Commit | Line | Data |
---|---|---|
d068ca2a MK |
1 | #ifndef __NET_SCHED_CODEL_IMPL_H |
2 | #define __NET_SCHED_CODEL_IMPL_H | |
3 | ||
4 | /* | |
5 | * Codel - The Controlled-Delay Active Queue Management algorithm | |
6 | * | |
7 | * Copyright (C) 2011-2012 Kathleen Nichols <nichols@pollere.com> | |
8 | * Copyright (C) 2011-2012 Van Jacobson <van@pollere.net> | |
9 | * Copyright (C) 2012 Michael D. Taht <dave.taht@bufferbloat.net> | |
10 | * Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com> | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or without | |
13 | * modification, are permitted provided that the following conditions | |
14 | * are met: | |
15 | * 1. Redistributions of source code must retain the above copyright | |
16 | * notice, this list of conditions, and the following disclaimer, | |
17 | * without modification. | |
18 | * 2. Redistributions in binary form must reproduce the above copyright | |
19 | * notice, this list of conditions and the following disclaimer in the | |
20 | * documentation and/or other materials provided with the distribution. | |
21 | * 3. The names of the authors may not be used to endorse or promote products | |
22 | * derived from this software without specific prior written permission. | |
23 | * | |
24 | * Alternatively, provided that this notice is retained in full, this | |
25 | * software may be distributed under the terms of the GNU General | |
26 | * Public License ("GPL") version 2, in which case the provisions of the | |
27 | * GPL apply INSTEAD OF those given above. | |
28 | * | |
29 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
30 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
31 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
32 | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT | |
33 | * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
34 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT | |
35 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | |
36 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
37 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | |
38 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
39 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH | |
40 | * DAMAGE. | |
41 | * | |
42 | */ | |
43 | ||
44 | /* Controlling Queue Delay (CoDel) algorithm | |
45 | * ========================================= | |
46 | * Source : Kathleen Nichols and Van Jacobson | |
47 | * http://queue.acm.org/detail.cfm?id=2209336 | |
48 | * | |
49 | * Implemented on linux by Dave Taht and Eric Dumazet | |
50 | */ | |
51 | ||
15fcb103 JK |
52 | #include <net/inet_ecn.h> |
53 | ||
d068ca2a MK |
54 | static void codel_params_init(struct codel_params *params) |
55 | { | |
56 | params->interval = MS2TIME(100); | |
57 | params->target = MS2TIME(5); | |
58 | params->ce_threshold = CODEL_DISABLED_THRESHOLD; | |
dfcb63ce THJ |
59 | params->ce_threshold_mask = 0; |
60 | params->ce_threshold_selector = 0; | |
d068ca2a MK |
61 | params->ecn = false; |
62 | } | |
63 | ||
64 | static void codel_vars_init(struct codel_vars *vars) | |
65 | { | |
66 | memset(vars, 0, sizeof(*vars)); | |
67 | } | |
68 | ||
69 | static void codel_stats_init(struct codel_stats *stats) | |
70 | { | |
71 | stats->maxpacket = 0; | |
72 | } | |
73 | ||
74 | /* | |
75 | * http://en.wikipedia.org/wiki/Methods_of_computing_square_roots#Iterative_methods_for_reciprocal_square_roots | |
76 | * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2) | |
77 | * | |
78 | * Here, invsqrt is a fixed point number (< 1.0), 32bit mantissa, aka Q0.32 | |
79 | */ | |
80 | static void codel_Newton_step(struct codel_vars *vars) | |
81 | { | |
82 | u32 invsqrt = ((u32)vars->rec_inv_sqrt) << REC_INV_SQRT_SHIFT; | |
83 | u32 invsqrt2 = ((u64)invsqrt * invsqrt) >> 32; | |
84 | u64 val = (3LL << 32) - ((u64)vars->count * invsqrt2); | |
85 | ||
86 | val >>= 2; /* avoid overflow in following multiply */ | |
87 | val = (val * invsqrt) >> (32 - 2 + 1); | |
88 | ||
89 | vars->rec_inv_sqrt = val >> REC_INV_SQRT_SHIFT; | |
90 | } | |
91 | ||
92 | /* | |
93 | * CoDel control_law is t + interval/sqrt(count) | |
94 | * We maintain in rec_inv_sqrt the reciprocal value of sqrt(count) to avoid | |
95 | * both sqrt() and divide operation. | |
96 | */ | |
97 | static codel_time_t codel_control_law(codel_time_t t, | |
98 | codel_time_t interval, | |
99 | u32 rec_inv_sqrt) | |
100 | { | |
101 | return t + reciprocal_scale(interval, rec_inv_sqrt << REC_INV_SQRT_SHIFT); | |
102 | } | |
103 | ||
104 | static bool codel_should_drop(const struct sk_buff *skb, | |
105 | void *ctx, | |
106 | struct codel_vars *vars, | |
107 | struct codel_params *params, | |
108 | struct codel_stats *stats, | |
109 | codel_skb_len_t skb_len_func, | |
110 | codel_skb_time_t skb_time_func, | |
111 | u32 *backlog, | |
112 | codel_time_t now) | |
113 | { | |
114 | bool ok_to_drop; | |
115 | u32 skb_len; | |
116 | ||
117 | if (!skb) { | |
118 | vars->first_above_time = 0; | |
119 | return false; | |
120 | } | |
121 | ||
122 | skb_len = skb_len_func(skb); | |
123 | vars->ldelay = now - skb_time_func(skb); | |
124 | ||
125 | if (unlikely(skb_len > stats->maxpacket)) | |
126 | stats->maxpacket = skb_len; | |
127 | ||
128 | if (codel_time_before(vars->ldelay, params->target) || | |
129 | *backlog <= params->mtu) { | |
130 | /* went below - stay below for at least interval */ | |
131 | vars->first_above_time = 0; | |
132 | return false; | |
133 | } | |
134 | ok_to_drop = false; | |
135 | if (vars->first_above_time == 0) { | |
136 | /* just went above from below. If we stay above | |
137 | * for at least interval we'll say it's ok to drop | |
138 | */ | |
139 | vars->first_above_time = now + params->interval; | |
140 | } else if (codel_time_after(now, vars->first_above_time)) { | |
141 | ok_to_drop = true; | |
142 | } | |
143 | return ok_to_drop; | |
144 | } | |
145 | ||
146 | static struct sk_buff *codel_dequeue(void *ctx, | |
147 | u32 *backlog, | |
148 | struct codel_params *params, | |
149 | struct codel_vars *vars, | |
150 | struct codel_stats *stats, | |
151 | codel_skb_len_t skb_len_func, | |
152 | codel_skb_time_t skb_time_func, | |
153 | codel_skb_drop_t drop_func, | |
154 | codel_skb_dequeue_t dequeue_func) | |
155 | { | |
156 | struct sk_buff *skb = dequeue_func(vars, ctx); | |
157 | codel_time_t now; | |
158 | bool drop; | |
159 | ||
160 | if (!skb) { | |
161 | vars->dropping = false; | |
162 | return skb; | |
163 | } | |
164 | now = codel_get_time(); | |
165 | drop = codel_should_drop(skb, ctx, vars, params, stats, | |
166 | skb_len_func, skb_time_func, backlog, now); | |
167 | if (vars->dropping) { | |
168 | if (!drop) { | |
169 | /* sojourn time below target - leave dropping state */ | |
170 | vars->dropping = false; | |
171 | } else if (codel_time_after_eq(now, vars->drop_next)) { | |
172 | /* It's time for the next drop. Drop the current | |
173 | * packet and dequeue the next. The dequeue might | |
174 | * take us out of dropping state. | |
175 | * If not, schedule the next drop. | |
176 | * A large backlog might result in drop rates so high | |
177 | * that the next drop should happen now, | |
178 | * hence the while loop. | |
179 | */ | |
180 | while (vars->dropping && | |
181 | codel_time_after_eq(now, vars->drop_next)) { | |
182 | vars->count++; /* dont care of possible wrap | |
183 | * since there is no more divide | |
184 | */ | |
185 | codel_Newton_step(vars); | |
186 | if (params->ecn && INET_ECN_set_ce(skb)) { | |
187 | stats->ecn_mark++; | |
188 | vars->drop_next = | |
189 | codel_control_law(vars->drop_next, | |
190 | params->interval, | |
191 | vars->rec_inv_sqrt); | |
192 | goto end; | |
193 | } | |
194 | stats->drop_len += skb_len_func(skb); | |
195 | drop_func(skb, ctx); | |
196 | stats->drop_count++; | |
197 | skb = dequeue_func(vars, ctx); | |
198 | if (!codel_should_drop(skb, ctx, | |
199 | vars, params, stats, | |
200 | skb_len_func, | |
201 | skb_time_func, | |
202 | backlog, now)) { | |
203 | /* leave dropping state */ | |
204 | vars->dropping = false; | |
205 | } else { | |
206 | /* and schedule the next drop */ | |
207 | vars->drop_next = | |
208 | codel_control_law(vars->drop_next, | |
209 | params->interval, | |
210 | vars->rec_inv_sqrt); | |
211 | } | |
212 | } | |
213 | } | |
214 | } else if (drop) { | |
215 | u32 delta; | |
216 | ||
217 | if (params->ecn && INET_ECN_set_ce(skb)) { | |
218 | stats->ecn_mark++; | |
219 | } else { | |
220 | stats->drop_len += skb_len_func(skb); | |
221 | drop_func(skb, ctx); | |
222 | stats->drop_count++; | |
223 | ||
224 | skb = dequeue_func(vars, ctx); | |
225 | drop = codel_should_drop(skb, ctx, vars, params, | |
226 | stats, skb_len_func, | |
227 | skb_time_func, backlog, now); | |
228 | } | |
229 | vars->dropping = true; | |
230 | /* if min went above target close to when we last went below it | |
231 | * assume that the drop rate that controlled the queue on the | |
232 | * last cycle is a good starting point to control it now. | |
233 | */ | |
234 | delta = vars->count - vars->lastcount; | |
235 | if (delta > 1 && | |
236 | codel_time_before(now - vars->drop_next, | |
237 | 16 * params->interval)) { | |
238 | vars->count = delta; | |
239 | /* we dont care if rec_inv_sqrt approximation | |
240 | * is not very precise : | |
241 | * Next Newton steps will correct it quadratically. | |
242 | */ | |
243 | codel_Newton_step(vars); | |
244 | } else { | |
245 | vars->count = 1; | |
246 | vars->rec_inv_sqrt = ~0U >> REC_INV_SQRT_SHIFT; | |
247 | } | |
248 | vars->lastcount = vars->count; | |
249 | vars->drop_next = codel_control_law(now, params->interval, | |
250 | vars->rec_inv_sqrt); | |
251 | } | |
252 | end: | |
e72aeb9e ED |
253 | if (skb && codel_time_after(vars->ldelay, params->ce_threshold)) { |
254 | bool set_ce = true; | |
255 | ||
dfcb63ce THJ |
256 | if (params->ce_threshold_mask) { |
257 | int dsfield = skb_get_dsfield(skb); | |
e72aeb9e | 258 | |
dfcb63ce THJ |
259 | set_ce = (dsfield >= 0 && |
260 | (((u8)dsfield & params->ce_threshold_mask) == | |
261 | params->ce_threshold_selector)); | |
e72aeb9e ED |
262 | } |
263 | if (set_ce && INET_ECN_set_ce(skb)) | |
264 | stats->ce_mark++; | |
265 | } | |
d068ca2a MK |
266 | return skb; |
267 | } | |
268 | ||
269 | #endif |