1 /*- 2 * Copyright (c) 2007-2008 3 * Swinburne University of Technology, Melbourne, Australia. 4 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 5 * Copyright (c) 2010 The FreeBSD Foundation 6 * All rights reserved. 7 * 8 * This software was developed at the Centre for Advanced Internet 9 * Architectures, Swinburne University of Technology, by Lawrence Stewart and 10 * James Healy, made possible in part by a grant from the Cisco University 11 * Research Program Fund at Community Foundation Silicon Valley. 12 * 13 * Portions of this software were developed at the Centre for Advanced 14 * Internet Architectures, Swinburne University of Technology, Melbourne, 15 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 16 * 17 * Redistribution and use in source and binary forms, with or without 18 * modification, are permitted provided that the following conditions 19 * are met: 20 * 1. Redistributions of source code must retain the above copyright 21 * notice, this list of conditions and the following disclaimer. 22 * 2. Redistributions in binary form must reproduce the above copyright 23 * notice, this list of conditions and the following disclaimer in the 24 * documentation and/or other materials provided with the distribution. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * $FreeBSD$ 39 */ 40 41 /* 42 * This software was first released in 2007 by James Healy and Lawrence Stewart 43 * whilst working on the NewTCP research project at Swinburne University of 44 * Technology's Centre for Advanced Internet Architectures, Melbourne, 45 * Australia, which was made possible in part by a grant from the Cisco 46 * University Research Program Fund at Community Foundation Silicon Valley. 47 * More details are available at: 48 * http://caia.swin.edu.au/urp/newtcp/ 49 */ 50 51 #ifndef _NETINET_CC_CC_H_ 52 #define _NETINET_CC_CC_H_ 53 54 #ifdef _KERNEL 55 56 MALLOC_DECLARE(M_CC_MEM); 57 58 /* Global CC vars. */ 59 extern STAILQ_HEAD(cc_head, cc_algo) cc_list; 60 extern const int tcprexmtthresh; 61 62 /* Per-netstack bits. */ 63 VNET_DECLARE(struct cc_algo *, default_cc_ptr); 64 #define V_default_cc_ptr VNET(default_cc_ptr) 65 66 VNET_DECLARE(int, cc_do_abe); 67 #define V_cc_do_abe VNET(cc_do_abe) 68 69 VNET_DECLARE(int, cc_abe_frlossreduce); 70 #define V_cc_abe_frlossreduce VNET(cc_abe_frlossreduce) 71 72 /* Define the new net.inet.tcp.cc sysctl tree. */ 73 SYSCTL_DECL(_net_inet_tcp_cc); 74 75 /* For CC modules that use hystart++ */ 76 extern uint32_t hystart_lowcwnd; 77 extern uint32_t hystart_minrtt_thresh; 78 extern uint32_t hystart_maxrtt_thresh; 79 extern uint32_t hystart_n_rttsamples; 80 extern uint32_t hystart_css_growth_div; 81 extern uint32_t hystart_css_rounds; 82 extern uint32_t hystart_bblogs; 83 84 /* CC housekeeping functions. */ 85 int cc_register_algo(struct cc_algo *add_cc); 86 int cc_deregister_algo(struct cc_algo *remove_cc); 87 88 /* 89 * Wrapper around transport structs that contain same-named congestion 90 * control variables. Allows algos to be shared amongst multiple CC aware 91 * transprots. 92 */ 93 struct cc_var { 94 void *cc_data; /* Per-connection private CC algorithm data. */ 95 int bytes_this_ack; /* # bytes acked by the current ACK. */ 96 tcp_seq curack; /* Most recent ACK. */ 97 uint32_t flags; /* Flags for cc_var (see below) */ 98 int type; /* Indicates which ptr is valid in ccvc. */ 99 union ccv_container { 100 struct tcpcb *tcp; 101 struct sctp_nets *sctp; 102 } ccvc; 103 uint16_t nsegs; /* # segments coalesced into current chain. */ 104 uint8_t labc; /* Dont use system abc use passed in */ 105 }; 106 107 /* cc_var flags. */ 108 #define CCF_ABC_SENTAWND 0x0001 /* ABC counted cwnd worth of bytes? */ 109 #define CCF_CWND_LIMITED 0x0002 /* Are we currently cwnd limited? */ 110 #define CCF_USE_LOCAL_ABC 0x0004 /* Dont use the system l_abc val */ 111 #define CCF_ACKNOW 0x0008 /* Will this ack be sent now? */ 112 #define CCF_IPHDR_CE 0x0010 /* Does this packet set CE bit? */ 113 #define CCF_TCPHDR_CWR 0x0020 /* Does this packet set CWR bit? */ 114 #define CCF_MAX_CWND 0x0040 /* Have we reached maximum cwnd? */ 115 #define CCF_CHG_MAX_CWND 0x0080 /* Cubic max_cwnd changed, for K */ 116 #define CCF_USR_IWND 0x0100 /* User specified initial window */ 117 #define CCF_USR_IWND_INIT_NSEG 0x0200 /* Convert segs to bytes on conn init */ 118 #define CCF_HYSTART_ALLOWED 0x0400 /* If the CC supports it Hystart is allowed */ 119 #define CCF_HYSTART_CAN_SH_CWND 0x0800 /* Can hystart when going CSS -> CA slam the cwnd */ 120 #define CCF_HYSTART_CONS_SSTH 0x1000 /* Should hystart use the more conservative ssthresh */ 121 122 /* ACK types passed to the ack_received() hook. */ 123 #define CC_ACK 0x0001 /* Regular in sequence ACK. */ 124 #define CC_DUPACK 0x0002 /* Duplicate ACK. */ 125 #define CC_PARTIALACK 0x0004 /* Not yet. */ 126 #define CC_SACK 0x0008 /* Not yet. */ 127 #endif /* _KERNEL */ 128 129 /* 130 * Congestion signal types passed to the cong_signal() hook. The highest order 8 131 * bits (0x01000000 - 0x80000000) are reserved for CC algos to declare their own 132 * congestion signal types. 133 */ 134 #define CC_ECN 0x00000001 /* ECN marked packet received. */ 135 #define CC_RTO 0x00000002 /* RTO fired. */ 136 #define CC_RTO_ERR 0x00000004 /* RTO fired in error. */ 137 #define CC_NDUPACK 0x00000008 /* Threshold of dupack's reached. */ 138 139 #define CC_SIGPRIVMASK 0xFF000000 /* Mask to check if sig is private. */ 140 141 #ifdef _KERNEL 142 /* 143 * Structure to hold data and function pointers that together represent a 144 * congestion control algorithm. 145 */ 146 struct cc_algo { 147 char name[TCP_CA_NAME_MAX]; 148 149 /* Init global module state on kldload. */ 150 int (*mod_init)(void); 151 152 /* Cleanup global module state on kldunload. */ 153 int (*mod_destroy)(void); 154 155 /* Return the size of the void pointer the CC needs for state */ 156 size_t (*cc_data_sz)(void); 157 158 /* 159 * Init CC state for a new control block. The CC 160 * module may be passed a NULL ptr indicating that 161 * it must allocate the memory. If it is passed a 162 * non-null pointer it is pre-allocated memory by 163 * the caller and the cb_init is expected to use that memory. 164 * It is not expected to fail if memory is passed in and 165 * all currently defined modules do not. 166 */ 167 int (*cb_init)(struct cc_var *ccv, void *ptr); 168 169 /* Cleanup CC state for a terminating control block. */ 170 void (*cb_destroy)(struct cc_var *ccv); 171 172 /* Init variables for a newly established connection. */ 173 void (*conn_init)(struct cc_var *ccv); 174 175 /* Called on receipt of an ack. */ 176 void (*ack_received)(struct cc_var *ccv, uint16_t type); 177 178 /* Called on detection of a congestion signal. */ 179 void (*cong_signal)(struct cc_var *ccv, uint32_t type); 180 181 /* Called after exiting congestion recovery. */ 182 void (*post_recovery)(struct cc_var *ccv); 183 184 /* Called when data transfer resumes after an idle period. */ 185 void (*after_idle)(struct cc_var *ccv); 186 187 /* Called for an additional ECN processing apart from RFC3168. */ 188 void (*ecnpkt_handler)(struct cc_var *ccv); 189 190 /* Called when a new "round" begins, if the transport is tracking rounds. */ 191 void (*newround)(struct cc_var *ccv, uint32_t round_cnt); 192 193 /* 194 * Called when a RTT sample is made (fas = flight at send, if you dont have it 195 * send the cwnd in). 196 */ 197 void (*rttsample)(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32_t fas); 198 199 /* Called for {get|set}sockopt() on a TCP socket with TCP_CCALGOOPT. */ 200 int (*ctl_output)(struct cc_var *, struct sockopt *, void *); 201 202 STAILQ_ENTRY (cc_algo) entries; 203 u_int cc_refcount; 204 uint8_t flags; 205 }; 206 207 #define CC_MODULE_BEING_REMOVED 0x01 /* The module is being removed */ 208 209 /* Macro to obtain the CC algo's struct ptr. */ 210 #define CC_ALGO(tp) ((tp)->cc_algo) 211 212 /* Macro to obtain the CC algo's data ptr. */ 213 #define CC_DATA(tp) ((tp)->ccv->cc_data) 214 215 /* Macro to obtain the system default CC algo's struct ptr. */ 216 #define CC_DEFAULT_ALGO() V_default_cc_ptr 217 218 extern struct rwlock cc_list_lock; 219 #define CC_LIST_LOCK_INIT() rw_init(&cc_list_lock, "cc_list") 220 #define CC_LIST_LOCK_DESTROY() rw_destroy(&cc_list_lock) 221 #define CC_LIST_RLOCK() rw_rlock(&cc_list_lock) 222 #define CC_LIST_RUNLOCK() rw_runlock(&cc_list_lock) 223 #define CC_LIST_WLOCK() rw_wlock(&cc_list_lock) 224 #define CC_LIST_WUNLOCK() rw_wunlock(&cc_list_lock) 225 #define CC_LIST_LOCK_ASSERT() rw_assert(&cc_list_lock, RA_LOCKED) 226 227 #define CC_ALGOOPT_LIMIT 2048 228 229 /* 230 * These routines give NewReno behavior to the caller 231 * they require no state and can be used by any other CC 232 * module that wishes to use NewReno type behaviour (along 233 * with anything else they may add on, pre or post call). 234 */ 235 void newreno_cc_post_recovery(struct cc_var *); 236 void newreno_cc_after_idle(struct cc_var *); 237 void newreno_cc_cong_signal(struct cc_var *, uint32_t ); 238 void newreno_cc_ack_received(struct cc_var *, uint16_t); 239 240 /* Called to temporarily keep an algo from going away during change */ 241 void cc_refer(struct cc_algo *algo); 242 /* Called to release the temporary hold */ 243 void cc_release(struct cc_algo *algo); 244 245 /* Called to attach a CC algorithm to a tcpcb */ 246 void cc_attach(struct tcpcb *, struct cc_algo *); 247 /* Called to detach a CC algorithm from a tcpcb */ 248 void cc_detach(struct tcpcb *); 249 250 #endif /* _KERNEL */ 251 #endif /* _NETINET_CC_CC_H_ */ 252