1 | /* |
2 | * TCP Veno congestion control |
3 | * |
4 | * This is based on the congestion detection/avoidance scheme described in |
5 | * C. P. Fu, S. C. Liew. |
6 | * "TCP Veno: TCP Enhancement for Transmission over Wireless Access Networks." |
7 | * IEEE Journal on Selected Areas in Communication, |
8 | * Feb. 2003. |
9 | * See http://www.ie.cuhk.edu.hk/fileadmin/staff_upload/soung/Journal/J3.pdf |
10 | */ |
11 | |
12 | #include <linux/mm.h> |
13 | #include <linux/module.h> |
14 | #include <linux/skbuff.h> |
15 | #include <linux/inet_diag.h> |
16 | |
17 | #include <net/tcp.h> |
18 | |
19 | /* Default values of the Veno variables, in fixed-point representation |
20 | * with V_PARAM_SHIFT bits to the right of the binary point. |
21 | */ |
22 | #define V_PARAM_SHIFT 1 |
23 | static const int beta = 3 << V_PARAM_SHIFT; |
24 | |
25 | /* Veno variables */ |
26 | struct veno { |
27 | u8 doing_veno_now; /* if true, do veno for this rtt */ |
28 | u16 cntrtt; /* # of rtts measured within last rtt */ |
29 | u32 minrtt; /* min of rtts measured within last rtt (in usec) */ |
30 | u32 basertt; /* the min of all Veno rtt measurements seen (in usec) */ |
31 | u32 inc; /* decide whether to increase cwnd */ |
32 | u32 diff; /* calculate the diff rate */ |
33 | }; |
34 | |
35 | /* There are several situations when we must "re-start" Veno: |
36 | * |
37 | * o when a connection is established |
38 | * o after an RTO |
39 | * o after fast recovery |
40 | * o when we send a packet and there is no outstanding |
41 | * unacknowledged data (restarting an idle connection) |
42 | * |
43 | */ |
44 | static inline void veno_enable(struct sock *sk) |
45 | { |
46 | struct veno *veno = inet_csk_ca(sk); |
47 | |
48 | /* turn on Veno */ |
49 | veno->doing_veno_now = 1; |
50 | |
51 | veno->minrtt = 0x7fffffff; |
52 | } |
53 | |
54 | static inline void veno_disable(struct sock *sk) |
55 | { |
56 | struct veno *veno = inet_csk_ca(sk); |
57 | |
58 | /* turn off Veno */ |
59 | veno->doing_veno_now = 0; |
60 | } |
61 | |
62 | static void tcp_veno_init(struct sock *sk) |
63 | { |
64 | struct veno *veno = inet_csk_ca(sk); |
65 | |
66 | veno->basertt = 0x7fffffff; |
67 | veno->inc = 1; |
68 | veno_enable(sk); |
69 | } |
70 | |
71 | /* Do rtt sampling needed for Veno. */ |
72 | static void tcp_veno_pkts_acked(struct sock *sk, |
73 | const struct ack_sample *sample) |
74 | { |
75 | struct veno *veno = inet_csk_ca(sk); |
76 | u32 vrtt; |
77 | |
78 | if (sample->rtt_us < 0) |
79 | return; |
80 | |
81 | /* Never allow zero rtt or baseRTT */ |
82 | vrtt = sample->rtt_us + 1; |
83 | |
84 | /* Filter to find propagation delay: */ |
85 | if (vrtt < veno->basertt) |
86 | veno->basertt = vrtt; |
87 | |
88 | /* Find the min rtt during the last rtt to find |
89 | * the current prop. delay + queuing delay: |
90 | */ |
91 | veno->minrtt = min(veno->minrtt, vrtt); |
92 | veno->cntrtt++; |
93 | } |
94 | |
95 | static void tcp_veno_state(struct sock *sk, u8 ca_state) |
96 | { |
97 | if (ca_state == TCP_CA_Open) |
98 | veno_enable(sk); |
99 | else |
100 | veno_disable(sk); |
101 | } |
102 | |
103 | /* |
104 | * If the connection is idle and we are restarting, |
105 | * then we don't want to do any Veno calculations |
106 | * until we get fresh rtt samples. So when we |
107 | * restart, we reset our Veno state to a clean |
108 | * state. After we get acks for this flight of |
109 | * packets, _then_ we can make Veno calculations |
110 | * again. |
111 | */ |
112 | static void tcp_veno_cwnd_event(struct sock *sk, enum tcp_ca_event event) |
113 | { |
114 | if (event == CA_EVENT_CWND_RESTART || event == CA_EVENT_TX_START) |
115 | tcp_veno_init(sk); |
116 | } |
117 | |
118 | static void tcp_veno_cong_avoid(struct sock *sk, u32 ack, u32 acked) |
119 | { |
120 | struct tcp_sock *tp = tcp_sk(sk); |
121 | struct veno *veno = inet_csk_ca(sk); |
122 | |
123 | if (!veno->doing_veno_now) { |
124 | tcp_reno_cong_avoid(sk, ack, acked); |
125 | return; |
126 | } |
127 | |
128 | /* limited by applications */ |
129 | if (!tcp_is_cwnd_limited(sk)) |
130 | return; |
131 | |
132 | /* We do the Veno calculations only if we got enough rtt samples */ |
133 | if (veno->cntrtt <= 2) { |
134 | /* We don't have enough rtt samples to do the Veno |
135 | * calculation, so we'll behave like Reno. |
136 | */ |
137 | tcp_reno_cong_avoid(sk, ack, acked); |
138 | } else { |
139 | u64 target_cwnd; |
140 | u32 rtt; |
141 | |
142 | /* We have enough rtt samples, so, using the Veno |
143 | * algorithm, we determine the state of the network. |
144 | */ |
145 | |
146 | rtt = veno->minrtt; |
147 | |
148 | target_cwnd = (u64)tp->snd_cwnd * veno->basertt; |
149 | target_cwnd <<= V_PARAM_SHIFT; |
150 | do_div(target_cwnd, rtt); |
151 | |
152 | veno->diff = (tp->snd_cwnd << V_PARAM_SHIFT) - target_cwnd; |
153 | |
154 | if (tcp_in_slow_start(tp)) { |
155 | /* Slow start. */ |
156 | tcp_slow_start(tp, acked); |
157 | } else { |
158 | /* Congestion avoidance. */ |
159 | if (veno->diff < beta) { |
160 | /* In the "non-congestive state", increase cwnd |
161 | * every rtt. |
162 | */ |
163 | tcp_cong_avoid_ai(tp, tp->snd_cwnd, 1); |
164 | } else { |
165 | /* In the "congestive state", increase cwnd |
166 | * every other rtt. |
167 | */ |
168 | if (tp->snd_cwnd_cnt >= tp->snd_cwnd) { |
169 | if (veno->inc && |
170 | tp->snd_cwnd < tp->snd_cwnd_clamp) { |
171 | tp->snd_cwnd++; |
172 | veno->inc = 0; |
173 | } else |
174 | veno->inc = 1; |
175 | tp->snd_cwnd_cnt = 0; |
176 | } else |
177 | tp->snd_cwnd_cnt++; |
178 | } |
179 | } |
180 | if (tp->snd_cwnd < 2) |
181 | tp->snd_cwnd = 2; |
182 | else if (tp->snd_cwnd > tp->snd_cwnd_clamp) |
183 | tp->snd_cwnd = tp->snd_cwnd_clamp; |
184 | } |
185 | /* Wipe the slate clean for the next rtt. */ |
186 | /* veno->cntrtt = 0; */ |
187 | veno->minrtt = 0x7fffffff; |
188 | } |
189 | |
190 | /* Veno MD phase */ |
191 | static u32 tcp_veno_ssthresh(struct sock *sk) |
192 | { |
193 | const struct tcp_sock *tp = tcp_sk(sk); |
194 | struct veno *veno = inet_csk_ca(sk); |
195 | |
196 | if (veno->diff < beta) |
197 | /* in "non-congestive state", cut cwnd by 1/5 */ |
198 | return max(tp->snd_cwnd * 4 / 5, 2U); |
199 | else |
200 | /* in "congestive state", cut cwnd by 1/2 */ |
201 | return max(tp->snd_cwnd >> 1U, 2U); |
202 | } |
203 | |
204 | static struct tcp_congestion_ops tcp_veno __read_mostly = { |
205 | .init = tcp_veno_init, |
206 | .ssthresh = tcp_veno_ssthresh, |
207 | .undo_cwnd = tcp_reno_undo_cwnd, |
208 | .cong_avoid = tcp_veno_cong_avoid, |
209 | .pkts_acked = tcp_veno_pkts_acked, |
210 | .set_state = tcp_veno_state, |
211 | .cwnd_event = tcp_veno_cwnd_event, |
212 | |
213 | .owner = THIS_MODULE, |
214 | .name = "veno" , |
215 | }; |
216 | |
217 | static int __init tcp_veno_register(void) |
218 | { |
219 | BUILD_BUG_ON(sizeof(struct veno) > ICSK_CA_PRIV_SIZE); |
220 | tcp_register_congestion_control(&tcp_veno); |
221 | return 0; |
222 | } |
223 | |
224 | static void __exit tcp_veno_unregister(void) |
225 | { |
226 | tcp_unregister_congestion_control(&tcp_veno); |
227 | } |
228 | |
229 | module_init(tcp_veno_register); |
230 | module_exit(tcp_veno_unregister); |
231 | |
232 | MODULE_AUTHOR("Bin Zhou, Cheng Peng Fu" ); |
233 | MODULE_LICENSE("GPL" ); |
234 | MODULE_DESCRIPTION("TCP Veno" ); |
235 | |