read returns EAGAIN on a _blocking_ socket, but should not?

read returns EAGAIN on a _blocking_ socket, but should not?

[Marc Lehmann]

Hi!

The SuS documents EAGAIN for read(2) as:

   [EAGAIN]
       The O_NONBLOCK flag is set for the file descriptor and the thread would
       be delayed.

And the linux manpage for read(2) says something similar. Yet, I do get
EAGAIN when reading from a blocking socket. Here is a short strace:

   socket(PF_FILE, SOCK_STREAM, 0)         = 3
   connect(3, {sa_family=AF_FILE, path="/serv"}, 110) = 0
   write(3, "\0\3", 2)                     = 2
   write(3, "NEW", 3)                      = 3
... many writes and reads omitted
   sendmsg(3, {msg_name(0)=NULL, msg_iov(1)=[{"\0", 1}], msg_controllen=20, {cmsg_len=20, cmsg_level=SOL_SOCKET, cmsg_type=SCM_RIGHTS, {3}}, msg_flags=MSG_OOB|MSG_DONTROUTE}, 0)                            = 1
   read(3, "\0\3", 2)            = 2
   read(3, 0x51d030, 3)           = -1 EAGAIN (Resource temporarily unavailable)
   write(2, "protocol error: unexpected eof f"..., 44)                 = 44

(the sendmsg call in the source, incidentally, specifies "0" as flags
argument, not MSG_OOB|MSG_DONTROUTE, but this is likely irrelevant).

Here is the server side for the last exchange above:

   recvmsg(9, {msg_name(0)=NULL, msg_iov(1)=[{"\0", 1}], msg_controllen=24, {cmsg_len=20, cmsg_level=SOL_SOCKET, cmsg_type=SCM_RIGHTS, {10}}, msg_flags=0}, 0) = 1
... ioctls on other fds omitted
   write(9, "\0\3", 2) = 2
   write(9, "END", 3) = 3

Now, the first backtrace creates a unix domain socket, connects to an
existing one, passes a file descriptor and then read() returns with
EAGAIN. Nowhere does it set the socket to blocking.

The full backtraces can be found as http://data.plan9.de/x.20407 (client
getting EAGAIN) and at http://data.plan9.de/x.20406 (server writing).

I am looking for the obvious logical error I made, but I can only
come up with the conclusion that read returning EAGAIN on a blocking
socket must be a kernel bug. It is also interesting that this code and
the server/client exchanges worked reliably _until_ I started to pass
filehandles from client to server.

I am using linux 2.6.14 on x86_64.

Thanks for your time and any insights you can give!

-- 
                The choice of a
      -----==-     _GNU_
      ----==-- _       generation     Marc Lehmann
      ---==---(_)__  __ ____  __      pcg@goof.com
      --==---/ / _ \/ // /\ \/ /      http://schmorp.de/
      -=====/_/_//_/\_,_/ /_/\_\      XX11-RIPE

Re: read returns EAGAIN on a _blocking_ socket, but should not? (solved)

[Marc Lehmann]

On Sun, Dec 25, 2005 at 04:16:11PM +0100, Marc Lehmann <schmorp@schmorp.de> wrote:
> Hi!
> 
> The SuS documents EAGAIN for read(2) as:

Argl, false alarm, very very sorry :(

>    sendmsg(3, {msg_name(0)=NULL, msg_iov(1)=[{"\0", 1}], msg_controllen=20, {cmsg_len=20, cmsg_level=SOL_SOCKET, cmsg_type=SCM_RIGHTS, {3}}, msg_flags=MSG_OOB|MSG_DONTROUTE}, 0)                            = 1

Due to a bug, this sendmsg call passes the wrong fd (3), which the other
process then sets to non-blocking.

Again, sorry :(

-- 
                The choice of a
      -----==-     _GNU_
      ----==-- _       generation     Marc Lehmann
      ---==---(_)__  __ ____  __      pcg@goof.com
      --==---/ / _ \/ // /\ \/ /      http://schmorp.de/
      -=====/_/_//_/\_,_/ /_/\_\      XX11-RIPE