/* * Copyright (c) 2010-2021 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in * compliance with the License. The rights granted to you under the License * may not be used to create, or enable the creation or redistribution of, * unlawful or unlicensed copies of an Apple operating system, or to * circumvent, violate, or enable the circumvention or violation of, any * terms of an Apple operating system software license agreement. * * Please obtain a copy of the License at * http://www.opensource.apple.com/apsl/ and read it before using this file. * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ #include "tcp_includes.h" #include #include #include #include #include #include #include #include #include /* This file implements an alternate TCP congestion control algorithm * for background transport developed by LEDBAT working group at IETF and * described in draft: draft-ietf-ledbat-congestion-02 * * Currently, it also implements LEDBAT++ as described in draft * draft-irtf-iccrg-ledbat-plus-plus-01. */ #define GAIN_CONSTANT (16) #define DEFER_SLOWDOWN_DURATION (30 * 1000) /* 30s */ int tcp_ledbat_init(struct tcpcb *tp); int tcp_ledbat_cleanup(struct tcpcb *tp); void tcp_ledbat_cwnd_init(struct tcpcb *tp); void tcp_ledbat_congestion_avd(struct tcpcb *tp, struct tcphdr *th); void tcp_ledbat_ack_rcvd(struct tcpcb *tp, struct tcphdr *th); static void ledbat_pp_ack_rcvd(struct tcpcb *tp, uint32_t bytes_acked); void tcp_ledbat_pre_fr(struct tcpcb *tp); void tcp_ledbat_post_fr(struct tcpcb *tp, struct tcphdr *th); void tcp_ledbat_after_idle(struct tcpcb *tp); void tcp_ledbat_after_timeout(struct tcpcb *tp); static int tcp_ledbat_delay_ack(struct tcpcb *tp, struct tcphdr *th); void tcp_ledbat_switch_cc(struct tcpcb *tp); struct tcp_cc_algo tcp_cc_ledbat = { .name = "ledbat", .init = tcp_ledbat_init, .cleanup = tcp_ledbat_cleanup, .cwnd_init = tcp_ledbat_cwnd_init, .congestion_avd = tcp_ledbat_congestion_avd, .ack_rcvd = tcp_ledbat_ack_rcvd, .pre_fr = tcp_ledbat_pre_fr, .post_fr = tcp_ledbat_post_fr, .after_idle = tcp_ledbat_after_idle, .after_timeout = tcp_ledbat_after_timeout, .delay_ack = tcp_ledbat_delay_ack, .switch_to = tcp_ledbat_switch_cc }; static void update_cwnd(struct tcpcb *tp, uint32_t update, bool is_incr) { uint32_t max_allowed_cwnd = 0, flight_size = 0; uint32_t base_rtt = get_base_rtt(tp); uint32_t curr_rtt = tcp_use_min_curr_rtt ? tp->curr_rtt_min : tp->t_rttcur; /* If we do not have a good RTT measurement yet, increment * congestion window by the default value. */ if (base_rtt == 0 || curr_rtt == 0) { tp->snd_cwnd += update; goto check_max; } if (curr_rtt <= (base_rtt + target_qdelay)) { /* * Delay decreased or remained the same, we can increase * the congestion window according to RFC 3465. * * Move background slow-start threshold to current * congestion window so that the next time (after some idle * period), we can attempt to do slow-start till here if there * is no increase in rtt */ if (tp->bg_ssthresh < tp->snd_cwnd) { tp->bg_ssthresh = tp->snd_cwnd; } tp->snd_cwnd += update; tp->snd_cwnd = tcp_round_to(tp->snd_cwnd, tp->t_maxseg); } else { if (tcp_ledbat_plus_plus) { VERIFY(is_incr == false); tp->snd_cwnd -= update; } else { /* In response to an increase in rtt, reduce the congestion * window by one-eighth. This will help to yield immediately * to a competing stream. */ uint32_t redwin; redwin = tp->snd_cwnd >> 3; tp->snd_cwnd -= redwin; } if (tp->snd_cwnd < bg_ss_fltsz * tp->t_maxseg) { tp->snd_cwnd = bg_ss_fltsz * tp->t_maxseg; } tp->snd_cwnd = tcp_round_to(tp->snd_cwnd, tp->t_maxseg); /* Lower background slow-start threshold so that the connection * will go into congestion avoidance phase */ if (tp->bg_ssthresh > tp->snd_cwnd) { tp->bg_ssthresh = tp->snd_cwnd; } } check_max: if (!tcp_ledbat_plus_plus) { /* Calculate the outstanding flight size and restrict the * congestion window to a factor of flight size. */ flight_size = tp->snd_max - tp->snd_una; max_allowed_cwnd = (tcp_ledbat_allowed_increase * tp->t_maxseg) + (flight_size << tcp_ledbat_tether_shift); tp->snd_cwnd = min(tp->snd_cwnd, max_allowed_cwnd); } else { tp->snd_cwnd = min(tp->snd_cwnd, TCP_MAXWIN << tp->snd_scale); } } static inline void tcp_ledbat_clear_state(struct tcpcb *tp) { tp->t_ccstate->ledbat_slowdown_events = 0; tp->t_ccstate->ledbat_slowdown_ts = 0; tp->t_ccstate->ledbat_slowdown_begin = 0; tp->t_ccstate->ledbat_md_bytes_acked = 0; } int tcp_ledbat_init(struct tcpcb *tp) { os_atomic_inc(&tcp_cc_ledbat.num_sockets, relaxed); tcp_ledbat_clear_state(tp); return 0; } int tcp_ledbat_cleanup(struct tcpcb *tp) { #pragma unused(tp) os_atomic_dec(&tcp_cc_ledbat.num_sockets, relaxed); return 0; } /* * Initialize the congestion window for a connection */ void tcp_ledbat_cwnd_init(struct tcpcb *tp) { tp->snd_cwnd = tp->t_maxseg * bg_ss_fltsz; tp->bg_ssthresh = tp->snd_ssthresh; } /* Function to handle an in-sequence ack which is fast-path processing * of an in sequence ack in tcp_input function (called as header prediction). * This gets called only during congestion avoidance phase. */ void tcp_ledbat_congestion_avd(struct tcpcb *tp, struct tcphdr *th) { int acked = 0; uint32_t incr = 0; acked = BYTES_ACKED(th, tp); if (tcp_ledbat_plus_plus) { ledbat_pp_ack_rcvd(tp, acked); } else { tp->t_bytes_acked += acked; if (tp->t_bytes_acked > tp->snd_cwnd) { tp->t_bytes_acked -= tp->snd_cwnd; incr = tp->t_maxseg; } if (tp->snd_cwnd < tp->snd_wnd && incr > 0) { update_cwnd(tp, incr, true); } } } /* * Compute the denominator * MIN(16, ceil(2 * TARGET / base)) */ static uint32_t ledbat_gain(uint32_t base_rtt) { return MIN(GAIN_CONSTANT, tcp_ceil(2 * target_qdelay / (double)base_rtt)); } /* * Congestion avoidance for ledbat++ */ static void ledbat_pp_congestion_avd(struct tcpcb *tp, uint32_t bytes_acked, uint32_t base_rtt, uint32_t curr_rtt, uint32_t now) { uint32_t update = 0; /* * Set the next slowdown time i.e. 9 times the duration * of previous slowdown except the initial slowdown. */ if (tp->t_ccstate->ledbat_slowdown_ts == 0) { uint32_t slowdown_duration = 0; if (tp->t_ccstate->ledbat_slowdown_events > 0) { slowdown_duration = now - tp->t_ccstate->ledbat_slowdown_begin; if (tp->bg_ssthresh > tp->snd_cwnd) { /* * Special case for slowdowns (other than initial) * where cwnd doesn't recover fully to previous * ssthresh */ slowdown_duration *= 2; } } tp->t_ccstate->ledbat_slowdown_ts = now + (9 * slowdown_duration); if (slowdown_duration == 0) { tp->t_ccstate->ledbat_slowdown_ts += (2 * (tp->t_srtt >> TCP_RTT_SHIFT)); } /* Reset the start */ tp->t_ccstate->ledbat_slowdown_begin = 0; /* On exit slow start due to higher qdelay, cap the ssthresh */ if (tp->bg_ssthresh > tp->snd_cwnd) { tp->bg_ssthresh = tp->snd_cwnd; } } if (curr_rtt <= base_rtt + target_qdelay) { /* Additive increase */ tp->t_bytes_acked += bytes_acked; if (tp->t_bytes_acked >= tp->snd_cwnd) { update = tp->t_maxseg; tp->t_bytes_acked -= tp->snd_cwnd; update_cwnd(tp, update, true); } } else { /* * Multiplicative decrease * W -= min(W * (qdelay/target - 1), W/2) (per RTT) * To calculate per bytes acked, it becomes * W -= min((qdelay/target - 1), 1/2) * bytes_acked */ uint32_t qdelay = curr_rtt > base_rtt ? (curr_rtt - base_rtt) : 0; tp->t_ccstate->ledbat_md_bytes_acked += bytes_acked; if (tp->t_ccstate->ledbat_md_bytes_acked >= tp->snd_cwnd) { update = (uint32_t)(MIN(((double)qdelay / target_qdelay - 1), 0.5) * (double)tp->snd_cwnd); tp->t_ccstate->ledbat_md_bytes_acked -= tp->snd_cwnd; update_cwnd(tp, update, false); if (tp->t_ccstate->ledbat_slowdown_ts != 0) { /* As the window has been reduced, defer the slowdown. */ tp->t_ccstate->ledbat_slowdown_ts = now + DEFER_SLOWDOWN_DURATION; } } } } /* * Different handling for ack received for ledbat++ */ static void ledbat_pp_ack_rcvd(struct tcpcb *tp, uint32_t bytes_acked) { uint32_t update = 0; const uint32_t base_rtt = get_base_rtt(tp); const uint32_t curr_rtt = tcp_use_min_curr_rtt ? tp->curr_rtt_min : tp->t_rttcur; const uint32_t ss_target = (uint32_t)(3 * target_qdelay / 4); struct tcp_globals *globals = tcp_get_globals(tp); /* * Slowdown period - first slowdown * is 2RTT after we exit initial slow start. * Subsequent slowdowns are after 9 times the * previous slow down durations. */ if (tp->t_ccstate->ledbat_slowdown_ts != 0 && tcp_globals_now(globals) >= tp->t_ccstate->ledbat_slowdown_ts) { if (tp->t_ccstate->ledbat_slowdown_begin == 0) { tp->t_ccstate->ledbat_slowdown_begin = tcp_globals_now(globals); tp->t_ccstate->ledbat_slowdown_events++; } if (tcp_globals_now(globals) < tp->t_ccstate->ledbat_slowdown_ts + (2 * (tp->t_srtt >> TCP_RTT_SHIFT))) { // Set cwnd to 2 packets and return if (tp->snd_cwnd > bg_ss_fltsz * tp->t_maxseg) { if (tp->bg_ssthresh < tp->snd_cwnd) { tp->bg_ssthresh = tp->snd_cwnd; } tp->snd_cwnd = bg_ss_fltsz * tp->t_maxseg; /* Reset total bytes acked */ tp->t_bytes_acked = 0; } return; } } if (curr_rtt == 0 || base_rtt == 0) { update = MIN(bytes_acked, TCP_CC_CWND_INIT_PKTS * tp->t_maxseg); update_cwnd(tp, update, true); } else if (tp->snd_cwnd < tp->bg_ssthresh && ((tp->t_ccstate->ledbat_slowdown_events > 0 && curr_rtt <= (base_rtt + target_qdelay)) || curr_rtt <= (base_rtt + ss_target))) { /* * Modified slow start with a dynamic GAIN * If the queuing delay is larger than 3/4 of the target * delay, exit slow start, iff, it is the initial slow start. * After the initial slow start, during CA, window growth * will be bound by ssthresh. */ tp->t_bytes_acked += bytes_acked; uint32_t gain_factor = ledbat_gain(base_rtt); if (tp->t_bytes_acked >= tp->t_maxseg * gain_factor) { update = MIN(tp->t_bytes_acked / gain_factor, TCP_CC_CWND_INIT_PKTS * tp->t_maxseg); tp->t_bytes_acked = 0; update_cwnd(tp, update, true); } /* Reset the next slowdown timestamp */ if (tp->t_ccstate->ledbat_slowdown_ts != 0) { tp->t_ccstate->ledbat_slowdown_ts = 0; } } else { /* Congestion avoidance */ ledbat_pp_congestion_avd(tp, bytes_acked, base_rtt, curr_rtt, tcp_globals_now(globals)); } } /* Function to process an ack. */ void tcp_ledbat_ack_rcvd(struct tcpcb *tp, struct tcphdr *th) { /* * RFC 3465 - Appropriate Byte Counting. * * If the window is currently less than ssthresh, * open the window by the number of bytes ACKed by * the last ACK, however clamp the window increase * to an upper limit "L". * * In congestion avoidance phase, open the window by * one segment each time "bytes_acked" grows to be * greater than or equal to the congestion window. */ uint32_t cw = tp->snd_cwnd; uint32_t incr = tp->t_maxseg; uint32_t acked = 0; acked = BYTES_ACKED(th, tp); if (tcp_ledbat_plus_plus) { ledbat_pp_ack_rcvd(tp, acked); return; } tp->t_bytes_acked += acked; if (cw >= tp->bg_ssthresh) { /* congestion-avoidance */ if (tp->t_bytes_acked < cw) { /* No need to increase yet. */ incr = 0; } } else { /* * If the user explicitly enables RFC3465 * use 2*SMSS for the "L" param. Otherwise * use the more conservative 1*SMSS. * * (See RFC 3465 2.3 Choosing the Limit) */ u_int abc_lim; abc_lim = (tp->snd_nxt == tp->snd_max) ? incr * 2 : incr; incr = ulmin(acked, abc_lim); } if (tp->t_bytes_acked >= cw) { tp->t_bytes_acked -= cw; } if (incr > 0) { update_cwnd(tp, incr, true); } } void tcp_ledbat_pre_fr(struct tcpcb *tp) { uint32_t win = min(tp->snd_wnd, tp->snd_cwnd); if (tp->t_flagsext & TF_CWND_NONVALIDATED) { tp->t_lossflightsize = tp->snd_max - tp->snd_una; win = max(tp->t_pipeack, tp->t_lossflightsize); } else { tp->t_lossflightsize = 0; } win = win / 2; win = tcp_round_to(win, tp->t_maxseg); if (win < 2 * tp->t_maxseg) { win = 2 * tp->t_maxseg; } tp->snd_ssthresh = win; if (tp->bg_ssthresh > tp->snd_ssthresh) { tp->bg_ssthresh = tp->snd_ssthresh; } tcp_cc_resize_sndbuf(tp); } void tcp_ledbat_post_fr(struct tcpcb *tp, struct tcphdr *th) { int32_t ss; if (th) { ss = tp->snd_max - th->th_ack; } else { ss = tp->snd_max - tp->snd_una; } /* * Complete ack. Inflate the congestion window to * ssthresh and exit fast recovery. * * Window inflation should have left us with approx. * snd_ssthresh outstanding data. But in case we * would be inclined to send a burst, better to do * it via the slow start mechanism. * * If the flight size is zero, then make congestion * window to be worth at least 2 segments to avoid * delayed acknowledgement (draft-ietf-tcpm-rfc3782-bis-05). */ if (ss < (int32_t)tp->snd_ssthresh) { tp->snd_cwnd = max(ss, tp->t_maxseg) + tp->t_maxseg; } else { tp->snd_cwnd = tp->snd_ssthresh; } tp->t_bytes_acked = 0; tp->t_ccstate->ledbat_md_bytes_acked = 0; } /* * Function to handle connections that have been idle for * some time. Slow start to get ack "clock" running again. * Clear base history after idle time. */ void tcp_ledbat_after_idle(struct tcpcb *tp) { tcp_ledbat_clear_state(tp); /* Reset the congestion window */ tp->snd_cwnd = tp->t_maxseg * bg_ss_fltsz; tp->t_bytes_acked = 0; tp->t_ccstate->ledbat_md_bytes_acked = 0; } /* Function to change the congestion window when the retransmit * timer fires. The behavior is the same as that for best-effort * TCP, reduce congestion window to one segment and start probing * the link using "slow start". The slow start threshold is set * to half of the current window. Lower the background slow start * threshold also. */ void tcp_ledbat_after_timeout(struct tcpcb *tp) { if (tp->t_state >= TCPS_ESTABLISHED) { tcp_ledbat_clear_state(tp); tcp_ledbat_pre_fr(tp); tp->snd_cwnd = tp->t_maxseg; } } /* * Indicate whether this ack should be delayed. * We can delay the ack if: * - our last ack wasn't a 0-sized window. * - the peer hasn't sent us a TH_PUSH data packet: if he did, take this * as a clue that we need to ACK without any delay. This helps higher * level protocols who won't send us more data even if the window is * open because their last "segment" hasn't been ACKed * Otherwise the receiver will ack every other full-sized segment or when the * delayed ack timer fires. This will help to generate better rtt estimates for * the other end if it is a ledbat sender. * */ static int tcp_ledbat_delay_ack(struct tcpcb *tp, struct tcphdr *th) { if (tcp_ack_strategy == TCP_ACK_STRATEGY_MODERN) { return tcp_cc_delay_ack(tp, th); } else { if ((tp->t_flags & TF_RXWIN0SENT) == 0 && (th->th_flags & TH_PUSH) == 0 && (tp->t_unacksegs == 1)) { return 1; } return 0; } } /* Change a connection to use ledbat. First, lower bg_ssthresh value * if it needs to be. */ void tcp_ledbat_switch_cc(struct tcpcb *tp) { uint32_t cwnd; tcp_ledbat_clear_state(tp); if (tp->bg_ssthresh == 0 || tp->bg_ssthresh > tp->snd_ssthresh) { tp->bg_ssthresh = tp->snd_ssthresh; } cwnd = min(tp->snd_wnd, tp->snd_cwnd); if (tp->snd_cwnd > tp->bg_ssthresh) { cwnd = cwnd / tp->t_maxseg; } else { cwnd = cwnd / 2 / tp->t_maxseg; } if (cwnd < bg_ss_fltsz) { cwnd = bg_ss_fltsz; } tp->snd_cwnd = cwnd * tp->t_maxseg; tp->t_bytes_acked = 0; os_atomic_inc(&tcp_cc_ledbat.num_sockets, relaxed); }