[LTP] [PATCH 1/2] syscalls/msgstress01: Fix the stop logic
Cyril Hrubis
chrubis@suse.cz
Thu May 23 15:31:31 CEST 2024
The stop flag didn't really work because:
- if queue is full msgsnd() will block in the kernel
- if queue is empty msgrcv() will block in the kernel
And if the other process from the reader-writer pair exits the queue
will be never changed and the test will get stuck and killed by watchdog
timer.
What we need to do is to use IPC_NOWAIT, retry manually (after short
usleep) and handle errors manually as well. In that case no processes
will sleep in kernel and setting the stop flag will actually stop the
test.
Signed-off-by: Cyril Hrubis <chrubis@suse.cz>
---
.../syscalls/ipc/msgstress/msgstress01.c | 23 ++++++++++++++++---
1 file changed, 20 insertions(+), 3 deletions(-)
diff --git a/testcases/kernel/syscalls/ipc/msgstress/msgstress01.c b/testcases/kernel/syscalls/ipc/msgstress/msgstress01.c
index 5c84957b3..62ffcf63b 100644
--- a/testcases/kernel/syscalls/ipc/msgstress/msgstress01.c
+++ b/testcases/kernel/syscalls/ipc/msgstress/msgstress01.c
@@ -101,8 +101,17 @@ static void writer(const int id, const int pos)
struct sysv_data *buff = &ipc_data[pos];
int iter = num_iterations;
- while (--iter >= 0 && !(*stop))
- SAFE_MSGSND(id, &buff->msg, 100, 0);
+ while (--iter >= 0 && !(*stop)) {
+ int size = msgsnd(id, &buff->msg, 100, IPC_NOWAIT);
+ if (size < 0) {
+ if (errno == EAGAIN) {
+ usleep(10);
+ continue;
+ }
+
+ tst_brk(TBROK | TERRNO, "msgsnd() failed");
+ }
+ }
}
static void reader(const int id, const int pos)
@@ -115,7 +124,15 @@ static void reader(const int id, const int pos)
while (--iter >= 0 && !(*stop)) {
memset(&msg_recv, 0, sizeof(struct sysv_msg));
- size = SAFE_MSGRCV(id, &msg_recv, 100, MSGTYPE, 0);
+ size = msgrcv(id, &msg_recv, 100, MSGTYPE, IPC_NOWAIT);
+ if (size < 0) {
+ if (errno == ENOMSG) {
+ usleep(10);
+ continue;
+ }
+
+ tst_brk(TBROK | TERRNO, "msgrcv() failed");
+ }
if (msg_recv.type != buff->msg.type) {
tst_res(TFAIL, "Received the wrong message type");
--
2.43.2
More information about the ltp
mailing list