msgrcv issue

[Michal Rokos]

Hello,

I've noticed following issue: having two or more processes that receive IPC messages via msgrcv() call are getting stuck when one process requesting msgtyp > 0 and other one(s) have msgtyp = 0.

Have you noticed something like this?

> uname -mnrsv
HP-UX ktest B.11.31 U ia64
> model
ia64 hp server rx7640

[Michal Rokos]

Attaching example: when sending message (via msg_s) on 15:54:48 and 15:55:00, messages are delivered to msg_r process on 15:55:00. So the process msg_r (with mtype=0) had been sleeping and was not woke up by OS on 15:54:48.

# ./msg_r
creating msgq...
Mon Apr 12 15:54:40 2010
msgq_id = 20971608, mtype = 999

# ./msg_r 20971608
reusing msgq [20971608]
Mon Apr 12 15:54:45 2010
msgq_id = 20971608, mtype = 0
Mon Apr 12 15:55:00 2010
received 1 [Hello]
Mon Apr 12 15:55:00 2010
msgq_id = 20971608, mtype = 0
Mon Apr 12 15:55:00 2010
received 1 [Hello]
Mon Apr 12 15:55:00 2010
msgq_id = 20971608, mtype = 0

# ./msg_s 20971608 1 'Hello'
Mon Apr 12 15:54:48 2010
msgq_id = 20971608, mtype = 1, mtext [Hello]
# ./msg_s 20971608 1 'Hello'
Mon Apr 12 15:55:00 2010
msgq_id = 20971608, mtype = 1, mtext [Hello]

[Michal Rokos]

> cat msg_r.c
#include
#include
#include
#include

typedef struct
{
long mtype;
char mtext[100];
} msgx_ent;

int main( int argc, char *argv[] )
{
int msgq_id;
time_t t;
msgx_ent msgx_buf;
int msg_len;

setvbuf(stdout, NULL, _IONBF, 0);

if ( argc > 1 )
{
printf("reusing msgq [%s]\n", argv[1]);
msgq_id = atoi(argv[1]);
}
else
{
printf("creating msgq...\n");
msgq_id = msgget(IPC_PRIVATE, 0664);
}

again:
t = time(NULL);
printf("%s\tmsgq_id = %d, mtype = %d\n",
ctime(&t), msgq_id, argc > 1 ? 0 : 999);

msg_len = msgrcv(msgq_id, &msgx_buf, 100, argc > 1 ? 0 : 999, 0);

t = time(NULL);
if ( msg_len < 0 )
{
printf("%s\tmsg_len = %d, errno = %d\n",
ctime(&t), msg_len, errno);
}
else
{
printf("%s\treceived %d [%s]\n",
ctime(&t), msgx_buf.mtype, msgx_buf.mtext);
}
goto again;

return 0;
}

[Michal Rokos]

> cat msg_s.c
#include
#include
#include
#include

typedef struct
{
long mtype;
char mtext[100];
} msgx_ent;

int main( int argc, char *argv[] )
{
int msgq_id;
time_t t;
msgx_ent msgx_buf;
int msg_len;

setvbuf(stdout, NULL, _IONBF, 0);

msgq_id = atoi(argv[1]);
msgx_buf.mtype = atoi(argv[2]);
strcpy(msgx_buf.mtext, argv[3]);

t = time(NULL);
printf("%s\tmsgq_id = %d, mtype = %ld, mtext [%s]\n",
ctime(&t), msgq_id, msgx_buf.mtype, msgx_buf.mtext);

msg_len = msgsnd(msgq_id, &msgx_buf, strlen(msgx_buf.mtext)+1, 0);

if ( msg_len < 0 )
{
printf("msg_len = %d, errno = %d\n", msg_len, errno);
}

return 0;
}

[Michal Rokos]

More investigation:
- issue persist on ELF-64 and ELF-32 binaries
- issue vanish when I send message with msgtype == 999 (the one that is specified in msgrcv() call)

Any idea what's wrong?

[James R. Ferguson]

Hi:

You might use 'tusc' to see some details. You don't offer any code, in particular the setting of 'msgflg' and any values of 'errno'.

Regards!

...JRF...

[James R. Ferguson · Accepted solution]

Hi (again):

You might look at patch PHKL_40306 or its successor PHKL_40367.

Regards!

...JRF...

[Michal Rokos]

Thank you for your help.
Indeed, this patch is not applied. I've just asked admin for installation. In case it solves the issue (or not), I will update the status here.

Thank you! Michal

[Michal Rokos]

I can confirm that patch, you've suggested, fixed the issue we had. Thank you very much for your help. Michal

[James R. Ferguson]

Hi (again):

> I can confirm that patch, you've suggested, fixed the issue we had. Thank you very much for your help. Michal

If you are happy with the answers you received, please read this about assigning points:

http://forums13.itrc.hp.com/service/forums/helptips.do?#28

Regards!

...JRF...