sch_skbprio.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * net/sched/sch_skbprio.c  SKB Priority Queue.
  4 *
  5 * Authors:	Nishanth Devarajan, <ndev2021@gmail.com>
  6 *		Cody Doucette, <doucette@bu.edu>
  7 *	        original idea by Michel Machado, Cody Doucette, and Qiaobin Fu
  8 */
  9
 10#include <linux/string.h>
 11#include <linux/module.h>
 12#include <linux/slab.h>
 13#include <linux/types.h>
 14#include <linux/kernel.h>
 15#include <linux/errno.h>
 16#include <linux/skbuff.h>
 17#include <net/pkt_sched.h>
 18#include <net/sch_generic.h>
 19#include <net/inet_ecn.h>
 20
 21/*		SKB Priority Queue
 22 *	=================================
 23 *
 24 * Skbprio (SKB Priority Queue) is a queueing discipline that prioritizes
 25 * packets according to their skb->priority field. Under congestion,
 26 * Skbprio drops already-enqueued lower priority packets to make space
 27 * available for higher priority packets; it was conceived as a solution
 28 * for denial-of-service defenses that need to route packets with different
 29 * priorities as a mean to overcome DoS attacks.
 30 */
 31
 32struct skbprio_sched_data {
 33	/* Queue state. */
 34	struct sk_buff_head qdiscs[SKBPRIO_MAX_PRIORITY];
 35	struct gnet_stats_queue qstats[SKBPRIO_MAX_PRIORITY];
 36	u16 highest_prio;
 37	u16 lowest_prio;
 38};
 39
 40static u16 calc_new_high_prio(const struct skbprio_sched_data *q)
 41{
 42	int prio;
 43
 44	for (prio = q->highest_prio - 1; prio >= q->lowest_prio; prio--) {
 45		if (!skb_queue_empty(&q->qdiscs[prio]))
 46			return prio;
 47	}
 48
 49	/* SKB queue is empty, return 0 (default highest priority setting). */
 50	return 0;
 51}
 52
 53static u16 calc_new_low_prio(const struct skbprio_sched_data *q)
 54{
 55	int prio;
 56
 57	for (prio = q->lowest_prio + 1; prio <= q->highest_prio; prio++) {
 58		if (!skb_queue_empty(&q->qdiscs[prio]))
 59			return prio;
 60	}
 61
 62	/* SKB queue is empty, return SKBPRIO_MAX_PRIORITY - 1
 63	 * (default lowest priority setting).
 64	 */
 65	return SKBPRIO_MAX_PRIORITY - 1;
 66}
 67
 68static int skbprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 69			  struct sk_buff **to_free)
 70{
 71	const unsigned int max_priority = SKBPRIO_MAX_PRIORITY - 1;
 72	struct skbprio_sched_data *q = qdisc_priv(sch);
 73	struct sk_buff_head *qdisc;
 74	struct sk_buff_head *lp_qdisc;
 75	struct sk_buff *to_drop;
 76	u16 prio, lp;
 77
 78	/* Obtain the priority of @skb. */
 79	prio = min(skb->priority, max_priority);
 80
 81	qdisc = &q->qdiscs[prio];
 82
 83	/* sch->limit can change under us from skbprio_change() */
 84	if (sch->q.qlen < READ_ONCE(sch->limit)) {
 85		__skb_queue_tail(qdisc, skb);
 86		qdisc_qstats_backlog_inc(sch, skb);
 87		q->qstats[prio].backlog += qdisc_pkt_len(skb);
 88
 89		/* Check to update highest and lowest priorities. */
 90		if (prio > q->highest_prio)
 91			q->highest_prio = prio;
 92
 93		if (prio < q->lowest_prio)
 94			q->lowest_prio = prio;
 95
 96		sch->q.qlen++;
 97		return NET_XMIT_SUCCESS;
 98	}
 99
100	/* If this packet has the lowest priority, drop it. */
101	lp = q->lowest_prio;
102	if (prio <= lp) {
103		q->qstats[prio].drops++;
104		q->qstats[prio].overlimits++;
105		return qdisc_drop(skb, sch, to_free);
106	}
107
108	__skb_queue_tail(qdisc, skb);
109	qdisc_qstats_backlog_inc(sch, skb);
110	q->qstats[prio].backlog += qdisc_pkt_len(skb);
111
112	/* Drop the packet at the tail of the lowest priority qdisc. */
113	lp_qdisc = &q->qdiscs[lp];
114	to_drop = __skb_dequeue_tail(lp_qdisc);
115	BUG_ON(!to_drop);
116	qdisc_qstats_backlog_dec(sch, to_drop);
117	qdisc_drop(to_drop, sch, to_free);
118
119	q->qstats[lp].backlog -= qdisc_pkt_len(to_drop);
120	q->qstats[lp].drops++;
121	q->qstats[lp].overlimits++;
122
123	/* Check to update highest and lowest priorities. */
124	if (skb_queue_empty(lp_qdisc)) {
125		if (q->lowest_prio == q->highest_prio) {
126			q->lowest_prio = prio;
127			q->highest_prio = prio;
128		} else {
129			q->lowest_prio = calc_new_low_prio(q);
130		}
131	}
132
133	if (prio > q->highest_prio)
134		q->highest_prio = prio;
135
136	return NET_XMIT_CN;
137}
138
139static struct sk_buff *skbprio_dequeue(struct Qdisc *sch)
140{
141	struct skbprio_sched_data *q = qdisc_priv(sch);
142	struct sk_buff_head *hpq = &q->qdiscs[q->highest_prio];
143	struct sk_buff *skb = __skb_dequeue(hpq);
144
145	if (unlikely(!skb))
146		return NULL;
147
148	sch->q.qlen--;
149	qdisc_qstats_backlog_dec(sch, skb);
150	qdisc_bstats_update(sch, skb);
151
152	q->qstats[q->highest_prio].backlog -= qdisc_pkt_len(skb);
153
154	/* Update highest priority field. */
155	if (skb_queue_empty(hpq)) {
156		if (q->lowest_prio == q->highest_prio) {
157			q->highest_prio = 0;
158			q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
159		} else {
160			q->highest_prio = calc_new_high_prio(q);
161		}
162	}
163	return skb;
164}
165
166static int skbprio_change(struct Qdisc *sch, struct nlattr *opt,
167			struct netlink_ext_ack *extack)
168{
169	struct tc_skbprio_qopt *ctl = nla_data(opt);
170
171	if (opt->nla_len != nla_attr_size(sizeof(*ctl)))
172		return -EINVAL;
173
174	WRITE_ONCE(sch->limit, ctl->limit);
175	return 0;
176}
177
178static int skbprio_init(struct Qdisc *sch, struct nlattr *opt,
179			struct netlink_ext_ack *extack)
180{
181	struct skbprio_sched_data *q = qdisc_priv(sch);
182	int prio;
183
184	/* Initialise all queues, one for each possible priority. */
185	for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
186		__skb_queue_head_init(&q->qdiscs[prio]);
187
188	memset(&q->qstats, 0, sizeof(q->qstats));
189	q->highest_prio = 0;
190	q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
191	sch->limit = 64;
192	if (!opt)
193		return 0;
194
195	return skbprio_change(sch, opt, extack);
196}
197
198static int skbprio_dump(struct Qdisc *sch, struct sk_buff *skb)
199{
200	struct tc_skbprio_qopt opt;
201
202	opt.limit = READ_ONCE(sch->limit);
203
204	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
205		return -1;
206
207	return skb->len;
208}
209
210static void skbprio_reset(struct Qdisc *sch)
211{
212	struct skbprio_sched_data *q = qdisc_priv(sch);
213	int prio;
214
215	for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
216		__skb_queue_purge(&q->qdiscs[prio]);
217
218	memset(&q->qstats, 0, sizeof(q->qstats));
219	q->highest_prio = 0;
220	q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
221}
222
223static void skbprio_destroy(struct Qdisc *sch)
224{
225	struct skbprio_sched_data *q = qdisc_priv(sch);
226	int prio;
227
228	for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
229		__skb_queue_purge(&q->qdiscs[prio]);
230}
231
232static struct Qdisc *skbprio_leaf(struct Qdisc *sch, unsigned long arg)
233{
234	return NULL;
235}
236
237static unsigned long skbprio_find(struct Qdisc *sch, u32 classid)
238{
239	return 0;
240}
241
242static int skbprio_dump_class(struct Qdisc *sch, unsigned long cl,
243			     struct sk_buff *skb, struct tcmsg *tcm)
244{
245	tcm->tcm_handle |= TC_H_MIN(cl);
246	return 0;
247}
248
249static int skbprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
250				   struct gnet_dump *d)
251{
252	struct skbprio_sched_data *q = qdisc_priv(sch);
253	if (gnet_stats_copy_queue(d, NULL, &q->qstats[cl - 1],
254		q->qstats[cl - 1].qlen) < 0)
255		return -1;
256	return 0;
257}
258
259static void skbprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
260{
261	unsigned int i;
262
263	if (arg->stop)
264		return;
265
266	for (i = 0; i < SKBPRIO_MAX_PRIORITY; i++) {
267		if (!tc_qdisc_stats_dump(sch, i + 1, arg))
268			break;
269	}
270}
271
272static const struct Qdisc_class_ops skbprio_class_ops = {
273	.leaf		=	skbprio_leaf,
274	.find		=	skbprio_find,
275	.dump		=	skbprio_dump_class,
276	.dump_stats	=	skbprio_dump_class_stats,
277	.walk		=	skbprio_walk,
278};
279
280static struct Qdisc_ops skbprio_qdisc_ops __read_mostly = {
281	.cl_ops		=	&skbprio_class_ops,
282	.id		=	"skbprio",
283	.priv_size	=	sizeof(struct skbprio_sched_data),
284	.enqueue	=	skbprio_enqueue,
285	.dequeue	=	skbprio_dequeue,
286	.peek		=	qdisc_peek_dequeued,
287	.init		=	skbprio_init,
288	.reset		=	skbprio_reset,
289	.change		=	skbprio_change,
290	.dump		=	skbprio_dump,
291	.destroy	=	skbprio_destroy,
292	.owner		=	THIS_MODULE,
293};
294MODULE_ALIAS_NET_SCH("skbprio");
295
296static int __init skbprio_module_init(void)
297{
298	return register_qdisc(&skbprio_qdisc_ops);
299}
300
301static void __exit skbprio_module_exit(void)
302{
303	unregister_qdisc(&skbprio_qdisc_ops);
304}
305
306module_init(skbprio_module_init)
307module_exit(skbprio_module_exit)
308
309MODULE_LICENSE("GPL");
310MODULE_DESCRIPTION("SKB priority based scheduling qdisc");