Linux Socket Problem?

Dear all,

I am writing a program for communcating data among
4 machines. I use UDP/IP, and for each pair of machines
I use 4 different sockets: one for sending msg, one for
recving msg, the other two for sending and recving acks.
Only the socket for recving msg is set to be asynchronous
(by fcntl), others are blocking. To avoid message
interrupts, I implement two macros called BEGINCS and
ENDCS, for blocking and unblocking SIGIO and SIGALRM
signals resp. (the only signals occurring should be these
two, and also the SIGSEGV, which will be captured in
another part of my program).

Here is the C++ code for getting an ack after a process
sends out the message (some declarations omitted):

int arrived = 0;
while (arrived != 1) {
BEGINCS;
FD_ZERO(&readfds);
FD_SET(ackrfd[toproc], &readfds);
polltime.tv_sec = 0;
polltime.tv_usec = 0;
res = select(maxackrfd, &readfds, NULL, NULL, &polltime);
if (res == -1) {
printf("Error in select(), exiting\n");
exit(1);
}
if (FD_ISSET(ackrfd[toproc], &readfds) != 0)
arrived = 1;
ENDCS;
}
if (arrived == 1) {
recv_again:
BEGINCS;
s = sizeof(from);
// The ack is always 4 bytes
res = recvfrom(ackrfd[toproc], (char *)&rep, 4, 0,
(struct sockaddr *)&from, (socklen_t *)&s);
if ((res == -1) && (errno == EINTR)) {
printf("Interrupted when recv, try again\n");
goto recv_again;
}
if (res > -1) {
success = 1;
}
else {
printf("send problem\n");
}
}

The problem is that, I found the processes sometimes
get waiting forever at the recvfrom() call. It seems that
the select() has reported that the socket has a message
received, but it is not the case when attempting to get it.

I am using Linux Redhat 7.3 (kernel 2.4.18-3). Do I
miss something in the program, or is there any bug in
Linux sockets causing this problem?

Any suggestion here is welcomed. Thanks!

--
LaBird (Benny).
Email: Pls remove all nos. from my above email addr.

LaBird wrote:

> Dear all,
>
> I am writing a program for communcating data among
> 4 machines. I use UDP/IP, and for each pair of machines
> I use 4 different sockets: one for sending msg, one for
> recving msg, the other two for sending and recving acks.
> Only the socket for recving msg is set to be asynchronous
> (by fcntl), others are blocking. To avoid message
> interrupts, I implement two macros called BEGINCS and
> ENDCS, for blocking and unblocking SIGIO and SIGALRM
> signals resp. (the only signals occurring should be these
> two, and also the SIGSEGV, which will be captured in
> another part of my program).

select() should always be used with non-blocking sockets. That is a general
sockets programming rule.

--
Phil Frisbie, Jr.
Hawk Software
http://www.hawksoft.com

Dear Phil,

Thanks. I modified this code from an open-source
program segment. I believe the reason it is
written this way (but not just calling recv or recvfrom
without select) is to give a chance for the program
to get other signals and serve them between select()
(when some messages have arrived) and recvfrom()
(to get the actual message). If we just have written
recvfrom() without select(), but using BEGINCS (which
blocks other signals) before recvfrom() to prevent
message from being interrupted, there is no chance
for the signals from being served. This is just my
understanding on why the program is written that way.

After prolonged debugging efforts, I've just found out
the problem of the code. The error is due to a wrong
value of maxackrfd being passed in select().

--
LaBird (Benny).
Email: Pls remove all nos. from my above email addr.


"Phil Frisbie, Jr." <(E-Mail Removed)> wrote in message
news:SasZb.2472$_(E-Mail Removed)...
> select() should always be used with non-blocking sockets. That is a
general
> sockets programming rule.
>
> --
> Phil Frisbie, Jr.
> Hawk Software
> http://www.hawksoft.com
>