]>
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; |
4b549a2e ED |
63 | struct codel_params cparams; |
64 | struct codel_stats cstats; | |
65 | u32 drop_overlimit; | |
66 | u32 new_flow_count; | |
67 | ||
68 | struct list_head new_flows; /* list of new flows */ | |
69 | struct list_head old_flows; /* list of old flows */ | |
70 | }; | |
71 | ||
72 | static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q, | |
342db221 | 73 | struct sk_buff *skb) |
4b549a2e | 74 | { |
342db221 | 75 | u32 hash = skb_get_hash_perturb(skb, q->perturbation); |
8fc54f68 DB |
76 | |
77 | return reciprocal_scale(hash, q->flows_cnt); | |
4b549a2e ED |
78 | } |
79 | ||
80 | static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch, | |
81 | int *qerr) | |
82 | { | |
83 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
25d8c0d5 | 84 | struct tcf_proto *filter; |
4b549a2e ED |
85 | struct tcf_result res; |
86 | int result; | |
87 | ||
88 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
89 | TC_H_MIN(skb->priority) > 0 && | |
90 | TC_H_MIN(skb->priority) <= q->flows_cnt) | |
91 | return TC_H_MIN(skb->priority); | |
92 | ||
69204cf7 | 93 | filter = rcu_dereference_bh(q->filter_list); |
25d8c0d5 | 94 | if (!filter) |
4b549a2e ED |
95 | return fq_codel_hash(q, skb) + 1; |
96 | ||
97 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
3b3ae880 | 98 | result = tc_classify(skb, filter, &res, false); |
4b549a2e ED |
99 | if (result >= 0) { |
100 | #ifdef CONFIG_NET_CLS_ACT | |
101 | switch (result) { | |
102 | case TC_ACT_STOLEN: | |
103 | case TC_ACT_QUEUED: | |
104 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; | |
105 | case TC_ACT_SHOT: | |
106 | return 0; | |
107 | } | |
108 | #endif | |
109 | if (TC_H_MIN(res.classid) <= q->flows_cnt) | |
110 | return TC_H_MIN(res.classid); | |
111 | } | |
112 | return 0; | |
113 | } | |
114 | ||
115 | /* helper functions : might be changed when/if skb use a standard list_head */ | |
116 | ||
117 | /* remove one skb from head of slot queue */ | |
118 | static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow) | |
119 | { | |
120 | struct sk_buff *skb = flow->head; | |
121 | ||
122 | flow->head = skb->next; | |
123 | skb->next = NULL; | |
124 | return skb; | |
125 | } | |
126 | ||
127 | /* add skb to flow queue (tail add) */ | |
128 | static inline void flow_queue_add(struct fq_codel_flow *flow, | |
129 | struct sk_buff *skb) | |
130 | { | |
131 | if (flow->head == NULL) | |
132 | flow->head = skb; | |
133 | else | |
134 | flow->tail->next = skb; | |
135 | flow->tail = skb; | |
136 | skb->next = NULL; | |
137 | } | |
138 | ||
9d18562a | 139 | static unsigned int fq_codel_drop(struct Qdisc *sch, unsigned int max_packets) |
4b549a2e ED |
140 | { |
141 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
142 | struct sk_buff *skb; | |
143 | unsigned int maxbacklog = 0, idx = 0, i, len; | |
144 | struct fq_codel_flow *flow; | |
9d18562a | 145 | unsigned int threshold; |
4b549a2e | 146 | |
9d18562a | 147 | /* Queue is full! Find the fat flow and drop packet(s) from it. |
4b549a2e ED |
148 | * This might sound expensive, but with 1024 flows, we scan |
149 | * 4KB of memory, and we dont need to handle a complex tree | |
150 | * in fast path (packet queue/enqueue) with many cache misses. | |
9d18562a ED |
151 | * In stress mode, we'll try to drop 64 packets from the flow, |
152 | * amortizing this linear lookup to one cache line per drop. | |
4b549a2e ED |
153 | */ |
154 | for (i = 0; i < q->flows_cnt; i++) { | |
155 | if (q->backlogs[i] > maxbacklog) { | |
156 | maxbacklog = q->backlogs[i]; | |
157 | idx = i; | |
158 | } | |
159 | } | |
9d18562a ED |
160 | |
161 | /* Our goal is to drop half of this fat flow backlog */ | |
162 | threshold = maxbacklog >> 1; | |
163 | ||
4b549a2e | 164 | flow = &q->flows[idx]; |
9d18562a ED |
165 | len = 0; |
166 | i = 0; | |
167 | do { | |
168 | skb = dequeue_head(flow); | |
169 | len += qdisc_pkt_len(skb); | |
170 | kfree_skb(skb); | |
171 | } while (++i < max_packets && len < threshold); | |
172 | ||
173 | flow->dropped += i; | |
4b549a2e | 174 | q->backlogs[idx] -= len; |
9d18562a ED |
175 | sch->qstats.drops += i; |
176 | sch->qstats.backlog -= len; | |
177 | sch->q.qlen -= i; | |
4b549a2e ED |
178 | return idx; |
179 | } | |
180 | ||
c0afd9ce WC |
181 | static unsigned int fq_codel_qdisc_drop(struct Qdisc *sch) |
182 | { | |
183 | unsigned int prev_backlog; | |
184 | ||
185 | prev_backlog = sch->qstats.backlog; | |
9d18562a | 186 | fq_codel_drop(sch, 1U); |
c0afd9ce WC |
187 | return prev_backlog - sch->qstats.backlog; |
188 | } | |
189 | ||
4b549a2e ED |
190 | static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
191 | { | |
192 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
9d18562a | 193 | unsigned int idx, prev_backlog, prev_qlen; |
4b549a2e ED |
194 | struct fq_codel_flow *flow; |
195 | int uninitialized_var(ret); | |
196 | ||
197 | idx = fq_codel_classify(skb, sch, &ret); | |
198 | if (idx == 0) { | |
199 | if (ret & __NET_XMIT_BYPASS) | |
25331d6c | 200 | qdisc_qstats_drop(sch); |
4b549a2e ED |
201 | kfree_skb(skb); |
202 | return ret; | |
203 | } | |
204 | idx--; | |
205 | ||
206 | codel_set_enqueue_time(skb); | |
207 | flow = &q->flows[idx]; | |
208 | flow_queue_add(flow, skb); | |
209 | q->backlogs[idx] += qdisc_pkt_len(skb); | |
25331d6c | 210 | qdisc_qstats_backlog_inc(sch, skb); |
4b549a2e ED |
211 | |
212 | if (list_empty(&flow->flowchain)) { | |
213 | list_add_tail(&flow->flowchain, &q->new_flows); | |
4b549a2e ED |
214 | q->new_flow_count++; |
215 | flow->deficit = q->quantum; | |
216 | flow->dropped = 0; | |
217 | } | |
cd68ddd4 | 218 | if (++sch->q.qlen <= sch->limit) |
4b549a2e ED |
219 | return NET_XMIT_SUCCESS; |
220 | ||
2ccccf5f | 221 | prev_backlog = sch->qstats.backlog; |
9d18562a ED |
222 | prev_qlen = sch->q.qlen; |
223 | ||
224 | /* fq_codel_drop() is quite expensive, as it performs a linear search | |
225 | * in q->backlogs[] to find a fat flow. | |
226 | * So instead of dropping a single packet, drop half of its backlog | |
227 | * with a 64 packets limit to not add a too big cpu spike here. | |
4b549a2e | 228 | */ |
9d18562a ED |
229 | ret = fq_codel_drop(sch, q->drop_batch_size); |
230 | ||
231 | q->drop_overlimit += prev_qlen - sch->q.qlen; | |
4b549a2e | 232 | |
9d18562a ED |
233 | /* As we dropped packet(s), better let upper stack know this */ |
234 | qdisc_tree_reduce_backlog(sch, prev_qlen - sch->q.qlen, | |
235 | prev_backlog - sch->qstats.backlog); | |
236 | ||
237 | return ret == idx ? NET_XMIT_CN : NET_XMIT_SUCCESS; | |
4b549a2e ED |
238 | } |
239 | ||
240 | /* This is the specific function called from codel_dequeue() | |
241 | * to dequeue a packet from queue. Note: backlog is handled in | |
242 | * codel, we dont need to reduce it here. | |
243 | */ | |
79bdc4c8 | 244 | static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx) |
4b549a2e | 245 | { |
79bdc4c8 | 246 | struct Qdisc *sch = ctx; |
865ec552 | 247 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
4b549a2e ED |
248 | struct fq_codel_flow *flow; |
249 | struct sk_buff *skb = NULL; | |
250 | ||
251 | flow = container_of(vars, struct fq_codel_flow, cvars); | |
252 | if (flow->head) { | |
253 | skb = dequeue_head(flow); | |
865ec552 | 254 | q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb); |
4b549a2e | 255 | sch->q.qlen--; |
79bdc4c8 | 256 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
4b549a2e ED |
257 | } |
258 | return skb; | |
259 | } | |
260 | ||
79bdc4c8 MK |
261 | static void drop_func(struct sk_buff *skb, void *ctx) |
262 | { | |
263 | struct Qdisc *sch = ctx; | |
264 | ||
265 | qdisc_drop(skb, sch); | |
266 | } | |
267 | ||
4b549a2e ED |
268 | static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch) |
269 | { | |
270 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
271 | struct sk_buff *skb; | |
272 | struct fq_codel_flow *flow; | |
273 | struct list_head *head; | |
274 | u32 prev_drop_count, prev_ecn_mark; | |
2ccccf5f | 275 | unsigned int prev_backlog; |
4b549a2e ED |
276 | |
277 | begin: | |
278 | head = &q->new_flows; | |
279 | if (list_empty(head)) { | |
280 | head = &q->old_flows; | |
281 | if (list_empty(head)) | |
282 | return NULL; | |
283 | } | |
284 | flow = list_first_entry(head, struct fq_codel_flow, flowchain); | |
285 | ||
286 | if (flow->deficit <= 0) { | |
287 | flow->deficit += q->quantum; | |
288 | list_move_tail(&flow->flowchain, &q->old_flows); | |
289 | goto begin; | |
290 | } | |
291 | ||
292 | prev_drop_count = q->cstats.drop_count; | |
293 | prev_ecn_mark = q->cstats.ecn_mark; | |
2ccccf5f | 294 | prev_backlog = sch->qstats.backlog; |
4b549a2e | 295 | |
79bdc4c8 MK |
296 | skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams, |
297 | &flow->cvars, &q->cstats, qdisc_pkt_len, | |
298 | codel_get_enqueue_time, drop_func, dequeue_func); | |
4b549a2e ED |
299 | |
300 | flow->dropped += q->cstats.drop_count - prev_drop_count; | |
301 | flow->dropped += q->cstats.ecn_mark - prev_ecn_mark; | |
302 | ||
303 | if (!skb) { | |
304 | /* force a pass through old_flows to prevent starvation */ | |
305 | if ((head == &q->new_flows) && !list_empty(&q->old_flows)) | |
306 | list_move_tail(&flow->flowchain, &q->old_flows); | |
307 | else | |
308 | list_del_init(&flow->flowchain); | |
309 | goto begin; | |
310 | } | |
311 | qdisc_bstats_update(sch, skb); | |
312 | flow->deficit -= qdisc_pkt_len(skb); | |
2ccccf5f | 313 | /* We cant call qdisc_tree_reduce_backlog() if our qlen is 0, |
4b549a2e ED |
314 | * or HTB crashes. Defer it for next round. |
315 | */ | |
316 | if (q->cstats.drop_count && sch->q.qlen) { | |
2ccccf5f WC |
317 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, |
318 | q->cstats.drop_len); | |
4b549a2e | 319 | q->cstats.drop_count = 0; |
2ccccf5f | 320 | q->cstats.drop_len = 0; |
4b549a2e ED |
321 | } |
322 | return skb; | |
323 | } | |
324 | ||
325 | static void fq_codel_reset(struct Qdisc *sch) | |
326 | { | |
3d0e0af4 ED |
327 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
328 | int i; | |
4b549a2e | 329 | |
3d0e0af4 ED |
330 | INIT_LIST_HEAD(&q->new_flows); |
331 | INIT_LIST_HEAD(&q->old_flows); | |
332 | for (i = 0; i < q->flows_cnt; i++) { | |
333 | struct fq_codel_flow *flow = q->flows + i; | |
334 | ||
335 | while (flow->head) { | |
336 | struct sk_buff *skb = dequeue_head(flow); | |
337 | ||
338 | qdisc_qstats_backlog_dec(sch, skb); | |
339 | kfree_skb(skb); | |
340 | } | |
341 | ||
342 | INIT_LIST_HEAD(&flow->flowchain); | |
343 | codel_vars_init(&flow->cvars); | |
344 | } | |
345 | memset(q->backlogs, 0, q->flows_cnt * sizeof(u32)); | |
346 | sch->q.qlen = 0; | |
4b549a2e ED |
347 | } |
348 | ||
349 | static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = { | |
350 | [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 }, | |
351 | [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 }, | |
352 | [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 }, | |
353 | [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 }, | |
354 | [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 }, | |
355 | [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 }, | |
80ba92fa | 356 | [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 }, |
9d18562a | 357 | [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 }, |
4b549a2e ED |
358 | }; |
359 | ||
360 | static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt) | |
361 | { | |
362 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
363 | struct nlattr *tb[TCA_FQ_CODEL_MAX + 1]; | |
364 | int err; | |
365 | ||
366 | if (!opt) | |
367 | return -EINVAL; | |
368 | ||
369 | err = nla_parse_nested(tb, TCA_FQ_CODEL_MAX, opt, fq_codel_policy); | |
370 | if (err < 0) | |
371 | return err; | |
372 | if (tb[TCA_FQ_CODEL_FLOWS]) { | |
373 | if (q->flows) | |
374 | return -EINVAL; | |
375 | q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]); | |
376 | if (!q->flows_cnt || | |
377 | q->flows_cnt > 65536) | |
378 | return -EINVAL; | |
379 | } | |
380 | sch_tree_lock(sch); | |
381 | ||
382 | if (tb[TCA_FQ_CODEL_TARGET]) { | |
383 | u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]); | |
384 | ||
385 | q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT; | |
386 | } | |
387 | ||
80ba92fa ED |
388 | if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) { |
389 | u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]); | |
390 | ||
391 | q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT; | |
392 | } | |
393 | ||
4b549a2e ED |
394 | if (tb[TCA_FQ_CODEL_INTERVAL]) { |
395 | u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]); | |
396 | ||
397 | q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT; | |
398 | } | |
399 | ||
400 | if (tb[TCA_FQ_CODEL_LIMIT]) | |
401 | sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]); | |
402 | ||
403 | if (tb[TCA_FQ_CODEL_ECN]) | |
404 | q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]); | |
405 | ||
406 | if (tb[TCA_FQ_CODEL_QUANTUM]) | |
407 | q->quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM])); | |
408 | ||
9d18562a ED |
409 | if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]) |
410 | q->drop_batch_size = min(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])); | |
411 | ||
4b549a2e ED |
412 | while (sch->q.qlen > sch->limit) { |
413 | struct sk_buff *skb = fq_codel_dequeue(sch); | |
414 | ||
2ccccf5f | 415 | q->cstats.drop_len += qdisc_pkt_len(skb); |
4b549a2e ED |
416 | kfree_skb(skb); |
417 | q->cstats.drop_count++; | |
418 | } | |
2ccccf5f | 419 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len); |
4b549a2e | 420 | q->cstats.drop_count = 0; |
2ccccf5f | 421 | q->cstats.drop_len = 0; |
4b549a2e ED |
422 | |
423 | sch_tree_unlock(sch); | |
424 | return 0; | |
425 | } | |
426 | ||
427 | static void *fq_codel_zalloc(size_t sz) | |
428 | { | |
429 | void *ptr = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN); | |
430 | ||
431 | if (!ptr) | |
432 | ptr = vzalloc(sz); | |
433 | return ptr; | |
434 | } | |
435 | ||
436 | static void fq_codel_free(void *addr) | |
437 | { | |
4cb28970 | 438 | kvfree(addr); |
4b549a2e ED |
439 | } |
440 | ||
441 | static void fq_codel_destroy(struct Qdisc *sch) | |
442 | { | |
443 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
444 | ||
445 | tcf_destroy_chain(&q->filter_list); | |
446 | fq_codel_free(q->backlogs); | |
447 | fq_codel_free(q->flows); | |
448 | } | |
449 | ||
450 | static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt) | |
451 | { | |
452 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
453 | int i; | |
454 | ||
455 | sch->limit = 10*1024; | |
456 | q->flows_cnt = 1024; | |
9d18562a | 457 | q->drop_batch_size = 64; |
4b549a2e | 458 | q->quantum = psched_mtu(qdisc_dev(sch)); |
63862b5b | 459 | q->perturbation = prandom_u32(); |
4b549a2e ED |
460 | INIT_LIST_HEAD(&q->new_flows); |
461 | INIT_LIST_HEAD(&q->old_flows); | |
79bdc4c8 | 462 | codel_params_init(&q->cparams); |
4b549a2e ED |
463 | codel_stats_init(&q->cstats); |
464 | q->cparams.ecn = true; | |
79bdc4c8 | 465 | q->cparams.mtu = psched_mtu(qdisc_dev(sch)); |
4b549a2e ED |
466 | |
467 | if (opt) { | |
468 | int err = fq_codel_change(sch, opt); | |
469 | if (err) | |
470 | return err; | |
471 | } | |
472 | ||
473 | if (!q->flows) { | |
474 | q->flows = fq_codel_zalloc(q->flows_cnt * | |
475 | sizeof(struct fq_codel_flow)); | |
476 | if (!q->flows) | |
477 | return -ENOMEM; | |
478 | q->backlogs = fq_codel_zalloc(q->flows_cnt * sizeof(u32)); | |
479 | if (!q->backlogs) { | |
480 | fq_codel_free(q->flows); | |
481 | return -ENOMEM; | |
482 | } | |
483 | for (i = 0; i < q->flows_cnt; i++) { | |
484 | struct fq_codel_flow *flow = q->flows + i; | |
485 | ||
486 | INIT_LIST_HEAD(&flow->flowchain); | |
b379135c | 487 | codel_vars_init(&flow->cvars); |
4b549a2e ED |
488 | } |
489 | } | |
490 | if (sch->limit >= 1) | |
491 | sch->flags |= TCQ_F_CAN_BYPASS; | |
492 | else | |
493 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
494 | return 0; | |
495 | } | |
496 | ||
497 | static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb) | |
498 | { | |
499 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
500 | struct nlattr *opts; | |
501 | ||
502 | opts = nla_nest_start(skb, TCA_OPTIONS); | |
503 | if (opts == NULL) | |
504 | goto nla_put_failure; | |
505 | ||
506 | if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET, | |
507 | codel_time_to_us(q->cparams.target)) || | |
508 | nla_put_u32(skb, TCA_FQ_CODEL_LIMIT, | |
509 | sch->limit) || | |
510 | nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL, | |
511 | codel_time_to_us(q->cparams.interval)) || | |
512 | nla_put_u32(skb, TCA_FQ_CODEL_ECN, | |
513 | q->cparams.ecn) || | |
514 | nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM, | |
515 | q->quantum) || | |
9d18562a ED |
516 | nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE, |
517 | q->drop_batch_size) || | |
4b549a2e ED |
518 | nla_put_u32(skb, TCA_FQ_CODEL_FLOWS, |
519 | q->flows_cnt)) | |
520 | goto nla_put_failure; | |
521 | ||
80ba92fa ED |
522 | if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD && |
523 | nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD, | |
524 | codel_time_to_us(q->cparams.ce_threshold))) | |
525 | goto nla_put_failure; | |
526 | ||
d59b7d80 | 527 | return nla_nest_end(skb, opts); |
4b549a2e ED |
528 | |
529 | nla_put_failure: | |
530 | return -1; | |
531 | } | |
532 | ||
533 | static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
534 | { | |
535 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
536 | struct tc_fq_codel_xstats st = { | |
537 | .type = TCA_FQ_CODEL_XSTATS_QDISC, | |
4b549a2e ED |
538 | }; |
539 | struct list_head *pos; | |
540 | ||
669d67bf SL |
541 | st.qdisc_stats.maxpacket = q->cstats.maxpacket; |
542 | st.qdisc_stats.drop_overlimit = q->drop_overlimit; | |
543 | st.qdisc_stats.ecn_mark = q->cstats.ecn_mark; | |
544 | st.qdisc_stats.new_flow_count = q->new_flow_count; | |
80ba92fa | 545 | st.qdisc_stats.ce_mark = q->cstats.ce_mark; |
669d67bf | 546 | |
4b549a2e ED |
547 | list_for_each(pos, &q->new_flows) |
548 | st.qdisc_stats.new_flows_len++; | |
549 | ||
550 | list_for_each(pos, &q->old_flows) | |
551 | st.qdisc_stats.old_flows_len++; | |
552 | ||
553 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
554 | } | |
555 | ||
556 | static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg) | |
557 | { | |
558 | return NULL; | |
559 | } | |
560 | ||
561 | static unsigned long fq_codel_get(struct Qdisc *sch, u32 classid) | |
562 | { | |
563 | return 0; | |
564 | } | |
565 | ||
566 | static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent, | |
567 | u32 classid) | |
568 | { | |
569 | /* we cannot bypass queue discipline anymore */ | |
570 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
571 | return 0; | |
572 | } | |
573 | ||
574 | static void fq_codel_put(struct Qdisc *q, unsigned long cl) | |
575 | { | |
576 | } | |
577 | ||
25d8c0d5 JF |
578 | static struct tcf_proto __rcu **fq_codel_find_tcf(struct Qdisc *sch, |
579 | unsigned long cl) | |
4b549a2e ED |
580 | { |
581 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
582 | ||
583 | if (cl) | |
584 | return NULL; | |
585 | return &q->filter_list; | |
586 | } | |
587 | ||
588 | static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl, | |
589 | struct sk_buff *skb, struct tcmsg *tcm) | |
590 | { | |
591 | tcm->tcm_handle |= TC_H_MIN(cl); | |
592 | return 0; | |
593 | } | |
594 | ||
595 | static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl, | |
596 | struct gnet_dump *d) | |
597 | { | |
598 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
599 | u32 idx = cl - 1; | |
600 | struct gnet_stats_queue qs = { 0 }; | |
601 | struct tc_fq_codel_xstats xstats; | |
602 | ||
603 | if (idx < q->flows_cnt) { | |
604 | const struct fq_codel_flow *flow = &q->flows[idx]; | |
605 | const struct sk_buff *skb = flow->head; | |
606 | ||
607 | memset(&xstats, 0, sizeof(xstats)); | |
608 | xstats.type = TCA_FQ_CODEL_XSTATS_CLASS; | |
609 | xstats.class_stats.deficit = flow->deficit; | |
610 | xstats.class_stats.ldelay = | |
611 | codel_time_to_us(flow->cvars.ldelay); | |
612 | xstats.class_stats.count = flow->cvars.count; | |
613 | xstats.class_stats.lastcount = flow->cvars.lastcount; | |
614 | xstats.class_stats.dropping = flow->cvars.dropping; | |
615 | if (flow->cvars.dropping) { | |
616 | codel_tdiff_t delta = flow->cvars.drop_next - | |
617 | codel_get_time(); | |
618 | ||
619 | xstats.class_stats.drop_next = (delta >= 0) ? | |
620 | codel_time_to_us(delta) : | |
621 | -codel_time_to_us(-delta); | |
622 | } | |
623 | while (skb) { | |
624 | qs.qlen++; | |
625 | skb = skb->next; | |
626 | } | |
627 | qs.backlog = q->backlogs[idx]; | |
628 | qs.drops = flow->dropped; | |
629 | } | |
b0ab6f92 | 630 | if (gnet_stats_copy_queue(d, NULL, &qs, 0) < 0) |
4b549a2e ED |
631 | return -1; |
632 | if (idx < q->flows_cnt) | |
633 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
634 | return 0; | |
635 | } | |
636 | ||
637 | static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
638 | { | |
639 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
640 | unsigned int i; | |
641 | ||
642 | if (arg->stop) | |
643 | return; | |
644 | ||
645 | for (i = 0; i < q->flows_cnt; i++) { | |
646 | if (list_empty(&q->flows[i].flowchain) || | |
647 | arg->count < arg->skip) { | |
648 | arg->count++; | |
649 | continue; | |
650 | } | |
651 | if (arg->fn(sch, i + 1, arg) < 0) { | |
652 | arg->stop = 1; | |
653 | break; | |
654 | } | |
655 | arg->count++; | |
656 | } | |
657 | } | |
658 | ||
659 | static const struct Qdisc_class_ops fq_codel_class_ops = { | |
660 | .leaf = fq_codel_leaf, | |
661 | .get = fq_codel_get, | |
662 | .put = fq_codel_put, | |
663 | .tcf_chain = fq_codel_find_tcf, | |
664 | .bind_tcf = fq_codel_bind, | |
665 | .unbind_tcf = fq_codel_put, | |
666 | .dump = fq_codel_dump_class, | |
667 | .dump_stats = fq_codel_dump_class_stats, | |
668 | .walk = fq_codel_walk, | |
669 | }; | |
670 | ||
671 | static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = { | |
672 | .cl_ops = &fq_codel_class_ops, | |
673 | .id = "fq_codel", | |
674 | .priv_size = sizeof(struct fq_codel_sched_data), | |
675 | .enqueue = fq_codel_enqueue, | |
676 | .dequeue = fq_codel_dequeue, | |
677 | .peek = qdisc_peek_dequeued, | |
c0afd9ce | 678 | .drop = fq_codel_qdisc_drop, |
4b549a2e ED |
679 | .init = fq_codel_init, |
680 | .reset = fq_codel_reset, | |
681 | .destroy = fq_codel_destroy, | |
682 | .change = fq_codel_change, | |
683 | .dump = fq_codel_dump, | |
684 | .dump_stats = fq_codel_dump_stats, | |
685 | .owner = THIS_MODULE, | |
686 | }; | |
687 | ||
688 | static int __init fq_codel_module_init(void) | |
689 | { | |
690 | return register_qdisc(&fq_codel_qdisc_ops); | |
691 | } | |
692 | ||
693 | static void __exit fq_codel_module_exit(void) | |
694 | { | |
695 | unregister_qdisc(&fq_codel_qdisc_ops); | |
696 | } | |
697 | ||
698 | module_init(fq_codel_module_init) | |
699 | module_exit(fq_codel_module_exit) | |
700 | MODULE_AUTHOR("Eric Dumazet"); | |
701 | MODULE_LICENSE("GPL"); |