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 ED |
26 | #include <net/codel.h> |
27 | ||
28 | /* Fair Queue CoDel. | |
29 | * | |
30 | * Principles : | |
31 | * Packets are classified (internal classifier or external) on flows. | |
32 | * This is a Stochastic model (as we use a hash, several flows | |
33 | * might be hashed on same slot) | |
34 | * Each flow has a CoDel managed queue. | |
35 | * Flows are linked onto two (Round Robin) lists, | |
36 | * so that new flows have priority on old ones. | |
37 | * | |
38 | * For a given flow, packets are not reordered (CoDel uses a FIFO) | |
39 | * head drops only. | |
40 | * ECN capability is on by default. | |
41 | * Low memory footprint (64 bytes per flow) | |
42 | */ | |
43 | ||
44 | struct fq_codel_flow { | |
45 | struct sk_buff *head; | |
46 | struct sk_buff *tail; | |
47 | struct list_head flowchain; | |
48 | int deficit; | |
49 | u32 dropped; /* number of drops (or ECN marks) on this flow */ | |
50 | struct codel_vars cvars; | |
51 | }; /* please try to keep this structure <= 64 bytes */ | |
52 | ||
53 | struct fq_codel_sched_data { | |
25d8c0d5 | 54 | struct tcf_proto __rcu *filter_list; /* optional external classifier */ |
4b549a2e ED |
55 | struct fq_codel_flow *flows; /* Flows table [flows_cnt] */ |
56 | u32 *backlogs; /* backlog table [flows_cnt] */ | |
57 | u32 flows_cnt; /* number of flows */ | |
58 | u32 perturbation; /* hash perturbation */ | |
59 | u32 quantum; /* psched_mtu(qdisc_dev(sch)); */ | |
60 | struct codel_params cparams; | |
61 | struct codel_stats cstats; | |
62 | u32 drop_overlimit; | |
63 | u32 new_flow_count; | |
64 | ||
65 | struct list_head new_flows; /* list of new flows */ | |
66 | struct list_head old_flows; /* list of old flows */ | |
67 | }; | |
68 | ||
69 | static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q, | |
342db221 | 70 | struct sk_buff *skb) |
4b549a2e | 71 | { |
342db221 | 72 | u32 hash = skb_get_hash_perturb(skb, q->perturbation); |
8fc54f68 DB |
73 | |
74 | return reciprocal_scale(hash, q->flows_cnt); | |
4b549a2e ED |
75 | } |
76 | ||
77 | static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch, | |
78 | int *qerr) | |
79 | { | |
80 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
25d8c0d5 | 81 | struct tcf_proto *filter; |
4b549a2e ED |
82 | struct tcf_result res; |
83 | int result; | |
84 | ||
85 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
86 | TC_H_MIN(skb->priority) > 0 && | |
87 | TC_H_MIN(skb->priority) <= q->flows_cnt) | |
88 | return TC_H_MIN(skb->priority); | |
89 | ||
69204cf7 | 90 | filter = rcu_dereference_bh(q->filter_list); |
25d8c0d5 | 91 | if (!filter) |
4b549a2e ED |
92 | return fq_codel_hash(q, skb) + 1; |
93 | ||
94 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
3b3ae880 | 95 | result = tc_classify(skb, filter, &res, false); |
4b549a2e ED |
96 | if (result >= 0) { |
97 | #ifdef CONFIG_NET_CLS_ACT | |
98 | switch (result) { | |
99 | case TC_ACT_STOLEN: | |
100 | case TC_ACT_QUEUED: | |
101 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; | |
102 | case TC_ACT_SHOT: | |
103 | return 0; | |
104 | } | |
105 | #endif | |
106 | if (TC_H_MIN(res.classid) <= q->flows_cnt) | |
107 | return TC_H_MIN(res.classid); | |
108 | } | |
109 | return 0; | |
110 | } | |
111 | ||
112 | /* helper functions : might be changed when/if skb use a standard list_head */ | |
113 | ||
114 | /* remove one skb from head of slot queue */ | |
115 | static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow) | |
116 | { | |
117 | struct sk_buff *skb = flow->head; | |
118 | ||
119 | flow->head = skb->next; | |
120 | skb->next = NULL; | |
121 | return skb; | |
122 | } | |
123 | ||
124 | /* add skb to flow queue (tail add) */ | |
125 | static inline void flow_queue_add(struct fq_codel_flow *flow, | |
126 | struct sk_buff *skb) | |
127 | { | |
128 | if (flow->head == NULL) | |
129 | flow->head = skb; | |
130 | else | |
131 | flow->tail->next = skb; | |
132 | flow->tail = skb; | |
133 | skb->next = NULL; | |
134 | } | |
135 | ||
136 | static unsigned int fq_codel_drop(struct Qdisc *sch) | |
137 | { | |
138 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
139 | struct sk_buff *skb; | |
140 | unsigned int maxbacklog = 0, idx = 0, i, len; | |
141 | struct fq_codel_flow *flow; | |
142 | ||
143 | /* Queue is full! Find the fat flow and drop packet from it. | |
144 | * This might sound expensive, but with 1024 flows, we scan | |
145 | * 4KB of memory, and we dont need to handle a complex tree | |
146 | * in fast path (packet queue/enqueue) with many cache misses. | |
147 | */ | |
148 | for (i = 0; i < q->flows_cnt; i++) { | |
149 | if (q->backlogs[i] > maxbacklog) { | |
150 | maxbacklog = q->backlogs[i]; | |
151 | idx = i; | |
152 | } | |
153 | } | |
154 | flow = &q->flows[idx]; | |
155 | skb = dequeue_head(flow); | |
156 | len = qdisc_pkt_len(skb); | |
157 | q->backlogs[idx] -= len; | |
4b549a2e | 158 | sch->q.qlen--; |
25331d6c JF |
159 | qdisc_qstats_drop(sch); |
160 | qdisc_qstats_backlog_dec(sch, skb); | |
052cbda4 | 161 | kfree_skb(skb); |
4b549a2e ED |
162 | flow->dropped++; |
163 | return idx; | |
164 | } | |
165 | ||
c0afd9ce WC |
166 | static unsigned int fq_codel_qdisc_drop(struct Qdisc *sch) |
167 | { | |
168 | unsigned int prev_backlog; | |
169 | ||
170 | prev_backlog = sch->qstats.backlog; | |
171 | fq_codel_drop(sch); | |
172 | return prev_backlog - sch->qstats.backlog; | |
173 | } | |
174 | ||
4b549a2e ED |
175 | static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
176 | { | |
177 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
2ccccf5f | 178 | unsigned int idx, prev_backlog; |
4b549a2e ED |
179 | struct fq_codel_flow *flow; |
180 | int uninitialized_var(ret); | |
181 | ||
182 | idx = fq_codel_classify(skb, sch, &ret); | |
183 | if (idx == 0) { | |
184 | if (ret & __NET_XMIT_BYPASS) | |
25331d6c | 185 | qdisc_qstats_drop(sch); |
4b549a2e ED |
186 | kfree_skb(skb); |
187 | return ret; | |
188 | } | |
189 | idx--; | |
190 | ||
191 | codel_set_enqueue_time(skb); | |
192 | flow = &q->flows[idx]; | |
193 | flow_queue_add(flow, skb); | |
194 | q->backlogs[idx] += qdisc_pkt_len(skb); | |
25331d6c | 195 | qdisc_qstats_backlog_inc(sch, skb); |
4b549a2e ED |
196 | |
197 | if (list_empty(&flow->flowchain)) { | |
198 | list_add_tail(&flow->flowchain, &q->new_flows); | |
4b549a2e ED |
199 | q->new_flow_count++; |
200 | flow->deficit = q->quantum; | |
201 | flow->dropped = 0; | |
202 | } | |
cd68ddd4 | 203 | if (++sch->q.qlen <= sch->limit) |
4b549a2e ED |
204 | return NET_XMIT_SUCCESS; |
205 | ||
2ccccf5f | 206 | prev_backlog = sch->qstats.backlog; |
4b549a2e ED |
207 | q->drop_overlimit++; |
208 | /* Return Congestion Notification only if we dropped a packet | |
209 | * from this flow. | |
210 | */ | |
211 | if (fq_codel_drop(sch) == idx) | |
212 | return NET_XMIT_CN; | |
213 | ||
214 | /* As we dropped a packet, better let upper stack know this */ | |
2ccccf5f | 215 | qdisc_tree_reduce_backlog(sch, 1, prev_backlog - sch->qstats.backlog); |
4b549a2e ED |
216 | return NET_XMIT_SUCCESS; |
217 | } | |
218 | ||
219 | /* This is the specific function called from codel_dequeue() | |
220 | * to dequeue a packet from queue. Note: backlog is handled in | |
221 | * codel, we dont need to reduce it here. | |
222 | */ | |
79bdc4c8 | 223 | static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx) |
4b549a2e | 224 | { |
79bdc4c8 | 225 | struct Qdisc *sch = ctx; |
865ec552 | 226 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
4b549a2e ED |
227 | struct fq_codel_flow *flow; |
228 | struct sk_buff *skb = NULL; | |
229 | ||
230 | flow = container_of(vars, struct fq_codel_flow, cvars); | |
231 | if (flow->head) { | |
232 | skb = dequeue_head(flow); | |
865ec552 | 233 | q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb); |
4b549a2e | 234 | sch->q.qlen--; |
79bdc4c8 | 235 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
4b549a2e ED |
236 | } |
237 | return skb; | |
238 | } | |
239 | ||
79bdc4c8 MK |
240 | static void drop_func(struct sk_buff *skb, void *ctx) |
241 | { | |
242 | struct Qdisc *sch = ctx; | |
243 | ||
244 | qdisc_drop(skb, sch); | |
245 | } | |
246 | ||
4b549a2e ED |
247 | static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch) |
248 | { | |
249 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
250 | struct sk_buff *skb; | |
251 | struct fq_codel_flow *flow; | |
252 | struct list_head *head; | |
253 | u32 prev_drop_count, prev_ecn_mark; | |
2ccccf5f | 254 | unsigned int prev_backlog; |
4b549a2e ED |
255 | |
256 | begin: | |
257 | head = &q->new_flows; | |
258 | if (list_empty(head)) { | |
259 | head = &q->old_flows; | |
260 | if (list_empty(head)) | |
261 | return NULL; | |
262 | } | |
263 | flow = list_first_entry(head, struct fq_codel_flow, flowchain); | |
264 | ||
265 | if (flow->deficit <= 0) { | |
266 | flow->deficit += q->quantum; | |
267 | list_move_tail(&flow->flowchain, &q->old_flows); | |
268 | goto begin; | |
269 | } | |
270 | ||
271 | prev_drop_count = q->cstats.drop_count; | |
272 | prev_ecn_mark = q->cstats.ecn_mark; | |
2ccccf5f | 273 | prev_backlog = sch->qstats.backlog; |
4b549a2e | 274 | |
79bdc4c8 MK |
275 | skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams, |
276 | &flow->cvars, &q->cstats, qdisc_pkt_len, | |
277 | codel_get_enqueue_time, drop_func, dequeue_func); | |
4b549a2e ED |
278 | |
279 | flow->dropped += q->cstats.drop_count - prev_drop_count; | |
280 | flow->dropped += q->cstats.ecn_mark - prev_ecn_mark; | |
281 | ||
282 | if (!skb) { | |
283 | /* force a pass through old_flows to prevent starvation */ | |
284 | if ((head == &q->new_flows) && !list_empty(&q->old_flows)) | |
285 | list_move_tail(&flow->flowchain, &q->old_flows); | |
286 | else | |
287 | list_del_init(&flow->flowchain); | |
288 | goto begin; | |
289 | } | |
290 | qdisc_bstats_update(sch, skb); | |
291 | flow->deficit -= qdisc_pkt_len(skb); | |
2ccccf5f | 292 | /* We cant call qdisc_tree_reduce_backlog() if our qlen is 0, |
4b549a2e ED |
293 | * or HTB crashes. Defer it for next round. |
294 | */ | |
295 | if (q->cstats.drop_count && sch->q.qlen) { | |
2ccccf5f WC |
296 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, |
297 | q->cstats.drop_len); | |
4b549a2e | 298 | q->cstats.drop_count = 0; |
2ccccf5f | 299 | q->cstats.drop_len = 0; |
4b549a2e ED |
300 | } |
301 | return skb; | |
302 | } | |
303 | ||
304 | static void fq_codel_reset(struct Qdisc *sch) | |
305 | { | |
3d0e0af4 ED |
306 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
307 | int i; | |
4b549a2e | 308 | |
3d0e0af4 ED |
309 | INIT_LIST_HEAD(&q->new_flows); |
310 | INIT_LIST_HEAD(&q->old_flows); | |
311 | for (i = 0; i < q->flows_cnt; i++) { | |
312 | struct fq_codel_flow *flow = q->flows + i; | |
313 | ||
314 | while (flow->head) { | |
315 | struct sk_buff *skb = dequeue_head(flow); | |
316 | ||
317 | qdisc_qstats_backlog_dec(sch, skb); | |
318 | kfree_skb(skb); | |
319 | } | |
320 | ||
321 | INIT_LIST_HEAD(&flow->flowchain); | |
322 | codel_vars_init(&flow->cvars); | |
323 | } | |
324 | memset(q->backlogs, 0, q->flows_cnt * sizeof(u32)); | |
325 | sch->q.qlen = 0; | |
4b549a2e ED |
326 | } |
327 | ||
328 | static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = { | |
329 | [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 }, | |
330 | [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 }, | |
331 | [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 }, | |
332 | [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 }, | |
333 | [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 }, | |
334 | [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 }, | |
80ba92fa | 335 | [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 }, |
4b549a2e ED |
336 | }; |
337 | ||
338 | static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt) | |
339 | { | |
340 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
341 | struct nlattr *tb[TCA_FQ_CODEL_MAX + 1]; | |
342 | int err; | |
343 | ||
344 | if (!opt) | |
345 | return -EINVAL; | |
346 | ||
347 | err = nla_parse_nested(tb, TCA_FQ_CODEL_MAX, opt, fq_codel_policy); | |
348 | if (err < 0) | |
349 | return err; | |
350 | if (tb[TCA_FQ_CODEL_FLOWS]) { | |
351 | if (q->flows) | |
352 | return -EINVAL; | |
353 | q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]); | |
354 | if (!q->flows_cnt || | |
355 | q->flows_cnt > 65536) | |
356 | return -EINVAL; | |
357 | } | |
358 | sch_tree_lock(sch); | |
359 | ||
360 | if (tb[TCA_FQ_CODEL_TARGET]) { | |
361 | u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]); | |
362 | ||
363 | q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT; | |
364 | } | |
365 | ||
80ba92fa ED |
366 | if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) { |
367 | u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]); | |
368 | ||
369 | q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT; | |
370 | } | |
371 | ||
4b549a2e ED |
372 | if (tb[TCA_FQ_CODEL_INTERVAL]) { |
373 | u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]); | |
374 | ||
375 | q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT; | |
376 | } | |
377 | ||
378 | if (tb[TCA_FQ_CODEL_LIMIT]) | |
379 | sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]); | |
380 | ||
381 | if (tb[TCA_FQ_CODEL_ECN]) | |
382 | q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]); | |
383 | ||
384 | if (tb[TCA_FQ_CODEL_QUANTUM]) | |
385 | q->quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM])); | |
386 | ||
387 | while (sch->q.qlen > sch->limit) { | |
388 | struct sk_buff *skb = fq_codel_dequeue(sch); | |
389 | ||
2ccccf5f | 390 | q->cstats.drop_len += qdisc_pkt_len(skb); |
4b549a2e ED |
391 | kfree_skb(skb); |
392 | q->cstats.drop_count++; | |
393 | } | |
2ccccf5f | 394 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len); |
4b549a2e | 395 | q->cstats.drop_count = 0; |
2ccccf5f | 396 | q->cstats.drop_len = 0; |
4b549a2e ED |
397 | |
398 | sch_tree_unlock(sch); | |
399 | return 0; | |
400 | } | |
401 | ||
402 | static void *fq_codel_zalloc(size_t sz) | |
403 | { | |
404 | void *ptr = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN); | |
405 | ||
406 | if (!ptr) | |
407 | ptr = vzalloc(sz); | |
408 | return ptr; | |
409 | } | |
410 | ||
411 | static void fq_codel_free(void *addr) | |
412 | { | |
4cb28970 | 413 | kvfree(addr); |
4b549a2e ED |
414 | } |
415 | ||
416 | static void fq_codel_destroy(struct Qdisc *sch) | |
417 | { | |
418 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
419 | ||
420 | tcf_destroy_chain(&q->filter_list); | |
421 | fq_codel_free(q->backlogs); | |
422 | fq_codel_free(q->flows); | |
423 | } | |
424 | ||
425 | static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt) | |
426 | { | |
427 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
428 | int i; | |
429 | ||
430 | sch->limit = 10*1024; | |
431 | q->flows_cnt = 1024; | |
432 | q->quantum = psched_mtu(qdisc_dev(sch)); | |
63862b5b | 433 | q->perturbation = prandom_u32(); |
4b549a2e ED |
434 | INIT_LIST_HEAD(&q->new_flows); |
435 | INIT_LIST_HEAD(&q->old_flows); | |
79bdc4c8 | 436 | codel_params_init(&q->cparams); |
4b549a2e ED |
437 | codel_stats_init(&q->cstats); |
438 | q->cparams.ecn = true; | |
79bdc4c8 | 439 | q->cparams.mtu = psched_mtu(qdisc_dev(sch)); |
4b549a2e ED |
440 | |
441 | if (opt) { | |
442 | int err = fq_codel_change(sch, opt); | |
443 | if (err) | |
444 | return err; | |
445 | } | |
446 | ||
447 | if (!q->flows) { | |
448 | q->flows = fq_codel_zalloc(q->flows_cnt * | |
449 | sizeof(struct fq_codel_flow)); | |
450 | if (!q->flows) | |
451 | return -ENOMEM; | |
452 | q->backlogs = fq_codel_zalloc(q->flows_cnt * sizeof(u32)); | |
453 | if (!q->backlogs) { | |
454 | fq_codel_free(q->flows); | |
455 | return -ENOMEM; | |
456 | } | |
457 | for (i = 0; i < q->flows_cnt; i++) { | |
458 | struct fq_codel_flow *flow = q->flows + i; | |
459 | ||
460 | INIT_LIST_HEAD(&flow->flowchain); | |
b379135c | 461 | codel_vars_init(&flow->cvars); |
4b549a2e ED |
462 | } |
463 | } | |
464 | if (sch->limit >= 1) | |
465 | sch->flags |= TCQ_F_CAN_BYPASS; | |
466 | else | |
467 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
468 | return 0; | |
469 | } | |
470 | ||
471 | static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb) | |
472 | { | |
473 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
474 | struct nlattr *opts; | |
475 | ||
476 | opts = nla_nest_start(skb, TCA_OPTIONS); | |
477 | if (opts == NULL) | |
478 | goto nla_put_failure; | |
479 | ||
480 | if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET, | |
481 | codel_time_to_us(q->cparams.target)) || | |
482 | nla_put_u32(skb, TCA_FQ_CODEL_LIMIT, | |
483 | sch->limit) || | |
484 | nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL, | |
485 | codel_time_to_us(q->cparams.interval)) || | |
486 | nla_put_u32(skb, TCA_FQ_CODEL_ECN, | |
487 | q->cparams.ecn) || | |
488 | nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM, | |
489 | q->quantum) || | |
490 | nla_put_u32(skb, TCA_FQ_CODEL_FLOWS, | |
491 | q->flows_cnt)) | |
492 | goto nla_put_failure; | |
493 | ||
80ba92fa ED |
494 | if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD && |
495 | nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD, | |
496 | codel_time_to_us(q->cparams.ce_threshold))) | |
497 | goto nla_put_failure; | |
498 | ||
d59b7d80 | 499 | return nla_nest_end(skb, opts); |
4b549a2e ED |
500 | |
501 | nla_put_failure: | |
502 | return -1; | |
503 | } | |
504 | ||
505 | static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
506 | { | |
507 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
508 | struct tc_fq_codel_xstats st = { | |
509 | .type = TCA_FQ_CODEL_XSTATS_QDISC, | |
4b549a2e ED |
510 | }; |
511 | struct list_head *pos; | |
512 | ||
669d67bf SL |
513 | st.qdisc_stats.maxpacket = q->cstats.maxpacket; |
514 | st.qdisc_stats.drop_overlimit = q->drop_overlimit; | |
515 | st.qdisc_stats.ecn_mark = q->cstats.ecn_mark; | |
516 | st.qdisc_stats.new_flow_count = q->new_flow_count; | |
80ba92fa | 517 | st.qdisc_stats.ce_mark = q->cstats.ce_mark; |
669d67bf | 518 | |
4b549a2e ED |
519 | list_for_each(pos, &q->new_flows) |
520 | st.qdisc_stats.new_flows_len++; | |
521 | ||
522 | list_for_each(pos, &q->old_flows) | |
523 | st.qdisc_stats.old_flows_len++; | |
524 | ||
525 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
526 | } | |
527 | ||
528 | static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg) | |
529 | { | |
530 | return NULL; | |
531 | } | |
532 | ||
533 | static unsigned long fq_codel_get(struct Qdisc *sch, u32 classid) | |
534 | { | |
535 | return 0; | |
536 | } | |
537 | ||
538 | static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent, | |
539 | u32 classid) | |
540 | { | |
541 | /* we cannot bypass queue discipline anymore */ | |
542 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
543 | return 0; | |
544 | } | |
545 | ||
546 | static void fq_codel_put(struct Qdisc *q, unsigned long cl) | |
547 | { | |
548 | } | |
549 | ||
25d8c0d5 JF |
550 | static struct tcf_proto __rcu **fq_codel_find_tcf(struct Qdisc *sch, |
551 | unsigned long cl) | |
4b549a2e ED |
552 | { |
553 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
554 | ||
555 | if (cl) | |
556 | return NULL; | |
557 | return &q->filter_list; | |
558 | } | |
559 | ||
560 | static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl, | |
561 | struct sk_buff *skb, struct tcmsg *tcm) | |
562 | { | |
563 | tcm->tcm_handle |= TC_H_MIN(cl); | |
564 | return 0; | |
565 | } | |
566 | ||
567 | static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl, | |
568 | struct gnet_dump *d) | |
569 | { | |
570 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
571 | u32 idx = cl - 1; | |
572 | struct gnet_stats_queue qs = { 0 }; | |
573 | struct tc_fq_codel_xstats xstats; | |
574 | ||
575 | if (idx < q->flows_cnt) { | |
576 | const struct fq_codel_flow *flow = &q->flows[idx]; | |
577 | const struct sk_buff *skb = flow->head; | |
578 | ||
579 | memset(&xstats, 0, sizeof(xstats)); | |
580 | xstats.type = TCA_FQ_CODEL_XSTATS_CLASS; | |
581 | xstats.class_stats.deficit = flow->deficit; | |
582 | xstats.class_stats.ldelay = | |
583 | codel_time_to_us(flow->cvars.ldelay); | |
584 | xstats.class_stats.count = flow->cvars.count; | |
585 | xstats.class_stats.lastcount = flow->cvars.lastcount; | |
586 | xstats.class_stats.dropping = flow->cvars.dropping; | |
587 | if (flow->cvars.dropping) { | |
588 | codel_tdiff_t delta = flow->cvars.drop_next - | |
589 | codel_get_time(); | |
590 | ||
591 | xstats.class_stats.drop_next = (delta >= 0) ? | |
592 | codel_time_to_us(delta) : | |
593 | -codel_time_to_us(-delta); | |
594 | } | |
595 | while (skb) { | |
596 | qs.qlen++; | |
597 | skb = skb->next; | |
598 | } | |
599 | qs.backlog = q->backlogs[idx]; | |
600 | qs.drops = flow->dropped; | |
601 | } | |
b0ab6f92 | 602 | if (gnet_stats_copy_queue(d, NULL, &qs, 0) < 0) |
4b549a2e ED |
603 | return -1; |
604 | if (idx < q->flows_cnt) | |
605 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
606 | return 0; | |
607 | } | |
608 | ||
609 | static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
610 | { | |
611 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
612 | unsigned int i; | |
613 | ||
614 | if (arg->stop) | |
615 | return; | |
616 | ||
617 | for (i = 0; i < q->flows_cnt; i++) { | |
618 | if (list_empty(&q->flows[i].flowchain) || | |
619 | arg->count < arg->skip) { | |
620 | arg->count++; | |
621 | continue; | |
622 | } | |
623 | if (arg->fn(sch, i + 1, arg) < 0) { | |
624 | arg->stop = 1; | |
625 | break; | |
626 | } | |
627 | arg->count++; | |
628 | } | |
629 | } | |
630 | ||
631 | static const struct Qdisc_class_ops fq_codel_class_ops = { | |
632 | .leaf = fq_codel_leaf, | |
633 | .get = fq_codel_get, | |
634 | .put = fq_codel_put, | |
635 | .tcf_chain = fq_codel_find_tcf, | |
636 | .bind_tcf = fq_codel_bind, | |
637 | .unbind_tcf = fq_codel_put, | |
638 | .dump = fq_codel_dump_class, | |
639 | .dump_stats = fq_codel_dump_class_stats, | |
640 | .walk = fq_codel_walk, | |
641 | }; | |
642 | ||
643 | static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = { | |
644 | .cl_ops = &fq_codel_class_ops, | |
645 | .id = "fq_codel", | |
646 | .priv_size = sizeof(struct fq_codel_sched_data), | |
647 | .enqueue = fq_codel_enqueue, | |
648 | .dequeue = fq_codel_dequeue, | |
649 | .peek = qdisc_peek_dequeued, | |
c0afd9ce | 650 | .drop = fq_codel_qdisc_drop, |
4b549a2e ED |
651 | .init = fq_codel_init, |
652 | .reset = fq_codel_reset, | |
653 | .destroy = fq_codel_destroy, | |
654 | .change = fq_codel_change, | |
655 | .dump = fq_codel_dump, | |
656 | .dump_stats = fq_codel_dump_stats, | |
657 | .owner = THIS_MODULE, | |
658 | }; | |
659 | ||
660 | static int __init fq_codel_module_init(void) | |
661 | { | |
662 | return register_qdisc(&fq_codel_qdisc_ops); | |
663 | } | |
664 | ||
665 | static void __exit fq_codel_module_exit(void) | |
666 | { | |
667 | unregister_qdisc(&fq_codel_qdisc_ops); | |
668 | } | |
669 | ||
670 | module_init(fq_codel_module_init) | |
671 | module_exit(fq_codel_module_exit) | |
672 | MODULE_AUTHOR("Eric Dumazet"); | |
673 | MODULE_LICENSE("GPL"); |