Loading...
1/*
2 * net/sched/sch_red.c Random Early Detection queue.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 *
11 * Changes:
12 * J Hadi Salim 980914: computation fixes
13 * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
14 * J Hadi Salim 980816: ECN support
15 */
16
17#include <linux/module.h>
18#include <linux/types.h>
19#include <linux/kernel.h>
20#include <linux/skbuff.h>
21#include <net/pkt_sched.h>
22#include <net/inet_ecn.h>
23#include <net/red.h>
24
25
26/* Parameters, settable by user:
27 -----------------------------
28
29 limit - bytes (must be > qth_max + burst)
30
31 Hard limit on queue length, should be chosen >qth_max
32 to allow packet bursts. This parameter does not
33 affect the algorithms behaviour and can be chosen
34 arbitrarily high (well, less than ram size)
35 Really, this limit will never be reached
36 if RED works correctly.
37 */
38
39struct red_sched_data {
40 u32 limit; /* HARD maximal queue length */
41 unsigned char flags;
42 struct timer_list adapt_timer;
43 struct red_parms parms;
44 struct red_vars vars;
45 struct red_stats stats;
46 struct Qdisc *qdisc;
47};
48
49static inline int red_use_ecn(struct red_sched_data *q)
50{
51 return q->flags & TC_RED_ECN;
52}
53
54static inline int red_use_harddrop(struct red_sched_data *q)
55{
56 return q->flags & TC_RED_HARDDROP;
57}
58
59static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch)
60{
61 struct red_sched_data *q = qdisc_priv(sch);
62 struct Qdisc *child = q->qdisc;
63 int ret;
64
65 q->vars.qavg = red_calc_qavg(&q->parms,
66 &q->vars,
67 child->qstats.backlog);
68
69 if (red_is_idling(&q->vars))
70 red_end_of_idle_period(&q->vars);
71
72 switch (red_action(&q->parms, &q->vars, q->vars.qavg)) {
73 case RED_DONT_MARK:
74 break;
75
76 case RED_PROB_MARK:
77 sch->qstats.overlimits++;
78 if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
79 q->stats.prob_drop++;
80 goto congestion_drop;
81 }
82
83 q->stats.prob_mark++;
84 break;
85
86 case RED_HARD_MARK:
87 sch->qstats.overlimits++;
88 if (red_use_harddrop(q) || !red_use_ecn(q) ||
89 !INET_ECN_set_ce(skb)) {
90 q->stats.forced_drop++;
91 goto congestion_drop;
92 }
93
94 q->stats.forced_mark++;
95 break;
96 }
97
98 ret = qdisc_enqueue(skb, child);
99 if (likely(ret == NET_XMIT_SUCCESS)) {
100 sch->q.qlen++;
101 } else if (net_xmit_drop_count(ret)) {
102 q->stats.pdrop++;
103 sch->qstats.drops++;
104 }
105 return ret;
106
107congestion_drop:
108 qdisc_drop(skb, sch);
109 return NET_XMIT_CN;
110}
111
112static struct sk_buff *red_dequeue(struct Qdisc *sch)
113{
114 struct sk_buff *skb;
115 struct red_sched_data *q = qdisc_priv(sch);
116 struct Qdisc *child = q->qdisc;
117
118 skb = child->dequeue(child);
119 if (skb) {
120 qdisc_bstats_update(sch, skb);
121 sch->q.qlen--;
122 } else {
123 if (!red_is_idling(&q->vars))
124 red_start_of_idle_period(&q->vars);
125 }
126 return skb;
127}
128
129static struct sk_buff *red_peek(struct Qdisc *sch)
130{
131 struct red_sched_data *q = qdisc_priv(sch);
132 struct Qdisc *child = q->qdisc;
133
134 return child->ops->peek(child);
135}
136
137static unsigned int red_drop(struct Qdisc *sch)
138{
139 struct red_sched_data *q = qdisc_priv(sch);
140 struct Qdisc *child = q->qdisc;
141 unsigned int len;
142
143 if (child->ops->drop && (len = child->ops->drop(child)) > 0) {
144 q->stats.other++;
145 sch->qstats.drops++;
146 sch->q.qlen--;
147 return len;
148 }
149
150 if (!red_is_idling(&q->vars))
151 red_start_of_idle_period(&q->vars);
152
153 return 0;
154}
155
156static void red_reset(struct Qdisc *sch)
157{
158 struct red_sched_data *q = qdisc_priv(sch);
159
160 qdisc_reset(q->qdisc);
161 sch->q.qlen = 0;
162 red_restart(&q->vars);
163}
164
165static void red_destroy(struct Qdisc *sch)
166{
167 struct red_sched_data *q = qdisc_priv(sch);
168
169 del_timer_sync(&q->adapt_timer);
170 qdisc_destroy(q->qdisc);
171}
172
173static const struct nla_policy red_policy[TCA_RED_MAX + 1] = {
174 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) },
175 [TCA_RED_STAB] = { .len = RED_STAB_SIZE },
176 [TCA_RED_MAX_P] = { .type = NLA_U32 },
177};
178
179static int red_change(struct Qdisc *sch, struct nlattr *opt)
180{
181 struct red_sched_data *q = qdisc_priv(sch);
182 struct nlattr *tb[TCA_RED_MAX + 1];
183 struct tc_red_qopt *ctl;
184 struct Qdisc *child = NULL;
185 int err;
186 u32 max_P;
187
188 if (opt == NULL)
189 return -EINVAL;
190
191 err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy);
192 if (err < 0)
193 return err;
194
195 if (tb[TCA_RED_PARMS] == NULL ||
196 tb[TCA_RED_STAB] == NULL)
197 return -EINVAL;
198
199 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0;
200
201 ctl = nla_data(tb[TCA_RED_PARMS]);
202
203 if (ctl->limit > 0) {
204 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit);
205 if (IS_ERR(child))
206 return PTR_ERR(child);
207 }
208
209 sch_tree_lock(sch);
210 q->flags = ctl->flags;
211 q->limit = ctl->limit;
212 if (child) {
213 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
214 qdisc_destroy(q->qdisc);
215 q->qdisc = child;
216 }
217
218 red_set_parms(&q->parms,
219 ctl->qth_min, ctl->qth_max, ctl->Wlog,
220 ctl->Plog, ctl->Scell_log,
221 nla_data(tb[TCA_RED_STAB]),
222 max_P);
223 red_set_vars(&q->vars);
224
225 del_timer(&q->adapt_timer);
226 if (ctl->flags & TC_RED_ADAPTATIVE)
227 mod_timer(&q->adapt_timer, jiffies + HZ/2);
228
229 if (!q->qdisc->q.qlen)
230 red_start_of_idle_period(&q->vars);
231
232 sch_tree_unlock(sch);
233 return 0;
234}
235
236static inline void red_adaptative_timer(unsigned long arg)
237{
238 struct Qdisc *sch = (struct Qdisc *)arg;
239 struct red_sched_data *q = qdisc_priv(sch);
240 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
241
242 spin_lock(root_lock);
243 red_adaptative_algo(&q->parms, &q->vars);
244 mod_timer(&q->adapt_timer, jiffies + HZ/2);
245 spin_unlock(root_lock);
246}
247
248static int red_init(struct Qdisc *sch, struct nlattr *opt)
249{
250 struct red_sched_data *q = qdisc_priv(sch);
251
252 q->qdisc = &noop_qdisc;
253 setup_timer(&q->adapt_timer, red_adaptative_timer, (unsigned long)sch);
254 return red_change(sch, opt);
255}
256
257static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
258{
259 struct red_sched_data *q = qdisc_priv(sch);
260 struct nlattr *opts = NULL;
261 struct tc_red_qopt opt = {
262 .limit = q->limit,
263 .flags = q->flags,
264 .qth_min = q->parms.qth_min >> q->parms.Wlog,
265 .qth_max = q->parms.qth_max >> q->parms.Wlog,
266 .Wlog = q->parms.Wlog,
267 .Plog = q->parms.Plog,
268 .Scell_log = q->parms.Scell_log,
269 };
270
271 sch->qstats.backlog = q->qdisc->qstats.backlog;
272 opts = nla_nest_start(skb, TCA_OPTIONS);
273 if (opts == NULL)
274 goto nla_put_failure;
275 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) ||
276 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P))
277 goto nla_put_failure;
278 return nla_nest_end(skb, opts);
279
280nla_put_failure:
281 nla_nest_cancel(skb, opts);
282 return -EMSGSIZE;
283}
284
285static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
286{
287 struct red_sched_data *q = qdisc_priv(sch);
288 struct tc_red_xstats st = {
289 .early = q->stats.prob_drop + q->stats.forced_drop,
290 .pdrop = q->stats.pdrop,
291 .other = q->stats.other,
292 .marked = q->stats.prob_mark + q->stats.forced_mark,
293 };
294
295 return gnet_stats_copy_app(d, &st, sizeof(st));
296}
297
298static int red_dump_class(struct Qdisc *sch, unsigned long cl,
299 struct sk_buff *skb, struct tcmsg *tcm)
300{
301 struct red_sched_data *q = qdisc_priv(sch);
302
303 tcm->tcm_handle |= TC_H_MIN(1);
304 tcm->tcm_info = q->qdisc->handle;
305 return 0;
306}
307
308static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
309 struct Qdisc **old)
310{
311 struct red_sched_data *q = qdisc_priv(sch);
312
313 if (new == NULL)
314 new = &noop_qdisc;
315
316 sch_tree_lock(sch);
317 *old = q->qdisc;
318 q->qdisc = new;
319 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
320 qdisc_reset(*old);
321 sch_tree_unlock(sch);
322 return 0;
323}
324
325static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
326{
327 struct red_sched_data *q = qdisc_priv(sch);
328 return q->qdisc;
329}
330
331static unsigned long red_get(struct Qdisc *sch, u32 classid)
332{
333 return 1;
334}
335
336static void red_put(struct Qdisc *sch, unsigned long arg)
337{
338}
339
340static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
341{
342 if (!walker->stop) {
343 if (walker->count >= walker->skip)
344 if (walker->fn(sch, 1, walker) < 0) {
345 walker->stop = 1;
346 return;
347 }
348 walker->count++;
349 }
350}
351
352static const struct Qdisc_class_ops red_class_ops = {
353 .graft = red_graft,
354 .leaf = red_leaf,
355 .get = red_get,
356 .put = red_put,
357 .walk = red_walk,
358 .dump = red_dump_class,
359};
360
361static struct Qdisc_ops red_qdisc_ops __read_mostly = {
362 .id = "red",
363 .priv_size = sizeof(struct red_sched_data),
364 .cl_ops = &red_class_ops,
365 .enqueue = red_enqueue,
366 .dequeue = red_dequeue,
367 .peek = red_peek,
368 .drop = red_drop,
369 .init = red_init,
370 .reset = red_reset,
371 .destroy = red_destroy,
372 .change = red_change,
373 .dump = red_dump,
374 .dump_stats = red_dump_stats,
375 .owner = THIS_MODULE,
376};
377
378static int __init red_module_init(void)
379{
380 return register_qdisc(&red_qdisc_ops);
381}
382
383static void __exit red_module_exit(void)
384{
385 unregister_qdisc(&red_qdisc_ops);
386}
387
388module_init(red_module_init)
389module_exit(red_module_exit)
390
391MODULE_LICENSE("GPL");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/sch_red.c Random Early Detection queue.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Changes:
8 * J Hadi Salim 980914: computation fixes
9 * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
10 * J Hadi Salim 980816: ECN support
11 */
12
13#include <linux/module.h>
14#include <linux/types.h>
15#include <linux/kernel.h>
16#include <linux/skbuff.h>
17#include <net/pkt_sched.h>
18#include <net/pkt_cls.h>
19#include <net/inet_ecn.h>
20#include <net/red.h>
21
22
23/* Parameters, settable by user:
24 -----------------------------
25
26 limit - bytes (must be > qth_max + burst)
27
28 Hard limit on queue length, should be chosen >qth_max
29 to allow packet bursts. This parameter does not
30 affect the algorithms behaviour and can be chosen
31 arbitrarily high (well, less than ram size)
32 Really, this limit will never be reached
33 if RED works correctly.
34 */
35
36struct red_sched_data {
37 u32 limit; /* HARD maximal queue length */
38
39 unsigned char flags;
40 /* Non-flags in tc_red_qopt.flags. */
41 unsigned char userbits;
42
43 struct timer_list adapt_timer;
44 struct Qdisc *sch;
45 struct red_parms parms;
46 struct red_vars vars;
47 struct red_stats stats;
48 struct Qdisc *qdisc;
49 struct tcf_qevent qe_early_drop;
50 struct tcf_qevent qe_mark;
51};
52
53#define TC_RED_SUPPORTED_FLAGS (TC_RED_HISTORIC_FLAGS | TC_RED_NODROP)
54
55static inline int red_use_ecn(struct red_sched_data *q)
56{
57 return q->flags & TC_RED_ECN;
58}
59
60static inline int red_use_harddrop(struct red_sched_data *q)
61{
62 return q->flags & TC_RED_HARDDROP;
63}
64
65static int red_use_nodrop(struct red_sched_data *q)
66{
67 return q->flags & TC_RED_NODROP;
68}
69
70static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
71 struct sk_buff **to_free)
72{
73 struct red_sched_data *q = qdisc_priv(sch);
74 struct Qdisc *child = q->qdisc;
75 unsigned int len;
76 int ret;
77
78 q->vars.qavg = red_calc_qavg(&q->parms,
79 &q->vars,
80 child->qstats.backlog);
81
82 if (red_is_idling(&q->vars))
83 red_end_of_idle_period(&q->vars);
84
85 switch (red_action(&q->parms, &q->vars, q->vars.qavg)) {
86 case RED_DONT_MARK:
87 break;
88
89 case RED_PROB_MARK:
90 qdisc_qstats_overlimit(sch);
91 if (!red_use_ecn(q)) {
92 q->stats.prob_drop++;
93 goto congestion_drop;
94 }
95
96 if (INET_ECN_set_ce(skb)) {
97 q->stats.prob_mark++;
98 skb = tcf_qevent_handle(&q->qe_mark, sch, skb, to_free, &ret);
99 if (!skb)
100 return NET_XMIT_CN | ret;
101 } else if (!red_use_nodrop(q)) {
102 q->stats.prob_drop++;
103 goto congestion_drop;
104 }
105
106 /* Non-ECT packet in ECN nodrop mode: queue it. */
107 break;
108
109 case RED_HARD_MARK:
110 qdisc_qstats_overlimit(sch);
111 if (red_use_harddrop(q) || !red_use_ecn(q)) {
112 q->stats.forced_drop++;
113 goto congestion_drop;
114 }
115
116 if (INET_ECN_set_ce(skb)) {
117 q->stats.forced_mark++;
118 skb = tcf_qevent_handle(&q->qe_mark, sch, skb, to_free, &ret);
119 if (!skb)
120 return NET_XMIT_CN | ret;
121 } else if (!red_use_nodrop(q)) {
122 q->stats.forced_drop++;
123 goto congestion_drop;
124 }
125
126 /* Non-ECT packet in ECN nodrop mode: queue it. */
127 break;
128 }
129
130 len = qdisc_pkt_len(skb);
131 ret = qdisc_enqueue(skb, child, to_free);
132 if (likely(ret == NET_XMIT_SUCCESS)) {
133 sch->qstats.backlog += len;
134 sch->q.qlen++;
135 } else if (net_xmit_drop_count(ret)) {
136 q->stats.pdrop++;
137 qdisc_qstats_drop(sch);
138 }
139 return ret;
140
141congestion_drop:
142 skb = tcf_qevent_handle(&q->qe_early_drop, sch, skb, to_free, &ret);
143 if (!skb)
144 return NET_XMIT_CN | ret;
145
146 qdisc_drop(skb, sch, to_free);
147 return NET_XMIT_CN;
148}
149
150static struct sk_buff *red_dequeue(struct Qdisc *sch)
151{
152 struct sk_buff *skb;
153 struct red_sched_data *q = qdisc_priv(sch);
154 struct Qdisc *child = q->qdisc;
155
156 skb = child->dequeue(child);
157 if (skb) {
158 qdisc_bstats_update(sch, skb);
159 qdisc_qstats_backlog_dec(sch, skb);
160 sch->q.qlen--;
161 } else {
162 if (!red_is_idling(&q->vars))
163 red_start_of_idle_period(&q->vars);
164 }
165 return skb;
166}
167
168static struct sk_buff *red_peek(struct Qdisc *sch)
169{
170 struct red_sched_data *q = qdisc_priv(sch);
171 struct Qdisc *child = q->qdisc;
172
173 return child->ops->peek(child);
174}
175
176static void red_reset(struct Qdisc *sch)
177{
178 struct red_sched_data *q = qdisc_priv(sch);
179
180 qdisc_reset(q->qdisc);
181 red_restart(&q->vars);
182}
183
184static int red_offload(struct Qdisc *sch, bool enable)
185{
186 struct red_sched_data *q = qdisc_priv(sch);
187 struct net_device *dev = qdisc_dev(sch);
188 struct tc_red_qopt_offload opt = {
189 .handle = sch->handle,
190 .parent = sch->parent,
191 };
192
193 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
194 return -EOPNOTSUPP;
195
196 if (enable) {
197 opt.command = TC_RED_REPLACE;
198 opt.set.min = q->parms.qth_min >> q->parms.Wlog;
199 opt.set.max = q->parms.qth_max >> q->parms.Wlog;
200 opt.set.probability = q->parms.max_P;
201 opt.set.limit = q->limit;
202 opt.set.is_ecn = red_use_ecn(q);
203 opt.set.is_harddrop = red_use_harddrop(q);
204 opt.set.is_nodrop = red_use_nodrop(q);
205 opt.set.qstats = &sch->qstats;
206 } else {
207 opt.command = TC_RED_DESTROY;
208 }
209
210 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt);
211}
212
213static void red_destroy(struct Qdisc *sch)
214{
215 struct red_sched_data *q = qdisc_priv(sch);
216
217 tcf_qevent_destroy(&q->qe_mark, sch);
218 tcf_qevent_destroy(&q->qe_early_drop, sch);
219 del_timer_sync(&q->adapt_timer);
220 red_offload(sch, false);
221 qdisc_put(q->qdisc);
222}
223
224static const struct nla_policy red_policy[TCA_RED_MAX + 1] = {
225 [TCA_RED_UNSPEC] = { .strict_start_type = TCA_RED_FLAGS },
226 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) },
227 [TCA_RED_STAB] = { .len = RED_STAB_SIZE },
228 [TCA_RED_MAX_P] = { .type = NLA_U32 },
229 [TCA_RED_FLAGS] = NLA_POLICY_BITFIELD32(TC_RED_SUPPORTED_FLAGS),
230 [TCA_RED_EARLY_DROP_BLOCK] = { .type = NLA_U32 },
231 [TCA_RED_MARK_BLOCK] = { .type = NLA_U32 },
232};
233
234static int __red_change(struct Qdisc *sch, struct nlattr **tb,
235 struct netlink_ext_ack *extack)
236{
237 struct Qdisc *old_child = NULL, *child = NULL;
238 struct red_sched_data *q = qdisc_priv(sch);
239 struct nla_bitfield32 flags_bf;
240 struct tc_red_qopt *ctl;
241 unsigned char userbits;
242 unsigned char flags;
243 int err;
244 u32 max_P;
245 u8 *stab;
246
247 if (tb[TCA_RED_PARMS] == NULL ||
248 tb[TCA_RED_STAB] == NULL)
249 return -EINVAL;
250
251 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0;
252
253 ctl = nla_data(tb[TCA_RED_PARMS]);
254 stab = nla_data(tb[TCA_RED_STAB]);
255 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog,
256 ctl->Scell_log, stab))
257 return -EINVAL;
258
259 err = red_get_flags(ctl->flags, TC_RED_HISTORIC_FLAGS,
260 tb[TCA_RED_FLAGS], TC_RED_SUPPORTED_FLAGS,
261 &flags_bf, &userbits, extack);
262 if (err)
263 return err;
264
265 if (ctl->limit > 0) {
266 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit,
267 extack);
268 if (IS_ERR(child))
269 return PTR_ERR(child);
270
271 /* child is fifo, no need to check for noop_qdisc */
272 qdisc_hash_add(child, true);
273 }
274
275 sch_tree_lock(sch);
276
277 flags = (q->flags & ~flags_bf.selector) | flags_bf.value;
278 err = red_validate_flags(flags, extack);
279 if (err)
280 goto unlock_out;
281
282 q->flags = flags;
283 q->userbits = userbits;
284 q->limit = ctl->limit;
285 if (child) {
286 qdisc_tree_flush_backlog(q->qdisc);
287 old_child = q->qdisc;
288 q->qdisc = child;
289 }
290
291 red_set_parms(&q->parms,
292 ctl->qth_min, ctl->qth_max, ctl->Wlog,
293 ctl->Plog, ctl->Scell_log,
294 stab,
295 max_P);
296 red_set_vars(&q->vars);
297
298 del_timer(&q->adapt_timer);
299 if (ctl->flags & TC_RED_ADAPTATIVE)
300 mod_timer(&q->adapt_timer, jiffies + HZ/2);
301
302 if (!q->qdisc->q.qlen)
303 red_start_of_idle_period(&q->vars);
304
305 sch_tree_unlock(sch);
306
307 red_offload(sch, true);
308
309 if (old_child)
310 qdisc_put(old_child);
311 return 0;
312
313unlock_out:
314 sch_tree_unlock(sch);
315 if (child)
316 qdisc_put(child);
317 return err;
318}
319
320static inline void red_adaptative_timer(struct timer_list *t)
321{
322 struct red_sched_data *q = from_timer(q, t, adapt_timer);
323 struct Qdisc *sch = q->sch;
324 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
325
326 spin_lock(root_lock);
327 red_adaptative_algo(&q->parms, &q->vars);
328 mod_timer(&q->adapt_timer, jiffies + HZ/2);
329 spin_unlock(root_lock);
330}
331
332static int red_init(struct Qdisc *sch, struct nlattr *opt,
333 struct netlink_ext_ack *extack)
334{
335 struct red_sched_data *q = qdisc_priv(sch);
336 struct nlattr *tb[TCA_RED_MAX + 1];
337 int err;
338
339 q->qdisc = &noop_qdisc;
340 q->sch = sch;
341 timer_setup(&q->adapt_timer, red_adaptative_timer, 0);
342
343 if (!opt)
344 return -EINVAL;
345
346 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy,
347 extack);
348 if (err < 0)
349 return err;
350
351 err = __red_change(sch, tb, extack);
352 if (err)
353 return err;
354
355 err = tcf_qevent_init(&q->qe_early_drop, sch,
356 FLOW_BLOCK_BINDER_TYPE_RED_EARLY_DROP,
357 tb[TCA_RED_EARLY_DROP_BLOCK], extack);
358 if (err)
359 return err;
360
361 return tcf_qevent_init(&q->qe_mark, sch,
362 FLOW_BLOCK_BINDER_TYPE_RED_MARK,
363 tb[TCA_RED_MARK_BLOCK], extack);
364}
365
366static int red_change(struct Qdisc *sch, struct nlattr *opt,
367 struct netlink_ext_ack *extack)
368{
369 struct red_sched_data *q = qdisc_priv(sch);
370 struct nlattr *tb[TCA_RED_MAX + 1];
371 int err;
372
373 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy,
374 extack);
375 if (err < 0)
376 return err;
377
378 err = tcf_qevent_validate_change(&q->qe_early_drop,
379 tb[TCA_RED_EARLY_DROP_BLOCK], extack);
380 if (err)
381 return err;
382
383 err = tcf_qevent_validate_change(&q->qe_mark,
384 tb[TCA_RED_MARK_BLOCK], extack);
385 if (err)
386 return err;
387
388 return __red_change(sch, tb, extack);
389}
390
391static int red_dump_offload_stats(struct Qdisc *sch)
392{
393 struct tc_red_qopt_offload hw_stats = {
394 .command = TC_RED_STATS,
395 .handle = sch->handle,
396 .parent = sch->parent,
397 {
398 .stats.bstats = &sch->bstats,
399 .stats.qstats = &sch->qstats,
400 },
401 };
402
403 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_RED, &hw_stats);
404}
405
406static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
407{
408 struct red_sched_data *q = qdisc_priv(sch);
409 struct nlattr *opts = NULL;
410 struct tc_red_qopt opt = {
411 .limit = q->limit,
412 .flags = (q->flags & TC_RED_HISTORIC_FLAGS) |
413 q->userbits,
414 .qth_min = q->parms.qth_min >> q->parms.Wlog,
415 .qth_max = q->parms.qth_max >> q->parms.Wlog,
416 .Wlog = q->parms.Wlog,
417 .Plog = q->parms.Plog,
418 .Scell_log = q->parms.Scell_log,
419 };
420 int err;
421
422 err = red_dump_offload_stats(sch);
423 if (err)
424 goto nla_put_failure;
425
426 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
427 if (opts == NULL)
428 goto nla_put_failure;
429 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) ||
430 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P) ||
431 nla_put_bitfield32(skb, TCA_RED_FLAGS,
432 q->flags, TC_RED_SUPPORTED_FLAGS) ||
433 tcf_qevent_dump(skb, TCA_RED_MARK_BLOCK, &q->qe_mark) ||
434 tcf_qevent_dump(skb, TCA_RED_EARLY_DROP_BLOCK, &q->qe_early_drop))
435 goto nla_put_failure;
436 return nla_nest_end(skb, opts);
437
438nla_put_failure:
439 nla_nest_cancel(skb, opts);
440 return -EMSGSIZE;
441}
442
443static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
444{
445 struct red_sched_data *q = qdisc_priv(sch);
446 struct net_device *dev = qdisc_dev(sch);
447 struct tc_red_xstats st = {0};
448
449 if (sch->flags & TCQ_F_OFFLOADED) {
450 struct tc_red_qopt_offload hw_stats_request = {
451 .command = TC_RED_XSTATS,
452 .handle = sch->handle,
453 .parent = sch->parent,
454 {
455 .xstats = &q->stats,
456 },
457 };
458 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED,
459 &hw_stats_request);
460 }
461 st.early = q->stats.prob_drop + q->stats.forced_drop;
462 st.pdrop = q->stats.pdrop;
463 st.marked = q->stats.prob_mark + q->stats.forced_mark;
464
465 return gnet_stats_copy_app(d, &st, sizeof(st));
466}
467
468static int red_dump_class(struct Qdisc *sch, unsigned long cl,
469 struct sk_buff *skb, struct tcmsg *tcm)
470{
471 struct red_sched_data *q = qdisc_priv(sch);
472
473 tcm->tcm_handle |= TC_H_MIN(1);
474 tcm->tcm_info = q->qdisc->handle;
475 return 0;
476}
477
478static void red_graft_offload(struct Qdisc *sch,
479 struct Qdisc *new, struct Qdisc *old,
480 struct netlink_ext_ack *extack)
481{
482 struct tc_red_qopt_offload graft_offload = {
483 .handle = sch->handle,
484 .parent = sch->parent,
485 .child_handle = new->handle,
486 .command = TC_RED_GRAFT,
487 };
488
489 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, old,
490 TC_SETUP_QDISC_RED, &graft_offload, extack);
491}
492
493static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
494 struct Qdisc **old, struct netlink_ext_ack *extack)
495{
496 struct red_sched_data *q = qdisc_priv(sch);
497
498 if (new == NULL)
499 new = &noop_qdisc;
500
501 *old = qdisc_replace(sch, new, &q->qdisc);
502
503 red_graft_offload(sch, new, *old, extack);
504 return 0;
505}
506
507static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
508{
509 struct red_sched_data *q = qdisc_priv(sch);
510 return q->qdisc;
511}
512
513static unsigned long red_find(struct Qdisc *sch, u32 classid)
514{
515 return 1;
516}
517
518static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
519{
520 if (!walker->stop) {
521 tc_qdisc_stats_dump(sch, 1, walker);
522 }
523}
524
525static const struct Qdisc_class_ops red_class_ops = {
526 .graft = red_graft,
527 .leaf = red_leaf,
528 .find = red_find,
529 .walk = red_walk,
530 .dump = red_dump_class,
531};
532
533static struct Qdisc_ops red_qdisc_ops __read_mostly = {
534 .id = "red",
535 .priv_size = sizeof(struct red_sched_data),
536 .cl_ops = &red_class_ops,
537 .enqueue = red_enqueue,
538 .dequeue = red_dequeue,
539 .peek = red_peek,
540 .init = red_init,
541 .reset = red_reset,
542 .destroy = red_destroy,
543 .change = red_change,
544 .dump = red_dump,
545 .dump_stats = red_dump_stats,
546 .owner = THIS_MODULE,
547};
548
549static int __init red_module_init(void)
550{
551 return register_qdisc(&red_qdisc_ops);
552}
553
554static void __exit red_module_exit(void)
555{
556 unregister_qdisc(&red_qdisc_ops);
557}
558
559module_init(red_module_init)
560module_exit(red_module_exit)
561
562MODULE_LICENSE("GPL");