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	u16 bpf_num_ops;
 32	bool is_ebpf;
 33};
 34
 35static unsigned int bpf_net_id;
 36static struct tc_action_ops act_bpf_ops;
 37
 38static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
 39		   struct tcf_result *res)
 40{
 41	bool at_ingress = skb_at_tc_ingress(skb);
 42	struct tcf_bpf *prog = to_bpf(act);
 43	struct bpf_prog *filter;
 44	int action, filter_res;
 45
 46	tcf_lastuse_update(&prog->tcf_tm);
 47	bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
 48
 49	rcu_read_lock();
 50	filter = rcu_dereference(prog->filter);
 51	if (at_ingress) {
 52		__skb_push(skb, skb->mac_len);
 53		bpf_compute_data_end(skb);
 54		filter_res = BPF_PROG_RUN(filter, skb);
 55		__skb_pull(skb, skb->mac_len);
 56	} else {
 57		bpf_compute_data_end(skb);
 58		filter_res = BPF_PROG_RUN(filter, skb);
 59	}
 60	rcu_read_unlock();
 61
 62	/* A BPF program may overwrite the default action opcode.
 63	 * Similarly as in cls_bpf, if filter_res == -1 we use the
 64	 * default action specified from tc.
 65	 *
 66	 * In case a different well-known TC_ACT opcode has been
 67	 * returned, it will overwrite the default one.
 68	 *
 69	 * For everything else that is unkown, TC_ACT_UNSPEC is
 70	 * returned.
 71	 */
 72	switch (filter_res) {
 73	case TC_ACT_PIPE:
 74	case TC_ACT_RECLASSIFY:
 75	case TC_ACT_OK:
 76	case TC_ACT_REDIRECT:
 77		action = filter_res;
 78		break;
 79	case TC_ACT_SHOT:
 80		action = filter_res;
 81		qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
 82		break;
 83	case TC_ACT_UNSPEC:
 84		action = prog->tcf_action;
 85		break;
 86	default:
 87		action = TC_ACT_UNSPEC;
 88		break;
 89	}
 90
 91	return action;
 92}
 93
 94static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
 95{
 96	return !prog->bpf_ops;
 97}
 98
 99static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
100				 struct sk_buff *skb)
101{
102	struct nlattr *nla;
103
104	if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
105		return -EMSGSIZE;
106
107	nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
108			  sizeof(struct sock_filter));
109	if (nla == NULL)
110		return -EMSGSIZE;
111
112	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
113
114	return 0;
115}
116
117static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
118				  struct sk_buff *skb)
119{
120	struct nlattr *nla;
121
122	if (prog->bpf_name &&
123	    nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
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
135static 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  = prog->tcf_refcnt - ref,
143		.bindcnt = prog->tcf_bindcnt - bind,
144		.action  = prog->tcf_action,
145	};
146	struct tcf_t tm;
147	int ret;
148
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	return skb->len;
165
166nla_put_failure:
167	nlmsg_trim(skb, tp);
168	return -1;
169}
170
171static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
172	[TCA_ACT_BPF_PARMS]	= { .len = sizeof(struct tc_act_bpf) },
173	[TCA_ACT_BPF_FD]	= { .type = NLA_U32 },
174	[TCA_ACT_BPF_NAME]	= { .type = NLA_NUL_STRING,
175				    .len = ACT_BPF_NAME_LEN },
176	[TCA_ACT_BPF_OPS_LEN]	= { .type = NLA_U16 },
177	[TCA_ACT_BPF_OPS]	= { .type = NLA_BINARY,
178				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
179};
180
181static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
182{
183	struct sock_filter *bpf_ops;
184	struct sock_fprog_kern fprog_tmp;
185	struct bpf_prog *fp;
186	u16 bpf_size, bpf_num_ops;
187	int ret;
188
189	bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
190	if (bpf_num_ops	> BPF_MAXINSNS || bpf_num_ops == 0)
191		return -EINVAL;
192
193	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
194	if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
195		return -EINVAL;
196
197	bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
198	if (bpf_ops == NULL)
199		return -ENOMEM;
200
201	memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
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
220static 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
247static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
248{
249	if (cfg->is_ebpf)
250		bpf_prog_put(cfg->filter);
251	else
252		bpf_prog_destroy(cfg->filter);
253
254	kfree(cfg->bpf_ops);
255	kfree(cfg->bpf_name);
256}
257
258static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
259				  struct tcf_bpf_cfg *cfg)
260{
261	cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
262	/* updates to prog->filter are prevented, since it's called either
263	 * with rtnl lock or during final cleanup in rcu callback
264	 */
265	cfg->filter = rcu_dereference_protected(prog->filter, 1);
266
267	cfg->bpf_ops = prog->bpf_ops;
268	cfg->bpf_name = prog->bpf_name;
269}
270
271static int tcf_bpf_init(struct net *net, struct nlattr *nla,
272			struct nlattr *est, struct tc_action **act,
273			int replace, int bind)
274{
275	struct tc_action_net *tn = net_generic(net, bpf_net_id);
276	struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
277	struct tcf_bpf_cfg cfg, old;
278	struct tc_act_bpf *parm;
279	struct tcf_bpf *prog;
280	bool is_bpf, is_ebpf;
281	int ret, res = 0;
282
283	if (!nla)
284		return -EINVAL;
285
286	ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
287	if (ret < 0)
288		return ret;
289
290	if (!tb[TCA_ACT_BPF_PARMS])
291		return -EINVAL;
292
293	parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
294
295	if (!tcf_hash_check(tn, parm->index, act, bind)) {
296		ret = tcf_hash_create(tn, parm->index, est, act,
297				      &act_bpf_ops, bind, true);
298		if (ret < 0)
299			return ret;
300
301		res = ACT_P_CREATED;
302	} else {
303		/* Don't override defaults. */
304		if (bind)
305			return 0;
306
307		tcf_hash_release(*act, bind);
308		if (!replace)
309			return -EEXIST;
310	}
311
312	is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
313	is_ebpf = tb[TCA_ACT_BPF_FD];
314
315	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
316		ret = -EINVAL;
317		goto out;
318	}
319
320	memset(&cfg, 0, sizeof(cfg));
321
322	ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
323		       tcf_bpf_init_from_efd(tb, &cfg);
324	if (ret < 0)
325		goto out;
326
327	prog = to_bpf(*act);
328	ASSERT_RTNL();
329
330	if (res != ACT_P_CREATED)
331		tcf_bpf_prog_fill_cfg(prog, &old);
332
333	prog->bpf_ops = cfg.bpf_ops;
334	prog->bpf_name = cfg.bpf_name;
335
336	if (cfg.bpf_num_ops)
337		prog->bpf_num_ops = cfg.bpf_num_ops;
338
339	prog->tcf_action = parm->action;
340	rcu_assign_pointer(prog->filter, cfg.filter);
341
342	if (res == ACT_P_CREATED) {
343		tcf_hash_insert(tn, *act);
344	} else {
345		/* make sure the program being replaced is no longer executing */
346		synchronize_rcu();
347		tcf_bpf_cfg_cleanup(&old);
348	}
349
350	return res;
351out:
352	if (res == ACT_P_CREATED)
353		tcf_hash_cleanup(*act, est);
354
355	return ret;
356}
357
358static void tcf_bpf_cleanup(struct tc_action *act, int bind)
359{
360	struct tcf_bpf_cfg tmp;
361
362	tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
363	tcf_bpf_cfg_cleanup(&tmp);
364}
365
366static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
367			  struct netlink_callback *cb, int type,
368			  const struct tc_action_ops *ops)
369{
370	struct tc_action_net *tn = net_generic(net, bpf_net_id);
371
372	return tcf_generic_walker(tn, skb, cb, type, ops);
373}
374
375static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
376{
377	struct tc_action_net *tn = net_generic(net, bpf_net_id);
378
379	return tcf_hash_search(tn, a, index);
380}
381
382static struct tc_action_ops act_bpf_ops __read_mostly = {
383	.kind		=	"bpf",
384	.type		=	TCA_ACT_BPF,
385	.owner		=	THIS_MODULE,
386	.act		=	tcf_bpf,
387	.dump		=	tcf_bpf_dump,
388	.cleanup	=	tcf_bpf_cleanup,
389	.init		=	tcf_bpf_init,
390	.walk		=	tcf_bpf_walker,
391	.lookup		=	tcf_bpf_search,
392	.size		=	sizeof(struct tcf_bpf),
393};
394
395static __net_init int bpf_init_net(struct net *net)
396{
397	struct tc_action_net *tn = net_generic(net, bpf_net_id);
398
399	return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK);
400}
401
402static void __net_exit bpf_exit_net(struct net *net)
403{
404	struct tc_action_net *tn = net_generic(net, bpf_net_id);
405
406	tc_action_net_exit(tn);
407}
408
409static struct pernet_operations bpf_net_ops = {
410	.init = bpf_init_net,
411	.exit = bpf_exit_net,
412	.id   = &bpf_net_id,
413	.size = sizeof(struct tc_action_net),
414};
415
416static int __init bpf_init_module(void)
417{
418	return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
419}
420
421static void __exit bpf_cleanup_module(void)
422{
423	tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
424}
425
426module_init(bpf_init_module);
427module_exit(bpf_cleanup_module);
428
429MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
430MODULE_DESCRIPTION("TC BPF based action");
431MODULE_LICENSE("GPL v2");