Linux Audio

Check our new training course

Loading...
v5.4
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Berkeley Packet Filter based traffic classifier
  4 *
  5 * Might be used to classify traffic through flexible, user-defined and
  6 * possibly JIT-ed BPF filters for traffic control as an alternative to
  7 * ematches.
  8 *
  9 * (C) 2013 Daniel Borkmann <dborkman@redhat.com>
 10 */
 11
 12#include <linux/module.h>
 13#include <linux/types.h>
 14#include <linux/skbuff.h>
 15#include <linux/filter.h>
 16#include <linux/bpf.h>
 17#include <linux/idr.h>
 18
 19#include <net/rtnetlink.h>
 20#include <net/pkt_cls.h>
 21#include <net/sock.h>
 
 22
 23MODULE_LICENSE("GPL");
 24MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
 25MODULE_DESCRIPTION("TC BPF based classifier");
 26
 27#define CLS_BPF_NAME_LEN	256
 28#define CLS_BPF_SUPPORTED_GEN_FLAGS		\
 29	(TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)
 30
 31struct cls_bpf_head {
 32	struct list_head plist;
 33	struct idr handle_idr;
 34	struct rcu_head rcu;
 35};
 36
 37struct cls_bpf_prog {
 38	struct bpf_prog *filter;
 39	struct list_head link;
 40	struct tcf_result res;
 41	bool exts_integrated;
 42	u32 gen_flags;
 43	unsigned int in_hw_count;
 44	struct tcf_exts exts;
 45	u32 handle;
 46	u16 bpf_num_ops;
 47	struct sock_filter *bpf_ops;
 48	const char *bpf_name;
 49	struct tcf_proto *tp;
 50	struct rcu_work rwork;
 51};
 52
 53static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
 54	[TCA_BPF_CLASSID]	= { .type = NLA_U32 },
 55	[TCA_BPF_FLAGS]		= { .type = NLA_U32 },
 56	[TCA_BPF_FLAGS_GEN]	= { .type = NLA_U32 },
 57	[TCA_BPF_FD]		= { .type = NLA_U32 },
 58	[TCA_BPF_NAME]		= { .type = NLA_NUL_STRING,
 59				    .len = CLS_BPF_NAME_LEN },
 60	[TCA_BPF_OPS_LEN]	= { .type = NLA_U16 },
 61	[TCA_BPF_OPS]		= { .type = NLA_BINARY,
 62				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
 63};
 64
 65static int cls_bpf_exec_opcode(int code)
 66{
 67	switch (code) {
 68	case TC_ACT_OK:
 69	case TC_ACT_SHOT:
 70	case TC_ACT_STOLEN:
 71	case TC_ACT_TRAP:
 72	case TC_ACT_REDIRECT:
 73	case TC_ACT_UNSPEC:
 74		return code;
 75	default:
 76		return TC_ACT_UNSPEC;
 77	}
 78}
 79
 80static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 81			    struct tcf_result *res)
 
 82{
 83	struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
 84	bool at_ingress = skb_at_tc_ingress(skb);
 85	struct cls_bpf_prog *prog;
 86	int ret = -1;
 87
 88	/* Needed here for accessing maps. */
 89	rcu_read_lock();
 90	list_for_each_entry_rcu(prog, &head->plist, link) {
 91		int filter_res;
 92
 93		qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
 94
 95		if (tc_skip_sw(prog->gen_flags)) {
 96			filter_res = prog->exts_integrated ? TC_ACT_UNSPEC : 0;
 97		} else if (at_ingress) {
 98			/* It is safe to push/pull even if skb_shared() */
 99			__skb_push(skb, skb->mac_len);
100			bpf_compute_data_pointers(skb);
101			filter_res = BPF_PROG_RUN(prog->filter, skb);
102			__skb_pull(skb, skb->mac_len);
103		} else {
104			bpf_compute_data_pointers(skb);
105			filter_res = BPF_PROG_RUN(prog->filter, skb);
106		}
 
 
107
108		if (prog->exts_integrated) {
109			res->class   = 0;
110			res->classid = TC_H_MAJ(prog->res.classid) |
111				       qdisc_skb_cb(skb)->tc_classid;
112
113			ret = cls_bpf_exec_opcode(filter_res);
114			if (ret == TC_ACT_UNSPEC)
115				continue;
116			break;
117		}
118
119		if (filter_res == 0)
120			continue;
121		if (filter_res != -1) {
122			res->class   = 0;
123			res->classid = filter_res;
124		} else {
125			*res = prog->res;
126		}
127
128		ret = tcf_exts_exec(skb, &prog->exts, res);
129		if (ret < 0)
130			continue;
131
132		break;
133	}
134	rcu_read_unlock();
135
136	return ret;
137}
138
139static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
140{
141	return !prog->bpf_ops;
142}
143
144static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
145			       struct cls_bpf_prog *oldprog,
146			       struct netlink_ext_ack *extack)
147{
148	struct tcf_block *block = tp->chain->block;
149	struct tc_cls_bpf_offload cls_bpf = {};
150	struct cls_bpf_prog *obj;
151	bool skip_sw;
152	int err;
153
154	skip_sw = prog && tc_skip_sw(prog->gen_flags);
155	obj = prog ?: oldprog;
156
157	tc_cls_common_offload_init(&cls_bpf.common, tp, obj->gen_flags, extack);
158	cls_bpf.command = TC_CLSBPF_OFFLOAD;
159	cls_bpf.exts = &obj->exts;
160	cls_bpf.prog = prog ? prog->filter : NULL;
161	cls_bpf.oldprog = oldprog ? oldprog->filter : NULL;
162	cls_bpf.name = obj->bpf_name;
163	cls_bpf.exts_integrated = obj->exts_integrated;
164
165	if (oldprog && prog)
166		err = tc_setup_cb_replace(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
167					  skip_sw, &oldprog->gen_flags,
168					  &oldprog->in_hw_count,
169					  &prog->gen_flags, &prog->in_hw_count,
170					  true);
171	else if (prog)
172		err = tc_setup_cb_add(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
173				      skip_sw, &prog->gen_flags,
174				      &prog->in_hw_count, true);
175	else
176		err = tc_setup_cb_destroy(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
177					  skip_sw, &oldprog->gen_flags,
178					  &oldprog->in_hw_count, true);
179
180	if (prog && err) {
181		cls_bpf_offload_cmd(tp, oldprog, prog, extack);
182		return err;
183	}
184
185	if (prog && skip_sw && !(prog->gen_flags & TCA_CLS_FLAGS_IN_HW))
186		return -EINVAL;
187
188	return 0;
189}
190
191static u32 cls_bpf_flags(u32 flags)
192{
193	return flags & CLS_BPF_SUPPORTED_GEN_FLAGS;
194}
195
196static int cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
197			   struct cls_bpf_prog *oldprog,
198			   struct netlink_ext_ack *extack)
199{
200	if (prog && oldprog &&
201	    cls_bpf_flags(prog->gen_flags) !=
202	    cls_bpf_flags(oldprog->gen_flags))
203		return -EINVAL;
204
205	if (prog && tc_skip_hw(prog->gen_flags))
206		prog = NULL;
207	if (oldprog && tc_skip_hw(oldprog->gen_flags))
208		oldprog = NULL;
209	if (!prog && !oldprog)
210		return 0;
211
212	return cls_bpf_offload_cmd(tp, prog, oldprog, extack);
213}
214
215static void cls_bpf_stop_offload(struct tcf_proto *tp,
216				 struct cls_bpf_prog *prog,
217				 struct netlink_ext_ack *extack)
218{
219	int err;
220
221	err = cls_bpf_offload_cmd(tp, NULL, prog, extack);
222	if (err)
223		pr_err("Stopping hardware offload failed: %d\n", err);
224}
225
226static void cls_bpf_offload_update_stats(struct tcf_proto *tp,
227					 struct cls_bpf_prog *prog)
228{
229	struct tcf_block *block = tp->chain->block;
230	struct tc_cls_bpf_offload cls_bpf = {};
231
232	tc_cls_common_offload_init(&cls_bpf.common, tp, prog->gen_flags, NULL);
233	cls_bpf.command = TC_CLSBPF_STATS;
234	cls_bpf.exts = &prog->exts;
235	cls_bpf.prog = prog->filter;
236	cls_bpf.name = prog->bpf_name;
237	cls_bpf.exts_integrated = prog->exts_integrated;
238
239	tc_setup_cb_call(block, TC_SETUP_CLSBPF, &cls_bpf, false, true);
240}
241
242static int cls_bpf_init(struct tcf_proto *tp)
243{
244	struct cls_bpf_head *head;
245
246	head = kzalloc(sizeof(*head), GFP_KERNEL);
247	if (head == NULL)
248		return -ENOBUFS;
249
250	INIT_LIST_HEAD_RCU(&head->plist);
251	idr_init(&head->handle_idr);
252	rcu_assign_pointer(tp->root, head);
253
254	return 0;
255}
256
257static void cls_bpf_free_parms(struct cls_bpf_prog *prog)
258{
259	if (cls_bpf_is_ebpf(prog))
260		bpf_prog_put(prog->filter);
261	else
262		bpf_prog_destroy(prog->filter);
263
264	kfree(prog->bpf_name);
265	kfree(prog->bpf_ops);
266}
267
268static void __cls_bpf_delete_prog(struct cls_bpf_prog *prog)
269{
270	tcf_exts_destroy(&prog->exts);
271	tcf_exts_put_net(&prog->exts);
272
273	cls_bpf_free_parms(prog);
274	kfree(prog);
275}
276
277static void cls_bpf_delete_prog_work(struct work_struct *work)
278{
279	struct cls_bpf_prog *prog = container_of(to_rcu_work(work),
280						 struct cls_bpf_prog,
281						 rwork);
282	rtnl_lock();
283	__cls_bpf_delete_prog(prog);
284	rtnl_unlock();
285}
286
287static void __cls_bpf_delete(struct tcf_proto *tp, struct cls_bpf_prog *prog,
288			     struct netlink_ext_ack *extack)
289{
290	struct cls_bpf_head *head = rtnl_dereference(tp->root);
291
292	idr_remove(&head->handle_idr, prog->handle);
293	cls_bpf_stop_offload(tp, prog, extack);
294	list_del_rcu(&prog->link);
295	tcf_unbind_filter(tp, &prog->res);
296	if (tcf_exts_get_net(&prog->exts))
297		tcf_queue_work(&prog->rwork, cls_bpf_delete_prog_work);
298	else
299		__cls_bpf_delete_prog(prog);
300}
301
302static int cls_bpf_delete(struct tcf_proto *tp, void *arg, bool *last,
303			  bool rtnl_held, struct netlink_ext_ack *extack)
304{
305	struct cls_bpf_head *head = rtnl_dereference(tp->root);
306
307	__cls_bpf_delete(tp, arg, extack);
308	*last = list_empty(&head->plist);
309	return 0;
310}
311
312static void cls_bpf_destroy(struct tcf_proto *tp, bool rtnl_held,
313			    struct netlink_ext_ack *extack)
314{
315	struct cls_bpf_head *head = rtnl_dereference(tp->root);
316	struct cls_bpf_prog *prog, *tmp;
317
318	list_for_each_entry_safe(prog, tmp, &head->plist, link)
319		__cls_bpf_delete(tp, prog, extack);
320
321	idr_destroy(&head->handle_idr);
322	kfree_rcu(head, rcu);
323}
324
325static void *cls_bpf_get(struct tcf_proto *tp, u32 handle)
326{
327	struct cls_bpf_head *head = rtnl_dereference(tp->root);
328	struct cls_bpf_prog *prog;
329
330	list_for_each_entry(prog, &head->plist, link) {
331		if (prog->handle == handle)
332			return prog;
333	}
334
335	return NULL;
336}
337
338static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
339{
340	struct sock_filter *bpf_ops;
341	struct sock_fprog_kern fprog_tmp;
342	struct bpf_prog *fp;
343	u16 bpf_size, bpf_num_ops;
344	int ret;
345
346	bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
347	if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
348		return -EINVAL;
349
350	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
351	if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
352		return -EINVAL;
353
354	bpf_ops = kmemdup(nla_data(tb[TCA_BPF_OPS]), bpf_size, GFP_KERNEL);
355	if (bpf_ops == NULL)
356		return -ENOMEM;
357
358	fprog_tmp.len = bpf_num_ops;
359	fprog_tmp.filter = bpf_ops;
360
361	ret = bpf_prog_create(&fp, &fprog_tmp);
362	if (ret < 0) {
363		kfree(bpf_ops);
364		return ret;
365	}
366
367	prog->bpf_ops = bpf_ops;
368	prog->bpf_num_ops = bpf_num_ops;
369	prog->bpf_name = NULL;
370	prog->filter = fp;
371
372	return 0;
373}
374
375static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
376				 u32 gen_flags, const struct tcf_proto *tp)
377{
378	struct bpf_prog *fp;
379	char *name = NULL;
380	bool skip_sw;
381	u32 bpf_fd;
382
383	bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
384	skip_sw = gen_flags & TCA_CLS_FLAGS_SKIP_SW;
385
386	fp = bpf_prog_get_type_dev(bpf_fd, BPF_PROG_TYPE_SCHED_CLS, skip_sw);
387	if (IS_ERR(fp))
388		return PTR_ERR(fp);
389
390	if (tb[TCA_BPF_NAME]) {
391		name = nla_memdup(tb[TCA_BPF_NAME], GFP_KERNEL);
392		if (!name) {
393			bpf_prog_put(fp);
394			return -ENOMEM;
395		}
396	}
397
398	prog->bpf_ops = NULL;
399	prog->bpf_name = name;
400	prog->filter = fp;
401
402	if (fp->dst_needed)
403		tcf_block_netif_keep_dst(tp->chain->block);
404
405	return 0;
406}
407
408static int cls_bpf_set_parms(struct net *net, struct tcf_proto *tp,
409			     struct cls_bpf_prog *prog, unsigned long base,
410			     struct nlattr **tb, struct nlattr *est, bool ovr,
411			     struct netlink_ext_ack *extack)
412{
413	bool is_bpf, is_ebpf, have_exts = false;
414	u32 gen_flags = 0;
415	int ret;
416
417	is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
418	is_ebpf = tb[TCA_BPF_FD];
419	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf))
420		return -EINVAL;
421
422	ret = tcf_exts_validate(net, tp, tb, est, &prog->exts, ovr, true,
423				extack);
424	if (ret < 0)
425		return ret;
426
427	if (tb[TCA_BPF_FLAGS]) {
428		u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
429
430		if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT)
431			return -EINVAL;
432
433		have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
434	}
435	if (tb[TCA_BPF_FLAGS_GEN]) {
436		gen_flags = nla_get_u32(tb[TCA_BPF_FLAGS_GEN]);
437		if (gen_flags & ~CLS_BPF_SUPPORTED_GEN_FLAGS ||
438		    !tc_flags_valid(gen_flags))
439			return -EINVAL;
440	}
441
442	prog->exts_integrated = have_exts;
443	prog->gen_flags = gen_flags;
444
445	ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
446		       cls_bpf_prog_from_efd(tb, prog, gen_flags, tp);
447	if (ret < 0)
448		return ret;
449
450	if (tb[TCA_BPF_CLASSID]) {
451		prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
452		tcf_bind_filter(tp, &prog->res, base);
453	}
454
455	return 0;
456}
457
458static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
459			  struct tcf_proto *tp, unsigned long base,
460			  u32 handle, struct nlattr **tca,
461			  void **arg, bool ovr, bool rtnl_held,
462			  struct netlink_ext_ack *extack)
463{
464	struct cls_bpf_head *head = rtnl_dereference(tp->root);
 
465	struct cls_bpf_prog *oldprog = *arg;
466	struct nlattr *tb[TCA_BPF_MAX + 1];
 
467	struct cls_bpf_prog *prog;
 
468	int ret;
469
470	if (tca[TCA_OPTIONS] == NULL)
471		return -EINVAL;
472
473	ret = nla_parse_nested_deprecated(tb, TCA_BPF_MAX, tca[TCA_OPTIONS],
474					  bpf_policy, NULL);
475	if (ret < 0)
476		return ret;
477
478	prog = kzalloc(sizeof(*prog), GFP_KERNEL);
479	if (!prog)
480		return -ENOBUFS;
481
482	ret = tcf_exts_init(&prog->exts, net, TCA_BPF_ACT, TCA_BPF_POLICE);
483	if (ret < 0)
484		goto errout;
485
486	if (oldprog) {
487		if (handle && oldprog->handle != handle) {
488			ret = -EINVAL;
489			goto errout;
490		}
491	}
492
493	if (handle == 0) {
494		handle = 1;
495		ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
496				    INT_MAX, GFP_KERNEL);
497	} else if (!oldprog) {
498		ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
499				    handle, GFP_KERNEL);
500	}
501
502	if (ret)
503		goto errout;
504	prog->handle = handle;
505
506	ret = cls_bpf_set_parms(net, tp, prog, base, tb, tca[TCA_RATE], ovr,
507				extack);
 
 
 
 
 
 
 
508	if (ret < 0)
509		goto errout_idr;
510
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
511	ret = cls_bpf_offload(tp, prog, oldprog, extack);
512	if (ret)
513		goto errout_parms;
514
515	if (!tc_in_hw(prog->gen_flags))
516		prog->gen_flags |= TCA_CLS_FLAGS_NOT_IN_HW;
517
518	if (oldprog) {
519		idr_replace(&head->handle_idr, prog, handle);
520		list_replace_rcu(&oldprog->link, &prog->link);
521		tcf_unbind_filter(tp, &oldprog->res);
522		tcf_exts_get_net(&oldprog->exts);
523		tcf_queue_work(&oldprog->rwork, cls_bpf_delete_prog_work);
524	} else {
525		list_add_rcu(&prog->link, &head->plist);
526	}
527
528	*arg = prog;
529	return 0;
530
531errout_parms:
 
 
532	cls_bpf_free_parms(prog);
533errout_idr:
534	if (!oldprog)
535		idr_remove(&head->handle_idr, prog->handle);
536errout:
537	tcf_exts_destroy(&prog->exts);
538	kfree(prog);
539	return ret;
540}
541
542static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
543				 struct sk_buff *skb)
544{
545	struct nlattr *nla;
546
547	if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
548		return -EMSGSIZE;
549
550	nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
551			  sizeof(struct sock_filter));
552	if (nla == NULL)
553		return -EMSGSIZE;
554
555	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
556
557	return 0;
558}
559
560static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
561				  struct sk_buff *skb)
562{
563	struct nlattr *nla;
564
565	if (prog->bpf_name &&
566	    nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
567		return -EMSGSIZE;
568
569	if (nla_put_u32(skb, TCA_BPF_ID, prog->filter->aux->id))
570		return -EMSGSIZE;
571
572	nla = nla_reserve(skb, TCA_BPF_TAG, sizeof(prog->filter->tag));
573	if (nla == NULL)
574		return -EMSGSIZE;
575
576	memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
577
578	return 0;
579}
580
581static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, void *fh,
582			struct sk_buff *skb, struct tcmsg *tm, bool rtnl_held)
583{
584	struct cls_bpf_prog *prog = fh;
585	struct nlattr *nest;
586	u32 bpf_flags = 0;
587	int ret;
588
589	if (prog == NULL)
590		return skb->len;
591
592	tm->tcm_handle = prog->handle;
593
594	cls_bpf_offload_update_stats(tp, prog);
595
596	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
597	if (nest == NULL)
598		goto nla_put_failure;
599
600	if (prog->res.classid &&
601	    nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
602		goto nla_put_failure;
603
604	if (cls_bpf_is_ebpf(prog))
605		ret = cls_bpf_dump_ebpf_info(prog, skb);
606	else
607		ret = cls_bpf_dump_bpf_info(prog, skb);
608	if (ret)
609		goto nla_put_failure;
610
611	if (tcf_exts_dump(skb, &prog->exts) < 0)
612		goto nla_put_failure;
613
614	if (prog->exts_integrated)
615		bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
616	if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
617		goto nla_put_failure;
618	if (prog->gen_flags &&
619	    nla_put_u32(skb, TCA_BPF_FLAGS_GEN, prog->gen_flags))
620		goto nla_put_failure;
621
622	nla_nest_end(skb, nest);
623
624	if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
625		goto nla_put_failure;
626
627	return skb->len;
628
629nla_put_failure:
630	nla_nest_cancel(skb, nest);
631	return -1;
632}
633
634static void cls_bpf_bind_class(void *fh, u32 classid, unsigned long cl)
 
635{
636	struct cls_bpf_prog *prog = fh;
637
638	if (prog && prog->res.classid == classid)
639		prog->res.class = cl;
640}
641
642static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg,
643			 bool rtnl_held)
644{
645	struct cls_bpf_head *head = rtnl_dereference(tp->root);
646	struct cls_bpf_prog *prog;
647
648	list_for_each_entry(prog, &head->plist, link) {
649		if (arg->count < arg->skip)
650			goto skip;
651		if (arg->fn(tp, prog, arg) < 0) {
652			arg->stop = 1;
653			break;
654		}
655skip:
656		arg->count++;
657	}
658}
659
660static int cls_bpf_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
661			     void *cb_priv, struct netlink_ext_ack *extack)
662{
663	struct cls_bpf_head *head = rtnl_dereference(tp->root);
664	struct tcf_block *block = tp->chain->block;
665	struct tc_cls_bpf_offload cls_bpf = {};
666	struct cls_bpf_prog *prog;
667	int err;
668
669	list_for_each_entry(prog, &head->plist, link) {
670		if (tc_skip_hw(prog->gen_flags))
671			continue;
672
673		tc_cls_common_offload_init(&cls_bpf.common, tp, prog->gen_flags,
674					   extack);
675		cls_bpf.command = TC_CLSBPF_OFFLOAD;
676		cls_bpf.exts = &prog->exts;
677		cls_bpf.prog = add ? prog->filter : NULL;
678		cls_bpf.oldprog = add ? NULL : prog->filter;
679		cls_bpf.name = prog->bpf_name;
680		cls_bpf.exts_integrated = prog->exts_integrated;
681
682		err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSBPF,
683					    &cls_bpf, cb_priv, &prog->gen_flags,
684					    &prog->in_hw_count);
685		if (err)
686			return err;
687	}
688
689	return 0;
690}
691
692static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
693	.kind		=	"bpf",
694	.owner		=	THIS_MODULE,
695	.classify	=	cls_bpf_classify,
696	.init		=	cls_bpf_init,
697	.destroy	=	cls_bpf_destroy,
698	.get		=	cls_bpf_get,
699	.change		=	cls_bpf_change,
700	.delete		=	cls_bpf_delete,
701	.walk		=	cls_bpf_walk,
702	.reoffload	=	cls_bpf_reoffload,
703	.dump		=	cls_bpf_dump,
704	.bind_class	=	cls_bpf_bind_class,
705};
706
707static int __init cls_bpf_init_mod(void)
708{
709	return register_tcf_proto_ops(&cls_bpf_ops);
710}
711
712static void __exit cls_bpf_exit_mod(void)
713{
714	unregister_tcf_proto_ops(&cls_bpf_ops);
715}
716
717module_init(cls_bpf_init_mod);
718module_exit(cls_bpf_exit_mod);
v6.8
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Berkeley Packet Filter based traffic classifier
  4 *
  5 * Might be used to classify traffic through flexible, user-defined and
  6 * possibly JIT-ed BPF filters for traffic control as an alternative to
  7 * ematches.
  8 *
  9 * (C) 2013 Daniel Borkmann <dborkman@redhat.com>
 10 */
 11
 12#include <linux/module.h>
 13#include <linux/types.h>
 14#include <linux/skbuff.h>
 15#include <linux/filter.h>
 16#include <linux/bpf.h>
 17#include <linux/idr.h>
 18
 19#include <net/rtnetlink.h>
 20#include <net/pkt_cls.h>
 21#include <net/sock.h>
 22#include <net/tc_wrapper.h>
 23
 24MODULE_LICENSE("GPL");
 25MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
 26MODULE_DESCRIPTION("TC BPF based classifier");
 27
 28#define CLS_BPF_NAME_LEN	256
 29#define CLS_BPF_SUPPORTED_GEN_FLAGS		\
 30	(TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)
 31
 32struct cls_bpf_head {
 33	struct list_head plist;
 34	struct idr handle_idr;
 35	struct rcu_head rcu;
 36};
 37
 38struct cls_bpf_prog {
 39	struct bpf_prog *filter;
 40	struct list_head link;
 41	struct tcf_result res;
 42	bool exts_integrated;
 43	u32 gen_flags;
 44	unsigned int in_hw_count;
 45	struct tcf_exts exts;
 46	u32 handle;
 47	u16 bpf_num_ops;
 48	struct sock_filter *bpf_ops;
 49	const char *bpf_name;
 50	struct tcf_proto *tp;
 51	struct rcu_work rwork;
 52};
 53
 54static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
 55	[TCA_BPF_CLASSID]	= { .type = NLA_U32 },
 56	[TCA_BPF_FLAGS]		= { .type = NLA_U32 },
 57	[TCA_BPF_FLAGS_GEN]	= { .type = NLA_U32 },
 58	[TCA_BPF_FD]		= { .type = NLA_U32 },
 59	[TCA_BPF_NAME]		= { .type = NLA_NUL_STRING,
 60				    .len = CLS_BPF_NAME_LEN },
 61	[TCA_BPF_OPS_LEN]	= { .type = NLA_U16 },
 62	[TCA_BPF_OPS]		= { .type = NLA_BINARY,
 63				    .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
 64};
 65
 66static int cls_bpf_exec_opcode(int code)
 67{
 68	switch (code) {
 69	case TC_ACT_OK:
 70	case TC_ACT_SHOT:
 71	case TC_ACT_STOLEN:
 72	case TC_ACT_TRAP:
 73	case TC_ACT_REDIRECT:
 74	case TC_ACT_UNSPEC:
 75		return code;
 76	default:
 77		return TC_ACT_UNSPEC;
 78	}
 79}
 80
 81TC_INDIRECT_SCOPE int cls_bpf_classify(struct sk_buff *skb,
 82				       const struct tcf_proto *tp,
 83				       struct tcf_result *res)
 84{
 85	struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
 86	bool at_ingress = skb_at_tc_ingress(skb);
 87	struct cls_bpf_prog *prog;
 88	int ret = -1;
 89
 
 
 90	list_for_each_entry_rcu(prog, &head->plist, link) {
 91		int filter_res;
 92
 93		qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
 94
 95		if (tc_skip_sw(prog->gen_flags)) {
 96			filter_res = prog->exts_integrated ? TC_ACT_UNSPEC : 0;
 97		} else if (at_ingress) {
 98			/* It is safe to push/pull even if skb_shared() */
 99			__skb_push(skb, skb->mac_len);
100			bpf_compute_data_pointers(skb);
101			filter_res = bpf_prog_run(prog->filter, skb);
102			__skb_pull(skb, skb->mac_len);
103		} else {
104			bpf_compute_data_pointers(skb);
105			filter_res = bpf_prog_run(prog->filter, skb);
106		}
107		if (unlikely(!skb->tstamp && skb->mono_delivery_time))
108			skb->mono_delivery_time = 0;
109
110		if (prog->exts_integrated) {
111			res->class   = 0;
112			res->classid = TC_H_MAJ(prog->res.classid) |
113				       qdisc_skb_cb(skb)->tc_classid;
114
115			ret = cls_bpf_exec_opcode(filter_res);
116			if (ret == TC_ACT_UNSPEC)
117				continue;
118			break;
119		}
120
121		if (filter_res == 0)
122			continue;
123		if (filter_res != -1) {
124			res->class   = 0;
125			res->classid = filter_res;
126		} else {
127			*res = prog->res;
128		}
129
130		ret = tcf_exts_exec(skb, &prog->exts, res);
131		if (ret < 0)
132			continue;
133
134		break;
135	}
 
136
137	return ret;
138}
139
140static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
141{
142	return !prog->bpf_ops;
143}
144
145static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
146			       struct cls_bpf_prog *oldprog,
147			       struct netlink_ext_ack *extack)
148{
149	struct tcf_block *block = tp->chain->block;
150	struct tc_cls_bpf_offload cls_bpf = {};
151	struct cls_bpf_prog *obj;
152	bool skip_sw;
153	int err;
154
155	skip_sw = prog && tc_skip_sw(prog->gen_flags);
156	obj = prog ?: oldprog;
157
158	tc_cls_common_offload_init(&cls_bpf.common, tp, obj->gen_flags, extack);
159	cls_bpf.command = TC_CLSBPF_OFFLOAD;
160	cls_bpf.exts = &obj->exts;
161	cls_bpf.prog = prog ? prog->filter : NULL;
162	cls_bpf.oldprog = oldprog ? oldprog->filter : NULL;
163	cls_bpf.name = obj->bpf_name;
164	cls_bpf.exts_integrated = obj->exts_integrated;
165
166	if (oldprog && prog)
167		err = tc_setup_cb_replace(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
168					  skip_sw, &oldprog->gen_flags,
169					  &oldprog->in_hw_count,
170					  &prog->gen_flags, &prog->in_hw_count,
171					  true);
172	else if (prog)
173		err = tc_setup_cb_add(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
174				      skip_sw, &prog->gen_flags,
175				      &prog->in_hw_count, true);
176	else
177		err = tc_setup_cb_destroy(block, tp, TC_SETUP_CLSBPF, &cls_bpf,
178					  skip_sw, &oldprog->gen_flags,
179					  &oldprog->in_hw_count, true);
180
181	if (prog && err) {
182		cls_bpf_offload_cmd(tp, oldprog, prog, extack);
183		return err;
184	}
185
186	if (prog && skip_sw && !(prog->gen_flags & TCA_CLS_FLAGS_IN_HW))
187		return -EINVAL;
188
189	return 0;
190}
191
192static u32 cls_bpf_flags(u32 flags)
193{
194	return flags & CLS_BPF_SUPPORTED_GEN_FLAGS;
195}
196
197static int cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
198			   struct cls_bpf_prog *oldprog,
199			   struct netlink_ext_ack *extack)
200{
201	if (prog && oldprog &&
202	    cls_bpf_flags(prog->gen_flags) !=
203	    cls_bpf_flags(oldprog->gen_flags))
204		return -EINVAL;
205
206	if (prog && tc_skip_hw(prog->gen_flags))
207		prog = NULL;
208	if (oldprog && tc_skip_hw(oldprog->gen_flags))
209		oldprog = NULL;
210	if (!prog && !oldprog)
211		return 0;
212
213	return cls_bpf_offload_cmd(tp, prog, oldprog, extack);
214}
215
216static void cls_bpf_stop_offload(struct tcf_proto *tp,
217				 struct cls_bpf_prog *prog,
218				 struct netlink_ext_ack *extack)
219{
220	int err;
221
222	err = cls_bpf_offload_cmd(tp, NULL, prog, extack);
223	if (err)
224		pr_err("Stopping hardware offload failed: %d\n", err);
225}
226
227static void cls_bpf_offload_update_stats(struct tcf_proto *tp,
228					 struct cls_bpf_prog *prog)
229{
230	struct tcf_block *block = tp->chain->block;
231	struct tc_cls_bpf_offload cls_bpf = {};
232
233	tc_cls_common_offload_init(&cls_bpf.common, tp, prog->gen_flags, NULL);
234	cls_bpf.command = TC_CLSBPF_STATS;
235	cls_bpf.exts = &prog->exts;
236	cls_bpf.prog = prog->filter;
237	cls_bpf.name = prog->bpf_name;
238	cls_bpf.exts_integrated = prog->exts_integrated;
239
240	tc_setup_cb_call(block, TC_SETUP_CLSBPF, &cls_bpf, false, true);
241}
242
243static int cls_bpf_init(struct tcf_proto *tp)
244{
245	struct cls_bpf_head *head;
246
247	head = kzalloc(sizeof(*head), GFP_KERNEL);
248	if (head == NULL)
249		return -ENOBUFS;
250
251	INIT_LIST_HEAD_RCU(&head->plist);
252	idr_init(&head->handle_idr);
253	rcu_assign_pointer(tp->root, head);
254
255	return 0;
256}
257
258static void cls_bpf_free_parms(struct cls_bpf_prog *prog)
259{
260	if (cls_bpf_is_ebpf(prog))
261		bpf_prog_put(prog->filter);
262	else
263		bpf_prog_destroy(prog->filter);
264
265	kfree(prog->bpf_name);
266	kfree(prog->bpf_ops);
267}
268
269static void __cls_bpf_delete_prog(struct cls_bpf_prog *prog)
270{
271	tcf_exts_destroy(&prog->exts);
272	tcf_exts_put_net(&prog->exts);
273
274	cls_bpf_free_parms(prog);
275	kfree(prog);
276}
277
278static void cls_bpf_delete_prog_work(struct work_struct *work)
279{
280	struct cls_bpf_prog *prog = container_of(to_rcu_work(work),
281						 struct cls_bpf_prog,
282						 rwork);
283	rtnl_lock();
284	__cls_bpf_delete_prog(prog);
285	rtnl_unlock();
286}
287
288static void __cls_bpf_delete(struct tcf_proto *tp, struct cls_bpf_prog *prog,
289			     struct netlink_ext_ack *extack)
290{
291	struct cls_bpf_head *head = rtnl_dereference(tp->root);
292
293	idr_remove(&head->handle_idr, prog->handle);
294	cls_bpf_stop_offload(tp, prog, extack);
295	list_del_rcu(&prog->link);
296	tcf_unbind_filter(tp, &prog->res);
297	if (tcf_exts_get_net(&prog->exts))
298		tcf_queue_work(&prog->rwork, cls_bpf_delete_prog_work);
299	else
300		__cls_bpf_delete_prog(prog);
301}
302
303static int cls_bpf_delete(struct tcf_proto *tp, void *arg, bool *last,
304			  bool rtnl_held, struct netlink_ext_ack *extack)
305{
306	struct cls_bpf_head *head = rtnl_dereference(tp->root);
307
308	__cls_bpf_delete(tp, arg, extack);
309	*last = list_empty(&head->plist);
310	return 0;
311}
312
313static void cls_bpf_destroy(struct tcf_proto *tp, bool rtnl_held,
314			    struct netlink_ext_ack *extack)
315{
316	struct cls_bpf_head *head = rtnl_dereference(tp->root);
317	struct cls_bpf_prog *prog, *tmp;
318
319	list_for_each_entry_safe(prog, tmp, &head->plist, link)
320		__cls_bpf_delete(tp, prog, extack);
321
322	idr_destroy(&head->handle_idr);
323	kfree_rcu(head, rcu);
324}
325
326static void *cls_bpf_get(struct tcf_proto *tp, u32 handle)
327{
328	struct cls_bpf_head *head = rtnl_dereference(tp->root);
329	struct cls_bpf_prog *prog;
330
331	list_for_each_entry(prog, &head->plist, link) {
332		if (prog->handle == handle)
333			return prog;
334	}
335
336	return NULL;
337}
338
339static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
340{
341	struct sock_filter *bpf_ops;
342	struct sock_fprog_kern fprog_tmp;
343	struct bpf_prog *fp;
344	u16 bpf_size, bpf_num_ops;
345	int ret;
346
347	bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
348	if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
349		return -EINVAL;
350
351	bpf_size = bpf_num_ops * sizeof(*bpf_ops);
352	if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
353		return -EINVAL;
354
355	bpf_ops = kmemdup(nla_data(tb[TCA_BPF_OPS]), bpf_size, GFP_KERNEL);
356	if (bpf_ops == NULL)
357		return -ENOMEM;
358
359	fprog_tmp.len = bpf_num_ops;
360	fprog_tmp.filter = bpf_ops;
361
362	ret = bpf_prog_create(&fp, &fprog_tmp);
363	if (ret < 0) {
364		kfree(bpf_ops);
365		return ret;
366	}
367
368	prog->bpf_ops = bpf_ops;
369	prog->bpf_num_ops = bpf_num_ops;
370	prog->bpf_name = NULL;
371	prog->filter = fp;
372
373	return 0;
374}
375
376static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
377				 u32 gen_flags, const struct tcf_proto *tp)
378{
379	struct bpf_prog *fp;
380	char *name = NULL;
381	bool skip_sw;
382	u32 bpf_fd;
383
384	bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
385	skip_sw = gen_flags & TCA_CLS_FLAGS_SKIP_SW;
386
387	fp = bpf_prog_get_type_dev(bpf_fd, BPF_PROG_TYPE_SCHED_CLS, skip_sw);
388	if (IS_ERR(fp))
389		return PTR_ERR(fp);
390
391	if (tb[TCA_BPF_NAME]) {
392		name = nla_memdup(tb[TCA_BPF_NAME], GFP_KERNEL);
393		if (!name) {
394			bpf_prog_put(fp);
395			return -ENOMEM;
396		}
397	}
398
399	prog->bpf_ops = NULL;
400	prog->bpf_name = name;
401	prog->filter = fp;
402
403	if (fp->dst_needed)
404		tcf_block_netif_keep_dst(tp->chain->block);
405
406	return 0;
407}
408
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
409static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
410			  struct tcf_proto *tp, unsigned long base,
411			  u32 handle, struct nlattr **tca,
412			  void **arg, u32 flags,
413			  struct netlink_ext_ack *extack)
414{
415	struct cls_bpf_head *head = rtnl_dereference(tp->root);
416	bool is_bpf, is_ebpf, have_exts = false;
417	struct cls_bpf_prog *oldprog = *arg;
418	struct nlattr *tb[TCA_BPF_MAX + 1];
419	bool bound_to_filter = false;
420	struct cls_bpf_prog *prog;
421	u32 gen_flags = 0;
422	int ret;
423
424	if (tca[TCA_OPTIONS] == NULL)
425		return -EINVAL;
426
427	ret = nla_parse_nested_deprecated(tb, TCA_BPF_MAX, tca[TCA_OPTIONS],
428					  bpf_policy, NULL);
429	if (ret < 0)
430		return ret;
431
432	prog = kzalloc(sizeof(*prog), GFP_KERNEL);
433	if (!prog)
434		return -ENOBUFS;
435
436	ret = tcf_exts_init(&prog->exts, net, TCA_BPF_ACT, TCA_BPF_POLICE);
437	if (ret < 0)
438		goto errout;
439
440	if (oldprog) {
441		if (handle && oldprog->handle != handle) {
442			ret = -EINVAL;
443			goto errout;
444		}
445	}
446
447	if (handle == 0) {
448		handle = 1;
449		ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
450				    INT_MAX, GFP_KERNEL);
451	} else if (!oldprog) {
452		ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
453				    handle, GFP_KERNEL);
454	}
455
456	if (ret)
457		goto errout;
458	prog->handle = handle;
459
460	is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
461	is_ebpf = tb[TCA_BPF_FD];
462	if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
463		ret = -EINVAL;
464		goto errout_idr;
465	}
466
467	ret = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &prog->exts,
468				flags, extack);
469	if (ret < 0)
470		goto errout_idr;
471
472	if (tb[TCA_BPF_FLAGS]) {
473		u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
474
475		if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT) {
476			ret = -EINVAL;
477			goto errout_idr;
478		}
479
480		have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
481	}
482	if (tb[TCA_BPF_FLAGS_GEN]) {
483		gen_flags = nla_get_u32(tb[TCA_BPF_FLAGS_GEN]);
484		if (gen_flags & ~CLS_BPF_SUPPORTED_GEN_FLAGS ||
485		    !tc_flags_valid(gen_flags)) {
486			ret = -EINVAL;
487			goto errout_idr;
488		}
489	}
490
491	prog->exts_integrated = have_exts;
492	prog->gen_flags = gen_flags;
493
494	ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
495		cls_bpf_prog_from_efd(tb, prog, gen_flags, tp);
496	if (ret < 0)
497		goto errout_idr;
498
499	if (tb[TCA_BPF_CLASSID]) {
500		prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
501		tcf_bind_filter(tp, &prog->res, base);
502		bound_to_filter = true;
503	}
504
505	ret = cls_bpf_offload(tp, prog, oldprog, extack);
506	if (ret)
507		goto errout_parms;
508
509	if (!tc_in_hw(prog->gen_flags))
510		prog->gen_flags |= TCA_CLS_FLAGS_NOT_IN_HW;
511
512	if (oldprog) {
513		idr_replace(&head->handle_idr, prog, handle);
514		list_replace_rcu(&oldprog->link, &prog->link);
515		tcf_unbind_filter(tp, &oldprog->res);
516		tcf_exts_get_net(&oldprog->exts);
517		tcf_queue_work(&oldprog->rwork, cls_bpf_delete_prog_work);
518	} else {
519		list_add_rcu(&prog->link, &head->plist);
520	}
521
522	*arg = prog;
523	return 0;
524
525errout_parms:
526	if (bound_to_filter)
527		tcf_unbind_filter(tp, &prog->res);
528	cls_bpf_free_parms(prog);
529errout_idr:
530	if (!oldprog)
531		idr_remove(&head->handle_idr, prog->handle);
532errout:
533	tcf_exts_destroy(&prog->exts);
534	kfree(prog);
535	return ret;
536}
537
538static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
539				 struct sk_buff *skb)
540{
541	struct nlattr *nla;
542
543	if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
544		return -EMSGSIZE;
545
546	nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
547			  sizeof(struct sock_filter));
548	if (nla == NULL)
549		return -EMSGSIZE;
550
551	memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
552
553	return 0;
554}
555
556static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
557				  struct sk_buff *skb)
558{
559	struct nlattr *nla;
560
561	if (prog->bpf_name &&
562	    nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
563		return -EMSGSIZE;
564
565	if (nla_put_u32(skb, TCA_BPF_ID, prog->filter->aux->id))
566		return -EMSGSIZE;
567
568	nla = nla_reserve(skb, TCA_BPF_TAG, sizeof(prog->filter->tag));
569	if (nla == NULL)
570		return -EMSGSIZE;
571
572	memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
573
574	return 0;
575}
576
577static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, void *fh,
578			struct sk_buff *skb, struct tcmsg *tm, bool rtnl_held)
579{
580	struct cls_bpf_prog *prog = fh;
581	struct nlattr *nest;
582	u32 bpf_flags = 0;
583	int ret;
584
585	if (prog == NULL)
586		return skb->len;
587
588	tm->tcm_handle = prog->handle;
589
590	cls_bpf_offload_update_stats(tp, prog);
591
592	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
593	if (nest == NULL)
594		goto nla_put_failure;
595
596	if (prog->res.classid &&
597	    nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
598		goto nla_put_failure;
599
600	if (cls_bpf_is_ebpf(prog))
601		ret = cls_bpf_dump_ebpf_info(prog, skb);
602	else
603		ret = cls_bpf_dump_bpf_info(prog, skb);
604	if (ret)
605		goto nla_put_failure;
606
607	if (tcf_exts_dump(skb, &prog->exts) < 0)
608		goto nla_put_failure;
609
610	if (prog->exts_integrated)
611		bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
612	if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
613		goto nla_put_failure;
614	if (prog->gen_flags &&
615	    nla_put_u32(skb, TCA_BPF_FLAGS_GEN, prog->gen_flags))
616		goto nla_put_failure;
617
618	nla_nest_end(skb, nest);
619
620	if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
621		goto nla_put_failure;
622
623	return skb->len;
624
625nla_put_failure:
626	nla_nest_cancel(skb, nest);
627	return -1;
628}
629
630static void cls_bpf_bind_class(void *fh, u32 classid, unsigned long cl,
631			       void *q, unsigned long base)
632{
633	struct cls_bpf_prog *prog = fh;
634
635	tc_cls_bind_class(classid, cl, q, &prog->res, base);
 
636}
637
638static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg,
639			 bool rtnl_held)
640{
641	struct cls_bpf_head *head = rtnl_dereference(tp->root);
642	struct cls_bpf_prog *prog;
643
644	list_for_each_entry(prog, &head->plist, link) {
645		if (!tc_cls_stats_dump(tp, arg, prog))
 
 
 
646			break;
 
 
 
647	}
648}
649
650static int cls_bpf_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
651			     void *cb_priv, struct netlink_ext_ack *extack)
652{
653	struct cls_bpf_head *head = rtnl_dereference(tp->root);
654	struct tcf_block *block = tp->chain->block;
655	struct tc_cls_bpf_offload cls_bpf = {};
656	struct cls_bpf_prog *prog;
657	int err;
658
659	list_for_each_entry(prog, &head->plist, link) {
660		if (tc_skip_hw(prog->gen_flags))
661			continue;
662
663		tc_cls_common_offload_init(&cls_bpf.common, tp, prog->gen_flags,
664					   extack);
665		cls_bpf.command = TC_CLSBPF_OFFLOAD;
666		cls_bpf.exts = &prog->exts;
667		cls_bpf.prog = add ? prog->filter : NULL;
668		cls_bpf.oldprog = add ? NULL : prog->filter;
669		cls_bpf.name = prog->bpf_name;
670		cls_bpf.exts_integrated = prog->exts_integrated;
671
672		err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSBPF,
673					    &cls_bpf, cb_priv, &prog->gen_flags,
674					    &prog->in_hw_count);
675		if (err)
676			return err;
677	}
678
679	return 0;
680}
681
682static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
683	.kind		=	"bpf",
684	.owner		=	THIS_MODULE,
685	.classify	=	cls_bpf_classify,
686	.init		=	cls_bpf_init,
687	.destroy	=	cls_bpf_destroy,
688	.get		=	cls_bpf_get,
689	.change		=	cls_bpf_change,
690	.delete		=	cls_bpf_delete,
691	.walk		=	cls_bpf_walk,
692	.reoffload	=	cls_bpf_reoffload,
693	.dump		=	cls_bpf_dump,
694	.bind_class	=	cls_bpf_bind_class,
695};
696
697static int __init cls_bpf_init_mod(void)
698{
699	return register_tcf_proto_ops(&cls_bpf_ops);
700}
701
702static void __exit cls_bpf_exit_mod(void)
703{
704	unregister_tcf_proto_ops(&cls_bpf_ops);
705}
706
707module_init(cls_bpf_init_mod);
708module_exit(cls_bpf_exit_mod);