Hi,
I have a question regarding blocked signals:
Is the current implementation to ignore attempts to set SIG_IGN on
blocked signals correct?
The following code will go into an endless loop on kernels 2.6.10 and
2.4.25, which is IMHO not the behaviour one would expect.
--------------------
#include <signal.h>
#include <sys/types.h>
#include <unistd.h>
volatile int sig_received = 0;
void sigio_handler_ex (int signum, siginfo_t * siginfo, void * ucontext)
{
struct sigaction sigio_action;
sig_received++;
printf("handler %d\n",sig_received);
sigio_action.sa_handler = SIG_IGN;
sigio_action.sa_flags = 0;
sigemptyset(&sigio_action.sa_mask);
sigaction (SIGIO, &sigio_action, 0);
kill(getpid(),SIGIO);
sigio_action.sa_sigaction = sigio_handler_ex;
sigio_action.sa_flags = SA_SIGINFO;
sigemptyset(&sigio_action.sa_mask);
sigaction (SIGIO, &sigio_action, 0);
}
int main(int argc, char **argv) {
struct sigaction sigio_action;
sigio_action.sa_sigaction = sigio_handler_ex;
sigio_action.sa_flags = SA_SIGINFO;
sigemptyset(&sigio_action.sa_mask);
sigaction (SIGIO, &sigio_action, 0);
kill(getpid(),SIGIO);
while (! sig_received) {
printf("waiting for signal\n");
sleep(1);
}
kill(getpid(),SIGIO);
printf("%d signals handled\n",sig_received);
}
--------------------
In kernel 2.6.10/kernel/signal.c sig_ignored() I found this comment:
...
/*
* Blocked signals are never ignored, since the
* signal handler may change by the time it is
* unblocked.
*/
if (sigismember(&t->blocked, sig))
return 0;
...
so it seems this behaviour is intentional, but I don't understand
it. Why should it matter if a signal handler may change while blocked,
if it is ignored also, which is a user request?
The machine im writing this mail on runs with the above lines
commented out without any problems so far...
All this resulted from problems a customer had with implementing a
whole protocol-stack to a serially attached device in a
signal-handler. After the handler ran (with SIG_IGN) there was always
an extra SIGIO which triggered the handler again. Of course the real
fix was to move the protocol-stack out of the handler but still it
should have worked since it was a controlled environment (so there
wasn't even a race between entering the handler and setting
SIG_IGN). Oh and it worked for years under some realtime variant of
hp-unix.
Please be so kind to CC any answer to me directly since I'm
currently not subscribed to lkml.
Yours
Tobias Grundmann
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
