Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/sched/sch_choke.c CHOKE scheduler
4 *
5 * Copyright (c) 2011 Stephen Hemminger <shemminger@vyatta.com>
6 * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com>
7 */
8
9#include <linux/module.h>
10#include <linux/types.h>
11#include <linux/kernel.h>
12#include <linux/skbuff.h>
13#include <linux/vmalloc.h>
14#include <net/pkt_sched.h>
15#include <net/pkt_cls.h>
16#include <net/inet_ecn.h>
17#include <net/red.h>
18#include <net/flow_dissector.h>
19
20/*
21 CHOKe stateless AQM for fair bandwidth allocation
22 =================================================
23
24 CHOKe (CHOose and Keep for responsive flows, CHOose and Kill for
25 unresponsive flows) is a variant of RED that penalizes misbehaving flows but
26 maintains no flow state. The difference from RED is an additional step
27 during the enqueuing process. If average queue size is over the
28 low threshold (qmin), a packet is chosen at random from the queue.
29 If both the new and chosen packet are from the same flow, both
30 are dropped. Unlike RED, CHOKe is not really a "classful" qdisc because it
31 needs to access packets in queue randomly. It has a minimal class
32 interface to allow overriding the builtin flow classifier with
33 filters.
34
35 Source:
36 R. Pan, B. Prabhakar, and K. Psounis, "CHOKe, A Stateless
37 Active Queue Management Scheme for Approximating Fair Bandwidth Allocation",
38 IEEE INFOCOM, 2000.
39
40 A. Tang, J. Wang, S. Low, "Understanding CHOKe: Throughput and Spatial
41 Characteristics", IEEE/ACM Transactions on Networking, 2004
42
43 */
44
45/* Upper bound on size of sk_buff table (packets) */
46#define CHOKE_MAX_QUEUE (128*1024 - 1)
47
48struct choke_sched_data {
49/* Parameters */
50 u32 limit;
51 unsigned char flags;
52
53 struct red_parms parms;
54
55/* Variables */
56 struct red_vars vars;
57 struct {
58 u32 prob_drop; /* Early probability drops */
59 u32 prob_mark; /* Early probability marks */
60 u32 forced_drop; /* Forced drops, qavg > max_thresh */
61 u32 forced_mark; /* Forced marks, qavg > max_thresh */
62 u32 pdrop; /* Drops due to queue limits */
63 u32 matched; /* Drops to flow match */
64 } stats;
65
66 unsigned int head;
67 unsigned int tail;
68
69 unsigned int tab_mask; /* size - 1 */
70
71 struct sk_buff **tab;
72};
73
74/* number of elements in queue including holes */
75static unsigned int choke_len(const struct choke_sched_data *q)
76{
77 return (q->tail - q->head) & q->tab_mask;
78}
79
80/* Is ECN parameter configured */
81static int use_ecn(const struct choke_sched_data *q)
82{
83 return q->flags & TC_RED_ECN;
84}
85
86/* Should packets over max just be dropped (versus marked) */
87static int use_harddrop(const struct choke_sched_data *q)
88{
89 return q->flags & TC_RED_HARDDROP;
90}
91
92/* Move head pointer forward to skip over holes */
93static void choke_zap_head_holes(struct choke_sched_data *q)
94{
95 do {
96 q->head = (q->head + 1) & q->tab_mask;
97 if (q->head == q->tail)
98 break;
99 } while (q->tab[q->head] == NULL);
100}
101
102/* Move tail pointer backwards to reuse holes */
103static void choke_zap_tail_holes(struct choke_sched_data *q)
104{
105 do {
106 q->tail = (q->tail - 1) & q->tab_mask;
107 if (q->head == q->tail)
108 break;
109 } while (q->tab[q->tail] == NULL);
110}
111
112/* Drop packet from queue array by creating a "hole" */
113static void choke_drop_by_idx(struct Qdisc *sch, unsigned int idx,
114 struct sk_buff **to_free)
115{
116 struct choke_sched_data *q = qdisc_priv(sch);
117 struct sk_buff *skb = q->tab[idx];
118
119 q->tab[idx] = NULL;
120
121 if (idx == q->head)
122 choke_zap_head_holes(q);
123 if (idx == q->tail)
124 choke_zap_tail_holes(q);
125
126 --sch->q.qlen;
127 qdisc_qstats_backlog_dec(sch, skb);
128 qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(skb));
129 qdisc_drop(skb, sch, to_free);
130}
131
132struct choke_skb_cb {
133 u8 keys_valid;
134 struct flow_keys_digest keys;
135};
136
137static inline struct choke_skb_cb *choke_skb_cb(const struct sk_buff *skb)
138{
139 qdisc_cb_private_validate(skb, sizeof(struct choke_skb_cb));
140 return (struct choke_skb_cb *)qdisc_skb_cb(skb)->data;
141}
142
143/*
144 * Compare flow of two packets
145 * Returns true only if source and destination address and port match.
146 * false for special cases
147 */
148static bool choke_match_flow(struct sk_buff *skb1,
149 struct sk_buff *skb2)
150{
151 struct flow_keys temp;
152
153 if (skb1->protocol != skb2->protocol)
154 return false;
155
156 if (!choke_skb_cb(skb1)->keys_valid) {
157 choke_skb_cb(skb1)->keys_valid = 1;
158 skb_flow_dissect_flow_keys(skb1, &temp, 0);
159 make_flow_keys_digest(&choke_skb_cb(skb1)->keys, &temp);
160 }
161
162 if (!choke_skb_cb(skb2)->keys_valid) {
163 choke_skb_cb(skb2)->keys_valid = 1;
164 skb_flow_dissect_flow_keys(skb2, &temp, 0);
165 make_flow_keys_digest(&choke_skb_cb(skb2)->keys, &temp);
166 }
167
168 return !memcmp(&choke_skb_cb(skb1)->keys,
169 &choke_skb_cb(skb2)->keys,
170 sizeof(choke_skb_cb(skb1)->keys));
171}
172
173/*
174 * Select a packet at random from queue
175 * HACK: since queue can have holes from previous deletion; retry several
176 * times to find a random skb but then just give up and return the head
177 * Will return NULL if queue is empty (q->head == q->tail)
178 */
179static struct sk_buff *choke_peek_random(const struct choke_sched_data *q,
180 unsigned int *pidx)
181{
182 struct sk_buff *skb;
183 int retrys = 3;
184
185 do {
186 *pidx = (q->head + get_random_u32_below(choke_len(q))) & q->tab_mask;
187 skb = q->tab[*pidx];
188 if (skb)
189 return skb;
190 } while (--retrys > 0);
191
192 return q->tab[*pidx = q->head];
193}
194
195/*
196 * Compare new packet with random packet in queue
197 * returns true if matched and sets *pidx
198 */
199static bool choke_match_random(const struct choke_sched_data *q,
200 struct sk_buff *nskb,
201 unsigned int *pidx)
202{
203 struct sk_buff *oskb;
204
205 if (q->head == q->tail)
206 return false;
207
208 oskb = choke_peek_random(q, pidx);
209 return choke_match_flow(oskb, nskb);
210}
211
212static int choke_enqueue(struct sk_buff *skb, struct Qdisc *sch,
213 struct sk_buff **to_free)
214{
215 struct choke_sched_data *q = qdisc_priv(sch);
216 const struct red_parms *p = &q->parms;
217
218 choke_skb_cb(skb)->keys_valid = 0;
219 /* Compute average queue usage (see RED) */
220 q->vars.qavg = red_calc_qavg(p, &q->vars, sch->q.qlen);
221 if (red_is_idling(&q->vars))
222 red_end_of_idle_period(&q->vars);
223
224 /* Is queue small? */
225 if (q->vars.qavg <= p->qth_min)
226 q->vars.qcount = -1;
227 else {
228 unsigned int idx;
229
230 /* Draw a packet at random from queue and compare flow */
231 if (choke_match_random(q, skb, &idx)) {
232 q->stats.matched++;
233 choke_drop_by_idx(sch, idx, to_free);
234 goto congestion_drop;
235 }
236
237 /* Queue is large, always mark/drop */
238 if (q->vars.qavg > p->qth_max) {
239 q->vars.qcount = -1;
240
241 qdisc_qstats_overlimit(sch);
242 if (use_harddrop(q) || !use_ecn(q) ||
243 !INET_ECN_set_ce(skb)) {
244 q->stats.forced_drop++;
245 goto congestion_drop;
246 }
247
248 q->stats.forced_mark++;
249 } else if (++q->vars.qcount) {
250 if (red_mark_probability(p, &q->vars, q->vars.qavg)) {
251 q->vars.qcount = 0;
252 q->vars.qR = red_random(p);
253
254 qdisc_qstats_overlimit(sch);
255 if (!use_ecn(q) || !INET_ECN_set_ce(skb)) {
256 q->stats.prob_drop++;
257 goto congestion_drop;
258 }
259
260 q->stats.prob_mark++;
261 }
262 } else
263 q->vars.qR = red_random(p);
264 }
265
266 /* Admit new packet */
267 if (sch->q.qlen < q->limit) {
268 q->tab[q->tail] = skb;
269 q->tail = (q->tail + 1) & q->tab_mask;
270 ++sch->q.qlen;
271 qdisc_qstats_backlog_inc(sch, skb);
272 return NET_XMIT_SUCCESS;
273 }
274
275 q->stats.pdrop++;
276 return qdisc_drop(skb, sch, to_free);
277
278congestion_drop:
279 qdisc_drop(skb, sch, to_free);
280 return NET_XMIT_CN;
281}
282
283static struct sk_buff *choke_dequeue(struct Qdisc *sch)
284{
285 struct choke_sched_data *q = qdisc_priv(sch);
286 struct sk_buff *skb;
287
288 if (q->head == q->tail) {
289 if (!red_is_idling(&q->vars))
290 red_start_of_idle_period(&q->vars);
291 return NULL;
292 }
293
294 skb = q->tab[q->head];
295 q->tab[q->head] = NULL;
296 choke_zap_head_holes(q);
297 --sch->q.qlen;
298 qdisc_qstats_backlog_dec(sch, skb);
299 qdisc_bstats_update(sch, skb);
300
301 return skb;
302}
303
304static void choke_reset(struct Qdisc *sch)
305{
306 struct choke_sched_data *q = qdisc_priv(sch);
307
308 while (q->head != q->tail) {
309 struct sk_buff *skb = q->tab[q->head];
310
311 q->head = (q->head + 1) & q->tab_mask;
312 if (!skb)
313 continue;
314 rtnl_qdisc_drop(skb, sch);
315 }
316
317 if (q->tab)
318 memset(q->tab, 0, (q->tab_mask + 1) * sizeof(struct sk_buff *));
319 q->head = q->tail = 0;
320 red_restart(&q->vars);
321}
322
323static const struct nla_policy choke_policy[TCA_CHOKE_MAX + 1] = {
324 [TCA_CHOKE_PARMS] = { .len = sizeof(struct tc_red_qopt) },
325 [TCA_CHOKE_STAB] = { .len = RED_STAB_SIZE },
326 [TCA_CHOKE_MAX_P] = { .type = NLA_U32 },
327};
328
329
330static void choke_free(void *addr)
331{
332 kvfree(addr);
333}
334
335static int choke_change(struct Qdisc *sch, struct nlattr *opt,
336 struct netlink_ext_ack *extack)
337{
338 struct choke_sched_data *q = qdisc_priv(sch);
339 struct nlattr *tb[TCA_CHOKE_MAX + 1];
340 const struct tc_red_qopt *ctl;
341 int err;
342 struct sk_buff **old = NULL;
343 unsigned int mask;
344 u32 max_P;
345 u8 *stab;
346
347 if (opt == NULL)
348 return -EINVAL;
349
350 err = nla_parse_nested_deprecated(tb, TCA_CHOKE_MAX, opt,
351 choke_policy, NULL);
352 if (err < 0)
353 return err;
354
355 if (tb[TCA_CHOKE_PARMS] == NULL ||
356 tb[TCA_CHOKE_STAB] == NULL)
357 return -EINVAL;
358
359 max_P = nla_get_u32_default(tb[TCA_CHOKE_MAX_P], 0);
360
361 ctl = nla_data(tb[TCA_CHOKE_PARMS]);
362 stab = nla_data(tb[TCA_CHOKE_STAB]);
363 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Scell_log, stab))
364 return -EINVAL;
365
366 if (ctl->limit > CHOKE_MAX_QUEUE)
367 return -EINVAL;
368
369 mask = roundup_pow_of_two(ctl->limit + 1) - 1;
370 if (mask != q->tab_mask) {
371 struct sk_buff **ntab;
372
373 ntab = kvcalloc(mask + 1, sizeof(struct sk_buff *), GFP_KERNEL);
374 if (!ntab)
375 return -ENOMEM;
376
377 sch_tree_lock(sch);
378 old = q->tab;
379 if (old) {
380 unsigned int oqlen = sch->q.qlen, tail = 0;
381 unsigned dropped = 0;
382
383 while (q->head != q->tail) {
384 struct sk_buff *skb = q->tab[q->head];
385
386 q->head = (q->head + 1) & q->tab_mask;
387 if (!skb)
388 continue;
389 if (tail < mask) {
390 ntab[tail++] = skb;
391 continue;
392 }
393 dropped += qdisc_pkt_len(skb);
394 qdisc_qstats_backlog_dec(sch, skb);
395 --sch->q.qlen;
396 rtnl_qdisc_drop(skb, sch);
397 }
398 qdisc_tree_reduce_backlog(sch, oqlen - sch->q.qlen, dropped);
399 q->head = 0;
400 q->tail = tail;
401 }
402
403 q->tab_mask = mask;
404 q->tab = ntab;
405 } else
406 sch_tree_lock(sch);
407
408 WRITE_ONCE(q->flags, ctl->flags);
409 WRITE_ONCE(q->limit, ctl->limit);
410
411 red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
412 ctl->Plog, ctl->Scell_log,
413 stab,
414 max_P);
415 red_set_vars(&q->vars);
416
417 if (q->head == q->tail)
418 red_end_of_idle_period(&q->vars);
419
420 sch_tree_unlock(sch);
421 choke_free(old);
422 return 0;
423}
424
425static int choke_init(struct Qdisc *sch, struct nlattr *opt,
426 struct netlink_ext_ack *extack)
427{
428 return choke_change(sch, opt, extack);
429}
430
431static int choke_dump(struct Qdisc *sch, struct sk_buff *skb)
432{
433 struct choke_sched_data *q = qdisc_priv(sch);
434 u8 Wlog = READ_ONCE(q->parms.Wlog);
435 struct nlattr *opts = NULL;
436 struct tc_red_qopt opt = {
437 .limit = READ_ONCE(q->limit),
438 .flags = READ_ONCE(q->flags),
439 .qth_min = READ_ONCE(q->parms.qth_min) >> Wlog,
440 .qth_max = READ_ONCE(q->parms.qth_max) >> Wlog,
441 .Wlog = Wlog,
442 .Plog = READ_ONCE(q->parms.Plog),
443 .Scell_log = READ_ONCE(q->parms.Scell_log),
444 };
445
446 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
447 if (opts == NULL)
448 goto nla_put_failure;
449
450 if (nla_put(skb, TCA_CHOKE_PARMS, sizeof(opt), &opt) ||
451 nla_put_u32(skb, TCA_CHOKE_MAX_P, READ_ONCE(q->parms.max_P)))
452 goto nla_put_failure;
453 return nla_nest_end(skb, opts);
454
455nla_put_failure:
456 nla_nest_cancel(skb, opts);
457 return -EMSGSIZE;
458}
459
460static int choke_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
461{
462 struct choke_sched_data *q = qdisc_priv(sch);
463 struct tc_choke_xstats st = {
464 .early = q->stats.prob_drop + q->stats.forced_drop,
465 .marked = q->stats.prob_mark + q->stats.forced_mark,
466 .pdrop = q->stats.pdrop,
467 .matched = q->stats.matched,
468 };
469
470 return gnet_stats_copy_app(d, &st, sizeof(st));
471}
472
473static void choke_destroy(struct Qdisc *sch)
474{
475 struct choke_sched_data *q = qdisc_priv(sch);
476
477 choke_free(q->tab);
478}
479
480static struct sk_buff *choke_peek_head(struct Qdisc *sch)
481{
482 struct choke_sched_data *q = qdisc_priv(sch);
483
484 return (q->head != q->tail) ? q->tab[q->head] : NULL;
485}
486
487static struct Qdisc_ops choke_qdisc_ops __read_mostly = {
488 .id = "choke",
489 .priv_size = sizeof(struct choke_sched_data),
490
491 .enqueue = choke_enqueue,
492 .dequeue = choke_dequeue,
493 .peek = choke_peek_head,
494 .init = choke_init,
495 .destroy = choke_destroy,
496 .reset = choke_reset,
497 .change = choke_change,
498 .dump = choke_dump,
499 .dump_stats = choke_dump_stats,
500 .owner = THIS_MODULE,
501};
502MODULE_ALIAS_NET_SCH("choke");
503
504static int __init choke_module_init(void)
505{
506 return register_qdisc(&choke_qdisc_ops);
507}
508
509static void __exit choke_module_exit(void)
510{
511 unregister_qdisc(&choke_qdisc_ops);
512}
513
514module_init(choke_module_init)
515module_exit(choke_module_exit)
516
517MODULE_LICENSE("GPL");
518MODULE_DESCRIPTION("Choose and keep responsive flows scheduler");
1/*
2 * net/sched/sch_choke.c CHOKE scheduler
3 *
4 * Copyright (c) 2011 Stephen Hemminger <shemminger@vyatta.com>
5 * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * version 2 as published by the Free Software Foundation.
10 *
11 */
12
13#include <linux/module.h>
14#include <linux/types.h>
15#include <linux/kernel.h>
16#include <linux/skbuff.h>
17#include <linux/vmalloc.h>
18#include <net/pkt_sched.h>
19#include <net/pkt_cls.h>
20#include <net/inet_ecn.h>
21#include <net/red.h>
22#include <net/flow_dissector.h>
23
24/*
25 CHOKe stateless AQM for fair bandwidth allocation
26 =================================================
27
28 CHOKe (CHOose and Keep for responsive flows, CHOose and Kill for
29 unresponsive flows) is a variant of RED that penalizes misbehaving flows but
30 maintains no flow state. The difference from RED is an additional step
31 during the enqueuing process. If average queue size is over the
32 low threshold (qmin), a packet is chosen at random from the queue.
33 If both the new and chosen packet are from the same flow, both
34 are dropped. Unlike RED, CHOKe is not really a "classful" qdisc because it
35 needs to access packets in queue randomly. It has a minimal class
36 interface to allow overriding the builtin flow classifier with
37 filters.
38
39 Source:
40 R. Pan, B. Prabhakar, and K. Psounis, "CHOKe, A Stateless
41 Active Queue Management Scheme for Approximating Fair Bandwidth Allocation",
42 IEEE INFOCOM, 2000.
43
44 A. Tang, J. Wang, S. Low, "Understanding CHOKe: Throughput and Spatial
45 Characteristics", IEEE/ACM Transactions on Networking, 2004
46
47 */
48
49/* Upper bound on size of sk_buff table (packets) */
50#define CHOKE_MAX_QUEUE (128*1024 - 1)
51
52struct choke_sched_data {
53/* Parameters */
54 u32 limit;
55 unsigned char flags;
56
57 struct red_parms parms;
58
59/* Variables */
60 struct red_vars vars;
61 struct {
62 u32 prob_drop; /* Early probability drops */
63 u32 prob_mark; /* Early probability marks */
64 u32 forced_drop; /* Forced drops, qavg > max_thresh */
65 u32 forced_mark; /* Forced marks, qavg > max_thresh */
66 u32 pdrop; /* Drops due to queue limits */
67 u32 other; /* Drops due to drop() calls */
68 u32 matched; /* Drops to flow match */
69 } stats;
70
71 unsigned int head;
72 unsigned int tail;
73
74 unsigned int tab_mask; /* size - 1 */
75
76 struct sk_buff **tab;
77};
78
79/* number of elements in queue including holes */
80static unsigned int choke_len(const struct choke_sched_data *q)
81{
82 return (q->tail - q->head) & q->tab_mask;
83}
84
85/* Is ECN parameter configured */
86static int use_ecn(const struct choke_sched_data *q)
87{
88 return q->flags & TC_RED_ECN;
89}
90
91/* Should packets over max just be dropped (versus marked) */
92static int use_harddrop(const struct choke_sched_data *q)
93{
94 return q->flags & TC_RED_HARDDROP;
95}
96
97/* Move head pointer forward to skip over holes */
98static void choke_zap_head_holes(struct choke_sched_data *q)
99{
100 do {
101 q->head = (q->head + 1) & q->tab_mask;
102 if (q->head == q->tail)
103 break;
104 } while (q->tab[q->head] == NULL);
105}
106
107/* Move tail pointer backwards to reuse holes */
108static void choke_zap_tail_holes(struct choke_sched_data *q)
109{
110 do {
111 q->tail = (q->tail - 1) & q->tab_mask;
112 if (q->head == q->tail)
113 break;
114 } while (q->tab[q->tail] == NULL);
115}
116
117/* Drop packet from queue array by creating a "hole" */
118static void choke_drop_by_idx(struct Qdisc *sch, unsigned int idx,
119 struct sk_buff **to_free)
120{
121 struct choke_sched_data *q = qdisc_priv(sch);
122 struct sk_buff *skb = q->tab[idx];
123
124 q->tab[idx] = NULL;
125
126 if (idx == q->head)
127 choke_zap_head_holes(q);
128 if (idx == q->tail)
129 choke_zap_tail_holes(q);
130
131 qdisc_qstats_backlog_dec(sch, skb);
132 qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(skb));
133 qdisc_drop(skb, sch, to_free);
134 --sch->q.qlen;
135}
136
137struct choke_skb_cb {
138 u16 classid;
139 u8 keys_valid;
140 struct flow_keys_digest keys;
141};
142
143static inline struct choke_skb_cb *choke_skb_cb(const struct sk_buff *skb)
144{
145 qdisc_cb_private_validate(skb, sizeof(struct choke_skb_cb));
146 return (struct choke_skb_cb *)qdisc_skb_cb(skb)->data;
147}
148
149static inline void choke_set_classid(struct sk_buff *skb, u16 classid)
150{
151 choke_skb_cb(skb)->classid = classid;
152}
153
154/*
155 * Compare flow of two packets
156 * Returns true only if source and destination address and port match.
157 * false for special cases
158 */
159static bool choke_match_flow(struct sk_buff *skb1,
160 struct sk_buff *skb2)
161{
162 struct flow_keys temp;
163
164 if (skb1->protocol != skb2->protocol)
165 return false;
166
167 if (!choke_skb_cb(skb1)->keys_valid) {
168 choke_skb_cb(skb1)->keys_valid = 1;
169 skb_flow_dissect_flow_keys(skb1, &temp, 0);
170 make_flow_keys_digest(&choke_skb_cb(skb1)->keys, &temp);
171 }
172
173 if (!choke_skb_cb(skb2)->keys_valid) {
174 choke_skb_cb(skb2)->keys_valid = 1;
175 skb_flow_dissect_flow_keys(skb2, &temp, 0);
176 make_flow_keys_digest(&choke_skb_cb(skb2)->keys, &temp);
177 }
178
179 return !memcmp(&choke_skb_cb(skb1)->keys,
180 &choke_skb_cb(skb2)->keys,
181 sizeof(choke_skb_cb(skb1)->keys));
182}
183
184/*
185 * Select a packet at random from queue
186 * HACK: since queue can have holes from previous deletion; retry several
187 * times to find a random skb but then just give up and return the head
188 * Will return NULL if queue is empty (q->head == q->tail)
189 */
190static struct sk_buff *choke_peek_random(const struct choke_sched_data *q,
191 unsigned int *pidx)
192{
193 struct sk_buff *skb;
194 int retrys = 3;
195
196 do {
197 *pidx = (q->head + prandom_u32_max(choke_len(q))) & q->tab_mask;
198 skb = q->tab[*pidx];
199 if (skb)
200 return skb;
201 } while (--retrys > 0);
202
203 return q->tab[*pidx = q->head];
204}
205
206/*
207 * Compare new packet with random packet in queue
208 * returns true if matched and sets *pidx
209 */
210static bool choke_match_random(const struct choke_sched_data *q,
211 struct sk_buff *nskb,
212 unsigned int *pidx)
213{
214 struct sk_buff *oskb;
215
216 if (q->head == q->tail)
217 return false;
218
219 oskb = choke_peek_random(q, pidx);
220 return choke_match_flow(oskb, nskb);
221}
222
223static int choke_enqueue(struct sk_buff *skb, struct Qdisc *sch,
224 struct sk_buff **to_free)
225{
226 struct choke_sched_data *q = qdisc_priv(sch);
227 const struct red_parms *p = &q->parms;
228
229 choke_skb_cb(skb)->keys_valid = 0;
230 /* Compute average queue usage (see RED) */
231 q->vars.qavg = red_calc_qavg(p, &q->vars, sch->q.qlen);
232 if (red_is_idling(&q->vars))
233 red_end_of_idle_period(&q->vars);
234
235 /* Is queue small? */
236 if (q->vars.qavg <= p->qth_min)
237 q->vars.qcount = -1;
238 else {
239 unsigned int idx;
240
241 /* Draw a packet at random from queue and compare flow */
242 if (choke_match_random(q, skb, &idx)) {
243 q->stats.matched++;
244 choke_drop_by_idx(sch, idx, to_free);
245 goto congestion_drop;
246 }
247
248 /* Queue is large, always mark/drop */
249 if (q->vars.qavg > p->qth_max) {
250 q->vars.qcount = -1;
251
252 qdisc_qstats_overlimit(sch);
253 if (use_harddrop(q) || !use_ecn(q) ||
254 !INET_ECN_set_ce(skb)) {
255 q->stats.forced_drop++;
256 goto congestion_drop;
257 }
258
259 q->stats.forced_mark++;
260 } else if (++q->vars.qcount) {
261 if (red_mark_probability(p, &q->vars, q->vars.qavg)) {
262 q->vars.qcount = 0;
263 q->vars.qR = red_random(p);
264
265 qdisc_qstats_overlimit(sch);
266 if (!use_ecn(q) || !INET_ECN_set_ce(skb)) {
267 q->stats.prob_drop++;
268 goto congestion_drop;
269 }
270
271 q->stats.prob_mark++;
272 }
273 } else
274 q->vars.qR = red_random(p);
275 }
276
277 /* Admit new packet */
278 if (sch->q.qlen < q->limit) {
279 q->tab[q->tail] = skb;
280 q->tail = (q->tail + 1) & q->tab_mask;
281 ++sch->q.qlen;
282 qdisc_qstats_backlog_inc(sch, skb);
283 return NET_XMIT_SUCCESS;
284 }
285
286 q->stats.pdrop++;
287 return qdisc_drop(skb, sch, to_free);
288
289congestion_drop:
290 qdisc_drop(skb, sch, to_free);
291 return NET_XMIT_CN;
292}
293
294static struct sk_buff *choke_dequeue(struct Qdisc *sch)
295{
296 struct choke_sched_data *q = qdisc_priv(sch);
297 struct sk_buff *skb;
298
299 if (q->head == q->tail) {
300 if (!red_is_idling(&q->vars))
301 red_start_of_idle_period(&q->vars);
302 return NULL;
303 }
304
305 skb = q->tab[q->head];
306 q->tab[q->head] = NULL;
307 choke_zap_head_holes(q);
308 --sch->q.qlen;
309 qdisc_qstats_backlog_dec(sch, skb);
310 qdisc_bstats_update(sch, skb);
311
312 return skb;
313}
314
315static void choke_reset(struct Qdisc *sch)
316{
317 struct choke_sched_data *q = qdisc_priv(sch);
318
319 while (q->head != q->tail) {
320 struct sk_buff *skb = q->tab[q->head];
321
322 q->head = (q->head + 1) & q->tab_mask;
323 if (!skb)
324 continue;
325 rtnl_qdisc_drop(skb, sch);
326 }
327
328 sch->q.qlen = 0;
329 sch->qstats.backlog = 0;
330 memset(q->tab, 0, (q->tab_mask + 1) * sizeof(struct sk_buff *));
331 q->head = q->tail = 0;
332 red_restart(&q->vars);
333}
334
335static const struct nla_policy choke_policy[TCA_CHOKE_MAX + 1] = {
336 [TCA_CHOKE_PARMS] = { .len = sizeof(struct tc_red_qopt) },
337 [TCA_CHOKE_STAB] = { .len = RED_STAB_SIZE },
338 [TCA_CHOKE_MAX_P] = { .type = NLA_U32 },
339};
340
341
342static void choke_free(void *addr)
343{
344 kvfree(addr);
345}
346
347static int choke_change(struct Qdisc *sch, struct nlattr *opt,
348 struct netlink_ext_ack *extack)
349{
350 struct choke_sched_data *q = qdisc_priv(sch);
351 struct nlattr *tb[TCA_CHOKE_MAX + 1];
352 const struct tc_red_qopt *ctl;
353 int err;
354 struct sk_buff **old = NULL;
355 unsigned int mask;
356 u32 max_P;
357
358 if (opt == NULL)
359 return -EINVAL;
360
361 err = nla_parse_nested(tb, TCA_CHOKE_MAX, opt, choke_policy, NULL);
362 if (err < 0)
363 return err;
364
365 if (tb[TCA_CHOKE_PARMS] == NULL ||
366 tb[TCA_CHOKE_STAB] == NULL)
367 return -EINVAL;
368
369 max_P = tb[TCA_CHOKE_MAX_P] ? nla_get_u32(tb[TCA_CHOKE_MAX_P]) : 0;
370
371 ctl = nla_data(tb[TCA_CHOKE_PARMS]);
372
373 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog))
374 return -EINVAL;
375
376 if (ctl->limit > CHOKE_MAX_QUEUE)
377 return -EINVAL;
378
379 mask = roundup_pow_of_two(ctl->limit + 1) - 1;
380 if (mask != q->tab_mask) {
381 struct sk_buff **ntab;
382
383 ntab = kvmalloc_array((mask + 1), sizeof(struct sk_buff *), GFP_KERNEL | __GFP_ZERO);
384 if (!ntab)
385 return -ENOMEM;
386
387 sch_tree_lock(sch);
388 old = q->tab;
389 if (old) {
390 unsigned int oqlen = sch->q.qlen, tail = 0;
391 unsigned dropped = 0;
392
393 while (q->head != q->tail) {
394 struct sk_buff *skb = q->tab[q->head];
395
396 q->head = (q->head + 1) & q->tab_mask;
397 if (!skb)
398 continue;
399 if (tail < mask) {
400 ntab[tail++] = skb;
401 continue;
402 }
403 dropped += qdisc_pkt_len(skb);
404 qdisc_qstats_backlog_dec(sch, skb);
405 --sch->q.qlen;
406 rtnl_qdisc_drop(skb, sch);
407 }
408 qdisc_tree_reduce_backlog(sch, oqlen - sch->q.qlen, dropped);
409 q->head = 0;
410 q->tail = tail;
411 }
412
413 q->tab_mask = mask;
414 q->tab = ntab;
415 } else
416 sch_tree_lock(sch);
417
418 q->flags = ctl->flags;
419 q->limit = ctl->limit;
420
421 red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
422 ctl->Plog, ctl->Scell_log,
423 nla_data(tb[TCA_CHOKE_STAB]),
424 max_P);
425 red_set_vars(&q->vars);
426
427 if (q->head == q->tail)
428 red_end_of_idle_period(&q->vars);
429
430 sch_tree_unlock(sch);
431 choke_free(old);
432 return 0;
433}
434
435static int choke_init(struct Qdisc *sch, struct nlattr *opt,
436 struct netlink_ext_ack *extack)
437{
438 return choke_change(sch, opt, extack);
439}
440
441static int choke_dump(struct Qdisc *sch, struct sk_buff *skb)
442{
443 struct choke_sched_data *q = qdisc_priv(sch);
444 struct nlattr *opts = NULL;
445 struct tc_red_qopt opt = {
446 .limit = q->limit,
447 .flags = q->flags,
448 .qth_min = q->parms.qth_min >> q->parms.Wlog,
449 .qth_max = q->parms.qth_max >> q->parms.Wlog,
450 .Wlog = q->parms.Wlog,
451 .Plog = q->parms.Plog,
452 .Scell_log = q->parms.Scell_log,
453 };
454
455 opts = nla_nest_start(skb, TCA_OPTIONS);
456 if (opts == NULL)
457 goto nla_put_failure;
458
459 if (nla_put(skb, TCA_CHOKE_PARMS, sizeof(opt), &opt) ||
460 nla_put_u32(skb, TCA_CHOKE_MAX_P, q->parms.max_P))
461 goto nla_put_failure;
462 return nla_nest_end(skb, opts);
463
464nla_put_failure:
465 nla_nest_cancel(skb, opts);
466 return -EMSGSIZE;
467}
468
469static int choke_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
470{
471 struct choke_sched_data *q = qdisc_priv(sch);
472 struct tc_choke_xstats st = {
473 .early = q->stats.prob_drop + q->stats.forced_drop,
474 .marked = q->stats.prob_mark + q->stats.forced_mark,
475 .pdrop = q->stats.pdrop,
476 .other = q->stats.other,
477 .matched = q->stats.matched,
478 };
479
480 return gnet_stats_copy_app(d, &st, sizeof(st));
481}
482
483static void choke_destroy(struct Qdisc *sch)
484{
485 struct choke_sched_data *q = qdisc_priv(sch);
486
487 choke_free(q->tab);
488}
489
490static struct sk_buff *choke_peek_head(struct Qdisc *sch)
491{
492 struct choke_sched_data *q = qdisc_priv(sch);
493
494 return (q->head != q->tail) ? q->tab[q->head] : NULL;
495}
496
497static struct Qdisc_ops choke_qdisc_ops __read_mostly = {
498 .id = "choke",
499 .priv_size = sizeof(struct choke_sched_data),
500
501 .enqueue = choke_enqueue,
502 .dequeue = choke_dequeue,
503 .peek = choke_peek_head,
504 .init = choke_init,
505 .destroy = choke_destroy,
506 .reset = choke_reset,
507 .change = choke_change,
508 .dump = choke_dump,
509 .dump_stats = choke_dump_stats,
510 .owner = THIS_MODULE,
511};
512
513static int __init choke_module_init(void)
514{
515 return register_qdisc(&choke_qdisc_ops);
516}
517
518static void __exit choke_module_exit(void)
519{
520 unregister_qdisc(&choke_qdisc_ops);
521}
522
523module_init(choke_module_init)
524module_exit(choke_module_exit)
525
526MODULE_LICENSE("GPL");