Programming

In another part of the program, a file is opened using fopen(). Anyhow, I was wondering if using dup2() in the following snippet was a legal close-on-exec move.

Code:

 static int
 tel(char *user, char *tty, const char *what)
{
  pid_t   cpid, wpid;
  int     stats;

  cpid = fork();
  if (cpid < 0) {
    perror("fork failed");
    exit(EXIT_FAILURE);
  }

  if (cpid == 0) {
    dup2(0,STDIN_FILENO);
    dup2(1,STDOUT_FILENO);
    if((execlp("tel", "tel", user, tty, what, (char *)0)) < 0)
      perror("execlp tel failed");
    _exit(EXIT_FAILURE);
  }
 

  while ((wpid= wait(&stats)) != cpid && wpid != -1)
    ;

  return (WIFEXITED(stats) ? WEXITSTATUS(stats) : -2);

Eh?

0 == STDIN_FILENO

and

1 == STDOUT_FILENO

so what are you trying to achieve?

I would only call dup2 if the file descriptors were different.

Yes, if a file descriptor has the close on exec bit set, they will be closed when exec() is called, otherwise I'm not sure what you are getting at.

On the code snippet I posted, fork and exec get called like 50 times in a row. The guy that is helping me develop this program says that every once in a while he gets the error message

"Out of File Descriptors"

on the Unix machines at his Universities computers. I haven't gotten this error message yet. Anyhow, I figured this might because the program keeps opening new file descriptors everytime fork() and exec() are executed, but doesn't actually close any of them until the program exits.

I figure that if I do a close-on-exec, it might solve the problem.

1. What OS are you running on?
2. Roughly how many open file descriptors does a running process have?
3. What is the open files ulimit ( ulimit -n ) on the system where the errors happen?

1. no fork does open any new descriptors

2. exec is in another process so would not matter, kernel opens program and shared libraries on your behalf

1) The OS I'm running on is Fedora Core 6. However, the target systems are FreeBSD 5.x and OpenBSD 3.x. I even had some guy network engineer from ebay email me saying that he was able to get a clean compile of this engineering wonder on the Solaris machines at his work.

2)I have no idea. I don't have the permissions to run lsof on the target machines.

3)I was told the error happened on the computers that were running OpenBSD. I have no idea what the ulimit is on those machines. I guess I'll have to check later on this week.

I could have sworn to god that I read someone that fork and exec created new fd's in the book "Advanced Programming in the Unix Environment" by Stevens and Rago.