On Sat, Jul 27, 2019 at 11:24 AM Yann Droneaud ydroneaud@opteya.com wrote:
Hi,
Hi Yann, So sorry I missed your email. My email filters were messed up and I fixed them now but missed a bunch of emails :(
Le vendredi 26 juillet 2019 à 09:22 -0700, Suren Baghdasaryan a écrit :
This adds testing for polling on pidfd of a process being killed. Test runs 10000 iterations by default to stress test pidfd polling functionality. It accepts an optional command-line parameter to override the number or iterations to run. Specifically, it tests for:
- pidfd_open on a child process succeeds
- pidfd_send_signal on a child process succeeds
- polling on pidfd succeeds and returns exactly one event
- returned event is POLLIN
- event is received within 3 secs of the process being killed
10000 iterations was chosen because of the race condition being tested which is not consistently reproducible but usually is revealed after less than 2000 iterations. Reveals race fixed by commit b191d6491be6 ("pidfd: fix a poll race when setting exit_state")
Signed-off-by: Suren Baghdasaryan surenb@google.com
tools/testing/selftests/pidfd/.gitignore | 1 + tools/testing/selftests/pidfd/Makefile | 2 +- .../testing/selftests/pidfd/pidfd_poll_test.c | 117 ++++++++++++++++++ 3 files changed, 119 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/pidfd/pidfd_poll_test.c
diff --git a/tools/testing/selftests/pidfd/.gitignore b/tools/testing/selftests/pidfd/.gitignore index 16d84d117bc0..a67896347d34 100644 --- a/tools/testing/selftests/pidfd/.gitignore +++ b/tools/testing/selftests/pidfd/.gitignore @@ -1,2 +1,3 @@ pidfd_open_test +pidfd_poll_test pidfd_test diff --git a/tools/testing/selftests/pidfd/Makefile b/tools/testing/selftests/pidfd/Makefile index 720b2d884b3c..ed58b7108d18 100644 --- a/tools/testing/selftests/pidfd/Makefile +++ b/tools/testing/selftests/pidfd/Makefile @@ -1,7 +1,7 @@ # SPDX-License-Identifier: GPL-2.0-only CFLAGS += -g -I../../../../usr/include/ -lpthread
-TEST_GEN_PROGS := pidfd_test pidfd_open_test +TEST_GEN_PROGS := pidfd_test pidfd_open_test pidfd_poll_test
include ../lib.mk
diff --git a/tools/testing/selftests/pidfd/pidfd_poll_test.c b/tools/testing/selftests/pidfd/pidfd_poll_test.c new file mode 100644 index 000000000000..f1b62b91e53e --- /dev/null +++ b/tools/testing/selftests/pidfd/pidfd_poll_test.c @@ -0,0 +1,117 @@ +// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE +#include <errno.h> +#include <linux/types.h> +#include <linux/wait.h> +#include <poll.h> +#include <signal.h> +#include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <syscall.h> +#include <sys/wait.h> +#include <unistd.h>
+#include "pidfd.h" +#include "../kselftest.h"
+static bool timeout;
+static void handle_alarm(int sig) +{
timeout = true;+}
Not needed if poll() timeout is used instead.
+int main(int argc, char **argv) +{
struct pollfd fds;int iter, nevents;int nr_iterations = 10000;fds.events = POLLIN;if (argc > 2)ksft_exit_fail_msg("Unexpected command line argument\n");if (argc == 2) {nr_iterations = atoi(argv[1]);if (nr_iterations <= 0)ksft_exit_fail_msg("invalid input parameter %s\n",argv[1]);}ksft_print_msg("running pidfd poll test for %d iterations\n",nr_iterations);for (iter = 0; iter < nr_iterations; iter++) {int pidfd;int child_pid = fork();if (child_pid < 0) {if (errno == EAGAIN) {iter--;continue;}ksft_exit_fail_msg("%s - failed to fork a child process\n",strerror(errno));}if (child_pid == 0) {/* Child process just sleeps for a min and exits */sleep(60);Instead of relying on a timer for no reason, I would use the following:
while (1) pause();
Yeah, not much difference I think. If you don't strongly object I'll keep it this way.
exit(EXIT_SUCCESS);}/* Parent kills the child and waits for its death */pidfd = sys_pidfd_open(child_pid, 0);if (pidfd < 0)ksft_exit_fail_msg("%s - pidfd_open failed\n",strerror(errno));/* Setup 3 sec alarm - plenty of time */if (signal(SIGALRM, handle_alarm) == SIG_ERR)ksft_exit_fail_msg("%s - signal failed\n",strerror(errno));alarm(3);Would the poll() timeout be more simpler to use than relying on SIGALRM: no need to setup signal, no need for handler, no need for timeout variable.
Unfortunately that would not work because after timeout is passed poll() checks for the condition one last time, it sees that the process is dead (the condition is satisfied) and it returns nevents==1 (event did happen) instead of nevents==0 (poll timed out). I did implement it this way originally and was surprised that it never timed out even when I saw a 3 sec delay but then figured it out after going through the kernel code.
/* Send SIGKILL to the child */if (sys_pidfd_send_signal(pidfd, SIGKILL, NULL, 0))ksft_exit_fail_msg("%s - pidfd_send_signal failed\n",strerror(errno));/* Wait for the death notification */fds.fd = pidfd;nevents = poll(&fds, 1, -1);With
nevents = poll(&fds, 1, 3000);
/* Check for error conditions */if (nevents < 0)ksft_exit_fail_msg("%s - poll failed\n",strerror(errno));And if (nevents == 0) ksft_exit_fail_msg( "death notification wait timeout\n");
if (nevents != 1)ksft_exit_fail_msg("unexpected poll result: %d\n",nevents);if (!(fds.revents & POLLIN))ksft_exit_fail_msg("unexpected event type received: 0x%x\n",fds.revents);if (timeout)ksft_exit_fail_msg("death notification wait timeout\n");close(pidfd);// Wait for child to prevent zombiesif (waitpid(child_pid, NULL, 0) < 0)ksft_exit_fail_msg("%s - waitpid failed\n",strerror(errno));I feel safer now you defeated the zombies army :) Thanks.
Thanks for pointing it out!
}ksft_test_result_pass("pidfd poll test: pass\n");return ksft_exit_pass();+}
Regards
-- Yann Droneaud OPTEYA
Thanks, Suren.
-- To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe@android.com.