Index: head/sys/netinet/cc/cc_cdg.c =================================================================== --- head/sys/netinet/cc/cc_cdg.c (revision 307899) +++ head/sys/netinet/cc/cc_cdg.c (revision 307900) @@ -1,708 +1,699 @@ /*- * Copyright (c) 2009-2013 * Swinburne University of Technology, Melbourne, Australia * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by David Hayes, made * possible in part by a gift from The Cisco University Research Program Fund, * a corporate advised fund of Silicon Valley Community Foundation. Development * and testing were further assisted by a grant from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * CAIA Delay-Gradient (CDG) congestion control algorithm * * An implemention of the delay-gradient congestion control algorithm proposed * in the following paper: * * D. A. Hayes and G. Armitage, "Revisiting TCP Congestion Control using Delay * Gradients", in IFIP Networking, Valencia, Spain, 9-13 May 2011. * * Developed as part of the NewTCP research project at Swinburne University of * Technology's Centre for Advanced Internet Architectures, Melbourne, * Australia. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CDG_VERSION "0.1" #define CAST_PTR_INT(X) (*((int*)(X))) /* Private delay-gradient induced congestion control signal. */ #define CC_CDG_DELAY 0x01000000 /* NewReno window deflation factor on loss (as a percentage). */ #define RENO_BETA 50 /* Queue states. */ #define CDG_Q_EMPTY 1 #define CDG_Q_RISING 2 #define CDG_Q_FALLING 3 #define CDG_Q_FULL 4 #define CDG_Q_UNKNOWN 9999 /* Number of bit shifts used in probexp lookup table. */ #define EXP_PREC 15 /* Largest gradient represented in probexp lookup table. */ #define MAXGRAD 5 /* * Delay Precision Enhance - number of bit shifts used for qtrend related * integer arithmetic precision. */ #define D_P_E 7 struct qdiff_sample { long qdiff; STAILQ_ENTRY(qdiff_sample) qdiff_lnk; }; struct cdg { long max_qtrend; long min_qtrend; STAILQ_HEAD(minrtts_head, qdiff_sample) qdiffmin_q; STAILQ_HEAD(maxrtts_head, qdiff_sample) qdiffmax_q; long window_incr; /* rttcount for window increase when in congestion avoidance */ long rtt_count; /* maximum measured rtt within an rtt period */ int maxrtt_in_rtt; /* maximum measured rtt within prev rtt period */ int maxrtt_in_prevrtt; /* minimum measured rtt within an rtt period */ int minrtt_in_rtt; /* minimum measured rtt within prev rtt period */ int minrtt_in_prevrtt; /* consecutive congestion episode counter */ uint32_t consec_cong_cnt; /* when tracking a new reno type loss window */ uint32_t shadow_w; /* maximum number of samples in the moving average queue */ int sample_q_size; /* number of samples in the moving average queue */ int num_samples; /* estimate of the queue state of the path */ int queue_state; }; /* * Lookup table for: * (1 - exp(-x)) << EXP_PREC, where x = [0,MAXGRAD] in 2^-7 increments * * Note: probexp[0] is set to 10 (not 0) as a safety for very low increase * gradients. */ static const int probexp[641] = { 10,255,508,759,1008,1255,1501,1744,1985,2225,2463,2698,2932,3165,3395,3624, 3850,4075,4299,4520,4740,4958,5175,5389,5602,5814,6024,6232,6438,6643,6846, 7048,7248,7447,7644,7839,8033,8226,8417,8606,8794,8981,9166,9350,9532,9713, 9892,10070,10247,10422,10596,10769,10940,11110,11278,11445,11611,11776,11939, 12101,12262,12422,12580,12737,12893,13048,13201,13354,13505,13655,13803,13951, 14097,14243,14387,14530,14672,14813,14952,15091,15229,15365,15500,15635,15768, 15900,16032,16162,16291,16419,16547,16673,16798,16922,17046,17168,17289,17410, 17529,17648,17766,17882,17998,18113,18227,18340,18453,18564,18675,18784,18893, 19001,19108,19215,19320,19425,19529,19632,19734,19835,19936,20036,20135,20233, 20331,20427,20523,20619,20713,20807,20900,20993,21084,21175,21265,21355,21444, 21532,21619,21706,21792,21878,21962,22046,22130,22213,22295,22376,22457,22537, 22617,22696,22774,22852,22929,23006,23082,23157,23232,23306,23380,23453,23525, 23597,23669,23739,23810,23879,23949,24017,24085,24153,24220,24286,24352,24418, 24483,24547,24611,24675,24738,24800,24862,24924,24985,25045,25106,25165,25224, 25283,25341,25399,25456,25513,25570,25626,25681,25737,25791,25846,25899,25953, 26006,26059,26111,26163,26214,26265,26316,26366,26416,26465,26514,26563,26611, 26659,26707,26754,26801,26847,26893,26939,26984,27029,27074,27118,27162,27206, 27249,27292,27335,27377,27419,27460,27502,27543,27583,27624,27664,27703,27743, 27782,27821,27859,27897,27935,27973,28010,28047,28084,28121,28157,28193,28228, 28263,28299,28333,28368,28402,28436,28470,28503,28536,28569,28602,28634,28667, 28699,28730,28762,28793,28824,28854,28885,28915,28945,28975,29004,29034,29063, 29092,29120,29149,29177,29205,29232,29260,29287,29314,29341,29368,29394,29421, 29447,29472,29498,29524,29549,29574,29599,29623,29648,29672,29696,29720,29744, 29767,29791,29814,29837,29860,29882,29905,29927,29949,29971,29993,30014,30036, 30057,30078,30099,30120,30141,30161,30181,30201,30221,30241,30261,30280,30300, 30319,30338,30357,30376,30394,30413,30431,30449,30467,30485,30503,30521,30538, 30555,30573,30590,30607,30624,30640,30657,30673,30690,30706,30722,30738,30753, 30769,30785,30800,30815,30831,30846,30861,30876,30890,30905,30919,30934,30948, 30962,30976,30990,31004,31018,31031,31045,31058,31072,31085,31098,31111,31124, 31137,31149,31162,31174,31187,31199,31211,31223,31235,31247,31259,31271,31283, 31294,31306,31317,31328,31339,31351,31362,31373,31383,31394,31405,31416,31426, 31436,31447,31457,31467,31477,31487,31497,31507,31517,31527,31537,31546,31556, 31565,31574,31584,31593,31602,31611,31620,31629,31638,31647,31655,31664,31673, 31681,31690,31698,31706,31715,31723,31731,31739,31747,31755,31763,31771,31778, 31786,31794,31801,31809,31816,31824,31831,31838,31846,31853,31860,31867,31874, 31881,31888,31895,31902,31908,31915,31922,31928,31935,31941,31948,31954,31960, 31967,31973,31979,31985,31991,31997,32003,32009,32015,32021,32027,32033,32038, 32044,32050,32055,32061,32066,32072,32077,32083,32088,32093,32098,32104,32109, 32114,32119,32124,32129,32134,32139,32144,32149,32154,32158,32163,32168,32173, 32177,32182,32186,32191,32195,32200,32204,32209,32213,32217,32222,32226,32230, 32234,32238,32242,32247,32251,32255,32259,32263,32267,32270,32274,32278,32282, 32286,32290,32293,32297,32301,32304,32308,32311,32315,32318,32322,32325,32329, 32332,32336,32339,32342,32346,32349,32352,32356,32359,32362,32365,32368,32371, 32374,32377,32381,32384,32387,32389,32392,32395,32398,32401,32404,32407,32410, 32412,32415,32418,32421,32423,32426,32429,32431,32434,32437,32439,32442,32444, 32447,32449,32452,32454,32457,32459,32461,32464,32466,32469,32471,32473,32476, 32478,32480,32482,32485,32487,32489,32491,32493,32495,32497,32500,32502,32504, 32506,32508,32510,32512,32514,32516,32518,32520,32522,32524,32526,32527,32529, 32531,32533,32535,32537,32538,32540,32542,32544,32545,32547}; static uma_zone_t qdiffsample_zone; static MALLOC_DEFINE(M_CDG, "cdg data", "Per connection data required for the CDG congestion control algorithm"); static int ertt_id; static VNET_DEFINE(uint32_t, cdg_alpha_inc); static VNET_DEFINE(uint32_t, cdg_beta_delay); static VNET_DEFINE(uint32_t, cdg_beta_loss); static VNET_DEFINE(uint32_t, cdg_smoothing_factor); static VNET_DEFINE(uint32_t, cdg_exp_backoff_scale); static VNET_DEFINE(uint32_t, cdg_consec_cong); static VNET_DEFINE(uint32_t, cdg_hold_backoff); #define V_cdg_alpha_inc VNET(cdg_alpha_inc) #define V_cdg_beta_delay VNET(cdg_beta_delay) #define V_cdg_beta_loss VNET(cdg_beta_loss) #define V_cdg_smoothing_factor VNET(cdg_smoothing_factor) #define V_cdg_exp_backoff_scale VNET(cdg_exp_backoff_scale) #define V_cdg_consec_cong VNET(cdg_consec_cong) #define V_cdg_hold_backoff VNET(cdg_hold_backoff) /* Function prototypes. */ static int cdg_mod_init(void); static int cdg_mod_destroy(void); static void cdg_conn_init(struct cc_var *ccv); static int cdg_cb_init(struct cc_var *ccv); static void cdg_cb_destroy(struct cc_var *ccv); static void cdg_cong_signal(struct cc_var *ccv, uint32_t signal_type); static void cdg_ack_received(struct cc_var *ccv, uint16_t ack_type); struct cc_algo cdg_cc_algo = { .name = "cdg", .mod_init = cdg_mod_init, .ack_received = cdg_ack_received, .cb_destroy = cdg_cb_destroy, .cb_init = cdg_cb_init, .conn_init = cdg_conn_init, .cong_signal = cdg_cong_signal, .mod_destroy = cdg_mod_destroy }; /* Vnet created and being initialised. */ static void cdg_init_vnet(const void *unused __unused) { V_cdg_alpha_inc = 0; V_cdg_beta_delay = 70; V_cdg_beta_loss = 50; V_cdg_smoothing_factor = 8; V_cdg_exp_backoff_scale = 3; V_cdg_consec_cong = 5; V_cdg_hold_backoff = 5; } static int cdg_mod_init(void) { VNET_ITERATOR_DECL(v); ertt_id = khelp_get_id("ertt"); if (ertt_id <= 0) return (EINVAL); qdiffsample_zone = uma_zcreate("cdg_qdiffsample", sizeof(struct qdiff_sample), NULL, NULL, NULL, NULL, 0, 0); VNET_LIST_RLOCK(); VNET_FOREACH(v) { CURVNET_SET(v); cdg_init_vnet(NULL); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); cdg_cc_algo.post_recovery = newreno_cc_algo.post_recovery; cdg_cc_algo.after_idle = newreno_cc_algo.after_idle; return (0); } static int cdg_mod_destroy(void) { uma_zdestroy(qdiffsample_zone); return (0); } static int cdg_cb_init(struct cc_var *ccv) { struct cdg *cdg_data; cdg_data = malloc(sizeof(struct cdg), M_CDG, M_NOWAIT); if (cdg_data == NULL) return (ENOMEM); cdg_data->shadow_w = 0; cdg_data->max_qtrend = 0; cdg_data->min_qtrend = 0; cdg_data->queue_state = CDG_Q_UNKNOWN; cdg_data->maxrtt_in_rtt = 0; cdg_data->maxrtt_in_prevrtt = 0; cdg_data->minrtt_in_rtt = INT_MAX; cdg_data->minrtt_in_prevrtt = 0; cdg_data->window_incr = 0; cdg_data->rtt_count = 0; cdg_data->consec_cong_cnt = 0; cdg_data->sample_q_size = V_cdg_smoothing_factor; cdg_data->num_samples = 0; STAILQ_INIT(&cdg_data->qdiffmin_q); STAILQ_INIT(&cdg_data->qdiffmax_q); ccv->cc_data = cdg_data; return (0); } static void cdg_conn_init(struct cc_var *ccv) { struct cdg *cdg_data = ccv->cc_data; /* * Initialise the shadow_cwnd in case we are competing with loss based * flows from the start */ cdg_data->shadow_w = CCV(ccv, snd_cwnd); } static void cdg_cb_destroy(struct cc_var *ccv) { struct cdg *cdg_data; struct qdiff_sample *qds, *qds_n; cdg_data = ccv->cc_data; qds = STAILQ_FIRST(&cdg_data->qdiffmin_q); while (qds != NULL) { qds_n = STAILQ_NEXT(qds, qdiff_lnk); uma_zfree(qdiffsample_zone,qds); qds = qds_n; } qds = STAILQ_FIRST(&cdg_data->qdiffmax_q); while (qds != NULL) { qds_n = STAILQ_NEXT(qds, qdiff_lnk); uma_zfree(qdiffsample_zone,qds); qds = qds_n; } free(ccv->cc_data, M_CDG); } static int cdg_beta_handler(SYSCTL_HANDLER_ARGS) { if (req->newptr != NULL && (CAST_PTR_INT(req->newptr) == 0 || CAST_PTR_INT(req->newptr) > 100)) return (EINVAL); return (sysctl_handle_int(oidp, arg1, arg2, req)); } static int cdg_exp_backoff_scale_handler(SYSCTL_HANDLER_ARGS) { if (req->newptr != NULL && CAST_PTR_INT(req->newptr) < 1) return (EINVAL); return (sysctl_handle_int(oidp, arg1, arg2, req)); } static inline uint32_t cdg_window_decrease(struct cc_var *ccv, unsigned long owin, unsigned int beta) { return ((ulmin(CCV(ccv, snd_wnd), owin) * beta) / 100); } /* * Window increase function * This window increase function is independent of the initial window size * to ensure small window flows are not discriminated against (i.e. fairness). * It increases at 1pkt/rtt like Reno for alpha_inc rtts, and then 2pkts/rtt for * the next alpha_inc rtts, etc. */ static void cdg_window_increase(struct cc_var *ccv, int new_measurement) { struct cdg *cdg_data; int incr, s_w_incr; cdg_data = ccv->cc_data; incr = s_w_incr = 0; if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) { /* Slow start. */ incr = CCV(ccv, t_maxseg); s_w_incr = incr; cdg_data->window_incr = cdg_data->rtt_count = 0; } else { /* Congestion avoidance. */ if (new_measurement) { s_w_incr = CCV(ccv, t_maxseg); if (V_cdg_alpha_inc == 0) { incr = CCV(ccv, t_maxseg); } else { if (++cdg_data->rtt_count >= V_cdg_alpha_inc) { cdg_data->window_incr++; cdg_data->rtt_count = 0; } incr = CCV(ccv, t_maxseg) * cdg_data->window_incr; } } } if (cdg_data->shadow_w > 0) cdg_data->shadow_w = ulmin(cdg_data->shadow_w + s_w_incr, TCP_MAXWIN << CCV(ccv, snd_scale)); CCV(ccv, snd_cwnd) = ulmin(CCV(ccv, snd_cwnd) + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } static void cdg_cong_signal(struct cc_var *ccv, uint32_t signal_type) { struct cdg *cdg_data = ccv->cc_data; - uint32_t cwin; - u_int mss; - cwin = CCV(ccv, snd_cwnd); - mss = CCV(ccv, t_maxseg); - switch(signal_type) { case CC_CDG_DELAY: CCV(ccv, snd_ssthresh) = cdg_window_decrease(ccv, CCV(ccv, snd_cwnd), V_cdg_beta_delay); CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); CCV(ccv, snd_recover) = CCV(ccv, snd_max); cdg_data->window_incr = cdg_data->rtt_count = 0; ENTER_CONGRECOVERY(CCV(ccv, t_flags)); break; case CC_NDUPACK: /* * If already responding to congestion OR we have guessed no * queue in the path is full. */ if (IN_CONGRECOVERY(CCV(ccv, t_flags)) || cdg_data->queue_state < CDG_Q_FULL) { - CCV(ccv, snd_ssthresh) = cwin; + CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); CCV(ccv, snd_recover) = CCV(ccv, snd_max); } else { /* * Loss is likely to be congestion related. We have * inferred a queue full state, so have shadow window * react to loss as NewReno would. */ if (cdg_data->shadow_w > 0) cdg_data->shadow_w = cdg_window_decrease(ccv, cdg_data->shadow_w, RENO_BETA); CCV(ccv, snd_ssthresh) = max(cdg_data->shadow_w, - cdg_window_decrease(ccv, cwin, V_cdg_beta_loss)); - CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); + cdg_window_decrease(ccv, CCV(ccv, snd_cwnd), + V_cdg_beta_loss)); cdg_data->window_incr = cdg_data->rtt_count = 0; } ENTER_RECOVERY(CCV(ccv, t_flags)); - break; - case CC_RTO: - CCV(ccv, snd_ssthresh) = max(2*mss, cwin/2); - CCV(ccv, snd_cwnd) = mss; break; default: newreno_cc_algo.cong_signal(ccv, signal_type); break; } } /* * Using a negative exponential probabilistic backoff so that sources with * varying RTTs which share the same link will, on average, have the same * probability of backoff over time. * * Prob_backoff = 1 - exp(-qtrend / V_cdg_exp_backoff_scale), where * V_cdg_exp_backoff_scale is the average qtrend for the exponential backoff. */ static inline int prob_backoff(long qtrend) { int backoff, idx, p; backoff = (qtrend > ((MAXGRAD * V_cdg_exp_backoff_scale) << D_P_E)); if (!backoff) { if (V_cdg_exp_backoff_scale > 1) idx = (qtrend + V_cdg_exp_backoff_scale / 2) / V_cdg_exp_backoff_scale; else idx = qtrend; /* Backoff probability proportional to rate of queue growth. */ p = (INT_MAX / (1 << EXP_PREC)) * probexp[idx]; backoff = (random() < p); } return (backoff); } static inline void calc_moving_average(struct cdg *cdg_data, long qdiff_max, long qdiff_min) { struct qdiff_sample *qds; ++cdg_data->num_samples; if (cdg_data->num_samples > cdg_data->sample_q_size) { /* Minimum RTT. */ qds = STAILQ_FIRST(&cdg_data->qdiffmin_q); cdg_data->min_qtrend = cdg_data->min_qtrend + (qdiff_min - qds->qdiff) / cdg_data->sample_q_size; STAILQ_REMOVE_HEAD(&cdg_data->qdiffmin_q, qdiff_lnk); qds->qdiff = qdiff_min; STAILQ_INSERT_TAIL(&cdg_data->qdiffmin_q, qds, qdiff_lnk); /* Maximum RTT. */ qds = STAILQ_FIRST(&cdg_data->qdiffmax_q); cdg_data->max_qtrend = cdg_data->max_qtrend + (qdiff_max - qds->qdiff) / cdg_data->sample_q_size; STAILQ_REMOVE_HEAD(&cdg_data->qdiffmax_q, qdiff_lnk); qds->qdiff = qdiff_max; STAILQ_INSERT_TAIL(&cdg_data->qdiffmax_q, qds, qdiff_lnk); --cdg_data->num_samples; } else { qds = uma_zalloc(qdiffsample_zone, M_NOWAIT); if (qds != NULL) { cdg_data->min_qtrend = cdg_data->min_qtrend + qdiff_min / cdg_data->sample_q_size; qds->qdiff = qdiff_min; STAILQ_INSERT_TAIL(&cdg_data->qdiffmin_q, qds, qdiff_lnk); } qds = uma_zalloc(qdiffsample_zone, M_NOWAIT); if (qds) { cdg_data->max_qtrend = cdg_data->max_qtrend + qdiff_max / cdg_data->sample_q_size; qds->qdiff = qdiff_max; STAILQ_INSERT_TAIL(&cdg_data->qdiffmax_q, qds, qdiff_lnk); } } } static void cdg_ack_received(struct cc_var *ccv, uint16_t ack_type) { struct cdg *cdg_data; struct ertt *e_t; long qdiff_max, qdiff_min; int congestion, new_measurement, slowstart; cdg_data = ccv->cc_data; e_t = (struct ertt *)khelp_get_osd(CCV(ccv, osd), ertt_id); new_measurement = e_t->flags & ERTT_NEW_MEASUREMENT; congestion = 0; cdg_data->maxrtt_in_rtt = imax(e_t->rtt, cdg_data->maxrtt_in_rtt); cdg_data->minrtt_in_rtt = imin(e_t->rtt, cdg_data->minrtt_in_rtt); if (new_measurement) { slowstart = (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)); /* * Update smoothed gradient measurements. Since we are only * using one measurement per RTT, use max or min rtt_in_rtt. * This is also less noisy than a sample RTT measurement. Max * RTT measurements can have trouble due to OS issues. */ if (cdg_data->maxrtt_in_prevrtt) { qdiff_max = ((long)(cdg_data->maxrtt_in_rtt - cdg_data->maxrtt_in_prevrtt) << D_P_E ); qdiff_min = ((long)(cdg_data->minrtt_in_rtt - cdg_data->minrtt_in_prevrtt) << D_P_E ); calc_moving_average(cdg_data, qdiff_max, qdiff_min); /* Probabilistic backoff with respect to gradient. */ if (slowstart && qdiff_min > 0) congestion = prob_backoff(qdiff_min); else if (cdg_data->min_qtrend > 0) congestion = prob_backoff(cdg_data->min_qtrend); else if (slowstart && qdiff_max > 0) congestion = prob_backoff(qdiff_max); else if (cdg_data->max_qtrend > 0) congestion = prob_backoff(cdg_data->max_qtrend); /* Update estimate of queue state. */ if (cdg_data->min_qtrend > 0 && cdg_data->max_qtrend <= 0) { cdg_data->queue_state = CDG_Q_FULL; } else if (cdg_data->min_qtrend >= 0 && cdg_data->max_qtrend < 0) { cdg_data->queue_state = CDG_Q_EMPTY; cdg_data->shadow_w = 0; } else if (cdg_data->min_qtrend > 0 && cdg_data->max_qtrend > 0) { cdg_data->queue_state = CDG_Q_RISING; } else if (cdg_data->min_qtrend < 0 && cdg_data->max_qtrend < 0) { cdg_data->queue_state = CDG_Q_FALLING; } if (cdg_data->min_qtrend < 0 || cdg_data->max_qtrend < 0) cdg_data->consec_cong_cnt = 0; } cdg_data->minrtt_in_prevrtt = cdg_data->minrtt_in_rtt; cdg_data->minrtt_in_rtt = INT_MAX; cdg_data->maxrtt_in_prevrtt = cdg_data->maxrtt_in_rtt; cdg_data->maxrtt_in_rtt = 0; e_t->flags &= ~ERTT_NEW_MEASUREMENT; } if (congestion) { cdg_data->consec_cong_cnt++; if (!IN_RECOVERY(CCV(ccv, t_flags))) { if (cdg_data->consec_cong_cnt <= V_cdg_consec_cong) cdg_cong_signal(ccv, CC_CDG_DELAY); else /* * We have been backing off but the queue is not * falling. Assume we are competing with * loss-based flows and don't back off for the * next V_cdg_hold_backoff RTT periods. */ if (cdg_data->consec_cong_cnt >= V_cdg_consec_cong + V_cdg_hold_backoff) cdg_data->consec_cong_cnt = 0; /* Won't see effect until 2nd RTT. */ cdg_data->maxrtt_in_prevrtt = 0; /* * Resync shadow window in case we are competing with a * loss based flow */ cdg_data->shadow_w = ulmax(CCV(ccv, snd_cwnd), cdg_data->shadow_w); } } else if (ack_type == CC_ACK) cdg_window_increase(ccv, new_measurement); } /* When a vnet is created and being initialised, init the per-stack CDG vars. */ VNET_SYSINIT(cdg_init_vnet, SI_SUB_PROTO_BEGIN, SI_ORDER_FIRST, cdg_init_vnet, NULL); SYSCTL_DECL(_net_inet_tcp_cc_cdg); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, cdg, CTLFLAG_RW, NULL, "CAIA delay-gradient congestion control related settings"); SYSCTL_STRING(_net_inet_tcp_cc_cdg, OID_AUTO, version, CTLFLAG_RD, CDG_VERSION, sizeof(CDG_VERSION) - 1, "Current algorithm/implementation version number"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, alpha_inc, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_alpha_inc), 0, "Increment the window increase factor alpha by 1 MSS segment every " "alpha_inc RTTs during congestion avoidance mode."); SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, beta_delay, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_beta_delay), 70, &cdg_beta_handler, "IU", "Delay-based window decrease factor as a percentage " "(on delay-based backoff, w = w * beta_delay / 100)"); SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, beta_loss, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_beta_loss), 50, &cdg_beta_handler, "IU", "Loss-based window decrease factor as a percentage " "(on loss-based backoff, w = w * beta_loss / 100)"); SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, exp_backoff_scale, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_exp_backoff_scale), 2, &cdg_exp_backoff_scale_handler, "IU", "Scaling parameter for the probabilistic exponential backoff"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, smoothing_factor, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_smoothing_factor), 8, "Number of samples used for moving average smoothing (0 = no smoothing)"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, loss_compete_consec_cong, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_consec_cong), 5, "Number of consecutive delay-gradient based congestion episodes which will " "trigger loss based CC compatibility"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, loss_compete_hold_backoff, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_hold_backoff), 5, "Number of consecutive delay-gradient based congestion episodes to hold " "the window backoff for loss based CC compatibility"); DECLARE_CC_MODULE(cdg, &cdg_cc_algo); MODULE_DEPEND(cdg, ertt, 1, 1, 1); Index: head/sys/netinet/cc/cc_chd.c =================================================================== --- head/sys/netinet/cc/cc_chd.c (revision 307899) +++ head/sys/netinet/cc/cc_chd.c (revision 307900) @@ -1,506 +1,498 @@ /*- * Copyright (c) 2009-2010 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2010-2011 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by David Hayes and * Lawrence Stewart, made possible in part by a grant from the Cisco University * Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, Melbourne, Australia by * David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the CAIA-Hamilton delay based congestion control * algorithm, based on "Improved coexistence and loss tolerance for delay based * TCP congestion control" by D. A. Hayes and G. Armitage., in 35th Annual IEEE * Conference on Local Computer Networks (LCN 2010), Denver, Colorado, USA, * 11-14 October 2010. * * Originally released as part of the NewTCP research project at Swinburne * University of Technology's Centre for Advanced Internet Architectures, * Melbourne, Australia, which was made possible in part by a grant from the * Cisco University Research Program Fund at Community Foundation Silicon * Valley. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CAST_PTR_INT(X) (*((int*)(X))) /* * Private signal type for rate based congestion signal. * See for appropriate bit-range to use for private signals. */ #define CC_CHD_DELAY 0x02000000 /* Largest possible number returned by random(). */ #define RANDOM_MAX INT_MAX static void chd_ack_received(struct cc_var *ccv, uint16_t ack_type); static void chd_cb_destroy(struct cc_var *ccv); static int chd_cb_init(struct cc_var *ccv); static void chd_cong_signal(struct cc_var *ccv, uint32_t signal_type); static void chd_conn_init(struct cc_var *ccv); static int chd_mod_init(void); struct chd { /* * Shadow window - keeps track of what the NewReno congestion window * would have been if delay-based cwnd backoffs had not been made. This * functionality aids coexistence with loss-based TCP flows which may be * sharing links along the path. */ unsigned long shadow_w; /* * Loss-based TCP compatibility flag - When set, it turns on the shadow * window functionality. */ int loss_compete; /* The maximum round trip time seen within a measured rtt period. */ int maxrtt_in_rtt; /* The previous qdly that caused cwnd to backoff. */ int prev_backoff_qdly; }; static int ertt_id; static VNET_DEFINE(uint32_t, chd_qmin) = 5; static VNET_DEFINE(uint32_t, chd_pmax) = 50; static VNET_DEFINE(uint32_t, chd_loss_fair) = 1; static VNET_DEFINE(uint32_t, chd_use_max) = 1; static VNET_DEFINE(uint32_t, chd_qthresh) = 20; #define V_chd_qthresh VNET(chd_qthresh) #define V_chd_qmin VNET(chd_qmin) #define V_chd_pmax VNET(chd_pmax) #define V_chd_loss_fair VNET(chd_loss_fair) #define V_chd_use_max VNET(chd_use_max) static MALLOC_DEFINE(M_CHD, "chd data", "Per connection data required for the CHD congestion control algorithm"); struct cc_algo chd_cc_algo = { .name = "chd", .ack_received = chd_ack_received, .cb_destroy = chd_cb_destroy, .cb_init = chd_cb_init, .cong_signal = chd_cong_signal, .conn_init = chd_conn_init, .mod_init = chd_mod_init }; static __inline void chd_window_decrease(struct cc_var *ccv) { unsigned long win; win = min(CCV(ccv, snd_wnd), CCV(ccv, snd_cwnd)) / CCV(ccv, t_maxseg); win -= max((win / 2), 1); CCV(ccv, snd_ssthresh) = max(win, 2) * CCV(ccv, t_maxseg); } /* * Probabilistic backoff function. Returns 1 if we should backoff or 0 * otherwise. The calculation of p is similar to the calculation of p in cc_hd. */ static __inline int should_backoff(int qdly, int maxqdly, struct chd *chd_data) { unsigned long p, rand; rand = random(); if (qdly < V_chd_qthresh) { chd_data->loss_compete = 0; p = (((RANDOM_MAX / 100) * V_chd_pmax) / (V_chd_qthresh - V_chd_qmin)) * (qdly - V_chd_qmin); } else { if (qdly > V_chd_qthresh) { p = (((RANDOM_MAX / 100) * V_chd_pmax) / (maxqdly - V_chd_qthresh)) * (maxqdly - qdly); if (V_chd_loss_fair && rand < p) chd_data->loss_compete = 1; } else { p = (RANDOM_MAX / 100) * V_chd_pmax; chd_data->loss_compete = 0; } } return (rand < p); } static __inline void chd_window_increase(struct cc_var *ccv, int new_measurement) { struct chd *chd_data; int incr; chd_data = ccv->cc_data; incr = 0; if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) { /* Adapted from NewReno slow start. */ if (V_tcp_do_rfc3465) { /* In slow-start with ABC enabled. */ if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) { /* Not due to RTO. */ incr = min(ccv->bytes_this_ack, V_tcp_abc_l_var * CCV(ccv, t_maxseg)); } else { /* Due to RTO. */ incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); } } else incr = CCV(ccv, t_maxseg); } else { /* Congestion avoidance. */ if (V_tcp_do_rfc3465) { if (ccv->flags & CCF_ABC_SENTAWND) { ccv->flags &= ~CCF_ABC_SENTAWND; incr = CCV(ccv, t_maxseg); } } else if (new_measurement) incr = CCV(ccv, t_maxseg); } if (chd_data->shadow_w > 0) { /* Track NewReno window. */ chd_data->shadow_w = min(chd_data->shadow_w + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } CCV(ccv,snd_cwnd) = min(CCV(ccv, snd_cwnd) + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } /* * All ACK signals are used for timing measurements to determine delay-based * congestion. However, window increases are only performed when * ack_type == CC_ACK. */ static void chd_ack_received(struct cc_var *ccv, uint16_t ack_type) { struct chd *chd_data; struct ertt *e_t; int backoff, new_measurement, qdly, rtt; e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); chd_data = ccv->cc_data; new_measurement = e_t->flags & ERTT_NEW_MEASUREMENT; backoff = qdly = 0; chd_data->maxrtt_in_rtt = imax(e_t->rtt, chd_data->maxrtt_in_rtt); if (new_measurement) { /* * There is a new per RTT measurement, so check to see if there * is delay based congestion. */ rtt = V_chd_use_max ? chd_data->maxrtt_in_rtt : e_t->rtt; chd_data->maxrtt_in_rtt = 0; if (rtt && e_t->minrtt && !IN_RECOVERY(CCV(ccv, t_flags))) { qdly = rtt - e_t->minrtt; if (qdly > V_chd_qmin) { /* * Probabilistic delay based congestion * indication. */ backoff = should_backoff(qdly, e_t->maxrtt - e_t->minrtt, chd_data); } else chd_data->loss_compete = 0; } /* Reset per RTT measurement flag to start a new measurement. */ e_t->flags &= ~ERTT_NEW_MEASUREMENT; } if (backoff) { /* * Update shadow_w before delay based backoff. */ if (chd_data->loss_compete || qdly > chd_data->prev_backoff_qdly) { /* * Delay is higher than when we backed off previously, * so it is possible that this flow is competing with * loss based flows. */ chd_data->shadow_w = max(CCV(ccv, snd_cwnd), chd_data->shadow_w); } else { /* * Reset shadow_w, as it is probable that this flow is * not competing with loss based flows at the moment. */ chd_data->shadow_w = 0; } chd_data->prev_backoff_qdly = qdly; /* * Send delay-based congestion signal to the congestion signal * handler. */ chd_cong_signal(ccv, CC_CHD_DELAY); } else if (ack_type == CC_ACK) chd_window_increase(ccv, new_measurement); } static void chd_cb_destroy(struct cc_var *ccv) { if (ccv->cc_data != NULL) free(ccv->cc_data, M_CHD); } static int chd_cb_init(struct cc_var *ccv) { struct chd *chd_data; chd_data = malloc(sizeof(struct chd), M_CHD, M_NOWAIT); if (chd_data == NULL) return (ENOMEM); chd_data->shadow_w = 0; ccv->cc_data = chd_data; return (0); } static void chd_cong_signal(struct cc_var *ccv, uint32_t signal_type) { struct ertt *e_t; struct chd *chd_data; int qdly; - uint32_t cwin; - u_int mss; e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); chd_data = ccv->cc_data; qdly = imax(e_t->rtt, chd_data->maxrtt_in_rtt) - e_t->minrtt; - cwin = CCV(ccv, snd_cwnd); - mss = CCV(ccv, t_maxseg); switch(signal_type) { case CC_CHD_DELAY: chd_window_decrease(ccv); /* Set new ssthresh. */ CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); CCV(ccv, snd_recover) = CCV(ccv, snd_max); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); break; case CC_NDUPACK: /* Packet loss. */ /* * Only react to loss as a congestion signal if qdly > * V_chd_qthresh. If qdly is less than qthresh, presume that * this is a non congestion related loss. If qdly is greater * than qthresh, assume that we are competing with loss based * tcp flows and restore window from any unnecessary backoffs, * before the decrease. */ if (!IN_RECOVERY(CCV(ccv, t_flags)) && qdly > V_chd_qthresh) { if (chd_data->loss_compete) { CCV(ccv, snd_cwnd) = max(CCV(ccv, snd_cwnd), chd_data->shadow_w); } chd_window_decrease(ccv); } else { /* * This loss isn't congestion related, or already * recovering from congestion. */ CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); CCV(ccv, snd_recover) = CCV(ccv, snd_max); } if (chd_data->shadow_w > 0) { chd_data->shadow_w = max(chd_data->shadow_w / CCV(ccv, t_maxseg) / 2, 2) * CCV(ccv, t_maxseg); } ENTER_FASTRECOVERY(CCV(ccv, t_flags)); - break; - case CC_RTO: - CCV(ccv, snd_ssthresh) = max(2*mss, cwin/2); - CCV(ccv, snd_cwnd) = mss; break; default: newreno_cc_algo.cong_signal(ccv, signal_type); } } static void chd_conn_init(struct cc_var *ccv) { struct chd *chd_data; chd_data = ccv->cc_data; chd_data->prev_backoff_qdly = 0; chd_data->maxrtt_in_rtt = 0; chd_data->loss_compete = 0; /* * Initialise the shadow_cwnd to be equal to snd_cwnd in case we are * competing with loss based flows from the start. */ chd_data->shadow_w = CCV(ccv, snd_cwnd); } static int chd_mod_init(void) { ertt_id = khelp_get_id("ertt"); if (ertt_id <= 0) { printf("%s: h_ertt module not found\n", __func__); return (ENOENT); } chd_cc_algo.after_idle = newreno_cc_algo.after_idle; chd_cc_algo.post_recovery = newreno_cc_algo.post_recovery; return (0); } static int chd_loss_fair_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_chd_loss_fair; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else V_chd_loss_fair = new; } return (error); } static int chd_pmax_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_chd_pmax; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) == 0 || CAST_PTR_INT(req->newptr) > 100) error = EINVAL; else V_chd_pmax = new; } return (error); } static int chd_qthresh_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_chd_qthresh; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) <= V_chd_qmin) error = EINVAL; else V_chd_qthresh = new; } return (error); } SYSCTL_DECL(_net_inet_tcp_cc_chd); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, chd, CTLFLAG_RW, NULL, "CAIA Hamilton delay-based congestion control related settings"); SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, loss_fair, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(chd_loss_fair), 1, &chd_loss_fair_handler, "IU", "Flag to enable shadow window functionality."); SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, pmax, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(chd_pmax), 5, &chd_pmax_handler, "IU", "Per RTT maximum backoff probability as a percentage"); SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, queue_threshold, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(chd_qthresh), 20, &chd_qthresh_handler, "IU", "Queueing congestion threshold in ticks"); SYSCTL_UINT(_net_inet_tcp_cc_chd, OID_AUTO, queue_min, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_qmin), 5, "Minimum queueing delay threshold in ticks"); SYSCTL_UINT(_net_inet_tcp_cc_chd, OID_AUTO, use_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_use_max), 1, "Use the maximum RTT seen within the measurement period (RTT) " "as the basic delay measurement for the algorithm."); DECLARE_CC_MODULE(chd, &chd_cc_algo); MODULE_DEPEND(chd, ertt, 1, 1, 1); Index: head/sys/netinet/cc/cc_cubic.c =================================================================== --- head/sys/netinet/cc/cc_cubic.c (revision 307899) +++ head/sys/netinet/cc/cc_cubic.c (revision 307900) @@ -1,420 +1,411 @@ /*- * Copyright (c) 2008-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * All rights reserved. * * This software was developed by Lawrence Stewart while studying at the Centre * for Advanced Internet Architectures, Swinburne University of Technology, made * possible in part by a grant from the Cisco University Research Program Fund * at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the CUBIC congestion control algorithm for FreeBSD, * based on the Internet Draft "draft-rhee-tcpm-cubic-02" by Rhee, Xu and Ha. * Originally released as part of the NewTCP research project at Swinburne * University of Technology's Centre for Advanced Internet Architectures, * Melbourne, Australia, which was made possible in part by a grant from the * Cisco University Research Program Fund at Community Foundation Silicon * Valley. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static void cubic_ack_received(struct cc_var *ccv, uint16_t type); static void cubic_cb_destroy(struct cc_var *ccv); static int cubic_cb_init(struct cc_var *ccv); static void cubic_cong_signal(struct cc_var *ccv, uint32_t type); static void cubic_conn_init(struct cc_var *ccv); static int cubic_mod_init(void); static void cubic_post_recovery(struct cc_var *ccv); static void cubic_record_rtt(struct cc_var *ccv); static void cubic_ssthresh_update(struct cc_var *ccv); struct cubic { /* Cubic K in fixed point form with CUBIC_SHIFT worth of precision. */ int64_t K; /* Sum of RTT samples across an epoch in ticks. */ int64_t sum_rtt_ticks; /* cwnd at the most recent congestion event. */ unsigned long max_cwnd; /* cwnd at the previous congestion event. */ unsigned long prev_max_cwnd; /* Number of congestion events. */ uint32_t num_cong_events; /* Minimum observed rtt in ticks. */ int min_rtt_ticks; /* Mean observed rtt between congestion epochs. */ int mean_rtt_ticks; /* ACKs since last congestion event. */ int epoch_ack_count; /* Time of last congestion event in ticks. */ int t_last_cong; }; static MALLOC_DEFINE(M_CUBIC, "cubic data", "Per connection data required for the CUBIC congestion control algorithm"); struct cc_algo cubic_cc_algo = { .name = "cubic", .ack_received = cubic_ack_received, .cb_destroy = cubic_cb_destroy, .cb_init = cubic_cb_init, .cong_signal = cubic_cong_signal, .conn_init = cubic_conn_init, .mod_init = cubic_mod_init, .post_recovery = cubic_post_recovery, }; static void cubic_ack_received(struct cc_var *ccv, uint16_t type) { struct cubic *cubic_data; unsigned long w_tf, w_cubic_next; int ticks_since_cong; cubic_data = ccv->cc_data; cubic_record_rtt(ccv); /* * Regular ACK and we're not in cong/fast recovery and we're cwnd * limited and we're either not doing ABC or are slow starting or are * doing ABC and we've sent a cwnd's worth of bytes. */ if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && (ccv->flags & CCF_CWND_LIMITED) && (!V_tcp_do_rfc3465 || CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) || (V_tcp_do_rfc3465 && ccv->flags & CCF_ABC_SENTAWND))) { /* Use the logic in NewReno ack_received() for slow start. */ if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) || cubic_data->min_rtt_ticks == TCPTV_SRTTBASE) newreno_cc_algo.ack_received(ccv, type); else { ticks_since_cong = ticks - cubic_data->t_last_cong; /* * The mean RTT is used to best reflect the equations in * the I-D. Using min_rtt in the tf_cwnd calculation * causes w_tf to grow much faster than it should if the * RTT is dominated by network buffering rather than * propagation delay. */ w_tf = tf_cwnd(ticks_since_cong, cubic_data->mean_rtt_ticks, cubic_data->max_cwnd, CCV(ccv, t_maxseg)); w_cubic_next = cubic_cwnd(ticks_since_cong + cubic_data->mean_rtt_ticks, cubic_data->max_cwnd, CCV(ccv, t_maxseg), cubic_data->K); ccv->flags &= ~CCF_ABC_SENTAWND; if (w_cubic_next < w_tf) /* * TCP-friendly region, follow tf * cwnd growth. */ CCV(ccv, snd_cwnd) = w_tf; else if (CCV(ccv, snd_cwnd) < w_cubic_next) { /* * Concave or convex region, follow CUBIC * cwnd growth. */ if (V_tcp_do_rfc3465) CCV(ccv, snd_cwnd) = w_cubic_next; else CCV(ccv, snd_cwnd) += ((w_cubic_next - CCV(ccv, snd_cwnd)) * CCV(ccv, t_maxseg)) / CCV(ccv, snd_cwnd); } /* * If we're not in slow start and we're probing for a * new cwnd limit at the start of a connection * (happens when hostcache has a relevant entry), * keep updating our current estimate of the * max_cwnd. */ if (cubic_data->num_cong_events == 0 && cubic_data->max_cwnd < CCV(ccv, snd_cwnd)) cubic_data->max_cwnd = CCV(ccv, snd_cwnd); } } } static void cubic_cb_destroy(struct cc_var *ccv) { if (ccv->cc_data != NULL) free(ccv->cc_data, M_CUBIC); } static int cubic_cb_init(struct cc_var *ccv) { struct cubic *cubic_data; cubic_data = malloc(sizeof(struct cubic), M_CUBIC, M_NOWAIT|M_ZERO); if (cubic_data == NULL) return (ENOMEM); /* Init some key variables with sensible defaults. */ cubic_data->t_last_cong = ticks; cubic_data->min_rtt_ticks = TCPTV_SRTTBASE; cubic_data->mean_rtt_ticks = 1; ccv->cc_data = cubic_data; return (0); } /* * Perform any necessary tasks before we enter congestion recovery. */ static void cubic_cong_signal(struct cc_var *ccv, uint32_t type) { struct cubic *cubic_data; - uint32_t cwin; - u_int mss; cubic_data = ccv->cc_data; - cwin = CCV(ccv, snd_cwnd); - mss = CCV(ccv, t_maxseg); switch (type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { cubic_ssthresh_update(ccv); cubic_data->num_cong_events++; cubic_data->prev_max_cwnd = cubic_data->max_cwnd; - cubic_data->max_cwnd = cwin; - CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); + cubic_data->max_cwnd = CCV(ccv, snd_cwnd); } ENTER_RECOVERY(CCV(ccv, t_flags)); } break; case CC_ECN: if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { cubic_ssthresh_update(ccv); cubic_data->num_cong_events++; cubic_data->prev_max_cwnd = cubic_data->max_cwnd; - cubic_data->max_cwnd = cwin; + cubic_data->max_cwnd = CCV(ccv, snd_cwnd); cubic_data->t_last_cong = ticks; CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } break; case CC_RTO: /* * Grab the current time and record it so we know when the * most recent congestion event was. Only record it when the * timeout has fired more than once, as there is a reasonable * chance the first one is a false alarm and may not indicate * congestion. */ - if (CCV(ccv, t_rxtshift) >= 2) { + if (CCV(ccv, t_rxtshift) >= 2) cubic_data->num_cong_events++; cubic_data->t_last_cong = ticks; - cubic_ssthresh_update(ccv); - cubic_data->max_cwnd = cwin; - CCV(ccv, snd_cwnd) = mss; - } break; } } static void cubic_conn_init(struct cc_var *ccv) { struct cubic *cubic_data; cubic_data = ccv->cc_data; /* * Ensure we have a sane initial value for max_cwnd recorded. Without * this here bad things happen when entries from the TCP hostcache * get used. */ cubic_data->max_cwnd = CCV(ccv, snd_cwnd); } static int cubic_mod_init(void) { cubic_cc_algo.after_idle = newreno_cc_algo.after_idle; return (0); } /* * Perform any necessary tasks before we exit congestion recovery. */ static void cubic_post_recovery(struct cc_var *ccv) { struct cubic *cubic_data; int pipe; cubic_data = ccv->cc_data; pipe = 0; /* Fast convergence heuristic. */ if (cubic_data->max_cwnd < cubic_data->prev_max_cwnd) cubic_data->max_cwnd = (cubic_data->max_cwnd * CUBIC_FC_FACTOR) >> CUBIC_SHIFT; if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { /* * If inflight data is less than ssthresh, set cwnd * conservatively to avoid a burst of data, as suggested in * the NewReno RFC. Otherwise, use the CUBIC method. * * XXXLAS: Find a way to do this without needing curack */ if (V_tcp_do_rfc6675_pipe) pipe = tcp_compute_pipe(ccv->ccvc.tcp); else pipe = CCV(ccv, snd_max) - ccv->curack; if (pipe < CCV(ccv, snd_ssthresh)) CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg); else /* Update cwnd based on beta and adjusted max_cwnd. */ CCV(ccv, snd_cwnd) = max(1, ((CUBIC_BETA * cubic_data->max_cwnd) >> CUBIC_SHIFT)); } cubic_data->t_last_cong = ticks; /* Calculate the average RTT between congestion epochs. */ if (cubic_data->epoch_ack_count > 0 && cubic_data->sum_rtt_ticks >= cubic_data->epoch_ack_count) { cubic_data->mean_rtt_ticks = (int)(cubic_data->sum_rtt_ticks / cubic_data->epoch_ack_count); } cubic_data->epoch_ack_count = 0; cubic_data->sum_rtt_ticks = 0; cubic_data->K = cubic_k(cubic_data->max_cwnd / CCV(ccv, t_maxseg)); } /* * Record the min RTT and sum samples for the epoch average RTT calculation. */ static void cubic_record_rtt(struct cc_var *ccv) { struct cubic *cubic_data; int t_srtt_ticks; /* Ignore srtt until a min number of samples have been taken. */ if (CCV(ccv, t_rttupdated) >= CUBIC_MIN_RTT_SAMPLES) { cubic_data = ccv->cc_data; t_srtt_ticks = CCV(ccv, t_srtt) / TCP_RTT_SCALE; /* * Record the current SRTT as our minrtt if it's the smallest * we've seen or minrtt is currently equal to its initialised * value. * * XXXLAS: Should there be some hysteresis for minrtt? */ if ((t_srtt_ticks < cubic_data->min_rtt_ticks || cubic_data->min_rtt_ticks == TCPTV_SRTTBASE)) { cubic_data->min_rtt_ticks = max(1, t_srtt_ticks); /* * If the connection is within its first congestion * epoch, ensure we prime mean_rtt_ticks with a * reasonable value until the epoch average RTT is * calculated in cubic_post_recovery(). */ if (cubic_data->min_rtt_ticks > cubic_data->mean_rtt_ticks) cubic_data->mean_rtt_ticks = cubic_data->min_rtt_ticks; } /* Sum samples for epoch average RTT calculation. */ cubic_data->sum_rtt_ticks += t_srtt_ticks; cubic_data->epoch_ack_count++; } } /* * Update the ssthresh in the event of congestion. */ static void cubic_ssthresh_update(struct cc_var *ccv) { struct cubic *cubic_data; cubic_data = ccv->cc_data; /* * On the first congestion event, set ssthresh to cwnd * 0.5, on * subsequent congestion events, set it to cwnd * beta. */ if (cubic_data->num_cong_events == 0) CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd) >> 1; else CCV(ccv, snd_ssthresh) = ((u_long)CCV(ccv, snd_cwnd) * CUBIC_BETA) >> CUBIC_SHIFT; } DECLARE_CC_MODULE(cubic, &cubic_cc_algo); Index: head/sys/netinet/cc/cc_dctcp.c =================================================================== --- head/sys/netinet/cc/cc_dctcp.c (revision 307899) +++ head/sys/netinet/cc/cc_dctcp.c (revision 307900) @@ -1,475 +1,472 @@ /*- * Copyright (c) 2007-2008 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2014 Midori Kato * Copyright (c) 2014 The FreeBSD Foundation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the DCTCP algorithm for FreeBSD, based on * "Data Center TCP (DCTCP)" by M. Alizadeh, A. Greenberg, D. A. Maltz, * J. Padhye, P. Patel, B. Prabhakar, S. Sengupta, and M. Sridharan., * in ACM Conference on SIGCOMM 2010, New York, USA, * Originally released as the contribution of Microsoft Research project. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CAST_PTR_INT(X) (*((int*)(X))) #define MAX_ALPHA_VALUE 1024 static VNET_DEFINE(uint32_t, dctcp_alpha) = 0; #define V_dctcp_alpha VNET(dctcp_alpha) static VNET_DEFINE(uint32_t, dctcp_shift_g) = 4; #define V_dctcp_shift_g VNET(dctcp_shift_g) static VNET_DEFINE(uint32_t, dctcp_slowstart) = 0; #define V_dctcp_slowstart VNET(dctcp_slowstart) struct dctcp { int bytes_ecn; /* # of marked bytes during a RTT */ int bytes_total; /* # of acked bytes during a RTT */ int alpha; /* the fraction of marked bytes */ int ce_prev; /* CE state of the last segment */ int save_sndnxt; /* end sequence number of the current window */ int ece_curr; /* ECE flag in this segment */ int ece_prev; /* ECE flag in the last segment */ uint32_t num_cong_events; /* # of congestion events */ }; static MALLOC_DEFINE(M_dctcp, "dctcp data", "Per connection data required for the dctcp algorithm"); static void dctcp_ack_received(struct cc_var *ccv, uint16_t type); static void dctcp_after_idle(struct cc_var *ccv); static void dctcp_cb_destroy(struct cc_var *ccv); static int dctcp_cb_init(struct cc_var *ccv); static void dctcp_cong_signal(struct cc_var *ccv, uint32_t type); static void dctcp_conn_init(struct cc_var *ccv); static void dctcp_post_recovery(struct cc_var *ccv); static void dctcp_ecnpkt_handler(struct cc_var *ccv); static void dctcp_update_alpha(struct cc_var *ccv); struct cc_algo dctcp_cc_algo = { .name = "dctcp", .ack_received = dctcp_ack_received, .cb_destroy = dctcp_cb_destroy, .cb_init = dctcp_cb_init, .cong_signal = dctcp_cong_signal, .conn_init = dctcp_conn_init, .post_recovery = dctcp_post_recovery, .ecnpkt_handler = dctcp_ecnpkt_handler, .after_idle = dctcp_after_idle, }; static void dctcp_ack_received(struct cc_var *ccv, uint16_t type) { struct dctcp *dctcp_data; int bytes_acked = 0; dctcp_data = ccv->cc_data; if (CCV(ccv, t_flags) & TF_ECN_PERMIT) { /* * DCTCP doesn't treat receipt of ECN marked packet as a * congestion event. Thus, DCTCP always executes the ACK * processing out of congestion recovery. */ if (IN_CONGRECOVERY(CCV(ccv, t_flags))) { EXIT_CONGRECOVERY(CCV(ccv, t_flags)); newreno_cc_algo.ack_received(ccv, type); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } else newreno_cc_algo.ack_received(ccv, type); if (type == CC_DUPACK) bytes_acked = CCV(ccv, t_maxseg); if (type == CC_ACK) bytes_acked = ccv->bytes_this_ack; /* Update total bytes. */ dctcp_data->bytes_total += bytes_acked; /* Update total marked bytes. */ if (dctcp_data->ece_curr) { if (!dctcp_data->ece_prev && bytes_acked > CCV(ccv, t_maxseg)) { dctcp_data->bytes_ecn += (bytes_acked - CCV(ccv, t_maxseg)); } else dctcp_data->bytes_ecn += bytes_acked; dctcp_data->ece_prev = 1; } else { if (dctcp_data->ece_prev && bytes_acked > CCV(ccv, t_maxseg)) dctcp_data->bytes_ecn += CCV(ccv, t_maxseg); dctcp_data->ece_prev = 0; } dctcp_data->ece_curr = 0; /* * Update the fraction of marked bytes at the end of * current window size. */ if ((IN_FASTRECOVERY(CCV(ccv, t_flags)) && SEQ_GEQ(ccv->curack, CCV(ccv, snd_recover))) || (!IN_FASTRECOVERY(CCV(ccv, t_flags)) && SEQ_GT(ccv->curack, dctcp_data->save_sndnxt))) dctcp_update_alpha(ccv); } else newreno_cc_algo.ack_received(ccv, type); } static void dctcp_after_idle(struct cc_var *ccv) { struct dctcp *dctcp_data; dctcp_data = ccv->cc_data; /* Initialize internal parameters after idle time */ dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); dctcp_data->alpha = V_dctcp_alpha; dctcp_data->ece_curr = 0; dctcp_data->ece_prev = 0; dctcp_data->num_cong_events = 0; dctcp_cc_algo.after_idle = newreno_cc_algo.after_idle; } static void dctcp_cb_destroy(struct cc_var *ccv) { if (ccv->cc_data != NULL) free(ccv->cc_data, M_dctcp); } static int dctcp_cb_init(struct cc_var *ccv) { struct dctcp *dctcp_data; dctcp_data = malloc(sizeof(struct dctcp), M_dctcp, M_NOWAIT|M_ZERO); if (dctcp_data == NULL) return (ENOMEM); /* Initialize some key variables with sensible defaults. */ dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; /* * When alpha is set to 0 in the beginning, DCTCP sender transfers as * much data as possible until the value converges which may expand the * queueing delay at the switch. When alpha is set to 1, queueing delay * is kept small. * Throughput-sensitive applications should have alpha = 0 * Latency-sensitive applications should have alpha = 1 * * Note: DCTCP draft suggests initial alpha to be 1 but we've decided to * keep it 0 as default. */ dctcp_data->alpha = V_dctcp_alpha; dctcp_data->save_sndnxt = 0; dctcp_data->ce_prev = 0; dctcp_data->ece_curr = 0; dctcp_data->ece_prev = 0; dctcp_data->num_cong_events = 0; ccv->cc_data = dctcp_data; return (0); } /* * Perform any necessary tasks before we enter congestion recovery. */ static void dctcp_cong_signal(struct cc_var *ccv, uint32_t type) { struct dctcp *dctcp_data; - uint32_t cwin; - u_int mss; + u_int win, mss; dctcp_data = ccv->cc_data; - cwin = CCV(ccv, snd_cwnd); + win = CCV(ccv, snd_cwnd); mss = CCV(ccv, t_maxseg); switch (type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { CCV(ccv, snd_ssthresh) = mss * - max(cwin / 2 / mss, 2); + max(win / 2 / mss, 2); dctcp_data->num_cong_events++; } else { /* cwnd has already updated as congestion * recovery. Reverse cwnd value using * snd_cwnd_prev and recalculate snd_ssthresh */ - cwin = CCV(ccv, snd_cwnd_prev); + win = CCV(ccv, snd_cwnd_prev); CCV(ccv, snd_ssthresh) = - max(cwin / 2 / mss, 2) * mss; + max(win / 2 / mss, 2) * mss; } ENTER_RECOVERY(CCV(ccv, t_flags)); } break; case CC_ECN: /* * Save current snd_cwnd when the host encounters both * congestion recovery and fast recovery. */ - CCV(ccv, snd_cwnd_prev) = cwin; + CCV(ccv, snd_cwnd_prev) = win; if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { if (V_dctcp_slowstart && dctcp_data->num_cong_events++ == 0) { CCV(ccv, snd_ssthresh) = - mss * max(cwin / 2 / mss, 2); + mss * max(win / 2 / mss, 2); dctcp_data->alpha = MAX_ALPHA_VALUE; dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); } else - CCV(ccv, snd_ssthresh) = max((cwin - ((cwin * + CCV(ccv, snd_ssthresh) = max((win - ((win * dctcp_data->alpha) >> 11)) / mss, 2) * mss; CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } dctcp_data->ece_curr = 1; break; case CC_RTO: if (CCV(ccv, t_flags) & TF_ECN_PERMIT) { CCV(ccv, t_flags) |= TF_ECN_SND_CWR; dctcp_update_alpha(ccv); dctcp_data->save_sndnxt += CCV(ccv, t_maxseg); dctcp_data->num_cong_events++; - CCV(ccv, snd_ssthresh) = max(2 * mss, cwin / 2); - CCV(ccv, snd_cwnd) = mss; } break; } } static void dctcp_conn_init(struct cc_var *ccv) { struct dctcp *dctcp_data; dctcp_data = ccv->cc_data; if (CCV(ccv, t_flags) & TF_ECN_PERMIT) dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); } /* * Perform any necessary tasks before we exit congestion recovery. */ static void dctcp_post_recovery(struct cc_var *ccv) { dctcp_cc_algo.post_recovery = newreno_cc_algo.post_recovery; if (CCV(ccv, t_flags) & TF_ECN_PERMIT) dctcp_update_alpha(ccv); } /* * Execute an additional ECN processing using ECN field in IP header and the CWR * bit in TCP header. * * delay_ack == 0 - Delayed ACK disabled * delay_ack == 1 - Delayed ACK enabled */ static void dctcp_ecnpkt_handler(struct cc_var *ccv) { struct dctcp *dctcp_data; uint32_t ccflag; int delay_ack; dctcp_data = ccv->cc_data; ccflag = ccv->flags; delay_ack = 1; /* * DCTCP responses an ACK immediately when the CE state * in between this segment and the last segment is not same. */ if (ccflag & CCF_IPHDR_CE) { if (!dctcp_data->ce_prev && (ccflag & CCF_DELACK)) delay_ack = 0; dctcp_data->ce_prev = 1; CCV(ccv, t_flags) |= TF_ECN_SND_ECE; } else { if (dctcp_data->ce_prev && (ccflag & CCF_DELACK)) delay_ack = 0; dctcp_data->ce_prev = 0; CCV(ccv, t_flags) &= ~TF_ECN_SND_ECE; } /* DCTCP sets delayed ack when this segment sets the CWR flag. */ if ((ccflag & CCF_DELACK) && (ccflag & CCF_TCPHDR_CWR)) delay_ack = 1; if (delay_ack == 0) ccv->flags |= CCF_ACKNOW; else ccv->flags &= ~CCF_ACKNOW; } /* * Update the fraction of marked bytes represented as 'alpha'. * Also initialize several internal parameters at the end of this function. */ static void dctcp_update_alpha(struct cc_var *ccv) { struct dctcp *dctcp_data; int alpha_prev; dctcp_data = ccv->cc_data; alpha_prev = dctcp_data->alpha; dctcp_data->bytes_total = max(dctcp_data->bytes_total, 1); /* * Update alpha: alpha = (1 - g) * alpha + g * F. * Here: * g is weight factor * recommaded to be set to 1/16 * small g = slow convergence between competitive DCTCP flows * large g = impacts low utilization of bandwidth at switches * F is fraction of marked segments in last RTT * updated every RTT * Alpha must be round to 0 - MAX_ALPHA_VALUE. */ dctcp_data->alpha = min(alpha_prev - (alpha_prev >> V_dctcp_shift_g) + (dctcp_data->bytes_ecn << (10 - V_dctcp_shift_g)) / dctcp_data->bytes_total, MAX_ALPHA_VALUE); /* Initialize internal parameters for next alpha calculation */ dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); } static int dctcp_alpha_handler(SYSCTL_HANDLER_ARGS) { uint32_t new; int error; new = V_dctcp_alpha; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else { if (new > MAX_ALPHA_VALUE) V_dctcp_alpha = MAX_ALPHA_VALUE; else V_dctcp_alpha = new; } } return (error); } static int dctcp_shift_g_handler(SYSCTL_HANDLER_ARGS) { uint32_t new; int error; new = V_dctcp_shift_g; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else V_dctcp_shift_g = new; } return (error); } static int dctcp_slowstart_handler(SYSCTL_HANDLER_ARGS) { uint32_t new; int error; new = V_dctcp_slowstart; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else V_dctcp_slowstart = new; } return (error); } SYSCTL_DECL(_net_inet_tcp_cc_dctcp); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, dctcp, CTLFLAG_RW, NULL, "dctcp congestion control related settings"); SYSCTL_PROC(_net_inet_tcp_cc_dctcp, OID_AUTO, alpha, CTLFLAG_VNET|CTLTYPE_UINT|CTLFLAG_RW, &VNET_NAME(dctcp_alpha), 0, &dctcp_alpha_handler, "IU", "dctcp alpha parameter"); SYSCTL_PROC(_net_inet_tcp_cc_dctcp, OID_AUTO, shift_g, CTLFLAG_VNET|CTLTYPE_UINT|CTLFLAG_RW, &VNET_NAME(dctcp_shift_g), 4, &dctcp_shift_g_handler, "IU", "dctcp shift parameter"); SYSCTL_PROC(_net_inet_tcp_cc_dctcp, OID_AUTO, slowstart, CTLFLAG_VNET|CTLTYPE_UINT|CTLFLAG_RW, &VNET_NAME(dctcp_slowstart), 0, &dctcp_slowstart_handler, "IU", "half CWND reduction after the first slow start"); DECLARE_CC_MODULE(dctcp, &dctcp_cc_algo); Index: head/sys/netinet/cc/cc_htcp.c =================================================================== --- head/sys/netinet/cc/cc_htcp.c (revision 307899) +++ head/sys/netinet/cc/cc_htcp.c (revision 307900) @@ -1,534 +1,527 @@ /*- * Copyright (c) 2007-2008 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart and * James Healy, made possible in part by a grant from the Cisco University * Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the H-TCP congestion control algorithm for FreeBSD, * based on the Internet Draft "draft-leith-tcp-htcp-06.txt" by Leith and * Shorten. Originally released as part of the NewTCP research project at * Swinburne University of Technology's Centre for Advanced Internet * Architectures, Melbourne, Australia, which was made possible in part by a * grant from the Cisco University Research Program Fund at Community Foundation * Silicon Valley. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Fixed point math shifts. */ #define HTCP_SHIFT 8 #define HTCP_ALPHA_INC_SHIFT 4 #define HTCP_INIT_ALPHA 1 #define HTCP_DELTA_L hz /* 1 sec in ticks. */ #define HTCP_MINBETA 128 /* 0.5 << HTCP_SHIFT. */ #define HTCP_MAXBETA 204 /* ~0.8 << HTCP_SHIFT. */ #define HTCP_MINROWE 26 /* ~0.1 << HTCP_SHIFT. */ #define HTCP_MAXROWE 512 /* 2 << HTCP_SHIFT. */ /* RTT_ref (ms) used in the calculation of alpha if RTT scaling is enabled. */ #define HTCP_RTT_REF 100 /* Don't trust SRTT until this many samples have been taken. */ #define HTCP_MIN_RTT_SAMPLES 8 /* * HTCP_CALC_ALPHA performs a fixed point math calculation to determine the * value of alpha, based on the function defined in the HTCP spec. * * i.e. 1 + 10(delta - delta_l) + ((delta - delta_l) / 2) ^ 2 * * "diff" is passed in to the macro as "delta - delta_l" and is expected to be * in units of ticks. * * The joyousnous of fixed point maths means our function implementation looks a * little funky... * * In order to maintain some precision in the calculations, a fixed point shift * HTCP_ALPHA_INC_SHIFT is used to ensure the integer divisions don't * truncate the results too badly. * * The "16" value is the "1" term in the alpha function shifted up by * HTCP_ALPHA_INC_SHIFT * * The "160" value is the "10" multiplier in the alpha function multiplied by * 2^HTCP_ALPHA_INC_SHIFT * * Specifying these as constants reduces the computations required. After * up-shifting all the terms in the function and performing the required * calculations, we down-shift the final result by HTCP_ALPHA_INC_SHIFT to * ensure it is back in the correct range. * * The "hz" terms are required as kernels can be configured to run with * different tick timers, which we have to adjust for in the alpha calculation * (which originally was defined in terms of seconds). * * We also have to be careful to constrain the value of diff such that it won't * overflow whilst performing the calculation. The middle term i.e. (160 * diff) * / hz is the limiting factor in the calculation. We must constrain diff to be * less than the max size of an int divided by the constant 160 figure * i.e. diff < INT_MAX / 160 * * NB: Changing HTCP_ALPHA_INC_SHIFT will require you to MANUALLY update the * constants used in this function! */ #define HTCP_CALC_ALPHA(diff) \ ((\ (16) + \ ((160 * (diff)) / hz) + \ (((diff) / hz) * (((diff) << HTCP_ALPHA_INC_SHIFT) / (4 * hz))) \ ) >> HTCP_ALPHA_INC_SHIFT) static void htcp_ack_received(struct cc_var *ccv, uint16_t type); static void htcp_cb_destroy(struct cc_var *ccv); static int htcp_cb_init(struct cc_var *ccv); static void htcp_cong_signal(struct cc_var *ccv, uint32_t type); static int htcp_mod_init(void); static void htcp_post_recovery(struct cc_var *ccv); static void htcp_recalc_alpha(struct cc_var *ccv); static void htcp_recalc_beta(struct cc_var *ccv); static void htcp_record_rtt(struct cc_var *ccv); static void htcp_ssthresh_update(struct cc_var *ccv); struct htcp { /* cwnd before entering cong recovery. */ unsigned long prev_cwnd; /* cwnd additive increase parameter. */ int alpha; /* cwnd multiplicative decrease parameter. */ int beta; /* Largest rtt seen for the flow. */ int maxrtt; /* Shortest rtt seen for the flow. */ int minrtt; /* Time of last congestion event in ticks. */ int t_last_cong; }; static int htcp_rtt_ref; /* * The maximum number of ticks the value of diff can reach in * htcp_recalc_alpha() before alpha will stop increasing due to overflow. * See comment above HTCP_CALC_ALPHA for more info. */ static int htcp_max_diff = INT_MAX / ((1 << HTCP_ALPHA_INC_SHIFT) * 10); /* Per-netstack vars. */ static VNET_DEFINE(u_int, htcp_adaptive_backoff) = 0; static VNET_DEFINE(u_int, htcp_rtt_scaling) = 0; #define V_htcp_adaptive_backoff VNET(htcp_adaptive_backoff) #define V_htcp_rtt_scaling VNET(htcp_rtt_scaling) static MALLOC_DEFINE(M_HTCP, "htcp data", "Per connection data required for the HTCP congestion control algorithm"); struct cc_algo htcp_cc_algo = { .name = "htcp", .ack_received = htcp_ack_received, .cb_destroy = htcp_cb_destroy, .cb_init = htcp_cb_init, .cong_signal = htcp_cong_signal, .mod_init = htcp_mod_init, .post_recovery = htcp_post_recovery, }; static void htcp_ack_received(struct cc_var *ccv, uint16_t type) { struct htcp *htcp_data; htcp_data = ccv->cc_data; htcp_record_rtt(ccv); /* * Regular ACK and we're not in cong/fast recovery and we're cwnd * limited and we're either not doing ABC or are slow starting or are * doing ABC and we've sent a cwnd's worth of bytes. */ if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && (ccv->flags & CCF_CWND_LIMITED) && (!V_tcp_do_rfc3465 || CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) || (V_tcp_do_rfc3465 && ccv->flags & CCF_ABC_SENTAWND))) { htcp_recalc_beta(ccv); htcp_recalc_alpha(ccv); /* * Use the logic in NewReno ack_received() for slow start and * for the first HTCP_DELTA_L ticks after either the flow starts * or a congestion event (when alpha equals 1). */ if (htcp_data->alpha == 1 || CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) newreno_cc_algo.ack_received(ccv, type); else { if (V_tcp_do_rfc3465) { /* Increment cwnd by alpha segments. */ CCV(ccv, snd_cwnd) += htcp_data->alpha * CCV(ccv, t_maxseg); ccv->flags &= ~CCF_ABC_SENTAWND; } else /* * Increment cwnd by alpha/cwnd segments to * approximate an increase of alpha segments * per RTT. */ CCV(ccv, snd_cwnd) += (((htcp_data->alpha << HTCP_SHIFT) / (CCV(ccv, snd_cwnd) / CCV(ccv, t_maxseg))) * CCV(ccv, t_maxseg)) >> HTCP_SHIFT; } } } static void htcp_cb_destroy(struct cc_var *ccv) { if (ccv->cc_data != NULL) free(ccv->cc_data, M_HTCP); } static int htcp_cb_init(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = malloc(sizeof(struct htcp), M_HTCP, M_NOWAIT); if (htcp_data == NULL) return (ENOMEM); /* Init some key variables with sensible defaults. */ htcp_data->alpha = HTCP_INIT_ALPHA; htcp_data->beta = HTCP_MINBETA; htcp_data->maxrtt = TCPTV_SRTTBASE; htcp_data->minrtt = TCPTV_SRTTBASE; htcp_data->prev_cwnd = 0; htcp_data->t_last_cong = ticks; ccv->cc_data = htcp_data; return (0); } /* * Perform any necessary tasks before we enter congestion recovery. */ static void htcp_cong_signal(struct cc_var *ccv, uint32_t type) { struct htcp *htcp_data; - uint32_t cwin; - u_int mss; htcp_data = ccv->cc_data; - cwin = CCV(ccv, snd_cwnd); - mss = CCV(ccv, t_maxseg); switch (type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { /* * Apply hysteresis to maxrtt to ensure * reductions in the RTT are reflected in our * measurements. */ htcp_data->maxrtt = (htcp_data->minrtt + (htcp_data->maxrtt - htcp_data->minrtt) * 95) / 100; htcp_ssthresh_update(ccv); - CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); htcp_data->t_last_cong = ticks; - htcp_data->prev_cwnd = cwin; + htcp_data->prev_cwnd = CCV(ccv, snd_cwnd); } ENTER_RECOVERY(CCV(ccv, t_flags)); } break; case CC_ECN: if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { /* * Apply hysteresis to maxrtt to ensure reductions in * the RTT are reflected in our measurements. */ htcp_data->maxrtt = (htcp_data->minrtt + (htcp_data->maxrtt - htcp_data->minrtt) * 95) / 100; htcp_ssthresh_update(ccv); CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); htcp_data->t_last_cong = ticks; - htcp_data->prev_cwnd = cwin; + htcp_data->prev_cwnd = CCV(ccv, snd_cwnd); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } break; case CC_RTO: /* * Grab the current time and record it so we know when the * most recent congestion event was. Only record it when the * timeout has fired more than once, as there is a reasonable * chance the first one is a false alarm and may not indicate * congestion. */ if (CCV(ccv, t_rxtshift) >= 2) htcp_data->t_last_cong = ticks; - CCV(ccv, snd_ssthresh) = max(2 * mss, cwin / 2); - CCV(ccv, snd_cwnd) = mss; break; } } static int htcp_mod_init(void) { htcp_cc_algo.after_idle = newreno_cc_algo.after_idle; /* * HTCP_RTT_REF is defined in ms, and t_srtt in the tcpcb is stored in * units of TCP_RTT_SCALE*hz. Scale HTCP_RTT_REF to be in the same units * as t_srtt. */ htcp_rtt_ref = (HTCP_RTT_REF * TCP_RTT_SCALE * hz) / 1000; return (0); } /* * Perform any necessary tasks before we exit congestion recovery. */ static void htcp_post_recovery(struct cc_var *ccv) { int pipe; struct htcp *htcp_data; pipe = 0; htcp_data = ccv->cc_data; if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { /* * If inflight data is less than ssthresh, set cwnd * conservatively to avoid a burst of data, as suggested in the * NewReno RFC. Otherwise, use the HTCP method. * * XXXLAS: Find a way to do this without needing curack */ if (V_tcp_do_rfc6675_pipe) pipe = tcp_compute_pipe(ccv->ccvc.tcp); else pipe = CCV(ccv, snd_max) - ccv->curack; if (pipe < CCV(ccv, snd_ssthresh)) CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg); else CCV(ccv, snd_cwnd) = max(1, ((htcp_data->beta * htcp_data->prev_cwnd / CCV(ccv, t_maxseg)) >> HTCP_SHIFT)) * CCV(ccv, t_maxseg); } } static void htcp_recalc_alpha(struct cc_var *ccv) { struct htcp *htcp_data; int alpha, diff, now; htcp_data = ccv->cc_data; now = ticks; /* * If ticks has wrapped around (will happen approximately once every 49 * days on a machine with the default kern.hz=1000) and a flow straddles * the wrap point, our alpha calcs will be completely wrong. We cut our * losses and restart alpha from scratch by setting t_last_cong = now - * HTCP_DELTA_L. * * This does not deflate our cwnd at all. It simply slows the rate cwnd * is growing by until alpha regains the value it held prior to taking * this drastic measure. */ if (now < htcp_data->t_last_cong) htcp_data->t_last_cong = now - HTCP_DELTA_L; diff = now - htcp_data->t_last_cong - HTCP_DELTA_L; /* Cap alpha if the value of diff would overflow HTCP_CALC_ALPHA(). */ if (diff < htcp_max_diff) { /* * If it has been more than HTCP_DELTA_L ticks since congestion, * increase alpha according to the function defined in the spec. */ if (diff > 0) { alpha = HTCP_CALC_ALPHA(diff); /* * Adaptive backoff fairness adjustment: * 2 * (1 - beta) * alpha_raw */ if (V_htcp_adaptive_backoff) alpha = max(1, (2 * ((1 << HTCP_SHIFT) - htcp_data->beta) * alpha) >> HTCP_SHIFT); /* * RTT scaling: (RTT / RTT_ref) * alpha * alpha will be the raw value from HTCP_CALC_ALPHA() if * adaptive backoff is off, or the adjusted value if * adaptive backoff is on. */ if (V_htcp_rtt_scaling) alpha = max(1, (min(max(HTCP_MINROWE, (CCV(ccv, t_srtt) << HTCP_SHIFT) / htcp_rtt_ref), HTCP_MAXROWE) * alpha) >> HTCP_SHIFT); } else alpha = 1; htcp_data->alpha = alpha; } } static void htcp_recalc_beta(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = ccv->cc_data; /* * TCPTV_SRTTBASE is the initialised value of each connection's SRTT, so * we only calc beta if the connection's SRTT has been changed from its * initial value. beta is bounded to ensure it is always between * HTCP_MINBETA and HTCP_MAXBETA. */ if (V_htcp_adaptive_backoff && htcp_data->minrtt != TCPTV_SRTTBASE && htcp_data->maxrtt != TCPTV_SRTTBASE) htcp_data->beta = min(max(HTCP_MINBETA, (htcp_data->minrtt << HTCP_SHIFT) / htcp_data->maxrtt), HTCP_MAXBETA); else htcp_data->beta = HTCP_MINBETA; } /* * Record the minimum and maximum RTT seen for the connection. These are used in * the calculation of beta if adaptive backoff is enabled. */ static void htcp_record_rtt(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = ccv->cc_data; /* XXXLAS: Should there be some hysteresis for minrtt? */ /* * Record the current SRTT as our minrtt if it's the smallest we've seen * or minrtt is currently equal to its initialised value. Ignore SRTT * until a min number of samples have been taken. */ if ((CCV(ccv, t_srtt) < htcp_data->minrtt || htcp_data->minrtt == TCPTV_SRTTBASE) && (CCV(ccv, t_rttupdated) >= HTCP_MIN_RTT_SAMPLES)) htcp_data->minrtt = CCV(ccv, t_srtt); /* * Record the current SRTT as our maxrtt if it's the largest we've * seen. Ignore SRTT until a min number of samples have been taken. */ if (CCV(ccv, t_srtt) > htcp_data->maxrtt && CCV(ccv, t_rttupdated) >= HTCP_MIN_RTT_SAMPLES) htcp_data->maxrtt = CCV(ccv, t_srtt); } /* * Update the ssthresh in the event of congestion. */ static void htcp_ssthresh_update(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = ccv->cc_data; /* * On the first congestion event, set ssthresh to cwnd * 0.5, on * subsequent congestion events, set it to cwnd * beta. */ if (CCV(ccv, snd_ssthresh) == TCP_MAXWIN << TCP_MAX_WINSHIFT) CCV(ccv, snd_ssthresh) = ((u_long)CCV(ccv, snd_cwnd) * HTCP_MINBETA) >> HTCP_SHIFT; else { htcp_recalc_beta(ccv); CCV(ccv, snd_ssthresh) = ((u_long)CCV(ccv, snd_cwnd) * htcp_data->beta) >> HTCP_SHIFT; } } SYSCTL_DECL(_net_inet_tcp_cc_htcp); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, htcp, CTLFLAG_RW, NULL, "H-TCP related settings"); SYSCTL_UINT(_net_inet_tcp_cc_htcp, OID_AUTO, adaptive_backoff, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(htcp_adaptive_backoff), 0, "enable H-TCP adaptive backoff"); SYSCTL_UINT(_net_inet_tcp_cc_htcp, OID_AUTO, rtt_scaling, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(htcp_rtt_scaling), 0, "enable H-TCP RTT scaling"); DECLARE_CC_MODULE(htcp, &htcp_cc_algo); Index: head/sys/netinet/cc/cc_newreno.c =================================================================== --- head/sys/netinet/cc/cc_newreno.c (revision 307899) +++ head/sys/netinet/cc/cc_newreno.c (revision 307900) @@ -1,253 +1,244 @@ /*- * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. * Copyright (c) 2007-2008,2010 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart, James * Healy and David Hayes, made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * This software was first released in 2007 by James Healy and Lawrence Stewart * whilst working on the NewTCP research project at Swinburne University of * Technology's Centre for Advanced Internet Architectures, Melbourne, * Australia, which was made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include static void newreno_ack_received(struct cc_var *ccv, uint16_t type); static void newreno_after_idle(struct cc_var *ccv); static void newreno_cong_signal(struct cc_var *ccv, uint32_t type); static void newreno_post_recovery(struct cc_var *ccv); struct cc_algo newreno_cc_algo = { .name = "newreno", .ack_received = newreno_ack_received, .after_idle = newreno_after_idle, .cong_signal = newreno_cong_signal, .post_recovery = newreno_post_recovery, }; static void newreno_ack_received(struct cc_var *ccv, uint16_t type) { if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && (ccv->flags & CCF_CWND_LIMITED)) { u_int cw = CCV(ccv, snd_cwnd); u_int incr = CCV(ccv, t_maxseg); /* * Regular in-order ACK, open the congestion window. * Method depends on which congestion control state we're * in (slow start or cong avoid) and if ABC (RFC 3465) is * enabled. * * slow start: cwnd <= ssthresh * cong avoid: cwnd > ssthresh * * slow start and ABC (RFC 3465): * Grow cwnd exponentially by the amount of data * ACKed capping the max increment per ACK to * (abc_l_var * maxseg) bytes. * * slow start without ABC (RFC 5681): * Grow cwnd exponentially by maxseg per ACK. * * cong avoid and ABC (RFC 3465): * Grow cwnd linearly by maxseg per RTT for each * cwnd worth of ACKed data. * * cong avoid without ABC (RFC 5681): * Grow cwnd linearly by approximately maxseg per RTT using * maxseg^2 / cwnd per ACK as the increment. * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to * avoid capping cwnd. */ if (cw > CCV(ccv, snd_ssthresh)) { if (V_tcp_do_rfc3465) { if (ccv->flags & CCF_ABC_SENTAWND) ccv->flags &= ~CCF_ABC_SENTAWND; else incr = 0; } else incr = max((incr * incr / cw), 1); } else if (V_tcp_do_rfc3465) { /* * In slow-start with ABC enabled and no RTO in sight? * (Must not use abc_l_var > 1 if slow starting after * an RTO. On RTO, snd_nxt = snd_una, so the * snd_nxt == snd_max check is sufficient to * handle this). * * XXXLAS: Find a way to signal SS after RTO that * doesn't rely on tcpcb vars. */ if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) incr = min(ccv->bytes_this_ack, ccv->nsegs * V_tcp_abc_l_var * CCV(ccv, t_maxseg)); else incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); } /* ABC is on by default, so incr equals 0 frequently. */ if (incr > 0) CCV(ccv, snd_cwnd) = min(cw + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } } static void newreno_after_idle(struct cc_var *ccv) { int rw; /* * If we've been idle for more than one retransmit timeout the old * congestion window is no longer current and we have to reduce it to * the restart window before we can transmit again. * * The restart window is the initial window or the last CWND, whichever * is smaller. * * This is done to prevent us from flooding the path with a full CWND at * wirespeed, overloading router and switch buffers along the way. * * See RFC5681 Section 4.1. "Restarting Idle Connections". */ if (V_tcp_do_rfc3390) rw = min(4 * CCV(ccv, t_maxseg), max(2 * CCV(ccv, t_maxseg), 4380)); else rw = CCV(ccv, t_maxseg) * 2; CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd)); } /* * Perform any necessary tasks before we enter congestion recovery. */ static void newreno_cong_signal(struct cc_var *ccv, uint32_t type) { - uint32_t cwin; - u_int mss; + u_int win; - cwin = CCV(ccv, snd_cwnd); - mss = CCV(ccv, t_maxseg); - /* Catch algos which mistakenly leak private signal types. */ KASSERT((type & CC_SIGPRIVMASK) == 0, ("%s: congestion signal type 0x%08x is private\n", __func__, type)); - cwin = max(2*mss, cwin/2); + win = max(CCV(ccv, snd_cwnd) / 2 / CCV(ccv, t_maxseg), 2) * + CCV(ccv, t_maxseg); switch (type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { - if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { - CCV(ccv, snd_ssthresh) = cwin; - CCV(ccv, snd_cwnd) = cwin; - } + if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) + CCV(ccv, snd_ssthresh) = win; ENTER_RECOVERY(CCV(ccv, t_flags)); } break; case CC_ECN: if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { - CCV(ccv, snd_ssthresh) = cwin; - CCV(ccv, snd_cwnd) = cwin; + CCV(ccv, snd_ssthresh) = win; + CCV(ccv, snd_cwnd) = win; ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } - break; - case CC_RTO: - CCV(ccv, snd_ssthresh) = cwin; - CCV(ccv, snd_cwnd) = mss; break; } } /* * Perform any necessary tasks before we exit congestion recovery. */ static void newreno_post_recovery(struct cc_var *ccv) { int pipe; pipe = 0; if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { /* * Fast recovery will conclude after returning from this * function. Window inflation should have left us with * approximately snd_ssthresh outstanding data. But in case we * would be inclined to send a burst, better to do it via the * slow start mechanism. * * XXXLAS: Find a way to do this without needing curack */ if (V_tcp_do_rfc6675_pipe) pipe = tcp_compute_pipe(ccv->ccvc.tcp); else pipe = CCV(ccv, snd_max) - ccv->curack; if (pipe < CCV(ccv, snd_ssthresh)) CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg); else CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); } } DECLARE_CC_MODULE(newreno, &newreno_cc_algo); Index: head/sys/netinet/tcp_input.c =================================================================== --- head/sys/netinet/tcp_input.c (revision 307899) +++ head/sys/netinet/tcp_input.c (revision 307900) @@ -1,3918 +1,3900 @@ /*- * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. All rights reserved. * Copyright (c) 2007-2008,2010 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart, * James Healy and David Hayes, made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_tcpdebug.h" #include #include #ifdef TCP_HHOOK #include #endif #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include /* before tcp_seq.h, for tcp_random18() */ #include #include #include #include #include #define TCPSTATES /* for logging */ #include #include #include #include #include #include /* required for icmp_var.h */ #include /* for ICMP_BANDLIM */ #include #include #include #include #include #include #include #include #ifdef TCP_RFC7413 #include #endif #include #include #include #include #include #include #include #include #ifdef TCPPCAP #include #endif #include #ifdef TCPDEBUG #include #endif /* TCPDEBUG */ #ifdef TCP_OFFLOAD #include #endif #ifdef IPSEC #include #include #endif /*IPSEC*/ #include #include const int tcprexmtthresh = 3; int tcp_log_in_vain = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); VNET_DEFINE(int, blackhole) = 0; #define V_blackhole VNET(blackhole) SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(blackhole), 0, "Do not send RST on segments to closed ports"); VNET_DEFINE(int, tcp_delack_enabled) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_delack_enabled), 0, "Delay ACK to try and piggyback it onto a data packet"); VNET_DEFINE(int, drop_synfin) = 0; #define V_drop_synfin VNET(drop_synfin) SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(drop_synfin), 0, "Drop TCP packets with SYN+FIN set"); VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc6675_pipe), 0, "Use calculated pipe/in-flight bytes per RFC 6675"); VNET_DEFINE(int, tcp_do_rfc3042) = 1; #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3042), 0, "Enable RFC 3042 (Limited Transmit)"); VNET_DEFINE(int, tcp_do_rfc3390) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3390), 0, "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); VNET_DEFINE(int, tcp_initcwnd_segments) = 10; SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, "Slow-start flight size (initial congestion window) in number of segments"); VNET_DEFINE(int, tcp_do_rfc3465) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3465), 0, "Enable RFC 3465 (Appropriate Byte Counting)"); VNET_DEFINE(int, tcp_abc_l_var) = 2; SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_abc_l_var), 2, "Cap the max cwnd increment during slow-start to this number of segments"); static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); VNET_DEFINE(int, tcp_do_ecn) = 2; SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_ecn), 0, "TCP ECN support"); VNET_DEFINE(int, tcp_ecn_maxretries) = 1; SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ecn_maxretries), 0, "Max retries before giving up on ECN"); VNET_DEFINE(int, tcp_insecure_syn) = 0; #define V_tcp_insecure_syn VNET(tcp_insecure_syn) SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_syn), 0, "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); VNET_DEFINE(int, tcp_insecure_rst) = 0; #define V_tcp_insecure_rst VNET(tcp_insecure_rst) SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_rst), 0, "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); VNET_DEFINE(int, tcp_recvspace) = 1024*64; #define V_tcp_recvspace VNET(tcp_recvspace) SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_autorcvbuf), 0, "Enable automatic receive buffer sizing"); VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autorcvbuf_inc), 0, "Incrementor step size of automatic receive buffer"); VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autorcvbuf_max), 0, "Max size of automatic receive buffer"); VNET_DEFINE(struct inpcbhead, tcb); #define tcb6 tcb /* for KAME src sync over BSD*'s */ VNET_DEFINE(struct inpcbinfo, tcbinfo); /* * TCP statistics are stored in an array of counter(9)s, which size matches * size of struct tcpstat. TCP running connection count is a regular array. */ VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, "TCP connection counts by TCP state"); static void tcp_vnet_init(const void *unused) { COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); } VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, tcp_vnet_init, NULL); #ifdef VIMAGE static void tcp_vnet_uninit(const void *unused) { COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); VNET_PCPUSTAT_FREE(tcpstat); } VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, tcp_vnet_uninit, NULL); #endif /* VIMAGE */ /* * Kernel module interface for updating tcpstat. The argument is an index * into tcpstat treated as an array. */ void kmod_tcpstat_inc(int statnum) { counter_u64_add(VNET(tcpstat)[statnum], 1); } #ifdef TCP_HHOOK /* * Wrapper for the TCP established input helper hook. */ void hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) { struct tcp_hhook_data hhook_data; if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { hhook_data.tp = tp; hhook_data.th = th; hhook_data.to = to; hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, tp->osd); } } #endif /* * CC wrapper hook functions */ void cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, uint16_t type) { INP_WLOCK_ASSERT(tp->t_inpcb); tp->ccv->nsegs = nsegs; tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); if (tp->snd_cwnd <= tp->snd_wnd) tp->ccv->flags |= CCF_CWND_LIMITED; else tp->ccv->flags &= ~CCF_CWND_LIMITED; if (type == CC_ACK) { if (tp->snd_cwnd > tp->snd_ssthresh) { tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); if (tp->t_bytes_acked >= tp->snd_cwnd) { tp->t_bytes_acked -= tp->snd_cwnd; tp->ccv->flags |= CCF_ABC_SENTAWND; } } else { tp->ccv->flags &= ~CCF_ABC_SENTAWND; tp->t_bytes_acked = 0; } } if (CC_ALGO(tp)->ack_received != NULL) { /* XXXLAS: Find a way to live without this */ tp->ccv->curack = th->th_ack; CC_ALGO(tp)->ack_received(tp->ccv, type); } } void cc_conn_init(struct tcpcb *tp) { struct hc_metrics_lite metrics; struct inpcb *inp = tp->t_inpcb; u_int maxseg; int rtt; INP_WLOCK_ASSERT(tp->t_inpcb); tcp_hc_get(&inp->inp_inc, &metrics); maxseg = tcp_maxseg(tp); if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { tp->t_srtt = rtt; tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; TCPSTAT_INC(tcps_usedrtt); if (metrics.rmx_rttvar) { tp->t_rttvar = metrics.rmx_rttvar; TCPSTAT_INC(tcps_usedrttvar); } else { /* default variation is +- 1 rtt */ tp->t_rttvar = tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; } TCPT_RANGESET(tp->t_rxtcur, ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, tp->t_rttmin, TCPTV_REXMTMAX); } if (metrics.rmx_ssthresh) { /* * There's some sort of gateway or interface * buffer limit on the path. Use this to set * the slow start threshold, but set the * threshold to no less than 2*mss. */ tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); TCPSTAT_INC(tcps_usedssthresh); } /* * Set the initial slow-start flight size. * * RFC5681 Section 3.1 specifies the default conservative values. * RFC3390 specifies slightly more aggressive values. * RFC6928 increases it to ten segments. * Support for user specified value for initial flight size. * * If a SYN or SYN/ACK was lost and retransmitted, we have to * reduce the initial CWND to one segment as congestion is likely * requiring us to be cautious. */ if (tp->snd_cwnd == 1) tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ else if (V_tcp_initcwnd_segments) tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); else if (V_tcp_do_rfc3390) tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); else { /* Per RFC5681 Section 3.1 */ if (maxseg > 2190) tp->snd_cwnd = 2 * maxseg; else if (maxseg > 1095) tp->snd_cwnd = 3 * maxseg; else tp->snd_cwnd = 4 * maxseg; } if (CC_ALGO(tp)->conn_init != NULL) CC_ALGO(tp)->conn_init(tp->ccv); } void inline cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) { u_int maxseg; INP_WLOCK_ASSERT(tp->t_inpcb); switch(type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(tp->t_flags)) { tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_ECN_PERMIT) tp->t_flags |= TF_ECN_SND_CWR; } break; case CC_ECN: if (!IN_CONGRECOVERY(tp->t_flags)) { TCPSTAT_INC(tcps_ecn_rcwnd); tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_ECN_PERMIT) tp->t_flags |= TF_ECN_SND_CWR; } break; case CC_RTO: maxseg = tcp_maxseg(tp); tp->t_dupacks = 0; tp->t_bytes_acked = 0; EXIT_RECOVERY(tp->t_flags); - if (CC_ALGO(tp)->cong_signal == NULL) { - /* - * RFC5681 Section 3.1 - * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4) - */ - tp->snd_ssthresh = - max((tp->snd_max - tp->snd_una) / 2, 2 * maxseg); - tp->snd_cwnd = maxseg; - } + tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / + maxseg) * maxseg; + tp->snd_cwnd = maxseg; break; case CC_RTO_ERR: TCPSTAT_INC(tcps_sndrexmitbad); /* RTO was unnecessary, so reset everything. */ tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; tp->snd_recover = tp->snd_recover_prev; if (tp->t_flags & TF_WASFRECOVERY) ENTER_FASTRECOVERY(tp->t_flags); if (tp->t_flags & TF_WASCRECOVERY) ENTER_CONGRECOVERY(tp->t_flags); tp->snd_nxt = tp->snd_max; tp->t_flags &= ~TF_PREVVALID; tp->t_badrxtwin = 0; break; } if (CC_ALGO(tp)->cong_signal != NULL) { if (th != NULL) tp->ccv->curack = th->th_ack; CC_ALGO(tp)->cong_signal(tp->ccv, type); } } void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) { INP_WLOCK_ASSERT(tp->t_inpcb); /* XXXLAS: KASSERT that we're in recovery? */ if (CC_ALGO(tp)->post_recovery != NULL) { tp->ccv->curack = th->th_ack; CC_ALGO(tp)->post_recovery(tp->ccv); } /* XXXLAS: EXIT_RECOVERY ? */ tp->t_bytes_acked = 0; } #ifdef TCP_SIGNATURE static inline int tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) { int ret; tcp_fields_to_net(th); ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); tcp_fields_to_host(th); return (ret); } #endif /* * Indicate whether this ack should be delayed. We can delay the ack if * following conditions are met: * - There is no delayed ack timer in progress. * - Our last ack wasn't a 0-sized window. We never want to delay * the ack that opens up a 0-sized window. * - LRO wasn't used for this segment. We make sure by checking that the * segment size is not larger than the MSS. */ #define DELAY_ACK(tp, tlen) \ ((!tcp_timer_active(tp, TT_DELACK) && \ (tp->t_flags & TF_RXWIN0SENT) == 0) && \ (tlen <= tp->t_maxseg) && \ (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) static void inline cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) { INP_WLOCK_ASSERT(tp->t_inpcb); if (CC_ALGO(tp)->ecnpkt_handler != NULL) { switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: tp->ccv->flags |= CCF_IPHDR_CE; break; case IPTOS_ECN_ECT0: tp->ccv->flags &= ~CCF_IPHDR_CE; break; case IPTOS_ECN_ECT1: tp->ccv->flags &= ~CCF_IPHDR_CE; break; } if (th->th_flags & TH_CWR) tp->ccv->flags |= CCF_TCPHDR_CWR; else tp->ccv->flags &= ~CCF_TCPHDR_CWR; if (tp->t_flags & TF_DELACK) tp->ccv->flags |= CCF_DELACK; else tp->ccv->flags &= ~CCF_DELACK; CC_ALGO(tp)->ecnpkt_handler(tp->ccv); if (tp->ccv->flags & CCF_ACKNOW) tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); } } /* * TCP input handling is split into multiple parts: * tcp6_input is a thin wrapper around tcp_input for the extended * ip6_protox[] call format in ip6_input * tcp_input handles primary segment validation, inpcb lookup and * SYN processing on listen sockets * tcp_do_segment processes the ACK and text of the segment for * establishing, established and closing connections */ #ifdef INET6 int tcp6_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp; struct in6_ifaddr *ia6; struct ip6_hdr *ip6; IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); /* * draft-itojun-ipv6-tcp-to-anycast * better place to put this in? */ ip6 = mtod(m, struct ip6_hdr *); ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { struct ip6_hdr *ip6; ifa_free(&ia6->ia_ifa); ip6 = mtod(m, struct ip6_hdr *); icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); return (IPPROTO_DONE); } if (ia6) ifa_free(&ia6->ia_ifa); return (tcp_input(mp, offp, proto)); } #endif /* INET6 */ int tcp_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp; struct tcphdr *th = NULL; struct ip *ip = NULL; struct inpcb *inp = NULL; struct tcpcb *tp = NULL; struct socket *so = NULL; u_char *optp = NULL; int off0; int optlen = 0; #ifdef INET int len; #endif int tlen = 0, off; int drop_hdrlen; int thflags; int rstreason = 0; /* For badport_bandlim accounting purposes */ #ifdef TCP_SIGNATURE uint8_t sig_checked = 0; #endif uint8_t iptos; struct m_tag *fwd_tag = NULL; #ifdef INET6 struct ip6_hdr *ip6 = NULL; int isipv6; #else const void *ip6 = NULL; #endif /* INET6 */ struct tcpopt to; /* options in this segment */ char *s = NULL; /* address and port logging */ int ti_locked; #ifdef TCPDEBUG /* * The size of tcp_saveipgen must be the size of the max ip header, * now IPv6. */ u_char tcp_saveipgen[IP6_HDR_LEN]; struct tcphdr tcp_savetcp; short ostate = 0; #endif #ifdef INET6 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; #endif off0 = *offp; m = *mp; *mp = NULL; to.to_flags = 0; TCPSTAT_INC(tcps_rcvtotal); #ifdef INET6 if (isipv6) { /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); if (m == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in6_cksum_pseudo(ip6, tlen, IPPROTO_TCP, m->m_pkthdr.csum_data); th->th_sum ^= 0xffff; } else th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); if (th->th_sum) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } /* * Be proactive about unspecified IPv6 address in source. * As we use all-zero to indicate unbounded/unconnected pcb, * unspecified IPv6 address can be used to confuse us. * * Note that packets with unspecified IPv6 destination is * already dropped in ip6_input. */ if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { /* XXX stat */ goto drop; } iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { /* * Get IP and TCP header together in first mbuf. * Note: IP leaves IP header in first mbuf. */ if (off0 > sizeof (struct ip)) { ip_stripoptions(m); off0 = sizeof(struct ip); } if (m->m_len < sizeof (struct tcpiphdr)) { if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); tlen = ntohs(ip->ip_len) - off0; iptos = ip->ip_tos; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + tlen + IPPROTO_TCP)); th->th_sum ^= 0xffff; } else { struct ipovly *ipov = (struct ipovly *)ip; /* * Checksum extended TCP header and data. */ len = off0 + tlen; bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); ipov->ih_len = htons(tlen); th->th_sum = in_cksum(m, len); /* Reset length for SDT probes. */ ip->ip_len = htons(len); /* Reset TOS bits */ ip->ip_tos = iptos; /* Re-initialization for later version check */ ip->ip_v = IPVERSION; } if (th->th_sum) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } } #endif /* INET */ /* * Check that TCP offset makes sense, * pull out TCP options and adjust length. XXX */ off = th->th_off << 2; if (off < sizeof (struct tcphdr) || off > tlen) { TCPSTAT_INC(tcps_rcvbadoff); goto drop; } tlen -= off; /* tlen is used instead of ti->ti_len */ if (off > sizeof (struct tcphdr)) { #ifdef INET6 if (isipv6) { IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (m->m_len < sizeof(struct ip) + off) { if ((m = m_pullup(m, sizeof (struct ip) + off)) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); } } #endif optlen = off - sizeof (struct tcphdr); optp = (u_char *)(th + 1); } thflags = th->th_flags; /* * Convert TCP protocol specific fields to host format. */ tcp_fields_to_host(th); /* * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. */ drop_hdrlen = off0 + off; /* * Locate pcb for segment; if we're likely to add or remove a * connection then first acquire pcbinfo lock. There are three cases * where we might discover later we need a write lock despite the * flags: ACKs moving a connection out of the syncache, ACKs for a * connection in TIMEWAIT and SYNs not targeting a listening socket. */ if ((thflags & (TH_FIN | TH_RST)) != 0) { INP_INFO_RLOCK(&V_tcbinfo); ti_locked = TI_RLOCKED; } else ti_locked = TI_UNLOCKED; /* * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ if ( #ifdef INET6 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) #ifdef INET || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) #endif #endif #if defined(INET) && !defined(INET6) (m->m_flags & M_IP_NEXTHOP) #endif ) fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); findpcb: #ifdef INVARIANTS if (ti_locked == TI_RLOCKED) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); } else { INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); } #endif #ifdef INET6 if (isipv6 && fwd_tag != NULL) { struct sockaddr_in6 *next_hop6; next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); /* * Transparently forwarded. Pretend to be the destination. * Already got one like this? */ inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &next_hop6->sin6_addr, next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); } } else if (isipv6) { inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET if (fwd_tag != NULL) { struct sockaddr_in *next_hop; next_hop = (struct sockaddr_in *)(fwd_tag+1); /* * Transparently forwarded. Pretend to be the destination. * already got one like this? */ inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in_pcblookup(&V_tcbinfo, ip->ip_src, th->th_sport, next_hop->sin_addr, next_hop->sin_port ? ntohs(next_hop->sin_port) : th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); } } else inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); #endif /* INET */ /* * If the INPCB does not exist then all data in the incoming * segment is discarded and an appropriate RST is sent back. * XXX MRT Send RST using which routing table? */ if (inp == NULL) { /* * Log communication attempts to ports that are not * in use. */ if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || tcp_log_in_vain == 2) { if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) log(LOG_INFO, "%s; %s: Connection attempt " "to closed port\n", s, __func__); } /* * When blackholing do not respond with a RST but * completely ignore the segment and drop it. */ if ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole == 2) goto dropunlock; rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } INP_WLOCK_ASSERT(inp); /* * While waiting for inp lock during the lookup, another thread * can have dropped the inpcb, in which case we need to loop back * and try to find a new inpcb to deliver to. */ if (inp->inp_flags & INP_DROPPED) { INP_WUNLOCK(inp); inp = NULL; goto findpcb; } if ((inp->inp_flowtype == M_HASHTYPE_NONE) && (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && ((inp->inp_socket == NULL) || (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { inp->inp_flowid = m->m_pkthdr.flowid; inp->inp_flowtype = M_HASHTYPE_GET(m); } #ifdef IPSEC #ifdef INET6 if (isipv6 && ipsec6_in_reject(m, inp)) { goto dropunlock; } else #endif /* INET6 */ if (ipsec4_in_reject(m, inp) != 0) { goto dropunlock; } #endif /* IPSEC */ /* * Check the minimum TTL for socket. */ if (inp->inp_ip_minttl != 0) { #ifdef INET6 if (isipv6) { if (inp->inp_ip_minttl > ip6->ip6_hlim) goto dropunlock; } else #endif if (inp->inp_ip_minttl > ip->ip_ttl) goto dropunlock; } /* * A previous connection in TIMEWAIT state is supposed to catch stray * or duplicate segments arriving late. If this segment was a * legitimate new connection attempt, the old INPCB gets removed and * we can try again to find a listening socket. * * At this point, due to earlier optimism, we may hold only an inpcb * lock, and not the inpcbinfo write lock. If so, we need to try to * acquire it, or if that fails, acquire a reference on the inpcb, * drop all locks, acquire a global write lock, and then re-acquire * the inpcb lock. We may at that point discover that another thread * has tried to free the inpcb, in which case we need to loop back * and try to find a new inpcb to deliver to. * * XXXRW: It may be time to rethink timewait locking. */ relocked: if (inp->inp_flags & INP_TIMEWAIT) { if (ti_locked == TI_UNLOCKED) { if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { in_pcbref(inp); INP_WUNLOCK(inp); INP_INFO_RLOCK(&V_tcbinfo); ti_locked = TI_RLOCKED; INP_WLOCK(inp); if (in_pcbrele_wlocked(inp)) { inp = NULL; goto findpcb; } else if (inp->inp_flags & INP_DROPPED) { INP_WUNLOCK(inp); inp = NULL; goto findpcb; } } else ti_locked = TI_RLOCKED; } INP_INFO_RLOCK_ASSERT(&V_tcbinfo); if (thflags & TH_SYN) tcp_dooptions(&to, optp, optlen, TO_SYN); /* * NB: tcp_twcheck unlocks the INP and frees the mbuf. */ if (tcp_twcheck(inp, &to, th, m, tlen)) goto findpcb; INP_INFO_RUNLOCK(&V_tcbinfo); return (IPPROTO_DONE); } /* * The TCPCB may no longer exist if the connection is winding * down or it is in the CLOSED state. Either way we drop the * segment and send an appropriate response. */ tp = intotcpcb(inp); if (tp == NULL || tp->t_state == TCPS_CLOSED) { rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE) { tcp_offload_input(tp, m); m = NULL; /* consumed by the TOE driver */ goto dropunlock; } #endif /* * We've identified a valid inpcb, but it could be that we need an * inpcbinfo write lock but don't hold it. In this case, attempt to * acquire using the same strategy as the TIMEWAIT case above. If we * relock, we have to jump back to 'relocked' as the connection might * now be in TIMEWAIT. */ #ifdef INVARIANTS if ((thflags & (TH_FIN | TH_RST)) != 0) INP_INFO_RLOCK_ASSERT(&V_tcbinfo); #endif if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && !IS_FASTOPEN(tp->t_flags)))) { if (ti_locked == TI_UNLOCKED) { if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { in_pcbref(inp); INP_WUNLOCK(inp); INP_INFO_RLOCK(&V_tcbinfo); ti_locked = TI_RLOCKED; INP_WLOCK(inp); if (in_pcbrele_wlocked(inp)) { inp = NULL; goto findpcb; } else if (inp->inp_flags & INP_DROPPED) { INP_WUNLOCK(inp); inp = NULL; goto findpcb; } goto relocked; } else ti_locked = TI_RLOCKED; } INP_INFO_RLOCK_ASSERT(&V_tcbinfo); } #ifdef MAC INP_WLOCK_ASSERT(inp); if (mac_inpcb_check_deliver(inp, m)) goto dropunlock; #endif so = inp->inp_socket; KASSERT(so != NULL, ("%s: so == NULL", __func__)); #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) { ostate = tp->t_state; #ifdef INET6 if (isipv6) { bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); } else #endif bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); tcp_savetcp = *th; } #endif /* TCPDEBUG */ /* * When the socket is accepting connections (the INPCB is in LISTEN * state) we look into the SYN cache if this is a new connection * attempt or the completion of a previous one. */ KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), ("%s: so accepting but tp %p not listening", __func__, tp)); if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { struct in_conninfo inc; bzero(&inc, sizeof(inc)); #ifdef INET6 if (isipv6) { inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = ip6->ip6_src; inc.inc6_laddr = ip6->ip6_dst; } else #endif { inc.inc_faddr = ip->ip_src; inc.inc_laddr = ip->ip_dst; } inc.inc_fport = th->th_sport; inc.inc_lport = th->th_dport; inc.inc_fibnum = so->so_fibnum; /* * Check for an existing connection attempt in syncache if * the flag is only ACK. A successful lookup creates a new * socket appended to the listen queue in SYN_RECEIVED state. */ if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* * Parse the TCP options here because * syncookies need access to the reflected * timestamp. */ tcp_dooptions(&to, optp, optlen, 0); /* * NB: syncache_expand() doesn't unlock * inp and tcpinfo locks. */ if (!syncache_expand(&inc, &to, th, &so, m)) { /* * No syncache entry or ACK was not * for our SYN/ACK. Send a RST. * NB: syncache did its own logging * of the failure cause. */ rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } #ifdef TCP_RFC7413 tfo_socket_result: #endif if (so == NULL) { /* * We completed the 3-way handshake * but could not allocate a socket * either due to memory shortage, * listen queue length limits or * global socket limits. Send RST * or wait and have the remote end * retransmit the ACK for another * try. */ if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Socket allocation failed due to " "limits or memory shortage, %s\n", s, __func__, V_tcp_sc_rst_sock_fail ? "sending RST" : "try again"); if (V_tcp_sc_rst_sock_fail) { rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } else goto dropunlock; } /* * Socket is created in state SYN_RECEIVED. * Unlock the listen socket, lock the newly * created socket and update the tp variable. */ INP_WUNLOCK(inp); /* listen socket */ inp = sotoinpcb(so); /* * New connection inpcb is already locked by * syncache_expand(). */ INP_WLOCK_ASSERT(inp); tp = intotcpcb(inp); KASSERT(tp->t_state == TCPS_SYN_RECEIVED, ("%s: ", __func__)); #ifdef TCP_SIGNATURE if (sig_checked == 0) { tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0); if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, th, tp->t_flags)) { /* * In SYN_SENT state if it receives an * RST, it is allowed for further * processing. */ if ((thflags & TH_RST) == 0 || (tp->t_state == TCPS_SYN_SENT) == 0) goto dropunlock; } sig_checked = 1; } #endif /* * Process the segment and the data it * contains. tcp_do_segment() consumes * the mbuf chain and unlocks the inpcb. */ tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); return (IPPROTO_DONE); } /* * Segment flag validation for new connection attempts: * * Our (SYN|ACK) response was rejected. * Check with syncache and remove entry to prevent * retransmits. * * NB: syncache_chkrst does its own logging of failure * causes. */ if (thflags & TH_RST) { syncache_chkrst(&inc, th); goto dropunlock; } /* * We can't do anything without SYN. */ if ((thflags & TH_SYN) == 0) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN is missing, segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * (SYN|ACK) is bogus on a listen socket. */ if (thflags & TH_ACK) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|ACK invalid, segment rejected\n", s, __func__); syncache_badack(&inc); /* XXX: Not needed! */ TCPSTAT_INC(tcps_badsyn); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } /* * If the drop_synfin option is enabled, drop all * segments with both the SYN and FIN bits set. * This prevents e.g. nmap from identifying the * TCP/IP stack. * XXX: Poor reasoning. nmap has other methods * and is constantly refining its stack detection * strategies. * XXX: This is a violation of the TCP specification * and was used by RFC1644. */ if ((thflags & TH_FIN) && V_drop_synfin) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|FIN segment ignored (based on " "sysctl setting)\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * Segment's flags are (SYN) or (SYN|FIN). * * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored * as they do not affect the state of the TCP FSM. * The data pointed to by TH_URG and th_urp is ignored. */ KASSERT((thflags & (TH_RST|TH_ACK)) == 0, ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); KASSERT(thflags & (TH_SYN), ("%s: Listen socket: TH_SYN not set", __func__)); #ifdef INET6 /* * If deprecated address is forbidden, * we do not accept SYN to deprecated interface * address to prevent any new inbound connection from * getting established. * When we do not accept SYN, we send a TCP RST, * with deprecated source address (instead of dropping * it). We compromise it as it is much better for peer * to send a RST, and RST will be the final packet * for the exchange. * * If we do not forbid deprecated addresses, we accept * the SYN packet. RFC2462 does not suggest dropping * SYN in this case. * If we decipher RFC2462 5.5.4, it says like this: * 1. use of deprecated addr with existing * communication is okay - "SHOULD continue to be * used" * 2. use of it with new communication: * (2a) "SHOULD NOT be used if alternate address * with sufficient scope is available" * (2b) nothing mentioned otherwise. * Here we fall into (2b) case as we have no choice in * our source address selection - we must obey the peer. * * The wording in RFC2462 is confusing, and there are * multiple description text for deprecated address * handling - worse, they are not exactly the same. * I believe 5.5.4 is the best one, so we follow 5.5.4. */ if (isipv6 && !V_ip6_use_deprecated) { struct in6_ifaddr *ia6; ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia6 != NULL && (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { ifa_free(&ia6->ia_ifa); if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to deprecated " "IPv6 address rejected\n", s, __func__); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } if (ia6) ifa_free(&ia6->ia_ifa); } #endif /* INET6 */ /* * Basic sanity checks on incoming SYN requests: * Don't respond if the destination is a link layer * broadcast according to RFC1122 4.2.3.10, p. 104. * If it is from this socket it must be forged. * Don't respond if the source or destination is a * global or subnet broad- or multicast address. * Note that it is quite possible to receive unicast * link-layer packets with a broadcast IP address. Use * in_broadcast() to find them. */ if (m->m_flags & (M_BCAST|M_MCAST)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from broad- or multicast " "link layer address ignored\n", s, __func__); goto dropunlock; } #ifdef INET6 if (isipv6) { if (th->th_dport == th->th_sport && IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to/from self " "ignored\n", s, __func__); goto dropunlock; } if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to multicast " "address ignored\n", s, __func__); goto dropunlock; } } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (th->th_dport == th->th_sport && ip->ip_dst.s_addr == ip->ip_src.s_addr) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to self " "ignored\n", s, __func__); goto dropunlock; } if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to broad- " "or multicast address ignored\n", s, __func__); goto dropunlock; } } #endif /* * SYN appears to be valid. Create compressed TCP state * for syncache. */ #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); tcp_dooptions(&to, optp, optlen, TO_SYN); #ifdef TCP_RFC7413 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) goto tfo_socket_result; #else syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); #endif /* * Entry added to syncache and mbuf consumed. * Only the listen socket is unlocked by syncache_add(). */ if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; } INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); return (IPPROTO_DONE); } else if (tp->t_state == TCPS_LISTEN) { /* * When a listen socket is torn down the SO_ACCEPTCONN * flag is removed first while connections are drained * from the accept queue in a unlock/lock cycle of the * ACCEPT_LOCK, opening a race condition allowing a SYN * attempt go through unhandled. */ goto dropunlock; } #ifdef TCP_SIGNATURE if (sig_checked == 0) { tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0); if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, th, tp->t_flags)) { /* * In SYN_SENT state if it receives an RST, it is * allowed for further processing. */ if ((thflags & TH_RST) == 0 || (tp->t_state == TCPS_SYN_SENT) == 0) goto dropunlock; } sig_checked = 1; } #endif TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); /* * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later * state. tcp_do_segment() always consumes the mbuf chain, unlocks * the inpcb, and unlocks pcbinfo. */ tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); return (IPPROTO_DONE); dropwithreset: TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; } #ifdef INVARIANTS else { KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " "ti_locked: %d", __func__, ti_locked)); INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); } #endif if (inp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_WUNLOCK(inp); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); m = NULL; /* mbuf chain got consumed. */ goto drop; dropunlock: if (m != NULL) TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; } #ifdef INVARIANTS else { KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " "ti_locked: %d", __func__, ti_locked)); INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); } #endif if (inp != NULL) INP_WUNLOCK(inp); drop: INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); if (s != NULL) free(s, M_TCPLOG); if (m != NULL) m_freem(m); return (IPPROTO_DONE); } void tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, int ti_locked) { int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; int rstreason, todrop, win; uint32_t tiwin; uint16_t nsegs; char *s; struct in_conninfo *inc; struct mbuf *mfree; struct tcpopt to; #ifdef TCP_RFC7413 int tfo_syn; #endif #ifdef TCPDEBUG /* * The size of tcp_saveipgen must be the size of the max ip header, * now IPv6. */ u_char tcp_saveipgen[IP6_HDR_LEN]; struct tcphdr tcp_savetcp; short ostate = 0; #endif thflags = th->th_flags; inc = &tp->t_inpcb->inp_inc; tp->sackhint.last_sack_ack = 0; sack_changed = 0; nsegs = max(1, m->m_pkthdr.lro_nsegs); /* * If this is either a state-changing packet or current state isn't * established, we require a write lock on tcbinfo. Otherwise, we * allow the tcbinfo to be in either alocked or unlocked, as the * caller may have unnecessarily acquired a write lock due to a race. */ if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || tp->t_state != TCPS_ESTABLISHED) { KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " "SYN/FIN/RST/!EST", __func__, ti_locked)); INP_INFO_RLOCK_ASSERT(&V_tcbinfo); } else { #ifdef INVARIANTS if (ti_locked == TI_RLOCKED) INP_INFO_RLOCK_ASSERT(&V_tcbinfo); else { KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " "ti_locked: %d", __func__, ti_locked)); INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); } #endif } INP_WLOCK_ASSERT(tp->t_inpcb); KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", __func__)); #ifdef TCPPCAP /* Save segment, if requested. */ tcp_pcap_add(th, m, &(tp->t_inpkts)); #endif /* * Segment received on connection. * Reset idle time and keep-alive timer. * XXX: This should be done after segment * validation to ignore broken/spoofed segs. */ tp->t_rcvtime = ticks; /* * Scale up the window into a 32-bit value. * For the SYN_SENT state the scale is zero. */ tiwin = th->th_win << tp->snd_scale; /* * TCP ECN processing. */ if (tp->t_flags & TF_ECN_PERMIT) { if (thflags & TH_CWR) tp->t_flags &= ~TF_ECN_SND_ECE; switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: tp->t_flags |= TF_ECN_SND_ECE; TCPSTAT_INC(tcps_ecn_ce); break; case IPTOS_ECN_ECT0: TCPSTAT_INC(tcps_ecn_ect0); break; case IPTOS_ECN_ECT1: TCPSTAT_INC(tcps_ecn_ect1); break; } /* Process a packet differently from RFC3168. */ cc_ecnpkt_handler(tp, th, iptos); /* Congestion experienced. */ if (thflags & TH_ECE) { cc_cong_signal(tp, th, CC_ECN); } } /* * Parse options on any incoming segment. */ tcp_dooptions(&to, (u_char *)(th + 1), (th->th_off << 2) - sizeof(struct tcphdr), (thflags & TH_SYN) ? TO_SYN : 0); /* * If echoed timestamp is later than the current time, * fall back to non RFC1323 RTT calculation. Normalize * timestamp if syncookies were used when this connection * was established. */ if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { to.to_tsecr -= tp->ts_offset; if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) to.to_tsecr = 0; } /* * If timestamps were negotiated during SYN/ACK they should * appear on every segment during this session and vice versa. */ if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "no action\n", s, __func__); free(s, M_TCPLOG); } } if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp not expected, " "no action\n", s, __func__); free(s, M_TCPLOG); } } /* * Process options only when we get SYN/ACK back. The SYN case * for incoming connections is handled in tcp_syncache. * According to RFC1323 the window field in a SYN (i.e., a * or ) segment itself is never scaled. * XXX this is traditional behavior, may need to be cleaned up. */ if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) { tp->t_flags |= TF_RCVD_SCALE; tp->snd_scale = to.to_wscale; } /* * Initial send window. It will be updated with * the next incoming segment to the scaled value. */ tp->snd_wnd = th->th_win; if (to.to_flags & TOF_TS) { tp->t_flags |= TF_RCVD_TSTMP; tp->ts_recent = to.to_tsval; tp->ts_recent_age = tcp_ts_getticks(); } if (to.to_flags & TOF_MSS) tcp_mss(tp, to.to_mss); if ((tp->t_flags & TF_SACK_PERMIT) && (to.to_flags & TOF_SACKPERM) == 0) tp->t_flags &= ~TF_SACK_PERMIT; } /* * Header prediction: check for the two common cases * of a uni-directional data xfer. If the packet has * no control flags, is in-sequence, the window didn't * change and we're not retransmitting, it's a * candidate. If the length is zero and the ack moved * forward, we're the sender side of the xfer. Just * free the data acked & wake any higher level process * that was blocked waiting for space. If the length * is non-zero and the ack didn't move, we're the * receiver side. If we're getting packets in-order * (the reassembly queue is empty), add the data to * the socket buffer and note that we need a delayed ack. * Make sure that the hidden state-flags are also off. * Since we check for TCPS_ESTABLISHED first, it can only * be TH_NEEDSYN. */ if (tp->t_state == TCPS_ESTABLISHED && th->th_seq == tp->rcv_nxt && (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && tp->snd_nxt == tp->snd_max && tiwin && tiwin == tp->snd_wnd && ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && LIST_EMPTY(&tp->t_segq) && ((to.to_flags & TOF_TS) == 0 || TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { /* * If last ACK falls within this segment's sequence numbers, * record the timestamp. * NOTE that the test is modified according to the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } if (tlen == 0) { if (SEQ_GT(th->th_ack, tp->snd_una) && SEQ_LEQ(th->th_ack, tp->snd_max) && !IN_RECOVERY(tp->t_flags) && (to.to_flags & TOF_SACK) == 0 && TAILQ_EMPTY(&tp->snd_holes)) { /* * This is a pure ack for outstanding data. */ if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; TCPSTAT_INC(tcps_predack); /* * "bad retransmit" recovery. */ if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && (int)(ticks - tp->t_badrxtwin) < 0) { cc_cong_signal(tp, th, CC_RTO_ERR); } /* * Recalculate the transmit timer / rtt. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } acked = BYTES_THIS_ACK(tp, th); #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); sbdrop(&so->so_snd, acked); if (SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* * Let the congestion control algorithm update * congestion control related information. This * typically means increasing the congestion * window. */ cc_ack_received(tp, th, nsegs, CC_ACK); tp->snd_una = th->th_ack; /* * Pull snd_wl2 up to prevent seq wrap relative * to th_ack. */ tp->snd_wl2 = th->th_ack; tp->t_dupacks = 0; m_freem(m); /* * If all outstanding data are acked, stop * retransmit timer, otherwise restart timer * using current (possibly backed-off) value. * If process is waiting for space, * wakeup/selwakeup/signal. If data * are ready to send, let tcp_output * decide between more output or persist. */ #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); if (tp->snd_una == tp->snd_max) tcp_timer_activate(tp, TT_REXMT, 0); else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); sowwakeup(so); if (sbavail(&so->so_snd)) (void) tp->t_fb->tfb_tcp_output(tp); goto check_delack; } } else if (th->th_ack == tp->snd_una && tlen <= sbspace(&so->so_rcv)) { int newsize = 0; /* automatic sockbuf scaling */ /* * This is a pure, in-sequence data packet with * nothing on the reassembly queue and we have enough * buffer space to take it. */ if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; /* Clean receiver SACK report if present */ if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) tcp_clean_sackreport(tp); TCPSTAT_INC(tcps_preddat); tp->rcv_nxt += tlen; /* * Pull snd_wl1 up to prevent seq wrap relative to * th_seq. */ tp->snd_wl1 = th->th_seq; /* * Pull rcv_up up to prevent seq wrap relative to * rcv_nxt. */ tp->rcv_up = tp->rcv_nxt; TCPSTAT_ADD(tcps_rcvpack, nsegs); TCPSTAT_ADD(tcps_rcvbyte, tlen); #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); /* * Automatic sizing of receive socket buffer. Often the send * buffer size is not optimally adjusted to the actual network * conditions at hand (delay bandwidth product). Setting the * buffer size too small limits throughput on links with high * bandwidth and high delay (eg. trans-continental/oceanic links). * * On the receive side the socket buffer memory is only rarely * used to any significant extent. This allows us to be much * more aggressive in scaling the receive socket buffer. For * the case that the buffer space is actually used to a large * extent and we run out of kernel memory we can simply drop * the new segments; TCP on the sender will just retransmit it * later. Setting the buffer size too big may only consume too * much kernel memory if the application doesn't read() from * the socket or packet loss or reordering makes use of the * reassembly queue. * * The criteria to step up the receive buffer one notch are: * 1. Application has not set receive buffer size with * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. * 2. the number of bytes received during the time it takes * one timestamp to be reflected back to us (the RTT); * 3. received bytes per RTT is within seven eighth of the * current socket buffer size; * 4. receive buffer size has not hit maximal automatic size; * * This algorithm does one step per RTT at most and only if * we receive a bulk stream w/o packet losses or reorderings. * Shrinking the buffer during idle times is not necessary as * it doesn't consume any memory when idle. * * TODO: Only step up if the application is actually serving * the buffer to better manage the socket buffer resources. */ if (V_tcp_do_autorcvbuf && (to.to_flags & TOF_TS) && to.to_tsecr && (so->so_rcv.sb_flags & SB_AUTOSIZE)) { if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && to.to_tsecr - tp->rfbuf_ts < hz) { if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { newsize = min(so->so_rcv.sb_hiwat + V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); } /* Start over with next RTT. */ tp->rfbuf_ts = 0; tp->rfbuf_cnt = 0; } else tp->rfbuf_cnt += tlen; /* add up */ } /* Add data to socket buffer. */ SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { m_freem(m); } else { /* * Set new socket buffer size. * Give up when limit is reached. */ if (newsize) if (!sbreserve_locked(&so->so_rcv, newsize, so, NULL)) so->so_rcv.sb_flags &= ~SB_AUTOSIZE; m_adj(m, drop_hdrlen); /* delayed header drop */ sbappendstream_locked(&so->so_rcv, m, 0); } /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); if (DELAY_ACK(tp, tlen)) { tp->t_flags |= TF_DELACK; } else { tp->t_flags |= TF_ACKNOW; tp->t_fb->tfb_tcp_output(tp); } goto check_delack; } } /* * Calculate amount of space in receive window, * and then do TCP input processing. * Receive window is amount of space in rcv queue, * but not less than advertised window. */ win = sbspace(&so->so_rcv); if (win < 0) win = 0; tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); /* Reset receive buffer auto scaling when not in bulk receive mode. */ tp->rfbuf_ts = 0; tp->rfbuf_cnt = 0; switch (tp->t_state) { /* * If the state is SYN_RECEIVED: * if seg contains an ACK, but not for our SYN/ACK, send a RST. */ case TCPS_SYN_RECEIVED: if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } #ifdef TCP_RFC7413 if (IS_FASTOPEN(tp->t_flags)) { /* * When a TFO connection is in SYN_RECEIVED, the * only valid packets are the initial SYN, a * retransmit/copy of the initial SYN (possibly with * a subset of the original data), a valid ACK, a * FIN, or a RST. */ if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } else if (thflags & TH_SYN) { /* non-initial SYN is ignored */ if ((tcp_timer_active(tp, TT_DELACK) || tcp_timer_active(tp, TT_REXMT))) goto drop; } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { goto drop; } } #endif break; /* * If the state is SYN_SENT: * if seg contains an ACK, but not for our SYN, drop the input. * if seg contains a RST, then drop the connection. * if seg does not contain SYN, then drop it. * Otherwise this is an acceptable SYN segment * initialize tp->rcv_nxt and tp->irs * if seg contains ack then advance tp->snd_una * if seg contains an ECE and ECN support is enabled, the stream * is ECN capable. * if SYN has been acked change to ESTABLISHED else SYN_RCVD state * arrange for segment to be acked (eventually) * continue processing rest of data/controls, beginning with URG */ case TCPS_SYN_SENT: if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { TCP_PROBE5(connect__refused, NULL, tp, mtod(m, const char *), tp, th); tp = tcp_drop(tp, ECONNREFUSED); } if (thflags & TH_RST) goto drop; if (!(thflags & TH_SYN)) goto drop; tp->irs = th->th_seq; tcp_rcvseqinit(tp); if (thflags & TH_ACK) { TCPSTAT_INC(tcps_connects); soisconnected(so); #ifdef MAC mac_socketpeer_set_from_mbuf(m, so); #endif /* Do window scaling on this connection? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN << tp->rcv_scale); tp->snd_una++; /* SYN is acked */ /* * If there's data, delay ACK; if there's also a FIN * ACKNOW will be turned on later. */ if (DELAY_ACK(tp, tlen) && tlen != 0) tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); else tp->t_flags |= TF_ACKNOW; if ((thflags & TH_ECE) && V_tcp_do_ecn) { tp->t_flags |= TF_ECN_PERMIT; TCPSTAT_INC(tcps_ecn_shs); } /* * Received in SYN_SENT[*] state. * Transitions: * SYN_SENT --> ESTABLISHED * SYN_SENT* --> FIN_WAIT_1 */ tp->t_starttime = ticks; if (tp->t_flags & TF_NEEDFIN) { tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; thflags &= ~TH_SYN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(connect__established, NULL, tp, mtod(m, const char *), tp, th); cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } } else { /* * Received initial SYN in SYN-SENT[*] state => * simultaneous open. * If it succeeds, connection is * half-synchronized. * Otherwise, do 3-way handshake: * SYN-SENT -> SYN-RECEIVED * SYN-SENT* -> SYN-RECEIVED* */ tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); tcp_timer_activate(tp, TT_REXMT, 0); tcp_state_change(tp, TCPS_SYN_RECEIVED); } KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " "ti_locked %d", __func__, ti_locked)); INP_INFO_RLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tp->t_inpcb); /* * Advance th->th_seq to correspond to first data byte. * If data, trim to stay within window, * dropping FIN if necessary. */ th->th_seq++; if (tlen > tp->rcv_wnd) { todrop = tlen - tp->rcv_wnd; m_adj(m, -todrop); tlen = tp->rcv_wnd; thflags &= ~TH_FIN; TCPSTAT_INC(tcps_rcvpackafterwin); TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); } tp->snd_wl1 = th->th_seq - 1; tp->rcv_up = th->th_seq; /* * Client side of transaction: already sent SYN and data. * If the remote host used T/TCP to validate the SYN, * our data will be ACK'd; if so, enter normal data segment * processing in the middle of step 5, ack processing. * Otherwise, goto step 6. */ if (thflags & TH_ACK) goto process_ACK; goto step6; /* * If the state is LAST_ACK or CLOSING or TIME_WAIT: * do normal processing. * * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. */ case TCPS_LAST_ACK: case TCPS_CLOSING: break; /* continue normal processing */ } /* * States other than LISTEN or SYN_SENT. * First check the RST flag and sequence number since reset segments * are exempt from the timestamp and connection count tests. This * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix * below which allowed reset segments in half the sequence space * to fall though and be processed (which gives forged reset * segments with a random sequence number a 50 percent chance of * killing a connection). * Then check timestamp, if present. * Then check the connection count, if present. * Then check that at least some bytes of segment are within * receive window. If segment begins before rcv_nxt, * drop leading data (and SYN); if nothing left, just ack. */ if (thflags & TH_RST) { /* * RFC5961 Section 3.2 * * - RST drops connection only if SEG.SEQ == RCV.NXT. * - If RST is in window, we send challenge ACK. * * Note: to take into account delayed ACKs, we should * test against last_ack_sent instead of rcv_nxt. * Note 2: we handle special case of closed window, not * covered by the RFC. */ if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); KASSERT(ti_locked == TI_RLOCKED, ("%s: TH_RST ti_locked %d, th %p tp %p", __func__, ti_locked, th, tp)); KASSERT(tp->t_state != TCPS_SYN_SENT, ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", __func__, th, tp)); if (V_tcp_insecure_rst || tp->last_ack_sent == th->th_seq) { TCPSTAT_INC(tcps_drops); /* Drop the connection. */ switch (tp->t_state) { case TCPS_SYN_RECEIVED: so->so_error = ECONNREFUSED; goto close; case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: so->so_error = ECONNRESET; close: tcp_state_change(tp, TCPS_CLOSED); /* FALLTHROUGH */ default: tp = tcp_close(tp); } } else { TCPSTAT_INC(tcps_badrst); /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } } goto drop; } /* * RFC5961 Section 4.2 * Send challenge ACK for any SYN in synchronized state. */ if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && tp->t_state != TCPS_SYN_RECEIVED) { KASSERT(ti_locked == TI_RLOCKED, ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); INP_INFO_RLOCK_ASSERT(&V_tcbinfo); TCPSTAT_INC(tcps_badsyn); if (V_tcp_insecure_syn && SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { tp = tcp_drop(tp, ECONNRESET); rstreason = BANDLIM_UNLIMITED; } else { /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } goto drop; } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment * and it's less than ts_recent, drop it. */ if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { /* Check to see if ts_recent is over 24 days old. */ if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { /* * Invalidate ts_recent. If this segment updates * ts_recent, the age will be reset later and ts_recent * will get a valid value. If it does not, setting * ts_recent to zero will at least satisfy the * requirement that zero be placed in the timestamp * echo reply when ts_recent isn't valid. The * age isn't reset until we get a valid ts_recent * because we don't want out-of-order segments to be * dropped when ts_recent is old. */ tp->ts_recent = 0; } else { TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, tlen); TCPSTAT_INC(tcps_pawsdrop); if (tlen) goto dropafterack; goto drop; } } /* * In the SYN-RECEIVED state, validate that the packet belongs to * this connection before trimming the data to fit the receive * window. Check the sequence number versus IRS since we know * the sequence numbers haven't wrapped. This is a partial fix * for the "LAND" DoS attack. */ if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } todrop = tp->rcv_nxt - th->th_seq; if (todrop > 0) { if (thflags & TH_SYN) { thflags &= ~TH_SYN; th->th_seq++; if (th->th_urp > 1) th->th_urp--; else thflags &= ~TH_URG; todrop--; } /* * Following if statement from Stevens, vol. 2, p. 960. */ if (todrop > tlen || (todrop == tlen && (thflags & TH_FIN) == 0)) { /* * Any valid FIN must be to the left of the window. * At this point the FIN must be a duplicate or out * of sequence; drop it. */ thflags &= ~TH_FIN; /* * Send an ACK to resynchronize and drop any data. * But keep on processing for RST or ACK. */ tp->t_flags |= TF_ACKNOW; todrop = tlen; TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, todrop); } else { TCPSTAT_INC(tcps_rcvpartduppack); TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); } drop_hdrlen += todrop; /* drop from the top afterwards */ th->th_seq += todrop; tlen -= todrop; if (th->th_urp > todrop) th->th_urp -= todrop; else { thflags &= ~TH_URG; th->th_urp = 0; } } /* * If new data are received on a connection after the * user processes are gone, then RST the other end. */ if ((so->so_state & SS_NOFDREF) && tp->t_state > TCPS_CLOSE_WAIT && tlen) { KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); INP_INFO_RLOCK_ASSERT(&V_tcbinfo); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " "after socket was closed, " "sending RST and removing tcpcb\n", s, __func__, tcpstates[tp->t_state], tlen); free(s, M_TCPLOG); } tp = tcp_close(tp); TCPSTAT_INC(tcps_rcvafterclose); rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } /* * If segment ends after window, drop trailing data * (and PUSH and FIN); if nothing left, just ACK. */ todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); if (todrop > 0) { TCPSTAT_INC(tcps_rcvpackafterwin); if (todrop >= tlen) { TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); /* * If window is closed can only take segments at * window edge, and have to drop data and PUSH from * incoming segments. Continue processing, but * remember to ack. Otherwise, drop segment * and ack. */ if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { tp->t_flags |= TF_ACKNOW; TCPSTAT_INC(tcps_rcvwinprobe); } else goto dropafterack; } else TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); m_adj(m, -todrop); tlen -= todrop; thflags &= ~(TH_PUSH|TH_FIN); } /* * If last ACK falls within this segment's sequence numbers, * record its timestamp. * NOTE: * 1) That the test incorporates suggestions from the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). * 2) That updating only on newer timestamps interferes with * our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. * 3) That we modify the segment boundary check to be * Last.ACK.Sent <= SEG.SEQ + SEG.Len * instead of RFC1323's * Last.ACK.Sent < SEG.SEQ + SEG.Len, * This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated * Vol. 2 p.869. In such cases, we can still calculate the * RTT correctly when RCV.NXT == Last.ACK.Sent. */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN|TH_FIN)) != 0))) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN * flag is on (half-synchronized state), then queue data for * later processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_state == TCPS_SYN_RECEIVED || (tp->t_flags & TF_NEEDSYN)) { #ifdef TCP_RFC7413 if (tp->t_state == TCPS_SYN_RECEIVED && IS_FASTOPEN(tp->t_flags)) { tp->snd_wnd = tiwin; cc_conn_init(tp); } #endif goto step6; } else if (tp->t_flags & TF_ACKNOW) goto dropafterack; else goto drop; } /* * Ack processing. */ switch (tp->t_state) { /* * In SYN_RECEIVED state, the ack ACKs our SYN, so enter * ESTABLISHED state and continue processing. * The ACK was checked above. */ case TCPS_SYN_RECEIVED: TCPSTAT_INC(tcps_connects); soisconnected(so); /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; tp->snd_wnd = tiwin; } /* * Make transitions: * SYN-RECEIVED -> ESTABLISHED * SYN-RECEIVED* -> FIN-WAIT-1 */ tp->t_starttime = ticks; if (tp->t_flags & TF_NEEDFIN) { tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(accept__established, NULL, tp, mtod(m, const char *), tp, th); #ifdef TCP_RFC7413 if (tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; /* * Account for the ACK of our SYN prior to * regular ACK processing below. */ tp->snd_una++; } /* * TFO connections call cc_conn_init() during SYN * processing. Calling it again here for such * connections is not harmless as it would undo the * snd_cwnd reduction that occurs when a TFO SYN|ACK * is retransmitted. */ if (!IS_FASTOPEN(tp->t_flags)) #endif cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } /* * If segment contains data or ACK, will call tcp_reass() * later; if not, do so now to pass queued data to user. */ if (tlen == 0 && (thflags & TH_FIN) == 0) (void) tcp_reass(tp, (struct tcphdr *)0, 0, (struct mbuf *)0); tp->snd_wl1 = th->th_seq - 1; /* FALLTHROUGH */ /* * In ESTABLISHED state: drop duplicate ACKs; ACK out of range * ACKs. If the ack is in the range * tp->snd_una < th->th_ack <= tp->snd_max * then advance tp->snd_una to th->th_ack and drop * data from the retransmission queue. If this ACK reflects * more up to date window information we update our window information. */ case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: case TCPS_CLOSING: case TCPS_LAST_ACK: if (SEQ_GT(th->th_ack, tp->snd_max)) { TCPSTAT_INC(tcps_rcvacktoomuch); goto dropafterack; } if ((tp->t_flags & TF_SACK_PERMIT) && ((to.to_flags & TOF_SACK) || !TAILQ_EMPTY(&tp->snd_holes))) sack_changed = tcp_sack_doack(tp, &to, th->th_ack); else /* * Reset the value so that previous (valid) value * from the last ack with SACK doesn't get used. */ tp->sackhint.sacked_bytes = 0; #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif if (SEQ_LEQ(th->th_ack, tp->snd_una)) { u_int maxseg; maxseg = tcp_maxseg(tp); if (tlen == 0 && (tiwin == tp->snd_wnd || (tp->t_flags & TF_SACK_PERMIT))) { /* * If this is the first time we've seen a * FIN from the remote, this is not a * duplicate and it needs to be processed * normally. This happens during a * simultaneous close. */ if ((thflags & TH_FIN) && (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { tp->t_dupacks = 0; break; } TCPSTAT_INC(tcps_rcvdupack); /* * If we have outstanding data (other than * a window probe), this is a completely * duplicate ack (ie, window info didn't * change and FIN isn't set), * the ack is the biggest we've * seen and we've seen exactly our rexmt * threshold of them, assume a packet * has been dropped and retransmit it. * Kludge snd_nxt & the congestion * window so we send only this one * packet. * * We know we're losing at the current * window size so do congestion avoidance * (set ssthresh to half the current window * and pull our congestion window back to * the new ssthresh). * * Dup acks mean that packets have left the * network (they're now cached at the receiver) * so bump cwnd by the amount in the receiver * to keep a constant cwnd packets in the * network. * * When using TCP ECN, notify the peer that * we reduced the cwnd. */ /* * Following 2 kinds of acks should not affect * dupack counting: * 1) Old acks * 2) Acks with SACK but without any new SACK * information in them. These could result from * any anomaly in the network like a switch * duplicating packets or a possible DoS attack. */ if (th->th_ack != tp->snd_una || ((tp->t_flags & TF_SACK_PERMIT) && !sack_changed)) break; else if (!tcp_timer_active(tp, TT_REXMT)) tp->t_dupacks = 0; else if (++tp->t_dupacks > tcprexmtthresh || IN_FASTRECOVERY(tp->t_flags)) { cc_ack_received(tp, th, nsegs, CC_DUPACK); if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags)) { int awnd; /* * Compute the amount of data in flight first. * We can inject new data into the pipe iff * we have less than 1/2 the original window's * worth of data in flight. */ if (V_tcp_do_rfc6675_pipe) awnd = tcp_compute_pipe(tp); else awnd = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; if (awnd < tp->snd_ssthresh) { tp->snd_cwnd += maxseg; - /* - * RFC5681 Section 3.2 talks about cwnd - * inflation on additional dupacks and - * deflation on recovering from loss. - * - * We keep cwnd into check so that - * we don't have to 'deflate' it when we - * get out of recovery. - */ if (tp->snd_cwnd > tp->snd_ssthresh) tp->snd_cwnd = tp->snd_ssthresh; } } else tp->snd_cwnd += maxseg; (void) tp->t_fb->tfb_tcp_output(tp); goto drop; } else if (tp->t_dupacks == tcprexmtthresh) { tcp_seq onxt = tp->snd_nxt; /* * If we're doing sack, check to * see if we're already in sack * recovery. If we're not doing sack, * check to see if we're in newreno * recovery. */ if (tp->t_flags & TF_SACK_PERMIT) { if (IN_FASTRECOVERY(tp->t_flags)) { tp->t_dupacks = 0; break; } } else { if (SEQ_LEQ(th->th_ack, tp->snd_recover)) { tp->t_dupacks = 0; break; } } /* Congestion signal before ack. */ cc_cong_signal(tp, th, CC_NDUPACK); cc_ack_received(tp, th, nsegs, CC_DUPACK); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; if (tp->t_flags & TF_SACK_PERMIT) { TCPSTAT_INC( tcps_sack_recovery_episode); tp->sack_newdata = tp->snd_nxt; - if (CC_ALGO(tp)->cong_signal == NULL) - tp->snd_cwnd = maxseg; + tp->snd_cwnd = maxseg; (void) tp->t_fb->tfb_tcp_output(tp); goto drop; } tp->snd_nxt = th->th_ack; - if (CC_ALGO(tp)->cong_signal == NULL) - tp->snd_cwnd = maxseg; + tp->snd_cwnd = maxseg; (void) tp->t_fb->tfb_tcp_output(tp); KASSERT(tp->snd_limited <= 2, ("%s: tp->snd_limited too big", __func__)); - if (CC_ALGO(tp)->cong_signal == NULL) - tp->snd_cwnd = tp->snd_ssthresh + - maxseg * - (tp->t_dupacks - tp->snd_limited); + tp->snd_cwnd = tp->snd_ssthresh + + maxseg * + (tp->t_dupacks - tp->snd_limited); if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; goto drop; } else if (V_tcp_do_rfc3042) { /* * Process first and second duplicate * ACKs. Each indicates a segment * leaving the network, creating room * for more. Make sure we can send a * packet on reception of each duplicate * ACK by increasing snd_cwnd by one * segment. Restore the original * snd_cwnd after packet transmission. */ cc_ack_received(tp, th, nsegs, CC_DUPACK); uint32_t oldcwnd = tp->snd_cwnd; tcp_seq oldsndmax = tp->snd_max; u_int sent; int avail; KASSERT(tp->t_dupacks == 1 || tp->t_dupacks == 2, ("%s: dupacks not 1 or 2", __func__)); if (tp->t_dupacks == 1) tp->snd_limited = 0; tp->snd_cwnd = (tp->snd_nxt - tp->snd_una) + (tp->t_dupacks - tp->snd_limited) * maxseg; /* * Only call tcp_output when there * is new data available to be sent. * Otherwise we would send pure ACKs. */ SOCKBUF_LOCK(&so->so_snd); avail = sbavail(&so->so_snd) - (tp->snd_nxt - tp->snd_una); SOCKBUF_UNLOCK(&so->so_snd); if (avail > 0) (void) tp->t_fb->tfb_tcp_output(tp); sent = tp->snd_max - oldsndmax; if (sent > maxseg) { KASSERT((tp->t_dupacks == 2 && tp->snd_limited == 0) || (sent == maxseg + 1 && tp->t_flags & TF_SENTFIN), ("%s: sent too much", __func__)); tp->snd_limited = 2; } else if (sent > 0) ++tp->snd_limited; tp->snd_cwnd = oldcwnd; goto drop; } } break; } else { /* * This ack is advancing the left edge, reset the * counter. */ tp->t_dupacks = 0; /* * If this ack also has new SACK info, increment the * counter as per rfc6675. */ if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) tp->t_dupacks++; } KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("%s: th_ack <= snd_una", __func__)); /* * If the congestion window was inflated to account * for the other side's cached packets, retract it. */ if (IN_FASTRECOVERY(tp->t_flags)) { if (SEQ_LT(th->th_ack, tp->snd_recover)) { if (tp->t_flags & TF_SACK_PERMIT) tcp_sack_partialack(tp, th); else tcp_newreno_partial_ack(tp, th); } else cc_post_recovery(tp, th); } /* * If we reach this point, ACK is not a duplicate, * i.e., it ACKs something we sent. */ if (tp->t_flags & TF_NEEDSYN) { /* * T/TCP: Connection was half-synchronized, and our * SYN has been ACK'd (so connection is now fully * synchronized). Go to non-starred state, * increment snd_una for ACK of SYN, and check if * we can do window scaling. */ tp->t_flags &= ~TF_NEEDSYN; tp->snd_una++; /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; /* Send window already scaled. */ } } process_ACK: INP_WLOCK_ASSERT(tp->t_inpcb); acked = BYTES_THIS_ACK(tp, th); KASSERT(acked >= 0, ("%s: acked unexepectedly negative " "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, tp->snd_una, th->th_ack, tp, m)); TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); /* * If we just performed our first retransmit, and the ACK * arrives within our recovery window, then it was a mistake * to do the retransmit in the first place. Recover our * original cwnd and ssthresh, and proceed to transmit where * we left off. */ if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && (int)(ticks - tp->t_badrxtwin) < 0) cc_cong_signal(tp, th, CC_RTO_ERR); /* * If we have a timestamp reply, update smoothed * round trip time. If no timestamp is present but * transmit timer is running and timed sequence * number was acked, update smoothed round trip time. * Since we now have an rtt measurement, cancel the * timer backoff (cf., Phil Karn's retransmit alg.). * Recompute the initial retransmit timer. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } /* * If all outstanding data is acked, stop retransmit * timer and remember to restart (more output or persist). * If there is more data to be acked, restart retransmit * timer, using current (possibly backed-off) value. */ if (th->th_ack == tp->snd_max) { tcp_timer_activate(tp, TT_REXMT, 0); needoutput = 1; } else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); /* * If no data (only SYN) was ACK'd, * skip rest of ACK processing. */ if (acked == 0) goto step6; /* * Let the congestion control algorithm update congestion * control related information. This typically means increasing * the congestion window. */ cc_ack_received(tp, th, nsegs, CC_ACK); SOCKBUF_LOCK(&so->so_snd); if (acked > sbavail(&so->so_snd)) { if (tp->snd_wnd >= sbavail(&so->so_snd)) tp->snd_wnd -= sbavail(&so->so_snd); else tp->snd_wnd = 0; mfree = sbcut_locked(&so->so_snd, (int)sbavail(&so->so_snd)); ourfinisacked = 1; } else { mfree = sbcut_locked(&so->so_snd, acked); if (tp->snd_wnd >= (uint32_t) acked) tp->snd_wnd -= acked; else tp->snd_wnd = 0; ourfinisacked = 0; } /* NB: sowwakeup_locked() does an implicit unlock. */ sowwakeup_locked(so); m_freem(mfree); /* Detect una wraparound. */ if (!IN_RECOVERY(tp->t_flags) && SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* XXXLAS: Can this be moved up into cc_post_recovery? */ if (IN_RECOVERY(tp->t_flags) && SEQ_GEQ(th->th_ack, tp->snd_recover)) { EXIT_RECOVERY(tp->t_flags); } tp->snd_una = th->th_ack; if (tp->t_flags & TF_SACK_PERMIT) { if (SEQ_GT(tp->snd_una, tp->snd_recover)) tp->snd_recover = tp->snd_una; } if (SEQ_LT(tp->snd_nxt, tp->snd_una)) tp->snd_nxt = tp->snd_una; switch (tp->t_state) { /* * In FIN_WAIT_1 STATE in addition to the processing * for the ESTABLISHED state if our FIN is now acknowledged * then enter FIN_WAIT_2. */ case TCPS_FIN_WAIT_1: if (ourfinisacked) { /* * If we can't receive any more * data, then closing user can proceed. * Starting the timer is contrary to the * specification, but if we don't get a FIN * we'll hang forever. * * XXXjl: * we should release the tp also, and use a * compressed state. */ if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { soisdisconnected(so); tcp_timer_activate(tp, TT_2MSL, (tcp_fast_finwait2_recycle ? tcp_finwait2_timeout : TP_MAXIDLE(tp))); } tcp_state_change(tp, TCPS_FIN_WAIT_2); } break; /* * In CLOSING STATE in addition to the processing for * the ESTABLISHED state if the ACK acknowledges our FIN * then enter the TIME-WAIT state, otherwise ignore * the segment. */ case TCPS_CLOSING: if (ourfinisacked) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); tcp_twstart(tp); INP_INFO_RUNLOCK(&V_tcbinfo); m_freem(m); return; } break; /* * In LAST_ACK, we may still be waiting for data to drain * and/or to be acked, as well as for the ack of our FIN. * If our FIN is now acknowledged, delete the TCB, * enter the closed state and return. */ case TCPS_LAST_ACK: if (ourfinisacked) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); tp = tcp_close(tp); goto drop; } break; } } step6: INP_WLOCK_ASSERT(tp->t_inpcb); /* * Update window information. * Don't look at window if no ACK: TAC's send garbage on first SYN. */ if ((thflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { /* keep track of pure window updates */ if (tlen == 0 && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) TCPSTAT_INC(tcps_rcvwinupd); tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; tp->snd_wl2 = th->th_ack; if (tp->snd_wnd > tp->max_sndwnd) tp->max_sndwnd = tp->snd_wnd; needoutput = 1; } /* * Process segments with URG. */ if ((thflags & TH_URG) && th->th_urp && TCPS_HAVERCVDFIN(tp->t_state) == 0) { /* * This is a kludge, but if we receive and accept * random urgent pointers, we'll crash in * soreceive. It's hard to imagine someone * actually wanting to send this much urgent data. */ SOCKBUF_LOCK(&so->so_rcv); if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { th->th_urp = 0; /* XXX */ thflags &= ~TH_URG; /* XXX */ SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ goto dodata; /* XXX */ } /* * If this segment advances the known urgent pointer, * then mark the data stream. This should not happen * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since * a FIN has been received from the remote side. * In these states we ignore the URG. * * According to RFC961 (Assigned Protocols), * the urgent pointer points to the last octet * of urgent data. We continue, however, * to consider it to indicate the first octet * of data past the urgent section as the original * spec states (in one of two places). */ if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { tp->rcv_up = th->th_seq + th->th_urp; so->so_oobmark = sbavail(&so->so_rcv) + (tp->rcv_up - tp->rcv_nxt) - 1; if (so->so_oobmark == 0) so->so_rcv.sb_state |= SBS_RCVATMARK; sohasoutofband(so); tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); } SOCKBUF_UNLOCK(&so->so_rcv); /* * Remove out of band data so doesn't get presented to user. * This can happen independent of advancing the URG pointer, * but if two URG's are pending at once, some out-of-band * data may creep in... ick. */ if (th->th_urp <= (uint32_t)tlen && !(so->so_options & SO_OOBINLINE)) { /* hdr drop is delayed */ tcp_pulloutofband(so, th, m, drop_hdrlen); } } else { /* * If no out of band data is expected, * pull receive urgent pointer along * with the receive window. */ if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) tp->rcv_up = tp->rcv_nxt; } dodata: /* XXX */ INP_WLOCK_ASSERT(tp->t_inpcb); /* * Process the segment text, merging it into the TCP sequencing queue, * and arranging for acknowledgment of receipt if necessary. * This process logically involves adjusting tp->rcv_wnd as data * is presented to the user (this happens in tcp_usrreq.c, * case PRU_RCVD). If a FIN has already been received on this * connection then we just ignore the text. */ #ifdef TCP_RFC7413 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && IS_FASTOPEN(tp->t_flags)); #else #define tfo_syn (false) #endif if ((tlen || (thflags & TH_FIN) || tfo_syn) && TCPS_HAVERCVDFIN(tp->t_state) == 0) { tcp_seq save_start = th->th_seq; m_adj(m, drop_hdrlen); /* delayed header drop */ /* * Insert segment which includes th into TCP reassembly queue * with control block tp. Set thflags to whether reassembly now * includes a segment with FIN. This handles the common case * inline (segment is the next to be received on an established * connection, and the queue is empty), avoiding linkage into * and removal from the queue and repetition of various * conversions. * Set DELACK for segments received in order, but ack * immediately when segments are out of order (so * fast retransmit can work). */ if (th->th_seq == tp->rcv_nxt && LIST_EMPTY(&tp->t_segq) && (TCPS_HAVEESTABLISHED(tp->t_state) || tfo_syn)) { if (DELAY_ACK(tp, tlen) || tfo_syn) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt += tlen; thflags = th->th_flags & TH_FIN; TCPSTAT_INC(tcps_rcvpack); TCPSTAT_ADD(tcps_rcvbyte, tlen); SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) m_freem(m); else sbappendstream_locked(&so->so_rcv, m, 0); /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); } else { /* * XXX: Due to the header drop above "th" is * theoretically invalid by now. Fortunately * m_adj() doesn't actually frees any mbufs * when trimming from the head. */ thflags = tcp_reass(tp, th, &tlen, m); tp->t_flags |= TF_ACKNOW; } if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) tcp_update_sack_list(tp, save_start, save_start + tlen); #if 0 /* * Note the amount of data that peer has sent into * our window, in order to estimate the sender's * buffer size. * XXX: Unused. */ if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); else len = so->so_rcv.sb_hiwat; #endif } else { m_freem(m); thflags &= ~TH_FIN; } /* * If FIN is received ACK the FIN and let the user know * that the connection is closing. */ if (thflags & TH_FIN) { if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { socantrcvmore(so); /* * If connection is half-synchronized * (ie NEEDSYN flag on) then delay ACK, * so it may be piggybacked when SYN is sent. * Otherwise, since we received a FIN then no * more input can be expected, send ACK now. */ if (tp->t_flags & TF_NEEDSYN) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt++; } switch (tp->t_state) { /* * In SYN_RECEIVED and ESTABLISHED STATES * enter the CLOSE_WAIT state. */ case TCPS_SYN_RECEIVED: tp->t_starttime = ticks; /* FALLTHROUGH */ case TCPS_ESTABLISHED: tcp_state_change(tp, TCPS_CLOSE_WAIT); break; /* * If still in FIN_WAIT_1 STATE FIN has not been acked so * enter the CLOSING state. */ case TCPS_FIN_WAIT_1: tcp_state_change(tp, TCPS_CLOSING); break; /* * In FIN_WAIT_2 state enter the TIME_WAIT state, * starting the time-wait timer, turning off the other * standard timers. */ case TCPS_FIN_WAIT_2: INP_INFO_RLOCK_ASSERT(&V_tcbinfo); KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " "TCP_FIN_WAIT_2 ti_locked: %d", __func__, ti_locked)); tcp_twstart(tp); INP_INFO_RUNLOCK(&V_tcbinfo); return; } } if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); /* * Return any desired output. */ if (needoutput || (tp->t_flags & TF_ACKNOW)) (void) tp->t_fb->tfb_tcp_output(tp); check_delack: KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", __func__, ti_locked)); INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tp->t_inpcb); if (tp->t_flags & TF_DELACK) { tp->t_flags &= ~TF_DELACK; tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); } INP_WUNLOCK(tp->t_inpcb); return; dropafterack: /* * Generate an ACK dropping incoming segment if it occupies * sequence space, where the ACK reflects our state. * * We can now skip the test for the RST flag since all * paths to this code happen after packets containing * RST have been dropped. * * In the SYN-RECEIVED state, don't send an ACK unless the * segment we received passes the SYN-RECEIVED ACK test. * If it fails send a RST. This breaks the loop in the * "LAND" DoS attack, and also prevents an ACK storm * between two listening ports that have been sent forged * SYN segments, each with the source address of the other. */ if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && (SEQ_GT(tp->snd_una, th->th_ack) || SEQ_GT(th->th_ack, tp->snd_max)) ) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; tp->t_flags |= TF_ACKNOW; (void) tp->t_fb->tfb_tcp_output(tp); INP_WUNLOCK(tp->t_inpcb); m_freem(m); return; dropwithreset: if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; if (tp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_WUNLOCK(tp->t_inpcb); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); return; drop: if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK(&V_tcbinfo); ti_locked = TI_UNLOCKED; } #ifdef INVARIANTS else INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); #endif /* * Drop space held by incoming segment and return. */ #ifdef TCPDEBUG if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); if (tp != NULL) INP_WUNLOCK(tp->t_inpcb); m_freem(m); #ifndef TCP_RFC7413 #undef tfo_syn #endif } /* * Issue RST and make ACK acceptable to originator of segment. * The mbuf must still include the original packet header. * tp may be NULL. */ void tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int tlen, int rstreason) { #ifdef INET struct ip *ip; #endif #ifdef INET6 struct ip6_hdr *ip6; #endif if (tp != NULL) { INP_WLOCK_ASSERT(tp->t_inpcb); } /* Don't bother if destination was broadcast/multicast. */ if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) goto drop; #ifdef INET6 if (mtod(m, struct ip *)->ip_v == 6) { ip6 = mtod(m, struct ip6_hdr *); if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) goto drop; /* IPv6 anycast check is done at tcp6_input() */ } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip = mtod(m, struct ip *); if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) goto drop; } #endif /* Perform bandwidth limiting. */ if (badport_bandlim(rstreason) < 0) goto drop; /* tcp_respond consumes the mbuf chain. */ if (th->th_flags & TH_ACK) { tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, TH_RST); } else { if (th->th_flags & TH_SYN) tlen++; tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK); } return; drop: m_freem(m); } /* * Parse TCP options and place in tcpopt. */ void tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) { int opt, optlen; to->to_flags = 0; for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[0]; if (opt == TCPOPT_EOL) break; if (opt == TCPOPT_NOP) optlen = 1; else { if (cnt < 2) break; optlen = cp[1]; if (optlen < 2 || optlen > cnt) break; } switch (opt) { case TCPOPT_MAXSEG: if (optlen != TCPOLEN_MAXSEG) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_MSS; bcopy((char *)cp + 2, (char *)&to->to_mss, sizeof(to->to_mss)); to->to_mss = ntohs(to->to_mss); break; case TCPOPT_WINDOW: if (optlen != TCPOLEN_WINDOW) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_SCALE; to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); break; case TCPOPT_TIMESTAMP: if (optlen != TCPOLEN_TIMESTAMP) continue; to->to_flags |= TOF_TS; bcopy((char *)cp + 2, (char *)&to->to_tsval, sizeof(to->to_tsval)); to->to_tsval = ntohl(to->to_tsval); bcopy((char *)cp + 6, (char *)&to->to_tsecr, sizeof(to->to_tsecr)); to->to_tsecr = ntohl(to->to_tsecr); break; #ifdef TCP_SIGNATURE /* * XXX In order to reply to a host which has set the * TCP_SIGNATURE option in its initial SYN, we have to * record the fact that the option was observed here * for the syncache code to perform the correct response. */ case TCPOPT_SIGNATURE: if (optlen != TCPOLEN_SIGNATURE) continue; to->to_flags |= TOF_SIGNATURE; to->to_signature = cp + 2; break; #endif case TCPOPT_SACK_PERMITTED: if (optlen != TCPOLEN_SACK_PERMITTED) continue; if (!(flags & TO_SYN)) continue; if (!V_tcp_do_sack) continue; to->to_flags |= TOF_SACKPERM; break; case TCPOPT_SACK: if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) continue; if (flags & TO_SYN) continue; to->to_flags |= TOF_SACK; to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; to->to_sacks = cp + 2; TCPSTAT_INC(tcps_sack_rcv_blocks); break; #ifdef TCP_RFC7413 case TCPOPT_FAST_OPEN: if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && (optlen < TCPOLEN_FAST_OPEN_MIN) && (optlen > TCPOLEN_FAST_OPEN_MAX)) continue; if (!(flags & TO_SYN)) continue; if (!V_tcp_fastopen_enabled) continue; to->to_flags |= TOF_FASTOPEN; to->to_tfo_len = optlen - 2; to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; break; #endif default: continue; } } } /* * Pull out of band byte out of a segment so * it doesn't appear in the user's data queue. * It is still reflected in the segment length for * sequencing purposes. */ void tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) { int cnt = off + th->th_urp - 1; while (cnt >= 0) { if (m->m_len > cnt) { char *cp = mtod(m, caddr_t) + cnt; struct tcpcb *tp = sototcpcb(so); INP_WLOCK_ASSERT(tp->t_inpcb); tp->t_iobc = *cp; tp->t_oobflags |= TCPOOB_HAVEDATA; bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); m->m_len--; if (m->m_flags & M_PKTHDR) m->m_pkthdr.len--; return; } cnt -= m->m_len; m = m->m_next; if (m == NULL) break; } panic("tcp_pulloutofband"); } /* * Collect new round-trip time estimate * and update averages and current timeout. */ void tcp_xmit_timer(struct tcpcb *tp, int rtt) { int delta; INP_WLOCK_ASSERT(tp->t_inpcb); TCPSTAT_INC(tcps_rttupdated); tp->t_rttupdated++; if (tp->t_srtt != 0) { /* * srtt is stored as fixed point with 5 bits after the * binary point (i.e., scaled by 8). The following magic * is equivalent to the smoothing algorithm in rfc793 with * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed * point). Adjust rtt to origin 0. */ delta = ((rtt - 1) << TCP_DELTA_SHIFT) - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); if ((tp->t_srtt += delta) <= 0) tp->t_srtt = 1; /* * We accumulate a smoothed rtt variance (actually, a * smoothed mean difference), then set the retransmit * timer to smoothed rtt + 4 times the smoothed variance. * rttvar is stored as fixed point with 4 bits after the * binary point (scaled by 16). The following is * equivalent to rfc793 smoothing with an alpha of .75 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces * rfc793's wired-in beta. */ if (delta < 0) delta = -delta; delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); if ((tp->t_rttvar += delta) <= 0) tp->t_rttvar = 1; if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) tp->t_rttbest = tp->t_srtt + tp->t_rttvar; } else { /* * No rtt measurement yet - use the unsmoothed rtt. * Set the variance to half the rtt (so our first * retransmit happens at 3*rtt). */ tp->t_srtt = rtt << TCP_RTT_SHIFT; tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); tp->t_rttbest = tp->t_srtt + tp->t_rttvar; } tp->t_rtttime = 0; tp->t_rxtshift = 0; /* * the retransmit should happen at rtt + 4 * rttvar. * Because of the way we do the smoothing, srtt and rttvar * will each average +1/2 tick of bias. When we compute * the retransmit timer, we want 1/2 tick of rounding and * 1 extra tick because of +-1/2 tick uncertainty in the * firing of the timer. The bias will give us exactly the * 1.5 tick we need. But, because the bias is * statistical, we have to test that we don't drop below * the minimum feasible timer (which is 2 ticks). */ TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); /* * We received an ack for a packet that wasn't retransmitted; * it is probably safe to discard any error indications we've * received recently. This isn't quite right, but close enough * for now (a route might have failed after we sent a segment, * and the return path might not be symmetrical). */ tp->t_softerror = 0; } /* * Determine a reasonable value for maxseg size. * If the route is known, check route for mtu. * If none, use an mss that can be handled on the outgoing interface * without forcing IP to fragment. If no route is found, route has no mtu, * or the destination isn't local, use a default, hopefully conservative * size (usually 512 or the default IP max size, but no more than the mtu * of the interface), as we can't discover anything about intervening * gateways or networks. We also initialize the congestion/slow start * window to be a single segment if the destination isn't local. * While looking at the routing entry, we also initialize other path-dependent * parameters from pre-set or cached values in the routing entry. * * NOTE that resulting t_maxseg doesn't include space for TCP options or * IP options, e.g. IPSEC data, since length of this data may vary, and * thus it is calculated for every segment separately in tcp_output(). * * NOTE that this routine is only called when we process an incoming * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS * settings are handled in tcp_mssopt(). */ void tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) { int mss = 0; uint32_t maxmtu = 0; struct inpcb *inp = tp->t_inpcb; struct hc_metrics_lite metrics; #ifdef INET6 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; size_t min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : sizeof (struct tcpiphdr); #else const size_t min_protoh = sizeof(struct tcpiphdr); #endif INP_WLOCK_ASSERT(tp->t_inpcb); if (mtuoffer != -1) { KASSERT(offer == -1, ("%s: conflict", __func__)); offer = mtuoffer - min_protoh; } /* Initialize. */ #ifdef INET6 if (isipv6) { maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_v6mssdflt; } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { maxmtu = tcp_maxmtu(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_mssdflt; } #endif /* * No route to sender, stay with default mss and return. */ if (maxmtu == 0) { /* * In case we return early we need to initialize metrics * to a defined state as tcp_hc_get() would do for us * if there was no cache hit. */ if (metricptr != NULL) bzero(metricptr, sizeof(struct hc_metrics_lite)); return; } /* What have we got? */ switch (offer) { case 0: /* * Offer == 0 means that there was no MSS on the SYN * segment, in this case we use tcp_mssdflt as * already assigned to t_maxseg above. */ offer = tp->t_maxseg; break; case -1: /* * Offer == -1 means that we didn't receive SYN yet. */ /* FALLTHROUGH */ default: /* * Prevent DoS attack with too small MSS. Round up * to at least minmss. */ offer = max(offer, V_tcp_minmss); } /* * rmx information is now retrieved from tcp_hostcache. */ tcp_hc_get(&inp->inp_inc, &metrics); if (metricptr != NULL) bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); /* * If there's a discovered mtu in tcp hostcache, use it. * Else, use the link mtu. */ if (metrics.rmx_mtu) mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; else { #ifdef INET6 if (isipv6) { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in6_localaddr(&inp->in6p_faddr)) mss = min(mss, V_tcp_v6mssdflt); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in_localaddr(inp->inp_faddr)) mss = min(mss, V_tcp_mssdflt); } #endif /* * XXX - The above conditional (mss = maxmtu - min_protoh) * probably violates the TCP spec. * The problem is that, since we don't know the * other end's MSS, we are supposed to use a conservative * default. But, if we do that, then MTU discovery will * never actually take place, because the conservative * default is much less than the MTUs typically seen * on the Internet today. For the moment, we'll sweep * this under the carpet. * * The conservative default might not actually be a problem * if the only case this occurs is when sending an initial * SYN with options and data to a host we've never talked * to before. Then, they will reply with an MSS value which * will get recorded and the new parameters should get * recomputed. For Further Study. */ } mss = min(mss, offer); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ mss = max(mss, 64); tp->t_maxseg = mss; } void tcp_mss(struct tcpcb *tp, int offer) { int mss; uint32_t bufsize; struct inpcb *inp; struct socket *so; struct hc_metrics_lite metrics; struct tcp_ifcap cap; KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); bzero(&cap, sizeof(cap)); tcp_mss_update(tp, offer, -1, &metrics, &cap); mss = tp->t_maxseg; inp = tp->t_inpcb; /* * If there's a pipesize, change the socket buffer to that size, * don't change if sb_hiwat is different than default (then it * has been changed on purpose with setsockopt). * Make the socket buffers an integral number of mss units; * if the mss is larger than the socket buffer, decrease the mss. */ so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) bufsize = metrics.rmx_sendpipe; else bufsize = so->so_snd.sb_hiwat; if (bufsize < mss) mss = bufsize; else { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_snd.sb_hiwat) (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); } SOCKBUF_UNLOCK(&so->so_snd); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ tp->t_maxseg = max(mss, 64); SOCKBUF_LOCK(&so->so_rcv); if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) bufsize = metrics.rmx_recvpipe; else bufsize = so->so_rcv.sb_hiwat; if (bufsize > mss) { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_rcv.sb_hiwat) (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); } SOCKBUF_UNLOCK(&so->so_rcv); /* Check the interface for TSO capabilities. */ if (cap.ifcap & CSUM_TSO) { tp->t_flags |= TF_TSO; tp->t_tsomax = cap.tsomax; tp->t_tsomaxsegcount = cap.tsomaxsegcount; tp->t_tsomaxsegsize = cap.tsomaxsegsize; } } /* * Determine the MSS option to send on an outgoing SYN. */ int tcp_mssopt(struct in_conninfo *inc) { int mss = 0; uint32_t thcmtu = 0; uint32_t maxmtu = 0; size_t min_protoh; KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); #ifdef INET6 if (inc->inc_flags & INC_ISIPV6) { mss = V_tcp_v6mssdflt; maxmtu = tcp_maxmtu6(inc, NULL); min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = V_tcp_mssdflt; maxmtu = tcp_maxmtu(inc, NULL); min_protoh = sizeof(struct tcpiphdr); } #endif #if defined(INET6) || defined(INET) thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ #endif if (maxmtu && thcmtu) mss = min(maxmtu, thcmtu) - min_protoh; else if (maxmtu || thcmtu) mss = max(maxmtu, thcmtu) - min_protoh; return (mss); } /* * On a partial ack arrives, force the retransmission of the * next unacknowledged segment. Do not clear tp->t_dupacks. * By setting snd_nxt to ti_ack, this forces retransmission timer to * be started again. */ void tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) { tcp_seq onxt = tp->snd_nxt; uint32_t ocwnd = tp->snd_cwnd; u_int maxseg = tcp_maxseg(tp); INP_WLOCK_ASSERT(tp->t_inpcb); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; tp->snd_nxt = th->th_ack; /* * Set snd_cwnd to one segment beyond acknowledged offset. * (tp->snd_una has not yet been updated when this function is called.) */ tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); tp->t_flags |= TF_ACKNOW; (void) tp->t_fb->tfb_tcp_output(tp); tp->snd_cwnd = ocwnd; if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; /* * Partial window deflation. Relies on fact that tp->snd_una * not updated yet. */ if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); else tp->snd_cwnd = 0; tp->snd_cwnd += maxseg; } int tcp_compute_pipe(struct tcpcb *tp) { return (tp->snd_max - tp->snd_una + tp->sackhint.sack_bytes_rexmit - tp->sackhint.sacked_bytes); }