xref: /linux/net/sched/act_api.c (revision 6faadbbb7f9da70ce484f98f72223c20125a1009)
1 /*
2  * net/sched/act_api.c	Packet action API.
3  *
4  *		This program is free software; you can redistribute it and/or
5  *		modify it under the terms of the GNU General Public License
6  *		as published by the Free Software Foundation; either version
7  *		2 of the License, or (at your option) any later version.
8  *
9  * Author:	Jamal Hadi Salim
10  *
11  *
12  */
13 
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/slab.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/kmod.h>
22 #include <linux/err.h>
23 #include <linux/module.h>
24 #include <net/net_namespace.h>
25 #include <net/sock.h>
26 #include <net/sch_generic.h>
27 #include <net/pkt_cls.h>
28 #include <net/act_api.h>
29 #include <net/netlink.h>
30 
31 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
32 {
33 	u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
34 
35 	if (!tp)
36 		return -EINVAL;
37 	a->goto_chain = tcf_chain_get(tp->chain->block, chain_index, true);
38 	if (!a->goto_chain)
39 		return -ENOMEM;
40 	return 0;
41 }
42 
43 static void tcf_action_goto_chain_fini(struct tc_action *a)
44 {
45 	tcf_chain_put(a->goto_chain);
46 }
47 
48 static void tcf_action_goto_chain_exec(const struct tc_action *a,
49 				       struct tcf_result *res)
50 {
51 	const struct tcf_chain *chain = a->goto_chain;
52 
53 	res->goto_tp = rcu_dereference_bh(chain->filter_chain);
54 }
55 
56 static void free_tcf(struct rcu_head *head)
57 {
58 	struct tc_action *p = container_of(head, struct tc_action, tcfa_rcu);
59 
60 	free_percpu(p->cpu_bstats);
61 	free_percpu(p->cpu_qstats);
62 
63 	if (p->act_cookie) {
64 		kfree(p->act_cookie->data);
65 		kfree(p->act_cookie);
66 	}
67 	if (p->goto_chain)
68 		tcf_action_goto_chain_fini(p);
69 
70 	kfree(p);
71 }
72 
73 static void tcf_idr_remove(struct tcf_idrinfo *idrinfo, struct tc_action *p)
74 {
75 	spin_lock_bh(&idrinfo->lock);
76 	idr_remove_ext(&idrinfo->action_idr, p->tcfa_index);
77 	spin_unlock_bh(&idrinfo->lock);
78 	gen_kill_estimator(&p->tcfa_rate_est);
79 	/*
80 	 * gen_estimator est_timer() might access p->tcfa_lock
81 	 * or bstats, wait a RCU grace period before freeing p
82 	 */
83 	call_rcu(&p->tcfa_rcu, free_tcf);
84 }
85 
86 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
87 {
88 	int ret = 0;
89 
90 	if (p) {
91 		if (bind)
92 			p->tcfa_bindcnt--;
93 		else if (strict && p->tcfa_bindcnt > 0)
94 			return -EPERM;
95 
96 		p->tcfa_refcnt--;
97 		if (p->tcfa_bindcnt <= 0 && p->tcfa_refcnt <= 0) {
98 			if (p->ops->cleanup)
99 				p->ops->cleanup(p, bind);
100 			tcf_idr_remove(p->idrinfo, p);
101 			ret = ACT_P_DELETED;
102 		}
103 	}
104 
105 	return ret;
106 }
107 EXPORT_SYMBOL(__tcf_idr_release);
108 
109 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
110 			   struct netlink_callback *cb)
111 {
112 	int err = 0, index = -1, s_i = 0, n_i = 0;
113 	u32 act_flags = cb->args[2];
114 	unsigned long jiffy_since = cb->args[3];
115 	struct nlattr *nest;
116 	struct idr *idr = &idrinfo->action_idr;
117 	struct tc_action *p;
118 	unsigned long id = 1;
119 
120 	spin_lock_bh(&idrinfo->lock);
121 
122 	s_i = cb->args[0];
123 
124 	idr_for_each_entry_ext(idr, p, id) {
125 		index++;
126 		if (index < s_i)
127 			continue;
128 
129 		if (jiffy_since &&
130 		    time_after(jiffy_since,
131 			       (unsigned long)p->tcfa_tm.lastuse))
132 			continue;
133 
134 		nest = nla_nest_start(skb, n_i);
135 		if (!nest)
136 			goto nla_put_failure;
137 		err = tcf_action_dump_1(skb, p, 0, 0);
138 		if (err < 0) {
139 			index--;
140 			nlmsg_trim(skb, nest);
141 			goto done;
142 		}
143 		nla_nest_end(skb, nest);
144 		n_i++;
145 		if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
146 		    n_i >= TCA_ACT_MAX_PRIO)
147 			goto done;
148 	}
149 done:
150 	if (index >= 0)
151 		cb->args[0] = index + 1;
152 
153 	spin_unlock_bh(&idrinfo->lock);
154 	if (n_i) {
155 		if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
156 			cb->args[1] = n_i;
157 	}
158 	return n_i;
159 
160 nla_put_failure:
161 	nla_nest_cancel(skb, nest);
162 	goto done;
163 }
164 
165 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
166 			  const struct tc_action_ops *ops)
167 {
168 	struct nlattr *nest;
169 	int n_i = 0;
170 	int ret = -EINVAL;
171 	struct idr *idr = &idrinfo->action_idr;
172 	struct tc_action *p;
173 	unsigned long id = 1;
174 
175 	nest = nla_nest_start(skb, 0);
176 	if (nest == NULL)
177 		goto nla_put_failure;
178 	if (nla_put_string(skb, TCA_KIND, ops->kind))
179 		goto nla_put_failure;
180 
181 	idr_for_each_entry_ext(idr, p, id) {
182 		ret = __tcf_idr_release(p, false, true);
183 		if (ret == ACT_P_DELETED) {
184 			module_put(p->ops->owner);
185 			n_i++;
186 		} else if (ret < 0) {
187 			goto nla_put_failure;
188 		}
189 	}
190 	if (nla_put_u32(skb, TCA_FCNT, n_i))
191 		goto nla_put_failure;
192 	nla_nest_end(skb, nest);
193 
194 	return n_i;
195 nla_put_failure:
196 	nla_nest_cancel(skb, nest);
197 	return ret;
198 }
199 
200 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
201 		       struct netlink_callback *cb, int type,
202 		       const struct tc_action_ops *ops)
203 {
204 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
205 
206 	if (type == RTM_DELACTION) {
207 		return tcf_del_walker(idrinfo, skb, ops);
208 	} else if (type == RTM_GETACTION) {
209 		return tcf_dump_walker(idrinfo, skb, cb);
210 	} else {
211 		WARN(1, "tcf_generic_walker: unknown action %d\n", type);
212 		return -EINVAL;
213 	}
214 }
215 EXPORT_SYMBOL(tcf_generic_walker);
216 
217 static struct tc_action *tcf_idr_lookup(u32 index, struct tcf_idrinfo *idrinfo)
218 {
219 	struct tc_action *p = NULL;
220 
221 	spin_lock_bh(&idrinfo->lock);
222 	p = idr_find_ext(&idrinfo->action_idr, index);
223 	spin_unlock_bh(&idrinfo->lock);
224 
225 	return p;
226 }
227 
228 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
229 {
230 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
231 	struct tc_action *p = tcf_idr_lookup(index, idrinfo);
232 
233 	if (p) {
234 		*a = p;
235 		return 1;
236 	}
237 	return 0;
238 }
239 EXPORT_SYMBOL(tcf_idr_search);
240 
241 bool tcf_idr_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
242 		   int bind)
243 {
244 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
245 	struct tc_action *p = tcf_idr_lookup(index, idrinfo);
246 
247 	if (index && p) {
248 		if (bind)
249 			p->tcfa_bindcnt++;
250 		p->tcfa_refcnt++;
251 		*a = p;
252 		return true;
253 	}
254 	return false;
255 }
256 EXPORT_SYMBOL(tcf_idr_check);
257 
258 void tcf_idr_cleanup(struct tc_action *a, struct nlattr *est)
259 {
260 	if (est)
261 		gen_kill_estimator(&a->tcfa_rate_est);
262 	call_rcu(&a->tcfa_rcu, free_tcf);
263 }
264 EXPORT_SYMBOL(tcf_idr_cleanup);
265 
266 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
267 		   struct tc_action **a, const struct tc_action_ops *ops,
268 		   int bind, bool cpustats)
269 {
270 	struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
271 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
272 	struct idr *idr = &idrinfo->action_idr;
273 	int err = -ENOMEM;
274 	unsigned long idr_index;
275 
276 	if (unlikely(!p))
277 		return -ENOMEM;
278 	p->tcfa_refcnt = 1;
279 	if (bind)
280 		p->tcfa_bindcnt = 1;
281 
282 	if (cpustats) {
283 		p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
284 		if (!p->cpu_bstats) {
285 err1:
286 			kfree(p);
287 			return err;
288 		}
289 		p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
290 		if (!p->cpu_qstats) {
291 err2:
292 			free_percpu(p->cpu_bstats);
293 			goto err1;
294 		}
295 	}
296 	spin_lock_init(&p->tcfa_lock);
297 	/* user doesn't specify an index */
298 	if (!index) {
299 		idr_preload(GFP_KERNEL);
300 		spin_lock_bh(&idrinfo->lock);
301 		err = idr_alloc_ext(idr, NULL, &idr_index, 1, 0,
302 				    GFP_ATOMIC);
303 		spin_unlock_bh(&idrinfo->lock);
304 		idr_preload_end();
305 		if (err) {
306 err3:
307 			free_percpu(p->cpu_qstats);
308 			goto err2;
309 		}
310 		p->tcfa_index = idr_index;
311 	} else {
312 		idr_preload(GFP_KERNEL);
313 		spin_lock_bh(&idrinfo->lock);
314 		err = idr_alloc_ext(idr, NULL, NULL, index, index + 1,
315 				    GFP_ATOMIC);
316 		spin_unlock_bh(&idrinfo->lock);
317 		idr_preload_end();
318 		if (err)
319 			goto err3;
320 		p->tcfa_index = index;
321 	}
322 
323 	p->tcfa_tm.install = jiffies;
324 	p->tcfa_tm.lastuse = jiffies;
325 	p->tcfa_tm.firstuse = 0;
326 	if (est) {
327 		err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
328 					&p->tcfa_rate_est,
329 					&p->tcfa_lock, NULL, est);
330 		if (err) {
331 			goto err3;
332 		}
333 	}
334 
335 	p->idrinfo = idrinfo;
336 	p->ops = ops;
337 	INIT_LIST_HEAD(&p->list);
338 	*a = p;
339 	return 0;
340 }
341 EXPORT_SYMBOL(tcf_idr_create);
342 
343 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
344 {
345 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
346 
347 	spin_lock_bh(&idrinfo->lock);
348 	idr_replace_ext(&idrinfo->action_idr, a, a->tcfa_index);
349 	spin_unlock_bh(&idrinfo->lock);
350 }
351 EXPORT_SYMBOL(tcf_idr_insert);
352 
353 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
354 			 struct tcf_idrinfo *idrinfo)
355 {
356 	struct idr *idr = &idrinfo->action_idr;
357 	struct tc_action *p;
358 	int ret;
359 	unsigned long id = 1;
360 
361 	idr_for_each_entry_ext(idr, p, id) {
362 		ret = __tcf_idr_release(p, false, true);
363 		if (ret == ACT_P_DELETED)
364 			module_put(ops->owner);
365 		else if (ret < 0)
366 			return;
367 	}
368 	idr_destroy(&idrinfo->action_idr);
369 }
370 EXPORT_SYMBOL(tcf_idrinfo_destroy);
371 
372 static LIST_HEAD(act_base);
373 static DEFINE_RWLOCK(act_mod_lock);
374 
375 int tcf_register_action(struct tc_action_ops *act,
376 			struct pernet_operations *ops)
377 {
378 	struct tc_action_ops *a;
379 	int ret;
380 
381 	if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
382 		return -EINVAL;
383 
384 	/* We have to register pernet ops before making the action ops visible,
385 	 * otherwise tcf_action_init_1() could get a partially initialized
386 	 * netns.
387 	 */
388 	ret = register_pernet_subsys(ops);
389 	if (ret)
390 		return ret;
391 
392 	write_lock(&act_mod_lock);
393 	list_for_each_entry(a, &act_base, head) {
394 		if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
395 			write_unlock(&act_mod_lock);
396 			unregister_pernet_subsys(ops);
397 			return -EEXIST;
398 		}
399 	}
400 	list_add_tail(&act->head, &act_base);
401 	write_unlock(&act_mod_lock);
402 
403 	return 0;
404 }
405 EXPORT_SYMBOL(tcf_register_action);
406 
407 int tcf_unregister_action(struct tc_action_ops *act,
408 			  struct pernet_operations *ops)
409 {
410 	struct tc_action_ops *a;
411 	int err = -ENOENT;
412 
413 	write_lock(&act_mod_lock);
414 	list_for_each_entry(a, &act_base, head) {
415 		if (a == act) {
416 			list_del(&act->head);
417 			err = 0;
418 			break;
419 		}
420 	}
421 	write_unlock(&act_mod_lock);
422 	if (!err)
423 		unregister_pernet_subsys(ops);
424 	return err;
425 }
426 EXPORT_SYMBOL(tcf_unregister_action);
427 
428 /* lookup by name */
429 static struct tc_action_ops *tc_lookup_action_n(char *kind)
430 {
431 	struct tc_action_ops *a, *res = NULL;
432 
433 	if (kind) {
434 		read_lock(&act_mod_lock);
435 		list_for_each_entry(a, &act_base, head) {
436 			if (strcmp(kind, a->kind) == 0) {
437 				if (try_module_get(a->owner))
438 					res = a;
439 				break;
440 			}
441 		}
442 		read_unlock(&act_mod_lock);
443 	}
444 	return res;
445 }
446 
447 /* lookup by nlattr */
448 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
449 {
450 	struct tc_action_ops *a, *res = NULL;
451 
452 	if (kind) {
453 		read_lock(&act_mod_lock);
454 		list_for_each_entry(a, &act_base, head) {
455 			if (nla_strcmp(kind, a->kind) == 0) {
456 				if (try_module_get(a->owner))
457 					res = a;
458 				break;
459 			}
460 		}
461 		read_unlock(&act_mod_lock);
462 	}
463 	return res;
464 }
465 
466 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
467 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
468 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
469 		    int nr_actions, struct tcf_result *res)
470 {
471 	u32 jmp_prgcnt = 0;
472 	u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
473 	int i;
474 	int ret = TC_ACT_OK;
475 
476 	if (skb_skip_tc_classify(skb))
477 		return TC_ACT_OK;
478 
479 restart_act_graph:
480 	for (i = 0; i < nr_actions; i++) {
481 		const struct tc_action *a = actions[i];
482 
483 		if (jmp_prgcnt > 0) {
484 			jmp_prgcnt -= 1;
485 			continue;
486 		}
487 repeat:
488 		ret = a->ops->act(skb, a, res);
489 		if (ret == TC_ACT_REPEAT)
490 			goto repeat;	/* we need a ttl - JHS */
491 
492 		if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
493 			jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
494 			if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
495 				/* faulty opcode, stop pipeline */
496 				return TC_ACT_OK;
497 			} else {
498 				jmp_ttl -= 1;
499 				if (jmp_ttl > 0)
500 					goto restart_act_graph;
501 				else /* faulty graph, stop pipeline */
502 					return TC_ACT_OK;
503 			}
504 		} else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
505 			tcf_action_goto_chain_exec(a, res);
506 		}
507 
508 		if (ret != TC_ACT_PIPE)
509 			break;
510 	}
511 
512 	return ret;
513 }
514 EXPORT_SYMBOL(tcf_action_exec);
515 
516 int tcf_action_destroy(struct list_head *actions, int bind)
517 {
518 	struct tc_action *a, *tmp;
519 	int ret = 0;
520 
521 	list_for_each_entry_safe(a, tmp, actions, list) {
522 		ret = __tcf_idr_release(a, bind, true);
523 		if (ret == ACT_P_DELETED)
524 			module_put(a->ops->owner);
525 		else if (ret < 0)
526 			return ret;
527 	}
528 	return ret;
529 }
530 
531 int
532 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
533 {
534 	return a->ops->dump(skb, a, bind, ref);
535 }
536 
537 int
538 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
539 {
540 	int err = -EINVAL;
541 	unsigned char *b = skb_tail_pointer(skb);
542 	struct nlattr *nest;
543 
544 	if (nla_put_string(skb, TCA_KIND, a->ops->kind))
545 		goto nla_put_failure;
546 	if (tcf_action_copy_stats(skb, a, 0))
547 		goto nla_put_failure;
548 	if (a->act_cookie) {
549 		if (nla_put(skb, TCA_ACT_COOKIE, a->act_cookie->len,
550 			    a->act_cookie->data))
551 			goto nla_put_failure;
552 	}
553 
554 	nest = nla_nest_start(skb, TCA_OPTIONS);
555 	if (nest == NULL)
556 		goto nla_put_failure;
557 	err = tcf_action_dump_old(skb, a, bind, ref);
558 	if (err > 0) {
559 		nla_nest_end(skb, nest);
560 		return err;
561 	}
562 
563 nla_put_failure:
564 	nlmsg_trim(skb, b);
565 	return -1;
566 }
567 EXPORT_SYMBOL(tcf_action_dump_1);
568 
569 int tcf_action_dump(struct sk_buff *skb, struct list_head *actions,
570 		    int bind, int ref)
571 {
572 	struct tc_action *a;
573 	int err = -EINVAL;
574 	struct nlattr *nest;
575 
576 	list_for_each_entry(a, actions, list) {
577 		nest = nla_nest_start(skb, a->order);
578 		if (nest == NULL)
579 			goto nla_put_failure;
580 		err = tcf_action_dump_1(skb, a, bind, ref);
581 		if (err < 0)
582 			goto errout;
583 		nla_nest_end(skb, nest);
584 	}
585 
586 	return 0;
587 
588 nla_put_failure:
589 	err = -EINVAL;
590 errout:
591 	nla_nest_cancel(skb, nest);
592 	return err;
593 }
594 
595 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
596 {
597 	struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
598 	if (!c)
599 		return NULL;
600 
601 	c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
602 	if (!c->data) {
603 		kfree(c);
604 		return NULL;
605 	}
606 	c->len = nla_len(tb[TCA_ACT_COOKIE]);
607 
608 	return c;
609 }
610 
611 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
612 				    struct nlattr *nla, struct nlattr *est,
613 				    char *name, int ovr, int bind)
614 {
615 	struct tc_action *a;
616 	struct tc_action_ops *a_o;
617 	struct tc_cookie *cookie = NULL;
618 	char act_name[IFNAMSIZ];
619 	struct nlattr *tb[TCA_ACT_MAX + 1];
620 	struct nlattr *kind;
621 	int err;
622 
623 	if (name == NULL) {
624 		err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
625 		if (err < 0)
626 			goto err_out;
627 		err = -EINVAL;
628 		kind = tb[TCA_ACT_KIND];
629 		if (kind == NULL)
630 			goto err_out;
631 		if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
632 			goto err_out;
633 		if (tb[TCA_ACT_COOKIE]) {
634 			int cklen = nla_len(tb[TCA_ACT_COOKIE]);
635 
636 			if (cklen > TC_COOKIE_MAX_SIZE)
637 				goto err_out;
638 
639 			cookie = nla_memdup_cookie(tb);
640 			if (!cookie) {
641 				err = -ENOMEM;
642 				goto err_out;
643 			}
644 		}
645 	} else {
646 		err = -EINVAL;
647 		if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
648 			goto err_out;
649 	}
650 
651 	a_o = tc_lookup_action_n(act_name);
652 	if (a_o == NULL) {
653 #ifdef CONFIG_MODULES
654 		rtnl_unlock();
655 		request_module("act_%s", act_name);
656 		rtnl_lock();
657 
658 		a_o = tc_lookup_action_n(act_name);
659 
660 		/* We dropped the RTNL semaphore in order to
661 		 * perform the module load.  So, even if we
662 		 * succeeded in loading the module we have to
663 		 * tell the caller to replay the request.  We
664 		 * indicate this using -EAGAIN.
665 		 */
666 		if (a_o != NULL) {
667 			err = -EAGAIN;
668 			goto err_mod;
669 		}
670 #endif
671 		err = -ENOENT;
672 		goto err_out;
673 	}
674 
675 	/* backward compatibility for policer */
676 	if (name == NULL)
677 		err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind);
678 	else
679 		err = a_o->init(net, nla, est, &a, ovr, bind);
680 	if (err < 0)
681 		goto err_mod;
682 
683 	if (name == NULL && tb[TCA_ACT_COOKIE]) {
684 		if (a->act_cookie) {
685 			kfree(a->act_cookie->data);
686 			kfree(a->act_cookie);
687 		}
688 		a->act_cookie = cookie;
689 	}
690 
691 	/* module count goes up only when brand new policy is created
692 	 * if it exists and is only bound to in a_o->init() then
693 	 * ACT_P_CREATED is not returned (a zero is).
694 	 */
695 	if (err != ACT_P_CREATED)
696 		module_put(a_o->owner);
697 
698 	if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
699 		err = tcf_action_goto_chain_init(a, tp);
700 		if (err) {
701 			LIST_HEAD(actions);
702 
703 			list_add_tail(&a->list, &actions);
704 			tcf_action_destroy(&actions, bind);
705 			return ERR_PTR(err);
706 		}
707 	}
708 
709 	return a;
710 
711 err_mod:
712 	module_put(a_o->owner);
713 err_out:
714 	if (cookie) {
715 		kfree(cookie->data);
716 		kfree(cookie);
717 	}
718 	return ERR_PTR(err);
719 }
720 
721 static void cleanup_a(struct list_head *actions, int ovr)
722 {
723 	struct tc_action *a;
724 
725 	if (!ovr)
726 		return;
727 
728 	list_for_each_entry(a, actions, list)
729 		a->tcfa_refcnt--;
730 }
731 
732 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
733 		    struct nlattr *est, char *name, int ovr, int bind,
734 		    struct list_head *actions)
735 {
736 	struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
737 	struct tc_action *act;
738 	int err;
739 	int i;
740 
741 	err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, NULL);
742 	if (err < 0)
743 		return err;
744 
745 	for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
746 		act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind);
747 		if (IS_ERR(act)) {
748 			err = PTR_ERR(act);
749 			goto err;
750 		}
751 		act->order = i;
752 		if (ovr)
753 			act->tcfa_refcnt++;
754 		list_add_tail(&act->list, actions);
755 	}
756 
757 	/* Remove the temp refcnt which was necessary to protect against
758 	 * destroying an existing action which was being replaced
759 	 */
760 	cleanup_a(actions, ovr);
761 	return 0;
762 
763 err:
764 	tcf_action_destroy(actions, bind);
765 	return err;
766 }
767 
768 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
769 			  int compat_mode)
770 {
771 	int err = 0;
772 	struct gnet_dump d;
773 
774 	if (p == NULL)
775 		goto errout;
776 
777 	/* compat_mode being true specifies a call that is supposed
778 	 * to add additional backward compatibility statistic TLVs.
779 	 */
780 	if (compat_mode) {
781 		if (p->type == TCA_OLD_COMPAT)
782 			err = gnet_stats_start_copy_compat(skb, 0,
783 							   TCA_STATS,
784 							   TCA_XSTATS,
785 							   &p->tcfa_lock, &d,
786 							   TCA_PAD);
787 		else
788 			return 0;
789 	} else
790 		err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
791 					    &p->tcfa_lock, &d, TCA_ACT_PAD);
792 
793 	if (err < 0)
794 		goto errout;
795 
796 	if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
797 	    gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
798 	    gnet_stats_copy_queue(&d, p->cpu_qstats,
799 				  &p->tcfa_qstats,
800 				  p->tcfa_qstats.qlen) < 0)
801 		goto errout;
802 
803 	if (gnet_stats_finish_copy(&d) < 0)
804 		goto errout;
805 
806 	return 0;
807 
808 errout:
809 	return -1;
810 }
811 
812 static int tca_get_fill(struct sk_buff *skb, struct list_head *actions,
813 			u32 portid, u32 seq, u16 flags, int event, int bind,
814 			int ref)
815 {
816 	struct tcamsg *t;
817 	struct nlmsghdr *nlh;
818 	unsigned char *b = skb_tail_pointer(skb);
819 	struct nlattr *nest;
820 
821 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
822 	if (!nlh)
823 		goto out_nlmsg_trim;
824 	t = nlmsg_data(nlh);
825 	t->tca_family = AF_UNSPEC;
826 	t->tca__pad1 = 0;
827 	t->tca__pad2 = 0;
828 
829 	nest = nla_nest_start(skb, TCA_ACT_TAB);
830 	if (nest == NULL)
831 		goto out_nlmsg_trim;
832 
833 	if (tcf_action_dump(skb, actions, bind, ref) < 0)
834 		goto out_nlmsg_trim;
835 
836 	nla_nest_end(skb, nest);
837 
838 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
839 	return skb->len;
840 
841 out_nlmsg_trim:
842 	nlmsg_trim(skb, b);
843 	return -1;
844 }
845 
846 static int
847 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
848 	       struct list_head *actions, int event)
849 {
850 	struct sk_buff *skb;
851 
852 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
853 	if (!skb)
854 		return -ENOBUFS;
855 	if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
856 			 0, 0) <= 0) {
857 		kfree_skb(skb);
858 		return -EINVAL;
859 	}
860 
861 	return rtnl_unicast(skb, net, portid);
862 }
863 
864 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
865 					  struct nlmsghdr *n, u32 portid)
866 {
867 	struct nlattr *tb[TCA_ACT_MAX + 1];
868 	const struct tc_action_ops *ops;
869 	struct tc_action *a;
870 	int index;
871 	int err;
872 
873 	err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
874 	if (err < 0)
875 		goto err_out;
876 
877 	err = -EINVAL;
878 	if (tb[TCA_ACT_INDEX] == NULL ||
879 	    nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
880 		goto err_out;
881 	index = nla_get_u32(tb[TCA_ACT_INDEX]);
882 
883 	err = -EINVAL;
884 	ops = tc_lookup_action(tb[TCA_ACT_KIND]);
885 	if (!ops) /* could happen in batch of actions */
886 		goto err_out;
887 	err = -ENOENT;
888 	if (ops->lookup(net, &a, index) == 0)
889 		goto err_mod;
890 
891 	module_put(ops->owner);
892 	return a;
893 
894 err_mod:
895 	module_put(ops->owner);
896 err_out:
897 	return ERR_PTR(err);
898 }
899 
900 static int tca_action_flush(struct net *net, struct nlattr *nla,
901 			    struct nlmsghdr *n, u32 portid)
902 {
903 	struct sk_buff *skb;
904 	unsigned char *b;
905 	struct nlmsghdr *nlh;
906 	struct tcamsg *t;
907 	struct netlink_callback dcb;
908 	struct nlattr *nest;
909 	struct nlattr *tb[TCA_ACT_MAX + 1];
910 	const struct tc_action_ops *ops;
911 	struct nlattr *kind;
912 	int err = -ENOMEM;
913 
914 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
915 	if (!skb) {
916 		pr_debug("tca_action_flush: failed skb alloc\n");
917 		return err;
918 	}
919 
920 	b = skb_tail_pointer(skb);
921 
922 	err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
923 	if (err < 0)
924 		goto err_out;
925 
926 	err = -EINVAL;
927 	kind = tb[TCA_ACT_KIND];
928 	ops = tc_lookup_action(kind);
929 	if (!ops) /*some idjot trying to flush unknown action */
930 		goto err_out;
931 
932 	nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
933 			sizeof(*t), 0);
934 	if (!nlh)
935 		goto out_module_put;
936 	t = nlmsg_data(nlh);
937 	t->tca_family = AF_UNSPEC;
938 	t->tca__pad1 = 0;
939 	t->tca__pad2 = 0;
940 
941 	nest = nla_nest_start(skb, TCA_ACT_TAB);
942 	if (nest == NULL)
943 		goto out_module_put;
944 
945 	err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops);
946 	if (err <= 0)
947 		goto out_module_put;
948 
949 	nla_nest_end(skb, nest);
950 
951 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
952 	nlh->nlmsg_flags |= NLM_F_ROOT;
953 	module_put(ops->owner);
954 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
955 			     n->nlmsg_flags & NLM_F_ECHO);
956 	if (err > 0)
957 		return 0;
958 
959 	return err;
960 
961 out_module_put:
962 	module_put(ops->owner);
963 err_out:
964 	kfree_skb(skb);
965 	return err;
966 }
967 
968 static int
969 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
970 	       u32 portid)
971 {
972 	int ret;
973 	struct sk_buff *skb;
974 
975 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
976 	if (!skb)
977 		return -ENOBUFS;
978 
979 	if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
980 			 0, 1) <= 0) {
981 		kfree_skb(skb);
982 		return -EINVAL;
983 	}
984 
985 	/* now do the delete */
986 	ret = tcf_action_destroy(actions, 0);
987 	if (ret < 0) {
988 		kfree_skb(skb);
989 		return ret;
990 	}
991 
992 	ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
993 			     n->nlmsg_flags & NLM_F_ECHO);
994 	if (ret > 0)
995 		return 0;
996 	return ret;
997 }
998 
999 static int
1000 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1001 	      u32 portid, int event)
1002 {
1003 	int i, ret;
1004 	struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1005 	struct tc_action *act;
1006 	LIST_HEAD(actions);
1007 
1008 	ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, NULL);
1009 	if (ret < 0)
1010 		return ret;
1011 
1012 	if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1013 		if (tb[1] != NULL)
1014 			return tca_action_flush(net, tb[1], n, portid);
1015 		else
1016 			return -EINVAL;
1017 	}
1018 
1019 	for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1020 		act = tcf_action_get_1(net, tb[i], n, portid);
1021 		if (IS_ERR(act)) {
1022 			ret = PTR_ERR(act);
1023 			goto err;
1024 		}
1025 		act->order = i;
1026 		list_add_tail(&act->list, &actions);
1027 	}
1028 
1029 	if (event == RTM_GETACTION)
1030 		ret = tcf_get_notify(net, portid, n, &actions, event);
1031 	else { /* delete */
1032 		ret = tcf_del_notify(net, n, &actions, portid);
1033 		if (ret)
1034 			goto err;
1035 		return ret;
1036 	}
1037 err:
1038 	if (event != RTM_GETACTION)
1039 		tcf_action_destroy(&actions, 0);
1040 	return ret;
1041 }
1042 
1043 static int
1044 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
1045 	       u32 portid)
1046 {
1047 	struct sk_buff *skb;
1048 	int err = 0;
1049 
1050 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1051 	if (!skb)
1052 		return -ENOBUFS;
1053 
1054 	if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1055 			 RTM_NEWACTION, 0, 0) <= 0) {
1056 		kfree_skb(skb);
1057 		return -EINVAL;
1058 	}
1059 
1060 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1061 			     n->nlmsg_flags & NLM_F_ECHO);
1062 	if (err > 0)
1063 		err = 0;
1064 	return err;
1065 }
1066 
1067 static int tcf_action_add(struct net *net, struct nlattr *nla,
1068 			  struct nlmsghdr *n, u32 portid, int ovr)
1069 {
1070 	int ret = 0;
1071 	LIST_HEAD(actions);
1072 
1073 	ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, &actions);
1074 	if (ret)
1075 		return ret;
1076 
1077 	return tcf_add_notify(net, n, &actions, portid);
1078 }
1079 
1080 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1081 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1082 	[TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1083 			     .validation_data = &tcaa_root_flags_allowed },
1084 	[TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1085 };
1086 
1087 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1088 			 struct netlink_ext_ack *extack)
1089 {
1090 	struct net *net = sock_net(skb->sk);
1091 	struct nlattr *tca[TCA_ROOT_MAX + 1];
1092 	u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1093 	int ret = 0, ovr = 0;
1094 
1095 	if ((n->nlmsg_type != RTM_GETACTION) &&
1096 	    !netlink_capable(skb, CAP_NET_ADMIN))
1097 		return -EPERM;
1098 
1099 	ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1100 			  extack);
1101 	if (ret < 0)
1102 		return ret;
1103 
1104 	if (tca[TCA_ACT_TAB] == NULL) {
1105 		pr_notice("tc_ctl_action: received NO action attribs\n");
1106 		return -EINVAL;
1107 	}
1108 
1109 	/* n->nlmsg_flags & NLM_F_CREATE */
1110 	switch (n->nlmsg_type) {
1111 	case RTM_NEWACTION:
1112 		/* we are going to assume all other flags
1113 		 * imply create only if it doesn't exist
1114 		 * Note that CREATE | EXCL implies that
1115 		 * but since we want avoid ambiguity (eg when flags
1116 		 * is zero) then just set this
1117 		 */
1118 		if (n->nlmsg_flags & NLM_F_REPLACE)
1119 			ovr = 1;
1120 replay:
1121 		ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr);
1122 		if (ret == -EAGAIN)
1123 			goto replay;
1124 		break;
1125 	case RTM_DELACTION:
1126 		ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1127 				    portid, RTM_DELACTION);
1128 		break;
1129 	case RTM_GETACTION:
1130 		ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1131 				    portid, RTM_GETACTION);
1132 		break;
1133 	default:
1134 		BUG();
1135 	}
1136 
1137 	return ret;
1138 }
1139 
1140 static struct nlattr *find_dump_kind(struct nlattr **nla)
1141 {
1142 	struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1143 	struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1144 	struct nlattr *kind;
1145 
1146 	tb1 = nla[TCA_ACT_TAB];
1147 	if (tb1 == NULL)
1148 		return NULL;
1149 
1150 	if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1151 		      NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1152 		return NULL;
1153 
1154 	if (tb[1] == NULL)
1155 		return NULL;
1156 	if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1157 		return NULL;
1158 	kind = tb2[TCA_ACT_KIND];
1159 
1160 	return kind;
1161 }
1162 
1163 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1164 {
1165 	struct net *net = sock_net(skb->sk);
1166 	struct nlmsghdr *nlh;
1167 	unsigned char *b = skb_tail_pointer(skb);
1168 	struct nlattr *nest;
1169 	struct tc_action_ops *a_o;
1170 	int ret = 0;
1171 	struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1172 	struct nlattr *tb[TCA_ROOT_MAX + 1];
1173 	struct nlattr *count_attr = NULL;
1174 	unsigned long jiffy_since = 0;
1175 	struct nlattr *kind = NULL;
1176 	struct nla_bitfield32 bf;
1177 	u32 msecs_since = 0;
1178 	u32 act_count = 0;
1179 
1180 	ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1181 			  tcaa_policy, NULL);
1182 	if (ret < 0)
1183 		return ret;
1184 
1185 	kind = find_dump_kind(tb);
1186 	if (kind == NULL) {
1187 		pr_info("tc_dump_action: action bad kind\n");
1188 		return 0;
1189 	}
1190 
1191 	a_o = tc_lookup_action(kind);
1192 	if (a_o == NULL)
1193 		return 0;
1194 
1195 	cb->args[2] = 0;
1196 	if (tb[TCA_ROOT_FLAGS]) {
1197 		bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1198 		cb->args[2] = bf.value;
1199 	}
1200 
1201 	if (tb[TCA_ROOT_TIME_DELTA]) {
1202 		msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1203 	}
1204 
1205 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1206 			cb->nlh->nlmsg_type, sizeof(*t), 0);
1207 	if (!nlh)
1208 		goto out_module_put;
1209 
1210 	if (msecs_since)
1211 		jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1212 
1213 	t = nlmsg_data(nlh);
1214 	t->tca_family = AF_UNSPEC;
1215 	t->tca__pad1 = 0;
1216 	t->tca__pad2 = 0;
1217 	cb->args[3] = jiffy_since;
1218 	count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1219 	if (!count_attr)
1220 		goto out_module_put;
1221 
1222 	nest = nla_nest_start(skb, TCA_ACT_TAB);
1223 	if (nest == NULL)
1224 		goto out_module_put;
1225 
1226 	ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o);
1227 	if (ret < 0)
1228 		goto out_module_put;
1229 
1230 	if (ret > 0) {
1231 		nla_nest_end(skb, nest);
1232 		ret = skb->len;
1233 		act_count = cb->args[1];
1234 		memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1235 		cb->args[1] = 0;
1236 	} else
1237 		nlmsg_trim(skb, b);
1238 
1239 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1240 	if (NETLINK_CB(cb->skb).portid && ret)
1241 		nlh->nlmsg_flags |= NLM_F_MULTI;
1242 	module_put(a_o->owner);
1243 	return skb->len;
1244 
1245 out_module_put:
1246 	module_put(a_o->owner);
1247 	nlmsg_trim(skb, b);
1248 	return skb->len;
1249 }
1250 
1251 static int __init tc_action_init(void)
1252 {
1253 	rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1254 	rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1255 	rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1256 		      0);
1257 
1258 	return 0;
1259 }
1260 
1261 subsys_initcall(tc_action_init);
1262