blob: 5b4fe2ef7986450a0489f36dc3c44329d5b860b3 [file] [log] [blame]
b.liue9582032025-04-17 19:18:16 +08001// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Fair Queue CoDel discipline
4 *
5 * Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com>
6 */
7
8#include <linux/module.h>
9#include <linux/types.h>
10#include <linux/kernel.h>
11#include <linux/jiffies.h>
12#include <linux/string.h>
13#include <linux/in.h>
14#include <linux/errno.h>
15#include <linux/init.h>
16#include <linux/skbuff.h>
17#include <linux/jhash.h>
18#include <linux/slab.h>
19#include <linux/vmalloc.h>
20#include <net/netlink.h>
21#include <net/pkt_sched.h>
22#include <net/pkt_cls.h>
23#include <net/codel.h>
24#include <net/codel_impl.h>
25#include <net/codel_qdisc.h>
26
27/* Fair Queue CoDel.
28 *
29 * Principles :
30 * Packets are classified (internal classifier or external) on flows.
31 * This is a Stochastic model (as we use a hash, several flows
32 * might be hashed on same slot)
33 * Each flow has a CoDel managed queue.
34 * Flows are linked onto two (Round Robin) lists,
35 * so that new flows have priority on old ones.
36 *
37 * For a given flow, packets are not reordered (CoDel uses a FIFO)
38 * head drops only.
39 * ECN capability is on by default.
40 * Low memory footprint (64 bytes per flow)
41 */
42
43struct fq_codel_flow {
44 struct sk_buff *head;
45 struct sk_buff *tail;
46 struct list_head flowchain;
47 int deficit;
48 struct codel_vars cvars;
49}; /* please try to keep this structure <= 64 bytes */
50
51struct fq_codel_sched_data {
52 struct tcf_proto __rcu *filter_list; /* optional external classifier */
53 struct tcf_block *block;
54 struct fq_codel_flow *flows; /* Flows table [flows_cnt] */
55 u32 *backlogs; /* backlog table [flows_cnt] */
56 u32 flows_cnt; /* number of flows */
57 u32 quantum; /* psched_mtu(qdisc_dev(sch)); */
58 u32 drop_batch_size;
59 u32 memory_limit;
60 struct codel_params cparams;
61 struct codel_stats cstats;
62 u32 memory_usage;
63 u32 drop_overmemory;
64 u32 drop_overlimit;
65 u32 new_flow_count;
66
67 struct list_head new_flows; /* list of new flows */
68 struct list_head old_flows; /* list of old flows */
69};
70
71static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q,
72 struct sk_buff *skb)
73{
74 return reciprocal_scale(skb_get_hash(skb), q->flows_cnt);
75}
76
77static 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);
81 struct tcf_proto *filter;
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
90 filter = rcu_dereference_bh(q->filter_list);
91 if (!filter)
92 return fq_codel_hash(q, skb) + 1;
93
94 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
95 result = tcf_classify(skb, filter, &res, false);
96 if (result >= 0) {
97#ifdef CONFIG_NET_CLS_ACT
98 switch (result) {
99 case TC_ACT_STOLEN:
100 case TC_ACT_QUEUED:
101 case TC_ACT_TRAP:
102 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
103 /* fall through */
104 case TC_ACT_SHOT:
105 return 0;
106 }
107#endif
108 if (TC_H_MIN(res.classid) <= q->flows_cnt)
109 return TC_H_MIN(res.classid);
110 }
111 return 0;
112}
113
114/* helper functions : might be changed when/if skb use a standard list_head */
115
116/* remove one skb from head of slot queue */
117static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow)
118{
119 struct sk_buff *skb = flow->head;
120
121 flow->head = skb->next;
122 skb_mark_not_on_list(skb);
123 return skb;
124}
125
126/* add skb to flow queue (tail add) */
127static inline void flow_queue_add(struct fq_codel_flow *flow,
128 struct sk_buff *skb)
129{
130 if (flow->head == NULL)
131 flow->head = skb;
132 else
133 flow->tail->next = skb;
134 flow->tail = skb;
135 skb->next = NULL;
136}
137
138static unsigned int fq_codel_drop(struct Qdisc *sch, unsigned int max_packets,
139 struct sk_buff **to_free)
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;
145 unsigned int threshold;
146 unsigned int mem = 0;
147
148 /* Queue is full! Find the fat flow and drop packet(s) from it.
149 * This might sound expensive, but with 1024 flows, we scan
150 * 4KB of memory, and we dont need to handle a complex tree
151 * in fast path (packet queue/enqueue) with many cache misses.
152 * In stress mode, we'll try to drop 64 packets from the flow,
153 * amortizing this linear lookup to one cache line per drop.
154 */
155 for (i = 0; i < q->flows_cnt; i++) {
156 if (q->backlogs[i] > maxbacklog) {
157 maxbacklog = q->backlogs[i];
158 idx = i;
159 }
160 }
161
162 /* Our goal is to drop half of this fat flow backlog */
163 threshold = maxbacklog >> 1;
164
165 flow = &q->flows[idx];
166 len = 0;
167 i = 0;
168 do {
169 skb = dequeue_head(flow);
170 len += qdisc_pkt_len(skb);
171 mem += get_codel_cb(skb)->mem_usage;
172 __qdisc_drop(skb, to_free);
173 } while (++i < max_packets && len < threshold);
174
175 /* Tell codel to increase its signal strength also */
176 flow->cvars.count += i;
177 q->backlogs[idx] -= len;
178 q->memory_usage -= mem;
179 sch->qstats.drops += i;
180 sch->qstats.backlog -= len;
181 sch->q.qlen -= i;
182 return idx;
183}
184
185static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch,
186 struct sk_buff **to_free)
187{
188 struct fq_codel_sched_data *q = qdisc_priv(sch);
189 unsigned int idx, prev_backlog, prev_qlen;
190 struct fq_codel_flow *flow;
191 int ret;
192 unsigned int pkt_len;
193 bool memory_limited;
194
195 idx = fq_codel_classify(skb, sch, &ret);
196 if (idx == 0) {
197 if (ret & __NET_XMIT_BYPASS)
198 qdisc_qstats_drop(sch);
199 __qdisc_drop(skb, to_free);
200 return ret;
201 }
202 idx--;
203
204 codel_set_enqueue_time(skb);
205 flow = &q->flows[idx];
206 flow_queue_add(flow, skb);
207 q->backlogs[idx] += qdisc_pkt_len(skb);
208 qdisc_qstats_backlog_inc(sch, skb);
209
210 if (list_empty(&flow->flowchain)) {
211 list_add_tail(&flow->flowchain, &q->new_flows);
212 q->new_flow_count++;
213 flow->deficit = q->quantum;
214 }
215 get_codel_cb(skb)->mem_usage = skb->truesize;
216 q->memory_usage += get_codel_cb(skb)->mem_usage;
217 memory_limited = q->memory_usage > q->memory_limit;
218 if (++sch->q.qlen <= sch->limit && !memory_limited)
219 return NET_XMIT_SUCCESS;
220
221 prev_backlog = sch->qstats.backlog;
222 prev_qlen = sch->q.qlen;
223
224 /* save this packet length as it might be dropped by fq_codel_drop() */
225 pkt_len = qdisc_pkt_len(skb);
226 /* fq_codel_drop() is quite expensive, as it performs a linear search
227 * in q->backlogs[] to find a fat flow.
228 * So instead of dropping a single packet, drop half of its backlog
229 * with a 64 packets limit to not add a too big cpu spike here.
230 */
231 ret = fq_codel_drop(sch, q->drop_batch_size, to_free);
232
233 prev_qlen -= sch->q.qlen;
234 prev_backlog -= sch->qstats.backlog;
235 q->drop_overlimit += prev_qlen;
236 if (memory_limited)
237 q->drop_overmemory += prev_qlen;
238
239 /* As we dropped packet(s), better let upper stack know this.
240 * If we dropped a packet for this flow, return NET_XMIT_CN,
241 * but in this case, our parents wont increase their backlogs.
242 */
243 if (ret == idx) {
244 qdisc_tree_reduce_backlog(sch, prev_qlen - 1,
245 prev_backlog - pkt_len);
246 return NET_XMIT_CN;
247 }
248 qdisc_tree_reduce_backlog(sch, prev_qlen, prev_backlog);
249 return NET_XMIT_SUCCESS;
250}
251
252/* This is the specific function called from codel_dequeue()
253 * to dequeue a packet from queue. Note: backlog is handled in
254 * codel, we dont need to reduce it here.
255 */
256static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx)
257{
258 struct Qdisc *sch = ctx;
259 struct fq_codel_sched_data *q = qdisc_priv(sch);
260 struct fq_codel_flow *flow;
261 struct sk_buff *skb = NULL;
262
263 flow = container_of(vars, struct fq_codel_flow, cvars);
264 if (flow->head) {
265 skb = dequeue_head(flow);
266 q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb);
267 q->memory_usage -= get_codel_cb(skb)->mem_usage;
268 sch->q.qlen--;
269 sch->qstats.backlog -= qdisc_pkt_len(skb);
270 }
271 return skb;
272}
273
274static void drop_func(struct sk_buff *skb, void *ctx)
275{
276 struct Qdisc *sch = ctx;
277
278 kfree_skb(skb);
279 qdisc_qstats_drop(sch);
280}
281
282static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch)
283{
284 struct fq_codel_sched_data *q = qdisc_priv(sch);
285 struct sk_buff *skb;
286 struct fq_codel_flow *flow;
287 struct list_head *head;
288
289begin:
290 head = &q->new_flows;
291 if (list_empty(head)) {
292 head = &q->old_flows;
293 if (list_empty(head))
294 return NULL;
295 }
296 flow = list_first_entry(head, struct fq_codel_flow, flowchain);
297
298 if (flow->deficit <= 0) {
299 flow->deficit += q->quantum;
300 list_move_tail(&flow->flowchain, &q->old_flows);
301 goto begin;
302 }
303
304 skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams,
305 &flow->cvars, &q->cstats, qdisc_pkt_len,
306 codel_get_enqueue_time, drop_func, dequeue_func);
307
308 /* If our qlen is 0 qdisc_tree_reduce_backlog() will deactivate
309 * parent class, dequeue in parent qdisc will do the same if we
310 * return skb. Temporary increment qlen if we have skb.
311 */
312 if (q->cstats.drop_count) {
313 if (skb)
314 sch->q.qlen++;
315 qdisc_tree_reduce_backlog(sch, q->cstats.drop_count,
316 q->cstats.drop_len);
317 if (skb)
318 sch->q.qlen--;
319 q->cstats.drop_count = 0;
320 q->cstats.drop_len = 0;
321 }
322
323 if (!skb) {
324 /* force a pass through old_flows to prevent starvation */
325 if ((head == &q->new_flows) && !list_empty(&q->old_flows))
326 list_move_tail(&flow->flowchain, &q->old_flows);
327 else
328 list_del_init(&flow->flowchain);
329 goto begin;
330 }
331 qdisc_bstats_update(sch, skb);
332 flow->deficit -= qdisc_pkt_len(skb);
333 return skb;
334}
335
336static void fq_codel_flow_purge(struct fq_codel_flow *flow)
337{
338 rtnl_kfree_skbs(flow->head, flow->tail);
339 flow->head = NULL;
340}
341
342static void fq_codel_reset(struct Qdisc *sch)
343{
344 struct fq_codel_sched_data *q = qdisc_priv(sch);
345 int i;
346
347 INIT_LIST_HEAD(&q->new_flows);
348 INIT_LIST_HEAD(&q->old_flows);
349 for (i = 0; i < q->flows_cnt; i++) {
350 struct fq_codel_flow *flow = q->flows + i;
351
352 fq_codel_flow_purge(flow);
353 INIT_LIST_HEAD(&flow->flowchain);
354 codel_vars_init(&flow->cvars);
355 }
356 memset(q->backlogs, 0, q->flows_cnt * sizeof(u32));
357 sch->q.qlen = 0;
358 sch->qstats.backlog = 0;
359 q->memory_usage = 0;
360}
361
362static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = {
363 [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 },
364 [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 },
365 [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 },
366 [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 },
367 [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 },
368 [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 },
369 [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 },
370 [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 },
371 [TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 },
372};
373
374static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt,
375 struct netlink_ext_ack *extack)
376{
377 struct fq_codel_sched_data *q = qdisc_priv(sch);
378 struct nlattr *tb[TCA_FQ_CODEL_MAX + 1];
379 u32 quantum = 0;
380 int err;
381
382 if (!opt)
383 return -EINVAL;
384
385 err = nla_parse_nested_deprecated(tb, TCA_FQ_CODEL_MAX, opt,
386 fq_codel_policy, NULL);
387 if (err < 0)
388 return err;
389 if (tb[TCA_FQ_CODEL_FLOWS]) {
390 if (q->flows)
391 return -EINVAL;
392 q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]);
393 if (!q->flows_cnt ||
394 q->flows_cnt > 65536)
395 return -EINVAL;
396 }
397 if (tb[TCA_FQ_CODEL_QUANTUM]) {
398 quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM]));
399 if (quantum > FQ_CODEL_QUANTUM_MAX) {
400 NL_SET_ERR_MSG(extack, "Invalid quantum");
401 return -EINVAL;
402 }
403 }
404 sch_tree_lock(sch);
405
406 if (tb[TCA_FQ_CODEL_TARGET]) {
407 u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]);
408
409 q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT;
410 }
411
412 if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) {
413 u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]);
414
415 q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
416 }
417
418 if (tb[TCA_FQ_CODEL_INTERVAL]) {
419 u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]);
420
421 q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT;
422 }
423
424 if (tb[TCA_FQ_CODEL_LIMIT])
425 sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]);
426
427 if (tb[TCA_FQ_CODEL_ECN])
428 q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]);
429
430 if (quantum)
431 q->quantum = quantum;
432
433 if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])
434 q->drop_batch_size = max(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]));
435
436 if (tb[TCA_FQ_CODEL_MEMORY_LIMIT])
437 q->memory_limit = min(1U << 31, nla_get_u32(tb[TCA_FQ_CODEL_MEMORY_LIMIT]));
438
439 while (sch->q.qlen > sch->limit ||
440 q->memory_usage > q->memory_limit) {
441 struct sk_buff *skb = fq_codel_dequeue(sch);
442
443 q->cstats.drop_len += qdisc_pkt_len(skb);
444 rtnl_kfree_skbs(skb, skb);
445 q->cstats.drop_count++;
446 }
447 qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len);
448 q->cstats.drop_count = 0;
449 q->cstats.drop_len = 0;
450
451 sch_tree_unlock(sch);
452 return 0;
453}
454
455static void fq_codel_destroy(struct Qdisc *sch)
456{
457 struct fq_codel_sched_data *q = qdisc_priv(sch);
458
459 tcf_block_put(q->block);
460 kvfree(q->backlogs);
461 kvfree(q->flows);
462}
463
464static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt,
465 struct netlink_ext_ack *extack)
466{
467 struct fq_codel_sched_data *q = qdisc_priv(sch);
468 int i;
469 int err;
470
471 sch->limit = 10*1024;
472 q->flows_cnt = 1024;
473#ifdef CONFIG_X86_64
474 q->memory_limit = 32 << 20; /* 32 MBytes */
475#else
476 q->memory_limit = 4 << 20; /* 4 MBytes */
477#endif
478 q->drop_batch_size = 64;
479 q->quantum = psched_mtu(qdisc_dev(sch));
480 INIT_LIST_HEAD(&q->new_flows);
481 INIT_LIST_HEAD(&q->old_flows);
482 codel_params_init(&q->cparams);
483 codel_stats_init(&q->cstats);
484 q->cparams.ecn = true;
485 q->cparams.mtu = psched_mtu(qdisc_dev(sch));
486
487 if (opt) {
488 err = fq_codel_change(sch, opt, extack);
489 if (err)
490 goto init_failure;
491 }
492
493 err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
494 if (err)
495 goto init_failure;
496
497 if (!q->flows) {
498 q->flows = kvcalloc(q->flows_cnt,
499 sizeof(struct fq_codel_flow),
500 GFP_KERNEL);
501 if (!q->flows) {
502 err = -ENOMEM;
503 goto init_failure;
504 }
505 q->backlogs = kvcalloc(q->flows_cnt, sizeof(u32), GFP_KERNEL);
506 if (!q->backlogs) {
507 err = -ENOMEM;
508 goto alloc_failure;
509 }
510 for (i = 0; i < q->flows_cnt; i++) {
511 struct fq_codel_flow *flow = q->flows + i;
512
513 INIT_LIST_HEAD(&flow->flowchain);
514 codel_vars_init(&flow->cvars);
515 }
516 }
517 if (sch->limit >= 1)
518 sch->flags |= TCQ_F_CAN_BYPASS;
519 else
520 sch->flags &= ~TCQ_F_CAN_BYPASS;
521 return 0;
522
523alloc_failure:
524 kvfree(q->flows);
525 q->flows = NULL;
526init_failure:
527 q->flows_cnt = 0;
528 return err;
529}
530
531static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb)
532{
533 struct fq_codel_sched_data *q = qdisc_priv(sch);
534 struct nlattr *opts;
535
536 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
537 if (opts == NULL)
538 goto nla_put_failure;
539
540 if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET,
541 codel_time_to_us(q->cparams.target)) ||
542 nla_put_u32(skb, TCA_FQ_CODEL_LIMIT,
543 sch->limit) ||
544 nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL,
545 codel_time_to_us(q->cparams.interval)) ||
546 nla_put_u32(skb, TCA_FQ_CODEL_ECN,
547 q->cparams.ecn) ||
548 nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM,
549 q->quantum) ||
550 nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE,
551 q->drop_batch_size) ||
552 nla_put_u32(skb, TCA_FQ_CODEL_MEMORY_LIMIT,
553 q->memory_limit) ||
554 nla_put_u32(skb, TCA_FQ_CODEL_FLOWS,
555 q->flows_cnt))
556 goto nla_put_failure;
557
558 if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD &&
559 nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD,
560 codel_time_to_us(q->cparams.ce_threshold)))
561 goto nla_put_failure;
562
563 return nla_nest_end(skb, opts);
564
565nla_put_failure:
566 return -1;
567}
568
569static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
570{
571 struct fq_codel_sched_data *q = qdisc_priv(sch);
572 struct tc_fq_codel_xstats st = {
573 .type = TCA_FQ_CODEL_XSTATS_QDISC,
574 };
575 struct list_head *pos;
576
577 st.qdisc_stats.maxpacket = q->cstats.maxpacket;
578 st.qdisc_stats.drop_overlimit = q->drop_overlimit;
579 st.qdisc_stats.ecn_mark = q->cstats.ecn_mark;
580 st.qdisc_stats.new_flow_count = q->new_flow_count;
581 st.qdisc_stats.ce_mark = q->cstats.ce_mark;
582 st.qdisc_stats.memory_usage = q->memory_usage;
583 st.qdisc_stats.drop_overmemory = q->drop_overmemory;
584
585 sch_tree_lock(sch);
586 list_for_each(pos, &q->new_flows)
587 st.qdisc_stats.new_flows_len++;
588
589 list_for_each(pos, &q->old_flows)
590 st.qdisc_stats.old_flows_len++;
591 sch_tree_unlock(sch);
592
593 return gnet_stats_copy_app(d, &st, sizeof(st));
594}
595
596static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg)
597{
598 return NULL;
599}
600
601static unsigned long fq_codel_find(struct Qdisc *sch, u32 classid)
602{
603 return 0;
604}
605
606static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent,
607 u32 classid)
608{
609 return 0;
610}
611
612static void fq_codel_unbind(struct Qdisc *q, unsigned long cl)
613{
614}
615
616static struct tcf_block *fq_codel_tcf_block(struct Qdisc *sch, unsigned long cl,
617 struct netlink_ext_ack *extack)
618{
619 struct fq_codel_sched_data *q = qdisc_priv(sch);
620
621 if (cl)
622 return NULL;
623 return q->block;
624}
625
626static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl,
627 struct sk_buff *skb, struct tcmsg *tcm)
628{
629 tcm->tcm_handle |= TC_H_MIN(cl);
630 return 0;
631}
632
633static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl,
634 struct gnet_dump *d)
635{
636 struct fq_codel_sched_data *q = qdisc_priv(sch);
637 u32 idx = cl - 1;
638 struct gnet_stats_queue qs = { 0 };
639 struct tc_fq_codel_xstats xstats;
640
641 if (idx < q->flows_cnt) {
642 const struct fq_codel_flow *flow = &q->flows[idx];
643 const struct sk_buff *skb;
644
645 memset(&xstats, 0, sizeof(xstats));
646 xstats.type = TCA_FQ_CODEL_XSTATS_CLASS;
647 xstats.class_stats.deficit = flow->deficit;
648 xstats.class_stats.ldelay =
649 codel_time_to_us(flow->cvars.ldelay);
650 xstats.class_stats.count = flow->cvars.count;
651 xstats.class_stats.lastcount = flow->cvars.lastcount;
652 xstats.class_stats.dropping = flow->cvars.dropping;
653 if (flow->cvars.dropping) {
654 codel_tdiff_t delta = flow->cvars.drop_next -
655 codel_get_time();
656
657 xstats.class_stats.drop_next = (delta >= 0) ?
658 codel_time_to_us(delta) :
659 -codel_time_to_us(-delta);
660 }
661 if (flow->head) {
662 sch_tree_lock(sch);
663 skb = flow->head;
664 while (skb) {
665 qs.qlen++;
666 skb = skb->next;
667 }
668 sch_tree_unlock(sch);
669 }
670 qs.backlog = q->backlogs[idx];
671 qs.drops = 0;
672 }
673 if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0)
674 return -1;
675 if (idx < q->flows_cnt)
676 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
677 return 0;
678}
679
680static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg)
681{
682 struct fq_codel_sched_data *q = qdisc_priv(sch);
683 unsigned int i;
684
685 if (arg->stop)
686 return;
687
688 for (i = 0; i < q->flows_cnt; i++) {
689 if (list_empty(&q->flows[i].flowchain) ||
690 arg->count < arg->skip) {
691 arg->count++;
692 continue;
693 }
694 if (arg->fn(sch, i + 1, arg) < 0) {
695 arg->stop = 1;
696 break;
697 }
698 arg->count++;
699 }
700}
701
702static const struct Qdisc_class_ops fq_codel_class_ops = {
703 .leaf = fq_codel_leaf,
704 .find = fq_codel_find,
705 .tcf_block = fq_codel_tcf_block,
706 .bind_tcf = fq_codel_bind,
707 .unbind_tcf = fq_codel_unbind,
708 .dump = fq_codel_dump_class,
709 .dump_stats = fq_codel_dump_class_stats,
710 .walk = fq_codel_walk,
711};
712
713struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = {
714 .cl_ops = &fq_codel_class_ops,
715 .id = "fq_codel",
716 .priv_size = sizeof(struct fq_codel_sched_data),
717 .enqueue = fq_codel_enqueue,
718 .dequeue = fq_codel_dequeue,
719 .peek = qdisc_peek_dequeued,
720 .init = fq_codel_init,
721 .reset = fq_codel_reset,
722 .destroy = fq_codel_destroy,
723 .change = fq_codel_change,
724 .dump = fq_codel_dump,
725 .dump_stats = fq_codel_dump_stats,
726 .owner = THIS_MODULE,
727};
728EXPORT_SYMBOL(fq_codel_qdisc_ops);
729
730static int __init fq_codel_module_init(void)
731{
732 return register_qdisc(&fq_codel_qdisc_ops);
733}
734
735static void __exit fq_codel_module_exit(void)
736{
737 unregister_qdisc(&fq_codel_qdisc_ops);
738}
739
740module_init(fq_codel_module_init)
741module_exit(fq_codel_module_exit)
742MODULE_AUTHOR("Eric Dumazet");
743MODULE_LICENSE("GPL");