Jobs won't start using pdsh if script fd >= 10

[Jacek Konieczny]

After upgrading Upstart to 1.2 I found some of my jobs not starting (exiting with code '127'). After some investigation I found out that Upstart prepends 'exec 10<&-' to the scripts of the failing jobs, to close the input pipe. The problem is this won't work for some POSIX shells, which can handle only single-digit file descriptors this way. This works for the big and heavy bash, but won't for under PDKSH (used as /bin/sh in e.g. PLD Linux).

Can upstart be made to use some fixed file descriptor for this task? e.g. '3'?

[Jacek Konieczny (jajcus-jajcus)]

Attaching a quick hack (not an elegant solution) which solves the problem for me (the script fd is dup2()ed to fd #3 in the child process).

[Arkadiusz Miśkiewicz (arekm)]

Jacek also found this:

"Open files are represented by decimal numbers starting with zero. The largest possible value is implementation-dependent; however, all implementations support at least 0 to 9, inclusive, for use by the application. These numbers are called file descriptors. "

http://pubs.opengroup.org/onlinepubs/007908799/xcu/chap2.html

so posix shells are required to support file descriptiors 0-9 only.

[Scott James Remnant (scott)]

Thanks for the patch, actually I don't think that's a hacky way to do things, but there is an issue if the writing end of the error pipe happens to have that file descriptor - you could close it in the dup2() - I'll commit an expanded version of the patch

[Scott James Remnant (scott)]

http://bazaar.launchpad.net/~upstart-devel/upstart/trunk/revision/1280

[Yuri Zaporozhets (yuri-zaporozhets)]

Actually, the fix commited in 1280 does not work in all cases. Namely, when init has 7 (seven) files open. In this case the communication channel has the fd=8, and the first pipe call in job_process_run() returns (9, 10). This means that the script_fd parameter passed to job_process_spawn() also equals to 9. In such case the following code is screwed up completely

if (script_fd != -1) {
  int tmp = dup2(script_fd, JOB_PROCESS_SCRIPT_FD); /* dup2(9, 9) returns 9 */
  /*...*/
  close(script_fd); /* Close our current file descriptor */
  script_fd = tmp; /* ...which means a big trouble */
}

The correct way is, of course, not only to check script_fd against -1, but also check if it's already 9:

if ((script_fd != -1) && (script_fd != JOB_PROCESS_SCRIPT_FD)) {
  /*...*/
}

[Scott James Remnant (scott)]

Oh, I was apparently expecting dup2() to return -1 in the case of being given the same file descriptor

[Scott James Remnant (scott)]

http://bazaar.launchpad.net/~upstart-devel/upstart/trunk/revision/1308