Sending data in a signal

The well-known kill system call has been around for decades and is used to send a signal to another process. The most common use is to terminate or kill another process by sending the KILL or TERM signal but it can be used for a form of IPC, usually around giving the other process a “kick” to do something.

One thing that isn’t as well known is besides sending a signal to a process, you can send some data to it. This can either be an integer or a pointer and uses similar semantics to the known kill and signal handler. I came across this when there was a merge request for procps. The main changes are using sigqueue instead of kill in the sender and using a signal action not a signal handler in the receiver.

To illustrate this feature, I have a small set of programs called sender and receiver that will pass an integer between them.

The Sender

The sender program is extremely simple, use a random(ish) from time masked to two bytes, put it in the required union and send the lot to sendqueue.

# include <signal.h>
# include <stdlib.h>
# include <stdio.h>
# include <time.h>

int main(int argc, char **argv)
{
    union sigval sigval;
    pid_t pid;
    if (argc < 2 || (pid = atoi(argv[1])) < 0)
	return EXIT_FAILURE;
    sigval.sival_int = time(NULL);
    printf("sender: sending %d to PID %d\n",
        sigval.sival_int, pid);
    sigqueue(pid, SIGUSR1, sigval);
    return EXIT_SUCCESS;
}

The key lines are 12 and 15 where the random (ish) integer is stored in the sigval union and then sent to the other process with the sigqueue.

The receiver

The receiver just sets up the signal handler, sends its PID (so I know what to tell the sender) and sits in a sleeping loop.

# include <stdlib.h>
# include <sys/types.h>
# include <unistd.h>
# include <signal.h>

void signal_handler(int signum, siginfo_t *siginfo, void *ucontext)
{
    if (signum != SIGUSR1) return;
    if (siginfo->si_code != SI_QUEUE) return;
    printf("receiver: Got value %d\n",
        siginfo->si_int);
}

int main(int argc, char **argv)
{
    pid_t pid = getpid();
    struct sigaction signal_action;
    printf("receiver: PID is %d\n", pid);
    signal_action.sa_sigaction = signal_handler;
    sigemptyset (&signal_action.sa_mask);
    signal_action.sa_flags = SA_SIGINFO;
    sigaction(SIGUSR1, &signal_action, NULL);
    while(1) sleep(100);
    return EXIT_SUCCESS;
}

Lines 19-22 setup the signal handler. The main difference here is SA_SIGINFO used for the signal flags and sigaction references a sa_sigaction function rather than sa_handler.

We need to use a different function because the sigaction only is passed the signal number but we need more information, including the integer that the sender process stored in sigval.

Lines 6-12 are the signal handler function itself. It first checks that the receiver process got the correct signal (SIGUSR1 in this case) and that we got this signal from sigqueue because the type is SI_QUEUE. Checking the type of signal is important because different signals give you different data. For example, if you signalled this process with kill then si_int is undefined.

The result

As a proof of concept, the results are not terribly exciting. We see the sender say what it will be sending and the receiver saying it got it. It was useful to get some results, especially when things went wrong.

$ ./receiver &;
[1] 370216
receiver: PID is 370216
$ ./sender 370216
sender: sending 133 to PID 370216
receiver: Got value 133

Gotchas

While testing the two processes there were two gotchas I encountered.

GDB and the siginfo structure

The sigaction manual page shows a simple siginfo_t structure, however, when looking at what is passed to the signal handler, it’s much more complicated.

(gdb) p *siginfo
$2 = {si_signo = 10, si_errno = 0, si_code = -1, __pad0 = 0,
  _sifields = {_pad = {371539, 1000, 11, 32766, 0 <repeats 24 times>},
  _kill = {si_pid = 371539, si_uid = 1000},
  _timer = {si_tid = 371539, si_overrun = 1000, 
    si_sigval = {sival_int = 11, sival_ptr = 0x7ffe0000000b}},
  _rt = {si_pid = 371539, si_uid = 1000,
    si_sigval = {sival_int = 11, sival_ptr = 0x7ffe0000000b}},
  _sigchld = {si_pid = 371539, si_uid = 1000, si_status = 11, si_utime = 0, si_stime = 0},
  _sigfault = {si_addr = 0x3e80005ab53, si_addr_lsb = 11,
    _bounds = {_addr_bnd = {_lower = 0x0, _upper = 0x0}, _pkey = 0}},
  _sigpoll = {si_band = 4294967667539, si_fd = 11},
  _sigsys = {_call_addr = 0x3e80005ab53, _syscall = 11, _arch = 32766}}}

(gdb) p siginfo->_sifields._rt.si_sigval.sival_int
$3 = 11

So the integer is stored in a union in a structure in a structure. Much harder to find than just simply sival_int.

The pointer is just a pointer

So perhaps sending an integer is not enough. The sigval is a union with an integer and a pointer. Could a string be sent instead? I changed line 13 of the sender so it used a string instead of an integer.

    sigval.sival_ptr = "Hello, World!"

The receiver needed a minor adjustment to print out the string. I tried this and the receiver segmentation faulted. What was going on?

The issue is the set of system calls does a simple passing of the values. So if the sender sends a pointer to a string located at 0x1234567 then the receiver will have a pointer to the same location. When the receiver tries to dereference the sival_ptr, it is pointing to memory that is not owned by it but by another process (the sender) so it segmentation faults.

The solution would be to use shared memory between the processes. The signal queue would then use the pointer to the shared memory and, in theory, all would be well.