Re: Strangeness when write()-ing zero-length datagrams

["Michael T Kerrisk" <mtk-lists@gmx.net>]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="us-ascii", Size: 2638 bytes --]

Bert,

> > On Linux 2.4.26 and 2.6.7, if we connect a datagram socket and 
> > then use write() to send 0 bytes, a datagram is NOT generated 
> > (this happens in both the Unix and Internet domains).  
> > Datagrams are generated in this case for send() and sendto() 
> > though.
> 
> Some postings were made about this some time ago, 

I didn't find anything with google or archive seach -- do
you have any more specific reference on this thread?

> it appears SUS has an
> opinion on zero byte writes:
> 
> If nbyte is 0, write() will return 0 and have no other results if the 
> file is a regular file; otherwise, the results are unspecified.
> http://www.opengroup.org/onlinepubs/007908799/xsh/write.html
> 
> The send(2) manpage states:
> 
> The send() function is identical to sendto() with a null pointer dest_len
> argument, and to write() if no flags are used.
> 
> Which means that a zero byte send() without flags is unspecified.

Yes, but on Linux write(fd, buf, 0) != send(fd, buf, 0, 0), which
does seem inconsistent.  And as I noted, all the other implementations
I tested behave differently from Linux.

(Just so I'm sure we have our scenarios clear, I'm referring to 
the following:

HOST-A                     HOST-B

create datagram socket

bind socket to 
INADDR_ANY:port-x

                            create datagram socket

                            connect socket to 
                            HOST-B:port-x

                            Now the following behave 
                            differently:
                                write(fd, buf, 0);
                                send(fd, buf, 0, 0);

)

> Also note that it is hard if impossible to usefully inform userspace of
> the reception of a zery byte packet - returning 0 from recvfrom may 
> also mean an 'orderly shutdown'.

Am I missing something?  There isn't the notion of a true 
connection with datagram sockets, and so no notion of an orderly 
shutdown (I am assuming you are meaning some scenario where the 
peer calls shutdown() or close()).

It is straightforward to detect a zero-length datagram: recvfrom()
returns 0.

> Do you actually have a need for zero byte packets?

No, but I discovered this difference during some testing that I 
was doing, and then observed that it seems different from 
most (all?) other implementations.  It seemed worth documenting 
this fact.

Cheers,

Michael

-- 
Michael Kerrisk
mtk-lists@gmx.net

NEU: WLAN-Router für 0,- EUR* - auch für DSL-Wechsler!
GMX DSL = supergünstig & kabellos http://www.gmx.net/de/go/dsl