cls_bpf.c revision 1fab9abc56036256265f350c700d3d1251a47f0a 
1/*
2 * Berkeley Packet Filter based traffic classifier
3 *
4 * Might be used to classify traffic through flexible, user-defined and
5 * possibly JIT-ed BPF filters for traffic control as an alternative to
6 * ematches.
7 *
8 * (C) 2013 Daniel Borkmann <dborkman@redhat.com>
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
13 */
14
15#include <linux/module.h>
16#include <linux/types.h>
17#include <linux/skbuff.h>
18#include <linux/filter.h>
19#include <net/rtnetlink.h>
20#include <net/pkt_cls.h>
21#include <net/sock.h>
22
23MODULE_LICENSE("GPL");
24MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
25MODULE_DESCRIPTION("TC BPF based classifier");
26
27struct cls_bpf_head {
28	struct list_head plist;
29	u32 hgen;
30};
31
32struct cls_bpf_prog {
33	struct sk_filter *filter;
34	struct sock_filter *bpf_ops;
35	struct tcf_exts exts;
36	struct tcf_result res;
37	struct list_head link;
38	u32 handle;
39	u16 bpf_len;
40};
41
42static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
43	[TCA_BPF_CLASSID]	= { .type = NLA_U32 },
44	[TCA_BPF_OPS_LEN]	= { .type = NLA_U16 },
45	[TCA_BPF_OPS]		= { .type = NLA_BINARY,
46				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
47};
48
49static const struct tcf_ext_map bpf_ext_map = {
50	.action = TCA_BPF_ACT,
51	.police = TCA_BPF_POLICE,
52};
53
54static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
55			    struct tcf_result *res)
56{
57	struct cls_bpf_head *head = tp->root;
58	struct cls_bpf_prog *prog;
59	int ret;
60
61	list_for_each_entry(prog, &head->plist, link) {
62		int filter_res = SK_RUN_FILTER(prog->filter, skb);
63
64		if (filter_res == 0)
65			continue;
66
67		*res = prog->res;
68		if (filter_res != -1)
69			res->classid = filter_res;
70
71		ret = tcf_exts_exec(skb, &prog->exts, res);
72		if (ret < 0)
73			continue;
74
75		return ret;
76	}
77
78	return -1;
79}
80
81static int cls_bpf_init(struct tcf_proto *tp)
82{
83	struct cls_bpf_head *head;
84
85	head = kzalloc(sizeof(*head), GFP_KERNEL);
86	if (head == NULL)
87		return -ENOBUFS;
88
89	INIT_LIST_HEAD(&head->plist);
90	tp->root = head;
91
92	return 0;
93}
94
95static void cls_bpf_delete_prog(struct tcf_proto *tp, struct cls_bpf_prog *prog)
96{
97	tcf_unbind_filter(tp, &prog->res);
98	tcf_exts_destroy(tp, &prog->exts);
99
100	sk_unattached_filter_destroy(prog->filter);
101
102	kfree(prog->bpf_ops);
103	kfree(prog);
104}
105
106static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
107{
108	struct cls_bpf_head *head = tp->root;
109	struct cls_bpf_prog *prog, *todel = (struct cls_bpf_prog *) arg;
110
111	list_for_each_entry(prog, &head->plist, link) {
112		if (prog == todel) {
113			tcf_tree_lock(tp);
114			list_del(&prog->link);
115			tcf_tree_unlock(tp);
116
117			cls_bpf_delete_prog(tp, prog);
118			return 0;
119		}
120	}
121
122	return -ENOENT;
123}
124
125static void cls_bpf_destroy(struct tcf_proto *tp)
126{
127	struct cls_bpf_head *head = tp->root;
128	struct cls_bpf_prog *prog, *tmp;
129
130	list_for_each_entry_safe(prog, tmp, &head->plist, link) {
131		list_del(&prog->link);
132		cls_bpf_delete_prog(tp, prog);
133	}
134
135	kfree(head);
136}
137
138static unsigned long cls_bpf_get(struct tcf_proto *tp, u32 handle)
139{
140	struct cls_bpf_head *head = tp->root;
141	struct cls_bpf_prog *prog;
142	unsigned long ret = 0UL;
143
144	if (head == NULL)
145		return 0UL;
146
147	list_for_each_entry(prog, &head->plist, link) {
148		if (prog->handle == handle) {
149			ret = (unsigned long) prog;
150			break;
151		}
152	}
153
154	return ret;
155}
156
157static void cls_bpf_put(struct tcf_proto *tp, unsigned long f)
158{
159}
160
161static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
162				   struct cls_bpf_prog *prog,
163				   unsigned long base, struct nlattr **tb,
164				   struct nlattr *est)
165{
166	struct sock_filter *bpf_ops, *bpf_old;
167	struct tcf_exts exts;
168	struct sock_fprog tmp;
169	struct sk_filter *fp, *fp_old;
170	u16 bpf_size, bpf_len;
171	u32 classid;
172	int ret;
173
174	if (!tb[TCA_BPF_OPS_LEN] || !tb[TCA_BPF_OPS] || !tb[TCA_BPF_CLASSID])
175		return -EINVAL;
176
177	ret = tcf_exts_validate(net, tp, tb, est, &exts, &bpf_ext_map);
178	if (ret < 0)
179		return ret;
180
181	classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
182	bpf_len = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
183	if (bpf_len > BPF_MAXINSNS || bpf_len == 0) {
184		ret = -EINVAL;
185		goto errout;
186	}
187
188	bpf_size = bpf_len * sizeof(*bpf_ops);
189	bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
190	if (bpf_ops == NULL) {
191		ret = -ENOMEM;
192		goto errout;
193	}
194
195	memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
196
197	tmp.len = bpf_len;
198	tmp.filter = (struct sock_filter __user *) bpf_ops;
199
200	ret = sk_unattached_filter_create(&fp, &tmp);
201	if (ret)
202		goto errout_free;
203
204	tcf_tree_lock(tp);
205	fp_old = prog->filter;
206	bpf_old = prog->bpf_ops;
207
208	prog->bpf_len = bpf_len;
209	prog->bpf_ops = bpf_ops;
210	prog->filter = fp;
211	prog->res.classid = classid;
212	tcf_tree_unlock(tp);
213
214	tcf_bind_filter(tp, &prog->res, base);
215	tcf_exts_change(tp, &prog->exts, &exts);
216
217	if (fp_old)
218		sk_unattached_filter_destroy(fp_old);
219	if (bpf_old)
220		kfree(bpf_old);
221
222	return 0;
223
224errout_free:
225	kfree(bpf_ops);
226errout:
227	tcf_exts_destroy(tp, &exts);
228	return ret;
229}
230
231static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
232				   struct cls_bpf_head *head)
233{
234	unsigned int i = 0x80000000;
235
236	do {
237		if (++head->hgen == 0x7FFFFFFF)
238			head->hgen = 1;
239	} while (--i > 0 && cls_bpf_get(tp, head->hgen));
240	if (i == 0)
241		pr_err("Insufficient number of handles\n");
242
243	return i;
244}
245
246static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
247			  struct tcf_proto *tp, unsigned long base,
248			  u32 handle, struct nlattr **tca,
249			  unsigned long *arg)
250{
251	struct cls_bpf_head *head = tp->root;
252	struct cls_bpf_prog *prog = (struct cls_bpf_prog *) *arg;
253	struct nlattr *tb[TCA_BPF_MAX + 1];
254	int ret;
255
256	if (tca[TCA_OPTIONS] == NULL)
257		return -EINVAL;
258
259	ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
260	if (ret < 0)
261		return ret;
262
263	if (prog != NULL) {
264		if (handle && prog->handle != handle)
265			return -EINVAL;
266		return cls_bpf_modify_existing(net, tp, prog, base, tb,
267					       tca[TCA_RATE]);
268	}
269
270	prog = kzalloc(sizeof(*prog), GFP_KERNEL);
271	if (prog == NULL)
272		return -ENOBUFS;
273
274	if (handle == 0)
275		prog->handle = cls_bpf_grab_new_handle(tp, head);
276	else
277		prog->handle = handle;
278	if (prog->handle == 0) {
279		ret = -EINVAL;
280		goto errout;
281	}
282
283	ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE]);
284	if (ret < 0)
285		goto errout;
286
287	tcf_tree_lock(tp);
288	list_add(&prog->link, &head->plist);
289	tcf_tree_unlock(tp);
290
291	*arg = (unsigned long) prog;
292
293	return 0;
294errout:
295	if (*arg == 0UL && prog)
296		kfree(prog);
297
298	return ret;
299}
300
301static int cls_bpf_dump(struct tcf_proto *tp, unsigned long fh,
302			struct sk_buff *skb, struct tcmsg *tm)
303{
304	struct cls_bpf_prog *prog = (struct cls_bpf_prog *) fh;
305	struct nlattr *nest, *nla;
306
307	if (prog == NULL)
308		return skb->len;
309
310	tm->tcm_handle = prog->handle;
311
312	nest = nla_nest_start(skb, TCA_OPTIONS);
313	if (nest == NULL)
314		goto nla_put_failure;
315
316	if (nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
317		goto nla_put_failure;
318	if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_len))
319		goto nla_put_failure;
320
321	nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_len *
322			  sizeof(struct sock_filter));
323	if (nla == NULL)
324		goto nla_put_failure;
325
326	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
327
328	if (tcf_exts_dump(skb, &prog->exts, &bpf_ext_map) < 0)
329		goto nla_put_failure;
330
331	nla_nest_end(skb, nest);
332
333	if (tcf_exts_dump_stats(skb, &prog->exts, &bpf_ext_map) < 0)
334		goto nla_put_failure;
335
336	return skb->len;
337
338nla_put_failure:
339	nla_nest_cancel(skb, nest);
340	return -1;
341}
342
343static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
344{
345	struct cls_bpf_head *head = tp->root;
346	struct cls_bpf_prog *prog;
347
348	list_for_each_entry(prog, &head->plist, link) {
349		if (arg->count < arg->skip)
350			goto skip;
351		if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
352			arg->stop = 1;
353			break;
354		}
355skip:
356		arg->count++;
357	}
358}
359
360static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
361	.kind		=	"bpf",
362	.owner		=	THIS_MODULE,
363	.classify	=	cls_bpf_classify,
364	.init		=	cls_bpf_init,
365	.destroy	=	cls_bpf_destroy,
366	.get		=	cls_bpf_get,
367	.put		=	cls_bpf_put,
368	.change		=	cls_bpf_change,
369	.delete		=	cls_bpf_delete,
370	.walk		=	cls_bpf_walk,
371	.dump		=	cls_bpf_dump,
372};
373
374static int __init cls_bpf_init_mod(void)
375{
376	return register_tcf_proto_ops(&cls_bpf_ops);
377}
378
379static void __exit cls_bpf_exit_mod(void)
380{
381	unregister_tcf_proto_ops(&cls_bpf_ops);
382}
383
384module_init(cls_bpf_init_mod);
385module_exit(cls_bpf_exit_mod);
386