:
> The tpacket2_hdr doesn't suffer from this problem.

I hadn't noticed the new header version. I've grabbed the latest  
version of libpcap which utilizes tpacket2_hdr where available, and  
everything is fixed again.

Thanks
David Miller - Aug. 11, 2009, 1:06 a.m.
You can't make that change without breaking all existing
64-bit binaries running on 64-bit kernels.

Secondarily, we're already fully aware of this problem, that's why
there is a second version of the data structure named tpacket2_hdr
which was created in a way which fixes the problem without breaking
existing setups.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

--- a/include/linux/if_packet.h   2009-08-10 16:53:11.000000000 +0000
+++ b/include/linux/if_packet.h   2009-08-10 16:53:28.000000000 +0000
@@ -65,7 +65,7 @@  struct tpacket_auxdata

  struct tpacket_hdr
  {
-       unsigned long   tp_status;
+       unsigned int    tp_status;
  #define TP_STATUS_KERNEL       0
  #define TP_STATUS_USER         1
  #define TP_STATUS_COPY         2