Commit | Line | Data |
---|---|---|
ec97ecf1 MT |
1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* Flow Queue PIE discipline | |
3 | * | |
4 | * Copyright (C) 2019 Mohit P. Tahiliani <tahiliani@nitk.edu.in> | |
5 | * Copyright (C) 2019 Sachin D. Patil <sdp.sachin@gmail.com> | |
6 | * Copyright (C) 2019 V. Saicharan <vsaicharan1998@gmail.com> | |
7 | * Copyright (C) 2019 Mohit Bhasi <mohitbhasi1998@gmail.com> | |
8 | * Copyright (C) 2019 Leslie Monis <lesliemonis@gmail.com> | |
9 | * Copyright (C) 2019 Gautam Ramakrishnan <gautamramk@gmail.com> | |
10 | */ | |
11 | ||
12 | #include <linux/jhash.h> | |
9474c62a | 13 | #include <linux/module.h> |
ec97ecf1 MT |
14 | #include <linux/sizes.h> |
15 | #include <linux/vmalloc.h> | |
16 | #include <net/pkt_cls.h> | |
17 | #include <net/pie.h> | |
18 | ||
19 | /* Flow Queue PIE | |
20 | * | |
21 | * Principles: | |
22 | * - Packets are classified on flows. | |
23 | * - This is a Stochastic model (as we use a hash, several flows might | |
24 | * be hashed to the same slot) | |
25 | * - Each flow has a PIE managed queue. | |
26 | * - Flows are linked onto two (Round Robin) lists, | |
27 | * so that new flows have priority on old ones. | |
28 | * - For a given flow, packets are not reordered. | |
29 | * - Drops during enqueue only. | |
30 | * - ECN capability is off by default. | |
31 | * - ECN threshold (if ECN is enabled) is at 10% by default. | |
32 | * - Uses timestamps to calculate queue delay by default. | |
33 | */ | |
34 | ||
35 | /** | |
36 | * struct fq_pie_flow - contains data for each flow | |
37 | * @vars: pie vars associated with the flow | |
38 | * @deficit: number of remaining byte credits | |
39 | * @backlog: size of data in the flow | |
40 | * @qlen: number of packets in the flow | |
41 | * @flowchain: flowchain for the flow | |
42 | * @head: first packet in the flow | |
43 | * @tail: last packet in the flow | |
44 | */ | |
45 | struct fq_pie_flow { | |
46 | struct pie_vars vars; | |
47 | s32 deficit; | |
48 | u32 backlog; | |
49 | u32 qlen; | |
50 | struct list_head flowchain; | |
51 | struct sk_buff *head; | |
52 | struct sk_buff *tail; | |
53 | }; | |
54 | ||
55 | struct fq_pie_sched_data { | |
56 | struct tcf_proto __rcu *filter_list; /* optional external classifier */ | |
57 | struct tcf_block *block; | |
58 | struct fq_pie_flow *flows; | |
59 | struct Qdisc *sch; | |
60 | struct list_head old_flows; | |
61 | struct list_head new_flows; | |
62 | struct pie_params p_params; | |
63 | u32 ecn_prob; | |
64 | u32 flows_cnt; | |
8c21ab1b | 65 | u32 flows_cursor; |
ec97ecf1 MT |
66 | u32 quantum; |
67 | u32 memory_limit; | |
68 | u32 new_flow_count; | |
69 | u32 memory_usage; | |
70 | u32 overmemory; | |
71 | struct pie_stats stats; | |
72 | struct timer_list adapt_timer; | |
73 | }; | |
74 | ||
75 | static unsigned int fq_pie_hash(const struct fq_pie_sched_data *q, | |
76 | struct sk_buff *skb) | |
77 | { | |
78 | return reciprocal_scale(skb_get_hash(skb), q->flows_cnt); | |
79 | } | |
80 | ||
81 | static unsigned int fq_pie_classify(struct sk_buff *skb, struct Qdisc *sch, | |
82 | int *qerr) | |
83 | { | |
84 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
85 | struct tcf_proto *filter; | |
86 | struct tcf_result res; | |
87 | int result; | |
88 | ||
89 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
90 | TC_H_MIN(skb->priority) > 0 && | |
91 | TC_H_MIN(skb->priority) <= q->flows_cnt) | |
92 | return TC_H_MIN(skb->priority); | |
93 | ||
94 | filter = rcu_dereference_bh(q->filter_list); | |
95 | if (!filter) | |
96 | return fq_pie_hash(q, skb) + 1; | |
97 | ||
98 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
3aa26055 | 99 | result = tcf_classify(skb, NULL, filter, &res, false); |
ec97ecf1 MT |
100 | if (result >= 0) { |
101 | #ifdef CONFIG_NET_CLS_ACT | |
102 | switch (result) { | |
103 | case TC_ACT_STOLEN: | |
104 | case TC_ACT_QUEUED: | |
105 | case TC_ACT_TRAP: | |
106 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; | |
964201de | 107 | fallthrough; |
ec97ecf1 MT |
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 | /* add skb to flow queue (tail add) */ | |
119 | static inline void flow_queue_add(struct fq_pie_flow *flow, | |
120 | struct sk_buff *skb) | |
121 | { | |
122 | if (!flow->head) | |
123 | flow->head = skb; | |
124 | else | |
125 | flow->tail->next = skb; | |
126 | flow->tail = skb; | |
127 | skb->next = NULL; | |
128 | } | |
129 | ||
130 | static int fq_pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, | |
131 | struct sk_buff **to_free) | |
132 | { | |
ff9f17ce | 133 | enum skb_drop_reason reason = SKB_DROP_REASON_QDISC_OVERLIMIT; |
ec97ecf1 MT |
134 | struct fq_pie_sched_data *q = qdisc_priv(sch); |
135 | struct fq_pie_flow *sel_flow; | |
3f649ab7 | 136 | int ret; |
ec97ecf1 MT |
137 | u8 memory_limited = false; |
138 | u8 enqueue = false; | |
139 | u32 pkt_len; | |
140 | u32 idx; | |
141 | ||
142 | /* Classifies packet into corresponding flow */ | |
143 | idx = fq_pie_classify(skb, sch, &ret); | |
e70f7a11 DC |
144 | if (idx == 0) { |
145 | if (ret & __NET_XMIT_BYPASS) | |
146 | qdisc_qstats_drop(sch); | |
147 | __qdisc_drop(skb, to_free); | |
148 | return ret; | |
149 | } | |
150 | idx--; | |
ec97ecf1 | 151 | |
e70f7a11 | 152 | sel_flow = &q->flows[idx]; |
ec97ecf1 MT |
153 | /* Checks whether adding a new packet would exceed memory limit */ |
154 | get_pie_cb(skb)->mem_usage = skb->truesize; | |
155 | memory_limited = q->memory_usage > q->memory_limit + skb->truesize; | |
156 | ||
157 | /* Checks if the qdisc is full */ | |
158 | if (unlikely(qdisc_qlen(sch) >= sch->limit)) { | |
159 | q->stats.overlimit++; | |
160 | goto out; | |
161 | } else if (unlikely(memory_limited)) { | |
162 | q->overmemory++; | |
163 | } | |
164 | ||
ff9f17ce THJ |
165 | reason = SKB_DROP_REASON_QDISC_CONGESTED; |
166 | ||
ec97ecf1 MT |
167 | if (!pie_drop_early(sch, &q->p_params, &sel_flow->vars, |
168 | sel_flow->backlog, skb->len)) { | |
169 | enqueue = true; | |
170 | } else if (q->p_params.ecn && | |
171 | sel_flow->vars.prob <= (MAX_PROB / 100) * q->ecn_prob && | |
172 | INET_ECN_set_ce(skb)) { | |
173 | /* If packet is ecn capable, mark it if drop probability | |
174 | * is lower than the parameter ecn_prob, else drop it. | |
175 | */ | |
176 | q->stats.ecn_mark++; | |
177 | enqueue = true; | |
178 | } | |
179 | if (enqueue) { | |
180 | /* Set enqueue time only when dq_rate_estimator is disabled. */ | |
181 | if (!q->p_params.dq_rate_estimator) | |
182 | pie_set_enqueue_time(skb); | |
183 | ||
184 | pkt_len = qdisc_pkt_len(skb); | |
185 | q->stats.packets_in++; | |
186 | q->memory_usage += skb->truesize; | |
187 | sch->qstats.backlog += pkt_len; | |
188 | sch->q.qlen++; | |
189 | flow_queue_add(sel_flow, skb); | |
190 | if (list_empty(&sel_flow->flowchain)) { | |
191 | list_add_tail(&sel_flow->flowchain, &q->new_flows); | |
192 | q->new_flow_count++; | |
193 | sel_flow->deficit = q->quantum; | |
194 | sel_flow->qlen = 0; | |
195 | sel_flow->backlog = 0; | |
196 | } | |
197 | sel_flow->qlen++; | |
198 | sel_flow->backlog += pkt_len; | |
199 | return NET_XMIT_SUCCESS; | |
200 | } | |
201 | out: | |
202 | q->stats.dropped++; | |
203 | sel_flow->vars.accu_prob = 0; | |
ff9f17ce | 204 | qdisc_drop_reason(skb, sch, to_free, reason); |
ec97ecf1 MT |
205 | return NET_XMIT_CN; |
206 | } | |
207 | ||
ea23fbd2 | 208 | static const struct netlink_range_validation fq_pie_q_range = { |
cd2b8113 ED |
209 | .min = 1, |
210 | .max = 1 << 20, | |
211 | }; | |
212 | ||
ec97ecf1 MT |
213 | static const struct nla_policy fq_pie_policy[TCA_FQ_PIE_MAX + 1] = { |
214 | [TCA_FQ_PIE_LIMIT] = {.type = NLA_U32}, | |
215 | [TCA_FQ_PIE_FLOWS] = {.type = NLA_U32}, | |
216 | [TCA_FQ_PIE_TARGET] = {.type = NLA_U32}, | |
217 | [TCA_FQ_PIE_TUPDATE] = {.type = NLA_U32}, | |
218 | [TCA_FQ_PIE_ALPHA] = {.type = NLA_U32}, | |
219 | [TCA_FQ_PIE_BETA] = {.type = NLA_U32}, | |
cd2b8113 ED |
220 | [TCA_FQ_PIE_QUANTUM] = |
221 | NLA_POLICY_FULL_RANGE(NLA_U32, &fq_pie_q_range), | |
ec97ecf1 MT |
222 | [TCA_FQ_PIE_MEMORY_LIMIT] = {.type = NLA_U32}, |
223 | [TCA_FQ_PIE_ECN_PROB] = {.type = NLA_U32}, | |
224 | [TCA_FQ_PIE_ECN] = {.type = NLA_U32}, | |
225 | [TCA_FQ_PIE_BYTEMODE] = {.type = NLA_U32}, | |
226 | [TCA_FQ_PIE_DQ_RATE_ESTIMATOR] = {.type = NLA_U32}, | |
227 | }; | |
228 | ||
229 | static inline struct sk_buff *dequeue_head(struct fq_pie_flow *flow) | |
230 | { | |
231 | struct sk_buff *skb = flow->head; | |
232 | ||
233 | flow->head = skb->next; | |
234 | skb->next = NULL; | |
235 | return skb; | |
236 | } | |
237 | ||
238 | static struct sk_buff *fq_pie_qdisc_dequeue(struct Qdisc *sch) | |
239 | { | |
240 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
241 | struct sk_buff *skb = NULL; | |
242 | struct fq_pie_flow *flow; | |
243 | struct list_head *head; | |
244 | u32 pkt_len; | |
245 | ||
246 | begin: | |
247 | head = &q->new_flows; | |
248 | if (list_empty(head)) { | |
249 | head = &q->old_flows; | |
250 | if (list_empty(head)) | |
251 | return NULL; | |
252 | } | |
253 | ||
254 | flow = list_first_entry(head, struct fq_pie_flow, flowchain); | |
255 | /* Flow has exhausted all its credits */ | |
256 | if (flow->deficit <= 0) { | |
257 | flow->deficit += q->quantum; | |
258 | list_move_tail(&flow->flowchain, &q->old_flows); | |
259 | goto begin; | |
260 | } | |
261 | ||
262 | if (flow->head) { | |
263 | skb = dequeue_head(flow); | |
264 | pkt_len = qdisc_pkt_len(skb); | |
265 | sch->qstats.backlog -= pkt_len; | |
266 | sch->q.qlen--; | |
267 | qdisc_bstats_update(sch, skb); | |
268 | } | |
269 | ||
270 | if (!skb) { | |
271 | /* force a pass through old_flows to prevent starvation */ | |
272 | if (head == &q->new_flows && !list_empty(&q->old_flows)) | |
273 | list_move_tail(&flow->flowchain, &q->old_flows); | |
274 | else | |
275 | list_del_init(&flow->flowchain); | |
276 | goto begin; | |
277 | } | |
278 | ||
279 | flow->qlen--; | |
280 | flow->deficit -= pkt_len; | |
281 | flow->backlog -= pkt_len; | |
282 | q->memory_usage -= get_pie_cb(skb)->mem_usage; | |
283 | pie_process_dequeue(skb, &q->p_params, &flow->vars, flow->backlog); | |
284 | return skb; | |
285 | } | |
286 | ||
287 | static int fq_pie_change(struct Qdisc *sch, struct nlattr *opt, | |
288 | struct netlink_ext_ack *extack) | |
289 | { | |
290 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
291 | struct nlattr *tb[TCA_FQ_PIE_MAX + 1]; | |
292 | unsigned int len_dropped = 0; | |
293 | unsigned int num_dropped = 0; | |
294 | int err; | |
295 | ||
ec97ecf1 MT |
296 | err = nla_parse_nested(tb, TCA_FQ_PIE_MAX, opt, fq_pie_policy, extack); |
297 | if (err < 0) | |
298 | return err; | |
299 | ||
300 | sch_tree_lock(sch); | |
301 | if (tb[TCA_FQ_PIE_LIMIT]) { | |
302 | u32 limit = nla_get_u32(tb[TCA_FQ_PIE_LIMIT]); | |
303 | ||
13a9965d ED |
304 | WRITE_ONCE(q->p_params.limit, limit); |
305 | WRITE_ONCE(sch->limit, limit); | |
ec97ecf1 MT |
306 | } |
307 | if (tb[TCA_FQ_PIE_FLOWS]) { | |
308 | if (q->flows) { | |
309 | NL_SET_ERR_MSG_MOD(extack, | |
310 | "Number of flows cannot be changed"); | |
311 | goto flow_error; | |
312 | } | |
313 | q->flows_cnt = nla_get_u32(tb[TCA_FQ_PIE_FLOWS]); | |
3a62fed2 | 314 | if (!q->flows_cnt || q->flows_cnt > 65536) { |
ec97ecf1 | 315 | NL_SET_ERR_MSG_MOD(extack, |
3a62fed2 | 316 | "Number of flows must range in [1..65536]"); |
ec97ecf1 MT |
317 | goto flow_error; |
318 | } | |
319 | } | |
320 | ||
321 | /* convert from microseconds to pschedtime */ | |
322 | if (tb[TCA_FQ_PIE_TARGET]) { | |
323 | /* target is in us */ | |
324 | u32 target = nla_get_u32(tb[TCA_FQ_PIE_TARGET]); | |
325 | ||
326 | /* convert to pschedtime */ | |
13a9965d ED |
327 | WRITE_ONCE(q->p_params.target, |
328 | PSCHED_NS2TICKS((u64)target * NSEC_PER_USEC)); | |
ec97ecf1 MT |
329 | } |
330 | ||
331 | /* tupdate is in jiffies */ | |
332 | if (tb[TCA_FQ_PIE_TUPDATE]) | |
13a9965d ED |
333 | WRITE_ONCE(q->p_params.tupdate, |
334 | usecs_to_jiffies(nla_get_u32(tb[TCA_FQ_PIE_TUPDATE]))); | |
ec97ecf1 MT |
335 | |
336 | if (tb[TCA_FQ_PIE_ALPHA]) | |
13a9965d ED |
337 | WRITE_ONCE(q->p_params.alpha, |
338 | nla_get_u32(tb[TCA_FQ_PIE_ALPHA])); | |
ec97ecf1 MT |
339 | |
340 | if (tb[TCA_FQ_PIE_BETA]) | |
13a9965d ED |
341 | WRITE_ONCE(q->p_params.beta, |
342 | nla_get_u32(tb[TCA_FQ_PIE_BETA])); | |
ec97ecf1 MT |
343 | |
344 | if (tb[TCA_FQ_PIE_QUANTUM]) | |
13a9965d | 345 | WRITE_ONCE(q->quantum, nla_get_u32(tb[TCA_FQ_PIE_QUANTUM])); |
ec97ecf1 MT |
346 | |
347 | if (tb[TCA_FQ_PIE_MEMORY_LIMIT]) | |
13a9965d ED |
348 | WRITE_ONCE(q->memory_limit, |
349 | nla_get_u32(tb[TCA_FQ_PIE_MEMORY_LIMIT])); | |
ec97ecf1 MT |
350 | |
351 | if (tb[TCA_FQ_PIE_ECN_PROB]) | |
13a9965d ED |
352 | WRITE_ONCE(q->ecn_prob, |
353 | nla_get_u32(tb[TCA_FQ_PIE_ECN_PROB])); | |
ec97ecf1 MT |
354 | |
355 | if (tb[TCA_FQ_PIE_ECN]) | |
13a9965d ED |
356 | WRITE_ONCE(q->p_params.ecn, |
357 | nla_get_u32(tb[TCA_FQ_PIE_ECN])); | |
ec97ecf1 MT |
358 | |
359 | if (tb[TCA_FQ_PIE_BYTEMODE]) | |
13a9965d ED |
360 | WRITE_ONCE(q->p_params.bytemode, |
361 | nla_get_u32(tb[TCA_FQ_PIE_BYTEMODE])); | |
ec97ecf1 MT |
362 | |
363 | if (tb[TCA_FQ_PIE_DQ_RATE_ESTIMATOR]) | |
13a9965d ED |
364 | WRITE_ONCE(q->p_params.dq_rate_estimator, |
365 | nla_get_u32(tb[TCA_FQ_PIE_DQ_RATE_ESTIMATOR])); | |
ec97ecf1 MT |
366 | |
367 | /* Drop excess packets if new limit is lower */ | |
368 | while (sch->q.qlen > sch->limit) { | |
2d3cbfd6 | 369 | struct sk_buff *skb = qdisc_dequeue_internal(sch, false); |
ec97ecf1 | 370 | |
ec97ecf1 MT |
371 | len_dropped += qdisc_pkt_len(skb); |
372 | num_dropped += 1; | |
7a02ea65 | 373 | rtnl_kfree_skbs(skb, skb); |
ec97ecf1 MT |
374 | } |
375 | qdisc_tree_reduce_backlog(sch, num_dropped, len_dropped); | |
376 | ||
377 | sch_tree_unlock(sch); | |
378 | return 0; | |
379 | ||
380 | flow_error: | |
381 | sch_tree_unlock(sch); | |
382 | return -EINVAL; | |
383 | } | |
384 | ||
385 | static void fq_pie_timer(struct timer_list *t) | |
386 | { | |
41cb0855 | 387 | struct fq_pie_sched_data *q = timer_container_of(q, t, adapt_timer); |
8c21ab1b | 388 | unsigned long next, tupdate; |
ec97ecf1 MT |
389 | struct Qdisc *sch = q->sch; |
390 | spinlock_t *root_lock; /* to lock qdisc for probability calculations */ | |
8c21ab1b | 391 | int max_cnt, i; |
ec97ecf1 | 392 | |
d636fc5d | 393 | rcu_read_lock(); |
ec97ecf1 MT |
394 | root_lock = qdisc_lock(qdisc_root_sleeping(sch)); |
395 | spin_lock(root_lock); | |
396 | ||
8c21ab1b ED |
397 | /* Limit this expensive loop to 2048 flows per round. */ |
398 | max_cnt = min_t(int, q->flows_cnt - q->flows_cursor, 2048); | |
399 | for (i = 0; i < max_cnt; i++) { | |
400 | pie_calculate_probability(&q->p_params, | |
401 | &q->flows[q->flows_cursor].vars, | |
402 | q->flows[q->flows_cursor].backlog); | |
403 | q->flows_cursor++; | |
404 | } | |
ec97ecf1 | 405 | |
8c21ab1b ED |
406 | tupdate = q->p_params.tupdate; |
407 | next = 0; | |
408 | if (q->flows_cursor >= q->flows_cnt) { | |
409 | q->flows_cursor = 0; | |
410 | next = tupdate; | |
411 | } | |
412 | if (tupdate) | |
413 | mod_timer(&q->adapt_timer, jiffies + next); | |
ec97ecf1 | 414 | spin_unlock(root_lock); |
d636fc5d | 415 | rcu_read_unlock(); |
ec97ecf1 MT |
416 | } |
417 | ||
418 | static int fq_pie_init(struct Qdisc *sch, struct nlattr *opt, | |
419 | struct netlink_ext_ack *extack) | |
420 | { | |
421 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
422 | int err; | |
3a62fed2 | 423 | u32 idx; |
ec97ecf1 MT |
424 | |
425 | pie_params_init(&q->p_params); | |
426 | sch->limit = 10 * 1024; | |
427 | q->p_params.limit = sch->limit; | |
428 | q->quantum = psched_mtu(qdisc_dev(sch)); | |
429 | q->sch = sch; | |
430 | q->ecn_prob = 10; | |
431 | q->flows_cnt = 1024; | |
432 | q->memory_limit = SZ_32M; | |
433 | ||
434 | INIT_LIST_HEAD(&q->new_flows); | |
435 | INIT_LIST_HEAD(&q->old_flows); | |
4eef8b1f | 436 | timer_setup(&q->adapt_timer, fq_pie_timer, 0); |
ec97ecf1 MT |
437 | |
438 | if (opt) { | |
439 | err = fq_pie_change(sch, opt, extack); | |
440 | ||
441 | if (err) | |
442 | return err; | |
443 | } | |
444 | ||
445 | err = tcf_block_get(&q->block, &q->filter_list, sch, extack); | |
446 | if (err) | |
447 | goto init_failure; | |
448 | ||
449 | q->flows = kvcalloc(q->flows_cnt, sizeof(struct fq_pie_flow), | |
450 | GFP_KERNEL); | |
451 | if (!q->flows) { | |
452 | err = -ENOMEM; | |
453 | goto init_failure; | |
454 | } | |
455 | for (idx = 0; idx < q->flows_cnt; idx++) { | |
456 | struct fq_pie_flow *flow = q->flows + idx; | |
457 | ||
458 | INIT_LIST_HEAD(&flow->flowchain); | |
459 | pie_vars_init(&flow->vars); | |
460 | } | |
461 | ||
ec97ecf1 MT |
462 | mod_timer(&q->adapt_timer, jiffies + HZ / 2); |
463 | ||
464 | return 0; | |
465 | ||
466 | init_failure: | |
467 | q->flows_cnt = 0; | |
468 | ||
469 | return err; | |
470 | } | |
471 | ||
472 | static int fq_pie_dump(struct Qdisc *sch, struct sk_buff *skb) | |
473 | { | |
474 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
475 | struct nlattr *opts; | |
476 | ||
477 | opts = nla_nest_start(skb, TCA_OPTIONS); | |
478 | if (!opts) | |
479 | return -EMSGSIZE; | |
480 | ||
481 | /* convert target from pschedtime to us */ | |
13a9965d ED |
482 | if (nla_put_u32(skb, TCA_FQ_PIE_LIMIT, READ_ONCE(sch->limit)) || |
483 | nla_put_u32(skb, TCA_FQ_PIE_FLOWS, READ_ONCE(q->flows_cnt)) || | |
ec97ecf1 | 484 | nla_put_u32(skb, TCA_FQ_PIE_TARGET, |
13a9965d | 485 | ((u32)PSCHED_TICKS2NS(READ_ONCE(q->p_params.target))) / |
ec97ecf1 MT |
486 | NSEC_PER_USEC) || |
487 | nla_put_u32(skb, TCA_FQ_PIE_TUPDATE, | |
13a9965d ED |
488 | jiffies_to_usecs(READ_ONCE(q->p_params.tupdate))) || |
489 | nla_put_u32(skb, TCA_FQ_PIE_ALPHA, READ_ONCE(q->p_params.alpha)) || | |
490 | nla_put_u32(skb, TCA_FQ_PIE_BETA, READ_ONCE(q->p_params.beta)) || | |
491 | nla_put_u32(skb, TCA_FQ_PIE_QUANTUM, READ_ONCE(q->quantum)) || | |
492 | nla_put_u32(skb, TCA_FQ_PIE_MEMORY_LIMIT, | |
493 | READ_ONCE(q->memory_limit)) || | |
494 | nla_put_u32(skb, TCA_FQ_PIE_ECN_PROB, READ_ONCE(q->ecn_prob)) || | |
495 | nla_put_u32(skb, TCA_FQ_PIE_ECN, READ_ONCE(q->p_params.ecn)) || | |
496 | nla_put_u32(skb, TCA_FQ_PIE_BYTEMODE, READ_ONCE(q->p_params.bytemode)) || | |
ec97ecf1 | 497 | nla_put_u32(skb, TCA_FQ_PIE_DQ_RATE_ESTIMATOR, |
13a9965d | 498 | READ_ONCE(q->p_params.dq_rate_estimator))) |
ec97ecf1 MT |
499 | goto nla_put_failure; |
500 | ||
501 | return nla_nest_end(skb, opts); | |
502 | ||
503 | nla_put_failure: | |
504 | nla_nest_cancel(skb, opts); | |
505 | return -EMSGSIZE; | |
506 | } | |
507 | ||
508 | static int fq_pie_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
509 | { | |
510 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
511 | struct tc_fq_pie_xstats st = { | |
512 | .packets_in = q->stats.packets_in, | |
513 | .overlimit = q->stats.overlimit, | |
514 | .overmemory = q->overmemory, | |
515 | .dropped = q->stats.dropped, | |
516 | .ecn_mark = q->stats.ecn_mark, | |
517 | .new_flow_count = q->new_flow_count, | |
518 | .memory_usage = q->memory_usage, | |
519 | }; | |
520 | struct list_head *pos; | |
521 | ||
522 | sch_tree_lock(sch); | |
523 | list_for_each(pos, &q->new_flows) | |
524 | st.new_flows_len++; | |
525 | ||
526 | list_for_each(pos, &q->old_flows) | |
527 | st.old_flows_len++; | |
528 | sch_tree_unlock(sch); | |
529 | ||
530 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
531 | } | |
532 | ||
533 | static void fq_pie_reset(struct Qdisc *sch) | |
534 | { | |
535 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
3a62fed2 | 536 | u32 idx; |
ec97ecf1 MT |
537 | |
538 | INIT_LIST_HEAD(&q->new_flows); | |
539 | INIT_LIST_HEAD(&q->old_flows); | |
540 | for (idx = 0; idx < q->flows_cnt; idx++) { | |
541 | struct fq_pie_flow *flow = q->flows + idx; | |
542 | ||
543 | /* Removes all packets from flow */ | |
544 | rtnl_kfree_skbs(flow->head, flow->tail); | |
545 | flow->head = NULL; | |
546 | ||
547 | INIT_LIST_HEAD(&flow->flowchain); | |
548 | pie_vars_init(&flow->vars); | |
549 | } | |
ec97ecf1 MT |
550 | } |
551 | ||
552 | static void fq_pie_destroy(struct Qdisc *sch) | |
553 | { | |
554 | struct fq_pie_sched_data *q = qdisc_priv(sch); | |
555 | ||
556 | tcf_block_put(q->block); | |
61c24026 | 557 | q->p_params.tupdate = 0; |
8fa7292f | 558 | timer_delete_sync(&q->adapt_timer); |
ec97ecf1 MT |
559 | kvfree(q->flows); |
560 | } | |
561 | ||
562 | static struct Qdisc_ops fq_pie_qdisc_ops __read_mostly = { | |
563 | .id = "fq_pie", | |
564 | .priv_size = sizeof(struct fq_pie_sched_data), | |
565 | .enqueue = fq_pie_qdisc_enqueue, | |
566 | .dequeue = fq_pie_qdisc_dequeue, | |
567 | .peek = qdisc_peek_dequeued, | |
568 | .init = fq_pie_init, | |
569 | .destroy = fq_pie_destroy, | |
570 | .reset = fq_pie_reset, | |
571 | .change = fq_pie_change, | |
572 | .dump = fq_pie_dump, | |
573 | .dump_stats = fq_pie_dump_stats, | |
574 | .owner = THIS_MODULE, | |
575 | }; | |
9474c62a | 576 | MODULE_ALIAS_NET_SCH("fq_pie"); |
ec97ecf1 MT |
577 | |
578 | static int __init fq_pie_module_init(void) | |
579 | { | |
580 | return register_qdisc(&fq_pie_qdisc_ops); | |
581 | } | |
582 | ||
583 | static void __exit fq_pie_module_exit(void) | |
584 | { | |
585 | unregister_qdisc(&fq_pie_qdisc_ops); | |
586 | } | |
587 | ||
588 | module_init(fq_pie_module_init); | |
589 | module_exit(fq_pie_module_exit); | |
590 | ||
591 | MODULE_DESCRIPTION("Flow Queue Proportional Integral controller Enhanced (FQ-PIE)"); | |
592 | MODULE_AUTHOR("Mohit P. Tahiliani"); | |
593 | MODULE_LICENSE("GPL"); |