inet/tcp/tcp_output.c

4  * The contents of this file are subject to the terms of the
75  * This controls how tiny a write must be before we try to copy it
292 		/* If tiny tx and room in txq tail, pullup to save mblks. */
349 	 * Note that tcp_mss has been adjusted to take into account the
352 	 * they cannot be included in tcp_mss.  Thus we need to calculate
353 	 * the actual segment length when we need to send a segment which
379 		 * complete yet. We want to queue the data for transmission
380 		 * after entering ESTABLISHED state (RFC793). A jump to
411 		 * be sent out.  Otherwise, we need to take into account
424 			 * like the window is shrunk.  Do a check here to
475 	 * the smaller of 1 MSS and global tcp_naglim_def (default to be
488 	 * is set, then we have to force TCP not to send partial segment
491 	 * later. When tcp_zero_win_probe is set, TCP needs to send out
492 	 * something to do zero window probe.
508 	/* Pretend that all we were trying to send really got sent */
525 		 * If new data was sent, need to update the notsack
566  * Initial STREAMS write side put() procedure for sockets. It tries to
608  * Call by tcp_wput() to handle misc non M_DATA messages.
624 	 * get reflected back (in response to our own FLUSHRW) and get
626 	 * because a ref would have added but we need to check the state
647 /* tcp_wput_flush is called by tcp_wput_nondata to handle M_FLUSH messages. */
662 		 * know it, may come to an end.
696 	 * TODO: you can't just flush these, you have to increase rwnd for one
709  * tcp_wput_iocdata is called by tcp_wput_nondata to handle all M_IOCDATA
730 		 * use the CONN_IOCTLREF_* macros to hold off tcp_close until
814  * tcp_wput_ioctl is called by tcp_wput_nondata() to handle all M_IOCTL
843 		 * Either sockmod is about to be popped and the socket
845 		 * is about to be pushed so we could no longer use read-
866 	 * conn_ioctlref to hold off tcp_close until we're done here.
884  * This routine is called by tcp_wput() to handle all TPI requests.
919 			/* Set length to zero for reclamation below */
1025 		 * We used to M_ERROR.  Sending TNOTSUPPORT gives the user
1026 		 * to recover.
1126 	 * If ZEROCOPY has turned off, try not to send any zero-copy message
1182 	 * Check to see if this connection has been idled for some
1183 	 * time and no ACK is expected.  If it is, we need to slow
1184 	 * start again to get back the connection's "self-clock" as
1215 	 * determine if anything to send (Nagle).
1227 		 * mss < xmit_hiwater so there is no need to worry
1288 	/* see if we need to allocate a mblk for the headers */
1368 	 * If we ran out of memory, we pretend to have sent the packet
1399 	 * Try to force urgent data out on the wire. Even if we have unsent
1424  * descriptor, we take this to mean: "close the stream state NOW, close the tcp
1431  * to the application. 4.4BSD only returns EINTR and EWOULDBLOCK
1432  * errors to the application (from tsleep()) and not errors
1448 	 * stick around waiting for tcp_close_output to run and can therefore
1526 		 * socket wants to get rid of the socket, for example, if the
1578 			 * Check if we need to detach or just close
1603 		 * Need to cancel those timers which will not be used when
1604 		 * TCP is detached.  This has to be done before the conn_wq
1605 		 * is set to NULL.
1620 		 * successfully canceled. In this case we need to restart it
1631 	/* Detach did not complete. Still need to remove q from stream. */
1671 	/* Signal tcp_close() to finish closing. */
1712 	 * Check here to avoid sending zero-copy message down to IP when
1713 	 * ZEROCOPY capability has turned off. We only need to deal with
1715 	 * Since we have tried to backoff the tcp_xmit_head when turning
1751 	 * In the off-chance that the eager received and responded to
1753 	 * the ixa_pktlen. It would be better to fix the SYN/accept
1754 	 * multithreading scheme to avoid this complexity.
1783 	 * enough available data to initiate LSO transmission in the for(){}
1809 			 * Check whether be able to to do LSO for the current
1853 			 * Ignore this if we are going to send a
1872 					 * We are not supposed to send
1890 		 * The reason to adjust len here is that we need to set flags
1919 		 * Branch off to tcp_xmit_mp() if any of the VALID bits is
1922 		 * that the FIN flag needs to be set.  Note for this case,
2052 		 * There are four reasons to allocate a new hdr mblk:
2114 					 * told to break up larger writes
2122 					 * rounded up to nearest 64-byte;
2156 				 * order to remain within the b_cont limit.
2195 			/* Append LSO information to the mp. */
2215 			 * Make sure to clean up LSO information. Wherever a
2281 		 * so we have to check that and unset it first.
2297 	 * We do not have a good algorithm to update ssthresh at this time.
2325 			 * If we are going to create a DCE we'd better have
2368 	/* If a text string is passed in with the request, pass it to strlog. */
2431 		/* Track what we sent to the peer */
2450  * when RST is in response to an unexpected inbound packet for which
2453  * IPSEC NOTE : Try to send the reply with the same protection as it came
2454  * in.  We have the ip_recv_attr_t which is reversed to form the ip_xmit_attr_t.
2483 	 * If connp != NULL we use conn_ixa to keep IP_NEXTHOP and other
2487 	 * We get a safe copy of conn_ixa so we don't need to restore anything
2547 		 * address to TCP.  Similar checks are done below for IPv6.
2569 			uint8_t *to;
2571 			to = mp->b_rptr + ip_hdr_len - IPV6_HDR_LEN;
2572 			ovbcopy(ip6h, to, IPV6_HDR_LEN);
2632 			 * it is being called in response to a SYN segment
2654 		 * Apply IPsec based on how IPsec was applied to
2678 	 * to keep a local trace buffer in tcp_record_trace().
2691  * Generate a "no listener here" RST in response to an "unknown" segment.
2692  * connp is set by caller when RST is in response to an unexpected
2694  * Note that we are reusing the incoming mp to construct the outgoing RST.
2747 		    char *, "Could not reply with RST to mp(1)",
2749 		ip2dbg(("tcp_xmit_listeners_reset: not permitted to reply\n"));
2817 	 * TCP/IP header according to RFC 793 as it means
2819 	 * to get around some broken middle boxes/end hosts
2820 	 * out there, we allow the option value to be the
2828 	 * the only case we need to take care of is in
2839 	/* Update the offset to cover the additional word */
2872 		 * Set up all the bits to tell other side
2884 		 * Reset the MSS option value to be SMSS
2889 		 * is better for us to be more cautious.
2908 		 * updated to include the timestamps option.
2969 	 * Get IP set to checksum on our behalf
3021  * tcp_xmit_mp is called to return a pointer to an mblk chain complete with
3022  * ip and tcp header ready to pass down to IP.  If the mp passed in is
3023  * non-NULL, then up to max_to_send bytes of data will be dup'ed off that
3027  * routine is not usually called to send new data for the first time.  It
3028  * is mostly called out of the timer for retransmits, and to generate ACKs.
3060 	 * Note that tcp_mss has been adjusted to take into account the
3063 	 * they cannot be included in tcp_mss.  Thus we need to calculate
3064 	 * the actual segment length when we need to send a segment which
3130 	 * Use tcp_unsent to determine if the PUSH bit should be used assumes
3132 	 * to retransmit data the setting of the PUSH bit may appear some
3161 	/* Check if there is any special processing needs to be done. */
3168 			/* Need to do connection set up processing. */
3172 			/* Need to do connection tear down processing. */
3177 		 * Need to do urgent pointer processing.
3262  * to a segment.  If it returns B_FALSE, TCP should not respond.
3270 	 * TCP needs to protect itself from generating too many RSTs.
3274 	 * What we do here is to have a limit of tcp_rst_sent_rate RSTs
3295  * based on tcp_pipe.  Then it goes thru the notsack list to find eligible
3298  * all eligible segments, it checks to see if TCP can send some new segments
3327 	 * Limit the num of outstanding data in the network to be
3359 		 * All holes are filled.  Manipulate tcp_cwnd to send more
3361 		 * set to tcp_cwnd_ssthresh.
3404 		 * Update the send timestamp to avoid false retransmission.
3412 		 * Update tcp_rexmit_max to extend this SACK recovery phase.
3415 		 * to extend SACK recover phase to avoid starting another
3425  * tcp_ss_rexmit() is called to do slow start retransmission after a timeout
3429  * to be sent in one time to tcp_snd_burst, the burst variable.
3480 			 * Update the send timestamp to avoid false
3493 		 * the rest of the job to tcp_wput_data().  But we
3494 		 * need to check the send window first.  If the
3501 	/* Only call tcp_wput_data() if there is data to be sent. */
3514 	 * All sent data has been acknowledged or no data left to send, just
3515 	 * to return.
3534  * tcp_get_seg_mp() is called to get the pointer to a segment in the
3536  * sequence number is equal to last valid sequence number (tcp_snxt), the
3537  * returned mblk is the last valid mblk, and off is set to the length of
3545  *	int32_t *off: after the execution, *off will be the offset to
3546  *		the returned mblk which points to the requested seq no.
3547  *		It is the caller's responsibility to send in a non-null off.
3550  *	A mblk_t pointer pointing to the requested segment in send queue.
3578  * This routine adjusts next-to-send sequence number variables, in the
3599  * if the receiver shrinks the window, i.e. moves the right window to the
3640  * tcp_fill_header is called by tcp_send() to fill the outgoing TCP header