xref: /openbmc/linux/net/sched/sch_skbprio.c (revision 176f011b)
1 /*
2  * net/sched/sch_skbprio.c  SKB Priority 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:	Nishanth Devarajan, <ndev2021@gmail.com>
10  *		Cody Doucette, <doucette@bu.edu>
11  *	        original idea by Michel Machado, Cody Doucette, and Qiaobin Fu
12  */
13 
14 #include <linux/string.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <net/pkt_sched.h>
22 #include <net/sch_generic.h>
23 #include <net/inet_ecn.h>
24 
25 /*		SKB Priority Queue
26  *	=================================
27  *
28  * Skbprio (SKB Priority Queue) is a queueing discipline that prioritizes
29  * packets according to their skb->priority field. Under congestion,
30  * Skbprio drops already-enqueued lower priority packets to make space
31  * available for higher priority packets; it was conceived as a solution
32  * for denial-of-service defenses that need to route packets with different
33  * priorities as a mean to overcome DoS attacks.
34  */
35 
36 struct skbprio_sched_data {
37 	/* Queue state. */
38 	struct sk_buff_head qdiscs[SKBPRIO_MAX_PRIORITY];
39 	struct gnet_stats_queue qstats[SKBPRIO_MAX_PRIORITY];
40 	u16 highest_prio;
41 	u16 lowest_prio;
42 };
43 
44 static u16 calc_new_high_prio(const struct skbprio_sched_data *q)
45 {
46 	int prio;
47 
48 	for (prio = q->highest_prio - 1; prio >= q->lowest_prio; prio--) {
49 		if (!skb_queue_empty(&q->qdiscs[prio]))
50 			return prio;
51 	}
52 
53 	/* SKB queue is empty, return 0 (default highest priority setting). */
54 	return 0;
55 }
56 
57 static u16 calc_new_low_prio(const struct skbprio_sched_data *q)
58 {
59 	int prio;
60 
61 	for (prio = q->lowest_prio + 1; prio <= q->highest_prio; prio++) {
62 		if (!skb_queue_empty(&q->qdiscs[prio]))
63 			return prio;
64 	}
65 
66 	/* SKB queue is empty, return SKBPRIO_MAX_PRIORITY - 1
67 	 * (default lowest priority setting).
68 	 */
69 	return SKBPRIO_MAX_PRIORITY - 1;
70 }
71 
72 static int skbprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
73 			  struct sk_buff **to_free)
74 {
75 	const unsigned int max_priority = SKBPRIO_MAX_PRIORITY - 1;
76 	struct skbprio_sched_data *q = qdisc_priv(sch);
77 	struct sk_buff_head *qdisc;
78 	struct sk_buff_head *lp_qdisc;
79 	struct sk_buff *to_drop;
80 	u16 prio, lp;
81 
82 	/* Obtain the priority of @skb. */
83 	prio = min(skb->priority, max_priority);
84 
85 	qdisc = &q->qdiscs[prio];
86 	if (sch->q.qlen < sch->limit) {
87 		__skb_queue_tail(qdisc, skb);
88 		qdisc_qstats_backlog_inc(sch, skb);
89 		q->qstats[prio].backlog += qdisc_pkt_len(skb);
90 
91 		/* Check to update highest and lowest priorities. */
92 		if (prio > q->highest_prio)
93 			q->highest_prio = prio;
94 
95 		if (prio < q->lowest_prio)
96 			q->lowest_prio = prio;
97 
98 		sch->q.qlen++;
99 		return NET_XMIT_SUCCESS;
100 	}
101 
102 	/* If this packet has the lowest priority, drop it. */
103 	lp = q->lowest_prio;
104 	if (prio <= lp) {
105 		q->qstats[prio].drops++;
106 		q->qstats[prio].overlimits++;
107 		return qdisc_drop(skb, sch, to_free);
108 	}
109 
110 	__skb_queue_tail(qdisc, skb);
111 	qdisc_qstats_backlog_inc(sch, skb);
112 	q->qstats[prio].backlog += qdisc_pkt_len(skb);
113 
114 	/* Drop the packet at the tail of the lowest priority qdisc. */
115 	lp_qdisc = &q->qdiscs[lp];
116 	to_drop = __skb_dequeue_tail(lp_qdisc);
117 	BUG_ON(!to_drop);
118 	qdisc_qstats_backlog_dec(sch, to_drop);
119 	qdisc_drop(to_drop, sch, to_free);
120 
121 	q->qstats[lp].backlog -= qdisc_pkt_len(to_drop);
122 	q->qstats[lp].drops++;
123 	q->qstats[lp].overlimits++;
124 
125 	/* Check to update highest and lowest priorities. */
126 	if (skb_queue_empty(lp_qdisc)) {
127 		if (q->lowest_prio == q->highest_prio) {
128 			/* The incoming packet is the only packet in queue. */
129 			BUG_ON(sch->q.qlen != 1);
130 			q->lowest_prio = prio;
131 			q->highest_prio = prio;
132 		} else {
133 			q->lowest_prio = calc_new_low_prio(q);
134 		}
135 	}
136 
137 	if (prio > q->highest_prio)
138 		q->highest_prio = prio;
139 
140 	return NET_XMIT_CN;
141 }
142 
143 static struct sk_buff *skbprio_dequeue(struct Qdisc *sch)
144 {
145 	struct skbprio_sched_data *q = qdisc_priv(sch);
146 	struct sk_buff_head *hpq = &q->qdiscs[q->highest_prio];
147 	struct sk_buff *skb = __skb_dequeue(hpq);
148 
149 	if (unlikely(!skb))
150 		return NULL;
151 
152 	sch->q.qlen--;
153 	qdisc_qstats_backlog_dec(sch, skb);
154 	qdisc_bstats_update(sch, skb);
155 
156 	q->qstats[q->highest_prio].backlog -= qdisc_pkt_len(skb);
157 
158 	/* Update highest priority field. */
159 	if (skb_queue_empty(hpq)) {
160 		if (q->lowest_prio == q->highest_prio) {
161 			BUG_ON(sch->q.qlen);
162 			q->highest_prio = 0;
163 			q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
164 		} else {
165 			q->highest_prio = calc_new_high_prio(q);
166 		}
167 	}
168 	return skb;
169 }
170 
171 static int skbprio_change(struct Qdisc *sch, struct nlattr *opt,
172 			struct netlink_ext_ack *extack)
173 {
174 	struct tc_skbprio_qopt *ctl = nla_data(opt);
175 
176 	sch->limit = ctl->limit;
177 	return 0;
178 }
179 
180 static int skbprio_init(struct Qdisc *sch, struct nlattr *opt,
181 			struct netlink_ext_ack *extack)
182 {
183 	struct skbprio_sched_data *q = qdisc_priv(sch);
184 	int prio;
185 
186 	/* Initialise all queues, one for each possible priority. */
187 	for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
188 		__skb_queue_head_init(&q->qdiscs[prio]);
189 
190 	memset(&q->qstats, 0, sizeof(q->qstats));
191 	q->highest_prio = 0;
192 	q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
193 	sch->limit = 64;
194 	if (!opt)
195 		return 0;
196 
197 	return skbprio_change(sch, opt, extack);
198 }
199 
200 static int skbprio_dump(struct Qdisc *sch, struct sk_buff *skb)
201 {
202 	struct tc_skbprio_qopt opt;
203 
204 	opt.limit = sch->limit;
205 
206 	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
207 		return -1;
208 
209 	return skb->len;
210 }
211 
212 static void skbprio_reset(struct Qdisc *sch)
213 {
214 	struct skbprio_sched_data *q = qdisc_priv(sch);
215 	int prio;
216 
217 	sch->qstats.backlog = 0;
218 	sch->q.qlen = 0;
219 
220 	for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
221 		__skb_queue_purge(&q->qdiscs[prio]);
222 
223 	memset(&q->qstats, 0, sizeof(q->qstats));
224 	q->highest_prio = 0;
225 	q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
226 }
227 
228 static void skbprio_destroy(struct Qdisc *sch)
229 {
230 	struct skbprio_sched_data *q = qdisc_priv(sch);
231 	int prio;
232 
233 	for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
234 		__skb_queue_purge(&q->qdiscs[prio]);
235 }
236 
237 static struct Qdisc *skbprio_leaf(struct Qdisc *sch, unsigned long arg)
238 {
239 	return NULL;
240 }
241 
242 static unsigned long skbprio_find(struct Qdisc *sch, u32 classid)
243 {
244 	return 0;
245 }
246 
247 static int skbprio_dump_class(struct Qdisc *sch, unsigned long cl,
248 			     struct sk_buff *skb, struct tcmsg *tcm)
249 {
250 	tcm->tcm_handle |= TC_H_MIN(cl);
251 	return 0;
252 }
253 
254 static int skbprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
255 				   struct gnet_dump *d)
256 {
257 	struct skbprio_sched_data *q = qdisc_priv(sch);
258 	if (gnet_stats_copy_queue(d, NULL, &q->qstats[cl - 1],
259 		q->qstats[cl - 1].qlen) < 0)
260 		return -1;
261 	return 0;
262 }
263 
264 static void skbprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
265 {
266 	unsigned int i;
267 
268 	if (arg->stop)
269 		return;
270 
271 	for (i = 0; i < SKBPRIO_MAX_PRIORITY; i++) {
272 		if (arg->count < arg->skip) {
273 			arg->count++;
274 			continue;
275 		}
276 		if (arg->fn(sch, i + 1, arg) < 0) {
277 			arg->stop = 1;
278 			break;
279 		}
280 		arg->count++;
281 	}
282 }
283 
284 static const struct Qdisc_class_ops skbprio_class_ops = {
285 	.leaf		=	skbprio_leaf,
286 	.find		=	skbprio_find,
287 	.dump		=	skbprio_dump_class,
288 	.dump_stats	=	skbprio_dump_class_stats,
289 	.walk		=	skbprio_walk,
290 };
291 
292 static struct Qdisc_ops skbprio_qdisc_ops __read_mostly = {
293 	.cl_ops		=	&skbprio_class_ops,
294 	.id		=	"skbprio",
295 	.priv_size	=	sizeof(struct skbprio_sched_data),
296 	.enqueue	=	skbprio_enqueue,
297 	.dequeue	=	skbprio_dequeue,
298 	.peek		=	qdisc_peek_dequeued,
299 	.init		=	skbprio_init,
300 	.reset		=	skbprio_reset,
301 	.change		=	skbprio_change,
302 	.dump		=	skbprio_dump,
303 	.destroy	=	skbprio_destroy,
304 	.owner		=	THIS_MODULE,
305 };
306 
307 static int __init skbprio_module_init(void)
308 {
309 	return register_qdisc(&skbprio_qdisc_ops);
310 }
311 
312 static void __exit skbprio_module_exit(void)
313 {
314 	unregister_qdisc(&skbprio_qdisc_ops);
315 }
316 
317 module_init(skbprio_module_init)
318 module_exit(skbprio_module_exit)
319 
320 MODULE_LICENSE("GPL");
321