xref: /linux/net/sched/act_bpf.c (revision fbcf8a340150abd20bf44fc706362b0827157fe8)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
4  */
5 
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/kernel.h>
9 #include <linux/skbuff.h>
10 #include <linux/rtnetlink.h>
11 #include <linux/filter.h>
12 #include <linux/bpf.h>
13 
14 #include <net/netlink.h>
15 #include <net/sock.h>
16 #include <net/pkt_sched.h>
17 #include <net/pkt_cls.h>
18 
19 #include <linux/tc_act/tc_bpf.h>
20 #include <net/tc_act/tc_bpf.h>
21 
22 #define ACT_BPF_NAME_LEN	256
23 
24 struct tcf_bpf_cfg {
25 	struct bpf_prog *filter;
26 	struct sock_filter *bpf_ops;
27 	const char *bpf_name;
28 	u16 bpf_num_ops;
29 	bool is_ebpf;
30 };
31 
32 static unsigned int bpf_net_id;
33 static struct tc_action_ops act_bpf_ops;
34 
35 static int tcf_bpf_act(struct sk_buff *skb, const struct tc_action *act,
36 		       struct tcf_result *res)
37 {
38 	bool at_ingress = skb_at_tc_ingress(skb);
39 	struct tcf_bpf *prog = to_bpf(act);
40 	struct bpf_prog *filter;
41 	int action, filter_res;
42 
43 	tcf_lastuse_update(&prog->tcf_tm);
44 	bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
45 
46 	filter = rcu_dereference(prog->filter);
47 	if (at_ingress) {
48 		__skb_push(skb, skb->mac_len);
49 		bpf_compute_data_pointers(skb);
50 		filter_res = BPF_PROG_RUN(filter, skb);
51 		__skb_pull(skb, skb->mac_len);
52 	} else {
53 		bpf_compute_data_pointers(skb);
54 		filter_res = BPF_PROG_RUN(filter, skb);
55 	}
56 	if (skb_sk_is_prefetched(skb) && filter_res != TC_ACT_OK)
57 		skb_orphan(skb);
58 
59 	/* A BPF program may overwrite the default action opcode.
60 	 * Similarly as in cls_bpf, if filter_res == -1 we use the
61 	 * default action specified from tc.
62 	 *
63 	 * In case a different well-known TC_ACT opcode has been
64 	 * returned, it will overwrite the default one.
65 	 *
66 	 * For everything else that is unknown, TC_ACT_UNSPEC is
67 	 * returned.
68 	 */
69 	switch (filter_res) {
70 	case TC_ACT_PIPE:
71 	case TC_ACT_RECLASSIFY:
72 	case TC_ACT_OK:
73 	case TC_ACT_REDIRECT:
74 		action = filter_res;
75 		break;
76 	case TC_ACT_SHOT:
77 		action = filter_res;
78 		qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
79 		break;
80 	case TC_ACT_UNSPEC:
81 		action = prog->tcf_action;
82 		break;
83 	default:
84 		action = TC_ACT_UNSPEC;
85 		break;
86 	}
87 
88 	return action;
89 }
90 
91 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
92 {
93 	return !prog->bpf_ops;
94 }
95 
96 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
97 				 struct sk_buff *skb)
98 {
99 	struct nlattr *nla;
100 
101 	if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
102 		return -EMSGSIZE;
103 
104 	nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
105 			  sizeof(struct sock_filter));
106 	if (nla == NULL)
107 		return -EMSGSIZE;
108 
109 	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
110 
111 	return 0;
112 }
113 
114 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
115 				  struct sk_buff *skb)
116 {
117 	struct nlattr *nla;
118 
119 	if (prog->bpf_name &&
120 	    nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
121 		return -EMSGSIZE;
122 
123 	if (nla_put_u32(skb, TCA_ACT_BPF_ID, prog->filter->aux->id))
124 		return -EMSGSIZE;
125 
126 	nla = nla_reserve(skb, TCA_ACT_BPF_TAG, sizeof(prog->filter->tag));
127 	if (nla == NULL)
128 		return -EMSGSIZE;
129 
130 	memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
131 
132 	return 0;
133 }
134 
135 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
136 			int bind, int ref)
137 {
138 	unsigned char *tp = skb_tail_pointer(skb);
139 	struct tcf_bpf *prog = to_bpf(act);
140 	struct tc_act_bpf opt = {
141 		.index   = prog->tcf_index,
142 		.refcnt  = refcount_read(&prog->tcf_refcnt) - ref,
143 		.bindcnt = atomic_read(&prog->tcf_bindcnt) - bind,
144 	};
145 	struct tcf_t tm;
146 	int ret;
147 
148 	spin_lock_bh(&prog->tcf_lock);
149 	opt.action = prog->tcf_action;
150 	if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
151 		goto nla_put_failure;
152 
153 	if (tcf_bpf_is_ebpf(prog))
154 		ret = tcf_bpf_dump_ebpf_info(prog, skb);
155 	else
156 		ret = tcf_bpf_dump_bpf_info(prog, skb);
157 	if (ret)
158 		goto nla_put_failure;
159 
160 	tcf_tm_dump(&tm, &prog->tcf_tm);
161 	if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
162 			  TCA_ACT_BPF_PAD))
163 		goto nla_put_failure;
164 
165 	spin_unlock_bh(&prog->tcf_lock);
166 	return skb->len;
167 
168 nla_put_failure:
169 	spin_unlock_bh(&prog->tcf_lock);
170 	nlmsg_trim(skb, tp);
171 	return -1;
172 }
173 
174 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
175 	[TCA_ACT_BPF_PARMS]	= { .len = sizeof(struct tc_act_bpf) },
176 	[TCA_ACT_BPF_FD]	= { .type = NLA_U32 },
177 	[TCA_ACT_BPF_NAME]	= { .type = NLA_NUL_STRING,
178 				    .len = ACT_BPF_NAME_LEN },
179 	[TCA_ACT_BPF_OPS_LEN]	= { .type = NLA_U16 },
180 	[TCA_ACT_BPF_OPS]	= { .type = NLA_BINARY,
181 				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
182 };
183 
184 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
185 {
186 	struct sock_filter *bpf_ops;
187 	struct sock_fprog_kern fprog_tmp;
188 	struct bpf_prog *fp;
189 	u16 bpf_size, bpf_num_ops;
190 	int ret;
191 
192 	bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
193 	if (bpf_num_ops	> BPF_MAXINSNS || bpf_num_ops == 0)
194 		return -EINVAL;
195 
196 	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
197 	if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
198 		return -EINVAL;
199 
200 	bpf_ops = kmemdup(nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size, GFP_KERNEL);
201 	if (bpf_ops == NULL)
202 		return -ENOMEM;
203 
204 	fprog_tmp.len = bpf_num_ops;
205 	fprog_tmp.filter = bpf_ops;
206 
207 	ret = bpf_prog_create(&fp, &fprog_tmp);
208 	if (ret < 0) {
209 		kfree(bpf_ops);
210 		return ret;
211 	}
212 
213 	cfg->bpf_ops = bpf_ops;
214 	cfg->bpf_num_ops = bpf_num_ops;
215 	cfg->filter = fp;
216 	cfg->is_ebpf = false;
217 
218 	return 0;
219 }
220 
221 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
222 {
223 	struct bpf_prog *fp;
224 	char *name = NULL;
225 	u32 bpf_fd;
226 
227 	bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
228 
229 	fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
230 	if (IS_ERR(fp))
231 		return PTR_ERR(fp);
232 
233 	if (tb[TCA_ACT_BPF_NAME]) {
234 		name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL);
235 		if (!name) {
236 			bpf_prog_put(fp);
237 			return -ENOMEM;
238 		}
239 	}
240 
241 	cfg->bpf_name = name;
242 	cfg->filter = fp;
243 	cfg->is_ebpf = true;
244 
245 	return 0;
246 }
247 
248 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
249 {
250 	struct bpf_prog *filter = cfg->filter;
251 
252 	if (filter) {
253 		if (cfg->is_ebpf)
254 			bpf_prog_put(filter);
255 		else
256 			bpf_prog_destroy(filter);
257 	}
258 
259 	kfree(cfg->bpf_ops);
260 	kfree(cfg->bpf_name);
261 }
262 
263 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
264 				  struct tcf_bpf_cfg *cfg)
265 {
266 	cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
267 	/* updates to prog->filter are prevented, since it's called either
268 	 * with tcf lock or during final cleanup in rcu callback
269 	 */
270 	cfg->filter = rcu_dereference_protected(prog->filter, 1);
271 
272 	cfg->bpf_ops = prog->bpf_ops;
273 	cfg->bpf_name = prog->bpf_name;
274 }
275 
276 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
277 			struct nlattr *est, struct tc_action **act,
278 			struct tcf_proto *tp, u32 flags,
279 			struct netlink_ext_ack *extack)
280 {
281 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
282 	bool bind = flags & TCA_ACT_FLAGS_BIND;
283 	struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
284 	struct tcf_chain *goto_ch = NULL;
285 	struct tcf_bpf_cfg cfg, old;
286 	struct tc_act_bpf *parm;
287 	struct tcf_bpf *prog;
288 	bool is_bpf, is_ebpf;
289 	int ret, res = 0;
290 	u32 index;
291 
292 	if (!nla)
293 		return -EINVAL;
294 
295 	ret = nla_parse_nested_deprecated(tb, TCA_ACT_BPF_MAX, nla,
296 					  act_bpf_policy, NULL);
297 	if (ret < 0)
298 		return ret;
299 
300 	if (!tb[TCA_ACT_BPF_PARMS])
301 		return -EINVAL;
302 
303 	parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
304 	index = parm->index;
305 	ret = tcf_idr_check_alloc(tn, &index, act, bind);
306 	if (!ret) {
307 		ret = tcf_idr_create(tn, index, est, act,
308 				     &act_bpf_ops, bind, true, 0);
309 		if (ret < 0) {
310 			tcf_idr_cleanup(tn, index);
311 			return ret;
312 		}
313 
314 		res = ACT_P_CREATED;
315 	} else if (ret > 0) {
316 		/* Don't override defaults. */
317 		if (bind)
318 			return 0;
319 
320 		if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
321 			tcf_idr_release(*act, bind);
322 			return -EEXIST;
323 		}
324 	} else {
325 		return ret;
326 	}
327 
328 	ret = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
329 	if (ret < 0)
330 		goto release_idr;
331 
332 	is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
333 	is_ebpf = tb[TCA_ACT_BPF_FD];
334 
335 	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
336 		ret = -EINVAL;
337 		goto put_chain;
338 	}
339 
340 	memset(&cfg, 0, sizeof(cfg));
341 
342 	ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
343 		       tcf_bpf_init_from_efd(tb, &cfg);
344 	if (ret < 0)
345 		goto put_chain;
346 
347 	prog = to_bpf(*act);
348 
349 	spin_lock_bh(&prog->tcf_lock);
350 	if (res != ACT_P_CREATED)
351 		tcf_bpf_prog_fill_cfg(prog, &old);
352 
353 	prog->bpf_ops = cfg.bpf_ops;
354 	prog->bpf_name = cfg.bpf_name;
355 
356 	if (cfg.bpf_num_ops)
357 		prog->bpf_num_ops = cfg.bpf_num_ops;
358 
359 	goto_ch = tcf_action_set_ctrlact(*act, parm->action, goto_ch);
360 	rcu_assign_pointer(prog->filter, cfg.filter);
361 	spin_unlock_bh(&prog->tcf_lock);
362 
363 	if (goto_ch)
364 		tcf_chain_put_by_act(goto_ch);
365 
366 	if (res != ACT_P_CREATED) {
367 		/* make sure the program being replaced is no longer executing */
368 		synchronize_rcu();
369 		tcf_bpf_cfg_cleanup(&old);
370 	}
371 
372 	return res;
373 
374 put_chain:
375 	if (goto_ch)
376 		tcf_chain_put_by_act(goto_ch);
377 
378 release_idr:
379 	tcf_idr_release(*act, bind);
380 	return ret;
381 }
382 
383 static void tcf_bpf_cleanup(struct tc_action *act)
384 {
385 	struct tcf_bpf_cfg tmp;
386 
387 	tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
388 	tcf_bpf_cfg_cleanup(&tmp);
389 }
390 
391 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
392 			  struct netlink_callback *cb, int type,
393 			  const struct tc_action_ops *ops,
394 			  struct netlink_ext_ack *extack)
395 {
396 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
397 
398 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
399 }
400 
401 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
402 {
403 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
404 
405 	return tcf_idr_search(tn, a, index);
406 }
407 
408 static struct tc_action_ops act_bpf_ops __read_mostly = {
409 	.kind		=	"bpf",
410 	.id		=	TCA_ID_BPF,
411 	.owner		=	THIS_MODULE,
412 	.act		=	tcf_bpf_act,
413 	.dump		=	tcf_bpf_dump,
414 	.cleanup	=	tcf_bpf_cleanup,
415 	.init		=	tcf_bpf_init,
416 	.walk		=	tcf_bpf_walker,
417 	.lookup		=	tcf_bpf_search,
418 	.size		=	sizeof(struct tcf_bpf),
419 };
420 
421 static __net_init int bpf_init_net(struct net *net)
422 {
423 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
424 
425 	return tc_action_net_init(net, tn, &act_bpf_ops);
426 }
427 
428 static void __net_exit bpf_exit_net(struct list_head *net_list)
429 {
430 	tc_action_net_exit(net_list, bpf_net_id);
431 }
432 
433 static struct pernet_operations bpf_net_ops = {
434 	.init = bpf_init_net,
435 	.exit_batch = bpf_exit_net,
436 	.id   = &bpf_net_id,
437 	.size = sizeof(struct tc_action_net),
438 };
439 
440 static int __init bpf_init_module(void)
441 {
442 	return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
443 }
444 
445 static void __exit bpf_cleanup_module(void)
446 {
447 	tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
448 }
449 
450 module_init(bpf_init_module);
451 module_exit(bpf_cleanup_module);
452 
453 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
454 MODULE_DESCRIPTION("TC BPF based action");
455 MODULE_LICENSE("GPL v2");
456