Skip to contentSkip to conversation

User-space notification of process end

DISCUSSION Glen Turner 2 participants 2 messages Feb 3, 2004 – Feb 4, 2004

lore
Share & tools
lore
#0 lore raw

Hi,

I am writing a application which needs to know fairly promptly if a daemon has died. I'd prefer not to alter the daemon source code or to run the program as a non-daemon child of a daemon watcher process.

I tried using fnctl(..., F_NOTIFY, ...) as
follows

  f = open("/proc/123", O_RDONLY);
  signal(SIGIO, handler);
  fcntl(f, F_NOTIFY, DN_DELETE | DN_RENAME);
  F_ZERO(&f_set);
  F_SET(f, &f_set);
  select(1, NULL, NULL, &f_set, NULL);

hoping I'd see the /proc/<processid>/* files being removed at process end.

But procfs doesn't seem to support fnctl(.., F_NOTIFY, ...) for parameters other than DN_ACCESS. This doesn't seem to be limited to my code (the dnotify program, which has much better signal handling, has the same behavior).

Suggestions, particularly ones which don't require polling for the existence of the watched process, are welcome.

uname -r says 2.4.22-1.2149.nptl, which is the latest Fedora Core 1 kernel. I'm willing to try 2.6 if that supports F_NOTIFY on /proc.

Thanks,
Glen

Signature · 3 lines
Glen Turner         Tel: (08) 8303 3936 or +61 8 8303 3936 
Network Engineer          Email: [email protected]
Australian Academic & Research Network   www.aarnet.edu.au
#1 lore raw
Quoted 3 lines

I am writing a application which needs to know fairly promptly if a daemon has died. I'd prefer not to ... run the program as a non-daemon child of a daemon watcher process.

Why not a watcher process, if I may ask. That is, I'd be tempted to stash a copy of the daemon's executable, and in place of the original executalbe put a piece of code that fork/exec'd the stashed copy, waited for the exit, and notified you as need be.

The available operations on files in the /proc file system tend to be a very limited subset of those on regular files; so I am not surprised that F_NOTIFY didn't work.

Signature · 3 lines
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <[email protected]> 1.650.933.1373