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 * $FreeBSD$ 26 */ 27 28 #ifndef _NETINET_TCP_RACK_H_ 29 #define _NETINET_TCP_RACK_H_ 30 31 #define RACK_ACKED 0x000001/* The remote endpoint acked this */ 32 #define RACK_TO_REXT 0x000002/* A timeout occured on this sendmap entry */ 33 #define RACK_DEFERRED 0x000004/* We can't use this for RTT calc - not used */ 34 #define RACK_OVERMAX 0x000008/* We have more retran's then we can fit */ 35 #define RACK_SACK_PASSED 0x000010/* A sack was done above this block */ 36 #define RACK_WAS_SACKPASS 0x000020/* We retransmitted due to SACK pass */ 37 #define RACK_HAS_FIN 0x000040/* segment is sent with fin */ 38 #define RACK_TLP 0x000080/* segment sent as tail-loss-probe */ 39 #define RACK_RWND_COLLAPSED 0x000100/* The peer collapsed the rwnd on the segment */ 40 #define RACK_APP_LIMITED 0x000200/* We went app limited after this send */ 41 #define RACK_WAS_ACKED 0x000400/* a RTO undid the ack, but it already had a rtt calc done */ 42 #define RACK_HAS_SYN 0x000800/* SYN is on this guy */ 43 #define RACK_SENT_W_DSACK 0x001000/* Sent with a dsack */ 44 #define RACK_SENT_SP 0x002000/* sent in slow path */ 45 #define RACK_SENT_FP 0x004000/* sent in fast path */ 46 #define RACK_HAD_PUSH 0x008000/* Push was sent on original send */ 47 #define RACK_MUST_RXT 0x010000/* We must retransmit this rsm (non-sack/mtu chg)*/ 48 #define RACK_NUM_OF_RETRANS 3 49 50 #define RACK_INITIAL_RTO 1000000 /* 1 second in microseconds */ 51 52 #define RACK_REQ_AVG 3 /* Must be less than 256 */ 53 54 struct rack_sendmap { 55 TAILQ_ENTRY(rack_sendmap) r_tnext; /* Time of transmit based next */ 56 uint32_t r_start; /* Sequence number of the segment */ 57 uint32_t r_end; /* End seq, this is 1 beyond actually */ 58 uint32_t r_rtr_bytes; /* How many bytes have been retransmitted */ 59 uint32_t r_flags : 24, /* Flags as defined above */ 60 r_rtr_cnt : 8; /* Retran count, index this -1 to get time */ 61 struct mbuf *m; 62 uint32_t soff; 63 uint32_t orig_m_len; 64 uint32_t r_nseq_appl; /* If this one is app limited, this is the nxt seq limited */ 65 uint8_t r_dupack; /* Dup ack count */ 66 uint8_t r_in_tmap; /* Flag to see if its in the r_tnext array */ 67 uint8_t r_limit_type; /* is this entry counted against a limit? */ 68 uint8_t r_just_ret : 1, /* After sending, the next pkt was just returned, i.e. limited */ 69 r_one_out_nr : 1, /* Special case 1 outstanding and not in recovery */ 70 r_no_rtt_allowed : 1, /* No rtt measurement allowed */ 71 r_hw_tls : 1, 72 r_avail : 4; 73 uint64_t r_tim_lastsent[RACK_NUM_OF_RETRANS]; 74 uint64_t r_ack_arrival; /* This is the time of ack-arrival (if SACK'd) */ 75 RB_ENTRY(rack_sendmap) r_next; /* RB Tree next */ 76 uint32_t r_fas; /* Flight at send */ 77 }; 78 79 struct deferred_opt_list { 80 TAILQ_ENTRY(deferred_opt_list) next; 81 int optname; 82 uint64_t optval; 83 }; 84 85 /* 86 * Timestamps in the rack sendmap are now moving to be 87 * uint64_t's. This means that if you want a uint32_t 88 * usec timestamp (the old usecond timestamp) you simply have 89 * to cast it to uint32_t. The reason we do this is not for 90 * wrap, but we need to get back, at times, to the millisecond 91 * timestamp that is used in the TSTMP option. To do this we 92 * can use the rack_ts_to_msec() inline below which can take 93 * the 64bit ts and make into the correct timestamp millisecond 94 * wise. Thats not possible with the 32bit usecond timestamp since 95 * the seconds wrap too quickly to cover all bases. 96 * 97 * There are quite a few places in rack where I simply cast 98 * back to uint32_t and then end up using the TSTMP_XX() 99 * macros. This is ok, but we could do simple compares if 100 * we ever decided to move all of those variables to 64 bits 101 * as well. 102 */ 103 104 inline uint64_t 105 rack_to_usec_ts(struct timeval *tv) 106 { 107 return ((tv->tv_sec * HPTS_USEC_IN_SEC) + tv->tv_usec); 108 } 109 110 inline uint32_t 111 rack_ts_to_msec(uint64_t ts) 112 { 113 return((uint32_t)(ts / HPTS_MSEC_IN_SEC)); 114 } 115 116 117 RB_HEAD(rack_rb_tree_head, rack_sendmap); 118 TAILQ_HEAD(rack_head, rack_sendmap); 119 TAILQ_HEAD(def_opt_head, deferred_opt_list); 120 121 /* Map change logging */ 122 #define MAP_MERGE 0x01 123 #define MAP_SPLIT 0x02 124 #define MAP_NEW 0x03 125 #define MAP_SACK_M1 0x04 126 #define MAP_SACK_M2 0x05 127 #define MAP_SACK_M3 0x06 128 #define MAP_SACK_M4 0x07 129 #define MAP_SACK_M5 0x08 130 #define MAP_FREE 0x09 131 #define MAP_TRIM_HEAD 0x0a 132 133 #define RACK_LIMIT_TYPE_SPLIT 1 134 135 /* 136 * We use the rate sample structure to 137 * assist in single sack/ack rate and rtt 138 * calculation. In the future we will expand 139 * this in BBR to do forward rate sample 140 * b/w estimation. 141 */ 142 #define RACK_RTT_EMPTY 0x00000001 /* Nothing yet stored in RTT's */ 143 #define RACK_RTT_VALID 0x00000002 /* We have at least one valid RTT */ 144 struct rack_rtt_sample { 145 uint32_t rs_flags; 146 uint32_t rs_rtt_lowest; 147 uint32_t rs_rtt_highest; 148 uint32_t rs_rtt_cnt; 149 uint32_t rs_us_rtt; 150 int32_t confidence; 151 uint64_t rs_rtt_tot; 152 uint16_t rs_us_rtrcnt; 153 }; 154 155 #define RACK_LOG_TYPE_ACK 0x01 156 #define RACK_LOG_TYPE_OUT 0x02 157 #define RACK_LOG_TYPE_TO 0x03 158 #define RACK_LOG_TYPE_ALLOC 0x04 159 #define RACK_LOG_TYPE_FREE 0x05 160 161 /* 162 * Magic numbers for logging timeout events if the 163 * logging is enabled. 164 */ 165 #define RACK_TO_FRM_TMR 1 166 #define RACK_TO_FRM_TLP 2 167 #define RACK_TO_FRM_RACK 3 168 #define RACK_TO_FRM_KEEP 4 169 #define RACK_TO_FRM_PERSIST 5 170 #define RACK_TO_FRM_DELACK 6 171 172 struct rack_opts_stats { 173 uint64_t tcp_rack_tlp_reduce; 174 uint64_t tcp_rack_pace_always; 175 uint64_t tcp_rack_pace_reduce; 176 uint64_t tcp_rack_max_seg; 177 uint64_t tcp_rack_prr_sendalot; 178 uint64_t tcp_rack_min_to; 179 uint64_t tcp_rack_early_seg; 180 uint64_t tcp_rack_reord_thresh; 181 uint64_t tcp_rack_reord_fade; 182 uint64_t tcp_rack_tlp_thresh; 183 uint64_t tcp_rack_pkt_delay; 184 uint64_t tcp_rack_tlp_inc_var; 185 uint64_t tcp_tlp_use; 186 uint64_t tcp_rack_idle_reduce; 187 uint64_t tcp_rack_idle_reduce_high; 188 uint64_t rack_no_timer_in_hpts; 189 uint64_t tcp_rack_min_pace_seg; 190 uint64_t tcp_rack_pace_rate_ca; 191 uint64_t tcp_rack_rr; 192 uint64_t tcp_rack_do_detection; 193 uint64_t tcp_rack_rrr_no_conf_rate; 194 uint64_t tcp_initial_rate; 195 uint64_t tcp_initial_win; 196 uint64_t tcp_hdwr_pacing; 197 uint64_t tcp_gp_inc_ss; 198 uint64_t tcp_gp_inc_ca; 199 uint64_t tcp_gp_inc_rec; 200 uint64_t tcp_rack_force_max_seg; 201 uint64_t tcp_rack_pace_rate_ss; 202 uint64_t tcp_rack_pace_rate_rec; 203 /* Temp counters for dsack */ 204 uint64_t tcp_sack_path_1; 205 uint64_t tcp_sack_path_2a; 206 uint64_t tcp_sack_path_2b; 207 uint64_t tcp_sack_path_3; 208 uint64_t tcp_sack_path_4; 209 /* non temp counters */ 210 uint64_t tcp_rack_scwnd; 211 uint64_t tcp_rack_noprr; 212 uint64_t tcp_rack_cfg_rate; 213 uint64_t tcp_timely_dyn; 214 uint64_t tcp_rack_mbufq; 215 uint64_t tcp_fillcw; 216 uint64_t tcp_npush; 217 uint64_t tcp_lscwnd; 218 uint64_t tcp_profile; 219 uint64_t tcp_hdwr_rate_cap; 220 uint64_t tcp_pacing_rate_cap; 221 uint64_t tcp_pacing_up_only; 222 uint64_t tcp_use_cmp_acks; 223 uint64_t tcp_rack_abc_val; 224 uint64_t tcp_rec_abc_val; 225 uint64_t tcp_rack_measure_cnt; 226 uint64_t tcp_rack_delayed_ack; 227 uint64_t tcp_rack_rtt_use; 228 uint64_t tcp_data_after_close; 229 uint64_t tcp_defer_opt; 230 uint64_t tcp_rack_fastrsm_hack; 231 uint64_t tcp_rack_beta; 232 uint64_t tcp_rack_beta_ecn; 233 uint64_t tcp_rack_timer_slop; 234 uint64_t tcp_rack_dsack_opt; 235 }; 236 237 /* RTT shrink reasons */ 238 #define RACK_RTTS_INIT 0 239 #define RACK_RTTS_NEWRTT 1 240 #define RACK_RTTS_EXITPROBE 2 241 #define RACK_RTTS_ENTERPROBE 3 242 #define RACK_RTTS_REACHTARGET 4 243 #define RACK_RTTS_SEEHBP 5 244 #define RACK_RTTS_NOBACKOFF 6 245 #define RACK_RTTS_SAFETY 7 246 247 #define RACK_USE_BEG 1 248 #define RACK_USE_END 2 249 #define RACK_USE_END_OR_THACK 3 250 251 #define TLP_USE_ID 1 /* Internet draft behavior */ 252 #define TLP_USE_TWO_ONE 2 /* Use 2.1 behavior */ 253 #define TLP_USE_TWO_TWO 3 /* Use 2.2 behavior */ 254 #define RACK_MIN_BW 8000 /* 64kbps in Bps */ 255 256 /* Rack quality indicators for GPUT measurements */ 257 #define RACK_QUALITY_NONE 0 /* No quality stated */ 258 #define RACK_QUALITY_HIGH 1 /* A normal measurement of a GP RTT */ 259 #define RACK_QUALITY_APPLIMITED 2 /* An app limited case that may be of lower quality */ 260 #define RACK_QUALITY_PERSIST 3 /* A measurement where we went into persists */ 261 #define RACK_QUALITY_PROBERTT 4 /* A measurement where we went into or exited probe RTT */ 262 #define RACK_QUALITY_ALLACKED 5 /* All data is now acknowledged */ 263 264 265 #define MIN_GP_WIN 6 /* We need at least 6 MSS in a GP measurement */ 266 #ifdef _KERNEL 267 #define RACK_OPTS_SIZE (sizeof(struct rack_opts_stats)/sizeof(uint64_t)) 268 extern counter_u64_t rack_opts_arry[RACK_OPTS_SIZE]; 269 #define RACK_OPTS_ADD(name, amm) counter_u64_add(rack_opts_arry[(offsetof(struct rack_opts_stats, name)/sizeof(uint64_t))], (amm)) 270 #define RACK_OPTS_INC(name) RACK_OPTS_ADD(name, 1) 271 #endif 272 /* 273 * As we get each SACK we wade through the 274 * rc_map and mark off what is acked. 275 * We also increment rc_sacked as well. 276 * 277 * We also pay attention to missing entries 278 * based on the time and possibly mark them 279 * for retransmit. If we do and we are not already 280 * in recovery we enter recovery. In doing 281 * so we claer prr_delivered/holes_rxt and prr_sent_dur_rec. 282 * We also setup rc_next/rc_snd_nxt/rc_send_end so 283 * we will know where to send from. When not in 284 * recovery rc_next will be NULL and rc_snd_nxt should 285 * equal snd_max. 286 * 287 * Whenever we retransmit from recovery we increment 288 * rc_holes_rxt as we retran a block and mark it as retransmitted 289 * with the time it was sent. During non-recovery sending we 290 * add to our map and note the time down of any send expanding 291 * the rc_map at the tail and moving rc_snd_nxt up with snd_max. 292 * 293 * In recovery during SACK/ACK processing if a chunk has 294 * been retransmitted and it is now acked, we decrement rc_holes_rxt. 295 * When we retransmit from the scoreboard we use 296 * rc_next and rc_snd_nxt/rc_send_end to help us 297 * find what needs to be retran. 298 * 299 * To calculate pipe we simply take (snd_max - snd_una) + rc_holes_rxt 300 * This gets us the effect of RFC6675 pipe, counting twice for 301 * bytes retransmitted. 302 */ 303 304 #define TT_RACK_FR_TMR 0x2000 305 306 /* 307 * Locking for the rack control block. 308 * a) Locked by INP_WLOCK 309 * b) Locked by the hpts-mutex 310 * 311 */ 312 #define RACK_GP_HIST 4 /* How much goodput history do we maintain? */ 313 314 #define RACK_NUM_FSB_DEBUG 16 315 #ifdef _KERNEL 316 struct rack_fast_send_blk { 317 uint32_t left_to_send; 318 uint16_t tcp_ip_hdr_len; 319 uint8_t tcp_flags; 320 uint8_t hoplimit; 321 uint8_t *tcp_ip_hdr; 322 uint32_t recwin; 323 uint32_t off; 324 struct tcphdr *th; 325 struct udphdr *udp; 326 struct mbuf *m; 327 uint32_t o_m_len; 328 uint32_t rfo_apply_push : 1, 329 hw_tls : 1, 330 unused : 30; 331 }; 332 333 struct rack_control { 334 /* Second cache line 0x40 from tcp_rack */ 335 struct rack_rb_tree_head rc_mtree; /* Tree of all segments Lock(a) */ 336 struct rack_head rc_tmap; /* List in transmit order Lock(a) */ 337 struct rack_sendmap *rc_tlpsend; /* Remembered place for 338 * tlp_sending Lock(a) */ 339 struct rack_sendmap *rc_resend; /* something we have been asked to 340 * resend */ 341 struct rack_fast_send_blk fsb; /* The fast-send block */ 342 uint32_t timer_slop; 343 uint32_t input_pkt; 344 uint32_t saved_input_pkt; 345 uint32_t rc_hpts_flags; 346 uint32_t rc_fixed_pacing_rate_ca; 347 uint32_t rc_fixed_pacing_rate_rec; 348 uint32_t rc_fixed_pacing_rate_ss; 349 uint32_t cwnd_to_use; /* The cwnd in use */ 350 uint32_t rc_timer_exp; /* If a timer ticks of expiry */ 351 uint32_t rc_rack_min_rtt; /* lowest RTT seen Lock(a) */ 352 uint32_t rc_rack_largest_cwnd; /* Largest CWND we have seen Lock(a) */ 353 354 /* Third Cache line 0x80 */ 355 struct rack_head rc_free; /* Allocation array */ 356 uint64_t last_hw_bw_req; 357 uint64_t crte_prev_rate; 358 uint64_t bw_rate_cap; 359 uint32_t rc_reorder_ts; /* Last time we saw reordering Lock(a) */ 360 361 uint32_t rc_tlp_new_data; /* we need to send new-data on a TLP 362 * Lock(a) */ 363 uint32_t rc_prr_out; /* bytes sent during recovery Lock(a) */ 364 365 uint32_t rc_prr_recovery_fs; /* recovery fs point Lock(a) */ 366 367 uint32_t rc_prr_sndcnt; /* Prr sndcnt Lock(a) */ 368 369 uint32_t rc_sacked; /* Tot sacked on scoreboard Lock(a) */ 370 uint32_t last_sent_tlp_seq; /* Last tlp sequence that was retransmitted Lock(a) */ 371 372 uint32_t rc_prr_delivered; /* during recovery prr var Lock(a) */ 373 uint16_t rc_tlp_cnt_out; /* count of times we have sent a TLP without new data */ 374 uint16_t last_sent_tlp_len; /* Number of bytes in the last sent tlp */ 375 376 uint32_t rc_loss_count; /* How many bytes have been retransmitted 377 * Lock(a) */ 378 uint32_t rc_reorder_fade; /* Socket option value Lock(a) */ 379 380 /* Forth cache line 0xc0 */ 381 /* Times */ 382 383 uint32_t rc_rack_tmit_time; /* Rack transmit time Lock(a) */ 384 uint32_t rc_holes_rxt; /* Tot retraned from scoreboard Lock(a) */ 385 386 uint32_t rc_num_maps_alloced; /* Number of map blocks (sacks) we 387 * have allocated */ 388 uint32_t rc_rcvtime; /* When we last received data */ 389 uint32_t rc_num_split_allocs; /* num split map entries allocated */ 390 391 uint32_t rc_last_output_to; 392 uint32_t rc_went_idle_time; 393 394 struct rack_sendmap *rc_sacklast; /* sack remembered place 395 * Lock(a) */ 396 397 struct rack_sendmap *rc_first_appl; /* Pointer to first app limited */ 398 struct rack_sendmap *rc_end_appl; /* Pointer to last app limited */ 399 /* Cache line split 0x100 */ 400 struct sack_filter rack_sf; 401 /* Cache line split 0x140 */ 402 /* Flags for various things */ 403 uint32_t rc_pace_max_segs; 404 uint32_t rc_pace_min_segs; 405 uint32_t rc_app_limited_cnt; 406 uint16_t rack_per_of_gp_ss; /* 100 = 100%, so from 65536 = 655 x bw */ 407 uint16_t rack_per_of_gp_ca; /* 100 = 100%, so from 65536 = 655 x bw */ 408 uint16_t rack_per_of_gp_rec; /* 100 = 100%, so from 65536 = 655 x bw, 0=off */ 409 uint16_t rack_per_of_gp_probertt; /* 100 = 100%, so from 65536 = 655 x bw, 0=off */ 410 uint32_t rc_high_rwnd; 411 uint32_t ack_count; 412 uint32_t sack_count; 413 uint32_t sack_noextra_move; 414 uint32_t sack_moved_extra; 415 struct rack_rtt_sample rack_rs; 416 const struct tcp_hwrate_limit_table *crte; 417 uint32_t rc_agg_early; 418 uint32_t rc_agg_delayed; 419 uint32_t rc_tlp_rxt_last_time; 420 uint32_t rc_saved_cwnd; 421 uint64_t rc_gp_output_ts; /* chg*/ 422 uint64_t rc_gp_cumack_ts; /* chg*/ 423 struct timeval act_rcv_time; 424 struct timeval rc_last_time_decay; /* SAD time decay happened here */ 425 uint64_t gp_bw; 426 uint64_t init_rate; 427 #ifdef NETFLIX_SHARED_CWND 428 struct shared_cwnd *rc_scw; 429 #endif 430 uint64_t last_gp_comp_bw; 431 uint64_t last_max_bw; /* Our calculated max b/w last */ 432 struct time_filter_small rc_gp_min_rtt; 433 struct def_opt_head opt_list; 434 int32_t rc_rtt_diff; /* Timely style rtt diff of our gp_srtt */ 435 uint32_t rc_gp_srtt; /* Current GP srtt */ 436 uint32_t rc_prev_gp_srtt; /* Previous RTT */ 437 uint32_t rc_entry_gp_rtt; /* Entry to PRTT gp-rtt */ 438 uint32_t rc_loss_at_start; /* At measurement window where was our lost value */ 439 440 uint32_t dsack_round_end; /* In a round of seeing a DSACK */ 441 uint32_t current_round; /* Starting at zero */ 442 uint32_t roundends; /* acked value above which round ends */ 443 uint32_t num_dsack; /* Count of dsack's seen (1 per window)*/ 444 uint32_t forced_ack_ts; 445 uint32_t rc_lower_rtt_us_cts; /* Time our GP rtt was last lowered */ 446 uint32_t rc_time_probertt_entered; 447 uint32_t rc_time_probertt_starts; 448 uint32_t rc_lowest_us_rtt; 449 uint32_t rc_highest_us_rtt; 450 uint32_t rc_last_us_rtt; 451 uint32_t rc_time_of_last_probertt; 452 uint32_t rc_target_probertt_flight; 453 uint32_t rc_probertt_sndmax_atexit; /* Highest sent to in probe-rtt */ 454 uint32_t rc_cwnd_at_erec; 455 uint32_t rc_ssthresh_at_erec; 456 uint32_t dsack_byte_cnt; 457 uint32_t retran_during_recovery; 458 uint32_t rc_gp_lowrtt; /* Lowest rtt seen during GPUT measurement */ 459 uint32_t rc_gp_high_rwnd; /* Highest rwnd seen during GPUT measurement */ 460 uint32_t rc_snd_max_at_rto; /* For non-sack when the RTO occured what was snd-max */ 461 uint32_t rc_out_at_rto; 462 int32_t rc_scw_index; 463 uint32_t rc_tlp_threshold; /* Socket option value Lock(a) */ 464 uint32_t rc_last_timeout_snduna; 465 uint32_t last_tlp_acked_start; 466 uint32_t last_tlp_acked_end; 467 uint32_t challenge_ack_ts; 468 uint32_t challenge_ack_cnt; 469 uint32_t rc_min_to; /* Socket option value Lock(a) */ 470 uint32_t rc_pkt_delay; /* Socket option value Lock(a) */ 471 uint32_t persist_lost_ends; 472 struct newreno rc_saved_beta; /* 473 * For newreno cc: 474 * rc_saved_cc are the values we have had 475 * set by the user, if pacing is not happening 476 * (i.e. its early and we have not turned on yet 477 * or it was turned off). The minute pacing 478 * is turned on we pull out the values currently 479 * being used by newreno and replace them with 480 * these values, then save off the old values here, 481 * we also set the flag (if ecn_beta is set) to make 482 * new_reno do less of a backoff for ecn (think abe). 483 */ 484 uint16_t rc_early_recovery_segs; /* Socket option value Lock(a) */ 485 uint16_t rc_reorder_shift; /* Socket option value Lock(a) */ 486 uint8_t dsack_persist; 487 uint8_t rc_no_push_at_mrtt; /* No push when we exceed max rtt */ 488 uint8_t num_measurements; /* Number of measurements (up to 0xff, we freeze at 0xff) */ 489 uint8_t req_measurements; /* How many measurements are required? */ 490 uint8_t rc_tlp_cwnd_reduce; /* Socket option value Lock(a) */ 491 uint8_t rc_prr_sendalot;/* Socket option value Lock(a) */ 492 uint8_t rc_rate_sample_method; 493 }; 494 #endif 495 496 #define RACK_TIMELY_CNT_BOOST 5 /* At 5th increase boost */ 497 #define RACK_MINRTT_FILTER_TIM 10 /* Seconds */ 498 499 #define RACK_HYSTART_OFF 0 500 #define RACK_HYSTART_ON 1 /* hystart++ on */ 501 #define RACK_HYSTART_ON_W_SC 2 /* hystart++ on +Slam Cwnd */ 502 #define RACK_HYSTART_ON_W_SC_C 3 /* hystart++ on, 503 * Conservative ssthresh and 504 * +Slam cwnd 505 */ 506 507 #ifdef _KERNEL 508 509 struct tcp_rack { 510 /* First cache line 0x00 */ 511 TAILQ_ENTRY(tcp_rack) r_hpts; /* hptsi queue next Lock(b) */ 512 int32_t(*r_substate) (struct mbuf *, struct tcphdr *, 513 struct socket *, struct tcpcb *, struct tcpopt *, 514 int32_t, int32_t, uint32_t, int, int, uint8_t); /* Lock(a) */ 515 struct tcpcb *rc_tp; /* The tcpcb Lock(a) */ 516 struct inpcb *rc_inp; /* The inpcb Lock(a) */ 517 uint8_t rc_free_cnt; /* Number of free entries on the rc_free list 518 * Lock(a) */ 519 uint8_t client_bufferlvl; /* 0 - 5 normaly, less than or at 2 means its real low */ 520 uint8_t no_prr_addback : 1, 521 gp_ready : 1, 522 defer_options: 1, 523 fast_rsm_hack: 1, 524 rc_ack_can_sendout_data: 1, /* 525 * If set it will override pacing restrictions on not sending 526 * data when the pacing timer is running. I.e. you set this 527 * and an ACK will send data. Default is off and its only used 528 * without pacing when we are doing 5G speed up for there 529 * ack filtering. 530 */ 531 rc_pacing_cc_set: 1, /* 532 * If we are pacing (pace_always=1) and we have reached the 533 * point where we start pacing (fixed or gp has reached its 534 * magic gp_ready state) this flag indicates we have set in 535 * values to effect CC's backoff's. If pacing is turned off 536 * then we must restore the values saved in rc_saved_beta, 537 * if its going to gp_ready we need to copy the values into 538 * the CC module and set our flags. 539 * 540 * Note this only happens if the cc name is newreno (CCALGONAME_NEWRENO). 541 */ 542 543 rc_rack_tmr_std_based :1, 544 rc_rack_use_dsack: 1; 545 uint8_t rc_dsack_round_seen: 1, 546 rc_last_tlp_acked_set: 1, 547 rc_last_tlp_past_cumack: 1, 548 rc_last_sent_tlp_seq_valid: 1, 549 rc_last_sent_tlp_past_cumack: 1, 550 probe_not_answered: 1, 551 avail_bytes : 2; 552 uint32_t rc_rack_rtt; /* RACK-RTT Lock(a) */ 553 uint16_t r_mbuf_queue : 1, /* Do we do mbuf queue for non-paced */ 554 rtt_limit_mul : 4, /* muliply this by low rtt */ 555 r_limit_scw : 1, 556 r_must_retran : 1, /* For non-sack customers we hit an RTO and new data should be resends */ 557 r_use_cmp_ack: 1, /* Do we use compressed acks */ 558 r_ent_rec_ns: 1, /* We entered recovery and have not sent */ 559 r_might_revert: 1, /* Flag to find out if we might need to revert */ 560 r_fast_output: 1, /* Fast output is in progress we can skip the bulk of rack_output */ 561 r_fsb_inited: 1, 562 r_rack_hw_rate_caps: 1, 563 r_up_only: 1, 564 r_via_fill_cw : 1, 565 r_fill_less_agg : 1; 566 567 uint8_t rc_user_set_max_segs; /* Socket option value Lock(a) */ 568 uint8_t rc_labc; /* Appropriate Byte Counting Value */ 569 uint16_t forced_ack : 1, 570 rc_gp_incr : 1, 571 rc_gp_bwred : 1, 572 rc_gp_timely_inc_cnt : 3, 573 rc_gp_timely_dec_cnt : 3, 574 r_use_labc_for_rec: 1, 575 rc_highly_buffered: 1, /* The path is highly buffered */ 576 rc_dragged_bottom: 1, 577 rc_dack_mode : 1, /* Mac O/S emulation of d-ack */ 578 rc_dack_toggle : 1, /* For Mac O/S emulation of d-ack */ 579 rc_gp_filled : 1, 580 rc_is_spare : 1; 581 uint8_t r_state; /* Current rack state Lock(a) */ 582 uint8_t rc_tmr_stopped : 7, 583 t_timers_stopped : 1; 584 uint8_t rc_enobuf : 7, /* count of enobufs on connection provides */ 585 rc_on_min_to : 1; 586 uint8_t r_timer_override : 1, /* hpts override Lock(a) */ 587 r_is_v6 : 1, /* V6 pcb Lock(a) */ 588 rc_in_persist : 1, 589 rc_tlp_in_progress : 1, 590 rc_always_pace : 1, /* Socket option value Lock(a) */ 591 rc_pace_to_cwnd : 1, 592 rc_pace_fill_if_rttin_range : 1, 593 rc_srtt_measure_made : 1; 594 uint8_t app_limited_needs_set : 1, 595 use_fixed_rate : 1, 596 rc_has_collapsed : 1, 597 r_rep_attack : 1, 598 r_rep_reverse : 1, 599 rack_hdrw_pacing : 1, /* We are doing Hardware pacing */ 600 rack_hdw_pace_ena : 1, /* Is hardware pacing enabled? */ 601 rack_attempt_hdwr_pace : 1; /* Did we attempt hdwr pacing (if allowed) */ 602 uint8_t rack_tlp_threshold_use : 3, /* only 1, 2 and 3 used so far */ 603 rack_rec_nonrxt_use_cr : 1, 604 rack_enable_scwnd : 1, 605 rack_attempted_scwnd : 1, 606 rack_no_prr : 1, 607 rack_scwnd_is_idle : 1; 608 uint8_t rc_allow_data_af_clo: 1, 609 delayed_ack : 1, 610 set_pacing_done_a_iw : 1, 611 use_rack_rr : 1, 612 alloc_limit_reported : 1, 613 sack_attack_disable : 1, 614 do_detection : 1, 615 rc_force_max_seg : 1; 616 uint8_t r_early : 1, 617 r_late : 1, 618 r_wanted_output: 1, 619 r_rr_config : 2, 620 rc_avail_bit : 3; 621 uint16_t rc_init_win : 8, 622 rc_gp_rtt_set : 1, 623 rc_gp_dyn_mul : 1, 624 rc_gp_saw_rec : 1, 625 rc_gp_saw_ca : 1, 626 rc_gp_saw_ss : 1, 627 rc_gp_no_rec_chg : 1, 628 in_probe_rtt : 1, 629 measure_saw_probe_rtt : 1; 630 /* Cache line 2 0x40 */ 631 struct rack_control r_ctl; 632 } __aligned(CACHE_LINE_SIZE); 633 634 #endif 635 #endif 636