1 /*- 2 * Copyright (c) 2016-9 3 * Netflix Inc. All rights reserved. 4 * Author Randall R. Stewart 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 * 27 * $FreeBSD$ 28 */ 29 30 #ifndef _NETINET_TCP_BBR_H_ 31 #define _NETINET_TCP_BBR_H_ 32 33 #define BBR_INITIAL_RTO 1000000 /* 1 second in micro-seconds */ 34 /* Send map flags */ 35 #define BBR_ACKED 0x0001 /* The remote endpoint acked this */ 36 #define BBR_WAS_RENEGED 0x0002 /* The peer reneged the ack */ 37 #define BBR_RXT_CLEARED 0x0004 /* ACK Cleared by the RXT timer */ 38 #define BBR_OVERMAX 0x0008 /* We have more retran's then we can 39 * fit */ 40 #define BBR_SACK_PASSED 0x0010 /* A sack was done above this block */ 41 #define BBR_WAS_SACKPASS 0x0020 /* We retransmitted due to SACK pass */ 42 #define BBR_HAS_FIN 0x0040 /* segment is sent with fin */ 43 #define BBR_TLP 0x0080 /* segment sent as tail-loss-probe */ 44 #define BBR_HAS_SYN 0x0100 /* segment has the syn */ 45 #define BBR_MARKED_LOST 0x0200 /* 46 * This segments is lost and 47 * totaled into bbr->rc_ctl.rc_lost 48 */ 49 #define BBR_RWND_COLLAPSED 0x0400 /* The peer collapsed the rwnd on the segment */ 50 #define BBR_NUM_OF_RETRANS 7 51 52 /* Defines for socket options to set pacing overheads */ 53 #define BBR_INCL_ENET_OH 0x01 54 #define BBR_INCL_IP_OH 0x02 55 #define BBR_INCL_TCP_OH 0x03 56 57 /* 58 * With the addition of both measurement algorithms 59 * I had to move over the size of a 60 * cache line (unfortunately). For now there is 61 * no way around this. We may be able to cut back 62 * at some point I hope. 63 */ 64 struct bbr_sendmap { 65 TAILQ_ENTRY(bbr_sendmap) r_next; /* seq number arrayed next */ 66 TAILQ_ENTRY(bbr_sendmap) r_tnext; /* Time of tmit based next */ 67 uint32_t r_start; /* Sequence number of the segment */ 68 uint32_t r_end; /* End seq, this is 1 beyond actually */ 69 70 uint32_t r_rtr_bytes; /* How many bytes have been retransmitted */ 71 uint32_t r_delivered; /* Delivered amount at send */ 72 73 uint32_t r_del_time; /* The time of the last delivery update */ 74 uint8_t r_rtr_cnt:4, /* Retran count, index this -1 to get time 75 * sent */ 76 unused_bit:1, 77 r_is_drain:1, /* In a draining cycle */ 78 r_app_limited:1,/* We went app limited */ 79 r_ts_valid:1; /* Timestamp field is valid (r_del_ack_ts) */ 80 uint8_t r_dupack; /* Dup ack count */ 81 uint8_t r_in_tmap:1, /* Flag to see if its in the r_tnext array */ 82 r_is_smallmap:1,/* Was logged as a small-map send-map item */ 83 r_is_gain:1, /* Was in gain cycle */ 84 r_bbr_state:5; /* The BBR state at send */ 85 uint8_t r_limit_type; /* is this entry counted against a limit? */ 86 87 uint16_t r_flags; /* Flags as defined above */ 88 uint16_t r_spare16; 89 uint32_t r_del_ack_ts; /* At send what timestamp of peer was (if r_ts_valid set) */ 90 /****************Cache line*****************/ 91 uint32_t r_tim_lastsent[BBR_NUM_OF_RETRANS]; 92 /* 93 * Question, should we instead just grab the sending b/w 94 * from the filter with the gain and store it in a 95 * uint64_t instead? 96 */ 97 uint32_t r_first_sent_time; /* Time of first pkt in flight sent */ 98 uint32_t r_pacing_delay; /* pacing delay of this send */ 99 uint32_t r_flight_at_send; /* flight at the time of the send */ 100 #ifdef _KERNEL 101 } __aligned(CACHE_LINE_SIZE); 102 #else 103 }; 104 #endif 105 #define BBR_LIMIT_TYPE_SPLIT 1 106 107 TAILQ_HEAD(bbr_head, bbr_sendmap); 108 109 #define BBR_SEGMENT_TIME_SIZE 1500 /* How many bytes in time_between */ 110 111 #define BBR_MIN_SEG 1460 /* MSS size */ 112 #define BBR_MAX_GAIN_VALUE 0xffff 113 114 #define BBR_TIMER_FUDGE 1500 /* 1.5ms in micro seconds */ 115 116 /* BW twiddle secret codes */ 117 #define BBR_RED_BW_CONGSIG 0 /* We enter recovery and set using b/w */ 118 #define BBR_RED_BW_RATECAL 1 /* We are calculating the loss rate */ 119 #define BBR_RED_BW_USELRBW 2 /* We are dropping the lower b/w with 120 * cDR */ 121 #define BBR_RED_BW_SETHIGHLOSS 3 /* We have set our highloss value at 122 * exit from probe-rtt */ 123 #define BBR_RED_BW_PE_CLREARLY 4 /* We have decided to clear the 124 * reduction early */ 125 #define BBR_RED_BW_PE_CLAFDEL 5 /* We are clearing it on schedule 126 * delayed */ 127 #define BBR_RED_BW_REC_ENDCLL 6 /* Recover exits save high if needed 128 * an clear to start measuring */ 129 #define BBR_RED_BW_PE_NOEARLY_OUT 7 /* Set pkt epoch judged that we do not 130 * get out of jail early */ 131 /* For calculating a rate */ 132 #define BBR_CALC_BW 1 133 #define BBR_CALC_LOSS 2 134 135 #define BBR_RTT_BY_TIMESTAMP 0 136 #define BBR_RTT_BY_EXACTMATCH 1 137 #define BBR_RTT_BY_EARLIER_RET 2 138 #define BBR_RTT_BY_THIS_RETRAN 3 139 #define BBR_RTT_BY_SOME_RETRAN 4 140 #define BBR_RTT_BY_TSMATCHING 5 141 142 /* Markers to track where we enter persists from */ 143 #define BBR_PERSISTS_FROM_1 1 144 #define BBR_PERSISTS_FROM_2 2 145 #define BBR_PERSISTS_FROM_3 3 146 #define BBR_PERSISTS_FROM_4 4 147 #define BBR_PERSISTS_FROM_5 5 148 149 /* magic cookies to ask for the RTT */ 150 #define BBR_RTT_PROP 0 151 #define BBR_RTT_RACK 1 152 #define BBR_RTT_PKTRTT 2 153 #define BBR_SRTT 3 154 155 #define BBR_SACKED 0 156 #define BBR_CUM_ACKED 1 157 158 /* threshold in useconds where we consider we need a higher min cwnd */ 159 #define BBR_HIGH_SPEED 1000 160 #define BBR_HIGHSPEED_NUM_MSS 12 161 162 #define MAX_REDUCE_RXT 3 /* What is the maximum times we are willing to 163 * reduce b/w in RTX's. Setting this has a 164 * multiplicative effect e.g. if we are 165 * reducing by 20% then setting it to 3 means 166 * you will have reduced the b/w estimate by > 167 * 60% before you stop. */ 168 /* 169 * We use the rate sample structure to 170 * assist in single sack/ack rate and rtt 171 * calculation. In the future we will expand 172 * this in BBR to do forward rate sample 173 * b/w estimation. 174 */ 175 #define BBR_RS_RTT_EMPTY 0x00000001 /* Nothing yet stored in RTT's */ 176 #define BBR_RS_BW_EMPTY 0x00000002 /* Nothing yet stored in cDR */ 177 #define BBR_RS_RTT_VALID 0x00000004 /* We have at least one valid RTT */ 178 #define BBR_RS_BW_VAILD 0x00000008 /* We have a valid cDR */ 179 #define BBR_RS_EMPTY (BBR_RS_RTT_EMPTY|BBR_RS_BW_EMPTY) 180 struct bbr_rtt_sample { 181 uint32_t rs_flags; 182 uint32_t rs_rtt_lowest; 183 uint32_t rs_rtt_lowest_sendtime; 184 uint32_t rs_rtt_low_seq_start; 185 186 uint32_t rs_rtt_highest; 187 uint32_t rs_rtt_cnt; 188 189 uint64_t rs_rtt_tot; 190 uint32_t cur_rtt; 191 uint32_t cur_rtt_bytecnt; 192 193 uint32_t cur_rtt_rsmcnt; 194 uint32_t rc_crtt_set:1, 195 avail_bits:31; 196 uint64_t rs_cDR; 197 }; 198 199 /* RTT shrink reasons */ 200 #define BBR_RTTS_INIT 0 201 #define BBR_RTTS_NEWRTT 1 202 #define BBR_RTTS_RTTPROBE 2 203 #define BBR_RTTS_WASIDLE 3 204 #define BBR_RTTS_PERSIST 4 205 #define BBR_RTTS_REACHTAR 5 206 #define BBR_RTTS_ENTERPROBE 6 207 #define BBR_RTTS_SHRINK_PG 7 208 #define BBR_RTTS_SHRINK_PG_FINAL 8 209 #define BBR_RTTS_NEW_TARGET 9 210 #define BBR_RTTS_LEAVE_DRAIN 10 211 #define BBR_RTTS_RESETS_VALUES 11 212 213 #define BBR_NUM_RATES 5 214 /* Rate flags */ 215 #define BBR_RT_FLAG_FREE 0x00 /* Is on the free list */ 216 #define BBR_RT_FLAG_INUSE 0x01 /* Has been allocated */ 217 #define BBR_RT_FLAG_READY 0x02 /* Ready to initiate a measurement. */ 218 #define BBR_RT_FLAG_CAPPED_PRE 0x04 /* Ready to cap if we send the next segment */ 219 #define BBR_RT_FLAG_CAPPED 0x08 /* Measurement is capped */ 220 #define BBR_RT_FLAG_PASTFA 0x10 /* Past the first ack. */ 221 #define BBR_RT_FLAG_LIMITED 0x20 /* Saw application/cwnd or rwnd limited period */ 222 #define BBR_RT_SEEN_A_ACK 0x40 /* A ack has been saved */ 223 #define BBR_RT_PREV_RTT_SET 0x80 /* There was a RTT set in */ 224 #define BBR_RT_PREV_SEND_TIME 0x100 /* 225 *There was a RTT send time set that can be used 226 * no snd_limits 227 */ 228 #define BBR_RT_SET_GRADIENT 0x200 229 #define BBR_RT_TS_VALID 0x400 230 231 232 struct bbr_log { 233 union { 234 struct bbr_sendmap *rsm; /* For alloc/free */ 235 uint64_t sb_acc; /* For out/ack or t-o */ 236 }; 237 struct tcpcb *tp; 238 uint32_t t_flags; 239 uint32_t th_seq; 240 uint32_t th_ack; 241 uint32_t snd_una; 242 uint32_t snd_nxt; 243 uint32_t snd_max; 244 uint32_t snd_cwnd; 245 uint32_t snd_wnd; 246 uint32_t rc_lost; 247 uint32_t target_cwnd; /* UU */ 248 uint32_t inflight; /* UU */ 249 uint32_t applimited; /* UU */ 250 /* Things for BBR */ 251 uint32_t delivered; /* UU */ 252 uint64_t cur_del_rate; /* UU */ 253 uint64_t delRate; /* UU */ 254 uint64_t rttProp; /* UU */ 255 uint64_t lt_bw; /* UU */ 256 uint32_t timeStamp; 257 uint32_t time; 258 uint32_t slot; /* UU */ 259 uint32_t delayed_by; 260 uint32_t exp_del; 261 uint32_t pkts_out; 262 uint32_t new_win; 263 uint32_t hptsi_gain; /* UU */ 264 uint32_t cwnd_gain; /* UU */ 265 uint32_t epoch; /* UU */ 266 uint32_t lt_epoch; /* UU */ 267 /* Sack fun */ 268 uint32_t blk_start[4]; /* xx */ 269 uint32_t blk_end[4]; 270 uint32_t len; /* Timeout T3=1, TLP=2, RACK=3 */ 271 uint8_t type; 272 uint8_t n_sackblks; 273 uint8_t applied; /* UU */ 274 uint8_t inhpts; /* UU */ 275 uint8_t ininput; /* UU */ 276 uint8_t use_lt_bw; /* UU */ 277 }; 278 279 struct bbr_log_sysctl_out { 280 uint32_t bbr_log_at; 281 uint32_t bbr_log_max; 282 struct bbr_log entries[0]; 283 }; 284 285 /* 286 * Magic numbers for logging timeout events if the 287 * logging is enabled. 288 */ 289 #define BBR_TO_FRM_TMR 1 290 #define BBR_TO_FRM_TLP 2 291 #define BBR_TO_FRM_RACK 3 292 #define BBR_TO_FRM_KEEP 4 293 #define BBR_TO_FRM_PERSIST 5 294 #define BBR_TO_FRM_DELACK 6 295 296 #define BBR_SEES_STRETCH_ACK 1 297 #define BBR_SEES_COMPRESSED_ACKS 2 298 299 300 /* 301 * As we get each SACK we wade through the 302 * rc_map and mark off what is acked. 303 * We also increment rc_sacked as well. 304 * 305 * We also pay attention to missing entries 306 * based on the time and possibly mark them 307 * for retransmit. If we do and we are not already 308 * in recovery we enter recovery. In doing 309 * so we claer prr_delivered/holes_rxt and prr_sent_dur_rec. 310 * We also setup rc_next/rc_snd_nxt/rc_send_end so 311 * we will know where to send from. When not in 312 * recovery rc_next will be NULL and rc_snd_nxt should 313 * equal snd_max. 314 * 315 * Whenever we retransmit from recovery we increment 316 * rc_holes_rxt as we retran a block and mark it as retransmitted 317 * with the time it was sent. During non-recovery sending we 318 * add to our map and note the time down of any send expanding 319 * the rc_map at the tail and moving rc_snd_nxt up with snd_max. 320 * 321 * In recovery during SACK/ACK processing if a chunk has 322 * been retransmitted and it is now acked, we decrement rc_holes_rxt. 323 * When we retransmit from the scoreboard we use 324 * rc_next and rc_snd_nxt/rc_send_end to help us 325 * find what needs to be retran. 326 * 327 * To calculate pipe we simply take (snd_max - snd_una) + rc_holes_rxt 328 * This gets us the effect of RFC6675 pipe, counting twice for 329 * bytes retransmitted. 330 */ 331 332 #define TT_BBR_FR_TMR 0x2001 333 334 #define BBR_SCALE 8 335 #define BBR_UNIT (1 << BBR_SCALE) 336 337 #define BBR_NUM_RTTS_FOR_DEL_LIMIT 8 /* How many pkt-rtts do we keep 338 * Delivery rate for */ 339 #define BBR_NUM_RTTS_FOR_GOOG_DEL_LIMIT 10 /* How many pkt-rtts do we keep 340 * Delivery rate for google */ 341 342 #define BBR_SECONDS_NO_RTT 10 /* 10 seconds with no RTT shrinkage */ 343 #define BBR_PROBERTT_MAX 200 /* 200ms */ 344 #define BBR_PROBERTT_NUM_MSS 4 345 #define BBR_STARTUP_EPOCHS 3 346 #define USECS_IN_MSEC 1000 347 #define BBR_TIME_TO_SECONDS(a) (a / USECS_IN_SECOND) 348 #define BBR_TIME_TO_MILLI(a) (a / MS_IN_USEC) 349 350 351 /* BBR keeps time in usec's so we divide by 1000 and round up */ 352 #define BBR_TS_TO_MS(t) ((t+999)/MS_IN_USEC) 353 354 /* 355 * Locking for the rack control block. 356 * a) Locked by INP_WLOCK 357 * b) Locked by the hpts-mutex 358 * 359 */ 360 #define BBR_STATE_STARTUP 0x01 361 #define BBR_STATE_DRAIN 0x02 362 #define BBR_STATE_PROBE_BW 0x03 363 #define BBR_STATE_PROBE_RTT 0x04 364 #define BBR_STATE_IDLE_EXIT 0x05 365 366 /* Substate defines for STATE == PROBE_BW */ 367 #define BBR_SUB_GAIN 0 /* State 0 where we are 5/4 BBR_UNIT */ 368 #define BBR_SUB_DRAIN 1 /* State 1 where we are at 3/4 BBR_UNIT */ 369 #define BBR_SUB_LEVEL1 2 /* State 1 first BBR_UNIT */ 370 #define BBR_SUB_LEVEL2 3 /* State 2nd BBR_UNIT */ 371 #define BBR_SUB_LEVEL3 4 /* State 3rd BBR_UNIT */ 372 #define BBR_SUB_LEVEL4 5 /* State 4th BBR_UNIT */ 373 #define BBR_SUB_LEVEL5 6 /* State 5th BBR_UNIT */ 374 #define BBR_SUB_LEVEL6 7 /* State last BBR_UNIT */ 375 #define BBR_SUBSTATE_COUNT 8 376 377 /* Single remaining reduce log */ 378 #define BBR_REDUCE_AT_FR 5 379 380 #define BBR_BIG_LOG_SIZE 300000 381 382 struct bbr_stats { 383 uint64_t bbr_badfr; /* 0 */ 384 uint64_t bbr_badfr_bytes; /* 1 */ 385 uint64_t bbr_saw_oerr; /* 2 */ 386 uint64_t bbr_saw_emsgsiz; /* 3 */ 387 uint64_t bbr_reorder_seen; /* 4 */ 388 uint64_t bbr_tlp_tot; /* 5 */ 389 uint64_t bbr_tlp_newdata; /* 6 */ 390 uint64_t bbr_offset_recovery; /* 7 */ 391 uint64_t bbr_tlp_retran_fail; /* 8 */ 392 uint64_t bbr_to_tot; /* 9 */ 393 uint64_t bbr_to_arm_rack; /* 10 */ 394 uint64_t bbr_enter_probertt; /* 11 */ 395 uint64_t bbr_tlp_set; /* 12 */ 396 uint64_t bbr_resends_set; /* 13 */ 397 uint64_t bbr_force_output; /* 14 */ 398 uint64_t bbr_to_arm_tlp; /* 15 */ 399 uint64_t bbr_paced_segments; /* 16 */ 400 uint64_t bbr_saw_enobuf; /* 17 */ 401 uint64_t bbr_to_alloc_failed; /* 18 */ 402 uint64_t bbr_to_alloc_emerg; /* 19 */ 403 uint64_t bbr_sack_proc_all; /* 20 */ 404 uint64_t bbr_sack_proc_short; /* 21 */ 405 uint64_t bbr_sack_proc_restart; /* 22 */ 406 uint64_t bbr_to_alloc; /* 23 */ 407 uint64_t bbr_offset_drop; /* 24 */ 408 uint64_t bbr_runt_sacks; /* 25 */ 409 uint64_t bbr_sack_passed; /* 26 */ 410 uint64_t bbr_rlock_left_ret0; /* 27 */ 411 uint64_t bbr_rlock_left_ret1; /* 28 */ 412 uint64_t bbr_dynamic_rwnd; /* 29 */ 413 uint64_t bbr_static_rwnd; /* 30 */ 414 uint64_t bbr_sack_blocks; /* 31 */ 415 uint64_t bbr_sack_blocks_skip; /* 32 */ 416 uint64_t bbr_sack_search_both; /* 33 */ 417 uint64_t bbr_sack_search_fwd; /* 34 */ 418 uint64_t bbr_sack_search_back; /* 35 */ 419 uint64_t bbr_plain_acks; /* 36 */ 420 uint64_t bbr_acks_with_sacks; /* 37 */ 421 uint64_t bbr_progress_drops; /* 38 */ 422 uint64_t bbr_early; /* 39 */ 423 uint64_t bbr_reneges_seen; /* 40 */ 424 uint64_t bbr_persist_reneg; /* 41 */ 425 uint64_t bbr_dropped_af_data; /* 42 */ 426 uint64_t bbr_failed_mbuf_aloc; /* 43 */ 427 uint64_t bbr_cwnd_limited; /* 44 */ 428 uint64_t bbr_rwnd_limited; /* 45 */ 429 uint64_t bbr_app_limited; /* 46 */ 430 uint64_t bbr_force_timer_start; /* 47 */ 431 uint64_t bbr_hpts_min_time; /* 48 */ 432 uint64_t bbr_meets_tso_thresh; /* 49 */ 433 uint64_t bbr_miss_tso_rwnd; /* 50 */ 434 uint64_t bbr_miss_tso_cwnd; /* 51 */ 435 uint64_t bbr_miss_tso_app; /* 52 */ 436 uint64_t bbr_miss_retran; /* 53 */ 437 uint64_t bbr_miss_tlp; /* 54 */ 438 uint64_t bbr_miss_unknown; /* 55 */ 439 uint64_t bbr_hdwr_rl_add_ok; /* 56 */ 440 uint64_t bbr_hdwr_rl_add_fail; /* 57 */ 441 uint64_t bbr_hdwr_rl_mod_ok; /* 58 */ 442 uint64_t bbr_hdwr_rl_mod_fail; /* 59 */ 443 uint64_t bbr_collapsed_win; /* 60 */ 444 uint64_t bbr_alloc_limited; /* 61 */ 445 uint64_t bbr_alloc_limited_conns; /* 62 */ 446 uint64_t bbr_split_limited; /* 63 */ 447 }; 448 449 /* 450 * The structure bbr_opt_stats is a simple 451 * way to see how many options are being 452 * changed in the stack. 453 */ 454 struct bbr_opts_stats { 455 uint64_t tcp_bbr_pace_per_sec; 456 uint64_t tcp_bbr_pace_del_tar; 457 uint64_t tcp_bbr_pace_seg_max; 458 uint64_t tcp_bbr_pace_seg_min; 459 uint64_t tcp_bbr_pace_cross; 460 uint64_t tcp_bbr_drain_inc_extra; 461 uint64_t tcp_bbr_unlimited; 462 uint64_t tcp_bbr_iwintso; 463 uint64_t tcp_bbr_rec_over_hpts; 464 uint64_t tcp_bbr_recforce; 465 uint64_t tcp_bbr_startup_pg; 466 uint64_t tcp_bbr_drain_pg; 467 uint64_t tcp_bbr_rwnd_is_app; 468 uint64_t tcp_bbr_probe_rtt_int; 469 uint64_t tcp_bbr_one_retran; 470 uint64_t tcp_bbr_startup_loss_exit; 471 uint64_t tcp_bbr_use_lowgain; 472 uint64_t tcp_bbr_lowgain_thresh; 473 uint64_t tcp_bbr_lowgain_half; 474 uint64_t tcp_bbr_lowgain_fd; 475 uint64_t tcp_bbr_usedel_rate; 476 uint64_t tcp_bbr_min_rto; 477 uint64_t tcp_bbr_max_rto; 478 uint64_t tcp_rack_pace_max_seg; 479 uint64_t tcp_rack_min_to; 480 uint64_t tcp_rack_reord_thresh; 481 uint64_t tcp_rack_reord_fade; 482 uint64_t tcp_rack_tlp_thresh; 483 uint64_t tcp_rack_pkt_delay; 484 uint64_t tcp_bbr_startup_exit_epoch; 485 uint64_t tcp_bbr_ack_comp_alg; 486 uint64_t tcp_rack_cheat; 487 uint64_t tcp_iwnd_tso; 488 uint64_t tcp_utter_max_tso; 489 uint64_t tcp_hdwr_pacing; 490 uint64_t tcp_extra_state; 491 uint64_t tcp_floor_min_tso; 492 /* New */ 493 uint64_t tcp_bbr_algorithm; 494 uint64_t tcp_bbr_tslimits; 495 uint64_t tcp_bbr_probertt_len; 496 uint64_t tcp_bbr_probertt_gain; 497 uint64_t tcp_bbr_topaceout; 498 uint64_t tcp_use_rackcheat; 499 uint64_t tcp_delack; 500 uint64_t tcp_maxpeak; 501 uint64_t tcp_retran_wtso; 502 uint64_t tcp_data_ac; 503 uint64_t tcp_ts_raises; 504 uint64_t tcp_pacing_oh_tmr; 505 uint64_t tcp_pacing_oh; 506 uint64_t tcp_policer_det; 507 }; 508 509 510 #ifdef _KERNEL 511 #define BBR_STAT_SIZE (sizeof(struct bbr_stats)/sizeof(uint64_t)) 512 extern counter_u64_t bbr_stat_arry[BBR_STAT_SIZE]; 513 #define BBR_STAT_ADD(name, amm) counter_u64_add(bbr_stat_arry[(offsetof(struct bbr_stats, name)/sizeof(uint64_t))], (amm)) 514 #define BBR_STAT_INC(name) BBR_STAT_ADD(name, 1) 515 #define BBR_OPTS_SIZE (sizeof(struct bbr_stats)/sizeof(uint64_t)) 516 extern counter_u64_t bbr_opts_arry[BBR_OPTS_SIZE]; 517 #define BBR_OPTS_ADD(name, amm) counter_u64_add(bbr_opts_arry[(offsetof(struct bbr_opts_stats, name)/sizeof(uint64_t))], (amm)) 518 #define BBR_OPTS_INC(name) BBR_OPTS_ADD(name, 1) 519 #endif 520 521 #define BBR_NUM_LOSS_RATES 3 522 #define BBR_NUM_BW_RATES 3 523 524 #define BBR_RECOVERY_LOWRTT 1 525 #define BBR_RECOVERY_MEDRTT 2 526 #define BBR_RECOVERY_HIGHRTT 3 527 #define BBR_RECOVERY_EXTREMERTT 4 528 529 530 struct bbr_control { 531 /*******************************/ 532 /* Cache line 2 from bbr start */ 533 /*******************************/ 534 struct bbr_head rc_map; /* List of all segments Lock(a) */ 535 struct bbr_head rc_tmap; /* List in transmit order Lock(a) */ 536 struct bbr_sendmap *rc_resend; /* something we have been asked to 537 * resend */ 538 uint32_t rc_last_delay_val; /* How much we expect to delay Lock(a) */ 539 uint32_t rc_bbr_hptsi_gain:16, /* Current hptsi gain Lock(a) */ 540 rc_hpts_flags:16; /* flags on whats on the pacer wheel */ 541 542 uint32_t rc_delivered; /* BRR delivered amount Lock(a) */ 543 uint32_t rc_hptsi_agg_delay; /* How much time are we behind */ 544 545 uint32_t rc_flight_at_input; 546 uint32_t rc_lost_bytes; /* Total bytes currently marked lost */ 547 /*******************************/ 548 /* Cache line 3 from bbr start */ 549 /*******************************/ 550 struct time_filter rc_delrate; 551 /*******************************/ 552 /* Cache line 4 from bbr start */ 553 /*******************************/ 554 struct bbr_head rc_free; /* List of Free map entries Lock(a) */ 555 struct bbr_sendmap *rc_tlp_send; /* something we have been 556 * asked to resend */ 557 uint32_t rc_del_time; 558 uint32_t rc_target_at_state; /* Target for a state */ 559 560 uint16_t rc_free_cnt; /* Number of free entries on the rc_free list 561 * Lock(a) */ 562 uint16_t rc_startup_pg; 563 564 uint32_t cur_rtt; /* Last RTT from ack */ 565 566 567 uint32_t rc_went_idle_time; /* Used for persits to see if its 568 * probe-rtt qualified */ 569 uint32_t rc_pace_max_segs:17, /* How much in any single TSO we send Lock(a) */ 570 rc_pace_min_segs:15; /* The minimum single segment size before we enter persists */ 571 572 uint32_t rc_rtt_shrinks; /* Time of last rtt shrinkage Lock(a) */ 573 uint32_t r_app_limited_until; 574 uint32_t rc_timer_exp; /* If a timer ticks of expiry */ 575 uint32_t rc_rcv_epoch_start; /* Start time of the Epoch Lock(a) */ 576 577 /*******************************/ 578 /* Cache line 5 from bbr start */ 579 /*******************************/ 580 581 uint32_t rc_lost_at_pktepoch; /* what the lost value was at the last 582 * pkt-epoch */ 583 uint32_t r_measurement_count; /* count of measurement applied lock(a) */ 584 585 586 uint32_t rc_last_tlp_seq; /* Last tlp sequence Lock(a) */ 587 uint16_t rc_reorder_shift; /* Socket option value Lock(a) */ 588 uint16_t rc_pkt_delay; /* Socket option value Lock(a) */ 589 590 struct bbr_sendmap *rc_sacklast; /* sack remembered place 591 * Lock(a) */ 592 struct bbr_sendmap *rc_next; /* remembered place where we next 593 * retransmit at Lock(a) */ 594 595 uint32_t rc_sacked; /* Tot sacked on scoreboard Lock(a) */ 596 uint32_t rc_holes_rxt; /* Tot retraned from scoreboard Lock(a) */ 597 598 uint32_t rc_reorder_ts; /* Last time we saw reordering Lock(a) */ 599 uint32_t rc_init_rwnd; /* Initial rwnd when we transitioned */ 600 /*- --- 601 * used only initial and close 602 */ 603 uint32_t rc_high_rwnd; /* Highest rwnd seen */ 604 uint32_t rc_lowest_rtt; /* Smallest RTT we have seen */ 605 606 uint32_t rc_last_rtt; /* Last valid measured RTT that ack'd data */ 607 uint32_t bbr_cross_over; 608 609 /*******************************/ 610 /* Cache line 6 from bbr start */ 611 /*******************************/ 612 struct sack_filter bbr_sf; 613 614 /*******************************/ 615 /* Cache line 7 from bbr start */ 616 /*******************************/ 617 struct time_filter_small rc_rttprop; 618 uint32_t last_inbound_ts; /* Peers last timestamp */ 619 620 uint32_t rc_inc_tcp_oh: 1, 621 rc_inc_ip_oh: 1, 622 rc_inc_enet_oh:1, 623 rc_incr_tmrs:1, 624 restrict_growth:28; 625 uint32_t rc_lt_epoch_use; /* When we started lt-bw use Lock(a) */ 626 627 uint32_t rc_recovery_start; /* Time we start recovery Lock(a) */ 628 uint32_t rc_lt_del; /* Delivered at lt bw sampling start Lock(a) */ 629 630 uint64_t rc_bbr_cur_del_rate; /* Current measured delivery rate 631 * Lock(a) */ 632 633 /*******************************/ 634 /* Cache line 8 from bbr start */ 635 /*******************************/ 636 uint32_t rc_cwnd_on_ent; /* On entry to recovery the cwnd 637 * Lock(a) */ 638 uint32_t rc_agg_early; /* aggregate amount early */ 639 640 uint32_t rc_rcvtime; /* When we last received data Lock(a) */ 641 uint32_t rc_pkt_epoch_del; /* seq num that we need for RTT epoch */ 642 643 uint32_t rc_pkt_epoch; /* Epoch based on packet RTTs */ 644 uint32_t rc_pkt_epoch_time; /* Time we started the pkt epoch */ 645 646 uint32_t rc_pkt_epoch_rtt; /* RTT using the packet epoch */ 647 uint32_t rc_rtt_epoch; /* Current RTT epoch, it ticks every rttProp 648 * Lock(a) */ 649 uint32_t lowest_rtt; 650 uint32_t bbr_smallest_srtt_this_state; 651 652 uint32_t rc_lt_epoch; /* LT epoch start of bw_sampling */ 653 uint32_t rc_lost_at_startup; 654 655 uint32_t rc_bbr_state_atflight; 656 uint32_t rc_bbr_last_startup_epoch; /* Last startup epoch where we 657 * increased 20% */ 658 uint32_t rc_bbr_enters_probertt; /* Timestamp we entered 659 * probertt Lock(a) */ 660 uint32_t rc_lt_time; /* Time of lt sampling start Lock(a) */ 661 662 /*******************************/ 663 /* Cache line 9 from bbr start */ 664 /*******************************/ 665 uint64_t rc_lt_bw; /* LT bw calculated Lock(a) */ 666 uint64_t rc_bbr_lastbtlbw; /* For startup, what was last btlbw I 667 * saw to check the 20% gain Lock(a) */ 668 669 670 uint32_t rc_bbr_cwnd_gain; /* Current cwnd gain Lock(a) */ 671 uint32_t rc_pkt_epoch_loss_rate; /* pkt-epoch loss rate */ 672 673 uint32_t rc_saved_cwnd; /* Saved cwnd during Probe-rtt drain Lock(a) */ 674 uint32_t substate_pe; 675 676 uint32_t rc_lost; /* Number of bytes lost Lock(a) */ 677 uint32_t rc_exta_time_gd; /* How much extra time we got in d/g */ 678 679 uint32_t rc_lt_lost; /* Number of lt bytes lost at sampling start 680 * Lock(a) */ 681 uint32_t rc_bbr_state_time; 682 683 uint32_t rc_min_to; /* Socket option value Lock(a) */ 684 uint32_t rc_initial_hptsi_bw; /* Our initial startup bw Lock(a) */ 685 686 uint32_t bbr_lost_at_state; /* Temp counter debug lost value as we 687 * enter a state */ 688 /*******************************/ 689 /* Cache line 10 from bbr start */ 690 /*******************************/ 691 uint32_t rc_level_state_extra; 692 uint32_t rc_red_cwnd_pe; 693 const struct tcp_hwrate_limit_table *crte; 694 uint64_t red_bw; 695 696 uint32_t rc_probertt_int; 697 uint32_t rc_probertt_srttchktim; /* Time we last did a srtt 698 * check */ 699 uint32_t gain_epoch; /* Epoch we should be out of gain */ 700 uint32_t rc_min_rto_ms; 701 702 uint32_t rc_reorder_fade; /* Socket option value Lock(a) */ 703 uint32_t last_startup_measure; 704 705 int32_t bbr_hptsi_per_second; 706 int32_t bbr_hptsi_segments_delay_tar; 707 708 int32_t bbr_hptsi_segments_max; 709 uint32_t bbr_rttprobe_gain_val; 710 /*******************************/ 711 /* Cache line 11 from bbr start */ 712 /*******************************/ 713 uint32_t cur_rtt_send_time; /* Time we sent our rtt measured packet */ 714 uint32_t bbr_peer_tsratio; /* Our calculated ts ratio to multply */ 715 uint32_t bbr_ts_check_tstmp; /* When we filled it the TS that came on the ack */ 716 uint32_t bbr_ts_check_our_cts; /* When we filled it the cts of the send */ 717 uint32_t rc_tlp_rxt_last_time; 718 uint32_t bbr_smallest_srtt_state2; 719 uint32_t bbr_hdwr_cnt_noset_snt; /* count of hw pacing sends during delay */ 720 uint32_t startup_last_srtt; 721 uint32_t rc_ack_hdwr_delay; 722 uint32_t highest_hdwr_delay; /* Largest delay we have seen from hardware */ 723 uint32_t non_gain_extra; 724 uint32_t recovery_lr; /* The sum of the loss rate from the pe's during recovery */ 725 uint32_t last_in_probertt; 726 uint32_t flightsize_at_drain; /* In draining what was the last marked flight size */ 727 uint32_t rc_pe_of_prtt; /* PE we went into probe-rtt */ 728 uint32_t ts_in; /* ts that went with the last rtt */ 729 730 uint16_t rc_tlp_seg_send_cnt; /* Number of times we have TLP sent 731 * rc_last_tlp_seq Lock(a) */ 732 uint16_t rc_drain_pg; 733 uint32_t rc_num_maps_alloced; /* num send map entries allocated */ 734 uint32_t rc_num_split_allocs; /* num split map entries allocated */ 735 uint16_t rc_num_small_maps_alloced; /* Number of sack blocks 736 * allocated */ 737 uint16_t bbr_hptsi_bytes_min; 738 739 uint16_t bbr_hptsi_segments_floor; 740 uint16_t bbr_utter_max; 741 uint16_t bbr_google_discount; 742 743 }; 744 745 746 struct socket; 747 struct tcp_bbr { 748 /* First cache line 0x00 */ 749 int32_t(*r_substate) (struct mbuf *, struct tcphdr *, 750 struct socket *, struct tcpcb *, struct tcpopt *, 751 int32_t, int32_t, uint32_t, int32_t, int32_t); /* Lock(a) */ 752 struct tcpcb *rc_tp; /* The tcpcb Lock(a) */ 753 struct inpcb *rc_inp; /* The inpcb Lock(a) */ 754 struct timeval rc_tv; 755 uint32_t rc_pacer_started; /* Time we started the pacer */ 756 uint16_t no_pacing_until:8, /* No pacing until N packet epochs */ 757 ts_can_raise:1,/* TS b/w calculations can raise the bw higher */ 758 skip_gain:1, /* Skip the gain cycle (hardware pacing) */ 759 gain_is_limited:1, /* With hardware pacing we are limiting gain */ 760 output_error_seen:1, 761 oerror_cnt:4, 762 hw_pacing_set:1; /* long enough has passed for us to start pacing */ 763 uint16_t xxx_r_ack_count; /* During recovery count of ack's received 764 * that added data since output */ 765 uint16_t bbr_segs_rcvd; /* In Segment count since we sent a ack */ 766 767 uint8_t bbr_timer_src:4, /* Used for debugging Lock(a) */ 768 bbr_use_rack_cheat:1, /* Use the rack cheat */ 769 bbr_init_win_cheat:1, /* Send full IW for TSO */ 770 bbr_attempt_hdwr_pace:1,/* Try to do hardware pacing */ 771 bbr_hdrw_pacing:1; /* Hardware pacing is available */ 772 uint8_t bbr_hdw_pace_ena:1, /* Does the connection allow hardware pacing to be attempted */ 773 bbr_prev_in_rec:1, /* We were previously in recovery */ 774 pkt_conservation:1, 775 use_policer_detection:1, 776 xxx_bbr_hdw_pace_idx:4; /* If hardware pacing is on, index to slot in pace tbl */ 777 uint16_t r_wanted_output:1, 778 rtt_valid:1, 779 rc_timer_first:1, 780 rc_output_starts_timer:1, 781 rc_resends_use_tso:1, 782 rc_all_timers_stopped:1, 783 rc_loss_exit:1, 784 rc_ack_was_delayed:1, 785 rc_lt_is_sampling:1, 786 rc_filled_pipe:1, 787 rc_tlp_new_data:1, 788 rc_hit_state_1:1, 789 rc_ts_valid:1, 790 rc_prtt_set_ts:1, 791 rc_is_pkt_epoch_now:1, 792 rc_has_collapsed:1; 793 794 uint8_t r_state:4, /* Current bbr state Lock(a) */ 795 r_agg_early_set:1, /* Did we get called early */ 796 r_init_rtt:1, 797 r_use_policer:1, /* For google mode only */ 798 r_recovery_bw:1; 799 uint8_t r_timer_override:1, /* pacer override Lock(a) 0/1 */ 800 rc_in_persist:1, 801 rc_lt_use_bw:1, 802 rc_allow_data_af_clo:1, 803 rc_tlp_rtx_out:1, /* A TLP is in flight */ 804 rc_tlp_in_progress:1, /* a TLP timer is running needed? */ 805 rc_use_idle_restart:1; /* Do we restart fast after idle (persist or applim) */ 806 uint8_t rc_bbr_state:3, /* What is the major BBR state */ 807 rc_bbr_substate:3, /* For probeBW state */ 808 r_is_v6:1, 809 rc_past_init_win:1; 810 uint8_t rc_last_options; 811 uint8_t rc_tlp_threshold; /* Socket option value Lock(a) */ 812 uint8_t rc_max_rto_sec; 813 uint8_t rc_cwnd_limited:1, /* We are cwnd limited */ 814 rc_tmr_stopped:7; /* What timers have been stopped */ 815 uint8_t rc_use_google:1, 816 rc_use_ts_limit:1, 817 rc_ts_data_set:1, /* We have filled a set point to determine */ 818 rc_ts_clock_set:1, /* We have determined the ts type */ 819 rc_ts_cant_be_used:1, /* We determined we can't use ts values */ 820 rc_ack_is_cumack:1, 821 rc_no_pacing:1, 822 alloc_limit_reported:1; 823 uint8_t rc_init_win; 824 /* Cache line 2 0x40 */ 825 struct bbr_control r_ctl; 826 #ifdef _KERNEL 827 } __aligned(CACHE_LINE_SIZE); 828 #else 829 }; 830 #endif 831 832 #endif 833