> On 16. Jun 2019, at 19:56, malc <mlashley@xxxxxxxxx> wrote: > > I think you want 'man 3 sctp_sendmsg' [1] and one of the flags > supported there... namely SCTP_UNORDERED. SCTP_ADDR_OVER, SCTP_ABORT, > SCTP_EOF > See also [2] which I'm guessing is relevant to what you're /trying/ to achieve. Using MSG_NOSIGNAL is definitely not what should be usedin this context, but the first question is: If the user provides an invalid flag, does this result in sctp_sendmsg() to return -1 and errno to be set to EINVAL? The more relevant question is: Assume sendmsg() returns -1. Shouldn't then no message be sent out? This send call is atomic. So I would expect * either no error is indicated and the message is accepted by the kernel * or an error is indicated and the message is not accepted by the kernel. Do we agree on this? Does the linux implementation follows this rule? Best regards Michael > > Cheers, > malc. > > [1] https://linux.die.net/man/3/sctp_sendmsg > [2] https://sourceforge.net/p/lksctp/mailman/message/11557899/ > > > On Sun, Jun 16, 2019 at 2:40 PM Elmar Stellnberger <estellnb@xxxxxxxxxx> wrote: >> >> While sctp_send works well at me, sctp_sendmsg gives me an EINVAL as >> soon as I pass in some flags. Nonetheless it still sends the message in >> spite of returning the EINVAL. That looks like an error to me. Please >> investigate the issue. >> >> while( len > 0 && ( ( ret = sctp_sendmsg( sid, bufp, len, NULL, 0, 0, >> MSG_NOSIGNAL, (0), 1000, 0 ) ) >= 0 || ( ret==-1 && ( errno == EINTR || >> errno == ENOBUFS ) ) ) ) { // returns EINVAL if flags are passed >> if( ret >= 0 ) { bufp += ret; len -= ret; } else if( errno == >> ENOBUFS ) { struct timespec ts; ts.tv_sec=0; ts.tv_nsec=10000; >> nanosleep(&ts,NULL); } } >> if( ret == -1 ) perror ("error sending request to stream #(0)"); >> >> If you should be interested I can send the whole sctp test client >> program. >> >> >> >> >> >>
