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
 24struct 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
 32static unsigned int bpf_net_id;
 33static struct tc_action_ops act_bpf_ops;
 34
 35static 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	rcu_read_lock();
 47	filter = rcu_dereference(prog->filter);
 48	if (at_ingress) {
 49		__skb_push(skb, skb->mac_len);
 50		bpf_compute_data_pointers(skb);
 51		filter_res = BPF_PROG_RUN(filter, skb);
 52		__skb_pull(skb, skb->mac_len);
 53	} else {
 54		bpf_compute_data_pointers(skb);
 55		filter_res = BPF_PROG_RUN(filter, skb);
 56	}
 57	if (skb_sk_is_prefetched(skb) && filter_res != TC_ACT_OK)
 58		skb_orphan(skb);
 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	struct nlattr *nla;
120
121	if (prog->bpf_name &&
122	    nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
123		return -EMSGSIZE;
124
125	if (nla_put_u32(skb, TCA_ACT_BPF_ID, prog->filter->aux->id))
126		return -EMSGSIZE;
127
128	nla = nla_reserve(skb, TCA_ACT_BPF_TAG, sizeof(prog->filter->tag));
129	if (nla == NULL)
130		return -EMSGSIZE;
131
132	memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
133
134	return 0;
135}
136
137static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
138			int bind, int ref)
139{
140	unsigned char *tp = skb_tail_pointer(skb);
141	struct tcf_bpf *prog = to_bpf(act);
142	struct tc_act_bpf opt = {
143		.index   = prog->tcf_index,
144		.refcnt  = refcount_read(&prog->tcf_refcnt) - ref,
145		.bindcnt = atomic_read(&prog->tcf_bindcnt) - bind,
146	};
147	struct tcf_t tm;
148	int ret;
149
150	spin_lock_bh(&prog->tcf_lock);
151	opt.action = prog->tcf_action;
152	if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
153		goto nla_put_failure;
154
155	if (tcf_bpf_is_ebpf(prog))
156		ret = tcf_bpf_dump_ebpf_info(prog, skb);
157	else
158		ret = tcf_bpf_dump_bpf_info(prog, skb);
159	if (ret)
160		goto nla_put_failure;
161
162	tcf_tm_dump(&tm, &prog->tcf_tm);
163	if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
164			  TCA_ACT_BPF_PAD))
165		goto nla_put_failure;
166
167	spin_unlock_bh(&prog->tcf_lock);
168	return skb->len;
169
170nla_put_failure:
171	spin_unlock_bh(&prog->tcf_lock);
172	nlmsg_trim(skb, tp);
173	return -1;
174}
175
176static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
177	[TCA_ACT_BPF_PARMS]	= { .len = sizeof(struct tc_act_bpf) },
178	[TCA_ACT_BPF_FD]	= { .type = NLA_U32 },
179	[TCA_ACT_BPF_NAME]	= { .type = NLA_NUL_STRING,
180				    .len = ACT_BPF_NAME_LEN },
181	[TCA_ACT_BPF_OPS_LEN]	= { .type = NLA_U16 },
182	[TCA_ACT_BPF_OPS]	= { .type = NLA_BINARY,
183				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
184};
185
186static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
187{
188	struct sock_filter *bpf_ops;
189	struct sock_fprog_kern fprog_tmp;
190	struct bpf_prog *fp;
191	u16 bpf_size, bpf_num_ops;
192	int ret;
193
194	bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
195	if (bpf_num_ops	> BPF_MAXINSNS || bpf_num_ops == 0)
196		return -EINVAL;
197
198	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
199	if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
200		return -EINVAL;
201
202	bpf_ops = kmemdup(nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size, GFP_KERNEL);
203	if (bpf_ops == NULL)
204		return -ENOMEM;
205
206	fprog_tmp.len = bpf_num_ops;
207	fprog_tmp.filter = bpf_ops;
208
209	ret = bpf_prog_create(&fp, &fprog_tmp);
210	if (ret < 0) {
211		kfree(bpf_ops);
212		return ret;
213	}
214
215	cfg->bpf_ops = bpf_ops;
216	cfg->bpf_num_ops = bpf_num_ops;
217	cfg->filter = fp;
218	cfg->is_ebpf = false;
219
220	return 0;
221}
222
223static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
224{
225	struct bpf_prog *fp;
226	char *name = NULL;
227	u32 bpf_fd;
228
229	bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
230
231	fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
232	if (IS_ERR(fp))
233		return PTR_ERR(fp);
234
235	if (tb[TCA_ACT_BPF_NAME]) {
236		name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL);
237		if (!name) {
238			bpf_prog_put(fp);
239			return -ENOMEM;
240		}
241	}
242
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	struct bpf_prog *filter = cfg->filter;
253
254	if (filter) {
255		if (cfg->is_ebpf)
256			bpf_prog_put(filter);
257		else
258			bpf_prog_destroy(filter);
259	}
260
261	kfree(cfg->bpf_ops);
262	kfree(cfg->bpf_name);
263}
264
265static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
266				  struct tcf_bpf_cfg *cfg)
267{
268	cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
269	/* updates to prog->filter are prevented, since it's called either
270	 * with tcf lock or during final cleanup in rcu callback
271	 */
272	cfg->filter = rcu_dereference_protected(prog->filter, 1);
273
274	cfg->bpf_ops = prog->bpf_ops;
275	cfg->bpf_name = prog->bpf_name;
276}
277
278static int tcf_bpf_init(struct net *net, struct nlattr *nla,
279			struct nlattr *est, struct tc_action **act,
280			int replace, int bind, bool rtnl_held,
281			struct tcf_proto *tp, u32 flags,
282			struct netlink_ext_ack *extack)
283{
284	struct tc_action_net *tn = net_generic(net, bpf_net_id);
285	struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
286	struct tcf_chain *goto_ch = NULL;
287	struct tcf_bpf_cfg cfg, old;
288	struct tc_act_bpf *parm;
289	struct tcf_bpf *prog;
290	bool is_bpf, is_ebpf;
291	int ret, res = 0;
292	u32 index;
293
294	if (!nla)
295		return -EINVAL;
296
297	ret = nla_parse_nested_deprecated(tb, TCA_ACT_BPF_MAX, nla,
298					  act_bpf_policy, NULL);
299	if (ret < 0)
300		return ret;
301
302	if (!tb[TCA_ACT_BPF_PARMS])
303		return -EINVAL;
304
305	parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
306	index = parm->index;
307	ret = tcf_idr_check_alloc(tn, &index, act, bind);
308	if (!ret) {
309		ret = tcf_idr_create(tn, index, est, act,
310				     &act_bpf_ops, bind, true, 0);
311		if (ret < 0) {
312			tcf_idr_cleanup(tn, index);
313			return ret;
314		}
315
316		res = ACT_P_CREATED;
317	} else if (ret > 0) {
318		/* Don't override defaults. */
319		if (bind)
320			return 0;
321
322		if (!replace) {
323			tcf_idr_release(*act, bind);
324			return -EEXIST;
325		}
326	} else {
327		return ret;
328	}
329
330	ret = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
331	if (ret < 0)
332		goto release_idr;
333
334	is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
335	is_ebpf = tb[TCA_ACT_BPF_FD];
336
337	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
338		ret = -EINVAL;
339		goto put_chain;
340	}
341
342	memset(&cfg, 0, sizeof(cfg));
343
344	ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
345		       tcf_bpf_init_from_efd(tb, &cfg);
346	if (ret < 0)
347		goto put_chain;
348
349	prog = to_bpf(*act);
350
351	spin_lock_bh(&prog->tcf_lock);
352	if (res != ACT_P_CREATED)
353		tcf_bpf_prog_fill_cfg(prog, &old);
354
355	prog->bpf_ops = cfg.bpf_ops;
356	prog->bpf_name = cfg.bpf_name;
357
358	if (cfg.bpf_num_ops)
359		prog->bpf_num_ops = cfg.bpf_num_ops;
360
361	goto_ch = tcf_action_set_ctrlact(*act, parm->action, goto_ch);
362	rcu_assign_pointer(prog->filter, cfg.filter);
363	spin_unlock_bh(&prog->tcf_lock);
364
365	if (goto_ch)
366		tcf_chain_put_by_act(goto_ch);
367
368	if (res != ACT_P_CREATED) {
369		/* make sure the program being replaced is no longer executing */
370		synchronize_rcu();
371		tcf_bpf_cfg_cleanup(&old);
372	}
373
374	return res;
375
376put_chain:
377	if (goto_ch)
378		tcf_chain_put_by_act(goto_ch);
379
380release_idr:
381	tcf_idr_release(*act, bind);
382	return ret;
383}
384
385static void tcf_bpf_cleanup(struct tc_action *act)
386{
387	struct tcf_bpf_cfg tmp;
388
389	tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
390	tcf_bpf_cfg_cleanup(&tmp);
391}
392
393static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
394			  struct netlink_callback *cb, int type,
395			  const struct tc_action_ops *ops,
396			  struct netlink_ext_ack *extack)
397{
398	struct tc_action_net *tn = net_generic(net, bpf_net_id);
399
400	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
401}
402
403static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
404{
405	struct tc_action_net *tn = net_generic(net, bpf_net_id);
406
407	return tcf_idr_search(tn, a, index);
408}
409
410static struct tc_action_ops act_bpf_ops __read_mostly = {
411	.kind		=	"bpf",
412	.id		=	TCA_ID_BPF,
413	.owner		=	THIS_MODULE,
414	.act		=	tcf_bpf_act,
415	.dump		=	tcf_bpf_dump,
416	.cleanup	=	tcf_bpf_cleanup,
417	.init		=	tcf_bpf_init,
418	.walk		=	tcf_bpf_walker,
419	.lookup		=	tcf_bpf_search,
420	.size		=	sizeof(struct tcf_bpf),
421};
422
423static __net_init int bpf_init_net(struct net *net)
424{
425	struct tc_action_net *tn = net_generic(net, bpf_net_id);
426
427	return tc_action_net_init(net, tn, &act_bpf_ops);
428}
429
430static void __net_exit bpf_exit_net(struct list_head *net_list)
431{
432	tc_action_net_exit(net_list, bpf_net_id);
433}
434
435static struct pernet_operations bpf_net_ops = {
436	.init = bpf_init_net,
437	.exit_batch = bpf_exit_net,
438	.id   = &bpf_net_id,
439	.size = sizeof(struct tc_action_net),
440};
441
442static int __init bpf_init_module(void)
443{
444	return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
445}
446
447static void __exit bpf_cleanup_module(void)
448{
449	tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
450}
451
452module_init(bpf_init_module);
453module_exit(bpf_cleanup_module);
454
455MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
456MODULE_DESCRIPTION("TC BPF based action");
457MODULE_LICENSE("GPL v2");