master
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)
90/*
91 * Wrapper around transport structs that contain same-named congestion
92 * control variables. Allows algos to be shared amongst multiple CC aware
93 * transprots.
94 */
95struct cc_var {
96 void *cc_data; /* Per-connection private CC algorithm data. */
97 int bytes_this_ack; /* # bytes acked by the current ACK. */
98 tcp_seq curack; /* Most recent ACK. */
99 uint32_t flags; /* Flags for cc_var (see below) */
100 int type; /* Indicates which ptr is valid in ccvc. */
101 union ccv_container {
102 struct tcpcb *tcp;
103 struct sctp_nets *sctp;
104 } ccvc;
105 uint16_t nsegs; /* # segments coalesced into current chain. */
106 uint8_t labc; /* Dont use system abc use passed in */
107};
108
109/* cc_var flags. */
110#define CCF_ABC_SENTAWND 0x0001 /* ABC counted cwnd worth of bytes? */
111#define CCF_CWND_LIMITED 0x0002 /* Are we currently cwnd limited? */
112#define CCF_USE_LOCAL_ABC 0x0004 /* Dont use the system l_abc val */
113#define CCF_ACKNOW 0x0008 /* Will this ack be sent now? */
114#define CCF_IPHDR_CE 0x0010 /* Does this packet set CE bit? */
115#define CCF_TCPHDR_CWR 0x0020 /* Does this packet set CWR bit? */
116#define CCF_MAX_CWND 0x0040 /* Have we reached maximum cwnd? */
117#define CCF_CHG_MAX_CWND 0x0080 /* CUBIC max_cwnd changed, for K */
118#define CCF_USR_IWND 0x0100 /* User specified initial window */
119#define CCF_USR_IWND_INIT_NSEG 0x0200 /* Convert segs to bytes on conn init */
120#define CCF_HYSTART_ALLOWED 0x0400 /* If the CC supports it Hystart is allowed */
121#define CCF_HYSTART_CAN_SH_CWND 0x0800 /* Can hystart when going CSS -> CA slam the cwnd */
122#define CCF_HYSTART_CONS_SSTH 0x1000 /* Should hystart use the more conservative ssthresh */
123
124/* ACK types passed to the ack_received() hook. */
125#define CC_ACK 0x0001 /* Regular in sequence ACK. */
126#define CC_DUPACK 0x0002 /* Duplicate ACK. */
127#define CC_PARTIALACK 0x0004 /* Not yet. */
128#define CC_SACK 0x0008 /* Not yet. */
129#endif /* defined(_KERNEL) || defined(_WANT_TCPCB) */
130
131/*
132 * Congestion signal types passed to the cong_signal() hook. The highest order 8
133 * bits (0x01000000 - 0x80000000) are reserved for CC algos to declare their own
134 * congestion signal types.
135 */
136#define CC_ECN 0x00000001 /* ECN marked packet received. */
137#define CC_RTO 0x00000002 /* RTO fired. */
138#define CC_RTO_ERR 0x00000004 /* RTO fired in error. */
139#define CC_NDUPACK 0x00000008 /* Threshold of dupack's reached. */
140
141#define CC_SIGPRIVMASK 0xFF000000 /* Mask to check if sig is private. */
142
143#ifdef _KERNEL
144/*
145 * Structure to hold data and function pointers that together represent a
146 * congestion control algorithm.
147 */
148struct cc_algo {
149 char name[TCP_CA_NAME_MAX];
150
151 /* Init global module state on kldload. */
152 int (*mod_init)(void);
153
154 /* Cleanup global module state on kldunload. */
155 int (*mod_destroy)(void);
156
157 /* Return the size of the void pointer the CC needs for state */
158 size_t (*cc_data_sz)(void);
159
160 /*
161 * Init CC state for a new control block. The CC
162 * module may be passed a NULL ptr indicating that
163 * it must allocate the memory. If it is passed a
164 * non-null pointer it is pre-allocated memory by
165 * the caller and the cb_init is expected to use that memory.
166 * It is not expected to fail if memory is passed in and
167 * all currently defined modules do not.
168 */
169 int (*cb_init)(struct cc_var *ccv, void *ptr);
170
171 /* Cleanup CC state for a terminating control block. */
172 void (*cb_destroy)(struct cc_var *ccv);
173
174 /* Init variables for a newly established connection. */
175 void (*conn_init)(struct cc_var *ccv);
176
177 /* Called on receipt of an ack. */
178 void (*ack_received)(struct cc_var *ccv, uint16_t type);
179
180 /* Called on detection of a congestion signal. */
181 void (*cong_signal)(struct cc_var *ccv, uint32_t type);
182
183 /* Called after exiting congestion recovery. */
184 void (*post_recovery)(struct cc_var *ccv);
185
186 /* Called when data transfer resumes after an idle period. */
187 void (*after_idle)(struct cc_var *ccv);
188
189 /* Called for an additional ECN processing apart from RFC3168. */
190 void (*ecnpkt_handler)(struct cc_var *ccv);
191
192 /* Called when a new "round" begins, if the transport is tracking rounds. */
193 void (*newround)(struct cc_var *ccv, uint32_t round_cnt);
194
195 /*
196 * Called when a RTT sample is made (fas = flight at send, if you dont have it
197 * send the cwnd in).
198 */
199 void (*rttsample)(struct cc_var *ccv, uint32_t usec_rtt, uint32_t rxtcnt, uint32_t fas);
200
201 /* Called for {get|set}sockopt() on a TCP socket with TCP_CCALGOOPT. */
202 int (*ctl_output)(struct cc_var *, struct sockopt *, void *);
203
204 STAILQ_ENTRY (cc_algo) entries;
205 u_int cc_refcount;
206 uint8_t flags;
207};
208
209#define CC_MODULE_BEING_REMOVED 0x01 /* The module is being removed */
210
211/* Macro to obtain the CC algo's struct ptr. */
212#define CC_ALGO(tp) ((tp)->t_cc)
213
214/* Macro to obtain the CC algo's data ptr. */
215#define CC_DATA(tp) ((tp)->t_ccv.cc_data)
216
217/* Macro to obtain the system default CC algo's struct ptr. */
218#define CC_DEFAULT_ALGO() V_default_cc_ptr
219
220extern struct rwlock cc_list_lock;
221#define CC_LIST_LOCK_INIT() rw_init(&cc_list_lock, "cc_list")
222#define CC_LIST_LOCK_DESTROY() rw_destroy(&cc_list_lock)
223#define CC_LIST_RLOCK() rw_rlock(&cc_list_lock)
224#define CC_LIST_RUNLOCK() rw_runlock(&cc_list_lock)
225#define CC_LIST_WLOCK() rw_wlock(&cc_list_lock)
226#define CC_LIST_WUNLOCK() rw_wunlock(&cc_list_lock)
227#define CC_LIST_LOCK_ASSERT() rw_assert(&cc_list_lock, RA_LOCKED)
228
229#define CC_ALGOOPT_LIMIT 2048
230
231/*
232 * These routines give NewReno behavior to the caller
233 * they require no state and can be used by any other CC
234 * module that wishes to use NewReno type behaviour (along
235 * with anything else they may add on, pre or post call).
236 */
237void newreno_cc_post_recovery(struct cc_var *);
238void newreno_cc_after_idle(struct cc_var *);
239void newreno_cc_cong_signal(struct cc_var *, uint32_t );
240void newreno_cc_ack_received(struct cc_var *, uint16_t);
241
242/* Called to temporarily keep an algo from going away during change */
243void cc_refer(struct cc_algo *algo);
244/* Called to release the temporary hold */
245void cc_release(struct cc_algo *algo);
246
247/* Called to attach a CC algorithm to a tcpcb */
248void cc_attach(struct tcpcb *, struct cc_algo *);
249/* Called to detach a CC algorithm from a tcpcb */
250void cc_detach(struct tcpcb *);
251
252#endif /* _KERNEL */
253#endif /* _NETINET_CC_CC_H_ */