Project

General

Profile

Bug #4865

some osmo projects don't generate coredump upon SIGABRT

Added by pespin 13 days ago. Updated 8 days ago.

Status:
Feedback
Priority:
Normal
Assignee:
Target version:
-
Start date:
11/20/2020
Due date:
% Done:

90%

Spec Reference:

Description

See for instance osmo-bsc:

static void signal_handler(int signal)
{
    fprintf(stdout, "signal %u received\n", signal);

    switch (signal) {
    case SIGINT:
    case SIGTERM:
        bsc_shutdown_net(bsc_gsmnet);
        osmo_signal_dispatch(SS_L_GLOBAL, S_L_GLOBAL_SHUTDOWN, NULL);
        sleep(3);
        exit(0);
        break;
    case SIGABRT:
        /* in case of abort, we want to obtain a talloc report
         * and then return to the caller, who will abort the process */
    case SIGUSR1:
        talloc_report(tall_vty_ctx, stderr);
        talloc_report_full(tall_bsc_ctx, stderr);
        break;
    default:
        break;
    }
}

    signal(SIGINT, &signal_handler);
    signal(SIGTERM, &signal_handler);
    signal(SIGABRT, &signal_handler);
    signal(SIGUSR1, &signal_handler);
    signal(SIGUSR2, &signal_handler);

So when a process sends SIGABRT to it (killall -ABRT osmo-bsc), it will simply print talloc contexts and continue executing normally (and no coredump dumped). That's definetly not what we want here. Upon SIGABRT, we want the process to produce a coredump and exit.

In order to do so, after calling talloc_report we have to call the default SIGABRT handler.

That can be done by keeping reference to the older signal handler at startup when we voerride the signal:

sighandler_t default_sigabrt; /* global varialbe */
default_sigabrt = signal(SIGABRT, &signal_handler);

Then in our signal_handler:


static void signal_handler(int signal)
{
...
    case SIGABRT:
        talloc_report(tall_vty_ctx, stderr);
        talloc_report_full(tall_bsc_ctx, stderr);
        default_sigabrt(signal);
        break;
...
}

History

#1 Updated by laforge 13 days ago

Nice catch!

#2 Updated by pespin 8 days ago

So this issue only appears when sending plain SIGABRT, abort() should handle this gracefully. From "man abort":

DESCRIPTION
       The  abort()  function first unblocks the SIGABRT signal, and then raises that signal for the calling process (as though raise(3) was called).  This
       results in the abnormal termination of the process unless the SIGABRT signal is caught and the signal handler does not return (see longjmp(3)).

       If the SIGABRT signal is ignored, or caught by a handler that returns, the abort() function will still terminate  the  process.   It  does  this  by
       restoring the default disposition for SIGABRT and then raising the signal for a second time.

So abort takes care of calling SIG_DFL if our handler returned. But in the event a plain SIGBART is sent, the program will continue without generating a coredump (because the code generating the coredump is inside SIGABRT SIG_DFL).

So the best is probably to do this in the SIGABRT signal handler:

signal(SIGABRT, SIG_DFL);
raise (SIGABRT);

#3 Updated by pespin 8 days ago

Fixed (tested) here:
https://gerrit.osmocom.org/c/osmo-bsc/+/21336 main: generate coredump and exit upon SIGABRT received

I will now look at similar issue in other osmocom projects.

#4 Updated by pespin 8 days ago

  • Status changed from New to Feedback
  • % Done changed from 0 to 90

I submitted a bunch of patches to gerrit fixing same issue in the other osmocom projects.
Once merged ticket can be closed.

Also available in: Atom PDF

Add picture from clipboard (Maximum size: 48.8 MB)