[2726] in linux-net channel archive
tcp.c, FIONREAD bug?
daemon@ATHENA.MIT.EDU (Stefan Magdalinski)
Mon Apr 29 17:29:27 1996
Date: Mon, 29 Apr 1996 00:38:24 +0100 (BST)
From: Stefan Magdalinski <stefan@iandi.demon.co.uk>
To: "'LINUX-NET@vger.rutgers.edu'" <LINUX-NET@vger.rutgers.edu>
X-MAIL-From: owner-linux-net-outgoing@vger.rutgers.edu
X-RCPT-To: ik-linux@genesis.tcp.co.uk
Hi, I'm way out of my depth with low-level tcp stuff, but I have to
enquire about this. I've lurked for a while to see if anything pased
about this. It appears, and Randy Chapman (the guy porting Sun's JDK
to Linux) suggested this, that there's a bug in tcp.c, or at least
something that hasn't been implemented yet.
I've written a Java application that uses
TCP sockets, and intermittently checks to see if data is waiting by using
a call to available() in the Java.net library. available is purported to
return the number of bytes available to be read on that particular
socket, which indeed it does, on Solaris, and nearly under linux.
However, under linux (1.2.13, thru 1.3.91), available seems to exhibit
the following behaviour:
As soon as the value of available() becomes non-zero (i.e. some data is
on the socket), that value sticks, even if more data piles up, until at
least one call to read() has been performed on the socket. As an example
execution (each line represents the state at a call to available.
bytes on socket available() returns
0 0
0 0
5 5
8 5 //should be 8
Now, available is (according to Randy) a wrapper for the FIONREAD ioctl,
the code for which looks like this (from tcp.c)
---------------------------------------
case TIOCINQ:
#ifdef FIXME /* FIXME: */
case FIONREAD:
#endif
{
unsigned long amount;
if (sk->state == TCP_LISTEN)
return(-EINVAL);
lock_sock(sk);
amount = tcp_readable(sk);
release_sock(sk);
err=verify_area(VERIFY_WRITE,(void *)arg,
sizeof(int));
if(err)
return err;
put_user(amount, (int *)arg);
return(0);
}
----------------------------
I don't understand this, but the #define FIXME raises my suspicions.
Any ideas/comments?
I'd like to try and fix it myself, but I need a pointer or two in the
right direction.
*////////////////////////*
Stefan Magdalinski
I&I
stefan@iandi.demon.co.uk
*\\\\\\\\\\\\\\\\\\\\\\\\*
