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/inet_ecn.h>
20#include <net/red.h>
21#include <net/flow_dissector.h>
22
23/*
24 CHOKe stateless AQM for fair bandwidth allocation
25 =================================================
26
27 CHOKe (CHOose and Keep for responsive flows, CHOose and Kill for
28 unresponsive flows) is a variant of RED that penalizes misbehaving flows but
29 maintains no flow state. The difference from RED is an additional step
30 during the enqueuing process. If average queue size is over the
31 low threshold (qmin), a packet is chosen at random from the queue.
32 If both the new and chosen packet are from the same flow, both
33 are dropped. Unlike RED, CHOKe is not really a "classful" qdisc because it
34 needs to access packets in queue randomly. It has a minimal class
35 interface to allow overriding the builtin flow classifier with
36 filters.
37
38 Source:
39 R. Pan, B. Prabhakar, and K. Psounis, "CHOKe, A Stateless
40 Active Queue Management Scheme for Approximating Fair Bandwidth Allocation",
41 IEEE INFOCOM, 2000.
42
43 A. Tang, J. Wang, S. Low, "Understanding CHOKe: Throughput and Spatial
44 Characteristics", IEEE/ACM Transactions on Networking, 2004
45
46 */
47
48/* Upper bound on size of sk_buff table (packets) */
49#define CHOKE_MAX_QUEUE (128*1024 - 1)
50
51struct choke_sched_data {
52/* Parameters */
53 u32 limit;
54 unsigned char flags;
55
56 struct red_parms parms;
57
58/* Variables */
59 struct red_vars vars;
60 struct tcf_proto __rcu *filter_list;
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
154static u16 choke_get_classid(const struct sk_buff *skb)
155{
156 return choke_skb_cb(skb)->classid;
157}
158
159/*
160 * Compare flow of two packets
161 * Returns true only if source and destination address and port match.
162 * false for special cases
163 */
164static bool choke_match_flow(struct sk_buff *skb1,
165 struct sk_buff *skb2)
166{
167 struct flow_keys temp;
168
169 if (skb1->protocol != skb2->protocol)
170 return false;
171
172 if (!choke_skb_cb(skb1)->keys_valid) {
173 choke_skb_cb(skb1)->keys_valid = 1;
174 skb_flow_dissect_flow_keys(skb1, &temp, 0);
175 make_flow_keys_digest(&choke_skb_cb(skb1)->keys, &temp);
176 }
177
178 if (!choke_skb_cb(skb2)->keys_valid) {
179 choke_skb_cb(skb2)->keys_valid = 1;
180 skb_flow_dissect_flow_keys(skb2, &temp, 0);
181 make_flow_keys_digest(&choke_skb_cb(skb2)->keys, &temp);
182 }
183
184 return !memcmp(&choke_skb_cb(skb1)->keys,
185 &choke_skb_cb(skb2)->keys,
186 sizeof(choke_skb_cb(skb1)->keys));
187}
188
189/*
190 * Classify flow using either:
191 * 1. pre-existing classification result in skb
192 * 2. fast internal classification
193 * 3. use TC filter based classification
194 */
195static bool choke_classify(struct sk_buff *skb,
196 struct Qdisc *sch, int *qerr)
197
198{
199 struct choke_sched_data *q = qdisc_priv(sch);
200 struct tcf_result res;
201 struct tcf_proto *fl;
202 int result;
203
204 fl = rcu_dereference_bh(q->filter_list);
205 result = tc_classify(skb, fl, &res, false);
206 if (result >= 0) {
207#ifdef CONFIG_NET_CLS_ACT
208 switch (result) {
209 case TC_ACT_STOLEN:
210 case TC_ACT_QUEUED:
211 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
212 case TC_ACT_SHOT:
213 return false;
214 }
215#endif
216 choke_set_classid(skb, TC_H_MIN(res.classid));
217 return true;
218 }
219
220 return false;
221}
222
223/*
224 * Select a packet at random from queue
225 * HACK: since queue can have holes from previous deletion; retry several
226 * times to find a random skb but then just give up and return the head
227 * Will return NULL if queue is empty (q->head == q->tail)
228 */
229static struct sk_buff *choke_peek_random(const struct choke_sched_data *q,
230 unsigned int *pidx)
231{
232 struct sk_buff *skb;
233 int retrys = 3;
234
235 do {
236 *pidx = (q->head + prandom_u32_max(choke_len(q))) & q->tab_mask;
237 skb = q->tab[*pidx];
238 if (skb)
239 return skb;
240 } while (--retrys > 0);
241
242 return q->tab[*pidx = q->head];
243}
244
245/*
246 * Compare new packet with random packet in queue
247 * returns true if matched and sets *pidx
248 */
249static bool choke_match_random(const struct choke_sched_data *q,
250 struct sk_buff *nskb,
251 unsigned int *pidx)
252{
253 struct sk_buff *oskb;
254
255 if (q->head == q->tail)
256 return false;
257
258 oskb = choke_peek_random(q, pidx);
259 if (rcu_access_pointer(q->filter_list))
260 return choke_get_classid(nskb) == choke_get_classid(oskb);
261
262 return choke_match_flow(oskb, nskb);
263}
264
265static int choke_enqueue(struct sk_buff *skb, struct Qdisc *sch,
266 struct sk_buff **to_free)
267{
268 int ret = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
269 struct choke_sched_data *q = qdisc_priv(sch);
270 const struct red_parms *p = &q->parms;
271
272 if (rcu_access_pointer(q->filter_list)) {
273 /* If using external classifiers, get result and record it. */
274 if (!choke_classify(skb, sch, &ret))
275 goto other_drop; /* Packet was eaten by filter */
276 }
277
278 choke_skb_cb(skb)->keys_valid = 0;
279 /* Compute average queue usage (see RED) */
280 q->vars.qavg = red_calc_qavg(p, &q->vars, sch->q.qlen);
281 if (red_is_idling(&q->vars))
282 red_end_of_idle_period(&q->vars);
283
284 /* Is queue small? */
285 if (q->vars.qavg <= p->qth_min)
286 q->vars.qcount = -1;
287 else {
288 unsigned int idx;
289
290 /* Draw a packet at random from queue and compare flow */
291 if (choke_match_random(q, skb, &idx)) {
292 q->stats.matched++;
293 choke_drop_by_idx(sch, idx, to_free);
294 goto congestion_drop;
295 }
296
297 /* Queue is large, always mark/drop */
298 if (q->vars.qavg > p->qth_max) {
299 q->vars.qcount = -1;
300
301 qdisc_qstats_overlimit(sch);
302 if (use_harddrop(q) || !use_ecn(q) ||
303 !INET_ECN_set_ce(skb)) {
304 q->stats.forced_drop++;
305 goto congestion_drop;
306 }
307
308 q->stats.forced_mark++;
309 } else if (++q->vars.qcount) {
310 if (red_mark_probability(p, &q->vars, q->vars.qavg)) {
311 q->vars.qcount = 0;
312 q->vars.qR = red_random(p);
313
314 qdisc_qstats_overlimit(sch);
315 if (!use_ecn(q) || !INET_ECN_set_ce(skb)) {
316 q->stats.prob_drop++;
317 goto congestion_drop;
318 }
319
320 q->stats.prob_mark++;
321 }
322 } else
323 q->vars.qR = red_random(p);
324 }
325
326 /* Admit new packet */
327 if (sch->q.qlen < q->limit) {
328 q->tab[q->tail] = skb;
329 q->tail = (q->tail + 1) & q->tab_mask;
330 ++sch->q.qlen;
331 qdisc_qstats_backlog_inc(sch, skb);
332 return NET_XMIT_SUCCESS;
333 }
334
335 q->stats.pdrop++;
336 return qdisc_drop(skb, sch, to_free);
337
338congestion_drop:
339 qdisc_drop(skb, sch, to_free);
340 return NET_XMIT_CN;
341
342other_drop:
343 if (ret & __NET_XMIT_BYPASS)
344 qdisc_qstats_drop(sch);
345 __qdisc_drop(skb, to_free);
346 return ret;
347}
348
349static struct sk_buff *choke_dequeue(struct Qdisc *sch)
350{
351 struct choke_sched_data *q = qdisc_priv(sch);
352 struct sk_buff *skb;
353
354 if (q->head == q->tail) {
355 if (!red_is_idling(&q->vars))
356 red_start_of_idle_period(&q->vars);
357 return NULL;
358 }
359
360 skb = q->tab[q->head];
361 q->tab[q->head] = NULL;
362 choke_zap_head_holes(q);
363 --sch->q.qlen;
364 qdisc_qstats_backlog_dec(sch, skb);
365 qdisc_bstats_update(sch, skb);
366
367 return skb;
368}
369
370static void choke_reset(struct Qdisc *sch)
371{
372 struct choke_sched_data *q = qdisc_priv(sch);
373
374 while (q->head != q->tail) {
375 struct sk_buff *skb = q->tab[q->head];
376
377 q->head = (q->head + 1) & q->tab_mask;
378 if (!skb)
379 continue;
380 rtnl_qdisc_drop(skb, sch);
381 }
382
383 sch->q.qlen = 0;
384 sch->qstats.backlog = 0;
385 memset(q->tab, 0, (q->tab_mask + 1) * sizeof(struct sk_buff *));
386 q->head = q->tail = 0;
387 red_restart(&q->vars);
388}
389
390static const struct nla_policy choke_policy[TCA_CHOKE_MAX + 1] = {
391 [TCA_CHOKE_PARMS] = { .len = sizeof(struct tc_red_qopt) },
392 [TCA_CHOKE_STAB] = { .len = RED_STAB_SIZE },
393 [TCA_CHOKE_MAX_P] = { .type = NLA_U32 },
394};
395
396
397static void choke_free(void *addr)
398{
399 kvfree(addr);
400}
401
402static int choke_change(struct Qdisc *sch, struct nlattr *opt)
403{
404 struct choke_sched_data *q = qdisc_priv(sch);
405 struct nlattr *tb[TCA_CHOKE_MAX + 1];
406 const struct tc_red_qopt *ctl;
407 int err;
408 struct sk_buff **old = NULL;
409 unsigned int mask;
410 u32 max_P;
411
412 if (opt == NULL)
413 return -EINVAL;
414
415 err = nla_parse_nested(tb, TCA_CHOKE_MAX, opt, choke_policy);
416 if (err < 0)
417 return err;
418
419 if (tb[TCA_CHOKE_PARMS] == NULL ||
420 tb[TCA_CHOKE_STAB] == NULL)
421 return -EINVAL;
422
423 max_P = tb[TCA_CHOKE_MAX_P] ? nla_get_u32(tb[TCA_CHOKE_MAX_P]) : 0;
424
425 ctl = nla_data(tb[TCA_CHOKE_PARMS]);
426
427 if (ctl->limit > CHOKE_MAX_QUEUE)
428 return -EINVAL;
429
430 mask = roundup_pow_of_two(ctl->limit + 1) - 1;
431 if (mask != q->tab_mask) {
432 struct sk_buff **ntab;
433
434 ntab = kcalloc(mask + 1, sizeof(struct sk_buff *),
435 GFP_KERNEL | __GFP_NOWARN);
436 if (!ntab)
437 ntab = vzalloc((mask + 1) * sizeof(struct sk_buff *));
438 if (!ntab)
439 return -ENOMEM;
440
441 sch_tree_lock(sch);
442 old = q->tab;
443 if (old) {
444 unsigned int oqlen = sch->q.qlen, tail = 0;
445 unsigned dropped = 0;
446
447 while (q->head != q->tail) {
448 struct sk_buff *skb = q->tab[q->head];
449
450 q->head = (q->head + 1) & q->tab_mask;
451 if (!skb)
452 continue;
453 if (tail < mask) {
454 ntab[tail++] = skb;
455 continue;
456 }
457 dropped += qdisc_pkt_len(skb);
458 qdisc_qstats_backlog_dec(sch, skb);
459 --sch->q.qlen;
460 rtnl_qdisc_drop(skb, sch);
461 }
462 qdisc_tree_reduce_backlog(sch, oqlen - sch->q.qlen, dropped);
463 q->head = 0;
464 q->tail = tail;
465 }
466
467 q->tab_mask = mask;
468 q->tab = ntab;
469 } else
470 sch_tree_lock(sch);
471
472 q->flags = ctl->flags;
473 q->limit = ctl->limit;
474
475 red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
476 ctl->Plog, ctl->Scell_log,
477 nla_data(tb[TCA_CHOKE_STAB]),
478 max_P);
479 red_set_vars(&q->vars);
480
481 if (q->head == q->tail)
482 red_end_of_idle_period(&q->vars);
483
484 sch_tree_unlock(sch);
485 choke_free(old);
486 return 0;
487}
488
489static int choke_init(struct Qdisc *sch, struct nlattr *opt)
490{
491 return choke_change(sch, opt);
492}
493
494static int choke_dump(struct Qdisc *sch, struct sk_buff *skb)
495{
496 struct choke_sched_data *q = qdisc_priv(sch);
497 struct nlattr *opts = NULL;
498 struct tc_red_qopt opt = {
499 .limit = q->limit,
500 .flags = q->flags,
501 .qth_min = q->parms.qth_min >> q->parms.Wlog,
502 .qth_max = q->parms.qth_max >> q->parms.Wlog,
503 .Wlog = q->parms.Wlog,
504 .Plog = q->parms.Plog,
505 .Scell_log = q->parms.Scell_log,
506 };
507
508 opts = nla_nest_start(skb, TCA_OPTIONS);
509 if (opts == NULL)
510 goto nla_put_failure;
511
512 if (nla_put(skb, TCA_CHOKE_PARMS, sizeof(opt), &opt) ||
513 nla_put_u32(skb, TCA_CHOKE_MAX_P, q->parms.max_P))
514 goto nla_put_failure;
515 return nla_nest_end(skb, opts);
516
517nla_put_failure:
518 nla_nest_cancel(skb, opts);
519 return -EMSGSIZE;
520}
521
522static int choke_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
523{
524 struct choke_sched_data *q = qdisc_priv(sch);
525 struct tc_choke_xstats st = {
526 .early = q->stats.prob_drop + q->stats.forced_drop,
527 .marked = q->stats.prob_mark + q->stats.forced_mark,
528 .pdrop = q->stats.pdrop,
529 .other = q->stats.other,
530 .matched = q->stats.matched,
531 };
532
533 return gnet_stats_copy_app(d, &st, sizeof(st));
534}
535
536static void choke_destroy(struct Qdisc *sch)
537{
538 struct choke_sched_data *q = qdisc_priv(sch);
539
540 tcf_destroy_chain(&q->filter_list);
541 choke_free(q->tab);
542}
543
544static struct sk_buff *choke_peek_head(struct Qdisc *sch)
545{
546 struct choke_sched_data *q = qdisc_priv(sch);
547
548 return (q->head != q->tail) ? q->tab[q->head] : NULL;
549}
550
551static struct Qdisc_ops choke_qdisc_ops __read_mostly = {
552 .id = "choke",
553 .priv_size = sizeof(struct choke_sched_data),
554
555 .enqueue = choke_enqueue,
556 .dequeue = choke_dequeue,
557 .peek = choke_peek_head,
558 .init = choke_init,
559 .destroy = choke_destroy,
560 .reset = choke_reset,
561 .change = choke_change,
562 .dump = choke_dump,
563 .dump_stats = choke_dump_stats,
564 .owner = THIS_MODULE,
565};
566
567static int __init choke_module_init(void)
568{
569 return register_qdisc(&choke_qdisc_ops);
570}
571
572static void __exit choke_module_exit(void)
573{
574 unregister_qdisc(&choke_qdisc_ops);
575}
576
577module_init(choke_module_init)
578module_exit(choke_module_exit)
579
580MODULE_LICENSE("GPL");