1/*
2 * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 */
9
10#include <linux/module.h>
11#include <linux/init.h>
12#include <linux/kernel.h>
13#include <linux/skbuff.h>
14#include <linux/rtnetlink.h>
15#include <linux/filter.h>
16#include <linux/bpf.h>
17
18#include <net/netlink.h>
19#include <net/pkt_sched.h>
20
21#include <linux/tc_act/tc_bpf.h>
22#include <net/tc_act/tc_bpf.h>
23
24#define BPF_TAB_MASK		15
25#define ACT_BPF_NAME_LEN	256
26
27struct tcf_bpf_cfg {
28	struct bpf_prog *filter;
29	struct sock_filter *bpf_ops;
30	const char *bpf_name;
31	u32 bpf_fd;
32	u16 bpf_num_ops;
33	bool is_ebpf;
34};
35
36static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
37		   struct tcf_result *res)
38{
39	struct tcf_bpf *prog = act->priv;
40	struct bpf_prog *filter;
41	int action, filter_res;
42	bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
43
44	if (unlikely(!skb_mac_header_was_set(skb)))
45		return TC_ACT_UNSPEC;
46
47	tcf_lastuse_update(&prog->tcf_tm);
48	bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
49
50	rcu_read_lock();
51	filter = rcu_dereference(prog->filter);
52	if (at_ingress) {
53		__skb_push(skb, skb->mac_len);
54		filter_res = BPF_PROG_RUN(filter, skb);
55		__skb_pull(skb, skb->mac_len);
56	} else {
57		filter_res = BPF_PROG_RUN(filter, skb);
58	}
59	rcu_read_unlock();
60
61	/* A BPF program may overwrite the default action opcode.
62	 * Similarly as in cls_bpf, if filter_res == -1 we use the
63	 * default action specified from tc.
64	 *
65	 * In case a different well-known TC_ACT opcode has been
66	 * returned, it will overwrite the default one.
67	 *
68	 * For everything else that is unkown, TC_ACT_UNSPEC is
69	 * returned.
70	 */
71	switch (filter_res) {
72	case TC_ACT_PIPE:
73	case TC_ACT_RECLASSIFY:
74	case TC_ACT_OK:
75	case TC_ACT_REDIRECT:
76		action = filter_res;
77		break;
78	case TC_ACT_SHOT:
79		action = filter_res;
80		qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
81		break;
82	case TC_ACT_UNSPEC:
83		action = prog->tcf_action;
84		break;
85	default:
86		action = TC_ACT_UNSPEC;
87		break;
88	}
89
90	return action;
91}
92
93static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
94{
95	return !prog->bpf_ops;
96}
97
98static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
99				 struct sk_buff *skb)
100{
101	struct nlattr *nla;
102
103	if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
104		return -EMSGSIZE;
105
106	nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
107			  sizeof(struct sock_filter));
108	if (nla == NULL)
109		return -EMSGSIZE;
110
111	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
112
113	return 0;
114}
115
116static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
117				  struct sk_buff *skb)
118{
119	if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
120		return -EMSGSIZE;
121
122	if (prog->bpf_name &&
123	    nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
124		return -EMSGSIZE;
125
126	return 0;
127}
128
129static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
130			int bind, int ref)
131{
132	unsigned char *tp = skb_tail_pointer(skb);
133	struct tcf_bpf *prog = act->priv;
134	struct tc_act_bpf opt = {
135		.index   = prog->tcf_index,
136		.refcnt  = prog->tcf_refcnt - ref,
137		.bindcnt = prog->tcf_bindcnt - bind,
138		.action  = prog->tcf_action,
139	};
140	struct tcf_t tm;
141	int ret;
142
143	if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
144		goto nla_put_failure;
145
146	if (tcf_bpf_is_ebpf(prog))
147		ret = tcf_bpf_dump_ebpf_info(prog, skb);
148	else
149		ret = tcf_bpf_dump_bpf_info(prog, skb);
150	if (ret)
151		goto nla_put_failure;
152
153	tm.install = jiffies_to_clock_t(jiffies - prog->tcf_tm.install);
154	tm.lastuse = jiffies_to_clock_t(jiffies - prog->tcf_tm.lastuse);
155	tm.expires = jiffies_to_clock_t(prog->tcf_tm.expires);
156
157	if (nla_put(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm))
158		goto nla_put_failure;
159
160	return skb->len;
161
162nla_put_failure:
163	nlmsg_trim(skb, tp);
164	return -1;
165}
166
167static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
168	[TCA_ACT_BPF_PARMS]	= { .len = sizeof(struct tc_act_bpf) },
169	[TCA_ACT_BPF_FD]	= { .type = NLA_U32 },
170	[TCA_ACT_BPF_NAME]	= { .type = NLA_NUL_STRING, .len = ACT_BPF_NAME_LEN },
171	[TCA_ACT_BPF_OPS_LEN]	= { .type = NLA_U16 },
172	[TCA_ACT_BPF_OPS]	= { .type = NLA_BINARY,
173				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
174};
175
176static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
177{
178	struct sock_filter *bpf_ops;
179	struct sock_fprog_kern fprog_tmp;
180	struct bpf_prog *fp;
181	u16 bpf_size, bpf_num_ops;
182	int ret;
183
184	bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
185	if (bpf_num_ops	> BPF_MAXINSNS || bpf_num_ops == 0)
186		return -EINVAL;
187
188	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
189	if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
190		return -EINVAL;
191
192	bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
193	if (bpf_ops == NULL)
194		return -ENOMEM;
195
196	memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
197
198	fprog_tmp.len = bpf_num_ops;
199	fprog_tmp.filter = bpf_ops;
200
201	ret = bpf_prog_create(&fp, &fprog_tmp);
202	if (ret < 0) {
203		kfree(bpf_ops);
204		return ret;
205	}
206
207	cfg->bpf_ops = bpf_ops;
208	cfg->bpf_num_ops = bpf_num_ops;
209	cfg->filter = fp;
210	cfg->is_ebpf = false;
211
212	return 0;
213}
214
215static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
216{
217	struct bpf_prog *fp;
218	char *name = NULL;
219	u32 bpf_fd;
220
221	bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
222
223	fp = bpf_prog_get(bpf_fd);
224	if (IS_ERR(fp))
225		return PTR_ERR(fp);
226
227	if (fp->type != BPF_PROG_TYPE_SCHED_ACT) {
228		bpf_prog_put(fp);
229		return -EINVAL;
230	}
231
232	if (tb[TCA_ACT_BPF_NAME]) {
233		name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]),
234			       nla_len(tb[TCA_ACT_BPF_NAME]),
235			       GFP_KERNEL);
236		if (!name) {
237			bpf_prog_put(fp);
238			return -ENOMEM;
239		}
240	}
241
242	cfg->bpf_fd = bpf_fd;
243	cfg->bpf_name = name;
244	cfg->filter = fp;
245	cfg->is_ebpf = true;
246
247	return 0;
248}
249
250static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
251{
252	if (cfg->is_ebpf)
253		bpf_prog_put(cfg->filter);
254	else
255		bpf_prog_destroy(cfg->filter);
256
257	kfree(cfg->bpf_ops);
258	kfree(cfg->bpf_name);
259}
260
261static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
262				  struct tcf_bpf_cfg *cfg)
263{
264	cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
265	/* updates to prog->filter are prevented, since it's called either
266	 * with rtnl lock or during final cleanup in rcu callback
267	 */
268	cfg->filter = rcu_dereference_protected(prog->filter, 1);
269
270	cfg->bpf_ops = prog->bpf_ops;
271	cfg->bpf_name = prog->bpf_name;
272}
273
274static int tcf_bpf_init(struct net *net, struct nlattr *nla,
275			struct nlattr *est, struct tc_action *act,
276			int replace, int bind)
277{
278	struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
279	struct tcf_bpf_cfg cfg, old;
280	struct tc_act_bpf *parm;
281	struct tcf_bpf *prog;
282	bool is_bpf, is_ebpf;
283	int ret, res = 0;
284
285	if (!nla)
286		return -EINVAL;
287
288	ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
289	if (ret < 0)
290		return ret;
291
292	if (!tb[TCA_ACT_BPF_PARMS])
293		return -EINVAL;
294
295	parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
296
297	if (!tcf_hash_check(parm->index, act, bind)) {
298		ret = tcf_hash_create(parm->index, est, act,
299				      sizeof(*prog), bind, true);
300		if (ret < 0)
301			return ret;
302
303		res = ACT_P_CREATED;
304	} else {
305		/* Don't override defaults. */
306		if (bind)
307			return 0;
308
309		tcf_hash_release(act, bind);
310		if (!replace)
311			return -EEXIST;
312	}
313
314	is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
315	is_ebpf = tb[TCA_ACT_BPF_FD];
316
317	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
318		ret = -EINVAL;
319		goto out;
320	}
321
322	memset(&cfg, 0, sizeof(cfg));
323
324	ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
325		       tcf_bpf_init_from_efd(tb, &cfg);
326	if (ret < 0)
327		goto out;
328
329	prog = to_bpf(act);
330	ASSERT_RTNL();
331
332	if (res != ACT_P_CREATED)
333		tcf_bpf_prog_fill_cfg(prog, &old);
334
335	prog->bpf_ops = cfg.bpf_ops;
336	prog->bpf_name = cfg.bpf_name;
337
338	if (cfg.bpf_num_ops)
339		prog->bpf_num_ops = cfg.bpf_num_ops;
340	if (cfg.bpf_fd)
341		prog->bpf_fd = cfg.bpf_fd;
342
343	prog->tcf_action = parm->action;
344	rcu_assign_pointer(prog->filter, cfg.filter);
345
346	if (res == ACT_P_CREATED) {
347		tcf_hash_insert(act);
348	} else {
349		/* make sure the program being replaced is no longer executing */
350		synchronize_rcu();
351		tcf_bpf_cfg_cleanup(&old);
352	}
353
354	return res;
355out:
356	if (res == ACT_P_CREATED)
357		tcf_hash_cleanup(act, est);
358
359	return ret;
360}
361
362static void tcf_bpf_cleanup(struct tc_action *act, int bind)
363{
364	struct tcf_bpf_cfg tmp;
365
366	tcf_bpf_prog_fill_cfg(act->priv, &tmp);
367	tcf_bpf_cfg_cleanup(&tmp);
368}
369
370static struct tc_action_ops act_bpf_ops __read_mostly = {
371	.kind		=	"bpf",
372	.type		=	TCA_ACT_BPF,
373	.owner		=	THIS_MODULE,
374	.act		=	tcf_bpf,
375	.dump		=	tcf_bpf_dump,
376	.cleanup	=	tcf_bpf_cleanup,
377	.init		=	tcf_bpf_init,
378};
379
380static int __init bpf_init_module(void)
381{
382	return tcf_register_action(&act_bpf_ops, BPF_TAB_MASK);
383}
384
385static void __exit bpf_cleanup_module(void)
386{
387	tcf_unregister_action(&act_bpf_ops);
388}
389
390module_init(bpf_init_module);
391module_exit(bpf_cleanup_module);
392
393MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
394MODULE_DESCRIPTION("TC BPF based action");
395MODULE_LICENSE("GPL v2");
396