asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Chris Friesen]

I have a SOCK_STREAM PF_UNIX socket with no special options enabled using fcntl().

The default with sendmsg() and recvmsg() is to block.

With sndmsg(), I can use the MSG_DONTWAIT flag to enable non-blocking operation 
for that one call.

How do I do the equivalent for recvmsg()?

Thanks,

Chris

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Michael T Kerrisk]

> I have a SOCK_STREAM PF_UNIX socket with no special options enabled using
> fcntl().
> 
> The default with sendmsg() and recvmsg() is to block.
> 
> With sndmsg(), I can use the MSG_DONTWAIT flag to enable non-blocking
> operation for that one call.
> 
> How do I do the equivalent for recvmsg()?

MSG_DONTWAIT should also work with recvmsg().

Cheers

-- 
Michael Kerrisk
mtk-lists@gmx.net

"Sie haben neue Mails!" - Die GMX Toolbar informiert Sie beim Surfen!
Jetzt aktivieren unter http://www.gmx.net/info

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Chris Friesen]

Michael T Kerrisk wrote:

> MSG_DONTWAIT should also work with recvmsg().

Hmm...  Just tried it with a DGRAM socket, and it seems to work.  Any ideas why 
its not in the man pages?

Chris

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Andi Kleen]

On Thu, Jul 08, 2004 at 12:21:35PM -0400, Chris Friesen wrote:
> Michael T Kerrisk wrote:
> 
> >MSG_DONTWAIT should also work with recvmsg().
> 
> Hmm...  Just tried it with a DGRAM socket, and it seems to work.  Any ideas 
> why its not in the man pages?

Nobody ever added it? Just send a patch to aeb@cwi.nl

-Andi

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Chris Friesen]

[-- Attachment #1: Type: text/plain, Size: 1468 bytes --]

Andi Kleen wrote:
> On Thu, Jul 08, 2004 at 12:21:35PM -0400, Chris Friesen wrote:
>  > Michael T Kerrisk wrote:
>  >
>  > >MSG_DONTWAIT should also work with recvmsg().
>  >
>  > Hmm...  Just tried it with a DGRAM socket, and it seems to work.  Any 
> ideas
>  > why its not in the man pages?
> 
> Nobody ever added it? Just send a patch to aeb@cwi.nl

Sending patch as suggested.  Fundamentally, the delta is as follows, I've 
included an attachment with what I hope are the proper formatting codes (copied 
from send(2)).

--- recv.man    2004-07-08 15:43:17.000000000 -0400
+++ recv2.man   2004-07-08 15:47:29.000000000 -0400
@@ -67,6 +67,11 @@
                disconnect occurs, or the next data to be received is of a  dif-
                ferent type than that returned.

+       MSG_DONTWAIT
+              Enables  non-blocking  operation;  if the operation would block,
+              EAGAIN is returned (this can also be enabled  using  the  O_NON-
+              BLOCK with the F_SETFL fcntl(2)).
+
         MSG_NOSIGNAL
                This  flag  turns  off raising of SIGPIPE on stream sockets when
                the other end disappears.


Note also that there is a mention of MSG_DONTWAIT in the msg_flags field in the 
msghdr.  It gives the impression that one can *set* that field to cause the 
non-blocking behaviour.  My understanding is that the msg_flags field is a 
return value only.  Perhaps that portion should be reworded as well.

Chris

[-- Attachment #2: man2recv.diff --]
[-- Type: application/octet-stream, Size: 741 bytes --]

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Andries Brouwer]

On Thu, Jul 08, 2004 at 12:49:28PM -0400, Chris Friesen wrote:

> > > why its not in the man pages?
> >
> >Nobody ever added it? Just send a patch to aeb@cwi.nl
> 
> Sending patch as suggested.  Fundamentally, the delta is as follows, I've 
> included an attachment with what I hope are the proper formatting codes 
> (copied from send(2)).
> 
> --- recv.man    2004-07-08 15:43:17.000000000 -0400
> +++ recv2.man   2004-07-08 15:47:29.000000000 -0400
> @@ -67,6 +67,11 @@
>                disconnect occurs, or the next data to be received is of a  
>                dif-
>                ferent type than that returned.
> 
> +       MSG_DONTWAIT
> +              Enables  non-blocking  operation;  if the operation would 
> block,
> +              EAGAIN is returned (this can also be enabled  using  the  
> O_NON-
> +              BLOCK with the F_SETFL fcntl(2)).
> +
>         MSG_NOSIGNAL
>                This  flag  turns  off raising of SIGPIPE on stream sockets 
>                when
>                the other end disappears.
> 
> 
> Note also that there is a mention of MSG_DONTWAIT in the msg_flags field in 
> the msghdr.  It gives the impression that one can *set* that field to cause 
> the non-blocking behaviour.  My understanding is that the msg_flags field 
> is a return value only.  Perhaps that portion should be reworded as well.
> 
> Chris


Can you find the man-pages-1.67 package and construct a concrete patch?

(Don't know precisely what you want to do. The above gives a filename
recv.man, which would be recv.2 in my sources. But there is no MSG_NOSIGNAL
in recv.2, only in send.2. The fragment of text that you give occurs
already in recv.2.)

Andries

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Chris Friesen]

Andries Brouwer wrote:
> Can you find the man-pages-1.67 package and construct a concrete patch?
> 
> (Don't know precisely what you want to do. The above gives a filename
> recv.man, which would be recv.2 in my sources. But there is no MSG_NOSIGNAL
> in recv.2, only in send.2. The fragment of text that you give occurs
> already in recv.2.)

Sorry, the filenames have no meaning since I wasn't working from the package, 
but from the output of the "man" command.

Working from the webpage at:

http://homepages.cwi.nl/~aeb/linux/man2html/man2/recv.2.html

There is mention of MSG_DONTWAIT when discussing the "msg_flags" field in 
"struct msghdr".  It contains exactly the text proposed.  However, the msg_flags 
field is set on *return* of the call.  Every other flag discussed there 
"indicates" something.  The blurb for MSG_DONTWAIT says that it "enables" a 
behaviour, which doesn't make sense for a field set on call return.

There is no mention of MSG_DONTWAIT when discussing the "flags" parameter of the 
recv/recvfrom/recvmsg call.

Generally, I want to move the current blob from the "msg_flags" area to the 
"flags" area, and then add a different discussion for "msg_flags" if appropriate 
(I'm not sure about that part).

Do you still want me to generate an actual patch against man-pages-1.67, or is 
that description enough?

Chris

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Andries Brouwer]

On Thu, Jul 08, 2004 at 02:33:19PM -0400, Chris Friesen wrote:
> Andries Brouwer wrote:
> >Can you find the man-pages-1.67 package and construct a concrete patch?
> >
> >(Don't know precisely what you want to do. The above gives a filename
> >recv.man, which would be recv.2 in my sources. But there is no MSG_NOSIGNAL
> >in recv.2, only in send.2. The fragment of text that you give occurs
> >already in recv.2.)
> 
> Working from the webpage at:
> 
> http://homepages.cwi.nl/~aeb/linux/man2html/man2/recv.2.html

Ah, that is how I am punished for neglecting to update these html pages :-)
Will update them.

I was confused by your MSG_NOSIGNAL, but indeed, that is still on that page,
it was removed later.

Have moved the MSG_DONTWAIT part to after the MSG_WAITALL description.
[Will still check whether that is really appropriate.]

Thanks!

Andries

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Nivedita Singhvi]

Chris Friesen wrote:
> Michael T Kerrisk wrote:
> 
>> MSG_DONTWAIT should also work with recvmsg().
> 
> 
> Hmm...  Just tried it with a DGRAM socket, and it seems to work.  Any 
> ideas why its not in the man pages?

Think you are on an older version of the man pages.
It's been there for a while.

thanks,
Nivedita

Re: asymmetry with MSG_DONTWAIT in sendmsg() and recvmsg()

[Chris Friesen]

Nivedita Singhvi wrote:
> Chris Friesen wrote:
>  > Michael T Kerrisk wrote:
>  >
>  >> MSG_DONTWAIT should also work with recvmsg().
>  >
>  >
>  > Hmm...  Just tried it with a DGRAM socket, and it seems to work.  Any
>  > ideas why its not in the man pages?
> 
> Think you are on an older version of the man pages.
> It's been there for a while.

It's mentioned in the "msg_flags" section, which is set on call *return*.  It's 
not mentioned in the "flags" section.

Perhaps its just in the wrong part of the man page...

Chris