Chronological Thread 

< Chronological >      < Thread >

---

Dear Xu Yun,

maybe this flag was by accident set to MSG_PEEK, which enables you to read the packet but
it remains in the socket layer.

Best regards
Michael

On Aug 10, 2005, at 4:39 AM, Xu Yun,SCNB R&D NNC(BJ) wrote:

Dear Michael,

I found the problem. I made a mistake not to initialize the argument of flags of sctp_recvmsg(). I thought it's an 'out' argument. After I assigned the flags to 0 and then call sctp_recvmsg, everything runs perfect. Thank you very much for your help!

Best Regards,
Xu Yun

-----Original Message-----
From: Michael Tuexen
[mailto:Michael.Tuexen AT micmac.franken.de]
Sent: Tuesday, August 09, 2005 9:21 PM
To: Thomas Dreibholz
Cc: discussion; Xu Yun,SCNB R&D NNC(BJ)
Subject: Re: [SCTP-Discussion] The difference between ext_sendto() and sct p_sendx()

Hi Thomas,

could you check if there is a problem with sctp_recvmsg in the
socketapilib? Could it be that
it always peeks?

Thank you very much.

Best regards
Michael

On Aug 9, 2005, at 13:40 Uhr, Xu Yun,SCNB R&D NNC(BJ) wrote:

Hello Michael,

Thanks for your hints. I've tried sctp_sendmsg and it runs as we
expected. But the receiving problem is still there: if I calls
ext_recvfrom to receive message everything runs well, but when I turns
to call sctp_recvmsg I got endless received messages.
After checking the source code of ext_recvfrom and sctp_recvmsg, the
only major difference between the two function is the argument
'receiveNotifications' of ext_recvmsg2() function. What's the
difference behind this argument?

BTW: In my last mail, I cut some unrelated code (including some lines
about timeout operation) when I pasted it, and it may make you
confused :) So the timeout should be the problem. I paste the
waitfor_message() again:

void* waitfor_message(void *not_used)
{
int ret;
struct timeval timeout;
int idx;
fd_set rfds;
m3ua_mgmt_ntfy_t ntfy;
m3ua_user_ntfy_t u_ntfy;
unsigned char msg[1024];
unsigned int msglen = 0;

timeout.tv_sec = 0;
timeout.tv_usec = 0;
while(1) {
if (0 == timeout.tv_sec && 0 == timeout.tv_usec) {
timeout.tv_sec = 1;
timeout.tv_usec = 0;
}
rfds = readfds;
ret = ext_select(FD_SETSIZE, &rfds, NULL, NULL, &timeout);
if (0 == ret) {
timeout.tv_sec = 0;
timeout.tv_usec = 0;
m3timer_ckexpiry(NULL);
um3_timer_ckexpiry(NULL);
}
else if (0 < ret) {
for (idx = 0; idx < MAX_LOCAL_EP; idx++)
if (1 == from[idx].e_state &&
FD_ISSET(from[idx].sockid,&rfds))
get_message(idx, from[idx].sockid, msg, &msglen);
}
else {
printf("ERRNO:%d\n", errno);
perror("E_select_Fail");
}
while (-1 != m3ua_mgmt_ntfy(&ntfy)) {
um3_process_mgmt_ntfy(&ntfy);
}
while (-1 != m3ua_user_ntfy(&u_ntfy)) {
um3_process_user_ntfy(&u_ntfy);
}
}
return NULL;
}

Best Regards,
Xu Yun


-----Original Message-----
From: Michael Tuexen
[mailto:Michael.Tuexen AT micmac.franken.de]
Sent: Tuesday, August 09, 2005 5:30 PM
To: Xu Yun,SCNB R&D NNC(BJ)
Cc:
discussion AT sctp.de
Subject: Re: [SCTP-Discussion] The difference between ext_sendto() and
sctp_sendx()

Hi,

see my comments in-line.

Best regards
Michael

On Aug 9, 2005, at 8:47 Uhr, Xu Yun,SCNB R&D NNC(BJ) wrote:

Dear all,

I'm a very new user of this sctp stack. These days I'm trying to
integrate this stack to a m3ua stack (also open source, from
sourceforge). Everything looks well when I choose ext_sendto() to send
and ext_recvfrom() to receive m3ua messages except the ppid is always
'0'. The sctp

That is within the specification. If you want to specify the PPID you
should use the sctp_sendmsg call.

payload protocol id of M3ua should be '3'. So I turned to use
sctp_sendx() and sctp_recvmsg() providing ppid by cmsg, and I met a
serious problem: after receiving and sending several m3ua messages,
sctp layer starts reporting to receive a message endlessly even after
I killed the peer. Could somebody give me some hints? The following is
main part of receiving and sending:

int m3ua_sendmsg(m3_u32 assoc_id,
m3_u32 stream_id,
m3_u32 msglen,
m3_u8 *pmsg)
{
int ret;
int tid = assoc_table[assoc_id].to;
int fid = assoc_table[assoc_id].from;
struct sctp_sndrcvinfo info;

....

info.sinfo_ppid=3;
info.sinfo_stream=0;
info.sinfo_assoc_id=0;
info.sinfo_timetolive=60000;
info.sinfo_flags=MSG_PR_SCTP_TTL;

ret = sctp_sendx(from[fid].sockid,pmsg, msglen,(struct
sockaddr*)&to[tid].addr,1,&info,0);

Try sctp_sendmsg instead of sctp_sendx

if (-1 == ret) {
printf("ERRNO:%d\n", errno);
perror("E_sendto_Fail");
}
return 0;
}

void* waitfor_message(void *not_used)
{
int ret;
struct timeval timeout;
int idx;
fd_set rfds;
unsigned char msg[1024];
unsigned int msglen = 0;

timeout.tv_sec = 1;
timeout.tv_usec = 0;
while(1) {
rfds = readfds;
ret = ext_select(FD_SETSIZE, &rfds, NULL, NULL, &timeout);
if (0 == ret) {
timeout.tv_sec = 0;

Not timeout.tv_sec = 1?

timeout.tv_usec = 0;
m3timer_ckexpiry(NULL);
um3_timer_ckexpiry(NULL);
}
else if (0 < ret) {
for (idx = 0; idx < MAX_LOCAL_EP; idx++)
if (1 == from[idx].e_state &&
FD_ISSET(from[idx].sockid,&rfds))
get_message(idx, from[idx].sockid, msg, &msglen);
}
else {
perror("E_select_Fail");
}
...
...
}
return NULL;
}

After the first timeout of 1 second you will only wait for 0 seconds,
which means that
the ext_select call returns immediately. I do not think that this is
what you want.

int get_message( int fid,
int sockid,
unsigned char msg[1024],
unsigned int *pmsglen)
{
...
struct sockaddr_in addr;
int from_len = sizeof(struct sockaddr);
struct sctp_sndrcvinfo info;
int flags;

memset(&addr, 0, sizeof(struct sockaddr));

info.sinfo_ppid=3;
info.sinfo_stream=0;
info.sinfo_assoc_id=0;
info.sinfo_timetolive=60000;
info.sinfo_flags=MSG_PR_SCTP_TTL;

You do not have to set ppid and so on at the receiving side. The
structure will
be filled in the the SCTP implementation for the received user data.

*pmsglen=sctp_recvmsg(sockid,msg,1024,(struct
sockaddr*)&addr,&from_len,&info,&flags);
if (0 < *pmsglen) {
/* search for association for which data is received */
...
}
return 0;
}


Thank you in advance!

Best Regards,
Xu Yun
_______________________________________________
discussion mailing list
discussion AT sctp.de
http://www.sctp.de/mailman/listinfo/discussion

---

• RE: [SCTP-Discussion] The difference between ext_sendto() and sct p_sendx(), Xu Yun,SCNB R&D NNC(BJ), 08/10/2005
  • Re: [SCTP-Discussion] The difference between ext_sendto() and sct p_sendx(), Michael Tuexen, 08/10/2005

---

Archive powered by MHonArc 2.6.19+.