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 *
| 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 *
|
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
| 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
|
56
57/* Global CC vars. */
58extern STAILQ_HEAD(cc_head, cc_algo) cc_list;
59extern const int tcprexmtthresh;
60extern struct cc_algo newreno_cc_algo;
61
62/* Per-netstack bits. */
63VNET_DECLARE(struct cc_algo *, default_cc_ptr);
64#define V_default_cc_ptr VNET(default_cc_ptr)
65
66/* Define the new net.inet.tcp.cc sysctl tree. */
67SYSCTL_DECL(_net_inet_tcp_cc);
68
69/* CC housekeeping functions. */
70int cc_register_algo(struct cc_algo *add_cc);
71int cc_deregister_algo(struct cc_algo *remove_cc);
72
73/*
74 * Wrapper around transport structs that contain same-named congestion
75 * control variables. Allows algos to be shared amongst multiple CC aware
76 * transprots.
77 */
78struct cc_var {
79 void *cc_data; /* Per-connection private CC algorithm data. */
80 int bytes_this_ack; /* # bytes acked by the current ACK. */
81 tcp_seq curack; /* Most recent ACK. */
82 uint32_t flags; /* Flags for cc_var (see below) */
83 int type; /* Indicates which ptr is valid in ccvc. */
84 union ccv_container {
85 struct tcpcb *tcp;
86 struct sctp_nets *sctp;
87 } ccvc;
88};
89
90/* cc_var flags. */
91#define CCF_ABC_SENTAWND 0x0001 /* ABC counted cwnd worth of bytes? */
92#define CCF_CWND_LIMITED 0x0002 /* Are we currently cwnd limited? */
93#define CCF_DELACK 0x0004 /* Is this ack delayed? */
94#define CCF_ACKNOW 0x0008 /* Will this ack be sent now? */
95#define CCF_IPHDR_CE 0x0010 /* Does this packet set CE bit? */
96#define CCF_TCPHDR_CWR 0x0020 /* Does this packet set CWR bit? */
97
98/* ACK types passed to the ack_received() hook. */
99#define CC_ACK 0x0001 /* Regular in sequence ACK. */
100#define CC_DUPACK 0x0002 /* Duplicate ACK. */
101#define CC_PARTIALACK 0x0004 /* Not yet. */
102#define CC_SACK 0x0008 /* Not yet. */
103
104/*
105 * Congestion signal types passed to the cong_signal() hook. The highest order 8
106 * bits (0x01000000 - 0x80000000) are reserved for CC algos to declare their own
107 * congestion signal types.
108 */
109#define CC_ECN 0x00000001 /* ECN marked packet received. */
110#define CC_RTO 0x00000002 /* RTO fired. */
111#define CC_RTO_ERR 0x00000004 /* RTO fired in error. */
112#define CC_NDUPACK 0x00000008 /* Threshold of dupack's reached. */
113
114#define CC_SIGPRIVMASK 0xFF000000 /* Mask to check if sig is private. */
115
116/*
117 * Structure to hold data and function pointers that together represent a
118 * congestion control algorithm.
119 */
120struct cc_algo {
121 char name[TCP_CA_NAME_MAX];
122
123 /* Init global module state on kldload. */
124 int (*mod_init)(void);
125
126 /* Cleanup global module state on kldunload. */
127 int (*mod_destroy)(void);
128
129 /* Init CC state for a new control block. */
130 int (*cb_init)(struct cc_var *ccv);
131
132 /* Cleanup CC state for a terminating control block. */
133 void (*cb_destroy)(struct cc_var *ccv);
134
135 /* Init variables for a newly established connection. */
136 void (*conn_init)(struct cc_var *ccv);
137
138 /* Called on receipt of an ack. */
139 void (*ack_received)(struct cc_var *ccv, uint16_t type);
140
141 /* Called on detection of a congestion signal. */
142 void (*cong_signal)(struct cc_var *ccv, uint32_t type);
143
144 /* Called after exiting congestion recovery. */
145 void (*post_recovery)(struct cc_var *ccv);
146
147 /* Called when data transfer resumes after an idle period. */
148 void (*after_idle)(struct cc_var *ccv);
149
150 /* Called for an additional ECN processing apart from RFC3168. */
151 void (*ecnpkt_handler)(struct cc_var *ccv);
152
153 STAILQ_ENTRY (cc_algo) entries;
154};
155
156/* Macro to obtain the CC algo's struct ptr. */
157#define CC_ALGO(tp) ((tp)->cc_algo)
158
159/* Macro to obtain the CC algo's data ptr. */
160#define CC_DATA(tp) ((tp)->ccv->cc_data)
161
162/* Macro to obtain the system default CC algo's struct ptr. */
163#define CC_DEFAULT() V_default_cc_ptr
164
165extern struct rwlock cc_list_lock;
166#define CC_LIST_LOCK_INIT() rw_init(&cc_list_lock, "cc_list")
167#define CC_LIST_LOCK_DESTROY() rw_destroy(&cc_list_lock)
168#define CC_LIST_RLOCK() rw_rlock(&cc_list_lock)
169#define CC_LIST_RUNLOCK() rw_runlock(&cc_list_lock)
170#define CC_LIST_WLOCK() rw_wlock(&cc_list_lock)
171#define CC_LIST_WUNLOCK() rw_wunlock(&cc_list_lock)
172#define CC_LIST_LOCK_ASSERT() rw_assert(&cc_list_lock, RA_LOCKED)
173
| 57
58/* Global CC vars. */
59extern STAILQ_HEAD(cc_head, cc_algo) cc_list;
60extern const int tcprexmtthresh;
61extern struct cc_algo newreno_cc_algo;
62
63/* Per-netstack bits. */
64VNET_DECLARE(struct cc_algo *, default_cc_ptr);
65#define V_default_cc_ptr VNET(default_cc_ptr)
66
67/* Define the new net.inet.tcp.cc sysctl tree. */
68SYSCTL_DECL(_net_inet_tcp_cc);
69
70/* CC housekeeping functions. */
71int cc_register_algo(struct cc_algo *add_cc);
72int cc_deregister_algo(struct cc_algo *remove_cc);
73
74/*
75 * Wrapper around transport structs that contain same-named congestion
76 * control variables. Allows algos to be shared amongst multiple CC aware
77 * transprots.
78 */
79struct cc_var {
80 void *cc_data; /* Per-connection private CC algorithm data. */
81 int bytes_this_ack; /* # bytes acked by the current ACK. */
82 tcp_seq curack; /* Most recent ACK. */
83 uint32_t flags; /* Flags for cc_var (see below) */
84 int type; /* Indicates which ptr is valid in ccvc. */
85 union ccv_container {
86 struct tcpcb *tcp;
87 struct sctp_nets *sctp;
88 } ccvc;
89};
90
91/* cc_var flags. */
92#define CCF_ABC_SENTAWND 0x0001 /* ABC counted cwnd worth of bytes? */
93#define CCF_CWND_LIMITED 0x0002 /* Are we currently cwnd limited? */
94#define CCF_DELACK 0x0004 /* Is this ack delayed? */
95#define CCF_ACKNOW 0x0008 /* Will this ack be sent now? */
96#define CCF_IPHDR_CE 0x0010 /* Does this packet set CE bit? */
97#define CCF_TCPHDR_CWR 0x0020 /* Does this packet set CWR bit? */
98
99/* ACK types passed to the ack_received() hook. */
100#define CC_ACK 0x0001 /* Regular in sequence ACK. */
101#define CC_DUPACK 0x0002 /* Duplicate ACK. */
102#define CC_PARTIALACK 0x0004 /* Not yet. */
103#define CC_SACK 0x0008 /* Not yet. */
104
105/*
106 * Congestion signal types passed to the cong_signal() hook. The highest order 8
107 * bits (0x01000000 - 0x80000000) are reserved for CC algos to declare their own
108 * congestion signal types.
109 */
110#define CC_ECN 0x00000001 /* ECN marked packet received. */
111#define CC_RTO 0x00000002 /* RTO fired. */
112#define CC_RTO_ERR 0x00000004 /* RTO fired in error. */
113#define CC_NDUPACK 0x00000008 /* Threshold of dupack's reached. */
114
115#define CC_SIGPRIVMASK 0xFF000000 /* Mask to check if sig is private. */
116
117/*
118 * Structure to hold data and function pointers that together represent a
119 * congestion control algorithm.
120 */
121struct cc_algo {
122 char name[TCP_CA_NAME_MAX];
123
124 /* Init global module state on kldload. */
125 int (*mod_init)(void);
126
127 /* Cleanup global module state on kldunload. */
128 int (*mod_destroy)(void);
129
130 /* Init CC state for a new control block. */
131 int (*cb_init)(struct cc_var *ccv);
132
133 /* Cleanup CC state for a terminating control block. */
134 void (*cb_destroy)(struct cc_var *ccv);
135
136 /* Init variables for a newly established connection. */
137 void (*conn_init)(struct cc_var *ccv);
138
139 /* Called on receipt of an ack. */
140 void (*ack_received)(struct cc_var *ccv, uint16_t type);
141
142 /* Called on detection of a congestion signal. */
143 void (*cong_signal)(struct cc_var *ccv, uint32_t type);
144
145 /* Called after exiting congestion recovery. */
146 void (*post_recovery)(struct cc_var *ccv);
147
148 /* Called when data transfer resumes after an idle period. */
149 void (*after_idle)(struct cc_var *ccv);
150
151 /* Called for an additional ECN processing apart from RFC3168. */
152 void (*ecnpkt_handler)(struct cc_var *ccv);
153
154 STAILQ_ENTRY (cc_algo) entries;
155};
156
157/* Macro to obtain the CC algo's struct ptr. */
158#define CC_ALGO(tp) ((tp)->cc_algo)
159
160/* Macro to obtain the CC algo's data ptr. */
161#define CC_DATA(tp) ((tp)->ccv->cc_data)
162
163/* Macro to obtain the system default CC algo's struct ptr. */
164#define CC_DEFAULT() V_default_cc_ptr
165
166extern struct rwlock cc_list_lock;
167#define CC_LIST_LOCK_INIT() rw_init(&cc_list_lock, "cc_list")
168#define CC_LIST_LOCK_DESTROY() rw_destroy(&cc_list_lock)
169#define CC_LIST_RLOCK() rw_rlock(&cc_list_lock)
170#define CC_LIST_RUNLOCK() rw_runlock(&cc_list_lock)
171#define CC_LIST_WLOCK() rw_wlock(&cc_list_lock)
172#define CC_LIST_WUNLOCK() rw_wunlock(&cc_list_lock)
173#define CC_LIST_LOCK_ASSERT() rw_assert(&cc_list_lock, RA_LOCKED)
174
|