cvs commit: src/sys/kern uipc_socket.c

[Robert Watson]

On Tue, 7 Oct 2008, John Baldwin wrote:

> On Tuesday 07 October 2008 04:57:55 pm Robert Watson wrote:
>> rwatson     2008-10-07 20:57:55 UTC
>>
>>   FreeBSD src repository
>>
>>   Modified files:
>>     sys/kern             uipc_socket.c
>>   Log:
>>   SVN rev 183675 on 2008-10-07 20:57:55Z by rwatson
>>
>>   In soreceive_dgram, when a 0-length buffer is passed into recv(2) and
>>   no data is ready, return 0 rather than blocking or returning EAGAIN.
>>   This is consistent with the behavior of soreceive_generic (soreceive)
>>   in earlier versions of FreeBSD, and restores this behavior for UDP.
>>
>>   Discussed with: jhb, sam
>>   MFC after:      3 days
>
> I do find this behavior odd though.  I would expect
>
> 	recv(fd, NULL, 0)
>
> to discard the next packet from the socket if one is available rather than 
> returning success and not doing anything (and it seems that this is what it 
> does both before and now).  Similarly, I would expect recv(fd, NULL, 0) to 
> block on a blocking socket if there isn't a packet available.  It would be 
> orthogonal then to return EAGAIN in this case (no packet available, 
> zero-length user buffer) on a non-blocking socket.
>
> It seems that Solaris dropped this behavior (return 0) from their recv() 
> system call sometime after SunOS 4.0 from comments in the OpenSolaris 
> source. From reading __skb_recv_datagram(), it seems that Linux 2.6 returns 
> EAGAIN. NetBSD and OS X both have the odd behavior.  OpenBSD has the odd 
> behavior, but with a caveat of sorts having to do with control messages. 
> OpenBSD cvsweb annotate is down though, so I haven't found the reason for 
> their change.

Yes, I agree it's odd, and I'm not sure I like it.  I discovered the problem 
while writing edge-case regression tests for socket receive to better exercise 
soreceive_dgram, at first concluding it was a bug in soreceive_generic!  My 
feeling, though, is that I should leave behavior "compatible" for 7.1, and 
perhaps we should change it for 7.2.

Robert N M Watson
Computer Laboratory
University of Cambridge