1/*
2 * net/sched/police.c	Input police filter.
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 * 		J Hadi Salim (action changes)
11 */
12
13#include <linux/module.h>
14#include <linux/types.h>
15#include <linux/kernel.h>
16#include <linux/string.h>
17#include <linux/errno.h>
18#include <linux/skbuff.h>
19#include <linux/rtnetlink.h>
20#include <linux/init.h>
21#include <linux/slab.h>
22#include <net/act_api.h>
23#include <net/netlink.h>
24
25struct tcf_police {
26	struct tcf_common	common;
27	int			tcfp_result;
28	u32			tcfp_ewma_rate;
29	s64			tcfp_burst;
30	u32			tcfp_mtu;
31	s64			tcfp_toks;
32	s64			tcfp_ptoks;
33	s64			tcfp_mtu_ptoks;
34	s64			tcfp_t_c;
35	struct psched_ratecfg	rate;
36	bool			rate_present;
37	struct psched_ratecfg	peak;
38	bool			peak_present;
39};
40#define to_police(pc)	\
41	container_of(pc, struct tcf_police, common)
42
43#define POL_TAB_MASK     15
44
45/* old policer structure from before tc actions */
46struct tc_police_compat {
47	u32			index;
48	int			action;
49	u32			limit;
50	u32			burst;
51	u32			mtu;
52	struct tc_ratespec	rate;
53	struct tc_ratespec	peakrate;
54};
55
56/* Each policer is serialized by its individual spinlock */
57
58static int tcf_act_police_walker(struct sk_buff *skb, struct netlink_callback *cb,
59			      int type, struct tc_action *a)
60{
61	struct tcf_hashinfo *hinfo = a->ops->hinfo;
62	struct hlist_head *head;
63	struct tcf_common *p;
64	int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
65	struct nlattr *nest;
66
67	spin_lock_bh(&hinfo->lock);
68
69	s_i = cb->args[0];
70
71	for (i = 0; i < (POL_TAB_MASK + 1); i++) {
72		head = &hinfo->htab[tcf_hash(i, POL_TAB_MASK)];
73
74		hlist_for_each_entry_rcu(p, head, tcfc_head) {
75			index++;
76			if (index < s_i)
77				continue;
78			a->priv = p;
79			a->order = index;
80			nest = nla_nest_start(skb, a->order);
81			if (nest == NULL)
82				goto nla_put_failure;
83			if (type == RTM_DELACTION)
84				err = tcf_action_dump_1(skb, a, 0, 1);
85			else
86				err = tcf_action_dump_1(skb, a, 0, 0);
87			if (err < 0) {
88				index--;
89				nla_nest_cancel(skb, nest);
90				goto done;
91			}
92			nla_nest_end(skb, nest);
93			n_i++;
94		}
95	}
96done:
97	spin_unlock_bh(&hinfo->lock);
98	if (n_i)
99		cb->args[0] += n_i;
100	return n_i;
101
102nla_put_failure:
103	nla_nest_cancel(skb, nest);
104	goto done;
105}
106
107static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
108	[TCA_POLICE_RATE]	= { .len = TC_RTAB_SIZE },
109	[TCA_POLICE_PEAKRATE]	= { .len = TC_RTAB_SIZE },
110	[TCA_POLICE_AVRATE]	= { .type = NLA_U32 },
111	[TCA_POLICE_RESULT]	= { .type = NLA_U32 },
112};
113
114static int tcf_act_police_locate(struct net *net, struct nlattr *nla,
115				 struct nlattr *est, struct tc_action *a,
116				 int ovr, int bind)
117{
118	unsigned int h;
119	int ret = 0, err;
120	struct nlattr *tb[TCA_POLICE_MAX + 1];
121	struct tc_police *parm;
122	struct tcf_police *police;
123	struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
124	struct tcf_hashinfo *hinfo = a->ops->hinfo;
125	int size;
126
127	if (nla == NULL)
128		return -EINVAL;
129
130	err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy);
131	if (err < 0)
132		return err;
133
134	if (tb[TCA_POLICE_TBF] == NULL)
135		return -EINVAL;
136	size = nla_len(tb[TCA_POLICE_TBF]);
137	if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
138		return -EINVAL;
139	parm = nla_data(tb[TCA_POLICE_TBF]);
140
141	if (parm->index) {
142		if (tcf_hash_search(a, parm->index)) {
143			police = to_police(a->priv);
144			if (bind) {
145				police->tcf_bindcnt += 1;
146				police->tcf_refcnt += 1;
147				return 0;
148			}
149			if (ovr)
150				goto override;
151			/* not replacing */
152			return -EEXIST;
153		}
154	}
155
156	police = kzalloc(sizeof(*police), GFP_KERNEL);
157	if (police == NULL)
158		return -ENOMEM;
159	ret = ACT_P_CREATED;
160	police->tcf_refcnt = 1;
161	spin_lock_init(&police->tcf_lock);
162	if (bind)
163		police->tcf_bindcnt = 1;
164override:
165	if (parm->rate.rate) {
166		err = -ENOMEM;
167		R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE]);
168		if (R_tab == NULL)
169			goto failure;
170
171		if (parm->peakrate.rate) {
172			P_tab = qdisc_get_rtab(&parm->peakrate,
173					       tb[TCA_POLICE_PEAKRATE]);
174			if (P_tab == NULL)
175				goto failure;
176		}
177	}
178
179	spin_lock_bh(&police->tcf_lock);
180	if (est) {
181		err = gen_replace_estimator(&police->tcf_bstats, NULL,
182					    &police->tcf_rate_est,
183					    &police->tcf_lock, est);
184		if (err)
185			goto failure_unlock;
186	} else if (tb[TCA_POLICE_AVRATE] &&
187		   (ret == ACT_P_CREATED ||
188		    !gen_estimator_active(&police->tcf_bstats,
189					  &police->tcf_rate_est))) {
190		err = -EINVAL;
191		goto failure_unlock;
192	}
193
194	/* No failure allowed after this point */
195	police->tcfp_mtu = parm->mtu;
196	if (police->tcfp_mtu == 0) {
197		police->tcfp_mtu = ~0;
198		if (R_tab)
199			police->tcfp_mtu = 255 << R_tab->rate.cell_log;
200	}
201	if (R_tab) {
202		police->rate_present = true;
203		psched_ratecfg_precompute(&police->rate, &R_tab->rate, 0);
204		qdisc_put_rtab(R_tab);
205	} else {
206		police->rate_present = false;
207	}
208	if (P_tab) {
209		police->peak_present = true;
210		psched_ratecfg_precompute(&police->peak, &P_tab->rate, 0);
211		qdisc_put_rtab(P_tab);
212	} else {
213		police->peak_present = false;
214	}
215
216	if (tb[TCA_POLICE_RESULT])
217		police->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
218	police->tcfp_burst = PSCHED_TICKS2NS(parm->burst);
219	police->tcfp_toks = police->tcfp_burst;
220	if (police->peak_present) {
221		police->tcfp_mtu_ptoks = (s64) psched_l2t_ns(&police->peak,
222							     police->tcfp_mtu);
223		police->tcfp_ptoks = police->tcfp_mtu_ptoks;
224	}
225	police->tcf_action = parm->action;
226
227	if (tb[TCA_POLICE_AVRATE])
228		police->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
229
230	spin_unlock_bh(&police->tcf_lock);
231	if (ret != ACT_P_CREATED)
232		return ret;
233
234	police->tcfp_t_c = ktime_get_ns();
235	police->tcf_index = parm->index ? parm->index :
236		tcf_hash_new_index(hinfo);
237	h = tcf_hash(police->tcf_index, POL_TAB_MASK);
238	spin_lock_bh(&hinfo->lock);
239	hlist_add_head(&police->tcf_head, &hinfo->htab[h]);
240	spin_unlock_bh(&hinfo->lock);
241
242	a->priv = police;
243	return ret;
244
245failure_unlock:
246	spin_unlock_bh(&police->tcf_lock);
247failure:
248	qdisc_put_rtab(P_tab);
249	qdisc_put_rtab(R_tab);
250	if (ret == ACT_P_CREATED)
251		kfree(police);
252	return err;
253}
254
255static int tcf_act_police(struct sk_buff *skb, const struct tc_action *a,
256			  struct tcf_result *res)
257{
258	struct tcf_police *police = a->priv;
259	s64 now;
260	s64 toks;
261	s64 ptoks = 0;
262
263	spin_lock(&police->tcf_lock);
264
265	bstats_update(&police->tcf_bstats, skb);
266
267	if (police->tcfp_ewma_rate &&
268	    police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
269		police->tcf_qstats.overlimits++;
270		if (police->tcf_action == TC_ACT_SHOT)
271			police->tcf_qstats.drops++;
272		spin_unlock(&police->tcf_lock);
273		return police->tcf_action;
274	}
275
276	if (qdisc_pkt_len(skb) <= police->tcfp_mtu) {
277		if (!police->rate_present) {
278			spin_unlock(&police->tcf_lock);
279			return police->tcfp_result;
280		}
281
282		now = ktime_get_ns();
283		toks = min_t(s64, now - police->tcfp_t_c,
284			     police->tcfp_burst);
285		if (police->peak_present) {
286			ptoks = toks + police->tcfp_ptoks;
287			if (ptoks > police->tcfp_mtu_ptoks)
288				ptoks = police->tcfp_mtu_ptoks;
289			ptoks -= (s64) psched_l2t_ns(&police->peak,
290						     qdisc_pkt_len(skb));
291		}
292		toks += police->tcfp_toks;
293		if (toks > police->tcfp_burst)
294			toks = police->tcfp_burst;
295		toks -= (s64) psched_l2t_ns(&police->rate, qdisc_pkt_len(skb));
296		if ((toks|ptoks) >= 0) {
297			police->tcfp_t_c = now;
298			police->tcfp_toks = toks;
299			police->tcfp_ptoks = ptoks;
300			spin_unlock(&police->tcf_lock);
301			return police->tcfp_result;
302		}
303	}
304
305	police->tcf_qstats.overlimits++;
306	if (police->tcf_action == TC_ACT_SHOT)
307		police->tcf_qstats.drops++;
308	spin_unlock(&police->tcf_lock);
309	return police->tcf_action;
310}
311
312static int
313tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
314{
315	unsigned char *b = skb_tail_pointer(skb);
316	struct tcf_police *police = a->priv;
317	struct tc_police opt = {
318		.index = police->tcf_index,
319		.action = police->tcf_action,
320		.mtu = police->tcfp_mtu,
321		.burst = PSCHED_NS2TICKS(police->tcfp_burst),
322		.refcnt = police->tcf_refcnt - ref,
323		.bindcnt = police->tcf_bindcnt - bind,
324	};
325
326	if (police->rate_present)
327		psched_ratecfg_getrate(&opt.rate, &police->rate);
328	if (police->peak_present)
329		psched_ratecfg_getrate(&opt.peakrate, &police->peak);
330	if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
331		goto nla_put_failure;
332	if (police->tcfp_result &&
333	    nla_put_u32(skb, TCA_POLICE_RESULT, police->tcfp_result))
334		goto nla_put_failure;
335	if (police->tcfp_ewma_rate &&
336	    nla_put_u32(skb, TCA_POLICE_AVRATE, police->tcfp_ewma_rate))
337		goto nla_put_failure;
338	return skb->len;
339
340nla_put_failure:
341	nlmsg_trim(skb, b);
342	return -1;
343}
344
345MODULE_AUTHOR("Alexey Kuznetsov");
346MODULE_DESCRIPTION("Policing actions");
347MODULE_LICENSE("GPL");
348
349static struct tc_action_ops act_police_ops = {
350	.kind		=	"police",
351	.type		=	TCA_ID_POLICE,
352	.owner		=	THIS_MODULE,
353	.act		=	tcf_act_police,
354	.dump		=	tcf_act_police_dump,
355	.init		=	tcf_act_police_locate,
356	.walk		=	tcf_act_police_walker
357};
358
359static int __init
360police_init_module(void)
361{
362	return tcf_register_action(&act_police_ops, POL_TAB_MASK);
363}
364
365static void __exit
366police_cleanup_module(void)
367{
368	tcf_unregister_action(&act_police_ops);
369}
370
371module_init(police_init_module);
372module_exit(police_cleanup_module);
373