Hi Alexander,
On Thu, Aug 19, 2010 at 06:46:41PM +0300, Alexander Shishkin wrote:
> +AC_CHECK_FUNCS([eventfd])
please, see autoconf stuff and code for unshare(1) command. There
should be:
UTIL_CHECK_SYSCALL([eventfs])
and
AM_CONDITIONAL(BUILD_EVENTFD, ...)
> +usrbin_exec_PROGRAMS += cytune setarch eventfd
if BUILD_EVENTFD
usrbin_exec_PROGRAMS += eventfd
endif
our build system has to be usable on systems without eventfd(2).
> +$ eventfd -n PINGFD -w 'printf "ping received (%d)" $__EVENTFDVAL'
> +$ eventfd -n PINGFD -e 1
I don't like the -w options. The first horrible things is that the
variable name is hardcoded and second things is that this feature is
implemented by system() function. What about to implement "-w" as a
separate command (like "-e")?
I don't understand why your code uses fork(), you can control this
thing from shell by "&".
What about:
- setup a new event FD and starts a new shell or <command>:
$ eventfd -n <name> [<command> [<argv> ...]]
- read value from event FD and print to stdout:
$ eventfd -n <name> --watch
- write value to event FD
$ eventfd -n <name> --echo <value>
It would be also nice to support --fd <eventfd> as an alternative to
"-n" for watch and echo operations. The file descriptor is visible
for example in /proc/#/fd, so you don't have to depend on env.
variables, etc.
See my very primitive implementation below, you can:
$ eventfd -n PINGFD
$ xterm &
$ ./a -n PINGFD --wait | while read val; do \
echo "Event value: $val"; \
done
and in xterm:
$ eventfd -n PINGFD --echo 123
and in the original session you will see
Event value: 123
This all is possible without fork() and system().
> +{
> + struct pollfd fds = { .fd = fd, .events = POLLIN };
> +
> + while (poll(&fds, 1, -1) > 0) {
> + uint64_t efdval;
> + size_t len;
> + char s[BUFSIZ];
> +
> + len = read(fd, &efdval, sizeof efdval);
> + if (len != sizeof(uint64_t)) {
> + perror(_("short read from eventfd"));
> + exit(EXIT_FAILURE);
> + }
> +
> + snprintf(s, BUFSIZ, "%lu", efdval);
%"PRIu64"
....
> + fprintf(stderr, _("eventfd poll failed\n"));
> +}
please, use err() and warn() for error/rarning messages.
> +static int writer(char *varname, uint64_t echoval)
> +{
> + char *fdstr = getenv(varname);
> + int fd;
> + ssize_t len;
> +
> + if (!fdstr) {
> + fprintf(stderr, _("$%s is empty\n"), varname);
> + return EXIT_FAILURE;
> + }
> +
> + fd = atoi(fdstr);
I think we should be more paranoid with strings from env.variables or
command line -- what about strtoll()?
> + len = write(fd, &echoval, 8);
> + if (len == -1) {
> + perror(_("write failed"));
> + return EXIT_FAILURE;
> + }
> +
> + return EXIT_SUCCESS;
> +}
> +
> +static const struct option options[] = {
> + { "nofork", 0, 0, 'F' },
> + { "echo", 1, 0, 'e' },
> + { "varname", 1, 0, 'n' },
> + { "watchcmd", 1, 0, 'w' },
> + { "init", 1, 0, 'i' },
> + { "help", 0, 0, 'h' },
> + { NULL, 0, 0, 0 },
> +};
> +
> +static const char *optdescs[] = {
> + "exec the main process without forking",
> + "echo a number into eventfd",
> + "eventfd variable name",
> + "command to launch upon signal arrival",
> + "initial value of the eventfd counter",
> + "print help message"
> +};
> +
> +static void usage(int exitcode)
> +{
> + int i;
> +
> + printf(_("usage: %s [OPTIONS] <program> ...\n\n"
> + "Create, write and poll eventfds.\nOPTIONS:\n"),
> + program_invocation_short_name);
> + for (i = 0; options[i].name; i++)
> + printf(" -%c, --%-10s %s\n", options[i].val, options[i].name,
> + _(optdescs[i]));
>
Interesting idea :-) ... but please, use normal usage() with strings rather
than optdescs[] array. See for example sys-utils/fallocate.c.
The output from usage() should be to stderr in case of error
(EXIT_FAILURE).
> + exit(exitcode);
> +}
> +
> +static const char optstr[] = "Fe:n:w:i:h";
> +
> +int main(int argc, char *argv[])
> +{
> + uint64_t echoval = 0;
> + unsigned int initval = 0;
> + int fd, c, loptidx, nofork = 0;
> + char *varname = "EVENTFD";
> + char val[BUFSIZ];
> + char **nargv;
> + char *watchcmd = NULL;
> +
> + for (;;) {
> + c = getopt_long(argc, argv, optstr, options, NULL);
> + if (c == -1)
> + break;
> +
> + switch (c) {
> + case 'e':
> + echoval = strtoeventfd(optarg);
> + break;
> + case 'i':
> + initval = strtoul(optarg, NULL, 10);
> + break;
> + case 'n':
> + varname = strdup(optarg);
> + break;
> + case 'F':
> + nofork++;
> + break;
> + case 'w':
> + watchcmd = strdup(optarg);
if (!watchcmd)
err(EXIT_FAILURE, _("out of memory"));
...
Karel
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <poll.h>
#include <sys/eventfd.h>
#include <inttypes.h>
int main(int argc, char *argv[])
{
int idx = 1;
int fd = -1;
uint64_t val;
char *name = NULL;
if (strcmp(argv[idx], "-n") == 0) {
char *x = getenv((name = argv[++idx]));
if (x)
fd = atoi(x);
++idx;
}
if (strcmp(argv[idx], "--echo") == 0) {
/* write to event FD */
val = strtoumax(argv[++idx], NULL, 10);
write(fd, &val, 8);
} else if (strcmp(argv[idx], "--watch") == 0) {
/* read from event FD */
struct pollfd fds = { .fd = fd, .events = POLLIN };
while (poll(&fds, 1, -1) > 0) {
read(fd, &val, sizeof(val));
printf("%ju\n", val);
fflush(stdout);
}
} else if (fd == -1 && name) {
/* setup a new event FD and start shell */
char *nargv[] = { "/bin/bash", NULL };
char buf[BUFSIZ];
fd = eventfd(0, 0);
sprintf(buf, "%d", fd);
setenv(name, buf, 1);
return execv(nargv[0], nargv);
}
return EXIT_FAILURE;
}
--
To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
