sys/netinet/tcp_output.c

16  *    may be used to endorse or promote products derived from this software
138 	"Modify threshold for auto send buffer growth to account for SO_SNDLOWAT");
156     struct tcpopt *to, uint32_t len, int tso)
163 		hhook_data.to = to;
211 	struct tcpopt to;
253 	 * to send, then transmit; otherwise, investigate further.
269 	 * snd_nxt.  There may be SACK information that allows us to avoid
284 	 * to send out new data (when sendalot is 1), bypass this function.
290 	 * Still in sack recovery , reset rxmit flag to zero.
306 			 * snd_recover. Check to see if we can rexmit data
329 			KASSERT(off >= 0,("%s: sack block to the left of una : %d",
350 	 * and go to transmit state.
355 			 * If we still have some data to send, then
367 			 * to send then the probe will be the FIN
385 	 * to snd_max after the fast-retransmit.
388 	 * be set to snd_una, the offset will be 0, and the length may
411 			 * isn't really necessary), to work around a gcc
412 			 * optimization issue - to force gcc to compute
446 	 * Be careful not to send data and/or FIN on SYN segments.
447 	 * This measure is needed to prevent interoperability problems
482 		 * but we haven't been called to retransmit,
484 		 * after we sent into it.  If window shrank to 0,
486 		 * to (closed) window, and set the persist timer
490 		 * We also do a general check here to ensure that
492 		 * to send, but a 0-byte window. This makes sure
521 	 * makes it impossible to transmit any options which vary per generated
526 	 * the right thing below to provide length of just ip options and thus
527 	 * checking for ipoptlen is enough to decide if ip options are present.
585 	 *	- we need to retransmit
607 		 * us to flush a buffer queued with moretocome.  XXX
630 	 * Window updates are important when we close our window due to a
633 	 * remote end starts to send again the ACK clock takes over and
638 	 * update to be sent.  We also should avoid sending a flurry of
647 	 * In all other situations the ACK's to new incoming data will
708 	 * and we have not yet done so, then we need to send.
714 	 * In SACK, it is possible for tcp_output to fail to send a segment
727 	 * using ``persist'' packets is used to insure receipt of window
730 	 *	persisting		to move a small or zero window
736 	 *	is set when we are called to send a persist packet.
741 	 * If send window is too small, there is data to transmit, and no
742 	 * retransmit or persist is pending, then go to persist state.
754 	 * No reason to send a segment, just return.
770 	 * unless TCP set not to do any options.
790 	 * We only have to care about SYN and established connection
794 	to.to_flags = 0;
798 			to.to_mss = tcp_mssopt(&inp->inp_inc);
800 				to.to_mss -= V_tcp_udp_tunneling_overhead;
801 			to.to_flags |= TOF_MSS;
807 			 * have caused the original SYN or SYN|ACK to have
813 					to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN;
814 					to.to_tfo_cookie =
816 					to.to_flags |= TOF_FASTOPEN;
819 					to.to_tfo_len =
821 					to.to_tfo_cookie =
823 					to.to_flags |= TOF_FASTOPEN;
826 					 * If we wind up having more data to
838 			to.to_wscale = tp->request_r_scale;
839 			to.to_flags |= TOF_SCALE;
845 			to.to_tsval = curticks + tp->ts_offset;
846 			to.to_tsecr = tp->ts_recent;
847 			to.to_flags |= TOF_TS;
860 				to.to_flags |= TOF_SACKPERM;
863 				to.to_flags |= TOF_SACK;
864 				to.to_nsacks = tp->rcv_numsacks;
865 				to.to_sacks = (u_char *)tp->sackblks;
871 		 * Check that TCP_MD5SIG is enabled in tcpcb to
872 		 * account the size needed to set this TCP option.
875 			to.to_flags |= TOF_SIGNATURE;
879 		hdrlen += optlen = tcp_addoptions(&to, opt);
881 		 * If we wanted a TFO option to be added, but it was unable
882 		 * to fit, ensure no data is sent.
885 		    !(to.to_flags & TOF_FASTOPEN))
916 			 * Limit a TSO burst to prevent it from
968 			 * to clear the FIN flag on all but the
976 				 * Since we don't have enough space to put
1009 	 * This KASSERT is here to catch edge cases at a well defined place.
1015 	 * Grab a header mbuf, attaching a copy of data to
1076 		 * to the offset in the socket buffer chain.
1096 				 * shorten it to no longer need tso. Lets
1114 		 * give data to the user when a buffer fills or
1185 	 * If resending a FIN, be sure not to use a new sequence number.
1281 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1296 	 * a 0 window.  This may cause the remote transmitter to stall.  This
1297 	 * flag tells soreceive() to disable delayed acknowledgements when
1299 	 * to read more data than can be buffered prior to transmitting on
1312 		 * If no urgent pointer to send, then we pull
1313 		 * the urgent pointer to the left edge of the send window
1326 	if (to.to_flags & TOF_SIGNATURE) {
1334 		    (u_char *)(th + 1) + (to.to_signature - opt))) != 0) {
1347 		 * There is no need to fill in ip6_plen right now.
1408 	hhook_run_tcp_est_out(tp, th, &to, len, tso);
1413 	/* We're getting ready to send; log now. */
1424 	 * Fill in IP length and desired time to live and
1425 	 * send to IP level.  There should be a better way
1426 	 * to handle ttl and tos; we could keep them in
1427 	 * the template, but need a way to checksum without them.
1437 		 * user might want to change the value via setsockopt.
1445 		 * ip6_output() will set it properly; it's supposed to include
1486 	 * This might not be the best thing to do according to RFC3390
1490 	 * NB: Don't set DF on small MTU/MSS to have a safe fallback.
1555 			 * added new data to an empty socket buffer.
1604 			 * 1) A -> B: packet with enough data to fill the window
1610 			 * because it chose to send a packet. Unless tcp_output
1653 		 * due to a firewall reconfiguration cycle.  So we treat
1691 			 * to send segments changed to another or lowered
1771  * Insert TCP options according to the supplied parameters to the place
1783  * At minimum we need 10 bytes (to generate 1 SACK block).  If both
1788 tcp_addoptions(struct tcpopt *to, u_char *optp)
1793 		if ((to->to_flags & mask) != mask)
1797 		switch (to->to_flags & mask) {
1808 			to->to_mss = htons(to->to_mss);
1809 			bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss));
1810 			optp += sizeof(to->to_mss);
1822 			*optp++ = to->to_wscale;
1845 			to->to_tsval = htonl(to->to_tsval);
1846 			to->to_tsecr = htonl(to->to_tsecr);
1847 			bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval));
1848 			optp += sizeof(to->to_tsval);
1849 			bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr));
1850 			optp += sizeof(to->to_tsecr);
1861 				to->to_flags &= ~TOF_SIGNATURE;
1867 			to->to_signature = optp;
1875 			struct sackblk *sack = (struct sackblk *)to->to_sacks;
1886 			sackblks = min(to->to_nsacks,
1906 			/* XXX is there any point to aligning this option? */
1907 			total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len;
1909 				to->to_flags &= ~TOF_FASTOPEN;
1914 			if (to->to_tfo_len > 0) {
1915 				bcopy(to->to_tfo_cookie, optp, to->to_tfo_len);
1916 				optp += to->to_tfo_len;
1927 	/* Terminate and pad TCP options to a 4 byte boundary. */
1933 	 * According to RFC 793 (STD0007):
2032 			 * It is cheaper to just add the segments
2033 			 * than it is to take the cache miss to look
2122 	 * size is not optimally adjusted to the actual network conditions
2129 	 * The criteria to step up the send buffer one notch are:
2132 	 *  2. send buffer is filled to 7/8th with data (so we actually
2133 	 *     have data to make use of it);
2145 	 * what is strictly necessary to make full use of a given
2147 	 * to be much of an problem.  At worst we are trading wasting
2152 	 * with congestion window.  Requires another timer.  Has to