xref: /openbmc/linux/net/sched/sch_fq_codel.c (revision 887069f4)
1 // 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/slab.h>
18 #include <linux/vmalloc.h>
19 #include <net/netlink.h>
20 #include <net/pkt_sched.h>
21 #include <net/pkt_cls.h>
22 #include <net/codel.h>
23 #include <net/codel_impl.h>
24 #include <net/codel_qdisc.h>
25 
26 /*	Fair Queue CoDel.
27  *
28  * Principles :
29  * Packets are classified (internal classifier or external) on flows.
30  * This is a Stochastic model (as we use a hash, several flows
31  *			       might be hashed on same slot)
32  * Each flow has a CoDel managed queue.
33  * Flows are linked onto two (Round Robin) lists,
34  * so that new flows have priority on old ones.
35  *
36  * For a given flow, packets are not reordered (CoDel uses a FIFO)
37  * head drops only.
38  * ECN capability is on by default.
39  * Low memory footprint (64 bytes per flow)
40  */
41 
42 struct fq_codel_flow {
43 	struct sk_buff	  *head;
44 	struct sk_buff	  *tail;
45 	struct list_head  flowchain;
46 	int		  deficit;
47 	struct codel_vars cvars;
48 }; /* please try to keep this structure <= 64 bytes */
49 
50 struct fq_codel_sched_data {
51 	struct tcf_proto __rcu *filter_list; /* optional external classifier */
52 	struct tcf_block *block;
53 	struct fq_codel_flow *flows;	/* Flows table [flows_cnt] */
54 	u32		*backlogs;	/* backlog table [flows_cnt] */
55 	u32		flows_cnt;	/* number of flows */
56 	u32		quantum;	/* psched_mtu(qdisc_dev(sch)); */
57 	u32		drop_batch_size;
58 	u32		memory_limit;
59 	struct codel_params cparams;
60 	struct codel_stats cstats;
61 	u32		memory_usage;
62 	u32		drop_overmemory;
63 	u32		drop_overlimit;
64 	u32		new_flow_count;
65 
66 	struct list_head new_flows;	/* list of new flows */
67 	struct list_head old_flows;	/* list of old flows */
68 };
69 
70 static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q,
71 				  struct sk_buff *skb)
72 {
73 	return reciprocal_scale(skb_get_hash(skb), q->flows_cnt);
74 }
75 
76 static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch,
77 				      int *qerr)
78 {
79 	struct fq_codel_sched_data *q = qdisc_priv(sch);
80 	struct tcf_proto *filter;
81 	struct tcf_result res;
82 	int result;
83 
84 	if (TC_H_MAJ(skb->priority) == sch->handle &&
85 	    TC_H_MIN(skb->priority) > 0 &&
86 	    TC_H_MIN(skb->priority) <= q->flows_cnt)
87 		return TC_H_MIN(skb->priority);
88 
89 	filter = rcu_dereference_bh(q->filter_list);
90 	if (!filter)
91 		return fq_codel_hash(q, skb) + 1;
92 
93 	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
94 	result = tcf_classify(skb, NULL, filter, &res, false);
95 	if (result >= 0) {
96 #ifdef CONFIG_NET_CLS_ACT
97 		switch (result) {
98 		case TC_ACT_STOLEN:
99 		case TC_ACT_QUEUED:
100 		case TC_ACT_TRAP:
101 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
102 			fallthrough;
103 		case TC_ACT_SHOT:
104 			return 0;
105 		}
106 #endif
107 		if (TC_H_MIN(res.classid) <= q->flows_cnt)
108 			return TC_H_MIN(res.classid);
109 	}
110 	return 0;
111 }
112 
113 /* helper functions : might be changed when/if skb use a standard list_head */
114 
115 /* remove one skb from head of slot queue */
116 static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow)
117 {
118 	struct sk_buff *skb = flow->head;
119 
120 	flow->head = skb->next;
121 	skb_mark_not_on_list(skb);
122 	return skb;
123 }
124 
125 /* add skb to flow queue (tail add) */
126 static inline void flow_queue_add(struct fq_codel_flow *flow,
127 				  struct sk_buff *skb)
128 {
129 	if (flow->head == NULL)
130 		flow->head = skb;
131 	else
132 		flow->tail->next = skb;
133 	flow->tail = skb;
134 	skb->next = NULL;
135 }
136 
137 static unsigned int fq_codel_drop(struct Qdisc *sch, unsigned int max_packets,
138 				  struct sk_buff **to_free)
139 {
140 	struct fq_codel_sched_data *q = qdisc_priv(sch);
141 	struct sk_buff *skb;
142 	unsigned int maxbacklog = 0, idx = 0, i, len;
143 	struct fq_codel_flow *flow;
144 	unsigned int threshold;
145 	unsigned int mem = 0;
146 
147 	/* Queue is full! Find the fat flow and drop packet(s) from it.
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.
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.
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 	}
160 
161 	/* Our goal is to drop half of this fat flow backlog */
162 	threshold = maxbacklog >> 1;
163 
164 	flow = &q->flows[idx];
165 	len = 0;
166 	i = 0;
167 	do {
168 		skb = dequeue_head(flow);
169 		len += qdisc_pkt_len(skb);
170 		mem += get_codel_cb(skb)->mem_usage;
171 		__qdisc_drop(skb, to_free);
172 	} while (++i < max_packets && len < threshold);
173 
174 	/* Tell codel to increase its signal strength also */
175 	flow->cvars.count += i;
176 	q->backlogs[idx] -= len;
177 	q->memory_usage -= mem;
178 	sch->qstats.drops += i;
179 	sch->qstats.backlog -= len;
180 	sch->q.qlen -= i;
181 	return idx;
182 }
183 
184 static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch,
185 			    struct sk_buff **to_free)
186 {
187 	struct fq_codel_sched_data *q = qdisc_priv(sch);
188 	unsigned int idx, prev_backlog, prev_qlen;
189 	struct fq_codel_flow *flow;
190 	int ret;
191 	unsigned int pkt_len;
192 	bool memory_limited;
193 
194 	idx = fq_codel_classify(skb, sch, &ret);
195 	if (idx == 0) {
196 		if (ret & __NET_XMIT_BYPASS)
197 			qdisc_qstats_drop(sch);
198 		__qdisc_drop(skb, to_free);
199 		return ret;
200 	}
201 	idx--;
202 
203 	codel_set_enqueue_time(skb);
204 	flow = &q->flows[idx];
205 	flow_queue_add(flow, skb);
206 	q->backlogs[idx] += qdisc_pkt_len(skb);
207 	qdisc_qstats_backlog_inc(sch, skb);
208 
209 	if (list_empty(&flow->flowchain)) {
210 		list_add_tail(&flow->flowchain, &q->new_flows);
211 		q->new_flow_count++;
212 		flow->deficit = q->quantum;
213 	}
214 	get_codel_cb(skb)->mem_usage = skb->truesize;
215 	q->memory_usage += get_codel_cb(skb)->mem_usage;
216 	memory_limited = q->memory_usage > q->memory_limit;
217 	if (++sch->q.qlen <= sch->limit && !memory_limited)
218 		return NET_XMIT_SUCCESS;
219 
220 	prev_backlog = sch->qstats.backlog;
221 	prev_qlen = sch->q.qlen;
222 
223 	/* save this packet length as it might be dropped by fq_codel_drop() */
224 	pkt_len = qdisc_pkt_len(skb);
225 	/* fq_codel_drop() is quite expensive, as it performs a linear search
226 	 * in q->backlogs[] to find a fat flow.
227 	 * So instead of dropping a single packet, drop half of its backlog
228 	 * with a 64 packets limit to not add a too big cpu spike here.
229 	 */
230 	ret = fq_codel_drop(sch, q->drop_batch_size, to_free);
231 
232 	prev_qlen -= sch->q.qlen;
233 	prev_backlog -= sch->qstats.backlog;
234 	q->drop_overlimit += prev_qlen;
235 	if (memory_limited)
236 		q->drop_overmemory += prev_qlen;
237 
238 	/* As we dropped packet(s), better let upper stack know this.
239 	 * If we dropped a packet for this flow, return NET_XMIT_CN,
240 	 * but in this case, our parents wont increase their backlogs.
241 	 */
242 	if (ret == idx) {
243 		qdisc_tree_reduce_backlog(sch, prev_qlen - 1,
244 					  prev_backlog - pkt_len);
245 		return NET_XMIT_CN;
246 	}
247 	qdisc_tree_reduce_backlog(sch, prev_qlen, prev_backlog);
248 	return NET_XMIT_SUCCESS;
249 }
250 
251 /* This is the specific function called from codel_dequeue()
252  * to dequeue a packet from queue. Note: backlog is handled in
253  * codel, we dont need to reduce it here.
254  */
255 static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx)
256 {
257 	struct Qdisc *sch = ctx;
258 	struct fq_codel_sched_data *q = qdisc_priv(sch);
259 	struct fq_codel_flow *flow;
260 	struct sk_buff *skb = NULL;
261 
262 	flow = container_of(vars, struct fq_codel_flow, cvars);
263 	if (flow->head) {
264 		skb = dequeue_head(flow);
265 		q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb);
266 		q->memory_usage -= get_codel_cb(skb)->mem_usage;
267 		sch->q.qlen--;
268 		sch->qstats.backlog -= qdisc_pkt_len(skb);
269 	}
270 	return skb;
271 }
272 
273 static void drop_func(struct sk_buff *skb, void *ctx)
274 {
275 	struct Qdisc *sch = ctx;
276 
277 	kfree_skb(skb);
278 	qdisc_qstats_drop(sch);
279 }
280 
281 static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch)
282 {
283 	struct fq_codel_sched_data *q = qdisc_priv(sch);
284 	struct sk_buff *skb;
285 	struct fq_codel_flow *flow;
286 	struct list_head *head;
287 
288 begin:
289 	head = &q->new_flows;
290 	if (list_empty(head)) {
291 		head = &q->old_flows;
292 		if (list_empty(head))
293 			return NULL;
294 	}
295 	flow = list_first_entry(head, struct fq_codel_flow, flowchain);
296 
297 	if (flow->deficit <= 0) {
298 		flow->deficit += q->quantum;
299 		list_move_tail(&flow->flowchain, &q->old_flows);
300 		goto begin;
301 	}
302 
303 	skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams,
304 			    &flow->cvars, &q->cstats, qdisc_pkt_len,
305 			    codel_get_enqueue_time, drop_func, dequeue_func);
306 
307 	if (!skb) {
308 		/* force a pass through old_flows to prevent starvation */
309 		if ((head == &q->new_flows) && !list_empty(&q->old_flows))
310 			list_move_tail(&flow->flowchain, &q->old_flows);
311 		else
312 			list_del_init(&flow->flowchain);
313 		goto begin;
314 	}
315 	qdisc_bstats_update(sch, skb);
316 	flow->deficit -= qdisc_pkt_len(skb);
317 	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
318 	 * or HTB crashes. Defer it for next round.
319 	 */
320 	if (q->cstats.drop_count && sch->q.qlen) {
321 		qdisc_tree_reduce_backlog(sch, q->cstats.drop_count,
322 					  q->cstats.drop_len);
323 		q->cstats.drop_count = 0;
324 		q->cstats.drop_len = 0;
325 	}
326 	return skb;
327 }
328 
329 static void fq_codel_flow_purge(struct fq_codel_flow *flow)
330 {
331 	rtnl_kfree_skbs(flow->head, flow->tail);
332 	flow->head = NULL;
333 }
334 
335 static void fq_codel_reset(struct Qdisc *sch)
336 {
337 	struct fq_codel_sched_data *q = qdisc_priv(sch);
338 	int i;
339 
340 	INIT_LIST_HEAD(&q->new_flows);
341 	INIT_LIST_HEAD(&q->old_flows);
342 	for (i = 0; i < q->flows_cnt; i++) {
343 		struct fq_codel_flow *flow = q->flows + i;
344 
345 		fq_codel_flow_purge(flow);
346 		INIT_LIST_HEAD(&flow->flowchain);
347 		codel_vars_init(&flow->cvars);
348 	}
349 	memset(q->backlogs, 0, q->flows_cnt * sizeof(u32));
350 	sch->q.qlen = 0;
351 	sch->qstats.backlog = 0;
352 	q->memory_usage = 0;
353 }
354 
355 static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = {
356 	[TCA_FQ_CODEL_TARGET]	= { .type = NLA_U32 },
357 	[TCA_FQ_CODEL_LIMIT]	= { .type = NLA_U32 },
358 	[TCA_FQ_CODEL_INTERVAL]	= { .type = NLA_U32 },
359 	[TCA_FQ_CODEL_ECN]	= { .type = NLA_U32 },
360 	[TCA_FQ_CODEL_FLOWS]	= { .type = NLA_U32 },
361 	[TCA_FQ_CODEL_QUANTUM]	= { .type = NLA_U32 },
362 	[TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 },
363 	[TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 },
364 	[TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 },
365 };
366 
367 static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt,
368 			   struct netlink_ext_ack *extack)
369 {
370 	struct fq_codel_sched_data *q = qdisc_priv(sch);
371 	struct nlattr *tb[TCA_FQ_CODEL_MAX + 1];
372 	u32 quantum = 0;
373 	int err;
374 
375 	if (!opt)
376 		return -EINVAL;
377 
378 	err = nla_parse_nested_deprecated(tb, TCA_FQ_CODEL_MAX, opt,
379 					  fq_codel_policy, NULL);
380 	if (err < 0)
381 		return err;
382 	if (tb[TCA_FQ_CODEL_FLOWS]) {
383 		if (q->flows)
384 			return -EINVAL;
385 		q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]);
386 		if (!q->flows_cnt ||
387 		    q->flows_cnt > 65536)
388 			return -EINVAL;
389 	}
390 	if (tb[TCA_FQ_CODEL_QUANTUM]) {
391 		quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM]));
392 		if (quantum > FQ_CODEL_QUANTUM_MAX) {
393 			NL_SET_ERR_MSG(extack, "Invalid quantum");
394 			return -EINVAL;
395 		}
396 	}
397 	sch_tree_lock(sch);
398 
399 	if (tb[TCA_FQ_CODEL_TARGET]) {
400 		u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]);
401 
402 		q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT;
403 	}
404 
405 	if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) {
406 		u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]);
407 
408 		q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
409 	}
410 
411 	if (tb[TCA_FQ_CODEL_INTERVAL]) {
412 		u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]);
413 
414 		q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT;
415 	}
416 
417 	if (tb[TCA_FQ_CODEL_LIMIT])
418 		sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]);
419 
420 	if (tb[TCA_FQ_CODEL_ECN])
421 		q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]);
422 
423 	if (quantum)
424 		q->quantum = quantum;
425 
426 	if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])
427 		q->drop_batch_size = max(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]));
428 
429 	if (tb[TCA_FQ_CODEL_MEMORY_LIMIT])
430 		q->memory_limit = min(1U << 31, nla_get_u32(tb[TCA_FQ_CODEL_MEMORY_LIMIT]));
431 
432 	while (sch->q.qlen > sch->limit ||
433 	       q->memory_usage > q->memory_limit) {
434 		struct sk_buff *skb = fq_codel_dequeue(sch);
435 
436 		q->cstats.drop_len += qdisc_pkt_len(skb);
437 		rtnl_kfree_skbs(skb, skb);
438 		q->cstats.drop_count++;
439 	}
440 	qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len);
441 	q->cstats.drop_count = 0;
442 	q->cstats.drop_len = 0;
443 
444 	sch_tree_unlock(sch);
445 	return 0;
446 }
447 
448 static void fq_codel_destroy(struct Qdisc *sch)
449 {
450 	struct fq_codel_sched_data *q = qdisc_priv(sch);
451 
452 	tcf_block_put(q->block);
453 	kvfree(q->backlogs);
454 	kvfree(q->flows);
455 }
456 
457 static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt,
458 			 struct netlink_ext_ack *extack)
459 {
460 	struct fq_codel_sched_data *q = qdisc_priv(sch);
461 	int i;
462 	int err;
463 
464 	sch->limit = 10*1024;
465 	q->flows_cnt = 1024;
466 	q->memory_limit = 32 << 20; /* 32 MBytes */
467 	q->drop_batch_size = 64;
468 	q->quantum = psched_mtu(qdisc_dev(sch));
469 	INIT_LIST_HEAD(&q->new_flows);
470 	INIT_LIST_HEAD(&q->old_flows);
471 	codel_params_init(&q->cparams);
472 	codel_stats_init(&q->cstats);
473 	q->cparams.ecn = true;
474 	q->cparams.mtu = psched_mtu(qdisc_dev(sch));
475 
476 	if (opt) {
477 		err = fq_codel_change(sch, opt, extack);
478 		if (err)
479 			goto init_failure;
480 	}
481 
482 	err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
483 	if (err)
484 		goto init_failure;
485 
486 	if (!q->flows) {
487 		q->flows = kvcalloc(q->flows_cnt,
488 				    sizeof(struct fq_codel_flow),
489 				    GFP_KERNEL);
490 		if (!q->flows) {
491 			err = -ENOMEM;
492 			goto init_failure;
493 		}
494 		q->backlogs = kvcalloc(q->flows_cnt, sizeof(u32), GFP_KERNEL);
495 		if (!q->backlogs) {
496 			err = -ENOMEM;
497 			goto alloc_failure;
498 		}
499 		for (i = 0; i < q->flows_cnt; i++) {
500 			struct fq_codel_flow *flow = q->flows + i;
501 
502 			INIT_LIST_HEAD(&flow->flowchain);
503 			codel_vars_init(&flow->cvars);
504 		}
505 	}
506 	if (sch->limit >= 1)
507 		sch->flags |= TCQ_F_CAN_BYPASS;
508 	else
509 		sch->flags &= ~TCQ_F_CAN_BYPASS;
510 	return 0;
511 
512 alloc_failure:
513 	kvfree(q->flows);
514 	q->flows = NULL;
515 init_failure:
516 	q->flows_cnt = 0;
517 	return err;
518 }
519 
520 static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb)
521 {
522 	struct fq_codel_sched_data *q = qdisc_priv(sch);
523 	struct nlattr *opts;
524 
525 	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
526 	if (opts == NULL)
527 		goto nla_put_failure;
528 
529 	if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET,
530 			codel_time_to_us(q->cparams.target)) ||
531 	    nla_put_u32(skb, TCA_FQ_CODEL_LIMIT,
532 			sch->limit) ||
533 	    nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL,
534 			codel_time_to_us(q->cparams.interval)) ||
535 	    nla_put_u32(skb, TCA_FQ_CODEL_ECN,
536 			q->cparams.ecn) ||
537 	    nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM,
538 			q->quantum) ||
539 	    nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE,
540 			q->drop_batch_size) ||
541 	    nla_put_u32(skb, TCA_FQ_CODEL_MEMORY_LIMIT,
542 			q->memory_limit) ||
543 	    nla_put_u32(skb, TCA_FQ_CODEL_FLOWS,
544 			q->flows_cnt))
545 		goto nla_put_failure;
546 
547 	if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD &&
548 	    nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD,
549 			codel_time_to_us(q->cparams.ce_threshold)))
550 		goto nla_put_failure;
551 
552 	return nla_nest_end(skb, opts);
553 
554 nla_put_failure:
555 	return -1;
556 }
557 
558 static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
559 {
560 	struct fq_codel_sched_data *q = qdisc_priv(sch);
561 	struct tc_fq_codel_xstats st = {
562 		.type				= TCA_FQ_CODEL_XSTATS_QDISC,
563 	};
564 	struct list_head *pos;
565 
566 	st.qdisc_stats.maxpacket = q->cstats.maxpacket;
567 	st.qdisc_stats.drop_overlimit = q->drop_overlimit;
568 	st.qdisc_stats.ecn_mark = q->cstats.ecn_mark;
569 	st.qdisc_stats.new_flow_count = q->new_flow_count;
570 	st.qdisc_stats.ce_mark = q->cstats.ce_mark;
571 	st.qdisc_stats.memory_usage  = q->memory_usage;
572 	st.qdisc_stats.drop_overmemory = q->drop_overmemory;
573 
574 	sch_tree_lock(sch);
575 	list_for_each(pos, &q->new_flows)
576 		st.qdisc_stats.new_flows_len++;
577 
578 	list_for_each(pos, &q->old_flows)
579 		st.qdisc_stats.old_flows_len++;
580 	sch_tree_unlock(sch);
581 
582 	return gnet_stats_copy_app(d, &st, sizeof(st));
583 }
584 
585 static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg)
586 {
587 	return NULL;
588 }
589 
590 static unsigned long fq_codel_find(struct Qdisc *sch, u32 classid)
591 {
592 	return 0;
593 }
594 
595 static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent,
596 			      u32 classid)
597 {
598 	return 0;
599 }
600 
601 static void fq_codel_unbind(struct Qdisc *q, unsigned long cl)
602 {
603 }
604 
605 static struct tcf_block *fq_codel_tcf_block(struct Qdisc *sch, unsigned long cl,
606 					    struct netlink_ext_ack *extack)
607 {
608 	struct fq_codel_sched_data *q = qdisc_priv(sch);
609 
610 	if (cl)
611 		return NULL;
612 	return q->block;
613 }
614 
615 static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl,
616 			  struct sk_buff *skb, struct tcmsg *tcm)
617 {
618 	tcm->tcm_handle |= TC_H_MIN(cl);
619 	return 0;
620 }
621 
622 static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl,
623 				     struct gnet_dump *d)
624 {
625 	struct fq_codel_sched_data *q = qdisc_priv(sch);
626 	u32 idx = cl - 1;
627 	struct gnet_stats_queue qs = { 0 };
628 	struct tc_fq_codel_xstats xstats;
629 
630 	if (idx < q->flows_cnt) {
631 		const struct fq_codel_flow *flow = &q->flows[idx];
632 		const struct sk_buff *skb;
633 
634 		memset(&xstats, 0, sizeof(xstats));
635 		xstats.type = TCA_FQ_CODEL_XSTATS_CLASS;
636 		xstats.class_stats.deficit = flow->deficit;
637 		xstats.class_stats.ldelay =
638 			codel_time_to_us(flow->cvars.ldelay);
639 		xstats.class_stats.count = flow->cvars.count;
640 		xstats.class_stats.lastcount = flow->cvars.lastcount;
641 		xstats.class_stats.dropping = flow->cvars.dropping;
642 		if (flow->cvars.dropping) {
643 			codel_tdiff_t delta = flow->cvars.drop_next -
644 					      codel_get_time();
645 
646 			xstats.class_stats.drop_next = (delta >= 0) ?
647 				codel_time_to_us(delta) :
648 				-codel_time_to_us(-delta);
649 		}
650 		if (flow->head) {
651 			sch_tree_lock(sch);
652 			skb = flow->head;
653 			while (skb) {
654 				qs.qlen++;
655 				skb = skb->next;
656 			}
657 			sch_tree_unlock(sch);
658 		}
659 		qs.backlog = q->backlogs[idx];
660 		qs.drops = 0;
661 	}
662 	if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0)
663 		return -1;
664 	if (idx < q->flows_cnt)
665 		return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
666 	return 0;
667 }
668 
669 static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg)
670 {
671 	struct fq_codel_sched_data *q = qdisc_priv(sch);
672 	unsigned int i;
673 
674 	if (arg->stop)
675 		return;
676 
677 	for (i = 0; i < q->flows_cnt; i++) {
678 		if (list_empty(&q->flows[i].flowchain) ||
679 		    arg->count < arg->skip) {
680 			arg->count++;
681 			continue;
682 		}
683 		if (arg->fn(sch, i + 1, arg) < 0) {
684 			arg->stop = 1;
685 			break;
686 		}
687 		arg->count++;
688 	}
689 }
690 
691 static const struct Qdisc_class_ops fq_codel_class_ops = {
692 	.leaf		=	fq_codel_leaf,
693 	.find		=	fq_codel_find,
694 	.tcf_block	=	fq_codel_tcf_block,
695 	.bind_tcf	=	fq_codel_bind,
696 	.unbind_tcf	=	fq_codel_unbind,
697 	.dump		=	fq_codel_dump_class,
698 	.dump_stats	=	fq_codel_dump_class_stats,
699 	.walk		=	fq_codel_walk,
700 };
701 
702 static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = {
703 	.cl_ops		=	&fq_codel_class_ops,
704 	.id		=	"fq_codel",
705 	.priv_size	=	sizeof(struct fq_codel_sched_data),
706 	.enqueue	=	fq_codel_enqueue,
707 	.dequeue	=	fq_codel_dequeue,
708 	.peek		=	qdisc_peek_dequeued,
709 	.init		=	fq_codel_init,
710 	.reset		=	fq_codel_reset,
711 	.destroy	=	fq_codel_destroy,
712 	.change		=	fq_codel_change,
713 	.dump		=	fq_codel_dump,
714 	.dump_stats =	fq_codel_dump_stats,
715 	.owner		=	THIS_MODULE,
716 };
717 
718 static int __init fq_codel_module_init(void)
719 {
720 	return register_qdisc(&fq_codel_qdisc_ops);
721 }
722 
723 static void __exit fq_codel_module_exit(void)
724 {
725 	unregister_qdisc(&fq_codel_qdisc_ops);
726 }
727 
728 module_init(fq_codel_module_init)
729 module_exit(fq_codel_module_exit)
730 MODULE_AUTHOR("Eric Dumazet");
731 MODULE_LICENSE("GPL");
732 MODULE_DESCRIPTION("Fair Queue CoDel discipline");
733