xref: /openbmc/linux/net/sched/cls_bpf.c (revision d4fd6347)
1 /*
2  * Berkeley Packet Filter based traffic classifier
3  *
4  * Might be used to classify traffic through flexible, user-defined and
5  * possibly JIT-ed BPF filters for traffic control as an alternative to
6  * ematches.
7  *
8  * (C) 2013 Daniel Borkmann <dborkman@redhat.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License version 2 as
12  * published by the Free Software Foundation.
13  */
14 
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <linux/skbuff.h>
18 #include <linux/filter.h>
19 #include <linux/bpf.h>
20 #include <linux/idr.h>
21 
22 #include <net/rtnetlink.h>
23 #include <net/pkt_cls.h>
24 #include <net/sock.h>
25 
26 MODULE_LICENSE("GPL");
27 MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
28 MODULE_DESCRIPTION("TC BPF based classifier");
29 
30 #define CLS_BPF_NAME_LEN	256
31 #define CLS_BPF_SUPPORTED_GEN_FLAGS		\
32 	(TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)
33 
34 struct cls_bpf_head {
35 	struct list_head plist;
36 	struct idr handle_idr;
37 	struct rcu_head rcu;
38 };
39 
40 struct cls_bpf_prog {
41 	struct bpf_prog *filter;
42 	struct list_head link;
43 	struct tcf_result res;
44 	bool exts_integrated;
45 	u32 gen_flags;
46 	unsigned int in_hw_count;
47 	struct tcf_exts exts;
48 	u32 handle;
49 	u16 bpf_num_ops;
50 	struct sock_filter *bpf_ops;
51 	const char *bpf_name;
52 	struct tcf_proto *tp;
53 	struct rcu_work rwork;
54 };
55 
56 static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
57 	[TCA_BPF_CLASSID]	= { .type = NLA_U32 },
58 	[TCA_BPF_FLAGS]		= { .type = NLA_U32 },
59 	[TCA_BPF_FLAGS_GEN]	= { .type = NLA_U32 },
60 	[TCA_BPF_FD]		= { .type = NLA_U32 },
61 	[TCA_BPF_NAME]		= { .type = NLA_NUL_STRING,
62 				    .len = CLS_BPF_NAME_LEN },
63 	[TCA_BPF_OPS_LEN]	= { .type = NLA_U16 },
64 	[TCA_BPF_OPS]		= { .type = NLA_BINARY,
65 				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
66 };
67 
68 static int cls_bpf_exec_opcode(int code)
69 {
70 	switch (code) {
71 	case TC_ACT_OK:
72 	case TC_ACT_SHOT:
73 	case TC_ACT_STOLEN:
74 	case TC_ACT_TRAP:
75 	case TC_ACT_REDIRECT:
76 	case TC_ACT_UNSPEC:
77 		return code;
78 	default:
79 		return TC_ACT_UNSPEC;
80 	}
81 }
82 
83 static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
84 			    struct tcf_result *res)
85 {
86 	struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
87 	bool at_ingress = skb_at_tc_ingress(skb);
88 	struct cls_bpf_prog *prog;
89 	int ret = -1;
90 
91 	/* Needed here for accessing maps. */
92 	rcu_read_lock();
93 	list_for_each_entry_rcu(prog, &head->plist, link) {
94 		int filter_res;
95 
96 		qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
97 
98 		if (tc_skip_sw(prog->gen_flags)) {
99 			filter_res = prog->exts_integrated ? TC_ACT_UNSPEC : 0;
100 		} else if (at_ingress) {
101 			/* It is safe to push/pull even if skb_shared() */
102 			__skb_push(skb, skb->mac_len);
103 			bpf_compute_data_pointers(skb);
104 			filter_res = BPF_PROG_RUN(prog->filter, skb);
105 			__skb_pull(skb, skb->mac_len);
106 		} else {
107 			bpf_compute_data_pointers(skb);
108 			filter_res = BPF_PROG_RUN(prog->filter, skb);
109 		}
110 
111 		if (prog->exts_integrated) {
112 			res->class   = 0;
113 			res->classid = TC_H_MAJ(prog->res.classid) |
114 				       qdisc_skb_cb(skb)->tc_classid;
115 
116 			ret = cls_bpf_exec_opcode(filter_res);
117 			if (ret == TC_ACT_UNSPEC)
118 				continue;
119 			break;
120 		}
121 
122 		if (filter_res == 0)
123 			continue;
124 		if (filter_res != -1) {
125 			res->class   = 0;
126 			res->classid = filter_res;
127 		} else {
128 			*res = prog->res;
129 		}
130 
131 		ret = tcf_exts_exec(skb, &prog->exts, res);
132 		if (ret < 0)
133 			continue;
134 
135 		break;
136 	}
137 	rcu_read_unlock();
138 
139 	return ret;
140 }
141 
142 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
143 {
144 	return !prog->bpf_ops;
145 }
146 
147 static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
148 			       struct cls_bpf_prog *oldprog,
149 			       struct netlink_ext_ack *extack)
150 {
151 	struct tcf_block *block = tp->chain->block;
152 	struct tc_cls_bpf_offload cls_bpf = {};
153 	struct cls_bpf_prog *obj;
154 	bool skip_sw;
155 	int err;
156 
157 	skip_sw = prog && tc_skip_sw(prog->gen_flags);
158 	obj = prog ?: oldprog;
159 
160 	tc_cls_common_offload_init(&cls_bpf.common, tp, obj->gen_flags, extack);
161 	cls_bpf.command = TC_CLSBPF_OFFLOAD;
162 	cls_bpf.exts = &obj->exts;
163 	cls_bpf.prog = prog ? prog->filter : NULL;
164 	cls_bpf.oldprog = oldprog ? oldprog->filter : NULL;
165 	cls_bpf.name = obj->bpf_name;
166 	cls_bpf.exts_integrated = obj->exts_integrated;
167 
168 	if (oldprog)
169 		tcf_block_offload_dec(block, &oldprog->gen_flags);
170 
171 	err = tc_setup_cb_call(block, TC_SETUP_CLSBPF, &cls_bpf, skip_sw);
172 	if (prog) {
173 		if (err < 0) {
174 			cls_bpf_offload_cmd(tp, oldprog, prog, extack);
175 			return err;
176 		} else if (err > 0) {
177 			prog->in_hw_count = err;
178 			tcf_block_offload_inc(block, &prog->gen_flags);
179 		}
180 	}
181 
182 	if (prog && skip_sw && !(prog->gen_flags & TCA_CLS_FLAGS_IN_HW))
183 		return -EINVAL;
184 
185 	return 0;
186 }
187 
188 static u32 cls_bpf_flags(u32 flags)
189 {
190 	return flags & CLS_BPF_SUPPORTED_GEN_FLAGS;
191 }
192 
193 static int cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
194 			   struct cls_bpf_prog *oldprog,
195 			   struct netlink_ext_ack *extack)
196 {
197 	if (prog && oldprog &&
198 	    cls_bpf_flags(prog->gen_flags) !=
199 	    cls_bpf_flags(oldprog->gen_flags))
200 		return -EINVAL;
201 
202 	if (prog && tc_skip_hw(prog->gen_flags))
203 		prog = NULL;
204 	if (oldprog && tc_skip_hw(oldprog->gen_flags))
205 		oldprog = NULL;
206 	if (!prog && !oldprog)
207 		return 0;
208 
209 	return cls_bpf_offload_cmd(tp, prog, oldprog, extack);
210 }
211 
212 static void cls_bpf_stop_offload(struct tcf_proto *tp,
213 				 struct cls_bpf_prog *prog,
214 				 struct netlink_ext_ack *extack)
215 {
216 	int err;
217 
218 	err = cls_bpf_offload_cmd(tp, NULL, prog, extack);
219 	if (err)
220 		pr_err("Stopping hardware offload failed: %d\n", err);
221 }
222 
223 static void cls_bpf_offload_update_stats(struct tcf_proto *tp,
224 					 struct cls_bpf_prog *prog)
225 {
226 	struct tcf_block *block = tp->chain->block;
227 	struct tc_cls_bpf_offload cls_bpf = {};
228 
229 	tc_cls_common_offload_init(&cls_bpf.common, tp, prog->gen_flags, NULL);
230 	cls_bpf.command = TC_CLSBPF_STATS;
231 	cls_bpf.exts = &prog->exts;
232 	cls_bpf.prog = prog->filter;
233 	cls_bpf.name = prog->bpf_name;
234 	cls_bpf.exts_integrated = prog->exts_integrated;
235 
236 	tc_setup_cb_call(block, TC_SETUP_CLSBPF, &cls_bpf, false);
237 }
238 
239 static int cls_bpf_init(struct tcf_proto *tp)
240 {
241 	struct cls_bpf_head *head;
242 
243 	head = kzalloc(sizeof(*head), GFP_KERNEL);
244 	if (head == NULL)
245 		return -ENOBUFS;
246 
247 	INIT_LIST_HEAD_RCU(&head->plist);
248 	idr_init(&head->handle_idr);
249 	rcu_assign_pointer(tp->root, head);
250 
251 	return 0;
252 }
253 
254 static void cls_bpf_free_parms(struct cls_bpf_prog *prog)
255 {
256 	if (cls_bpf_is_ebpf(prog))
257 		bpf_prog_put(prog->filter);
258 	else
259 		bpf_prog_destroy(prog->filter);
260 
261 	kfree(prog->bpf_name);
262 	kfree(prog->bpf_ops);
263 }
264 
265 static void __cls_bpf_delete_prog(struct cls_bpf_prog *prog)
266 {
267 	tcf_exts_destroy(&prog->exts);
268 	tcf_exts_put_net(&prog->exts);
269 
270 	cls_bpf_free_parms(prog);
271 	kfree(prog);
272 }
273 
274 static void cls_bpf_delete_prog_work(struct work_struct *work)
275 {
276 	struct cls_bpf_prog *prog = container_of(to_rcu_work(work),
277 						 struct cls_bpf_prog,
278 						 rwork);
279 	rtnl_lock();
280 	__cls_bpf_delete_prog(prog);
281 	rtnl_unlock();
282 }
283 
284 static void __cls_bpf_delete(struct tcf_proto *tp, struct cls_bpf_prog *prog,
285 			     struct netlink_ext_ack *extack)
286 {
287 	struct cls_bpf_head *head = rtnl_dereference(tp->root);
288 
289 	idr_remove(&head->handle_idr, prog->handle);
290 	cls_bpf_stop_offload(tp, prog, extack);
291 	list_del_rcu(&prog->link);
292 	tcf_unbind_filter(tp, &prog->res);
293 	if (tcf_exts_get_net(&prog->exts))
294 		tcf_queue_work(&prog->rwork, cls_bpf_delete_prog_work);
295 	else
296 		__cls_bpf_delete_prog(prog);
297 }
298 
299 static int cls_bpf_delete(struct tcf_proto *tp, void *arg, bool *last,
300 			  bool rtnl_held, struct netlink_ext_ack *extack)
301 {
302 	struct cls_bpf_head *head = rtnl_dereference(tp->root);
303 
304 	__cls_bpf_delete(tp, arg, extack);
305 	*last = list_empty(&head->plist);
306 	return 0;
307 }
308 
309 static void cls_bpf_destroy(struct tcf_proto *tp, bool rtnl_held,
310 			    struct netlink_ext_ack *extack)
311 {
312 	struct cls_bpf_head *head = rtnl_dereference(tp->root);
313 	struct cls_bpf_prog *prog, *tmp;
314 
315 	list_for_each_entry_safe(prog, tmp, &head->plist, link)
316 		__cls_bpf_delete(tp, prog, extack);
317 
318 	idr_destroy(&head->handle_idr);
319 	kfree_rcu(head, rcu);
320 }
321 
322 static void *cls_bpf_get(struct tcf_proto *tp, u32 handle)
323 {
324 	struct cls_bpf_head *head = rtnl_dereference(tp->root);
325 	struct cls_bpf_prog *prog;
326 
327 	list_for_each_entry(prog, &head->plist, link) {
328 		if (prog->handle == handle)
329 			return prog;
330 	}
331 
332 	return NULL;
333 }
334 
335 static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
336 {
337 	struct sock_filter *bpf_ops;
338 	struct sock_fprog_kern fprog_tmp;
339 	struct bpf_prog *fp;
340 	u16 bpf_size, bpf_num_ops;
341 	int ret;
342 
343 	bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
344 	if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
345 		return -EINVAL;
346 
347 	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
348 	if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
349 		return -EINVAL;
350 
351 	bpf_ops = kmemdup(nla_data(tb[TCA_BPF_OPS]), bpf_size, GFP_KERNEL);
352 	if (bpf_ops == NULL)
353 		return -ENOMEM;
354 
355 	fprog_tmp.len = bpf_num_ops;
356 	fprog_tmp.filter = bpf_ops;
357 
358 	ret = bpf_prog_create(&fp, &fprog_tmp);
359 	if (ret < 0) {
360 		kfree(bpf_ops);
361 		return ret;
362 	}
363 
364 	prog->bpf_ops = bpf_ops;
365 	prog->bpf_num_ops = bpf_num_ops;
366 	prog->bpf_name = NULL;
367 	prog->filter = fp;
368 
369 	return 0;
370 }
371 
372 static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
373 				 u32 gen_flags, const struct tcf_proto *tp)
374 {
375 	struct bpf_prog *fp;
376 	char *name = NULL;
377 	bool skip_sw;
378 	u32 bpf_fd;
379 
380 	bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
381 	skip_sw = gen_flags & TCA_CLS_FLAGS_SKIP_SW;
382 
383 	fp = bpf_prog_get_type_dev(bpf_fd, BPF_PROG_TYPE_SCHED_CLS, skip_sw);
384 	if (IS_ERR(fp))
385 		return PTR_ERR(fp);
386 
387 	if (tb[TCA_BPF_NAME]) {
388 		name = nla_memdup(tb[TCA_BPF_NAME], GFP_KERNEL);
389 		if (!name) {
390 			bpf_prog_put(fp);
391 			return -ENOMEM;
392 		}
393 	}
394 
395 	prog->bpf_ops = NULL;
396 	prog->bpf_name = name;
397 	prog->filter = fp;
398 
399 	if (fp->dst_needed)
400 		tcf_block_netif_keep_dst(tp->chain->block);
401 
402 	return 0;
403 }
404 
405 static int cls_bpf_set_parms(struct net *net, struct tcf_proto *tp,
406 			     struct cls_bpf_prog *prog, unsigned long base,
407 			     struct nlattr **tb, struct nlattr *est, bool ovr,
408 			     struct netlink_ext_ack *extack)
409 {
410 	bool is_bpf, is_ebpf, have_exts = false;
411 	u32 gen_flags = 0;
412 	int ret;
413 
414 	is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
415 	is_ebpf = tb[TCA_BPF_FD];
416 	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf))
417 		return -EINVAL;
418 
419 	ret = tcf_exts_validate(net, tp, tb, est, &prog->exts, ovr, true,
420 				extack);
421 	if (ret < 0)
422 		return ret;
423 
424 	if (tb[TCA_BPF_FLAGS]) {
425 		u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
426 
427 		if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT)
428 			return -EINVAL;
429 
430 		have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
431 	}
432 	if (tb[TCA_BPF_FLAGS_GEN]) {
433 		gen_flags = nla_get_u32(tb[TCA_BPF_FLAGS_GEN]);
434 		if (gen_flags & ~CLS_BPF_SUPPORTED_GEN_FLAGS ||
435 		    !tc_flags_valid(gen_flags))
436 			return -EINVAL;
437 	}
438 
439 	prog->exts_integrated = have_exts;
440 	prog->gen_flags = gen_flags;
441 
442 	ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
443 		       cls_bpf_prog_from_efd(tb, prog, gen_flags, tp);
444 	if (ret < 0)
445 		return ret;
446 
447 	if (tb[TCA_BPF_CLASSID]) {
448 		prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
449 		tcf_bind_filter(tp, &prog->res, base);
450 	}
451 
452 	return 0;
453 }
454 
455 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
456 			  struct tcf_proto *tp, unsigned long base,
457 			  u32 handle, struct nlattr **tca,
458 			  void **arg, bool ovr, bool rtnl_held,
459 			  struct netlink_ext_ack *extack)
460 {
461 	struct cls_bpf_head *head = rtnl_dereference(tp->root);
462 	struct cls_bpf_prog *oldprog = *arg;
463 	struct nlattr *tb[TCA_BPF_MAX + 1];
464 	struct cls_bpf_prog *prog;
465 	int ret;
466 
467 	if (tca[TCA_OPTIONS] == NULL)
468 		return -EINVAL;
469 
470 	ret = nla_parse_nested_deprecated(tb, TCA_BPF_MAX, tca[TCA_OPTIONS],
471 					  bpf_policy, NULL);
472 	if (ret < 0)
473 		return ret;
474 
475 	prog = kzalloc(sizeof(*prog), GFP_KERNEL);
476 	if (!prog)
477 		return -ENOBUFS;
478 
479 	ret = tcf_exts_init(&prog->exts, net, TCA_BPF_ACT, TCA_BPF_POLICE);
480 	if (ret < 0)
481 		goto errout;
482 
483 	if (oldprog) {
484 		if (handle && oldprog->handle != handle) {
485 			ret = -EINVAL;
486 			goto errout;
487 		}
488 	}
489 
490 	if (handle == 0) {
491 		handle = 1;
492 		ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
493 				    INT_MAX, GFP_KERNEL);
494 	} else if (!oldprog) {
495 		ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
496 				    handle, GFP_KERNEL);
497 	}
498 
499 	if (ret)
500 		goto errout;
501 	prog->handle = handle;
502 
503 	ret = cls_bpf_set_parms(net, tp, prog, base, tb, tca[TCA_RATE], ovr,
504 				extack);
505 	if (ret < 0)
506 		goto errout_idr;
507 
508 	ret = cls_bpf_offload(tp, prog, oldprog, extack);
509 	if (ret)
510 		goto errout_parms;
511 
512 	if (!tc_in_hw(prog->gen_flags))
513 		prog->gen_flags |= TCA_CLS_FLAGS_NOT_IN_HW;
514 
515 	if (oldprog) {
516 		idr_replace(&head->handle_idr, prog, handle);
517 		list_replace_rcu(&oldprog->link, &prog->link);
518 		tcf_unbind_filter(tp, &oldprog->res);
519 		tcf_exts_get_net(&oldprog->exts);
520 		tcf_queue_work(&oldprog->rwork, cls_bpf_delete_prog_work);
521 	} else {
522 		list_add_rcu(&prog->link, &head->plist);
523 	}
524 
525 	*arg = prog;
526 	return 0;
527 
528 errout_parms:
529 	cls_bpf_free_parms(prog);
530 errout_idr:
531 	if (!oldprog)
532 		idr_remove(&head->handle_idr, prog->handle);
533 errout:
534 	tcf_exts_destroy(&prog->exts);
535 	kfree(prog);
536 	return ret;
537 }
538 
539 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
540 				 struct sk_buff *skb)
541 {
542 	struct nlattr *nla;
543 
544 	if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
545 		return -EMSGSIZE;
546 
547 	nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
548 			  sizeof(struct sock_filter));
549 	if (nla == NULL)
550 		return -EMSGSIZE;
551 
552 	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
553 
554 	return 0;
555 }
556 
557 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
558 				  struct sk_buff *skb)
559 {
560 	struct nlattr *nla;
561 
562 	if (prog->bpf_name &&
563 	    nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
564 		return -EMSGSIZE;
565 
566 	if (nla_put_u32(skb, TCA_BPF_ID, prog->filter->aux->id))
567 		return -EMSGSIZE;
568 
569 	nla = nla_reserve(skb, TCA_BPF_TAG, sizeof(prog->filter->tag));
570 	if (nla == NULL)
571 		return -EMSGSIZE;
572 
573 	memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
574 
575 	return 0;
576 }
577 
578 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, void *fh,
579 			struct sk_buff *skb, struct tcmsg *tm, bool rtnl_held)
580 {
581 	struct cls_bpf_prog *prog = fh;
582 	struct nlattr *nest;
583 	u32 bpf_flags = 0;
584 	int ret;
585 
586 	if (prog == NULL)
587 		return skb->len;
588 
589 	tm->tcm_handle = prog->handle;
590 
591 	cls_bpf_offload_update_stats(tp, prog);
592 
593 	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
594 	if (nest == NULL)
595 		goto nla_put_failure;
596 
597 	if (prog->res.classid &&
598 	    nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
599 		goto nla_put_failure;
600 
601 	if (cls_bpf_is_ebpf(prog))
602 		ret = cls_bpf_dump_ebpf_info(prog, skb);
603 	else
604 		ret = cls_bpf_dump_bpf_info(prog, skb);
605 	if (ret)
606 		goto nla_put_failure;
607 
608 	if (tcf_exts_dump(skb, &prog->exts) < 0)
609 		goto nla_put_failure;
610 
611 	if (prog->exts_integrated)
612 		bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
613 	if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
614 		goto nla_put_failure;
615 	if (prog->gen_flags &&
616 	    nla_put_u32(skb, TCA_BPF_FLAGS_GEN, prog->gen_flags))
617 		goto nla_put_failure;
618 
619 	nla_nest_end(skb, nest);
620 
621 	if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
622 		goto nla_put_failure;
623 
624 	return skb->len;
625 
626 nla_put_failure:
627 	nla_nest_cancel(skb, nest);
628 	return -1;
629 }
630 
631 static void cls_bpf_bind_class(void *fh, u32 classid, unsigned long cl)
632 {
633 	struct cls_bpf_prog *prog = fh;
634 
635 	if (prog && prog->res.classid == classid)
636 		prog->res.class = cl;
637 }
638 
639 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg,
640 			 bool rtnl_held)
641 {
642 	struct cls_bpf_head *head = rtnl_dereference(tp->root);
643 	struct cls_bpf_prog *prog;
644 
645 	list_for_each_entry(prog, &head->plist, link) {
646 		if (arg->count < arg->skip)
647 			goto skip;
648 		if (arg->fn(tp, prog, arg) < 0) {
649 			arg->stop = 1;
650 			break;
651 		}
652 skip:
653 		arg->count++;
654 	}
655 }
656 
657 static int cls_bpf_reoffload(struct tcf_proto *tp, bool add, tc_setup_cb_t *cb,
658 			     void *cb_priv, struct netlink_ext_ack *extack)
659 {
660 	struct cls_bpf_head *head = rtnl_dereference(tp->root);
661 	struct tcf_block *block = tp->chain->block;
662 	struct tc_cls_bpf_offload cls_bpf = {};
663 	struct cls_bpf_prog *prog;
664 	int err;
665 
666 	list_for_each_entry(prog, &head->plist, link) {
667 		if (tc_skip_hw(prog->gen_flags))
668 			continue;
669 
670 		tc_cls_common_offload_init(&cls_bpf.common, tp, prog->gen_flags,
671 					   extack);
672 		cls_bpf.command = TC_CLSBPF_OFFLOAD;
673 		cls_bpf.exts = &prog->exts;
674 		cls_bpf.prog = add ? prog->filter : NULL;
675 		cls_bpf.oldprog = add ? NULL : prog->filter;
676 		cls_bpf.name = prog->bpf_name;
677 		cls_bpf.exts_integrated = prog->exts_integrated;
678 
679 		err = cb(TC_SETUP_CLSBPF, &cls_bpf, cb_priv);
680 		if (err) {
681 			if (add && tc_skip_sw(prog->gen_flags))
682 				return err;
683 			continue;
684 		}
685 
686 		tc_cls_offload_cnt_update(block, &prog->in_hw_count,
687 					  &prog->gen_flags, add);
688 	}
689 
690 	return 0;
691 }
692 
693 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
694 	.kind		=	"bpf",
695 	.owner		=	THIS_MODULE,
696 	.classify	=	cls_bpf_classify,
697 	.init		=	cls_bpf_init,
698 	.destroy	=	cls_bpf_destroy,
699 	.get		=	cls_bpf_get,
700 	.change		=	cls_bpf_change,
701 	.delete		=	cls_bpf_delete,
702 	.walk		=	cls_bpf_walk,
703 	.reoffload	=	cls_bpf_reoffload,
704 	.dump		=	cls_bpf_dump,
705 	.bind_class	=	cls_bpf_bind_class,
706 };
707 
708 static int __init cls_bpf_init_mod(void)
709 {
710 	return register_tcf_proto_ops(&cls_bpf_ops);
711 }
712 
713 static void __exit cls_bpf_exit_mod(void)
714 {
715 	unregister_tcf_proto_ops(&cls_bpf_ops);
716 }
717 
718 module_init(cls_bpf_init_mod);
719 module_exit(cls_bpf_exit_mod);
720