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
fq_codel_hash(const struct fq_codel_sched_data * q,struct sk_buff * skb)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
fq_codel_classify(struct sk_buff * skb,struct Qdisc * sch,int * qerr)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 */
dequeue_head(struct fq_codel_flow * flow)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) */
flow_queue_add(struct fq_codel_flow * flow,struct sk_buff * skb)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
fq_codel_drop(struct Qdisc * sch,unsigned int max_packets,struct sk_buff ** to_free)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
fq_codel_enqueue(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free)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 */
dequeue_func(struct codel_vars * vars,void * ctx)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
drop_func(struct sk_buff * skb,void * ctx)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
fq_codel_dequeue(struct Qdisc * sch)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
fq_codel_flow_purge(struct fq_codel_flow * flow)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
fq_codel_reset(struct Qdisc * sch)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 q->memory_usage = 0;
351 }
352
353 static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = {
354 [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 },
355 [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 },
356 [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 },
357 [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 },
358 [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 },
359 [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 },
360 [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 },
361 [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 },
362 [TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 },
363 [TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR] = { .type = NLA_U8 },
364 [TCA_FQ_CODEL_CE_THRESHOLD_MASK] = { .type = NLA_U8 },
365 };
366
fq_codel_change(struct Qdisc * sch,struct nlattr * opt,struct netlink_ext_ack * extack)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 err = nla_parse_nested_deprecated(tb, TCA_FQ_CODEL_MAX, opt,
376 fq_codel_policy, NULL);
377 if (err < 0)
378 return err;
379 if (tb[TCA_FQ_CODEL_FLOWS]) {
380 if (q->flows)
381 return -EINVAL;
382 q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]);
383 if (!q->flows_cnt ||
384 q->flows_cnt > 65536)
385 return -EINVAL;
386 }
387 if (tb[TCA_FQ_CODEL_QUANTUM]) {
388 quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM]));
389 if (quantum > FQ_CODEL_QUANTUM_MAX) {
390 NL_SET_ERR_MSG(extack, "Invalid quantum");
391 return -EINVAL;
392 }
393 }
394 sch_tree_lock(sch);
395
396 if (tb[TCA_FQ_CODEL_TARGET]) {
397 u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]);
398
399 WRITE_ONCE(q->cparams.target,
400 (target * NSEC_PER_USEC) >> CODEL_SHIFT);
401 }
402
403 if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) {
404 u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]);
405
406 WRITE_ONCE(q->cparams.ce_threshold,
407 (val * NSEC_PER_USEC) >> CODEL_SHIFT);
408 }
409
410 if (tb[TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR])
411 WRITE_ONCE(q->cparams.ce_threshold_selector,
412 nla_get_u8(tb[TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR]));
413 if (tb[TCA_FQ_CODEL_CE_THRESHOLD_MASK])
414 WRITE_ONCE(q->cparams.ce_threshold_mask,
415 nla_get_u8(tb[TCA_FQ_CODEL_CE_THRESHOLD_MASK]));
416
417 if (tb[TCA_FQ_CODEL_INTERVAL]) {
418 u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]);
419
420 WRITE_ONCE(q->cparams.interval,
421 (interval * NSEC_PER_USEC) >> CODEL_SHIFT);
422 }
423
424 if (tb[TCA_FQ_CODEL_LIMIT])
425 WRITE_ONCE(sch->limit,
426 nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]));
427
428 if (tb[TCA_FQ_CODEL_ECN])
429 WRITE_ONCE(q->cparams.ecn,
430 !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]));
431
432 if (quantum)
433 WRITE_ONCE(q->quantum, quantum);
434
435 if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])
436 WRITE_ONCE(q->drop_batch_size,
437 max(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])));
438
439 if (tb[TCA_FQ_CODEL_MEMORY_LIMIT])
440 WRITE_ONCE(q->memory_limit,
441 min(1U << 31, nla_get_u32(tb[TCA_FQ_CODEL_MEMORY_LIMIT])));
442
443 while (sch->q.qlen > sch->limit ||
444 q->memory_usage > q->memory_limit) {
445 struct sk_buff *skb = fq_codel_dequeue(sch);
446
447 q->cstats.drop_len += qdisc_pkt_len(skb);
448 rtnl_kfree_skbs(skb, skb);
449 q->cstats.drop_count++;
450 }
451 qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len);
452 q->cstats.drop_count = 0;
453 q->cstats.drop_len = 0;
454
455 sch_tree_unlock(sch);
456 return 0;
457 }
458
fq_codel_destroy(struct Qdisc * sch)459 static void fq_codel_destroy(struct Qdisc *sch)
460 {
461 struct fq_codel_sched_data *q = qdisc_priv(sch);
462
463 tcf_block_put(q->block);
464 kvfree(q->backlogs);
465 kvfree(q->flows);
466 }
467
fq_codel_init(struct Qdisc * sch,struct nlattr * opt,struct netlink_ext_ack * extack)468 static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt,
469 struct netlink_ext_ack *extack)
470 {
471 struct fq_codel_sched_data *q = qdisc_priv(sch);
472 int i;
473 int err;
474
475 sch->limit = 10*1024;
476 q->flows_cnt = 1024;
477 q->memory_limit = 32 << 20; /* 32 MBytes */
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
523 alloc_failure:
524 kvfree(q->flows);
525 q->flows = NULL;
526 init_failure:
527 q->flows_cnt = 0;
528 return err;
529 }
530
fq_codel_dump(struct Qdisc * sch,struct sk_buff * skb)531 static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb)
532 {
533 struct fq_codel_sched_data *q = qdisc_priv(sch);
534 codel_time_t ce_threshold;
535 struct nlattr *opts;
536
537 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
538 if (opts == NULL)
539 goto nla_put_failure;
540
541 if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET,
542 codel_time_to_us(READ_ONCE(q->cparams.target))) ||
543 nla_put_u32(skb, TCA_FQ_CODEL_LIMIT,
544 READ_ONCE(sch->limit)) ||
545 nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL,
546 codel_time_to_us(READ_ONCE(q->cparams.interval))) ||
547 nla_put_u32(skb, TCA_FQ_CODEL_ECN,
548 READ_ONCE(q->cparams.ecn)) ||
549 nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM,
550 READ_ONCE(q->quantum)) ||
551 nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE,
552 READ_ONCE(q->drop_batch_size)) ||
553 nla_put_u32(skb, TCA_FQ_CODEL_MEMORY_LIMIT,
554 READ_ONCE(q->memory_limit)) ||
555 nla_put_u32(skb, TCA_FQ_CODEL_FLOWS,
556 READ_ONCE(q->flows_cnt)))
557 goto nla_put_failure;
558
559 ce_threshold = READ_ONCE(q->cparams.ce_threshold);
560 if (ce_threshold != CODEL_DISABLED_THRESHOLD) {
561 if (nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD,
562 codel_time_to_us(ce_threshold)))
563 goto nla_put_failure;
564 if (nla_put_u8(skb, TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR,
565 READ_ONCE(q->cparams.ce_threshold_selector)))
566 goto nla_put_failure;
567 if (nla_put_u8(skb, TCA_FQ_CODEL_CE_THRESHOLD_MASK,
568 READ_ONCE(q->cparams.ce_threshold_mask)))
569 goto nla_put_failure;
570 }
571
572 return nla_nest_end(skb, opts);
573
574 nla_put_failure:
575 return -1;
576 }
577
fq_codel_dump_stats(struct Qdisc * sch,struct gnet_dump * d)578 static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
579 {
580 struct fq_codel_sched_data *q = qdisc_priv(sch);
581 struct tc_fq_codel_xstats st = {
582 .type = TCA_FQ_CODEL_XSTATS_QDISC,
583 };
584 struct list_head *pos;
585
586 st.qdisc_stats.maxpacket = q->cstats.maxpacket;
587 st.qdisc_stats.drop_overlimit = q->drop_overlimit;
588 st.qdisc_stats.ecn_mark = q->cstats.ecn_mark;
589 st.qdisc_stats.new_flow_count = q->new_flow_count;
590 st.qdisc_stats.ce_mark = q->cstats.ce_mark;
591 st.qdisc_stats.memory_usage = q->memory_usage;
592 st.qdisc_stats.drop_overmemory = q->drop_overmemory;
593
594 sch_tree_lock(sch);
595 list_for_each(pos, &q->new_flows)
596 st.qdisc_stats.new_flows_len++;
597
598 list_for_each(pos, &q->old_flows)
599 st.qdisc_stats.old_flows_len++;
600 sch_tree_unlock(sch);
601
602 return gnet_stats_copy_app(d, &st, sizeof(st));
603 }
604
fq_codel_leaf(struct Qdisc * sch,unsigned long arg)605 static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg)
606 {
607 return NULL;
608 }
609
fq_codel_find(struct Qdisc * sch,u32 classid)610 static unsigned long fq_codel_find(struct Qdisc *sch, u32 classid)
611 {
612 return 0;
613 }
614
fq_codel_bind(struct Qdisc * sch,unsigned long parent,u32 classid)615 static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent,
616 u32 classid)
617 {
618 return 0;
619 }
620
fq_codel_unbind(struct Qdisc * q,unsigned long cl)621 static void fq_codel_unbind(struct Qdisc *q, unsigned long cl)
622 {
623 }
624
fq_codel_tcf_block(struct Qdisc * sch,unsigned long cl,struct netlink_ext_ack * extack)625 static struct tcf_block *fq_codel_tcf_block(struct Qdisc *sch, unsigned long cl,
626 struct netlink_ext_ack *extack)
627 {
628 struct fq_codel_sched_data *q = qdisc_priv(sch);
629
630 if (cl)
631 return NULL;
632 return q->block;
633 }
634
fq_codel_dump_class(struct Qdisc * sch,unsigned long cl,struct sk_buff * skb,struct tcmsg * tcm)635 static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl,
636 struct sk_buff *skb, struct tcmsg *tcm)
637 {
638 tcm->tcm_handle |= TC_H_MIN(cl);
639 return 0;
640 }
641
fq_codel_dump_class_stats(struct Qdisc * sch,unsigned long cl,struct gnet_dump * d)642 static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl,
643 struct gnet_dump *d)
644 {
645 struct fq_codel_sched_data *q = qdisc_priv(sch);
646 u32 idx = cl - 1;
647 struct gnet_stats_queue qs = { 0 };
648 struct tc_fq_codel_xstats xstats;
649
650 if (idx < q->flows_cnt) {
651 const struct fq_codel_flow *flow = &q->flows[idx];
652 const struct sk_buff *skb;
653
654 memset(&xstats, 0, sizeof(xstats));
655 xstats.type = TCA_FQ_CODEL_XSTATS_CLASS;
656 xstats.class_stats.deficit = flow->deficit;
657 xstats.class_stats.ldelay =
658 codel_time_to_us(flow->cvars.ldelay);
659 xstats.class_stats.count = flow->cvars.count;
660 xstats.class_stats.lastcount = flow->cvars.lastcount;
661 xstats.class_stats.dropping = flow->cvars.dropping;
662 if (flow->cvars.dropping) {
663 codel_tdiff_t delta = flow->cvars.drop_next -
664 codel_get_time();
665
666 xstats.class_stats.drop_next = (delta >= 0) ?
667 codel_time_to_us(delta) :
668 -codel_time_to_us(-delta);
669 }
670 if (flow->head) {
671 sch_tree_lock(sch);
672 skb = flow->head;
673 while (skb) {
674 qs.qlen++;
675 skb = skb->next;
676 }
677 sch_tree_unlock(sch);
678 }
679 qs.backlog = q->backlogs[idx];
680 qs.drops = 0;
681 }
682 if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0)
683 return -1;
684 if (idx < q->flows_cnt)
685 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
686 return 0;
687 }
688
fq_codel_walk(struct Qdisc * sch,struct qdisc_walker * arg)689 static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg)
690 {
691 struct fq_codel_sched_data *q = qdisc_priv(sch);
692 unsigned int i;
693
694 if (arg->stop)
695 return;
696
697 for (i = 0; i < q->flows_cnt; i++) {
698 if (list_empty(&q->flows[i].flowchain)) {
699 arg->count++;
700 continue;
701 }
702 if (!tc_qdisc_stats_dump(sch, i + 1, arg))
703 break;
704 }
705 }
706
707 static const struct Qdisc_class_ops fq_codel_class_ops = {
708 .leaf = fq_codel_leaf,
709 .find = fq_codel_find,
710 .tcf_block = fq_codel_tcf_block,
711 .bind_tcf = fq_codel_bind,
712 .unbind_tcf = fq_codel_unbind,
713 .dump = fq_codel_dump_class,
714 .dump_stats = fq_codel_dump_class_stats,
715 .walk = fq_codel_walk,
716 };
717
718 static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = {
719 .cl_ops = &fq_codel_class_ops,
720 .id = "fq_codel",
721 .priv_size = sizeof(struct fq_codel_sched_data),
722 .enqueue = fq_codel_enqueue,
723 .dequeue = fq_codel_dequeue,
724 .peek = qdisc_peek_dequeued,
725 .init = fq_codel_init,
726 .reset = fq_codel_reset,
727 .destroy = fq_codel_destroy,
728 .change = fq_codel_change,
729 .dump = fq_codel_dump,
730 .dump_stats = fq_codel_dump_stats,
731 .owner = THIS_MODULE,
732 };
733 MODULE_ALIAS_NET_SCH("fq_codel");
734
fq_codel_module_init(void)735 static int __init fq_codel_module_init(void)
736 {
737 return register_qdisc(&fq_codel_qdisc_ops);
738 }
739
fq_codel_module_exit(void)740 static void __exit fq_codel_module_exit(void)
741 {
742 unregister_qdisc(&fq_codel_qdisc_ops);
743 }
744
745 module_init(fq_codel_module_init)
746 module_exit(fq_codel_module_exit)
747 MODULE_AUTHOR("Eric Dumazet");
748 MODULE_LICENSE("GPL");
749 MODULE_DESCRIPTION("Fair Queue CoDel discipline");
750