xref: /freebsd/sys/netinet/sctp_cc_functions.c (revision c54a18d26b6afde430af32090857d70da0983532)
1 /*-
2  * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions are met:
6  *
7  * a) Redistributions of source code must retain the above copyright notice,
8  *   this list of conditions and the following disclaimer.
9  *
10  * b) Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in
12  *   the documentation and/or other materials provided with the distribution.
13  *
14  * c) Neither the name of Cisco Systems, Inc. nor the names of its
15  *    contributors may be used to endorse or promote products derived
16  *    from this software without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
20  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
22  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
28  * THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <netinet/sctp_os.h>
32 #include <netinet/sctp_var.h>
33 #include <netinet/sctp_sysctl.h>
34 #include <netinet/sctp_pcb.h>
35 #include <netinet/sctp_header.h>
36 #include <netinet/sctputil.h>
37 #include <netinet/sctp_output.h>
38 #include <netinet/sctp_input.h>
39 #include <netinet/sctp_indata.h>
40 #include <netinet/sctp_uio.h>
41 #include <netinet/sctp_timer.h>
42 #include <netinet/sctp_auth.h>
43 #include <netinet/sctp_asconf.h>
44 #include <netinet/sctp_cc_functions.h>
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 void
48 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
49 {
50 	/*
51 	 * We take the max of the burst limit times a MTU or the
52 	 * INITIAL_CWND. We then limit this to 4 MTU's of sending. cwnd must
53 	 * be at least 2 MTU.
54 	 */
55 	net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
56 	net->ssthresh = stcb->asoc.peers_rwnd;
57 
58 	if (sctp_logging_level & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) {
59 		sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
60 	}
61 }
62 
63 void
64 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb,
65     struct sctp_association *asoc)
66 {
67 	struct sctp_nets *net;
68 
69 	/*-
70 	 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) &&
71 	 * (net->fast_retran_loss_recovery == 0)))
72 	 */
73 	TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
74 		if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) {
75 			/* out of a RFC2582 Fast recovery window? */
76 			if (net->net_ack > 0) {
77 				/*
78 				 * per section 7.2.3, are there any
79 				 * destinations that had a fast retransmit
80 				 * to them. If so what we need to do is
81 				 * adjust ssthresh and cwnd.
82 				 */
83 				struct sctp_tmit_chunk *lchk;
84 				int old_cwnd = net->cwnd;
85 
86 				net->ssthresh = net->cwnd / 2;
87 				if (net->ssthresh < (net->mtu * 2)) {
88 					net->ssthresh = 2 * net->mtu;
89 				}
90 				net->cwnd = net->ssthresh;
91 				if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
92 					sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
93 					    SCTP_CWND_LOG_FROM_FR);
94 				}
95 				lchk = TAILQ_FIRST(&asoc->send_queue);
96 
97 				net->partial_bytes_acked = 0;
98 				/* Turn on fast recovery window */
99 				asoc->fast_retran_loss_recovery = 1;
100 				if (lchk == NULL) {
101 					/* Mark end of the window */
102 					asoc->fast_recovery_tsn = asoc->sending_seq - 1;
103 				} else {
104 					asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
105 				}
106 
107 				/*
108 				 * CMT fast recovery -- per destination
109 				 * recovery variable.
110 				 */
111 				net->fast_retran_loss_recovery = 1;
112 
113 				if (lchk == NULL) {
114 					/* Mark end of the window */
115 					net->fast_recovery_tsn = asoc->sending_seq - 1;
116 				} else {
117 					net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
118 				}
119 
120 				/*
121 				 * Disable Nonce Sum Checking and store the
122 				 * resync tsn
123 				 */
124 				asoc->nonce_sum_check = 0;
125 				asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1;
126 
127 				sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
128 				    stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32);
129 				sctp_timer_start(SCTP_TIMER_TYPE_SEND,
130 				    stcb->sctp_ep, stcb, net);
131 			}
132 		} else if (net->net_ack > 0) {
133 			/*
134 			 * Mark a peg that we WOULD have done a cwnd
135 			 * reduction but RFC2582 prevented this action.
136 			 */
137 			SCTP_STAT_INCR(sctps_fastretransinrtt);
138 		}
139 	}
140 }
141 
142 void
143 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb,
144     struct sctp_association *asoc,
145     int accum_moved, int reneged_all, int will_exit)
146 {
147 	struct sctp_nets *net;
148 
149 	/******************************/
150 	/* update cwnd and Early FR   */
151 	/******************************/
152 	TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
153 
154 #ifdef JANA_CMT_FAST_RECOVERY
155 		/*
156 		 * CMT fast recovery code. Need to debug.
157 		 */
158 		if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
159 			if (compare_with_wrap(asoc->last_acked_seq,
160 			    net->fast_recovery_tsn, MAX_TSN) ||
161 			    (asoc->last_acked_seq == net->fast_recovery_tsn) ||
162 			    compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) ||
163 			    (net->pseudo_cumack == net->fast_recovery_tsn)) {
164 				net->will_exit_fast_recovery = 1;
165 			}
166 		}
167 #endif
168 		if (sctp_early_fr) {
169 			/*
170 			 * So, first of all do we need to have a Early FR
171 			 * timer running?
172 			 */
173 			if (((TAILQ_FIRST(&asoc->sent_queue)) &&
174 			    (net->ref_count > 1) &&
175 			    (net->flight_size < net->cwnd)) ||
176 			    (reneged_all)) {
177 				/*
178 				 * yes, so in this case stop it if its
179 				 * running, and then restart it. Reneging
180 				 * all is a special case where we want to
181 				 * run the Early FR timer and then force the
182 				 * last few unacked to be sent, causing us
183 				 * to illicit a sack with gaps to force out
184 				 * the others.
185 				 */
186 				if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
187 					SCTP_STAT_INCR(sctps_earlyfrstpidsck2);
188 					sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
189 					    SCTP_FROM_SCTP_INDATA + SCTP_LOC_20);
190 				}
191 				SCTP_STAT_INCR(sctps_earlyfrstrid);
192 				sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net);
193 			} else {
194 				/* No, stop it if its running */
195 				if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
196 					SCTP_STAT_INCR(sctps_earlyfrstpidsck3);
197 					sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
198 					    SCTP_FROM_SCTP_INDATA + SCTP_LOC_21);
199 				}
200 			}
201 		}
202 		/* if nothing was acked on this destination skip it */
203 		if (net->net_ack == 0) {
204 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
205 				sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
206 			}
207 			continue;
208 		}
209 		if (net->net_ack2 > 0) {
210 			/*
211 			 * Karn's rule applies to clearing error count, this
212 			 * is optional.
213 			 */
214 			net->error_count = 0;
215 			if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) ==
216 			    SCTP_ADDR_NOT_REACHABLE) {
217 				/* addr came good */
218 				net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE;
219 				net->dest_state |= SCTP_ADDR_REACHABLE;
220 				sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb,
221 				    SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED);
222 				/* now was it the primary? if so restore */
223 				if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) {
224 					(void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net);
225 				}
226 			}
227 			/*
228 			 * JRS 5/14/07 - If CMT PF is on and the destination
229 			 * is in PF state, set the destination to active
230 			 * state and set the cwnd to one or two MTU's based
231 			 * on whether PF1 or PF2 is being used.
232 			 *
233 			 * Should we stop any running T3 timer here?
234 			 */
235 			if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) ==
236 			    SCTP_ADDR_PF)) {
237 				net->dest_state &= ~SCTP_ADDR_PF;
238 				net->cwnd = net->mtu * sctp_cmt_pf;
239 				SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
240 				    net, net->cwnd);
241 				/*
242 				 * Since the cwnd value is explicitly set,
243 				 * skip the code that updates the cwnd
244 				 * value.
245 				 */
246 				goto skip_cwnd_update;
247 			}
248 		}
249 #ifdef JANA_CMT_FAST_RECOVERY
250 		/*
251 		 * CMT fast recovery code
252 		 */
253 		/*
254 		 * if (sctp_cmt_on_off == 1 &&
255 		 * net->fast_retran_loss_recovery &&
256 		 * net->will_exit_fast_recovery == 0) { @@@ Do something }
257 		 * else if (sctp_cmt_on_off == 0 &&
258 		 * asoc->fast_retran_loss_recovery && will_exit == 0) {
259 		 */
260 #endif
261 
262 		if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) {
263 			/*
264 			 * If we are in loss recovery we skip any cwnd
265 			 * update
266 			 */
267 			goto skip_cwnd_update;
268 		}
269 		/*
270 		 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
271 		 * moved.
272 		 */
273 		if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) {
274 			/* If the cumulative ack moved we can proceed */
275 			if (net->cwnd <= net->ssthresh) {
276 				/* We are in slow start */
277 				if (net->flight_size + net->net_ack >= net->cwnd) {
278 					if (net->net_ack > (net->mtu * sctp_L2_abc_variable)) {
279 						net->cwnd += (net->mtu * sctp_L2_abc_variable);
280 						if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
281 							sctp_log_cwnd(stcb, net, net->mtu,
282 							    SCTP_CWND_LOG_FROM_SS);
283 						}
284 					} else {
285 						net->cwnd += net->net_ack;
286 						if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
287 							sctp_log_cwnd(stcb, net, net->net_ack,
288 							    SCTP_CWND_LOG_FROM_SS);
289 						}
290 					}
291 				} else {
292 					if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
293 						sctp_log_cwnd(stcb, net, net->net_ack,
294 						    SCTP_CWND_LOG_NOADV_SS);
295 					}
296 				}
297 			} else {
298 				/* We are in congestion avoidance */
299 				/*
300 				 * Add to pba
301 				 */
302 				net->partial_bytes_acked += net->net_ack;
303 
304 				if ((net->flight_size + net->net_ack >= net->cwnd) &&
305 				    (net->partial_bytes_acked >= net->cwnd)) {
306 					net->partial_bytes_acked -= net->cwnd;
307 					net->cwnd += net->mtu;
308 					if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
309 						sctp_log_cwnd(stcb, net, net->mtu,
310 						    SCTP_CWND_LOG_FROM_CA);
311 					}
312 				} else {
313 					if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
314 						sctp_log_cwnd(stcb, net, net->net_ack,
315 						    SCTP_CWND_LOG_NOADV_CA);
316 					}
317 				}
318 			}
319 		} else {
320 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
321 				sctp_log_cwnd(stcb, net, net->mtu,
322 				    SCTP_CWND_LOG_NO_CUMACK);
323 			}
324 		}
325 skip_cwnd_update:
326 		/*
327 		 * NOW, according to Karn's rule do we need to restore the
328 		 * RTO timer back? Check our net_ack2. If not set then we
329 		 * have a ambiguity.. i.e. all data ack'd was sent to more
330 		 * than one place.
331 		 */
332 		if (net->net_ack2) {
333 			/* restore any doubled timers */
334 			net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1;
335 			if (net->RTO < stcb->asoc.minrto) {
336 				net->RTO = stcb->asoc.minrto;
337 			}
338 			if (net->RTO > stcb->asoc.maxrto) {
339 				net->RTO = stcb->asoc.maxrto;
340 			}
341 		}
342 	}
343 }
344 
345 void
346 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net)
347 {
348 	int old_cwnd = net->cwnd;
349 
350 	net->ssthresh = max(net->cwnd / 2, 2 * net->mtu);
351 	net->cwnd = net->mtu;
352 	net->partial_bytes_acked = 0;
353 
354 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
355 		sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
356 	}
357 }
358 
359 void
360 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net)
361 {
362 	int old_cwnd = net->cwnd;
363 
364 	SCTP_STAT_INCR(sctps_ecnereducedcwnd);
365 	net->ssthresh = net->cwnd / 2;
366 	if (net->ssthresh < net->mtu) {
367 		net->ssthresh = net->mtu;
368 		/* here back off the timer as well, to slow us down */
369 		net->RTO <<= 1;
370 	}
371 	net->cwnd = net->ssthresh;
372 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
373 		sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
374 	}
375 }
376 
377 void
378 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb,
379     struct sctp_nets *net, struct sctp_pktdrop_chunk *cp,
380     uint32_t * bottle_bw, uint32_t * on_queue)
381 {
382 	uint32_t bw_avail;
383 	int rtt, incr;
384 	int old_cwnd = net->cwnd;
385 
386 	/* need real RTT for this calc */
387 	rtt = ((net->lastsa >> 2) + net->lastsv) >> 1;
388 	/* get bottle neck bw */
389 	*bottle_bw = ntohl(cp->bottle_bw);
390 	/* and whats on queue */
391 	*on_queue = ntohl(cp->current_onq);
392 	/*
393 	 * adjust the on-queue if our flight is more it could be that the
394 	 * router has not yet gotten data "in-flight" to it
395 	 */
396 	if (*on_queue < net->flight_size)
397 		*on_queue = net->flight_size;
398 	/* calculate the available space */
399 	bw_avail = (*bottle_bw * rtt) / 1000;
400 	if (bw_avail > *bottle_bw) {
401 		/*
402 		 * Cap the growth to no more than the bottle neck. This can
403 		 * happen as RTT slides up due to queues. It also means if
404 		 * you have more than a 1 second RTT with a empty queue you
405 		 * will be limited to the bottle_bw per second no matter if
406 		 * other points have 1/2 the RTT and you could get more
407 		 * out...
408 		 */
409 		bw_avail = *bottle_bw;
410 	}
411 	if (*on_queue > bw_avail) {
412 		/*
413 		 * No room for anything else don't allow anything else to be
414 		 * "added to the fire".
415 		 */
416 		int seg_inflight, seg_onqueue, my_portion;
417 
418 		net->partial_bytes_acked = 0;
419 
420 		/* how much are we over queue size? */
421 		incr = *on_queue - bw_avail;
422 		if (stcb->asoc.seen_a_sack_this_pkt) {
423 			/*
424 			 * undo any cwnd adjustment that the sack might have
425 			 * made
426 			 */
427 			net->cwnd = net->prev_cwnd;
428 		}
429 		/* Now how much of that is mine? */
430 		seg_inflight = net->flight_size / net->mtu;
431 		seg_onqueue = *on_queue / net->mtu;
432 		my_portion = (incr * seg_inflight) / seg_onqueue;
433 
434 		/* Have I made an adjustment already */
435 		if (net->cwnd > net->flight_size) {
436 			/*
437 			 * for this flight I made an adjustment we need to
438 			 * decrease the portion by a share our previous
439 			 * adjustment.
440 			 */
441 			int diff_adj;
442 
443 			diff_adj = net->cwnd - net->flight_size;
444 			if (diff_adj > my_portion)
445 				my_portion = 0;
446 			else
447 				my_portion -= diff_adj;
448 		}
449 		/*
450 		 * back down to the previous cwnd (assume we have had a sack
451 		 * before this packet). minus what ever portion of the
452 		 * overage is my fault.
453 		 */
454 		net->cwnd -= my_portion;
455 
456 		/* we will NOT back down more than 1 MTU */
457 		if (net->cwnd <= net->mtu) {
458 			net->cwnd = net->mtu;
459 		}
460 		/* force into CA */
461 		net->ssthresh = net->cwnd - 1;
462 	} else {
463 		/*
464 		 * Take 1/4 of the space left or max burst up .. whichever
465 		 * is less.
466 		 */
467 		incr = min((bw_avail - *on_queue) >> 2,
468 		    stcb->asoc.max_burst * net->mtu);
469 		net->cwnd += incr;
470 	}
471 	if (net->cwnd > bw_avail) {
472 		/* We can't exceed the pipe size */
473 		net->cwnd = bw_avail;
474 	}
475 	if (net->cwnd < net->mtu) {
476 		/* We always have 1 MTU */
477 		net->cwnd = net->mtu;
478 	}
479 	if (net->cwnd - old_cwnd != 0) {
480 		/* log only changes */
481 		if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
482 			sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
483 			    SCTP_CWND_LOG_FROM_SAT);
484 		}
485 	}
486 }
487 
488 void
489 sctp_cwnd_update_after_output(struct sctp_tcb *stcb,
490     struct sctp_nets *net, int burst_limit)
491 {
492 	int old_cwnd = net->cwnd;
493 
494 	if (net->ssthresh < net->cwnd)
495 		net->ssthresh = net->cwnd;
496 	net->cwnd = (net->flight_size + (burst_limit * net->mtu));
497 
498 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
499 		sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST);
500 	}
501 }
502 
503 void
504 sctp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp,
505     struct sctp_tcb *stcb, struct sctp_nets *net)
506 {
507 	int old_cwnd = net->cwnd;
508 
509 	sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED);
510 	/*
511 	 * make a small adjustment to cwnd and force to CA.
512 	 */
513 	if (net->cwnd > net->mtu)
514 		/* drop down one MTU after sending */
515 		net->cwnd -= net->mtu;
516 	if (net->cwnd < net->ssthresh)
517 		/* still in SS move to CA */
518 		net->ssthresh = net->cwnd - 1;
519 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
520 		sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR);
521 	}
522 }
523 
524 struct sctp_hs_raise_drop {
525 	int32_t cwnd;
526 	int32_t increase;
527 	int32_t drop_percent;
528 };
529 
530 #define SCTP_HS_TABLE_SIZE 73
531 
532 struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = {
533 	{38, 1, 50},		/* 0   */
534 	{118, 2, 44},		/* 1   */
535 	{221, 3, 41},		/* 2   */
536 	{347, 4, 38},		/* 3   */
537 	{495, 5, 37},		/* 4   */
538 	{663, 6, 35},		/* 5   */
539 	{851, 7, 34},		/* 6   */
540 	{1058, 8, 33},		/* 7   */
541 	{1284, 9, 32},		/* 8   */
542 	{1529, 10, 31},		/* 9   */
543 	{1793, 11, 30},		/* 10  */
544 	{2076, 12, 29},		/* 11  */
545 	{2378, 13, 28},		/* 12  */
546 	{2699, 14, 28},		/* 13  */
547 	{3039, 15, 27},		/* 14  */
548 	{3399, 16, 27},		/* 15  */
549 	{3778, 17, 26},		/* 16  */
550 	{4177, 18, 26},		/* 17  */
551 	{4596, 19, 25},		/* 18  */
552 	{5036, 20, 25},		/* 19  */
553 	{5497, 21, 24},		/* 20  */
554 	{5979, 22, 24},		/* 21  */
555 	{6483, 23, 23},		/* 22  */
556 	{7009, 24, 23},		/* 23  */
557 	{7558, 25, 22},		/* 24  */
558 	{8130, 26, 22},		/* 25  */
559 	{8726, 27, 22},		/* 26  */
560 	{9346, 28, 21},		/* 27  */
561 	{9991, 29, 21},		/* 28  */
562 	{10661, 30, 21},	/* 29  */
563 	{11358, 31, 20},	/* 30  */
564 	{12082, 32, 20},	/* 31  */
565 	{12834, 33, 20},	/* 32  */
566 	{13614, 34, 19},	/* 33  */
567 	{14424, 35, 19},	/* 34  */
568 	{15265, 36, 19},	/* 35  */
569 	{16137, 37, 19},	/* 36  */
570 	{17042, 38, 18},	/* 37  */
571 	{17981, 39, 18},	/* 38  */
572 	{18955, 40, 18},	/* 39  */
573 	{19965, 41, 17},	/* 40  */
574 	{21013, 42, 17},	/* 41  */
575 	{22101, 43, 17},	/* 42  */
576 	{23230, 44, 17},	/* 43  */
577 	{24402, 45, 16},	/* 44  */
578 	{25618, 46, 16},	/* 45  */
579 	{26881, 47, 16},	/* 46  */
580 	{28193, 48, 16},	/* 47  */
581 	{29557, 49, 15},	/* 48  */
582 	{30975, 50, 15},	/* 49  */
583 	{32450, 51, 15},	/* 50  */
584 	{33986, 52, 15},	/* 51  */
585 	{35586, 53, 14},	/* 52  */
586 	{37253, 54, 14},	/* 53  */
587 	{38992, 55, 14},	/* 54  */
588 	{40808, 56, 14},	/* 55  */
589 	{42707, 57, 13},	/* 56  */
590 	{44694, 58, 13},	/* 57  */
591 	{46776, 59, 13},	/* 58  */
592 	{48961, 60, 13},	/* 59  */
593 	{51258, 61, 13},	/* 60  */
594 	{53677, 62, 12},	/* 61  */
595 	{56230, 63, 12},	/* 62  */
596 	{58932, 64, 12},	/* 63  */
597 	{61799, 65, 12},	/* 64  */
598 	{64851, 66, 11},	/* 65  */
599 	{68113, 67, 11},	/* 66  */
600 	{71617, 68, 11},	/* 67  */
601 	{75401, 69, 10},	/* 68  */
602 	{79517, 70, 10},	/* 69  */
603 	{84035, 71, 10},	/* 70  */
604 	{89053, 72, 10},	/* 71  */
605 	{94717, 73, 9}		/* 72  */
606 };
607 
608 static void
609 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net)
610 {
611 	int cur_val, i, indx, incr;
612 
613 	cur_val = net->cwnd >> 10;
614 	indx = SCTP_HS_TABLE_SIZE - 1;
615 #ifdef SCTP_DEBUG
616 	printf("HS CC CAlled.\n");
617 #endif
618 	if (cur_val < sctp_cwnd_adjust[0].cwnd) {
619 		/* normal mode */
620 		if (net->net_ack > net->mtu) {
621 			net->cwnd += net->mtu;
622 			if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
623 				sctp_log_cwnd(stcb, net, net->mtu, SCTP_CWND_LOG_FROM_SS);
624 			}
625 		} else {
626 			net->cwnd += net->net_ack;
627 			if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
628 				sctp_log_cwnd(stcb, net, net->net_ack, SCTP_CWND_LOG_FROM_SS);
629 			}
630 		}
631 	} else {
632 		for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) {
633 			if (cur_val < sctp_cwnd_adjust[i].cwnd) {
634 				indx = i;
635 				break;
636 			}
637 		}
638 		net->last_hs_used = indx;
639 		incr = ((sctp_cwnd_adjust[indx].increase) << 10);
640 		net->cwnd += incr;
641 		if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
642 			sctp_log_cwnd(stcb, net, incr, SCTP_CWND_LOG_FROM_SS);
643 		}
644 	}
645 }
646 
647 static void
648 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net)
649 {
650 	int cur_val, i, indx;
651 	int old_cwnd = net->cwnd;
652 
653 	cur_val = net->cwnd >> 10;
654 	indx = net->last_hs_used;
655 	if (cur_val < sctp_cwnd_adjust[0].cwnd) {
656 		/* normal mode */
657 		net->ssthresh = net->cwnd / 2;
658 		if (net->ssthresh < (net->mtu * 2)) {
659 			net->ssthresh = 2 * net->mtu;
660 		}
661 		net->cwnd = net->ssthresh;
662 	} else {
663 		/* drop by the proper amount */
664 		net->ssthresh = net->cwnd - (int)((net->cwnd / 100) *
665 		    sctp_cwnd_adjust[net->last_hs_used].drop_percent);
666 		net->cwnd = net->ssthresh;
667 		/* now where are we */
668 		indx = net->last_hs_used;
669 		cur_val = net->cwnd >> 10;
670 		/* reset where we are in the table */
671 		if (cur_val < sctp_cwnd_adjust[0].cwnd) {
672 			/* feel out of hs */
673 			net->last_hs_used = 0;
674 		} else {
675 			for (i = indx; i >= 1; i--) {
676 				if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) {
677 					break;
678 				}
679 			}
680 			net->last_hs_used = indx;
681 		}
682 	}
683 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
684 		sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR);
685 	}
686 }
687 
688 void
689 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb,
690     struct sctp_association *asoc)
691 {
692 	struct sctp_nets *net;
693 
694 	/*
695 	 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) &&
696 	 * (net->fast_retran_loss_recovery == 0)))
697 	 */
698 	TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
699 		if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) {
700 			/* out of a RFC2582 Fast recovery window? */
701 			if (net->net_ack > 0) {
702 				/*
703 				 * per section 7.2.3, are there any
704 				 * destinations that had a fast retransmit
705 				 * to them. If so what we need to do is
706 				 * adjust ssthresh and cwnd.
707 				 */
708 				struct sctp_tmit_chunk *lchk;
709 
710 				sctp_hs_cwnd_decrease(stcb, net);
711 
712 				lchk = TAILQ_FIRST(&asoc->send_queue);
713 
714 				net->partial_bytes_acked = 0;
715 				/* Turn on fast recovery window */
716 				asoc->fast_retran_loss_recovery = 1;
717 				if (lchk == NULL) {
718 					/* Mark end of the window */
719 					asoc->fast_recovery_tsn = asoc->sending_seq - 1;
720 				} else {
721 					asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
722 				}
723 
724 				/*
725 				 * CMT fast recovery -- per destination
726 				 * recovery variable.
727 				 */
728 				net->fast_retran_loss_recovery = 1;
729 
730 				if (lchk == NULL) {
731 					/* Mark end of the window */
732 					net->fast_recovery_tsn = asoc->sending_seq - 1;
733 				} else {
734 					net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
735 				}
736 
737 				/*
738 				 * Disable Nonce Sum Checking and store the
739 				 * resync tsn
740 				 */
741 				asoc->nonce_sum_check = 0;
742 				asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1;
743 
744 				sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
745 				    stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32);
746 				sctp_timer_start(SCTP_TIMER_TYPE_SEND,
747 				    stcb->sctp_ep, stcb, net);
748 			}
749 		} else if (net->net_ack > 0) {
750 			/*
751 			 * Mark a peg that we WOULD have done a cwnd
752 			 * reduction but RFC2582 prevented this action.
753 			 */
754 			SCTP_STAT_INCR(sctps_fastretransinrtt);
755 		}
756 	}
757 }
758 
759 void
760 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb,
761     struct sctp_association *asoc,
762     int accum_moved, int reneged_all, int will_exit)
763 {
764 	struct sctp_nets *net;
765 
766 	/******************************/
767 	/* update cwnd and Early FR   */
768 	/******************************/
769 	TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
770 
771 #ifdef JANA_CMT_FAST_RECOVERY
772 		/*
773 		 * CMT fast recovery code. Need to debug.
774 		 */
775 		if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
776 			if (compare_with_wrap(asoc->last_acked_seq,
777 			    net->fast_recovery_tsn, MAX_TSN) ||
778 			    (asoc->last_acked_seq == net->fast_recovery_tsn) ||
779 			    compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) ||
780 			    (net->pseudo_cumack == net->fast_recovery_tsn)) {
781 				net->will_exit_fast_recovery = 1;
782 			}
783 		}
784 #endif
785 		if (sctp_early_fr) {
786 			/*
787 			 * So, first of all do we need to have a Early FR
788 			 * timer running?
789 			 */
790 			if (((TAILQ_FIRST(&asoc->sent_queue)) &&
791 			    (net->ref_count > 1) &&
792 			    (net->flight_size < net->cwnd)) ||
793 			    (reneged_all)) {
794 				/*
795 				 * yes, so in this case stop it if its
796 				 * running, and then restart it. Reneging
797 				 * all is a special case where we want to
798 				 * run the Early FR timer and then force the
799 				 * last few unacked to be sent, causing us
800 				 * to illicit a sack with gaps to force out
801 				 * the others.
802 				 */
803 				if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
804 					SCTP_STAT_INCR(sctps_earlyfrstpidsck2);
805 					sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
806 					    SCTP_FROM_SCTP_INDATA + SCTP_LOC_20);
807 				}
808 				SCTP_STAT_INCR(sctps_earlyfrstrid);
809 				sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net);
810 			} else {
811 				/* No, stop it if its running */
812 				if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
813 					SCTP_STAT_INCR(sctps_earlyfrstpidsck3);
814 					sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
815 					    SCTP_FROM_SCTP_INDATA + SCTP_LOC_21);
816 				}
817 			}
818 		}
819 		/* if nothing was acked on this destination skip it */
820 		if (net->net_ack == 0) {
821 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
822 				sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
823 			}
824 			continue;
825 		}
826 		if (net->net_ack2 > 0) {
827 			/*
828 			 * Karn's rule applies to clearing error count, this
829 			 * is optional.
830 			 */
831 			net->error_count = 0;
832 			if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) ==
833 			    SCTP_ADDR_NOT_REACHABLE) {
834 				/* addr came good */
835 				net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE;
836 				net->dest_state |= SCTP_ADDR_REACHABLE;
837 				sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb,
838 				    SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED);
839 				/* now was it the primary? if so restore */
840 				if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) {
841 					(void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net);
842 				}
843 			}
844 			/*
845 			 * JRS 5/14/07 - If CMT PF is on and the destination
846 			 * is in PF state, set the destination to active
847 			 * state and set the cwnd to one or two MTU's based
848 			 * on whether PF1 or PF2 is being used.
849 			 *
850 			 * Should we stop any running T3 timer here?
851 			 */
852 			if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) ==
853 			    SCTP_ADDR_PF)) {
854 				net->dest_state &= ~SCTP_ADDR_PF;
855 				net->cwnd = net->mtu * sctp_cmt_pf;
856 				SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
857 				    net, net->cwnd);
858 				/*
859 				 * Since the cwnd value is explicitly set,
860 				 * skip the code that updates the cwnd
861 				 * value.
862 				 */
863 				goto skip_cwnd_update;
864 			}
865 		}
866 #ifdef JANA_CMT_FAST_RECOVERY
867 		/*
868 		 * CMT fast recovery code
869 		 */
870 		/*
871 		 * if (sctp_cmt_on_off == 1 &&
872 		 * net->fast_retran_loss_recovery &&
873 		 * net->will_exit_fast_recovery == 0) { @@@ Do something }
874 		 * else if (sctp_cmt_on_off == 0 &&
875 		 * asoc->fast_retran_loss_recovery && will_exit == 0) {
876 		 */
877 #endif
878 
879 		if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) {
880 			/*
881 			 * If we are in loss recovery we skip any cwnd
882 			 * update
883 			 */
884 			goto skip_cwnd_update;
885 		}
886 		/*
887 		 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
888 		 * moved.
889 		 */
890 		if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) {
891 			/* If the cumulative ack moved we can proceed */
892 			if (net->cwnd <= net->ssthresh) {
893 				/* We are in slow start */
894 				if (net->flight_size + net->net_ack >= net->cwnd) {
895 
896 					sctp_hs_cwnd_increase(stcb, net);
897 
898 				} else {
899 					if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
900 						sctp_log_cwnd(stcb, net, net->net_ack,
901 						    SCTP_CWND_LOG_NOADV_SS);
902 					}
903 				}
904 			} else {
905 				/* We are in congestion avoidance */
906 				net->partial_bytes_acked += net->net_ack;
907 				if ((net->flight_size + net->net_ack >= net->cwnd) &&
908 				    (net->partial_bytes_acked >= net->cwnd)) {
909 					net->partial_bytes_acked -= net->cwnd;
910 					net->cwnd += net->mtu;
911 					if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
912 						sctp_log_cwnd(stcb, net, net->mtu,
913 						    SCTP_CWND_LOG_FROM_CA);
914 					}
915 				} else {
916 					if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
917 						sctp_log_cwnd(stcb, net, net->net_ack,
918 						    SCTP_CWND_LOG_NOADV_CA);
919 					}
920 				}
921 			}
922 		} else {
923 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
924 				sctp_log_cwnd(stcb, net, net->mtu,
925 				    SCTP_CWND_LOG_NO_CUMACK);
926 			}
927 		}
928 skip_cwnd_update:
929 		/*
930 		 * NOW, according to Karn's rule do we need to restore the
931 		 * RTO timer back? Check our net_ack2. If not set then we
932 		 * have a ambiguity.. i.e. all data ack'd was sent to more
933 		 * than one place.
934 		 */
935 		if (net->net_ack2) {
936 			/* restore any doubled timers */
937 			net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1;
938 			if (net->RTO < stcb->asoc.minrto) {
939 				net->RTO = stcb->asoc.minrto;
940 			}
941 			if (net->RTO > stcb->asoc.maxrto) {
942 				net->RTO = stcb->asoc.maxrto;
943 			}
944 		}
945 	}
946 }
947 
948 
949 /*
950  * H-TCP congestion control. The algorithm is detailed in:
951  * R.N.Shorten, D.J.Leith:
952  *   "H-TCP: TCP for high-speed and long-distance networks"
953  *   Proc. PFLDnet, Argonne, 2004.
954  * http://www.hamilton.ie/net/htcp3.pdf
955  */
956 
957 
958 static int use_rtt_scaling = 1;
959 static int use_bandwidth_switch = 1;
960 
961 static inline int
962 between(uint32_t seq1, uint32_t seq2, uint32_t seq3)
963 {
964 	return seq3 - seq2 >= seq1 - seq2;
965 }
966 
967 static inline uint32_t
968 htcp_cong_time(struct htcp *ca)
969 {
970 	return sctp_get_tick_count() - ca->last_cong;
971 }
972 
973 static inline uint32_t
974 htcp_ccount(struct htcp *ca)
975 {
976 	return htcp_cong_time(ca) / ca->minRTT;
977 }
978 
979 static inline void
980 htcp_reset(struct htcp *ca)
981 {
982 	ca->undo_last_cong = ca->last_cong;
983 	ca->undo_maxRTT = ca->maxRTT;
984 	ca->undo_old_maxB = ca->old_maxB;
985 	ca->last_cong = sctp_get_tick_count();
986 }
987 
988 #ifdef SCTP_NOT_USED
989 
990 static uint32_t
991 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net)
992 {
993 	net->htcp_ca.last_cong = net->htcp_ca.undo_last_cong;
994 	net->htcp_ca.maxRTT = net->htcp_ca.undo_maxRTT;
995 	net->htcp_ca.old_maxB = net->htcp_ca.undo_old_maxB;
996 	return max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->htcp_ca.beta) * net->mtu);
997 }
998 
999 #endif
1000 
1001 static inline void
1002 measure_rtt(struct sctp_tcb *stcb, struct sctp_nets *net)
1003 {
1004 	uint32_t srtt = net->lastsa >> 3;
1005 
1006 	/* keep track of minimum RTT seen so far, minRTT is zero at first */
1007 	if (net->htcp_ca.minRTT > srtt || !net->htcp_ca.minRTT)
1008 		net->htcp_ca.minRTT = srtt;
1009 
1010 	/* max RTT */
1011 	if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->htcp_ca) > 3) {
1012 		if (net->htcp_ca.maxRTT < net->htcp_ca.minRTT)
1013 			net->htcp_ca.maxRTT = net->htcp_ca.minRTT;
1014 		if (net->htcp_ca.maxRTT < srtt && srtt <= net->htcp_ca.maxRTT + MSEC_TO_TICKS(20))
1015 			net->htcp_ca.maxRTT = srtt;
1016 	}
1017 }
1018 
1019 static void
1020 measure_achieved_throughput(struct sctp_tcb *stcb, struct sctp_nets *net)
1021 {
1022 	uint32_t now = sctp_get_tick_count();
1023 
1024 	if (net->fast_retran_ip == 0)
1025 		net->htcp_ca.bytes_acked = net->net_ack;
1026 
1027 	if (!use_bandwidth_switch)
1028 		return;
1029 
1030 	/* achieved throughput calculations */
1031 	/* JRS - not 100% sure of this statement */
1032 	if (net->fast_retran_ip == 1) {
1033 		net->htcp_ca.bytecount = 0;
1034 		net->htcp_ca.lasttime = now;
1035 		return;
1036 	}
1037 	net->htcp_ca.bytecount += net->net_ack;
1038 
1039 	if (net->htcp_ca.bytecount >= net->cwnd - ((net->htcp_ca.alpha >> 7 ? : 1) * net->mtu)
1040 	    && now - net->htcp_ca.lasttime >= net->htcp_ca.minRTT
1041 	    && net->htcp_ca.minRTT > 0) {
1042 		uint32_t cur_Bi = net->htcp_ca.bytecount / net->mtu * hz / (now - net->htcp_ca.lasttime);
1043 
1044 		if (htcp_ccount(&net->htcp_ca) <= 3) {
1045 			/* just after backoff */
1046 			net->htcp_ca.minB = net->htcp_ca.maxB = net->htcp_ca.Bi = cur_Bi;
1047 		} else {
1048 			net->htcp_ca.Bi = (3 * net->htcp_ca.Bi + cur_Bi) / 4;
1049 			if (net->htcp_ca.Bi > net->htcp_ca.maxB)
1050 				net->htcp_ca.maxB = net->htcp_ca.Bi;
1051 			if (net->htcp_ca.minB > net->htcp_ca.maxB)
1052 				net->htcp_ca.minB = net->htcp_ca.maxB;
1053 		}
1054 		net->htcp_ca.bytecount = 0;
1055 		net->htcp_ca.lasttime = now;
1056 	}
1057 }
1058 
1059 static inline void
1060 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT)
1061 {
1062 	if (use_bandwidth_switch) {
1063 		uint32_t maxB = ca->maxB;
1064 		uint32_t old_maxB = ca->old_maxB;
1065 
1066 		ca->old_maxB = ca->maxB;
1067 
1068 		if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) {
1069 			ca->beta = BETA_MIN;
1070 			ca->modeswitch = 0;
1071 			return;
1072 		}
1073 	}
1074 	if (ca->modeswitch && minRTT > (uint32_t) MSEC_TO_TICKS(10) && maxRTT) {
1075 		ca->beta = (minRTT << 7) / maxRTT;
1076 		if (ca->beta < BETA_MIN)
1077 			ca->beta = BETA_MIN;
1078 		else if (ca->beta > BETA_MAX)
1079 			ca->beta = BETA_MAX;
1080 	} else {
1081 		ca->beta = BETA_MIN;
1082 		ca->modeswitch = 1;
1083 	}
1084 }
1085 
1086 static inline void
1087 htcp_alpha_update(struct htcp *ca)
1088 {
1089 	uint32_t minRTT = ca->minRTT;
1090 	uint32_t factor = 1;
1091 	uint32_t diff = htcp_cong_time(ca);
1092 
1093 	if (diff > (uint32_t) hz) {
1094 		diff -= hz;
1095 		factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz;
1096 	}
1097 	if (use_rtt_scaling && minRTT) {
1098 		uint32_t scale = (hz << 3) / (10 * minRTT);
1099 
1100 		scale = min(max(scale, 1U << 2), 10U << 3);	/* clamping ratio to
1101 								 * interval [0.5,10]<<3 */
1102 		factor = (factor << 3) / scale;
1103 		if (!factor)
1104 			factor = 1;
1105 	}
1106 	ca->alpha = 2 * factor * ((1 << 7) - ca->beta);
1107 	if (!ca->alpha)
1108 		ca->alpha = ALPHA_BASE;
1109 }
1110 
1111 /* After we have the rtt data to calculate beta, we'd still prefer to wait one
1112  * rtt before we adjust our beta to ensure we are working from a consistent
1113  * data.
1114  *
1115  * This function should be called when we hit a congestion event since only at
1116  * that point do we really have a real sense of maxRTT (the queues en route
1117  * were getting just too full now).
1118  */
1119 static void
1120 htcp_param_update(struct sctp_tcb *stcb, struct sctp_nets *net)
1121 {
1122 	uint32_t minRTT = net->htcp_ca.minRTT;
1123 	uint32_t maxRTT = net->htcp_ca.maxRTT;
1124 
1125 	htcp_beta_update(&net->htcp_ca, minRTT, maxRTT);
1126 	htcp_alpha_update(&net->htcp_ca);
1127 
1128 	/*
1129 	 * add slowly fading memory for maxRTT to accommodate routing
1130 	 * changes etc
1131 	 */
1132 	if (minRTT > 0 && maxRTT > minRTT)
1133 		net->htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100;
1134 }
1135 
1136 static uint32_t
1137 htcp_recalc_ssthresh(struct sctp_tcb *stcb, struct sctp_nets *net)
1138 {
1139 	htcp_param_update(stcb, net);
1140 	return max(((net->cwnd / net->mtu * net->htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu);
1141 }
1142 
1143 static void
1144 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net)
1145 {
1146 	/*-
1147 	 * How to handle these functions?
1148          *	if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question.
1149 	 *		return;
1150 	 */
1151 	if (net->cwnd <= net->ssthresh) {
1152 		/* We are in slow start */
1153 		if (net->flight_size + net->net_ack >= net->cwnd) {
1154 			if (net->net_ack > (net->mtu * sctp_L2_abc_variable)) {
1155 				net->cwnd += (net->mtu * sctp_L2_abc_variable);
1156 				if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1157 					sctp_log_cwnd(stcb, net, net->mtu,
1158 					    SCTP_CWND_LOG_FROM_SS);
1159 				}
1160 			} else {
1161 				net->cwnd += net->net_ack;
1162 				if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1163 					sctp_log_cwnd(stcb, net, net->net_ack,
1164 					    SCTP_CWND_LOG_FROM_SS);
1165 				}
1166 			}
1167 		} else {
1168 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1169 				sctp_log_cwnd(stcb, net, net->net_ack,
1170 				    SCTP_CWND_LOG_NOADV_SS);
1171 			}
1172 		}
1173 	} else {
1174 		measure_rtt(stcb, net);
1175 
1176 		/*
1177 		 * In dangerous area, increase slowly. In theory this is
1178 		 * net->cwnd += alpha / net->cwnd
1179 		 */
1180 		/* What is snd_cwnd_cnt?? */
1181 		if (((net->partial_bytes_acked / net->mtu * net->htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) {
1182 			/*-
1183 			 * Does SCTP have a cwnd clamp?
1184 			 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS).
1185 			 */
1186 			net->cwnd += net->mtu;
1187 			net->partial_bytes_acked = 0;
1188 			htcp_alpha_update(&net->htcp_ca);
1189 			if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1190 				sctp_log_cwnd(stcb, net, net->mtu,
1191 				    SCTP_CWND_LOG_FROM_CA);
1192 			}
1193 		} else {
1194 			net->partial_bytes_acked += net->net_ack;
1195 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1196 				sctp_log_cwnd(stcb, net, net->net_ack,
1197 				    SCTP_CWND_LOG_NOADV_CA);
1198 			}
1199 		}
1200 
1201 		net->htcp_ca.bytes_acked = net->mtu;
1202 	}
1203 }
1204 
1205 #ifdef SCTP_NOT_USED
1206 /* Lower bound on congestion window. */
1207 static uint32_t
1208 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net)
1209 {
1210 	return net->ssthresh;
1211 }
1212 
1213 #endif
1214 
1215 static void
1216 htcp_init(struct sctp_tcb *stcb, struct sctp_nets *net)
1217 {
1218 	memset(&net->htcp_ca, 0, sizeof(struct htcp));
1219 	net->htcp_ca.alpha = ALPHA_BASE;
1220 	net->htcp_ca.beta = BETA_MIN;
1221 	net->htcp_ca.bytes_acked = net->mtu;
1222 	net->htcp_ca.last_cong = sctp_get_tick_count();
1223 }
1224 
1225 void
1226 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
1227 {
1228 	/*
1229 	 * We take the max of the burst limit times a MTU or the
1230 	 * INITIAL_CWND. We then limit this to 4 MTU's of sending.
1231 	 */
1232 	net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
1233 	net->ssthresh = stcb->asoc.peers_rwnd;
1234 	htcp_init(stcb, net);
1235 
1236 	if (sctp_logging_level & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) {
1237 		sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
1238 	}
1239 }
1240 
1241 void
1242 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb,
1243     struct sctp_association *asoc,
1244     int accum_moved, int reneged_all, int will_exit)
1245 {
1246 	struct sctp_nets *net;
1247 
1248 	/******************************/
1249 	/* update cwnd and Early FR   */
1250 	/******************************/
1251 	TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
1252 
1253 #ifdef JANA_CMT_FAST_RECOVERY
1254 		/*
1255 		 * CMT fast recovery code. Need to debug.
1256 		 */
1257 		if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
1258 			if (compare_with_wrap(asoc->last_acked_seq,
1259 			    net->fast_recovery_tsn, MAX_TSN) ||
1260 			    (asoc->last_acked_seq == net->fast_recovery_tsn) ||
1261 			    compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) ||
1262 			    (net->pseudo_cumack == net->fast_recovery_tsn)) {
1263 				net->will_exit_fast_recovery = 1;
1264 			}
1265 		}
1266 #endif
1267 		if (sctp_early_fr) {
1268 			/*
1269 			 * So, first of all do we need to have a Early FR
1270 			 * timer running?
1271 			 */
1272 			if (((TAILQ_FIRST(&asoc->sent_queue)) &&
1273 			    (net->ref_count > 1) &&
1274 			    (net->flight_size < net->cwnd)) ||
1275 			    (reneged_all)) {
1276 				/*
1277 				 * yes, so in this case stop it if its
1278 				 * running, and then restart it. Reneging
1279 				 * all is a special case where we want to
1280 				 * run the Early FR timer and then force the
1281 				 * last few unacked to be sent, causing us
1282 				 * to illicit a sack with gaps to force out
1283 				 * the others.
1284 				 */
1285 				if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
1286 					SCTP_STAT_INCR(sctps_earlyfrstpidsck2);
1287 					sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
1288 					    SCTP_FROM_SCTP_INDATA + SCTP_LOC_20);
1289 				}
1290 				SCTP_STAT_INCR(sctps_earlyfrstrid);
1291 				sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net);
1292 			} else {
1293 				/* No, stop it if its running */
1294 				if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
1295 					SCTP_STAT_INCR(sctps_earlyfrstpidsck3);
1296 					sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
1297 					    SCTP_FROM_SCTP_INDATA + SCTP_LOC_21);
1298 				}
1299 			}
1300 		}
1301 		/* if nothing was acked on this destination skip it */
1302 		if (net->net_ack == 0) {
1303 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1304 				sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
1305 			}
1306 			continue;
1307 		}
1308 		if (net->net_ack2 > 0) {
1309 			/*
1310 			 * Karn's rule applies to clearing error count, this
1311 			 * is optional.
1312 			 */
1313 			net->error_count = 0;
1314 			if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) ==
1315 			    SCTP_ADDR_NOT_REACHABLE) {
1316 				/* addr came good */
1317 				net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE;
1318 				net->dest_state |= SCTP_ADDR_REACHABLE;
1319 				sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb,
1320 				    SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED);
1321 				/* now was it the primary? if so restore */
1322 				if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) {
1323 					(void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net);
1324 				}
1325 			}
1326 			/*
1327 			 * JRS 5/14/07 - If CMT PF is on and the destination
1328 			 * is in PF state, set the destination to active
1329 			 * state and set the cwnd to one or two MTU's based
1330 			 * on whether PF1 or PF2 is being used.
1331 			 *
1332 			 * Should we stop any running T3 timer here?
1333 			 */
1334 			if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) ==
1335 			    SCTP_ADDR_PF)) {
1336 				net->dest_state &= ~SCTP_ADDR_PF;
1337 				net->cwnd = net->mtu * sctp_cmt_pf;
1338 				SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
1339 				    net, net->cwnd);
1340 				/*
1341 				 * Since the cwnd value is explicitly set,
1342 				 * skip the code that updates the cwnd
1343 				 * value.
1344 				 */
1345 				goto skip_cwnd_update;
1346 			}
1347 		}
1348 #ifdef JANA_CMT_FAST_RECOVERY
1349 		/*
1350 		 * CMT fast recovery code
1351 		 */
1352 		/*
1353 		 * if (sctp_cmt_on_off == 1 &&
1354 		 * net->fast_retran_loss_recovery &&
1355 		 * net->will_exit_fast_recovery == 0) { @@@ Do something }
1356 		 * else if (sctp_cmt_on_off == 0 &&
1357 		 * asoc->fast_retran_loss_recovery && will_exit == 0) {
1358 		 */
1359 #endif
1360 
1361 		if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) {
1362 			/*
1363 			 * If we are in loss recovery we skip any cwnd
1364 			 * update
1365 			 */
1366 			goto skip_cwnd_update;
1367 		}
1368 		/*
1369 		 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
1370 		 * moved.
1371 		 */
1372 		if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) {
1373 			htcp_cong_avoid(stcb, net);
1374 			measure_achieved_throughput(stcb, net);
1375 		} else {
1376 			if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1377 				sctp_log_cwnd(stcb, net, net->mtu,
1378 				    SCTP_CWND_LOG_NO_CUMACK);
1379 			}
1380 		}
1381 skip_cwnd_update:
1382 		/*
1383 		 * NOW, according to Karn's rule do we need to restore the
1384 		 * RTO timer back? Check our net_ack2. If not set then we
1385 		 * have a ambiguity.. i.e. all data ack'd was sent to more
1386 		 * than one place.
1387 		 */
1388 		if (net->net_ack2) {
1389 			/* restore any doubled timers */
1390 			net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1;
1391 			if (net->RTO < stcb->asoc.minrto) {
1392 				net->RTO = stcb->asoc.minrto;
1393 			}
1394 			if (net->RTO > stcb->asoc.maxrto) {
1395 				net->RTO = stcb->asoc.maxrto;
1396 			}
1397 		}
1398 	}
1399 }
1400 
1401 void
1402 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb,
1403     struct sctp_association *asoc)
1404 {
1405 	struct sctp_nets *net;
1406 
1407 	/*
1408 	 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) &&
1409 	 * (net->fast_retran_loss_recovery == 0)))
1410 	 */
1411 	TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
1412 		if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) {
1413 			/* out of a RFC2582 Fast recovery window? */
1414 			if (net->net_ack > 0) {
1415 				/*
1416 				 * per section 7.2.3, are there any
1417 				 * destinations that had a fast retransmit
1418 				 * to them. If so what we need to do is
1419 				 * adjust ssthresh and cwnd.
1420 				 */
1421 				struct sctp_tmit_chunk *lchk;
1422 				int old_cwnd = net->cwnd;
1423 
1424 				/* JRS - reset as if state were changed */
1425 				htcp_reset(&net->htcp_ca);
1426 				net->ssthresh = htcp_recalc_ssthresh(stcb, net);
1427 				net->cwnd = net->ssthresh;
1428 				if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1429 					sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
1430 					    SCTP_CWND_LOG_FROM_FR);
1431 				}
1432 				lchk = TAILQ_FIRST(&asoc->send_queue);
1433 
1434 				net->partial_bytes_acked = 0;
1435 				/* Turn on fast recovery window */
1436 				asoc->fast_retran_loss_recovery = 1;
1437 				if (lchk == NULL) {
1438 					/* Mark end of the window */
1439 					asoc->fast_recovery_tsn = asoc->sending_seq - 1;
1440 				} else {
1441 					asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
1442 				}
1443 
1444 				/*
1445 				 * CMT fast recovery -- per destination
1446 				 * recovery variable.
1447 				 */
1448 				net->fast_retran_loss_recovery = 1;
1449 
1450 				if (lchk == NULL) {
1451 					/* Mark end of the window */
1452 					net->fast_recovery_tsn = asoc->sending_seq - 1;
1453 				} else {
1454 					net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
1455 				}
1456 
1457 				/*
1458 				 * Disable Nonce Sum Checking and store the
1459 				 * resync tsn
1460 				 */
1461 				asoc->nonce_sum_check = 0;
1462 				asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1;
1463 
1464 				sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
1465 				    stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32);
1466 				sctp_timer_start(SCTP_TIMER_TYPE_SEND,
1467 				    stcb->sctp_ep, stcb, net);
1468 			}
1469 		} else if (net->net_ack > 0) {
1470 			/*
1471 			 * Mark a peg that we WOULD have done a cwnd
1472 			 * reduction but RFC2582 prevented this action.
1473 			 */
1474 			SCTP_STAT_INCR(sctps_fastretransinrtt);
1475 		}
1476 	}
1477 }
1478 
1479 void
1480 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb,
1481     struct sctp_nets *net)
1482 {
1483 	int old_cwnd = net->cwnd;
1484 
1485 	/* JRS - reset as if the state were being changed to timeout */
1486 	htcp_reset(&net->htcp_ca);
1487 	net->ssthresh = htcp_recalc_ssthresh(stcb, net);
1488 	net->cwnd = net->mtu;
1489 	net->partial_bytes_acked = 0;
1490 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1491 		sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
1492 	}
1493 }
1494 
1495 void
1496 sctp_htcp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp,
1497     struct sctp_tcb *stcb, struct sctp_nets *net)
1498 {
1499 	int old_cwnd;
1500 
1501 	old_cwnd = net->cwnd;
1502 
1503 	sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED);
1504 	net->htcp_ca.last_cong = sctp_get_tick_count();
1505 	/*
1506 	 * make a small adjustment to cwnd and force to CA.
1507 	 */
1508 	if (net->cwnd > net->mtu)
1509 		/* drop down one MTU after sending */
1510 		net->cwnd -= net->mtu;
1511 	if (net->cwnd < net->ssthresh)
1512 		/* still in SS move to CA */
1513 		net->ssthresh = net->cwnd - 1;
1514 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1515 		sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR);
1516 	}
1517 }
1518 
1519 void
1520 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb,
1521     struct sctp_nets *net)
1522 {
1523 	int old_cwnd;
1524 
1525 	old_cwnd = net->cwnd;
1526 
1527 	/* JRS - reset hctp as if state changed */
1528 	htcp_reset(&net->htcp_ca);
1529 	SCTP_STAT_INCR(sctps_ecnereducedcwnd);
1530 	net->ssthresh = htcp_recalc_ssthresh(stcb, net);
1531 	if (net->ssthresh < net->mtu) {
1532 		net->ssthresh = net->mtu;
1533 		/* here back off the timer as well, to slow us down */
1534 		net->RTO <<= 1;
1535 	}
1536 	net->cwnd = net->ssthresh;
1537 	if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1538 		sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
1539 	}
1540 }
1541