xref: /openbmc/linux/net/sched/act_bpf.c (revision 78beef62)
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/pkt_sched.h>
16 #include <net/pkt_cls.h>
17 
18 #include <linux/tc_act/tc_bpf.h>
19 #include <net/tc_act/tc_bpf.h>
20 
21 #define ACT_BPF_NAME_LEN	256
22 
23 struct tcf_bpf_cfg {
24 	struct bpf_prog *filter;
25 	struct sock_filter *bpf_ops;
26 	const char *bpf_name;
27 	u16 bpf_num_ops;
28 	bool is_ebpf;
29 };
30 
31 static unsigned int bpf_net_id;
32 static struct tc_action_ops act_bpf_ops;
33 
34 static int tcf_bpf_act(struct sk_buff *skb, const struct tc_action *act,
35 		       struct tcf_result *res)
36 {
37 	bool at_ingress = skb_at_tc_ingress(skb);
38 	struct tcf_bpf *prog = to_bpf(act);
39 	struct bpf_prog *filter;
40 	int action, filter_res;
41 
42 	tcf_lastuse_update(&prog->tcf_tm);
43 	bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
44 
45 	rcu_read_lock();
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 	rcu_read_unlock();
57 
58 	/* A BPF program may overwrite the default action opcode.
59 	 * Similarly as in cls_bpf, if filter_res == -1 we use the
60 	 * default action specified from tc.
61 	 *
62 	 * In case a different well-known TC_ACT opcode has been
63 	 * returned, it will overwrite the default one.
64 	 *
65 	 * For everything else that is unkown, TC_ACT_UNSPEC is
66 	 * returned.
67 	 */
68 	switch (filter_res) {
69 	case TC_ACT_PIPE:
70 	case TC_ACT_RECLASSIFY:
71 	case TC_ACT_OK:
72 	case TC_ACT_REDIRECT:
73 		action = filter_res;
74 		break;
75 	case TC_ACT_SHOT:
76 		action = filter_res;
77 		qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
78 		break;
79 	case TC_ACT_UNSPEC:
80 		action = prog->tcf_action;
81 		break;
82 	default:
83 		action = TC_ACT_UNSPEC;
84 		break;
85 	}
86 
87 	return action;
88 }
89 
90 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
91 {
92 	return !prog->bpf_ops;
93 }
94 
95 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
96 				 struct sk_buff *skb)
97 {
98 	struct nlattr *nla;
99 
100 	if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
101 		return -EMSGSIZE;
102 
103 	nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
104 			  sizeof(struct sock_filter));
105 	if (nla == NULL)
106 		return -EMSGSIZE;
107 
108 	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
109 
110 	return 0;
111 }
112 
113 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
114 				  struct sk_buff *skb)
115 {
116 	struct nlattr *nla;
117 
118 	if (prog->bpf_name &&
119 	    nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
120 		return -EMSGSIZE;
121 
122 	if (nla_put_u32(skb, TCA_ACT_BPF_ID, prog->filter->aux->id))
123 		return -EMSGSIZE;
124 
125 	nla = nla_reserve(skb, TCA_ACT_BPF_TAG, sizeof(prog->filter->tag));
126 	if (nla == NULL)
127 		return -EMSGSIZE;
128 
129 	memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
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  = refcount_read(&prog->tcf_refcnt) - ref,
142 		.bindcnt = atomic_read(&prog->tcf_bindcnt) - bind,
143 	};
144 	struct tcf_t tm;
145 	int ret;
146 
147 	spin_lock_bh(&prog->tcf_lock);
148 	opt.action = prog->tcf_action;
149 	if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
150 		goto nla_put_failure;
151 
152 	if (tcf_bpf_is_ebpf(prog))
153 		ret = tcf_bpf_dump_ebpf_info(prog, skb);
154 	else
155 		ret = tcf_bpf_dump_bpf_info(prog, skb);
156 	if (ret)
157 		goto nla_put_failure;
158 
159 	tcf_tm_dump(&tm, &prog->tcf_tm);
160 	if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
161 			  TCA_ACT_BPF_PAD))
162 		goto nla_put_failure;
163 
164 	spin_unlock_bh(&prog->tcf_lock);
165 	return skb->len;
166 
167 nla_put_failure:
168 	spin_unlock_bh(&prog->tcf_lock);
169 	nlmsg_trim(skb, tp);
170 	return -1;
171 }
172 
173 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
174 	[TCA_ACT_BPF_PARMS]	= { .len = sizeof(struct tc_act_bpf) },
175 	[TCA_ACT_BPF_FD]	= { .type = NLA_U32 },
176 	[TCA_ACT_BPF_NAME]	= { .type = NLA_NUL_STRING,
177 				    .len = ACT_BPF_NAME_LEN },
178 	[TCA_ACT_BPF_OPS_LEN]	= { .type = NLA_U16 },
179 	[TCA_ACT_BPF_OPS]	= { .type = NLA_BINARY,
180 				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
181 };
182 
183 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
184 {
185 	struct sock_filter *bpf_ops;
186 	struct sock_fprog_kern fprog_tmp;
187 	struct bpf_prog *fp;
188 	u16 bpf_size, bpf_num_ops;
189 	int ret;
190 
191 	bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
192 	if (bpf_num_ops	> BPF_MAXINSNS || bpf_num_ops == 0)
193 		return -EINVAL;
194 
195 	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
196 	if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
197 		return -EINVAL;
198 
199 	bpf_ops = kmemdup(nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size, GFP_KERNEL);
200 	if (bpf_ops == NULL)
201 		return -ENOMEM;
202 
203 	fprog_tmp.len = bpf_num_ops;
204 	fprog_tmp.filter = bpf_ops;
205 
206 	ret = bpf_prog_create(&fp, &fprog_tmp);
207 	if (ret < 0) {
208 		kfree(bpf_ops);
209 		return ret;
210 	}
211 
212 	cfg->bpf_ops = bpf_ops;
213 	cfg->bpf_num_ops = bpf_num_ops;
214 	cfg->filter = fp;
215 	cfg->is_ebpf = false;
216 
217 	return 0;
218 }
219 
220 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
221 {
222 	struct bpf_prog *fp;
223 	char *name = NULL;
224 	u32 bpf_fd;
225 
226 	bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
227 
228 	fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
229 	if (IS_ERR(fp))
230 		return PTR_ERR(fp);
231 
232 	if (tb[TCA_ACT_BPF_NAME]) {
233 		name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL);
234 		if (!name) {
235 			bpf_prog_put(fp);
236 			return -ENOMEM;
237 		}
238 	}
239 
240 	cfg->bpf_name = name;
241 	cfg->filter = fp;
242 	cfg->is_ebpf = true;
243 
244 	return 0;
245 }
246 
247 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
248 {
249 	struct bpf_prog *filter = cfg->filter;
250 
251 	if (filter) {
252 		if (cfg->is_ebpf)
253 			bpf_prog_put(filter);
254 		else
255 			bpf_prog_destroy(filter);
256 	}
257 
258 	kfree(cfg->bpf_ops);
259 	kfree(cfg->bpf_name);
260 }
261 
262 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
263 				  struct tcf_bpf_cfg *cfg)
264 {
265 	cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
266 	/* updates to prog->filter are prevented, since it's called either
267 	 * with tcf lock or during final cleanup in rcu callback
268 	 */
269 	cfg->filter = rcu_dereference_protected(prog->filter, 1);
270 
271 	cfg->bpf_ops = prog->bpf_ops;
272 	cfg->bpf_name = prog->bpf_name;
273 }
274 
275 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
276 			struct nlattr *est, struct tc_action **act,
277 			int replace, int bind, bool rtnl_held,
278 			struct tcf_proto *tp, struct netlink_ext_ack *extack)
279 {
280 	struct tc_action_net *tn = net_generic(net, bpf_net_id);
281 	struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
282 	struct tcf_chain *goto_ch = NULL;
283 	struct tcf_bpf_cfg cfg, old;
284 	struct tc_act_bpf *parm;
285 	struct tcf_bpf *prog;
286 	bool is_bpf, is_ebpf;
287 	int ret, res = 0;
288 	u32 index;
289 
290 	if (!nla)
291 		return -EINVAL;
292 
293 	ret = nla_parse_nested_deprecated(tb, TCA_ACT_BPF_MAX, nla,
294 					  act_bpf_policy, NULL);
295 	if (ret < 0)
296 		return ret;
297 
298 	if (!tb[TCA_ACT_BPF_PARMS])
299 		return -EINVAL;
300 
301 	parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
302 	index = parm->index;
303 	ret = tcf_idr_check_alloc(tn, &index, act, bind);
304 	if (!ret) {
305 		ret = tcf_idr_create(tn, index, est, act,
306 				     &act_bpf_ops, bind, true);
307 		if (ret < 0) {
308 			tcf_idr_cleanup(tn, index);
309 			return ret;
310 		}
311 
312 		res = ACT_P_CREATED;
313 	} else if (ret > 0) {
314 		/* Don't override defaults. */
315 		if (bind)
316 			return 0;
317 
318 		if (!replace) {
319 			tcf_idr_release(*act, bind);
320 			return -EEXIST;
321 		}
322 	} else {
323 		return ret;
324 	}
325 
326 	ret = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
327 	if (ret < 0)
328 		goto release_idr;
329 
330 	is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
331 	is_ebpf = tb[TCA_ACT_BPF_FD];
332 
333 	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
334 		ret = -EINVAL;
335 		goto put_chain;
336 	}
337 
338 	memset(&cfg, 0, sizeof(cfg));
339 
340 	ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
341 		       tcf_bpf_init_from_efd(tb, &cfg);
342 	if (ret < 0)
343 		goto put_chain;
344 
345 	prog = to_bpf(*act);
346 
347 	spin_lock_bh(&prog->tcf_lock);
348 	if (res != ACT_P_CREATED)
349 		tcf_bpf_prog_fill_cfg(prog, &old);
350 
351 	prog->bpf_ops = cfg.bpf_ops;
352 	prog->bpf_name = cfg.bpf_name;
353 
354 	if (cfg.bpf_num_ops)
355 		prog->bpf_num_ops = cfg.bpf_num_ops;
356 
357 	goto_ch = tcf_action_set_ctrlact(*act, parm->action, goto_ch);
358 	rcu_assign_pointer(prog->filter, cfg.filter);
359 	spin_unlock_bh(&prog->tcf_lock);
360 
361 	if (goto_ch)
362 		tcf_chain_put_by_act(goto_ch);
363 
364 	if (res == ACT_P_CREATED) {
365 		tcf_idr_insert(tn, *act);
366 	} else {
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