1/*-
  2 * Copyright (c) 2016-2020 Netflix, Inc.
  3 *
  4 * Redistribution and use in source and binary forms, with or without
  5 * modification, are permitted provided that the following conditions
  6 * are met:
  7 * 1. Redistributions of source code must retain the above copyright
  8 *    notice, this list of conditions and the following disclaimer.
  9 * 2. Redistributions in binary form must reproduce the above copyright
 10 *    notice, this list of conditions and the following disclaimer in the
 11 *    documentation and/or other materials provided with the distribution.
 12 *
 13 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 16 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 23 * SUCH DAMAGE.
 24 */
 25
 26#ifndef _NETINET_TCP_RACK_H_
 27#define _NETINET_TCP_RACK_H_
 28
 29#define RACK_ACKED	    0x000001/* The remote endpoint acked this */
 30#define RACK_TO_REXT	    0x000002/* A timeout occurred on this sendmap entry */
 31#define RACK_DEFERRED	    0x000004/* We can't use this for RTT calc - not used */
 32#define RACK_OVERMAX	    0x000008/* We have more retran's then we can fit */
 33#define RACK_SACK_PASSED    0x000010/* A sack was done above this block */
 34#define RACK_WAS_SACKPASS   0x000020/* We retransmitted due to SACK pass */
 35#define RACK_HAS_FIN	    0x000040/* segment is sent with fin */
 36#define RACK_TLP	    0x000080/* segment sent as tail-loss-probe */
 37#define RACK_RWND_COLLAPSED 0x000100/* The peer collapsed the rwnd on the segment */
 38#define RACK_APP_LIMITED    0x000200/* We went app limited after this send */
 39#define RACK_WAS_ACKED	    0x000400/* a RTO undid the ack, but it already had a rtt calc done */
 40#define RACK_HAS_SYN	    0x000800/* SYN is on this guy */
 41#define RACK_SENT_W_DSACK   0x001000/* Sent with a dsack */
 42#define RACK_SENT_SP	    0x002000/* sent in slow path */
 43#define RACK_SENT_FP        0x004000/* sent in fast path */
 44#define RACK_HAD_PUSH	    0x008000/* Push was sent on original send */
 45#define RACK_MUST_RXT	    0x010000/* We must retransmit this rsm (non-sack/mtu chg)*/
 46#define RACK_IN_GP_WIN	    0x020000/* Send was in GP window when sent */
 47#define RACK_SHUFFLED	    0x040000/* The RSM was shuffled some data from one to another */
 48#define RACK_MERGED	    0x080000/* The RSM was merged */
 49#define RACK_PMTU_CHG	    0x100000/* The path mtu changed on this guy */
 50#define RACK_STRADDLE	    0x200000/* The seq straddles the bucket line */
 51#define RACK_NUM_OF_RETRANS 3
 52
 53#define RACK_INITIAL_RTO 1000000 /* 1 second in microseconds */
 54
 55#define RACK_REQ_AVG 3 	/* Must be less than 256 */
 56
 57struct rack_sendmap {
 58	TAILQ_ENTRY(rack_sendmap) next;
 59	TAILQ_ENTRY(rack_sendmap) r_tnext;	/* Time of transmit based next */
 60	uint32_t bindex;
 61	uint32_t r_start;	/* Sequence number of the segment */
 62	uint32_t r_end;		/* End seq, this is 1 beyond actually */
 63	uint32_t r_rtr_bytes;	/* How many bytes have been retransmitted */
 64	uint32_t r_flags : 24,	/* Flags as defined above */
 65		 r_rtr_cnt : 8;	/* Retran count, index this -1 to get time */
 66	struct mbuf *m;
 67	uint32_t soff;
 68	uint32_t orig_m_len;	/* The original mbuf len when we sent (can update) */
 69	uint32_t orig_t_space;	/* The original trailing space when we sent (can update) */
 70	uint32_t r_nseq_appl;	/* If this one is app limited, this is the nxt seq limited */
 71	uint8_t r_dupack;	/* Dup ack count */
 72	uint8_t r_in_tmap;	/* Flag to see if its in the r_tnext array */
 73	uint8_t r_limit_type;	/* is this entry counted against a limit? */
 74	uint8_t r_just_ret : 1, /* After sending, the next pkt was just returned, i.e. limited  */
 75		r_one_out_nr : 1,	/* Special case 1 outstanding and not in recovery */
 76		r_no_rtt_allowed : 1, /* No rtt measurement allowed */
 77		r_hw_tls : 1,
 78		r_avail : 4;
 79	uint64_t r_tim_lastsent[RACK_NUM_OF_RETRANS];
 80	uint64_t r_ack_arrival;	/* This is the time of ack-arrival (if SACK'd) */
 81	uint32_t r_fas;		/* Flight at send */
 82	uint8_t r_bas;		/* The burst size (burst at send = bas)  */
 83};
 84
 85struct deferred_opt_list {
 86	TAILQ_ENTRY(deferred_opt_list) next;
 87	int optname;
 88	uint64_t optval;
 89};
 90
 91/*
 92 * Timestamps in the rack sendmap are now moving to be
 93 * uint64_t's. This means that if you want a uint32_t
 94 * usec timestamp (the old usecond timestamp) you simply have
 95 * to cast it to uint32_t. The reason we do this is not for
 96 * wrap, but we need to get back, at times, to the millisecond
 97 * timestamp that is used in the TSTMP option. To do this we
 98 * can use the rack_ts_to_msec() inline below which can take
 99 * the 64bit ts and make into the correct timestamp millisecond
100 * wise. Thats not possible with the 32bit usecond timestamp since
101 * the seconds wrap too quickly to cover all bases.
102 *
103 * There are quite a few places in rack where I simply cast
104 * back to uint32_t and then end up using the TSTMP_XX()
105 * macros. This is ok, but we could do simple compares if
106 * we ever decided to move all of those variables to 64 bits
107 * as well.
108 */
109
110static inline uint64_t
111rack_to_usec_ts(struct timeval *tv)
112{
113	return ((tv->tv_sec * HPTS_USEC_IN_SEC) + tv->tv_usec);
114}
115
116static inline uint32_t
117rack_ts_to_msec(uint64_t ts)
118{
119	return((uint32_t)(ts / HPTS_MSEC_IN_SEC));
120}
121
122
123TAILQ_HEAD(rack_head, rack_sendmap);
124TAILQ_HEAD(def_opt_head, deferred_opt_list);
125
126/* Map change logging */
127#define MAP_MERGE	0x01
128#define MAP_SPLIT	0x02
129#define MAP_NEW		0x03
130#define MAP_SACK_M1	0x04
131#define MAP_SACK_M2	0x05
132#define MAP_SACK_M3	0x06
133#define MAP_SACK_M4	0x07
134#define MAP_SACK_M5	0x08
135#define MAP_FREE	0x09
136#define MAP_TRIM_HEAD	0x0a
137
138#define RACK_LIMIT_TYPE_SPLIT	1
139
140/*
141 * We use the rate sample structure to
142 * assist in single sack/ack rate and rtt
143 * calculation. In the future we will expand
144 * this in BBR to do forward rate sample
145 * b/w estimation.
146 */
147#define RACK_RTT_EMPTY 0x00000001	/* Nothing yet stored in RTT's */
148#define RACK_RTT_VALID 0x00000002	/* We have at least one valid RTT */
149struct rack_rtt_sample {
150	uint32_t rs_flags;
151	uint32_t rs_rtt_lowest;
152	uint32_t rs_rtt_highest;
153	uint32_t rs_rtt_cnt;
154	uint32_t rs_us_rtt;
155	int32_t  confidence;
156	uint64_t rs_rtt_tot;
157	uint16_t rs_us_rtrcnt;
158};
159
160#define RACK_LOG_TYPE_ACK	0x01
161#define RACK_LOG_TYPE_OUT	0x02
162#define RACK_LOG_TYPE_TO	0x03
163#define RACK_LOG_TYPE_ALLOC     0x04
164#define RACK_LOG_TYPE_FREE      0x05
165
166/*
167 * Magic numbers for logging timeout events if the
168 * logging is enabled.
169 */
170#define RACK_TO_FRM_TMR  1
171#define RACK_TO_FRM_TLP  2
172#define RACK_TO_FRM_RACK 3
173#define RACK_TO_FRM_KEEP 4
174#define RACK_TO_FRM_PERSIST 5
175#define RACK_TO_FRM_DELACK 6
176
177struct rack_opts_stats {
178	uint64_t tcp_rack_tlp_reduce;
179	uint64_t tcp_rack_pace_always;
180	uint64_t tcp_rack_pace_reduce;
181	uint64_t tcp_rack_max_seg;
182	uint64_t tcp_rack_prr_sendalot;
183	uint64_t tcp_rack_min_to;
184	uint64_t tcp_rack_early_seg;
185	uint64_t tcp_rack_reord_thresh;
186	uint64_t tcp_rack_reord_fade;
187	uint64_t tcp_rack_tlp_thresh;
188	uint64_t tcp_rack_pkt_delay;
189	uint64_t tcp_rack_tlp_inc_var;
190	uint64_t tcp_tlp_use;
191	uint64_t tcp_rack_idle_reduce;
192	uint64_t tcp_rack_idle_reduce_high;
193	uint64_t rack_no_timer_in_hpts;
194	uint64_t tcp_rack_min_pace_seg;
195	uint64_t tcp_rack_pace_rate_ca;
196	uint64_t tcp_rack_rr;
197	uint64_t tcp_rack_do_detection;
198	uint64_t tcp_rack_rrr_no_conf_rate;
199	uint64_t tcp_initial_rate;
200	uint64_t tcp_initial_win;
201	uint64_t tcp_hdwr_pacing;
202	uint64_t tcp_gp_inc_ss;
203	uint64_t tcp_gp_inc_ca;
204	uint64_t tcp_gp_inc_rec;
205	uint64_t tcp_rack_force_max_seg;
206	uint64_t tcp_rack_pace_rate_ss;
207	uint64_t tcp_rack_pace_rate_rec;
208	/* Temp counters for dsack */
209	uint64_t tcp_sack_path_1; /* not used */
210	uint64_t tcp_sack_path_2a; /* not used */
211	uint64_t tcp_sack_path_2b; /* not used */
212	uint64_t tcp_sack_path_3; /* not used */
213	uint64_t tcp_sack_path_4; /* not used */
214	/* non temp counters */
215	uint64_t tcp_rack_scwnd;
216	uint64_t tcp_rack_noprr;
217	uint64_t tcp_rack_cfg_rate;
218	uint64_t tcp_timely_dyn;
219	uint64_t tcp_rack_mbufq;
220	uint64_t tcp_fillcw;
221	uint64_t tcp_npush;
222	uint64_t tcp_lscwnd;
223	uint64_t tcp_profile;
224	uint64_t tcp_hdwr_rate_cap;
225	uint64_t tcp_pacing_rate_cap;
226	uint64_t tcp_pacing_up_only;
227	uint64_t tcp_use_cmp_acks;
228	uint64_t tcp_rack_abc_val;
229	uint64_t tcp_rec_abc_val;
230	uint64_t tcp_rack_measure_cnt;
231	uint64_t tcp_rack_delayed_ack;
232	uint64_t tcp_rack_rtt_use;
233	uint64_t tcp_data_after_close;
234	uint64_t tcp_defer_opt;
235	uint64_t tcp_rxt_clamp;
236	uint64_t tcp_rack_beta;
237	uint64_t tcp_rack_beta_ecn;
238	uint64_t tcp_rack_timer_slop;
239	uint64_t tcp_rack_dsack_opt;
240	uint64_t tcp_rack_hi_beta;
241	uint64_t tcp_split_limit;
242	uint64_t tcp_rack_pacing_divisor;
243	uint64_t tcp_rack_min_seg;
244	uint64_t tcp_dgp_in_rec;
245};
246
247/* RTT shrink reasons */
248#define RACK_RTTS_INIT     0
249#define RACK_RTTS_NEWRTT   1
250#define RACK_RTTS_EXITPROBE 2
251#define RACK_RTTS_ENTERPROBE 3
252#define RACK_RTTS_REACHTARGET 4
253#define RACK_RTTS_SEEHBP 5
254#define RACK_RTTS_NOBACKOFF 6
255#define RACK_RTTS_SAFETY 7
256
257#define RACK_USE_BEG 1
258#define RACK_USE_END 2
259#define RACK_USE_END_OR_THACK 3
260
261#define TLP_USE_ID	1	/* Internet draft behavior */
262#define TLP_USE_TWO_ONE 2	/* Use 2.1 behavior */
263#define TLP_USE_TWO_TWO 3	/* Use 2.2 behavior */
264#define RACK_MIN_BW 8000	/* 64kbps in Bps */
265
266/* Rack quality indicators for GPUT measurements */
267#define RACK_QUALITY_NONE	0	/* No quality stated */
268#define RACK_QUALITY_HIGH 	1	/* A normal measurement of a GP RTT */
269#define RACK_QUALITY_APPLIMITED	2 	/* An app limited case that may be of lower quality */
270#define RACK_QUALITY_PERSIST	3	/* A measurement where we went into persists */
271#define RACK_QUALITY_PROBERTT	4	/* A measurement where we went into or exited probe RTT */
272#define RACK_QUALITY_ALLACKED	5	/* All data is now acknowledged */
273
274#define MIN_GP_WIN 6	/* We need at least 6 MSS in a GP measurement */
275#ifdef _KERNEL
276#define RACK_OPTS_SIZE (sizeof(struct rack_opts_stats)/sizeof(uint64_t))
277extern counter_u64_t rack_opts_arry[RACK_OPTS_SIZE];
278#define RACK_OPTS_ADD(name, amm) counter_u64_add(rack_opts_arry[(offsetof(struct rack_opts_stats, name)/sizeof(uint64_t))], (amm))
279#define RACK_OPTS_INC(name) RACK_OPTS_ADD(name, 1)
280#endif
281/*
282 * As we get each SACK we wade through the
283 * rc_map and mark off what is acked.
284 * We also increment rc_sacked as well.
285 *
286 * We also pay attention to missing entries
287 * based on the time and possibly mark them
288 * for retransmit. If we do and we are not already
289 * in recovery we enter recovery. In doing
290 * so we claer prr_delivered/holes_rxt and prr_sent_dur_rec.
291 * We also setup rc_next/rc_snd_nxt/rc_send_end so
292 * we will know where to send from. When not in
293 * recovery rc_next will be NULL and rc_snd_nxt should
294 * equal snd_max.
295 *
296 * Whenever we retransmit from recovery we increment
297 * rc_holes_rxt as we retran a block and mark it as retransmitted
298 * with the time it was sent. During non-recovery sending we
299 * add to our map and note the time down of any send expanding
300 * the rc_map at the tail and moving rc_snd_nxt up with snd_max.
301 *
302 * In recovery during SACK/ACK processing if a chunk has
303 * been retransmitted and it is now acked, we decrement rc_holes_rxt.
304 * When we retransmit from the scoreboard we use
305 * rc_next and rc_snd_nxt/rc_send_end to help us
306 * find what needs to be retran.
307 *
308 * To calculate pipe we simply take (snd_max - snd_una) + rc_holes_rxt
309 * This gets us the effect of RFC6675 pipe, counting twice for
310 * bytes retransmitted.
311 */
312
313#define TT_RACK_FR_TMR	0x2000
314
315/*
316 * Locking for the rack control block.
317 * a) Locked by INP_WLOCK
318 * b) Locked by the hpts-mutex
319 *
320 */
321#define RACK_GP_HIST 4	/* How much goodput history do we maintain? */
322
323#define RACK_NUM_FSB_DEBUG 16
324#ifdef _KERNEL
325struct rack_fast_send_blk {
326	uint32_t left_to_send;
327	uint16_t tcp_ip_hdr_len;
328	uint8_t tcp_flags;
329	uint8_t hoplimit;
330	uint8_t *tcp_ip_hdr;
331	uint32_t recwin;
332	uint32_t off;
333	struct tcphdr *th;
334	struct udphdr *udp;
335	struct mbuf *m;
336	uint32_t o_m_len;
337	uint32_t o_t_len;
338	uint32_t rfo_apply_push : 1,
339		hw_tls : 1,
340		unused : 30;
341};
342
343struct tailq_hash;
344
345struct rack_control {
346	/* Second cache line 0x40 from tcp_rack */
347	struct tailq_hash *tqh; /* Tree of all segments Lock(a) */
348	struct rack_head rc_tmap;	/* List in transmit order Lock(a) */
349	struct rack_sendmap *rc_tlpsend;	/* Remembered place for
350						 * tlp_sending Lock(a) */
351	struct rack_sendmap *rc_resend;	/* something we have been asked to
352					 * resend */
353	struct rack_fast_send_blk fsb;	/* The fast-send block */
354	uint32_t timer_slop;
355	uint16_t pace_len_divisor;
356	uint16_t rc_user_set_min_segs;
357	uint32_t rc_hpts_flags;
358	uint32_t rc_fixed_pacing_rate_ca;
359	uint32_t rc_fixed_pacing_rate_rec;
360	uint32_t rc_fixed_pacing_rate_ss;
361	uint32_t cwnd_to_use;	/* The cwnd in use */
362	uint32_t rc_timer_exp;	/* If a timer ticks of expiry */
363	uint32_t rc_rack_min_rtt;	/* lowest RTT seen Lock(a) */
364	uint32_t rc_rack_largest_cwnd;	/* Largest CWND we have seen Lock(a) */
365
366	/* Third Cache line 0x80 */
367	struct rack_head rc_free;	/* Allocation array */
368	uint64_t last_hw_bw_req;
369	uint64_t crte_prev_rate;
370	uint64_t bw_rate_cap;
371	uint64_t last_cumack_advance; /* Last time cumack moved forward */
372	uint32_t rc_reorder_ts;	/* Last time we saw reordering Lock(a) */
373
374	uint32_t rc_tlp_new_data;	/* we need to send new-data on a TLP
375					 * Lock(a) */
376	uint32_t rc_prr_out;	/* bytes sent during recovery Lock(a) */
377
378	uint32_t rc_prr_recovery_fs;	/* recovery fs point Lock(a) */
379
380	uint32_t rc_prr_sndcnt;	/* Prr sndcnt Lock(a) */
381
382	uint32_t rc_sacked;	/* Tot sacked on scoreboard Lock(a) */
383	uint32_t last_sent_tlp_seq;	/* Last tlp sequence that was retransmitted Lock(a) */
384
385	uint32_t rc_prr_delivered;	/* during recovery prr var Lock(a) */
386
387	uint16_t rc_tlp_cnt_out;	/* count of times we have sent a TLP without new data */
388	uint16_t last_sent_tlp_len;	/* Number of bytes in the last sent tlp */
389
390	uint32_t rc_loss_count;	/* How many bytes have been retransmitted
391				 * Lock(a) */
392	uint32_t rc_reorder_fade;	/* Socket option value Lock(a) */
393
394	/* Forth cache line 0xc0  */
395	/* Times */
396
397	uint32_t rc_rack_tmit_time;	/* Rack transmit time Lock(a) */
398	uint32_t rc_holes_rxt;	/* Tot retraned from scoreboard Lock(a) */
399
400	uint32_t rc_num_maps_alloced;	/* Number of map blocks (sacks) we
401					 * have allocated */
402	uint32_t rc_rcvtime;	/* When we last received data */
403	uint32_t rc_num_split_allocs;	/* num split map entries allocated */
404	uint32_t rc_split_limit;	/* Limit from control var can be set by socket opt */
405
406	uint32_t rc_last_output_to;
407	uint32_t rc_went_idle_time;
408
409	struct rack_sendmap *rc_sacklast;	/* sack remembered place
410						 * Lock(a) */
411
412	struct rack_sendmap *rc_first_appl;	/* Pointer to first app limited */
413	struct rack_sendmap *rc_end_appl;	/* Pointer to last app limited */
414	/* Cache line split 0x100 */
415	struct sack_filter rack_sf;
416	/* Cache line split 0x140 */
417	/* Flags for various things */
418	uint32_t rc_pace_max_segs;
419	uint32_t rc_pace_min_segs;
420	uint32_t rc_app_limited_cnt;
421	uint16_t rack_per_of_gp_ss; /* 100 = 100%, so from 65536 = 655 x bw  */
422	uint16_t rack_per_of_gp_ca; /* 100 = 100%, so from 65536 = 655 x bw  */
423	uint16_t rack_per_of_gp_rec; /* 100 = 100%, so from 65536 = 655 x bw, 0=off */
424	uint16_t rack_per_of_gp_probertt; /* 100 = 100%, so from 65536 = 655 x bw, 0=off */
425	uint32_t rc_high_rwnd;
426	uint32_t ack_count;
427	uint32_t sack_count;
428	uint32_t sack_noextra_move;
429	uint32_t sack_moved_extra;
430	struct rack_rtt_sample rack_rs;
431	const struct tcp_hwrate_limit_table *crte;
432	uint32_t rc_agg_early;
433	uint32_t rc_agg_delayed;
434	uint32_t rc_tlp_rxt_last_time;
435	uint32_t rc_saved_cwnd;
436	uint64_t rc_gp_output_ts; /* chg*/
437	uint64_t rc_gp_cumack_ts; /* chg*/
438	struct timeval act_rcv_time;
439	struct timeval rc_last_time_decay;	/* SAD time decay happened here */
440	uint64_t gp_bw;
441	uint64_t init_rate;
442#ifdef NETFLIX_SHARED_CWND
443	struct shared_cwnd *rc_scw;
444#endif
445	uint64_t last_gp_comp_bw;
446	uint64_t last_max_bw;	/* Our calculated max b/w last */
447	struct time_filter_small rc_gp_min_rtt;
448	struct def_opt_head opt_list;
449	uint64_t lt_bw_time;	/* Total time with data outstanding (lt_bw = long term bandwidth)  */
450	uint64_t lt_bw_bytes;	/* Total bytes acked */
451	uint64_t lt_timemark;	/* 64 bit timestamp when we started sending */
452	struct tcp_sendfile_track *rc_last_sft;
453	uint32_t lt_seq;	/* Seq at start of lt_bw gauge */
454	int32_t rc_rtt_diff;		/* Timely style rtt diff of our gp_srtt */
455	uint64_t last_sndbytes;
456	uint64_t last_snd_rxt_bytes;
457	uint64_t rxt_threshold;
458	uint64_t last_tmit_time_acked;	/* Holds the last cumack point's last send time */
459	uint32_t last_rnd_rxt_clamped;
460	uint32_t num_of_clamps_applied;
461	uint32_t clamp_options;
462	uint32_t max_clamps;
463
464	uint32_t rc_gp_srtt;		/* Current GP srtt */
465	uint32_t rc_prev_gp_srtt;	/* Previous RTT */
466	uint32_t rc_entry_gp_rtt;	/* Entry to PRTT gp-rtt */
467	uint32_t rc_loss_at_start;	/* At measurement window where was our lost value */
468
469	uint32_t dsack_round_end;	/* In a round of seeing a DSACK */
470	uint32_t current_round;		/* Starting at zero */
471	uint32_t roundends;		/* acked value above which round ends */
472	uint32_t num_dsack;		/* Count of dsack's seen  (1 per window)*/
473	uint32_t forced_ack_ts;
474 	uint32_t last_collapse_point;	/* Last point peer collapsed too */
475	uint32_t high_collapse_point;
476	uint32_t rc_lower_rtt_us_cts;	/* Time our GP rtt was last lowered */
477	uint32_t rc_time_probertt_entered;
478	uint32_t rc_time_probertt_starts;
479	uint32_t rc_lowest_us_rtt;
480	uint32_t rc_highest_us_rtt;
481	uint32_t rc_last_us_rtt;
482	uint32_t rc_time_of_last_probertt;
483	uint32_t rc_target_probertt_flight;
484	uint32_t rc_probertt_sndmax_atexit;	/* Highest sent to in probe-rtt */
485	uint32_t rc_cwnd_at_erec;
486	uint32_t rc_ssthresh_at_erec;
487	uint32_t dsack_byte_cnt;
488	uint32_t retran_during_recovery;
489	uint32_t rc_gp_lowrtt;			/* Lowest rtt seen during GPUT measurement */
490	uint32_t rc_gp_high_rwnd;		/* Highest rwnd seen during GPUT measurement */
491	uint32_t rc_snd_max_at_rto;	/* For non-sack when the RTO occurred what was snd-max */
492	uint32_t rc_out_at_rto;
493	int32_t rc_scw_index;
494	uint32_t rc_tlp_threshold;	/* Socket option value Lock(a) */
495	uint32_t rc_last_timeout_snduna;
496	uint32_t last_tlp_acked_start;
497	uint32_t last_tlp_acked_end;
498	uint32_t challenge_ack_ts;
499	uint32_t challenge_ack_cnt;
500	uint32_t rc_min_to;	/* Socket option value Lock(a) */
501	uint32_t rc_pkt_delay;	/* Socket option value Lock(a) */
502	uint32_t persist_lost_ends;
503	uint32_t ack_during_sd;
504	uint32_t input_pkt;
505	uint32_t saved_input_pkt;
506	uint32_t saved_rxt_clamp_val; 	/* The encoded value we used to setup clamping */
507	struct newreno rc_saved_beta;	/*
508					 * For newreno cc:
509					 * rc_saved_cc are the values we have had
510					 * set by the user, if pacing is not happening
511					 * (i.e. its early and we have not turned on yet
512					 *  or it was turned off). The minute pacing
513					 * is turned on we pull out the values currently
514					 * being used by newreno and replace them with
515					 * these values, then save off the old values here,
516					 * we also set the flag (if ecn_beta is set) to make
517					 * new_reno do less of a backoff for ecn (think abe).
518					 */
519	uint16_t rc_early_recovery_segs;	/* Socket option value Lock(a) */
520	uint16_t rc_reorder_shift;	/* Socket option value Lock(a) */
521	uint8_t rack_per_upper_bound_ss;
522	uint8_t rack_per_upper_bound_ca;
523	uint8_t dsack_persist;
524	uint8_t rc_no_push_at_mrtt;	/* No push when we exceed max rtt */
525	uint8_t num_measurements;	/* Number of measurements (up to 0xff, we freeze at 0xff)  */
526	uint8_t req_measurements;	/* How many measurements are required? */
527	uint8_t saved_hibeta;
528	uint8_t rc_tlp_cwnd_reduce;	/* Socket option value Lock(a) */
529	uint8_t rc_prr_sendalot;/* Socket option value Lock(a) */
530	uint8_t rc_rate_sample_method;
531	uint8_t rc_dgp_bl_agg;		/* Buffer Level aggression during DGP */
532	uint8_t full_dgp_in_rec;	/* Flag to say if we do full DGP in recovery */
533	uint8_t client_suggested_maxseg;	/* Not sure what to do with this yet */
534	uint8_t pacing_discount_amm;	/*
535					 * This is a multipler to the base discount that
536					 * can be used to increase the discount.
537					 */
538	uint8_t already_had_a_excess;
539};
540#endif
541
542/* DGP with no buffer level mitigations */
543#define DGP_LEVEL0	0
544
545/*
546 * DGP with buffer level mitigation where BL:4 caps fillcw and BL:5
547 * turns off fillcw.
548 */
549#define DGP_LEVEL1	1
550
551/*
552 * DGP with buffer level mitigation where BL:3 caps fillcw and BL:4 turns off fillcw
553 * and BL:5 reduces by 10%
554 */
555#define DGP_LEVEL2	2
556
557/*
558 * DGP with buffer level mitigation where BL:2 caps fillcw and BL:3 turns off
559 * fillcw  BL:4 reduces by 10% and BL:5 reduces by 20%
560 */
561#define DGP_LEVEL3	3
562
563/* Hybrid pacing log defines */
564#define HYBRID_LOG_NO_ROOM	0	/* No room for the clients request */
565#define HYBRID_LOG_TURNED_OFF	1	/* Turned off hybrid pacing */
566#define HYBRID_LOG_NO_PACING	2	/* Failed to set pacing on */
567#define HYBRID_LOG_RULES_SET	3	/* Hybrid pacing for this chunk is set */
568#define HYBRID_LOG_NO_RANGE	4	/* In DGP mode, no range found */
569#define HYBRID_LOG_RULES_APP	5	/* The specified rules were applied */
570#define HYBRID_LOG_REQ_COMP	6	/* The request completed */
571#define HYBRID_LOG_BW_MEASURE	7	/* Follow up b/w measurements to the previous completed log */
572#define HYBRID_LOG_RATE_CAP	8	/* We had a rate cap apply */
573#define HYBRID_LOG_CAP_CALC	9	/* How we calculate the cap */
574#define HYBRID_LOG_ISSAME	10	/* Same as before  -- temp */
575#define HYBRID_LOG_ALLSENT	11	/* We sent it all no more rate-cap */
576#define HYBRID_LOG_OUTOFTIME	12	/* We are past the deadline DGP */
577#define HYBRID_LOG_CAPERROR	13	/* Hit one of the TSNH cases */
578#define HYBRID_LOG_EXTEND	14	/* We extended the end */
579#define HYBRID_LOG_SENT_LOST	15	/* A closing sent/lost report */
580
581#define RACK_TIMELY_CNT_BOOST 5	/* At 5th increase boost */
582#define RACK_MINRTT_FILTER_TIM 10 /* Seconds */
583
584#define RACK_HYSTART_OFF	0
585#define RACK_HYSTART_ON		1	/* hystart++ on */
586#define RACK_HYSTART_ON_W_SC	2	/* hystart++ on +Slam Cwnd */
587#define RACK_HYSTART_ON_W_SC_C	3	/* hystart++ on,
588					 * Conservative ssthresh and
589					 * +Slam cwnd
590					 */
591
592#define MAX_USER_SET_SEG 0x3f	/* The max we can set is 63 which is probably too many */
593
594#ifdef _KERNEL
595
596struct tcp_rack {
597	/* First cache line 0x00 */
598	TAILQ_ENTRY(tcp_rack) r_hpts;	/* hptsi queue next Lock(b) */
599	int32_t(*r_substate) (struct mbuf *, struct tcphdr *,
600	    struct socket *, struct tcpcb *, struct tcpopt *,
601	    int32_t, int32_t, uint32_t, int, int, uint8_t);	/* Lock(a) */
602	struct tcpcb *rc_tp;	/* The tcpcb Lock(a) */
603	struct inpcb *rc_inp;	/* The inpcb Lock(a) */
604	uint8_t rc_free_cnt;	/* Number of free entries on the rc_free list
605				 * Lock(a) */
606	uint8_t client_bufferlvl : 3, /* Expected range [0,5]: 0=unset, 1=low/empty */
607		rack_deferred_inited : 1,
608	        /* ******************************************************************** */
609	        /* Note for details of next two fields see rack_init_retransmit_rate()  */
610	        /* ******************************************************************** */
611		full_size_rxt: 1,
612		shape_rxt_to_pacing_min : 1,
613	        /* ******************************************************************** */
614		rc_ack_required: 1,
615		r_pacing_discount : 1;
616	uint8_t no_prr_addback : 1,
617		gp_ready : 1,
618		defer_options: 1,
619		excess_rxt_on: 1,	/* Are actions on for excess retransmissions? */
620		rc_ack_can_sendout_data: 1, /*
621					     * If set it will override pacing restrictions on not sending
622					     * data when the pacing timer is running. I.e. you set this
623					     * and an ACK will send data. Default is off and its only used
624					     * without pacing when we are doing 5G speed up for there
625					     * ack filtering.
626					     */
627		rc_pacing_cc_set: 1,	     /*
628					      * If we are pacing (pace_always=1) and we have reached the
629					      * point where we start pacing (fixed or gp has reached its
630					      * magic gp_ready state) this flag indicates we have set in
631					      * values to effect CC's backoff's. If pacing is turned off
632					      * then we must restore the values saved in rc_saved_beta,
633					      * if its going to gp_ready we need to copy the values into
634					      * the CC module and set our flags.
635					      *
636					      * Note this only happens if the cc name is newreno (CCALGONAME_NEWRENO).
637					      */
638
639		rc_rack_tmr_std_based :1,
640		rc_rack_use_dsack: 1;
641	uint8_t rc_dsack_round_seen: 1,
642		rc_last_tlp_acked_set: 1,
643		rc_last_tlp_past_cumack: 1,
644		rc_last_sent_tlp_seq_valid: 1,
645		rc_last_sent_tlp_past_cumack: 1,
646		probe_not_answered: 1,
647		rack_hibeta : 1,
648		lt_bw_up : 1;
649	uint32_t rc_rack_rtt;	/* RACK-RTT Lock(a) */
650	uint16_t r_mbuf_queue : 1,	/* Do we do mbuf queue for non-paced */
651		 rtt_limit_mul : 4,	/* muliply this by low rtt */
652		 r_limit_scw : 1,
653		 r_must_retran : 1,	/* For non-sack customers we hit an RTO and new data should be resends */
654		 r_use_cmp_ack: 1,	/* Do we use compressed acks */
655		 r_ent_rec_ns: 1,	/* We entered recovery and have not sent */
656		 r_might_revert: 1,	/* Flag to find out if we might need to revert */
657		 r_fast_output: 1, 	/* Fast output is in progress we can skip the bulk of rack_output */
658		 r_fsb_inited: 1,
659		 r_rack_hw_rate_caps: 1,
660		 r_up_only: 1,
661		 r_via_fill_cw : 1,
662		 r_fill_less_agg : 1;
663
664	uint8_t rc_user_set_max_segs : 7,	/* Socket option value Lock(a) */
665		rc_fillcw_apply_discount;
666	uint8_t rc_labc;		/* Appropriate Byte Counting Value */
667	uint16_t forced_ack : 1,
668		rc_gp_incr : 1,
669		rc_gp_bwred : 1,
670		rc_gp_timely_inc_cnt : 3,
671		rc_gp_timely_dec_cnt : 3,
672		r_use_labc_for_rec: 1,
673		rc_highly_buffered: 1,		/* The path is highly buffered */
674		rc_dragged_bottom: 1,
675		rc_pace_dnd : 1,		/* The pace do not disturb bit */
676		rc_avali2 : 1,
677		rc_gp_filled : 1,
678		rc_hw_nobuf : 1;
679	uint8_t r_state : 4, 	/* Current rack state Lock(a) */
680		rc_catch_up : 1,	/* catch up mode in dgp */
681		rc_hybrid_mode : 1,	/* We are in hybrid mode */
682		rc_suspicious : 1,	/* Suspect sacks have been given */
683		rc_new_rnd_needed: 1;
684	uint8_t rc_tmr_stopped : 7,
685		t_timers_stopped : 1;
686	uint8_t rc_enobuf : 7,	/* count of enobufs on connection provides */
687		rc_on_min_to : 1;
688	uint8_t r_timer_override : 1,	/* hpts override Lock(a) */
689		r_is_v6 : 1,	/* V6 pcb Lock(a)  */
690		rc_in_persist : 1,
691		rc_tlp_in_progress : 1,
692		rc_always_pace : 1,	/* Socket option value Lock(a) */
693		rc_pace_to_cwnd : 1,
694		rc_pace_fill_if_rttin_range : 1,
695		rc_srtt_measure_made : 1;
696	uint8_t app_limited_needs_set : 1,
697		use_fixed_rate : 1,
698		rc_has_collapsed : 1,
699		r_cwnd_was_clamped : 1,
700		r_clamped_gets_lower : 1,
701		rack_hdrw_pacing : 1,  /* We are doing Hardware pacing */
702		rack_hdw_pace_ena : 1, /* Is hardware pacing enabled? */
703		rack_attempt_hdwr_pace : 1; /* Did we attempt hdwr pacing (if allowed) */
704	uint8_t rack_tlp_threshold_use : 3,	/* only 1, 2 and 3 used so far */
705		rack_rec_nonrxt_use_cr : 1,
706		rack_enable_scwnd : 1,
707		rack_attempted_scwnd : 1,
708		rack_no_prr : 1,
709		rack_scwnd_is_idle : 1;
710	uint8_t rc_allow_data_af_clo: 1,
711		delayed_ack : 1,
712		set_pacing_done_a_iw : 1,
713		use_rack_rr : 1,
714		alloc_limit_reported : 1,
715		sack_attack_disable : 1,
716		do_detection : 1,
717		rc_force_max_seg : 1;
718	uint8_t r_early : 1,
719		r_late : 1,
720		r_wanted_output: 1,
721		r_rr_config : 2,
722		r_persist_lt_bw_off : 1,
723		r_collapse_point_valid : 1,
724		dgp_on : 1;
725	uint16_t rc_init_win : 8,
726		rc_gp_rtt_set : 1,
727		rc_gp_dyn_mul : 1,
728		rc_gp_saw_rec : 1,
729		rc_gp_saw_ca : 1,
730		rc_gp_saw_ss : 1,
731		rc_gp_no_rec_chg : 1,
732		in_probe_rtt : 1,
733		measure_saw_probe_rtt : 1;
734	/* Cache line 2 0x40 */
735	struct rack_control r_ctl;
736}        __aligned(CACHE_LINE_SIZE);
737
738#endif
739#endif