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
39/*
40 * This software was first released in 2007 by James Healy and Lawrence Stewart
41 * whilst working on the NewTCP research project at Swinburne University of
42 * Technology's Centre for Advanced Internet Architectures, Melbourne,
43 * Australia, which was made possible in part by a grant from the Cisco
44 * University Research Program Fund at Community Foundation Silicon Valley.
45 * More details are available at:
46 * http://caia.swin.edu.au/urp/newtcp/
47 */
48
49#ifndef _NETINET_CC_CC_H_
50#define _NETINET_CC_CC_H_
51
52#ifdef _KERNEL
53
54MALLOC_DECLARE(M_CC_MEM);
55
56/* Global CC vars. */
57extern STAILQ_HEAD(cc_head, cc_algo) cc_list;
58extern const int tcprexmtthresh;
59
60/* Per-netstack bits. */
61VNET_DECLARE(struct cc_algo *, default_cc_ptr);
62#define V_default_cc_ptr VNET(default_cc_ptr)
63
64VNET_DECLARE(int, cc_do_abe);
65#define V_cc_do_abe VNET(cc_do_abe)
66
67VNET_DECLARE(int, cc_abe_frlossreduce);
68#define V_cc_abe_frlossreduce VNET(cc_abe_frlossreduce)
69
70/* Define the new net.inet.tcp.cc sysctl tree. */
71#ifdef _SYS_SYSCTL_H_
72SYSCTL_DECL(_net_inet_tcp_cc);
73#endif
74
75/* For CC modules that use hystart++ */
76extern uint32_t hystart_lowcwnd;
77extern uint32_t hystart_minrtt_thresh;
78extern uint32_t hystart_maxrtt_thresh;
79extern uint32_t hystart_n_rttsamples;
80extern uint32_t hystart_css_growth_div;
81extern uint32_t hystart_css_rounds;
82extern uint32_t hystart_bblogs;
83
84/* CC housekeeping functions. */
85int cc_register_algo(struct cc_algo *add_cc);
86int cc_deregister_algo(struct cc_algo *remove_cc);
87#endif /* _KERNEL */
88
89#if defined(_KERNEL) || defined(_WANT_TCPCB)
90struct cc_var {
91 void *cc_data; /* Per-connection private CC algorithm data. */
92 int bytes_this_ack; /* # bytes acked by the current ACK. */
93 tcp_seq curack; /* Most recent ACK. */
94 uint32_t flags; /* Flags for cc_var (see below) */
95 struct tcpcb *tp; /* Pointer to tcpcb */
96 uint16_t nsegs; /* # segments coalesced into current chain. */
97 uint8_t labc; /* Dont use system abc use passed in */
98};
99
100/* cc_var flags. */
101#define CCF_ABC_SENTAWND 0x0001 /* ABC counted cwnd worth of bytes? */
102#define CCF_CWND_LIMITED 0x0002 /* Are we currently cwnd limited? */
103#define CCF_USE_LOCAL_ABC 0x0004 /* Dont use the system l_abc val */
104#define CCF_ACKNOW 0x0008 /* Will this ack be sent now? */
105#define CCF_IPHDR_CE 0x0010 /* Does this packet set CE bit? */
106#define CCF_TCPHDR_CWR 0x0020 /* Does this packet set CWR bit? */
107#define CCF_UNUSED1 0x0040
108#define CCF_UNUSED2 0x0080
109#define CCF_UNUSED3 0x0100
110#define CCF_UNUSED4 0x0200
111#define CCF_HYSTART_ALLOWED 0x0400 /* If the CC supports it Hystart is allowed */
112#define CCF_HYSTART_CAN_SH_CWND 0x0800 /* Can hystart when going CSS -> CA slam the cwnd */
113#define CCF_HYSTART_CONS_SSTH 0x1000 /* Should hystart use the more conservative ssthresh */
114
115#endif /* defined(_KERNEL) || defined(_WANT_TCPCB) */
116typedef enum {
117#if defined(_KERNEL) || defined(_WANT_TCPCB)
118 /* ACK types passed to the ack_received() hook. */
119 CC_ACK = 0x0001, /* Regular in sequence ACK. */
120 CC_DUPACK = 0x0002, /* Duplicate ACK. */
121 CC_PARTIALACK = 0x0004, /* Not yet. */
122 CC_SACK = 0x0008, /* Not yet. */
123#endif /* defined(_KERNEL) || defined(_WANT_TCPCB) */
124 /* Congestion signal types passed to the cong_signal() hook. */
125 CC_ECN = 0x0100, /* ECN marked packet received. */
126 CC_RTO = 0x0200, /* RTO fired. */
127 CC_RTO_ERR = 0x0400, /* RTO fired in error. */
128 CC_NDUPACK = 0x0800, /* Threshold of dupack's reached. */
129 /*
130 * The highest order 8 bits (0x01000000 - 0x80000000) are reserved
131 * for CC algos to declare their own congestion signal types.
132 */
133 CC_SIGPRIVMASK = 0xFF000000 /* Mask to check if sig is private. */
134} ccsignal_t;
135
136#ifdef _KERNEL
137/*
138 * Structure to hold data and function pointers that together represent a
139 * congestion control algorithm.
140 */
141struct cc_algo {
142 char name[TCP_CA_NAME_MAX];
143
144 /* Init global module state on kldload. */
145 int (*mod_init)(void);
146
147 /* Cleanup global module state on kldunload. */
148 int (*mod_destroy)(void);
149
150 /* Return the size of the void pointer the CC needs for state */
151 size_t (*cc_data_sz)(void);
152
153 /*
154 * Init CC state for a new control block. The CC
155 * module may be passed a NULL ptr indicating that
156 * it must allocate the memory. If it is passed a
157 * non-null pointer it is pre-allocated memory by
158 * the caller and the cb_init is expected to use that memory.
159 * It is not expected to fail if memory is passed in and
160 * all currently defined modules do not.
161 */
162 int (*cb_init)(struct cc_var *ccv, void *ptr);
163
164 /* Cleanup CC state for a terminating control block. */
165 void (*cb_destroy)(struct cc_var *ccv);
166
167 /* Init variables for a newly established connection. */
168 void (*conn_init)(struct cc_var *ccv);
169
170 /* Called on receipt of an ack. */
171 void (*ack_received)(struct cc_var *ccv, ccsignal_t type);
172
173 /* Called on detection of a congestion signal. */
174 void (*cong_signal)(struct cc_var *ccv, ccsignal_t type);
175
176 /* Called after exiting congestion recovery. */
177 void (*post_recovery)(struct cc_var *ccv);
178
179 /* Called when data transfer resumes after an idle period. */
180 void (*after_idle)(struct cc_var *ccv);
181
182 /* Called for an additional ECN processing apart from RFC3168. */
183 void (*ecnpkt_handler)(struct cc_var *ccv);
184
185 /* Called when a new "round" begins, if the transport is tracking rounds. */
186 void (*newround)(struct cc_var *ccv, uint32_t round_cnt);
187
188 /*
189 * Called when a RTT sample is made (fas = flight at send, if you dont have it
190 * send the cwnd in).
191 */
192 void (*rttsample)(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32_t fas);
193
194 /* Called for {get|set}sockopt() on a TCP socket with TCP_CCALGOOPT. */
195 int (*ctl_output)(struct cc_var *, struct sockopt *, void *);
196
197 STAILQ_ENTRY (cc_algo) entries;
198 u_int cc_refcount;
199 uint8_t flags;
200};
201
202#define CC_MODULE_BEING_REMOVED 0x01 /* The module is being removed */
203
204/* Macro to obtain the CC algo's struct ptr. */
205#define CC_ALGO(tp) ((tp)->t_cc)
206
207/* Macro to obtain the CC algo's data ptr. */
208#define CC_DATA(tp) ((tp)->t_ccv.cc_data)
209
210/* Macro to obtain the system default CC algo's struct ptr. */
211#define CC_DEFAULT_ALGO() V_default_cc_ptr
212
213extern struct rwlock cc_list_lock;
214#define CC_LIST_LOCK_INIT() rw_init(&cc_list_lock, "cc_list")
215#define CC_LIST_LOCK_DESTROY() rw_destroy(&cc_list_lock)
216#define CC_LIST_RLOCK() rw_rlock(&cc_list_lock)
217#define CC_LIST_RUNLOCK() rw_runlock(&cc_list_lock)
218#define CC_LIST_WLOCK() rw_wlock(&cc_list_lock)
219#define CC_LIST_WUNLOCK() rw_wunlock(&cc_list_lock)
220#define CC_LIST_LOCK_ASSERT() rw_assert(&cc_list_lock, RA_LOCKED)
221
222#define CC_ALGOOPT_LIMIT 2048
223
224/*
225 * These routines give NewReno behavior to the caller
226 * they require no state and can be used by any other CC
227 * module that wishes to use NewReno type behaviour (along
228 * with anything else they may add on, pre or post call).
229 */
230void newreno_cc_post_recovery(struct cc_var *);
231void newreno_cc_after_idle(struct cc_var *);
232void newreno_cc_cong_signal(struct cc_var *, ccsignal_t);
233void newreno_cc_ack_received(struct cc_var *, ccsignal_t);
234u_int newreno_cc_cwnd_on_multiplicative_decrease(struct cc_var *ccv, uint32_t mss);
235u_int newreno_cc_cwnd_in_cong_avoid(struct cc_var *ccv);
236u_int newreno_cc_cwnd_in_slow_start(struct cc_var *ccv);
237
238/* Called to temporarily keep an algo from going away during change */
239void cc_refer(struct cc_algo *algo);
240/* Called to release the temporary hold */
241void cc_release(struct cc_algo *algo);
242
243/* Called to attach a CC algorithm to a tcpcb */
244void cc_attach(struct tcpcb *, struct cc_algo *);
245/* Called to detach a CC algorithm from a tcpcb */
246void cc_detach(struct tcpcb *);
247
248#endif /* _KERNEL */
249#endif /* _NETINET_CC_CC_H_ */