Commit | Line | Data |
---|---|---|
4b549a2e ED |
1 | /* |
2 | * Fair Queue CoDel discipline | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
80ba92fa | 9 | * Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com> |
4b549a2e ED |
10 | */ |
11 | ||
12 | #include <linux/module.h> | |
13 | #include <linux/types.h> | |
14 | #include <linux/kernel.h> | |
15 | #include <linux/jiffies.h> | |
16 | #include <linux/string.h> | |
17 | #include <linux/in.h> | |
18 | #include <linux/errno.h> | |
19 | #include <linux/init.h> | |
20 | #include <linux/skbuff.h> | |
21 | #include <linux/jhash.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/vmalloc.h> | |
24 | #include <net/netlink.h> | |
25 | #include <net/pkt_sched.h> | |
4b549a2e | 26 | #include <net/codel.h> |
d068ca2a MK |
27 | #include <net/codel_impl.h> |
28 | #include <net/codel_qdisc.h> | |
4b549a2e ED |
29 | |
30 | /* Fair Queue CoDel. | |
31 | * | |
32 | * Principles : | |
33 | * Packets are classified (internal classifier or external) on flows. | |
34 | * This is a Stochastic model (as we use a hash, several flows | |
35 | * might be hashed on same slot) | |
36 | * Each flow has a CoDel managed queue. | |
37 | * Flows are linked onto two (Round Robin) lists, | |
38 | * so that new flows have priority on old ones. | |
39 | * | |
40 | * For a given flow, packets are not reordered (CoDel uses a FIFO) | |
41 | * head drops only. | |
42 | * ECN capability is on by default. | |
43 | * Low memory footprint (64 bytes per flow) | |
44 | */ | |
45 | ||
46 | struct fq_codel_flow { | |
47 | struct sk_buff *head; | |
48 | struct sk_buff *tail; | |
49 | struct list_head flowchain; | |
50 | int deficit; | |
51 | u32 dropped; /* number of drops (or ECN marks) on this flow */ | |
52 | struct codel_vars cvars; | |
53 | }; /* please try to keep this structure <= 64 bytes */ | |
54 | ||
55 | struct fq_codel_sched_data { | |
25d8c0d5 | 56 | struct tcf_proto __rcu *filter_list; /* optional external classifier */ |
4b549a2e ED |
57 | struct fq_codel_flow *flows; /* Flows table [flows_cnt] */ |
58 | u32 *backlogs; /* backlog table [flows_cnt] */ | |
59 | u32 flows_cnt; /* number of flows */ | |
60 | u32 perturbation; /* hash perturbation */ | |
61 | u32 quantum; /* psched_mtu(qdisc_dev(sch)); */ | |
9d18562a | 62 | u32 drop_batch_size; |
95b58430 | 63 | u32 memory_limit; |
4b549a2e ED |
64 | struct codel_params cparams; |
65 | struct codel_stats cstats; | |
95b58430 ED |
66 | u32 memory_usage; |
67 | u32 drop_overmemory; | |
4b549a2e ED |
68 | u32 drop_overlimit; |
69 | u32 new_flow_count; | |
70 | ||
71 | struct list_head new_flows; /* list of new flows */ | |
72 | struct list_head old_flows; /* list of old flows */ | |
73 | }; | |
74 | ||
75 | static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q, | |
342db221 | 76 | struct sk_buff *skb) |
4b549a2e | 77 | { |
342db221 | 78 | u32 hash = skb_get_hash_perturb(skb, q->perturbation); |
8fc54f68 DB |
79 | |
80 | return reciprocal_scale(hash, q->flows_cnt); | |
4b549a2e ED |
81 | } |
82 | ||
83 | static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch, | |
84 | int *qerr) | |
85 | { | |
86 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
25d8c0d5 | 87 | struct tcf_proto *filter; |
4b549a2e ED |
88 | struct tcf_result res; |
89 | int result; | |
90 | ||
91 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
92 | TC_H_MIN(skb->priority) > 0 && | |
93 | TC_H_MIN(skb->priority) <= q->flows_cnt) | |
94 | return TC_H_MIN(skb->priority); | |
95 | ||
69204cf7 | 96 | filter = rcu_dereference_bh(q->filter_list); |
25d8c0d5 | 97 | if (!filter) |
4b549a2e ED |
98 | return fq_codel_hash(q, skb) + 1; |
99 | ||
100 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
3b3ae880 | 101 | result = tc_classify(skb, filter, &res, false); |
4b549a2e ED |
102 | if (result >= 0) { |
103 | #ifdef CONFIG_NET_CLS_ACT | |
104 | switch (result) { | |
105 | case TC_ACT_STOLEN: | |
106 | case TC_ACT_QUEUED: | |
107 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; | |
108 | case TC_ACT_SHOT: | |
109 | return 0; | |
110 | } | |
111 | #endif | |
112 | if (TC_H_MIN(res.classid) <= q->flows_cnt) | |
113 | return TC_H_MIN(res.classid); | |
114 | } | |
115 | return 0; | |
116 | } | |
117 | ||
118 | /* helper functions : might be changed when/if skb use a standard list_head */ | |
119 | ||
120 | /* remove one skb from head of slot queue */ | |
121 | static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow) | |
122 | { | |
123 | struct sk_buff *skb = flow->head; | |
124 | ||
125 | flow->head = skb->next; | |
126 | skb->next = NULL; | |
127 | return skb; | |
128 | } | |
129 | ||
130 | /* add skb to flow queue (tail add) */ | |
131 | static inline void flow_queue_add(struct fq_codel_flow *flow, | |
132 | struct sk_buff *skb) | |
133 | { | |
134 | if (flow->head == NULL) | |
135 | flow->head = skb; | |
136 | else | |
137 | flow->tail->next = skb; | |
138 | flow->tail = skb; | |
139 | skb->next = NULL; | |
140 | } | |
141 | ||
520ac30f ED |
142 | static unsigned int fq_codel_drop(struct Qdisc *sch, unsigned int max_packets, |
143 | struct sk_buff **to_free) | |
4b549a2e ED |
144 | { |
145 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
146 | struct sk_buff *skb; | |
147 | unsigned int maxbacklog = 0, idx = 0, i, len; | |
148 | struct fq_codel_flow *flow; | |
9d18562a | 149 | unsigned int threshold; |
95b58430 | 150 | unsigned int mem = 0; |
4b549a2e | 151 | |
9d18562a | 152 | /* Queue is full! Find the fat flow and drop packet(s) from it. |
4b549a2e ED |
153 | * This might sound expensive, but with 1024 flows, we scan |
154 | * 4KB of memory, and we dont need to handle a complex tree | |
155 | * in fast path (packet queue/enqueue) with many cache misses. | |
9d18562a ED |
156 | * In stress mode, we'll try to drop 64 packets from the flow, |
157 | * amortizing this linear lookup to one cache line per drop. | |
4b549a2e ED |
158 | */ |
159 | for (i = 0; i < q->flows_cnt; i++) { | |
160 | if (q->backlogs[i] > maxbacklog) { | |
161 | maxbacklog = q->backlogs[i]; | |
162 | idx = i; | |
163 | } | |
164 | } | |
9d18562a ED |
165 | |
166 | /* Our goal is to drop half of this fat flow backlog */ | |
167 | threshold = maxbacklog >> 1; | |
168 | ||
4b549a2e | 169 | flow = &q->flows[idx]; |
9d18562a ED |
170 | len = 0; |
171 | i = 0; | |
172 | do { | |
173 | skb = dequeue_head(flow); | |
174 | len += qdisc_pkt_len(skb); | |
95b58430 | 175 | mem += skb->truesize; |
520ac30f | 176 | __qdisc_drop(skb, to_free); |
9d18562a ED |
177 | } while (++i < max_packets && len < threshold); |
178 | ||
179 | flow->dropped += i; | |
4b549a2e | 180 | q->backlogs[idx] -= len; |
95b58430 | 181 | q->memory_usage -= mem; |
9d18562a ED |
182 | sch->qstats.drops += i; |
183 | sch->qstats.backlog -= len; | |
184 | sch->q.qlen -= i; | |
4b549a2e ED |
185 | return idx; |
186 | } | |
187 | ||
520ac30f ED |
188 | static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
189 | struct sk_buff **to_free) | |
4b549a2e ED |
190 | { |
191 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
9d18562a | 192 | unsigned int idx, prev_backlog, prev_qlen; |
4b549a2e ED |
193 | struct fq_codel_flow *flow; |
194 | int uninitialized_var(ret); | |
80e509db | 195 | unsigned int pkt_len; |
95b58430 | 196 | bool memory_limited; |
4b549a2e ED |
197 | |
198 | idx = fq_codel_classify(skb, sch, &ret); | |
199 | if (idx == 0) { | |
200 | if (ret & __NET_XMIT_BYPASS) | |
25331d6c | 201 | qdisc_qstats_drop(sch); |
520ac30f | 202 | __qdisc_drop(skb, to_free); |
4b549a2e ED |
203 | return ret; |
204 | } | |
205 | idx--; | |
206 | ||
207 | codel_set_enqueue_time(skb); | |
208 | flow = &q->flows[idx]; | |
209 | flow_queue_add(flow, skb); | |
210 | q->backlogs[idx] += qdisc_pkt_len(skb); | |
25331d6c | 211 | qdisc_qstats_backlog_inc(sch, skb); |
4b549a2e ED |
212 | |
213 | if (list_empty(&flow->flowchain)) { | |
214 | list_add_tail(&flow->flowchain, &q->new_flows); | |
4b549a2e ED |
215 | q->new_flow_count++; |
216 | flow->deficit = q->quantum; | |
217 | flow->dropped = 0; | |
218 | } | |
95b58430 ED |
219 | q->memory_usage += skb->truesize; |
220 | memory_limited = q->memory_usage > q->memory_limit; | |
221 | if (++sch->q.qlen <= sch->limit && !memory_limited) | |
4b549a2e ED |
222 | return NET_XMIT_SUCCESS; |
223 | ||
2ccccf5f | 224 | prev_backlog = sch->qstats.backlog; |
9d18562a ED |
225 | prev_qlen = sch->q.qlen; |
226 | ||
80e509db ED |
227 | /* save this packet length as it might be dropped by fq_codel_drop() */ |
228 | pkt_len = qdisc_pkt_len(skb); | |
9d18562a ED |
229 | /* fq_codel_drop() is quite expensive, as it performs a linear search |
230 | * in q->backlogs[] to find a fat flow. | |
231 | * So instead of dropping a single packet, drop half of its backlog | |
232 | * with a 64 packets limit to not add a too big cpu spike here. | |
4b549a2e | 233 | */ |
520ac30f | 234 | ret = fq_codel_drop(sch, q->drop_batch_size, to_free); |
9d18562a | 235 | |
80e509db ED |
236 | prev_qlen -= sch->q.qlen; |
237 | prev_backlog -= sch->qstats.backlog; | |
238 | q->drop_overlimit += prev_qlen; | |
95b58430 | 239 | if (memory_limited) |
80e509db | 240 | q->drop_overmemory += prev_qlen; |
9d18562a | 241 | |
80e509db ED |
242 | /* As we dropped packet(s), better let upper stack know this. |
243 | * If we dropped a packet for this flow, return NET_XMIT_CN, | |
244 | * but in this case, our parents wont increase their backlogs. | |
245 | */ | |
246 | if (ret == idx) { | |
247 | qdisc_tree_reduce_backlog(sch, prev_qlen - 1, | |
248 | prev_backlog - pkt_len); | |
249 | return NET_XMIT_CN; | |
250 | } | |
251 | qdisc_tree_reduce_backlog(sch, prev_qlen, prev_backlog); | |
252 | return NET_XMIT_SUCCESS; | |
4b549a2e ED |
253 | } |
254 | ||
255 | /* This is the specific function called from codel_dequeue() | |
256 | * to dequeue a packet from queue. Note: backlog is handled in | |
257 | * codel, we dont need to reduce it here. | |
258 | */ | |
79bdc4c8 | 259 | static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx) |
4b549a2e | 260 | { |
79bdc4c8 | 261 | struct Qdisc *sch = ctx; |
865ec552 | 262 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
4b549a2e ED |
263 | struct fq_codel_flow *flow; |
264 | struct sk_buff *skb = NULL; | |
265 | ||
266 | flow = container_of(vars, struct fq_codel_flow, cvars); | |
267 | if (flow->head) { | |
268 | skb = dequeue_head(flow); | |
865ec552 | 269 | q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb); |
77f57761 | 270 | q->memory_usage -= skb->truesize; |
4b549a2e | 271 | sch->q.qlen--; |
79bdc4c8 | 272 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
4b549a2e ED |
273 | } |
274 | return skb; | |
275 | } | |
276 | ||
79bdc4c8 MK |
277 | static void drop_func(struct sk_buff *skb, void *ctx) |
278 | { | |
279 | struct Qdisc *sch = ctx; | |
280 | ||
520ac30f ED |
281 | kfree_skb(skb); |
282 | qdisc_qstats_drop(sch); | |
79bdc4c8 MK |
283 | } |
284 | ||
4b549a2e ED |
285 | static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch) |
286 | { | |
287 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
288 | struct sk_buff *skb; | |
289 | struct fq_codel_flow *flow; | |
290 | struct list_head *head; | |
291 | u32 prev_drop_count, prev_ecn_mark; | |
2ccccf5f | 292 | unsigned int prev_backlog; |
4b549a2e ED |
293 | |
294 | begin: | |
295 | head = &q->new_flows; | |
296 | if (list_empty(head)) { | |
297 | head = &q->old_flows; | |
298 | if (list_empty(head)) | |
299 | return NULL; | |
300 | } | |
301 | flow = list_first_entry(head, struct fq_codel_flow, flowchain); | |
302 | ||
303 | if (flow->deficit <= 0) { | |
304 | flow->deficit += q->quantum; | |
305 | list_move_tail(&flow->flowchain, &q->old_flows); | |
306 | goto begin; | |
307 | } | |
308 | ||
309 | prev_drop_count = q->cstats.drop_count; | |
310 | prev_ecn_mark = q->cstats.ecn_mark; | |
2ccccf5f | 311 | prev_backlog = sch->qstats.backlog; |
4b549a2e | 312 | |
79bdc4c8 MK |
313 | skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams, |
314 | &flow->cvars, &q->cstats, qdisc_pkt_len, | |
315 | codel_get_enqueue_time, drop_func, dequeue_func); | |
4b549a2e ED |
316 | |
317 | flow->dropped += q->cstats.drop_count - prev_drop_count; | |
318 | flow->dropped += q->cstats.ecn_mark - prev_ecn_mark; | |
319 | ||
320 | if (!skb) { | |
321 | /* force a pass through old_flows to prevent starvation */ | |
322 | if ((head == &q->new_flows) && !list_empty(&q->old_flows)) | |
323 | list_move_tail(&flow->flowchain, &q->old_flows); | |
324 | else | |
325 | list_del_init(&flow->flowchain); | |
326 | goto begin; | |
327 | } | |
328 | qdisc_bstats_update(sch, skb); | |
329 | flow->deficit -= qdisc_pkt_len(skb); | |
2ccccf5f | 330 | /* We cant call qdisc_tree_reduce_backlog() if our qlen is 0, |
4b549a2e ED |
331 | * or HTB crashes. Defer it for next round. |
332 | */ | |
333 | if (q->cstats.drop_count && sch->q.qlen) { | |
2ccccf5f WC |
334 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, |
335 | q->cstats.drop_len); | |
4b549a2e | 336 | q->cstats.drop_count = 0; |
2ccccf5f | 337 | q->cstats.drop_len = 0; |
4b549a2e ED |
338 | } |
339 | return skb; | |
340 | } | |
341 | ||
ece5d4c7 ED |
342 | static void fq_codel_flow_purge(struct fq_codel_flow *flow) |
343 | { | |
344 | rtnl_kfree_skbs(flow->head, flow->tail); | |
345 | flow->head = NULL; | |
346 | } | |
347 | ||
4b549a2e ED |
348 | static void fq_codel_reset(struct Qdisc *sch) |
349 | { | |
3d0e0af4 ED |
350 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
351 | int i; | |
4b549a2e | 352 | |
3d0e0af4 ED |
353 | INIT_LIST_HEAD(&q->new_flows); |
354 | INIT_LIST_HEAD(&q->old_flows); | |
355 | for (i = 0; i < q->flows_cnt; i++) { | |
356 | struct fq_codel_flow *flow = q->flows + i; | |
357 | ||
ece5d4c7 | 358 | fq_codel_flow_purge(flow); |
3d0e0af4 ED |
359 | INIT_LIST_HEAD(&flow->flowchain); |
360 | codel_vars_init(&flow->cvars); | |
361 | } | |
362 | memset(q->backlogs, 0, q->flows_cnt * sizeof(u32)); | |
363 | sch->q.qlen = 0; | |
ece5d4c7 | 364 | sch->qstats.backlog = 0; |
77f57761 | 365 | q->memory_usage = 0; |
4b549a2e ED |
366 | } |
367 | ||
368 | static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = { | |
369 | [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 }, | |
370 | [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 }, | |
371 | [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 }, | |
372 | [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 }, | |
373 | [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 }, | |
374 | [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 }, | |
80ba92fa | 375 | [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 }, |
9d18562a | 376 | [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 }, |
95b58430 | 377 | [TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 }, |
4b549a2e ED |
378 | }; |
379 | ||
380 | static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt) | |
381 | { | |
382 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
383 | struct nlattr *tb[TCA_FQ_CODEL_MAX + 1]; | |
384 | int err; | |
385 | ||
386 | if (!opt) | |
387 | return -EINVAL; | |
388 | ||
389 | err = nla_parse_nested(tb, TCA_FQ_CODEL_MAX, opt, fq_codel_policy); | |
390 | if (err < 0) | |
391 | return err; | |
392 | if (tb[TCA_FQ_CODEL_FLOWS]) { | |
393 | if (q->flows) | |
394 | return -EINVAL; | |
395 | q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]); | |
396 | if (!q->flows_cnt || | |
397 | q->flows_cnt > 65536) | |
398 | return -EINVAL; | |
399 | } | |
400 | sch_tree_lock(sch); | |
401 | ||
402 | if (tb[TCA_FQ_CODEL_TARGET]) { | |
403 | u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]); | |
404 | ||
405 | q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT; | |
406 | } | |
407 | ||
80ba92fa ED |
408 | if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) { |
409 | u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]); | |
410 | ||
411 | q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT; | |
412 | } | |
413 | ||
4b549a2e ED |
414 | if (tb[TCA_FQ_CODEL_INTERVAL]) { |
415 | u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]); | |
416 | ||
417 | q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT; | |
418 | } | |
419 | ||
420 | if (tb[TCA_FQ_CODEL_LIMIT]) | |
421 | sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]); | |
422 | ||
423 | if (tb[TCA_FQ_CODEL_ECN]) | |
424 | q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]); | |
425 | ||
426 | if (tb[TCA_FQ_CODEL_QUANTUM]) | |
427 | q->quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM])); | |
428 | ||
9d18562a ED |
429 | if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]) |
430 | q->drop_batch_size = min(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])); | |
431 | ||
95b58430 ED |
432 | if (tb[TCA_FQ_CODEL_MEMORY_LIMIT]) |
433 | q->memory_limit = min(1U << 31, nla_get_u32(tb[TCA_FQ_CODEL_MEMORY_LIMIT])); | |
434 | ||
435 | while (sch->q.qlen > sch->limit || | |
436 | q->memory_usage > q->memory_limit) { | |
4b549a2e ED |
437 | struct sk_buff *skb = fq_codel_dequeue(sch); |
438 | ||
2ccccf5f | 439 | q->cstats.drop_len += qdisc_pkt_len(skb); |
ece5d4c7 | 440 | rtnl_kfree_skbs(skb, skb); |
4b549a2e ED |
441 | q->cstats.drop_count++; |
442 | } | |
2ccccf5f | 443 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len); |
4b549a2e | 444 | q->cstats.drop_count = 0; |
2ccccf5f | 445 | q->cstats.drop_len = 0; |
4b549a2e ED |
446 | |
447 | sch_tree_unlock(sch); | |
448 | return 0; | |
449 | } | |
450 | ||
451 | static void *fq_codel_zalloc(size_t sz) | |
452 | { | |
453 | void *ptr = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN); | |
454 | ||
455 | if (!ptr) | |
456 | ptr = vzalloc(sz); | |
457 | return ptr; | |
458 | } | |
459 | ||
460 | static void fq_codel_free(void *addr) | |
461 | { | |
4cb28970 | 462 | kvfree(addr); |
4b549a2e ED |
463 | } |
464 | ||
465 | static void fq_codel_destroy(struct Qdisc *sch) | |
466 | { | |
467 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
468 | ||
469 | tcf_destroy_chain(&q->filter_list); | |
470 | fq_codel_free(q->backlogs); | |
471 | fq_codel_free(q->flows); | |
472 | } | |
473 | ||
474 | static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt) | |
475 | { | |
476 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
477 | int i; | |
478 | ||
479 | sch->limit = 10*1024; | |
480 | q->flows_cnt = 1024; | |
95b58430 | 481 | q->memory_limit = 32 << 20; /* 32 MBytes */ |
9d18562a | 482 | q->drop_batch_size = 64; |
4b549a2e | 483 | q->quantum = psched_mtu(qdisc_dev(sch)); |
63862b5b | 484 | q->perturbation = prandom_u32(); |
4b549a2e ED |
485 | INIT_LIST_HEAD(&q->new_flows); |
486 | INIT_LIST_HEAD(&q->old_flows); | |
79bdc4c8 | 487 | codel_params_init(&q->cparams); |
4b549a2e ED |
488 | codel_stats_init(&q->cstats); |
489 | q->cparams.ecn = true; | |
79bdc4c8 | 490 | q->cparams.mtu = psched_mtu(qdisc_dev(sch)); |
4b549a2e ED |
491 | |
492 | if (opt) { | |
493 | int err = fq_codel_change(sch, opt); | |
494 | if (err) | |
495 | return err; | |
496 | } | |
497 | ||
498 | if (!q->flows) { | |
499 | q->flows = fq_codel_zalloc(q->flows_cnt * | |
500 | sizeof(struct fq_codel_flow)); | |
501 | if (!q->flows) | |
502 | return -ENOMEM; | |
503 | q->backlogs = fq_codel_zalloc(q->flows_cnt * sizeof(u32)); | |
504 | if (!q->backlogs) { | |
505 | fq_codel_free(q->flows); | |
506 | return -ENOMEM; | |
507 | } | |
508 | for (i = 0; i < q->flows_cnt; i++) { | |
509 | struct fq_codel_flow *flow = q->flows + i; | |
510 | ||
511 | INIT_LIST_HEAD(&flow->flowchain); | |
b379135c | 512 | codel_vars_init(&flow->cvars); |
4b549a2e ED |
513 | } |
514 | } | |
515 | if (sch->limit >= 1) | |
516 | sch->flags |= TCQ_F_CAN_BYPASS; | |
517 | else | |
518 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
519 | return 0; | |
520 | } | |
521 | ||
522 | static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb) | |
523 | { | |
524 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
525 | struct nlattr *opts; | |
526 | ||
527 | opts = nla_nest_start(skb, TCA_OPTIONS); | |
528 | if (opts == NULL) | |
529 | goto nla_put_failure; | |
530 | ||
531 | if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET, | |
532 | codel_time_to_us(q->cparams.target)) || | |
533 | nla_put_u32(skb, TCA_FQ_CODEL_LIMIT, | |
534 | sch->limit) || | |
535 | nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL, | |
536 | codel_time_to_us(q->cparams.interval)) || | |
537 | nla_put_u32(skb, TCA_FQ_CODEL_ECN, | |
538 | q->cparams.ecn) || | |
539 | nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM, | |
540 | q->quantum) || | |
9d18562a ED |
541 | nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE, |
542 | q->drop_batch_size) || | |
95b58430 ED |
543 | nla_put_u32(skb, TCA_FQ_CODEL_MEMORY_LIMIT, |
544 | q->memory_limit) || | |
4b549a2e ED |
545 | nla_put_u32(skb, TCA_FQ_CODEL_FLOWS, |
546 | q->flows_cnt)) | |
547 | goto nla_put_failure; | |
548 | ||
80ba92fa ED |
549 | if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD && |
550 | nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD, | |
551 | codel_time_to_us(q->cparams.ce_threshold))) | |
552 | goto nla_put_failure; | |
553 | ||
d59b7d80 | 554 | return nla_nest_end(skb, opts); |
4b549a2e ED |
555 | |
556 | nla_put_failure: | |
557 | return -1; | |
558 | } | |
559 | ||
560 | static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
561 | { | |
562 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
563 | struct tc_fq_codel_xstats st = { | |
564 | .type = TCA_FQ_CODEL_XSTATS_QDISC, | |
4b549a2e ED |
565 | }; |
566 | struct list_head *pos; | |
567 | ||
669d67bf SL |
568 | st.qdisc_stats.maxpacket = q->cstats.maxpacket; |
569 | st.qdisc_stats.drop_overlimit = q->drop_overlimit; | |
570 | st.qdisc_stats.ecn_mark = q->cstats.ecn_mark; | |
571 | st.qdisc_stats.new_flow_count = q->new_flow_count; | |
80ba92fa | 572 | st.qdisc_stats.ce_mark = q->cstats.ce_mark; |
95b58430 ED |
573 | st.qdisc_stats.memory_usage = q->memory_usage; |
574 | st.qdisc_stats.drop_overmemory = q->drop_overmemory; | |
669d67bf | 575 | |
edb09eb1 | 576 | sch_tree_lock(sch); |
4b549a2e ED |
577 | list_for_each(pos, &q->new_flows) |
578 | st.qdisc_stats.new_flows_len++; | |
579 | ||
580 | list_for_each(pos, &q->old_flows) | |
581 | st.qdisc_stats.old_flows_len++; | |
edb09eb1 | 582 | sch_tree_unlock(sch); |
4b549a2e ED |
583 | |
584 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
585 | } | |
586 | ||
587 | static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg) | |
588 | { | |
589 | return NULL; | |
590 | } | |
591 | ||
592 | static unsigned long fq_codel_get(struct Qdisc *sch, u32 classid) | |
593 | { | |
594 | return 0; | |
595 | } | |
596 | ||
597 | static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent, | |
598 | u32 classid) | |
599 | { | |
600 | /* we cannot bypass queue discipline anymore */ | |
601 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
602 | return 0; | |
603 | } | |
604 | ||
605 | static void fq_codel_put(struct Qdisc *q, unsigned long cl) | |
606 | { | |
607 | } | |
608 | ||
25d8c0d5 JF |
609 | static struct tcf_proto __rcu **fq_codel_find_tcf(struct Qdisc *sch, |
610 | unsigned long cl) | |
4b549a2e ED |
611 | { |
612 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
613 | ||
614 | if (cl) | |
615 | return NULL; | |
616 | return &q->filter_list; | |
617 | } | |
618 | ||
619 | static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl, | |
620 | struct sk_buff *skb, struct tcmsg *tcm) | |
621 | { | |
622 | tcm->tcm_handle |= TC_H_MIN(cl); | |
623 | return 0; | |
624 | } | |
625 | ||
626 | static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl, | |
627 | struct gnet_dump *d) | |
628 | { | |
629 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
630 | u32 idx = cl - 1; | |
631 | struct gnet_stats_queue qs = { 0 }; | |
632 | struct tc_fq_codel_xstats xstats; | |
633 | ||
634 | if (idx < q->flows_cnt) { | |
635 | const struct fq_codel_flow *flow = &q->flows[idx]; | |
edb09eb1 | 636 | const struct sk_buff *skb; |
4b549a2e ED |
637 | |
638 | memset(&xstats, 0, sizeof(xstats)); | |
639 | xstats.type = TCA_FQ_CODEL_XSTATS_CLASS; | |
640 | xstats.class_stats.deficit = flow->deficit; | |
641 | xstats.class_stats.ldelay = | |
642 | codel_time_to_us(flow->cvars.ldelay); | |
643 | xstats.class_stats.count = flow->cvars.count; | |
644 | xstats.class_stats.lastcount = flow->cvars.lastcount; | |
645 | xstats.class_stats.dropping = flow->cvars.dropping; | |
646 | if (flow->cvars.dropping) { | |
647 | codel_tdiff_t delta = flow->cvars.drop_next - | |
648 | codel_get_time(); | |
649 | ||
650 | xstats.class_stats.drop_next = (delta >= 0) ? | |
651 | codel_time_to_us(delta) : | |
652 | -codel_time_to_us(-delta); | |
653 | } | |
edb09eb1 ED |
654 | if (flow->head) { |
655 | sch_tree_lock(sch); | |
656 | skb = flow->head; | |
657 | while (skb) { | |
658 | qs.qlen++; | |
659 | skb = skb->next; | |
660 | } | |
661 | sch_tree_unlock(sch); | |
4b549a2e ED |
662 | } |
663 | qs.backlog = q->backlogs[idx]; | |
664 | qs.drops = flow->dropped; | |
665 | } | |
aafddbf0 | 666 | if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0) |
4b549a2e ED |
667 | return -1; |
668 | if (idx < q->flows_cnt) | |
669 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
670 | return 0; | |
671 | } | |
672 | ||
673 | static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
674 | { | |
675 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
676 | unsigned int i; | |
677 | ||
678 | if (arg->stop) | |
679 | return; | |
680 | ||
681 | for (i = 0; i < q->flows_cnt; i++) { | |
682 | if (list_empty(&q->flows[i].flowchain) || | |
683 | arg->count < arg->skip) { | |
684 | arg->count++; | |
685 | continue; | |
686 | } | |
687 | if (arg->fn(sch, i + 1, arg) < 0) { | |
688 | arg->stop = 1; | |
689 | break; | |
690 | } | |
691 | arg->count++; | |
692 | } | |
693 | } | |
694 | ||
695 | static const struct Qdisc_class_ops fq_codel_class_ops = { | |
696 | .leaf = fq_codel_leaf, | |
697 | .get = fq_codel_get, | |
698 | .put = fq_codel_put, | |
699 | .tcf_chain = fq_codel_find_tcf, | |
700 | .bind_tcf = fq_codel_bind, | |
701 | .unbind_tcf = fq_codel_put, | |
702 | .dump = fq_codel_dump_class, | |
703 | .dump_stats = fq_codel_dump_class_stats, | |
704 | .walk = fq_codel_walk, | |
705 | }; | |
706 | ||
707 | static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = { | |
708 | .cl_ops = &fq_codel_class_ops, | |
709 | .id = "fq_codel", | |
710 | .priv_size = sizeof(struct fq_codel_sched_data), | |
711 | .enqueue = fq_codel_enqueue, | |
712 | .dequeue = fq_codel_dequeue, | |
713 | .peek = qdisc_peek_dequeued, | |
4b549a2e ED |
714 | .init = fq_codel_init, |
715 | .reset = fq_codel_reset, | |
716 | .destroy = fq_codel_destroy, | |
717 | .change = fq_codel_change, | |
718 | .dump = fq_codel_dump, | |
719 | .dump_stats = fq_codel_dump_stats, | |
720 | .owner = THIS_MODULE, | |
721 | }; | |
722 | ||
723 | static int __init fq_codel_module_init(void) | |
724 | { | |
725 | return register_qdisc(&fq_codel_qdisc_ops); | |
726 | } | |
727 | ||
728 | static void __exit fq_codel_module_exit(void) | |
729 | { | |
730 | unregister_qdisc(&fq_codel_qdisc_ops); | |
731 | } | |
732 | ||
733 | module_init(fq_codel_module_init) | |
734 | module_exit(fq_codel_module_exit) | |
735 | MODULE_AUTHOR("Eric Dumazet"); | |
736 | MODULE_LICENSE("GPL"); |