xref: /openbmc/linux/net/sched/act_bpf.c (revision bc5aa3a0)
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 
27 struct 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 
36 static int bpf_net_id;
37 static struct tc_action_ops act_bpf_ops;
38 
39 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
40 		   struct tcf_result *res)
41 {
42 	struct tcf_bpf *prog = to_bpf(act);
43 	struct bpf_prog *filter;
44 	int action, filter_res;
45 	bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
46 
47 	if (unlikely(!skb_mac_header_was_set(skb)))
48 		return TC_ACT_UNSPEC;
49 
50 	tcf_lastuse_update(&prog->tcf_tm);
51 	bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
52 
53 	rcu_read_lock();
54 	filter = rcu_dereference(prog->filter);
55 	if (at_ingress) {
56 		__skb_push(skb, skb->mac_len);
57 		bpf_compute_data_end(skb);
58 		filter_res = BPF_PROG_RUN(filter, skb);
59 		__skb_pull(skb, skb->mac_len);
60 	} else {
61 		bpf_compute_data_end(skb);
62 		filter_res = BPF_PROG_RUN(filter, skb);
63 	}
64 	rcu_read_unlock();
65 
66 	/* A BPF program may overwrite the default action opcode.
67 	 * Similarly as in cls_bpf, if filter_res == -1 we use the
68 	 * default action specified from tc.
69 	 *
70 	 * In case a different well-known TC_ACT opcode has been
71 	 * returned, it will overwrite the default one.
72 	 *
73 	 * For everything else that is unkown, TC_ACT_UNSPEC is
74 	 * returned.
75 	 */
76 	switch (filter_res) {
77 	case TC_ACT_PIPE:
78 	case TC_ACT_RECLASSIFY:
79 	case TC_ACT_OK:
80 	case TC_ACT_REDIRECT:
81 		action = filter_res;
82 		break;
83 	case TC_ACT_SHOT:
84 		action = filter_res;
85 		qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
86 		break;
87 	case TC_ACT_UNSPEC:
88 		action = prog->tcf_action;
89 		break;
90 	default:
91 		action = TC_ACT_UNSPEC;
92 		break;
93 	}
94 
95 	return action;
96 }
97 
98 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
99 {
100 	return !prog->bpf_ops;
101 }
102 
103 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
104 				 struct sk_buff *skb)
105 {
106 	struct nlattr *nla;
107 
108 	if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
109 		return -EMSGSIZE;
110 
111 	nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
112 			  sizeof(struct sock_filter));
113 	if (nla == NULL)
114 		return -EMSGSIZE;
115 
116 	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
117 
118 	return 0;
119 }
120 
121 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
122 				  struct sk_buff *skb)
123 {
124 	if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
125 		return -EMSGSIZE;
126 
127 	if (prog->bpf_name &&
128 	    nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
129 		return -EMSGSIZE;
130 
131 	return 0;
132 }
133 
134 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
135 			int bind, int ref)
136 {
137 	unsigned char *tp = skb_tail_pointer(skb);
138 	struct tcf_bpf *prog = to_bpf(act);
139 	struct tc_act_bpf opt = {
140 		.index   = prog->tcf_index,
141 		.refcnt  = prog->tcf_refcnt - ref,
142 		.bindcnt = prog->tcf_bindcnt - bind,
143 		.action  = prog->tcf_action,
144 	};
145 	struct tcf_t tm;
146 	int ret;
147 
148 	if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
149 		goto nla_put_failure;
150 
151 	if (tcf_bpf_is_ebpf(prog))
152 		ret = tcf_bpf_dump_ebpf_info(prog, skb);
153 	else
154 		ret = tcf_bpf_dump_bpf_info(prog, skb);
155 	if (ret)
156 		goto nla_put_failure;
157 
158 	tcf_tm_dump(&tm, &prog->tcf_tm);
159 	if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
160 			  TCA_ACT_BPF_PAD))
161 		goto nla_put_failure;
162 
163 	return skb->len;
164 
165 nla_put_failure:
166 	nlmsg_trim(skb, tp);
167 	return -1;
168 }
169 
170 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
171 	[TCA_ACT_BPF_PARMS]	= { .len = sizeof(struct tc_act_bpf) },
172 	[TCA_ACT_BPF_FD]	= { .type = NLA_U32 },
173 	[TCA_ACT_BPF_NAME]	= { .type = NLA_NUL_STRING,
174 				    .len = ACT_BPF_NAME_LEN },
175 	[TCA_ACT_BPF_OPS_LEN]	= { .type = NLA_U16 },
176 	[TCA_ACT_BPF_OPS]	= { .type = NLA_BINARY,
177 				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
178 };
179 
180 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
181 {
182 	struct sock_filter *bpf_ops;
183 	struct sock_fprog_kern fprog_tmp;
184 	struct bpf_prog *fp;
185 	u16 bpf_size, bpf_num_ops;
186 	int ret;
187 
188 	bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
189 	if (bpf_num_ops	> BPF_MAXINSNS || bpf_num_ops == 0)
190 		return -EINVAL;
191 
192 	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
193 	if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
194 		return -EINVAL;
195 
196 	bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
197 	if (bpf_ops == NULL)
198 		return -ENOMEM;
199 
200 	memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
201 
202 	fprog_tmp.len = bpf_num_ops;
203 	fprog_tmp.filter = bpf_ops;
204 
205 	ret = bpf_prog_create(&fp, &fprog_tmp);
206 	if (ret < 0) {
207 		kfree(bpf_ops);
208 		return ret;
209 	}
210 
211 	cfg->bpf_ops = bpf_ops;
212 	cfg->bpf_num_ops = bpf_num_ops;
213 	cfg->filter = fp;
214 	cfg->is_ebpf = false;
215 
216 	return 0;
217 }
218 
219 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
220 {
221 	struct bpf_prog *fp;
222 	char *name = NULL;
223 	u32 bpf_fd;
224 
225 	bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
226 
227 	fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
228 	if (IS_ERR(fp))
229 		return PTR_ERR(fp);
230 
231 	if (tb[TCA_ACT_BPF_NAME]) {
232 		name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]),
233 			       nla_len(tb[TCA_ACT_BPF_NAME]),
234 			       GFP_KERNEL);
235 		if (!name) {
236 			bpf_prog_put(fp);
237 			return -ENOMEM;
238 		}
239 	}
240 
241 	cfg->bpf_fd = bpf_fd;
242 	cfg->bpf_name = name;
243 	cfg->filter = fp;
244 	cfg->is_ebpf = true;
245 
246 	return 0;
247 }
248 
249 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
250 {
251 	if (cfg->is_ebpf)
252 		bpf_prog_put(cfg->filter);
253 	else
254 		bpf_prog_destroy(cfg->filter);
255 
256 	kfree(cfg->bpf_ops);
257 	kfree(cfg->bpf_name);
258 }
259 
260 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
261 				  struct tcf_bpf_cfg *cfg)
262 {
263 	cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
264 	/* updates to prog->filter are prevented, since it's called either
265 	 * with rtnl lock or during final cleanup in rcu callback
266 	 */
267 	cfg->filter = rcu_dereference_protected(prog->filter, 1);
268 
269 	cfg->bpf_ops = prog->bpf_ops;
270 	cfg->bpf_name = prog->bpf_name;
271 }
272 
273 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
274 			struct nlattr *est, struct tc_action **act,
275 			int replace, int bind)
276 {
277 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
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(tn, parm->index, act, bind)) {
298 		ret = tcf_hash_create(tn, parm->index, est, act,
299 				      &act_bpf_ops, 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(tn, *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;
355 out:
356 	if (res == ACT_P_CREATED)
357 		tcf_hash_cleanup(*act, est);
358 
359 	return ret;
360 }
361 
362 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
363 {
364 	struct tcf_bpf_cfg tmp;
365 
366 	tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
367 	tcf_bpf_cfg_cleanup(&tmp);
368 }
369 
370 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
371 			  struct netlink_callback *cb, int type,
372 			  const struct tc_action_ops *ops)
373 {
374 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
375 
376 	return tcf_generic_walker(tn, skb, cb, type, ops);
377 }
378 
379 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
380 {
381 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
382 
383 	return tcf_hash_search(tn, a, index);
384 }
385 
386 static struct tc_action_ops act_bpf_ops __read_mostly = {
387 	.kind		=	"bpf",
388 	.type		=	TCA_ACT_BPF,
389 	.owner		=	THIS_MODULE,
390 	.act		=	tcf_bpf,
391 	.dump		=	tcf_bpf_dump,
392 	.cleanup	=	tcf_bpf_cleanup,
393 	.init		=	tcf_bpf_init,
394 	.walk		=	tcf_bpf_walker,
395 	.lookup		=	tcf_bpf_search,
396 	.size		=	sizeof(struct tcf_bpf),
397 };
398 
399 static __net_init int bpf_init_net(struct net *net)
400 {
401 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
402 
403 	return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK);
404 }
405 
406 static void __net_exit bpf_exit_net(struct net *net)
407 {
408 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
409 
410 	tc_action_net_exit(tn);
411 }
412 
413 static struct pernet_operations bpf_net_ops = {
414 	.init = bpf_init_net,
415 	.exit = bpf_exit_net,
416 	.id   = &bpf_net_id,
417 	.size = sizeof(struct tc_action_net),
418 };
419 
420 static int __init bpf_init_module(void)
421 {
422 	return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
423 }
424 
425 static void __exit bpf_cleanup_module(void)
426 {
427 	tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
428 }
429 
430 module_init(bpf_init_module);
431 module_exit(bpf_cleanup_module);
432 
433 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
434 MODULE_DESCRIPTION("TC BPF based action");
435 MODULE_LICENSE("GPL v2");
436