xref: /linux/net/sched/act_api.c (revision 478eb5abb51931a152abab068f8a717b7ff480fd)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/act_api.c	Packet action API.
4  *
5  * Author:	Jamal Hadi Salim
6  */
7 
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/string.h>
11 #include <linux/errno.h>
12 #include <linux/slab.h>
13 #include <linux/skbuff.h>
14 #include <linux/init.h>
15 #include <linux/kmod.h>
16 #include <linux/err.h>
17 #include <linux/module.h>
18 #include <net/net_namespace.h>
19 #include <net/sock.h>
20 #include <net/sch_generic.h>
21 #include <net/pkt_cls.h>
22 #include <net/tc_act/tc_pedit.h>
23 #include <net/act_api.h>
24 #include <net/netlink.h>
25 #include <net/flow_offload.h>
26 #include <net/tc_wrapper.h>
27 
28 #ifdef CONFIG_INET
29 DEFINE_STATIC_KEY_FALSE(tcf_frag_xmit_count);
30 EXPORT_SYMBOL_GPL(tcf_frag_xmit_count);
31 #endif
32 
33 int tcf_dev_queue_xmit(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb))
34 {
35 #ifdef CONFIG_INET
36 	if (static_branch_unlikely(&tcf_frag_xmit_count))
37 		return sch_frag_xmit_hook(skb, xmit);
38 #endif
39 
40 	return xmit(skb);
41 }
42 EXPORT_SYMBOL_GPL(tcf_dev_queue_xmit);
43 
44 static void tcf_action_goto_chain_exec(const struct tcf_chain *chain,
45 				       struct tcf_result *res)
46 {
47 	res->goto_tp = rcu_dereference_bh(chain->filter_chain);
48 }
49 
50 static void tcf_free_cookie_rcu(struct rcu_head *p)
51 {
52 	struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
53 
54 	kfree(cookie->data);
55 	kfree(cookie);
56 }
57 
58 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
59 				  struct tc_cookie *new_cookie)
60 {
61 	struct tc_cookie *old;
62 
63 	old = unrcu_pointer(xchg(old_cookie, RCU_INITIALIZER(new_cookie)));
64 	if (old)
65 		call_rcu(&old->rcu, tcf_free_cookie_rcu);
66 }
67 
68 int tcf_action_check_ctrlact(int action, struct tcf_proto *tp,
69 			     struct tcf_chain **newchain,
70 			     struct netlink_ext_ack *extack)
71 {
72 	int opcode = TC_ACT_EXT_OPCODE(action), ret = -EINVAL;
73 	u32 chain_index;
74 
75 	if (!opcode)
76 		ret = action > TC_ACT_VALUE_MAX ? -EINVAL : 0;
77 	else if (opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC)
78 		ret = 0;
79 	if (ret) {
80 		NL_SET_ERR_MSG(extack, "invalid control action");
81 		goto end;
82 	}
83 
84 	if (TC_ACT_EXT_CMP(action, TC_ACT_GOTO_CHAIN)) {
85 		chain_index = action & TC_ACT_EXT_VAL_MASK;
86 		if (!tp || !newchain) {
87 			ret = -EINVAL;
88 			NL_SET_ERR_MSG(extack,
89 				       "can't goto NULL proto/chain");
90 			goto end;
91 		}
92 		*newchain = tcf_chain_get_by_act(tp->chain->block, chain_index);
93 		if (!*newchain) {
94 			ret = -ENOMEM;
95 			NL_SET_ERR_MSG(extack,
96 				       "can't allocate goto_chain");
97 		}
98 	}
99 end:
100 	return ret;
101 }
102 EXPORT_SYMBOL(tcf_action_check_ctrlact);
103 
104 struct tcf_chain *tcf_action_set_ctrlact(struct tc_action *a, int action,
105 					 struct tcf_chain *goto_chain)
106 {
107 	a->tcfa_action = action;
108 	goto_chain = rcu_replace_pointer(a->goto_chain, goto_chain, 1);
109 	return goto_chain;
110 }
111 EXPORT_SYMBOL(tcf_action_set_ctrlact);
112 
113 static void free_tcf(struct tc_action *p)
114 {
115 	struct tcf_chain *chain = rcu_dereference_protected(p->goto_chain, 1);
116 
117 	free_percpu(p->cpu_bstats);
118 	free_percpu(p->cpu_bstats_hw);
119 	free_percpu(p->cpu_qstats);
120 
121 	tcf_set_action_cookie(&p->user_cookie, NULL);
122 	if (chain)
123 		tcf_chain_put_by_act(chain);
124 
125 	kfree_rcu(p, tcfa_rcu);
126 }
127 
128 static void offload_action_hw_count_set(struct tc_action *act,
129 					u32 hw_count)
130 {
131 	act->in_hw_count = hw_count;
132 }
133 
134 static void offload_action_hw_count_inc(struct tc_action *act,
135 					u32 hw_count)
136 {
137 	act->in_hw_count += hw_count;
138 }
139 
140 static void offload_action_hw_count_dec(struct tc_action *act,
141 					u32 hw_count)
142 {
143 	act->in_hw_count = act->in_hw_count > hw_count ?
144 			   act->in_hw_count - hw_count : 0;
145 }
146 
147 static unsigned int tcf_offload_act_num_actions_single(struct tc_action *act)
148 {
149 	unsigned int count;
150 
151 	if (is_tcf_pedit(act)) {
152 		spin_lock_bh(&act->tcfa_lock);
153 		count = tcf_pedit_nkeys_locked(act);
154 		spin_unlock_bh(&act->tcfa_lock);
155 		return count;
156 	}
157 	return 1;
158 }
159 
160 static bool tc_act_skip_hw(u32 flags)
161 {
162 	return (flags & TCA_ACT_FLAGS_SKIP_HW) ? true : false;
163 }
164 
165 static bool tc_act_skip_sw(u32 flags)
166 {
167 	return (flags & TCA_ACT_FLAGS_SKIP_SW) ? true : false;
168 }
169 
170 /* SKIP_HW and SKIP_SW are mutually exclusive flags. */
171 static bool tc_act_flags_valid(u32 flags)
172 {
173 	flags &= TCA_ACT_FLAGS_SKIP_HW | TCA_ACT_FLAGS_SKIP_SW;
174 
175 	return flags ^ (TCA_ACT_FLAGS_SKIP_HW | TCA_ACT_FLAGS_SKIP_SW);
176 }
177 
178 static int offload_action_init(struct flow_offload_action *fl_action,
179 			       struct tc_action *act,
180 			       enum offload_act_command  cmd,
181 			       struct netlink_ext_ack *extack)
182 {
183 	int err;
184 
185 	fl_action->extack = extack;
186 	fl_action->command = cmd;
187 	fl_action->index = act->tcfa_index;
188 	fl_action->cookie = (unsigned long)act;
189 
190 	if (act->ops->offload_act_setup) {
191 		spin_lock_bh(&act->tcfa_lock);
192 		err = act->ops->offload_act_setup(act, fl_action, NULL,
193 						  false, extack);
194 		spin_unlock_bh(&act->tcfa_lock);
195 		return err;
196 	}
197 
198 	return -EOPNOTSUPP;
199 }
200 
201 static int tcf_action_offload_cmd_ex(struct flow_offload_action *fl_act,
202 				     u32 *hw_count)
203 {
204 	int err;
205 
206 	err = flow_indr_dev_setup_offload(NULL, NULL, TC_SETUP_ACT,
207 					  fl_act, NULL, NULL);
208 	if (err < 0)
209 		return err;
210 
211 	if (hw_count)
212 		*hw_count = err;
213 
214 	return 0;
215 }
216 
217 static int tcf_action_offload_cmd_cb_ex(struct flow_offload_action *fl_act,
218 					u32 *hw_count,
219 					flow_indr_block_bind_cb_t *cb,
220 					void *cb_priv)
221 {
222 	int err;
223 
224 	err = cb(NULL, NULL, cb_priv, TC_SETUP_ACT, NULL, fl_act, NULL);
225 	if (err < 0)
226 		return err;
227 
228 	if (hw_count)
229 		*hw_count = 1;
230 
231 	return 0;
232 }
233 
234 static int tcf_action_offload_cmd(struct flow_offload_action *fl_act,
235 				  u32 *hw_count,
236 				  flow_indr_block_bind_cb_t *cb,
237 				  void *cb_priv)
238 {
239 	return cb ? tcf_action_offload_cmd_cb_ex(fl_act, hw_count,
240 						 cb, cb_priv) :
241 		    tcf_action_offload_cmd_ex(fl_act, hw_count);
242 }
243 
244 static int tcf_action_offload_add_ex(struct tc_action *action,
245 				     struct netlink_ext_ack *extack,
246 				     flow_indr_block_bind_cb_t *cb,
247 				     void *cb_priv)
248 {
249 	bool skip_sw = tc_act_skip_sw(action->tcfa_flags);
250 	struct tc_action *actions[TCA_ACT_MAX_PRIO] = {
251 		[0] = action,
252 	};
253 	struct flow_offload_action *fl_action;
254 	u32 in_hw_count = 0;
255 	int num, err = 0;
256 
257 	if (tc_act_skip_hw(action->tcfa_flags))
258 		return 0;
259 
260 	num = tcf_offload_act_num_actions_single(action);
261 	fl_action = offload_action_alloc(num);
262 	if (!fl_action)
263 		return -ENOMEM;
264 
265 	err = offload_action_init(fl_action, action, FLOW_ACT_REPLACE, extack);
266 	if (err)
267 		goto fl_err;
268 
269 	err = tc_setup_action(&fl_action->action, actions, 0, extack);
270 	if (err) {
271 		NL_SET_ERR_MSG_MOD(extack,
272 				   "Failed to setup tc actions for offload");
273 		goto fl_err;
274 	}
275 
276 	err = tcf_action_offload_cmd(fl_action, &in_hw_count, cb, cb_priv);
277 	if (!err)
278 		cb ? offload_action_hw_count_inc(action, in_hw_count) :
279 		     offload_action_hw_count_set(action, in_hw_count);
280 
281 	if (skip_sw && !tc_act_in_hw(action))
282 		err = -EINVAL;
283 
284 	tc_cleanup_offload_action(&fl_action->action);
285 
286 fl_err:
287 	kfree(fl_action);
288 
289 	return err;
290 }
291 
292 /* offload the tc action after it is inserted */
293 static int tcf_action_offload_add(struct tc_action *action,
294 				  struct netlink_ext_ack *extack)
295 {
296 	return tcf_action_offload_add_ex(action, extack, NULL, NULL);
297 }
298 
299 int tcf_action_update_hw_stats(struct tc_action *action)
300 {
301 	struct flow_offload_action fl_act = {};
302 	int err;
303 
304 	err = offload_action_init(&fl_act, action, FLOW_ACT_STATS, NULL);
305 	if (err)
306 		return err;
307 
308 	err = tcf_action_offload_cmd(&fl_act, NULL, NULL, NULL);
309 	if (!err) {
310 		preempt_disable();
311 		tcf_action_stats_update(action, fl_act.stats.bytes,
312 					fl_act.stats.pkts,
313 					fl_act.stats.drops,
314 					fl_act.stats.lastused,
315 					true);
316 		preempt_enable();
317 		action->used_hw_stats = fl_act.stats.used_hw_stats;
318 		action->used_hw_stats_valid = true;
319 	} else {
320 		return -EOPNOTSUPP;
321 	}
322 
323 	return 0;
324 }
325 EXPORT_SYMBOL(tcf_action_update_hw_stats);
326 
327 static int tcf_action_offload_del_ex(struct tc_action *action,
328 				     flow_indr_block_bind_cb_t *cb,
329 				     void *cb_priv)
330 {
331 	struct flow_offload_action fl_act = {};
332 	u32 in_hw_count = 0;
333 	int err = 0;
334 
335 	if (!tc_act_in_hw(action))
336 		return 0;
337 
338 	err = offload_action_init(&fl_act, action, FLOW_ACT_DESTROY, NULL);
339 	if (err)
340 		return err;
341 
342 	err = tcf_action_offload_cmd(&fl_act, &in_hw_count, cb, cb_priv);
343 	if (err < 0)
344 		return err;
345 
346 	if (!cb && action->in_hw_count != in_hw_count)
347 		return -EINVAL;
348 
349 	/* do not need to update hw state when deleting action */
350 	if (cb && in_hw_count)
351 		offload_action_hw_count_dec(action, in_hw_count);
352 
353 	return 0;
354 }
355 
356 static int tcf_action_offload_del(struct tc_action *action)
357 {
358 	return tcf_action_offload_del_ex(action, NULL, NULL);
359 }
360 
361 static void tcf_action_cleanup(struct tc_action *p)
362 {
363 	tcf_action_offload_del(p);
364 	if (p->ops->cleanup)
365 		p->ops->cleanup(p);
366 
367 	gen_kill_estimator(&p->tcfa_rate_est);
368 	free_tcf(p);
369 }
370 
371 static int __tcf_action_put(struct tc_action *p, bool bind)
372 {
373 	struct tcf_idrinfo *idrinfo = p->idrinfo;
374 
375 	if (refcount_dec_and_mutex_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
376 		if (bind)
377 			atomic_dec(&p->tcfa_bindcnt);
378 		idr_remove(&idrinfo->action_idr, p->tcfa_index);
379 		mutex_unlock(&idrinfo->lock);
380 
381 		tcf_action_cleanup(p);
382 		return 1;
383 	}
384 
385 	if (bind)
386 		atomic_dec(&p->tcfa_bindcnt);
387 
388 	return 0;
389 }
390 
391 static int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
392 {
393 	int ret = 0;
394 
395 	/* Release with strict==1 and bind==0 is only called through act API
396 	 * interface (classifiers always bind). Only case when action with
397 	 * positive reference count and zero bind count can exist is when it was
398 	 * also created with act API (unbinding last classifier will destroy the
399 	 * action if it was created by classifier). So only case when bind count
400 	 * can be changed after initial check is when unbound action is
401 	 * destroyed by act API while classifier binds to action with same id
402 	 * concurrently. This result either creation of new action(same behavior
403 	 * as before), or reusing existing action if concurrent process
404 	 * increments reference count before action is deleted. Both scenarios
405 	 * are acceptable.
406 	 */
407 	if (p) {
408 		if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
409 			return -EPERM;
410 
411 		if (__tcf_action_put(p, bind))
412 			ret = ACT_P_DELETED;
413 	}
414 
415 	return ret;
416 }
417 
418 int tcf_idr_release(struct tc_action *a, bool bind)
419 {
420 	const struct tc_action_ops *ops = a->ops;
421 	int ret;
422 
423 	ret = __tcf_idr_release(a, bind, false);
424 	if (ret == ACT_P_DELETED)
425 		module_put(ops->owner);
426 	return ret;
427 }
428 EXPORT_SYMBOL(tcf_idr_release);
429 
430 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
431 {
432 	struct tc_cookie *user_cookie;
433 	u32 cookie_len = 0;
434 
435 	rcu_read_lock();
436 	user_cookie = rcu_dereference(act->user_cookie);
437 
438 	if (user_cookie)
439 		cookie_len = nla_total_size(user_cookie->len);
440 	rcu_read_unlock();
441 
442 	return  nla_total_size(0) /* action number nested */
443 		+ nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
444 		+ cookie_len /* TCA_ACT_COOKIE */
445 		+ nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_HW_STATS */
446 		/* TCA_ACT_USED_HW_STATS */
447 		+ nla_total_size(sizeof(struct nla_bitfield32))
448 		+ nla_total_size(sizeof(u32)) /* TCA_ACT_IN_HW_COUNT */
449 		+ nla_total_size(0) /* TCA_ACT_STATS nested */
450 		+ nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_FLAGS */
451 		/* TCA_STATS_BASIC */
452 		+ nla_total_size_64bit(sizeof(struct gnet_stats_basic))
453 		/* TCA_STATS_BASIC_HW */
454 		+ nla_total_size_64bit(sizeof(struct gnet_stats_basic))
455 		/* TCA_STATS_PKT64, emitted by both of the basic copies above */
456 		+ 2 * nla_total_size_64bit(sizeof(u64))
457 		/* TCA_STATS_RATE_EST */
458 		+ nla_total_size_64bit(sizeof(struct gnet_stats_rate_est))
459 		/* TCA_STATS_RATE_EST64 */
460 		+ nla_total_size_64bit(sizeof(struct gnet_stats_rate_est64))
461 		/* TCA_STATS_QUEUE */
462 		+ nla_total_size_64bit(sizeof(struct gnet_stats_queue))
463 		+ nla_total_size(0) /* TCA_ACT_OPTIONS nested */
464 		/* TCA_GACT_TM; actions dump their tcf_t with nla_put_64bit(),
465 		 * which may emit an extra NLA_PAD attribute.
466 		 */
467 		+ nla_total_size_64bit(sizeof(struct tcf_t));
468 }
469 
470 static size_t tcf_action_full_attrs_size(size_t sz)
471 {
472 	return NLMSG_HDRLEN                     /* struct nlmsghdr */
473 		+ sizeof(struct tcamsg)
474 		+ nla_total_size(0)             /* TCA_ACT_TAB nested */
475 		+ sz;
476 }
477 
478 static size_t tcf_action_fill_size(const struct tc_action *act)
479 {
480 	size_t sz = tcf_action_shared_attrs_size(act);
481 
482 	if (act->ops->get_fill_size)
483 		return act->ops->get_fill_size(act) + sz;
484 	return sz;
485 }
486 
487 static int
488 tcf_action_dump_terse(struct sk_buff *skb, struct tc_action *a, bool from_act)
489 {
490 	unsigned char *b = skb_tail_pointer(skb);
491 	struct tc_cookie *cookie;
492 
493 	if (nla_put_string(skb, TCA_ACT_KIND, a->ops->kind))
494 		goto nla_put_failure;
495 	if (tcf_action_copy_stats(skb, a, 0))
496 		goto nla_put_failure;
497 	if (from_act && nla_put_u32(skb, TCA_ACT_INDEX, a->tcfa_index))
498 		goto nla_put_failure;
499 
500 	rcu_read_lock();
501 	cookie = rcu_dereference(a->user_cookie);
502 	if (cookie) {
503 		if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
504 			rcu_read_unlock();
505 			goto nla_put_failure;
506 		}
507 	}
508 	rcu_read_unlock();
509 
510 	return 0;
511 
512 nla_put_failure:
513 	nlmsg_trim(skb, b);
514 	return -1;
515 }
516 
517 static int
518 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
519 {
520 	unsigned char *b = skb_tail_pointer(skb);
521 	struct nlattr *nest;
522 	int err = -EINVAL;
523 	u32 flags;
524 
525 	if (tcf_action_dump_terse(skb, a, false))
526 		goto nla_put_failure;
527 
528 	if (a->hw_stats != TCA_ACT_HW_STATS_ANY &&
529 	    nla_put_bitfield32(skb, TCA_ACT_HW_STATS,
530 			       a->hw_stats, TCA_ACT_HW_STATS_ANY))
531 		goto nla_put_failure;
532 
533 	if (a->used_hw_stats_valid &&
534 	    nla_put_bitfield32(skb, TCA_ACT_USED_HW_STATS,
535 			       a->used_hw_stats, TCA_ACT_HW_STATS_ANY))
536 		goto nla_put_failure;
537 
538 	flags = a->tcfa_flags & TCA_ACT_FLAGS_USER_MASK;
539 	if (flags &&
540 	    nla_put_bitfield32(skb, TCA_ACT_FLAGS,
541 			       flags, flags))
542 		goto nla_put_failure;
543 
544 	if (nla_put_u32(skb, TCA_ACT_IN_HW_COUNT, a->in_hw_count))
545 		goto nla_put_failure;
546 
547 	nest = nla_nest_start_noflag(skb, TCA_ACT_OPTIONS);
548 	if (nest == NULL)
549 		goto nla_put_failure;
550 	err = tcf_action_dump_old(skb, a, bind, ref);
551 	if (err > 0) {
552 		nla_nest_end(skb, nest);
553 		return err;
554 	}
555 
556 nla_put_failure:
557 	nlmsg_trim(skb, b);
558 	return -1;
559 }
560 
561 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
562 			   struct netlink_callback *cb)
563 {
564 	int err = 0, index = -1, s_i = 0, n_i = 0;
565 	u32 act_flags = cb->args[2];
566 	unsigned long jiffy_since = cb->args[3];
567 	struct nlattr *nest;
568 	struct idr *idr = &idrinfo->action_idr;
569 	struct tc_action *p;
570 	unsigned long id = 1;
571 	unsigned long tmp;
572 
573 	mutex_lock(&idrinfo->lock);
574 
575 	s_i = cb->args[0];
576 
577 	idr_for_each_entry_ul(idr, p, tmp, id) {
578 		index++;
579 		if (index < s_i)
580 			continue;
581 		if (IS_ERR(p))
582 			continue;
583 
584 		if (jiffy_since &&
585 		    time_after(jiffy_since,
586 			       (unsigned long)p->tcfa_tm.lastuse))
587 			continue;
588 
589 		tcf_action_update_hw_stats(p);
590 
591 		nest = nla_nest_start_noflag(skb, n_i);
592 		if (!nest) {
593 			index--;
594 			goto nla_put_failure;
595 		}
596 		err = (act_flags & TCA_ACT_FLAG_TERSE_DUMP) ?
597 			tcf_action_dump_terse(skb, p, true) :
598 			tcf_action_dump_1(skb, p, 0, 0);
599 		if (err < 0) {
600 			index--;
601 			nlmsg_trim(skb, nest);
602 			goto done;
603 		}
604 		nla_nest_end(skb, nest);
605 		n_i++;
606 		if (!(act_flags & TCA_ACT_FLAG_LARGE_DUMP_ON) &&
607 		    n_i >= TCA_ACT_MAX_PRIO)
608 			goto done;
609 	}
610 done:
611 	if (index >= 0)
612 		cb->args[0] = index + 1;
613 
614 	mutex_unlock(&idrinfo->lock);
615 	if (n_i) {
616 		if (act_flags & TCA_ACT_FLAG_LARGE_DUMP_ON)
617 			cb->args[1] = n_i;
618 	}
619 	return n_i;
620 
621 nla_put_failure:
622 	nla_nest_cancel(skb, nest);
623 	goto done;
624 }
625 
626 static int tcf_idr_release_unsafe(struct tc_action *p)
627 {
628 	if (atomic_read(&p->tcfa_bindcnt) > 0)
629 		return -EPERM;
630 
631 	if (refcount_dec_and_test(&p->tcfa_refcnt)) {
632 		idr_remove(&p->idrinfo->action_idr, p->tcfa_index);
633 		tcf_action_cleanup(p);
634 		return ACT_P_DELETED;
635 	}
636 
637 	return 0;
638 }
639 
640 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
641 			  const struct tc_action_ops *ops,
642 			  struct netlink_ext_ack *extack)
643 {
644 	struct nlattr *nest;
645 	int n_i = 0;
646 	int ret = -EINVAL;
647 	struct idr *idr = &idrinfo->action_idr;
648 	struct tc_action *p;
649 	unsigned long id = 1;
650 	unsigned long tmp;
651 
652 	nest = nla_nest_start_noflag(skb, 0);
653 	if (nest == NULL)
654 		goto nla_put_failure;
655 	if (nla_put_string(skb, TCA_ACT_KIND, ops->kind))
656 		goto nla_put_failure;
657 
658 	ret = 0;
659 	mutex_lock(&idrinfo->lock);
660 	idr_for_each_entry_ul(idr, p, tmp, id) {
661 		if (IS_ERR(p))
662 			continue;
663 		ret = tcf_idr_release_unsafe(p);
664 		if (ret == ACT_P_DELETED)
665 			module_put(ops->owner);
666 		else if (ret < 0)
667 			break;
668 		n_i++;
669 	}
670 	mutex_unlock(&idrinfo->lock);
671 	if (ret < 0) {
672 		if (n_i)
673 			NL_SET_ERR_MSG(extack, "Unable to flush all TC actions");
674 		else
675 			goto nla_put_failure;
676 	}
677 
678 	ret = nla_put_u32(skb, TCA_FCNT, n_i);
679 	if (ret)
680 		goto nla_put_failure;
681 	nla_nest_end(skb, nest);
682 
683 	return n_i;
684 nla_put_failure:
685 	nla_nest_cancel(skb, nest);
686 	return ret;
687 }
688 
689 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
690 		       struct netlink_callback *cb, int type,
691 		       const struct tc_action_ops *ops,
692 		       struct netlink_ext_ack *extack)
693 {
694 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
695 
696 	if (type == RTM_DELACTION) {
697 		return tcf_del_walker(idrinfo, skb, ops, extack);
698 	} else if (type == RTM_GETACTION) {
699 		return tcf_dump_walker(idrinfo, skb, cb);
700 	} else {
701 		WARN(1, "tcf_generic_walker: unknown command %d\n", type);
702 		NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
703 		return -EINVAL;
704 	}
705 }
706 EXPORT_SYMBOL(tcf_generic_walker);
707 
708 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
709 {
710 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
711 	struct tc_action *p;
712 
713 	mutex_lock(&idrinfo->lock);
714 	p = idr_find(&idrinfo->action_idr, index);
715 	if (IS_ERR(p))
716 		p = NULL;
717 	else if (p)
718 		refcount_inc(&p->tcfa_refcnt);
719 	mutex_unlock(&idrinfo->lock);
720 
721 	if (p) {
722 		*a = p;
723 		return true;
724 	}
725 	return false;
726 }
727 EXPORT_SYMBOL(tcf_idr_search);
728 
729 static int __tcf_generic_walker(struct net *net, struct sk_buff *skb,
730 				struct netlink_callback *cb, int type,
731 				const struct tc_action_ops *ops,
732 				struct netlink_ext_ack *extack)
733 {
734 	struct tc_action_net *tn = net_generic(net, ops->net_id);
735 
736 	if (unlikely(ops->walk))
737 		return ops->walk(net, skb, cb, type, ops, extack);
738 
739 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
740 }
741 
742 static int __tcf_idr_search(struct net *net,
743 			    const struct tc_action_ops *ops,
744 			    struct tc_action **a, u32 index)
745 {
746 	struct tc_action_net *tn = net_generic(net, ops->net_id);
747 
748 	if (unlikely(ops->lookup))
749 		return ops->lookup(net, a, index);
750 
751 	return tcf_idr_search(tn, a, index);
752 }
753 
754 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
755 {
756 	struct tc_action *p;
757 	int ret = 0;
758 
759 	mutex_lock(&idrinfo->lock);
760 	p = idr_find(&idrinfo->action_idr, index);
761 	if (!p) {
762 		mutex_unlock(&idrinfo->lock);
763 		return -ENOENT;
764 	}
765 
766 	if (!atomic_read(&p->tcfa_bindcnt)) {
767 		if (refcount_dec_and_test(&p->tcfa_refcnt)) {
768 			struct module *owner = p->ops->owner;
769 
770 			WARN_ON(p != idr_remove(&idrinfo->action_idr,
771 						p->tcfa_index));
772 			mutex_unlock(&idrinfo->lock);
773 
774 			tcf_action_cleanup(p);
775 			module_put(owner);
776 			return 0;
777 		}
778 		ret = 0;
779 	} else {
780 		ret = -EPERM;
781 	}
782 
783 	mutex_unlock(&idrinfo->lock);
784 	return ret;
785 }
786 
787 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
788 		   struct tc_action **a, const struct tc_action_ops *ops,
789 		   int bind, bool cpustats, u32 flags)
790 {
791 	struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
792 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
793 	int err = -ENOMEM;
794 
795 	if (unlikely(!p))
796 		return -ENOMEM;
797 	refcount_set(&p->tcfa_refcnt, 1);
798 	if (bind)
799 		atomic_set(&p->tcfa_bindcnt, 1);
800 
801 	if (cpustats) {
802 		p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_sync);
803 		if (!p->cpu_bstats)
804 			goto err1;
805 		p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_sync);
806 		if (!p->cpu_bstats_hw)
807 			goto err2;
808 		p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
809 		if (!p->cpu_qstats)
810 			goto err3;
811 	}
812 	gnet_stats_basic_sync_init(&p->tcfa_bstats);
813 	gnet_stats_basic_sync_init(&p->tcfa_bstats_hw);
814 	spin_lock_init(&p->tcfa_lock);
815 	p->tcfa_index = index;
816 	p->tcfa_tm.install = jiffies;
817 	p->tcfa_tm.lastuse = jiffies;
818 	p->tcfa_tm.firstuse = 0;
819 	p->tcfa_flags = flags;
820 	if (est) {
821 		err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
822 					&p->tcfa_rate_est,
823 					&p->tcfa_lock, false, est);
824 		if (err)
825 			goto err4;
826 	}
827 
828 	p->idrinfo = idrinfo;
829 	__module_get(ops->owner);
830 	p->ops = ops;
831 	*a = p;
832 	return 0;
833 err4:
834 	free_percpu(p->cpu_qstats);
835 err3:
836 	free_percpu(p->cpu_bstats_hw);
837 err2:
838 	free_percpu(p->cpu_bstats);
839 err1:
840 	kfree(p);
841 	return err;
842 }
843 EXPORT_SYMBOL(tcf_idr_create);
844 
845 int tcf_idr_create_from_flags(struct tc_action_net *tn, u32 index,
846 			      struct nlattr *est, struct tc_action **a,
847 			      const struct tc_action_ops *ops, int bind,
848 			      u32 flags)
849 {
850 	/* Set cpustats according to actions flags. */
851 	return tcf_idr_create(tn, index, est, a, ops, bind,
852 			      !(flags & TCA_ACT_FLAGS_NO_PERCPU_STATS), flags);
853 }
854 EXPORT_SYMBOL(tcf_idr_create_from_flags);
855 
856 /* Cleanup idr index that was allocated but not initialized. */
857 
858 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
859 {
860 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
861 
862 	mutex_lock(&idrinfo->lock);
863 	/* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
864 	WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
865 	mutex_unlock(&idrinfo->lock);
866 }
867 EXPORT_SYMBOL(tcf_idr_cleanup);
868 
869 /* Check if action with specified index exists. If actions is found, increments
870  * its reference and bind counters, and return 1. Otherwise insert temporary
871  * error pointer (to prevent concurrent users from inserting actions with same
872  * index) and return 0.
873  *
874  * May return -EAGAIN for binding actions in case of a parallel add/delete on
875  * the requested index.
876  */
877 
878 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
879 			struct tc_action **a, int bind)
880 {
881 	struct tcf_idrinfo *idrinfo = tn->idrinfo;
882 	struct tc_action *p;
883 	int ret;
884 	u32 max;
885 
886 	if (*index) {
887 		rcu_read_lock();
888 		p = idr_find(&idrinfo->action_idr, *index);
889 
890 		if (IS_ERR(p)) {
891 			/* This means that another process allocated
892 			 * index but did not assign the pointer yet.
893 			 */
894 			rcu_read_unlock();
895 			return -EAGAIN;
896 		}
897 
898 		if (!p) {
899 			/* Empty slot, try to allocate it */
900 			max = *index;
901 			rcu_read_unlock();
902 			goto new;
903 		}
904 
905 		if (!refcount_inc_not_zero(&p->tcfa_refcnt)) {
906 			/* Action was deleted in parallel */
907 			rcu_read_unlock();
908 			return -EAGAIN;
909 		}
910 
911 		if (bind)
912 			atomic_inc(&p->tcfa_bindcnt);
913 		*a = p;
914 
915 		rcu_read_unlock();
916 
917 		return 1;
918 	} else {
919 		/* Find a slot */
920 		*index = 1;
921 		max = UINT_MAX;
922 	}
923 
924 new:
925 	*a = NULL;
926 
927 	mutex_lock(&idrinfo->lock);
928 	ret = idr_alloc_u32(&idrinfo->action_idr, ERR_PTR(-EBUSY), index, max,
929 			    GFP_KERNEL);
930 	mutex_unlock(&idrinfo->lock);
931 
932 	/* N binds raced for action allocation,
933 	 * retry for all the ones that failed.
934 	 */
935 	if (ret == -ENOSPC && *index == max)
936 		ret = -EAGAIN;
937 
938 	return ret;
939 }
940 EXPORT_SYMBOL(tcf_idr_check_alloc);
941 
942 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
943 			 struct tcf_idrinfo *idrinfo)
944 {
945 	struct idr *idr = &idrinfo->action_idr;
946 	bool mutex_taken = false;
947 	struct tc_action *p;
948 	unsigned long id = 1;
949 	unsigned long tmp;
950 	int ret;
951 
952 	idr_for_each_entry_ul(idr, p, tmp, id) {
953 		if (IS_ERR(p))
954 			continue;
955 		if (tc_act_in_hw(p) && !mutex_taken) {
956 			rtnl_lock();
957 			mutex_taken = true;
958 		}
959 		ret = __tcf_idr_release(p, false, true);
960 		if (ret == ACT_P_DELETED)
961 			module_put(ops->owner);
962 		else if (ret < 0)
963 			return;
964 	}
965 	if (mutex_taken)
966 		rtnl_unlock();
967 	idr_destroy(&idrinfo->action_idr);
968 }
969 EXPORT_SYMBOL(tcf_idrinfo_destroy);
970 
971 static LIST_HEAD(act_base);
972 static DEFINE_RWLOCK(act_mod_lock);
973 /* since act ops id is stored in pernet subsystem list,
974  * then there is no way to walk through only all the action
975  * subsystem, so we keep tc action pernet ops id for
976  * reoffload to walk through.
977  */
978 static LIST_HEAD(act_pernet_id_list);
979 static DEFINE_MUTEX(act_id_mutex);
980 struct tc_act_pernet_id {
981 	struct list_head list;
982 	unsigned int id;
983 };
984 
985 static int tcf_pernet_add_id_list(unsigned int id)
986 {
987 	struct tc_act_pernet_id *id_ptr;
988 	int ret = 0;
989 
990 	mutex_lock(&act_id_mutex);
991 	list_for_each_entry(id_ptr, &act_pernet_id_list, list) {
992 		if (id_ptr->id == id) {
993 			ret = -EEXIST;
994 			goto err_out;
995 		}
996 	}
997 
998 	id_ptr = kzalloc_obj(*id_ptr);
999 	if (!id_ptr) {
1000 		ret = -ENOMEM;
1001 		goto err_out;
1002 	}
1003 	id_ptr->id = id;
1004 
1005 	list_add_tail(&id_ptr->list, &act_pernet_id_list);
1006 
1007 err_out:
1008 	mutex_unlock(&act_id_mutex);
1009 	return ret;
1010 }
1011 
1012 static void tcf_pernet_del_id_list(unsigned int id)
1013 {
1014 	struct tc_act_pernet_id *id_ptr;
1015 
1016 	mutex_lock(&act_id_mutex);
1017 	list_for_each_entry(id_ptr, &act_pernet_id_list, list) {
1018 		if (id_ptr->id == id) {
1019 			list_del(&id_ptr->list);
1020 			kfree(id_ptr);
1021 			break;
1022 		}
1023 	}
1024 	mutex_unlock(&act_id_mutex);
1025 }
1026 
1027 int tcf_register_action(struct tc_action_ops *act,
1028 			struct pernet_operations *ops)
1029 {
1030 	struct tc_action_ops *a;
1031 	int ret;
1032 
1033 	if (!act->act || !act->dump || !act->init)
1034 		return -EINVAL;
1035 
1036 	/* We have to register pernet ops before making the action ops visible,
1037 	 * otherwise tcf_action_init_1() could get a partially initialized
1038 	 * netns.
1039 	 */
1040 	ret = register_pernet_subsys(ops);
1041 	if (ret)
1042 		return ret;
1043 
1044 	if (ops->id) {
1045 		ret = tcf_pernet_add_id_list(*ops->id);
1046 		if (ret)
1047 			goto err_id;
1048 	}
1049 
1050 	write_lock(&act_mod_lock);
1051 	list_for_each_entry(a, &act_base, head) {
1052 		if (act->id == a->id || (strcmp(act->kind, a->kind) == 0)) {
1053 			ret = -EEXIST;
1054 			goto err_out;
1055 		}
1056 	}
1057 	list_add_tail(&act->head, &act_base);
1058 	write_unlock(&act_mod_lock);
1059 
1060 	return 0;
1061 
1062 err_out:
1063 	write_unlock(&act_mod_lock);
1064 	if (ops->id)
1065 		tcf_pernet_del_id_list(*ops->id);
1066 err_id:
1067 	unregister_pernet_subsys(ops);
1068 	return ret;
1069 }
1070 EXPORT_SYMBOL(tcf_register_action);
1071 
1072 int tcf_unregister_action(struct tc_action_ops *act,
1073 			  struct pernet_operations *ops)
1074 {
1075 	struct tc_action_ops *a;
1076 	int err = -ENOENT;
1077 
1078 	write_lock(&act_mod_lock);
1079 	list_for_each_entry(a, &act_base, head) {
1080 		if (a == act) {
1081 			list_del(&act->head);
1082 			err = 0;
1083 			break;
1084 		}
1085 	}
1086 	write_unlock(&act_mod_lock);
1087 	if (!err) {
1088 		unregister_pernet_subsys(ops);
1089 		if (ops->id)
1090 			tcf_pernet_del_id_list(*ops->id);
1091 	}
1092 	return err;
1093 }
1094 EXPORT_SYMBOL(tcf_unregister_action);
1095 
1096 /* lookup by name */
1097 static struct tc_action_ops *tc_lookup_action_n(char *kind)
1098 {
1099 	struct tc_action_ops *a, *res = NULL;
1100 
1101 	if (kind) {
1102 		read_lock(&act_mod_lock);
1103 		list_for_each_entry(a, &act_base, head) {
1104 			if (strcmp(kind, a->kind) == 0) {
1105 				if (try_module_get(a->owner))
1106 					res = a;
1107 				break;
1108 			}
1109 		}
1110 		read_unlock(&act_mod_lock);
1111 	}
1112 	return res;
1113 }
1114 
1115 /* lookup by nlattr */
1116 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
1117 {
1118 	struct tc_action_ops *a, *res = NULL;
1119 
1120 	if (kind) {
1121 		read_lock(&act_mod_lock);
1122 		list_for_each_entry(a, &act_base, head) {
1123 			if (nla_strcmp(kind, a->kind) == 0) {
1124 				if (try_module_get(a->owner))
1125 					res = a;
1126 				break;
1127 			}
1128 		}
1129 		read_unlock(&act_mod_lock);
1130 	}
1131 	return res;
1132 }
1133 
1134 /*TCA_ACT_MAX_PRIO is 32, there count up to 32 */
1135 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
1136 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
1137 		    int nr_actions, struct tcf_result *res)
1138 {
1139 	u32 jmp_prgcnt = 0;
1140 	u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
1141 	int i;
1142 	int ret = TC_ACT_OK;
1143 
1144 	if (skb_skip_tc_classify(skb))
1145 		return TC_ACT_OK;
1146 
1147 restart_act_graph:
1148 	for (i = 0; i < nr_actions; i++) {
1149 		const struct tc_action *a = actions[i];
1150 		int repeat_ttl;
1151 
1152 		if (jmp_prgcnt > 0) {
1153 			jmp_prgcnt -= 1;
1154 			continue;
1155 		}
1156 
1157 		if (tc_act_skip_sw(a->tcfa_flags))
1158 			continue;
1159 
1160 		repeat_ttl = 32;
1161 repeat:
1162 		ret = tc_act(skb, a, res);
1163 		if (unlikely(ret == TC_ACT_REPEAT)) {
1164 			if (--repeat_ttl != 0)
1165 				goto repeat;
1166 			/* suspicious opcode, stop pipeline */
1167 			net_warn_ratelimited("TC_ACT_REPEAT abuse ?\n");
1168 			return TC_ACT_OK;
1169 		}
1170 		if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
1171 			jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
1172 			if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
1173 				/* faulty opcode, stop pipeline */
1174 				return TC_ACT_OK;
1175 			} else {
1176 				jmp_ttl -= 1;
1177 				if (jmp_ttl > 0)
1178 					goto restart_act_graph;
1179 				else /* faulty graph, stop pipeline */
1180 					return TC_ACT_OK;
1181 			}
1182 		} else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
1183 			struct tcf_chain *chain = rcu_dereference_bh(a->goto_chain);
1184 
1185 			if (unlikely(!chain)) {
1186 				tcf_set_drop_reason(skb,
1187 						    SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1188 				return TC_ACT_SHOT;
1189 			}
1190 			tcf_action_goto_chain_exec(chain, res);
1191 		}
1192 
1193 		if (ret != TC_ACT_PIPE)
1194 			break;
1195 	}
1196 
1197 	return ret;
1198 }
1199 EXPORT_SYMBOL(tcf_action_exec);
1200 
1201 int tcf_action_destroy(struct tc_action *actions[], int bind)
1202 {
1203 	struct tc_action *a;
1204 	int ret = 0, i;
1205 
1206 	tcf_act_for_each_action(i, a, actions) {
1207 		actions[i] = NULL;
1208 		/* Drop our reference even if the action is still bound to a filter. */
1209 		ret = tcf_idr_release(a, bind);
1210 	}
1211 	return ret;
1212 }
1213 
1214 static int tcf_action_put(struct tc_action *p)
1215 {
1216 	return __tcf_action_put(p, false);
1217 }
1218 
1219 static void tcf_action_put_many(struct tc_action *actions[])
1220 {
1221 	struct tc_action *a;
1222 	int i;
1223 
1224 	tcf_act_for_each_action(i, a, actions) {
1225 		const struct tc_action_ops *ops = a->ops;
1226 		if (tcf_action_put(a))
1227 			module_put(ops->owner);
1228 	}
1229 }
1230 
1231 static void tca_put_bound_many(struct tc_action *actions[], int init_res[])
1232 {
1233 	struct tc_action *a;
1234 	int i;
1235 
1236 	tcf_act_for_each_action(i, a, actions) {
1237 		const struct tc_action_ops *ops = a->ops;
1238 
1239 		if (init_res[i] == ACT_P_CREATED)
1240 			continue;
1241 
1242 		if (tcf_action_put(a))
1243 			module_put(ops->owner);
1244 	}
1245 }
1246 
1247 int
1248 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
1249 {
1250 	return a->ops->dump(skb, a, bind, ref);
1251 }
1252 
1253 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
1254 		    int bind, int ref, bool terse)
1255 {
1256 	struct tc_action *a;
1257 	int err = -EINVAL, i;
1258 	struct nlattr *nest;
1259 
1260 	tcf_act_for_each_action(i, a, actions) {
1261 		nest = nla_nest_start_noflag(skb, i + 1);
1262 		if (nest == NULL)
1263 			goto nla_put_failure;
1264 		err = terse ? tcf_action_dump_terse(skb, a, false) :
1265 			tcf_action_dump_1(skb, a, bind, ref);
1266 		if (err < 0)
1267 			goto errout;
1268 		nla_nest_end(skb, nest);
1269 	}
1270 
1271 	return 0;
1272 
1273 nla_put_failure:
1274 	err = -EINVAL;
1275 errout:
1276 	nla_nest_cancel(skb, nest);
1277 	return err;
1278 }
1279 
1280 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
1281 {
1282 	struct tc_cookie *c = kzalloc_obj(*c);
1283 	if (!c)
1284 		return NULL;
1285 
1286 	c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
1287 	if (!c->data) {
1288 		kfree(c);
1289 		return NULL;
1290 	}
1291 	c->len = nla_len(tb[TCA_ACT_COOKIE]);
1292 
1293 	return c;
1294 }
1295 
1296 static u8 tcf_action_hw_stats_get(struct nlattr *hw_stats_attr)
1297 {
1298 	struct nla_bitfield32 hw_stats_bf;
1299 
1300 	/* If the user did not pass the attr, that means he does
1301 	 * not care about the type. Return "any" in that case
1302 	 * which is setting on all supported types.
1303 	 */
1304 	if (!hw_stats_attr)
1305 		return TCA_ACT_HW_STATS_ANY;
1306 	hw_stats_bf = nla_get_bitfield32(hw_stats_attr);
1307 	return hw_stats_bf.value;
1308 }
1309 
1310 static const struct nla_policy tcf_action_policy[TCA_ACT_MAX + 1] = {
1311 	[TCA_ACT_KIND]		= { .type = NLA_STRING },
1312 	[TCA_ACT_INDEX]		= { .type = NLA_U32 },
1313 	[TCA_ACT_COOKIE]	= { .type = NLA_BINARY,
1314 				    .len = TC_COOKIE_MAX_SIZE },
1315 	[TCA_ACT_OPTIONS]	= { .type = NLA_NESTED },
1316 	[TCA_ACT_FLAGS]		= NLA_POLICY_BITFIELD32(TCA_ACT_FLAGS_NO_PERCPU_STATS |
1317 							TCA_ACT_FLAGS_SKIP_HW |
1318 							TCA_ACT_FLAGS_SKIP_SW),
1319 	[TCA_ACT_HW_STATS]	= NLA_POLICY_BITFIELD32(TCA_ACT_HW_STATS_ANY),
1320 };
1321 
1322 void tcf_idr_insert_many(struct tc_action *actions[], int init_res[])
1323 {
1324 	struct tc_action *a;
1325 	int i;
1326 
1327 	tcf_act_for_each_action(i, a, actions) {
1328 		struct tcf_idrinfo *idrinfo;
1329 
1330 		if (init_res[i] == ACT_P_BOUND)
1331 			continue;
1332 
1333 		idrinfo = a->idrinfo;
1334 		mutex_lock(&idrinfo->lock);
1335 		/* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
1336 		idr_replace(&idrinfo->action_idr, a, a->tcfa_index);
1337 		mutex_unlock(&idrinfo->lock);
1338 	}
1339 }
1340 
1341 struct tc_action_ops *tc_action_load_ops(struct nlattr *nla, u32 flags,
1342 					 struct netlink_ext_ack *extack)
1343 {
1344 	bool police = flags & TCA_ACT_FLAGS_POLICE;
1345 	struct nlattr *tb[TCA_ACT_MAX + 1];
1346 	struct tc_action_ops *a_o;
1347 	char act_name[IFNAMSIZ];
1348 	struct nlattr *kind;
1349 	int err;
1350 
1351 	if (!police) {
1352 		err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1353 						  tcf_action_policy, extack);
1354 		if (err < 0)
1355 			return ERR_PTR(err);
1356 		err = -EINVAL;
1357 		kind = tb[TCA_ACT_KIND];
1358 		if (!kind) {
1359 			NL_SET_ERR_MSG(extack, "TC action kind must be specified");
1360 			return ERR_PTR(err);
1361 		}
1362 		if (nla_strscpy(act_name, kind, IFNAMSIZ) < 0) {
1363 			NL_SET_ERR_MSG(extack, "TC action name too long");
1364 			return ERR_PTR(err);
1365 		}
1366 	} else {
1367 		if (strscpy(act_name, "police", IFNAMSIZ) < 0) {
1368 			NL_SET_ERR_MSG(extack, "TC action name too long");
1369 			return ERR_PTR(-EINVAL);
1370 		}
1371 	}
1372 
1373 	a_o = tc_lookup_action_n(act_name);
1374 	if (a_o == NULL) {
1375 #ifdef CONFIG_MODULES
1376 		bool rtnl_held = !(flags & TCA_ACT_FLAGS_NO_RTNL);
1377 
1378 		if (rtnl_held)
1379 			rtnl_unlock();
1380 		request_module(NET_ACT_ALIAS_PREFIX "%s", act_name);
1381 		if (rtnl_held)
1382 			rtnl_lock();
1383 
1384 		a_o = tc_lookup_action_n(act_name);
1385 
1386 		/* We dropped the RTNL semaphore in order to
1387 		 * perform the module load.  So, even if we
1388 		 * succeeded in loading the module we have to
1389 		 * tell the caller to replay the request.  We
1390 		 * indicate this using -EAGAIN.
1391 		 */
1392 		if (a_o != NULL) {
1393 			module_put(a_o->owner);
1394 			return ERR_PTR(-EAGAIN);
1395 		}
1396 #endif
1397 		NL_SET_ERR_MSG(extack, "Failed to load TC action module");
1398 		return ERR_PTR(-ENOENT);
1399 	}
1400 
1401 	return a_o;
1402 }
1403 
1404 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
1405 				    struct nlattr *nla, struct nlattr *est,
1406 				    struct tc_action_ops *a_o, int *init_res,
1407 				    u32 flags, struct netlink_ext_ack *extack)
1408 {
1409 	bool police = flags & TCA_ACT_FLAGS_POLICE;
1410 	struct nla_bitfield32 userflags = { 0, 0 };
1411 	struct tc_cookie *user_cookie = NULL;
1412 	u8 hw_stats = TCA_ACT_HW_STATS_ANY;
1413 	struct nlattr *tb[TCA_ACT_MAX + 1];
1414 	struct tc_action *a;
1415 	int err;
1416 
1417 	/* backward compatibility for policer */
1418 	if (!police) {
1419 		err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1420 						  tcf_action_policy, extack);
1421 		if (err < 0)
1422 			return ERR_PTR(err);
1423 		if (tb[TCA_ACT_COOKIE]) {
1424 			user_cookie = nla_memdup_cookie(tb);
1425 			if (!user_cookie) {
1426 				NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
1427 				err = -ENOMEM;
1428 				goto err_out;
1429 			}
1430 		}
1431 		hw_stats = tcf_action_hw_stats_get(tb[TCA_ACT_HW_STATS]);
1432 		if (tb[TCA_ACT_FLAGS]) {
1433 			userflags = nla_get_bitfield32(tb[TCA_ACT_FLAGS]);
1434 			if (!tc_act_flags_valid(userflags.value)) {
1435 				err = -EINVAL;
1436 				goto err_out;
1437 			}
1438 		}
1439 
1440 		err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, tp,
1441 				userflags.value | flags, extack);
1442 	} else {
1443 		err = a_o->init(net, nla, est, &a, tp, userflags.value | flags,
1444 				extack);
1445 	}
1446 	if (err < 0)
1447 		goto err_out;
1448 	*init_res = err;
1449 
1450 	if (!police && tb[TCA_ACT_COOKIE])
1451 		tcf_set_action_cookie(&a->user_cookie, user_cookie);
1452 
1453 	if (!police)
1454 		a->hw_stats = hw_stats;
1455 
1456 	return a;
1457 
1458 err_out:
1459 	if (user_cookie) {
1460 		kfree(user_cookie->data);
1461 		kfree(user_cookie);
1462 	}
1463 	return ERR_PTR(err);
1464 }
1465 
1466 static bool tc_act_bind(u32 flags)
1467 {
1468 	return !!(flags & TCA_ACT_FLAGS_BIND);
1469 }
1470 
1471 /* Returns numbers of initialized actions or negative error. */
1472 
1473 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
1474 		    struct nlattr *est, struct tc_action *actions[],
1475 		    int init_res[], size_t *attr_size,
1476 		    u32 flags, u32 fl_flags,
1477 		    struct netlink_ext_ack *extack)
1478 {
1479 	struct tc_action_ops *ops[TCA_ACT_MAX_PRIO] = {};
1480 	struct nlattr *tb[TCA_ACT_MAX_PRIO + 2];
1481 	struct tc_action *act;
1482 	size_t sz = 0;
1483 	int err;
1484 	int i;
1485 
1486 	err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO + 1, nla, NULL,
1487 					  extack);
1488 	if (err < 0)
1489 		return err;
1490 
1491 	/* The nested attributes are parsed as types, but they are really an
1492 	 * array of actions. So we parse one more than we can handle, and return
1493 	 * an error if the last one is set (as that indicates that the request
1494 	 * contained more than the maximum number of actions).
1495 	 */
1496 	if (tb[TCA_ACT_MAX_PRIO + 1]) {
1497 		NL_SET_ERR_MSG_FMT(extack,
1498 				   "Only %d actions supported per filter",
1499 				   TCA_ACT_MAX_PRIO);
1500 		return -EINVAL;
1501 	}
1502 
1503 	for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1504 		struct tc_action_ops *a_o;
1505 
1506 		a_o = tc_action_load_ops(tb[i], flags, extack);
1507 		if (IS_ERR(a_o)) {
1508 			err = PTR_ERR(a_o);
1509 			goto err_mod;
1510 		}
1511 		ops[i - 1] = a_o;
1512 	}
1513 
1514 	for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1515 		act = tcf_action_init_1(net, tp, tb[i], est, ops[i - 1],
1516 					&init_res[i - 1], flags, extack);
1517 		if (IS_ERR(act)) {
1518 			err = PTR_ERR(act);
1519 			goto err;
1520 		}
1521 		sz += tcf_action_fill_size(act);
1522 		/* Start from index 0 */
1523 		actions[i - 1] = act;
1524 		if (tc_act_bind(flags)) {
1525 			bool skip_sw = tc_skip_sw(fl_flags);
1526 			bool skip_hw = tc_skip_hw(fl_flags);
1527 
1528 			if (tc_act_bind(act->tcfa_flags)) {
1529 				/* Action is created by classifier and is not
1530 				 * standalone. Check that the user did not set
1531 				 * any action flags different than the
1532 				 * classifier flags, and inherit the flags from
1533 				 * the classifier for the compatibility case
1534 				 * where no flags were specified at all.
1535 				 */
1536 				if ((tc_act_skip_sw(act->tcfa_flags) && !skip_sw) ||
1537 				    (tc_act_skip_hw(act->tcfa_flags) && !skip_hw)) {
1538 					NL_SET_ERR_MSG(extack,
1539 						       "Mismatch between action and filter offload flags");
1540 					err = -EINVAL;
1541 					goto err;
1542 				}
1543 				if (skip_sw)
1544 					act->tcfa_flags |= TCA_ACT_FLAGS_SKIP_SW;
1545 				if (skip_hw)
1546 					act->tcfa_flags |= TCA_ACT_FLAGS_SKIP_HW;
1547 				continue;
1548 			}
1549 
1550 			/* Action is standalone */
1551 			if (skip_sw != tc_act_skip_sw(act->tcfa_flags) ||
1552 			    skip_hw != tc_act_skip_hw(act->tcfa_flags)) {
1553 				NL_SET_ERR_MSG(extack,
1554 					       "Mismatch between action and filter offload flags");
1555 				err = -EINVAL;
1556 				goto err;
1557 			}
1558 		} else {
1559 			err = tcf_action_offload_add(act, extack);
1560 			if (tc_act_skip_sw(act->tcfa_flags) && err)
1561 				goto err;
1562 		}
1563 	}
1564 
1565 	/* We have to commit them all together, because if any error happened in
1566 	 * between, we could not handle the failure gracefully.
1567 	 */
1568 	tcf_idr_insert_many(actions, init_res);
1569 
1570 	*attr_size = tcf_action_full_attrs_size(sz);
1571 	err = i - 1;
1572 	goto err_mod;
1573 
1574 err:
1575 	tcf_action_destroy(actions, flags & TCA_ACT_FLAGS_BIND);
1576 err_mod:
1577 	for (i = 0; i < TCA_ACT_MAX_PRIO && ops[i]; i++)
1578 		module_put(ops[i]->owner);
1579 	return err;
1580 }
1581 
1582 void tcf_action_update_stats(struct tc_action *a, u64 bytes, u64 packets,
1583 			     u64 drops, bool hw)
1584 {
1585 	if (a->cpu_bstats) {
1586 		_bstats_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
1587 
1588 		this_cpu_add(a->cpu_qstats->drops, drops);
1589 
1590 		if (hw)
1591 			_bstats_update(this_cpu_ptr(a->cpu_bstats_hw),
1592 				       bytes, packets);
1593 		return;
1594 	}
1595 
1596 	_bstats_update(&a->tcfa_bstats, bytes, packets);
1597 	atomic_add(drops, &a->tcfa_drops);
1598 	if (hw)
1599 		_bstats_update(&a->tcfa_bstats_hw, bytes, packets);
1600 }
1601 EXPORT_SYMBOL(tcf_action_update_stats);
1602 
1603 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
1604 			  int compat_mode)
1605 {
1606 	struct gnet_stats_queue qstats = {0};
1607 	struct gnet_dump d;
1608 	int err = 0;
1609 
1610 	if (p == NULL)
1611 		goto errout;
1612 
1613 	/* compat_mode being true specifies a call that is supposed
1614 	 * to add additional backward compatibility statistic TLVs.
1615 	 */
1616 	if (compat_mode) {
1617 		if (p->type == TCA_OLD_COMPAT)
1618 			err = gnet_stats_start_copy_compat(skb, 0,
1619 							   TCA_STATS,
1620 							   TCA_XSTATS,
1621 							   &p->tcfa_lock, &d,
1622 							   TCA_PAD);
1623 		else
1624 			return 0;
1625 	} else
1626 		err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1627 					    &p->tcfa_lock, &d, TCA_ACT_PAD);
1628 
1629 	if (err < 0)
1630 		goto errout;
1631 
1632 	qstats.drops = atomic_read(&p->tcfa_drops);
1633 	qstats.overlimits = atomic_read(&p->tcfa_overlimits);
1634 
1635 	if (gnet_stats_copy_basic(&d, p->cpu_bstats,
1636 				  &p->tcfa_bstats, false) < 0 ||
1637 	    gnet_stats_copy_basic_hw(&d, p->cpu_bstats_hw,
1638 				     &p->tcfa_bstats_hw, false) < 0 ||
1639 	    gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1640 	    gnet_stats_copy_queue(&d, p->cpu_qstats,
1641 				  &qstats,
1642 				  qstats.qlen) < 0)
1643 		goto errout;
1644 
1645 	if (gnet_stats_finish_copy(&d) < 0)
1646 		goto errout;
1647 
1648 	return 0;
1649 
1650 errout:
1651 	return -1;
1652 }
1653 
1654 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1655 			u32 portid, u32 seq, u16 flags, int event, int bind,
1656 			int ref, struct netlink_ext_ack *extack)
1657 {
1658 	struct tcamsg *t;
1659 	struct nlmsghdr *nlh;
1660 	unsigned char *b = skb_tail_pointer(skb);
1661 	struct nlattr *nest;
1662 
1663 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1664 	if (!nlh)
1665 		goto out_nlmsg_trim;
1666 	t = nlmsg_data(nlh);
1667 	t->tca_family = AF_UNSPEC;
1668 	t->tca__pad1 = 0;
1669 	t->tca__pad2 = 0;
1670 
1671 	if (extack && extack->_msg &&
1672 	    nla_put_string(skb, TCA_ROOT_EXT_WARN_MSG, extack->_msg))
1673 		goto out_nlmsg_trim;
1674 
1675 	nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1676 	if (!nest)
1677 		goto out_nlmsg_trim;
1678 
1679 	if (tcf_action_dump(skb, actions, bind, ref, false) < 0)
1680 		goto out_nlmsg_trim;
1681 
1682 	nla_nest_end(skb, nest);
1683 
1684 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1685 
1686 	return skb->len;
1687 
1688 out_nlmsg_trim:
1689 	nlmsg_trim(skb, b);
1690 	return -1;
1691 }
1692 
1693 static int
1694 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1695 	       struct tc_action *actions[], size_t attr_size, int event,
1696 	       struct netlink_ext_ack *extack)
1697 {
1698 	struct sk_buff *skb;
1699 
1700 	skb = alloc_skb(max(attr_size, NLMSG_GOODSIZE), GFP_KERNEL);
1701 	if (!skb)
1702 		return -ENOBUFS;
1703 	if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1704 			 0, 1, NULL) <= 0) {
1705 		NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1706 		kfree_skb(skb);
1707 		return -EINVAL;
1708 	}
1709 
1710 	return rtnl_unicast(skb, net, portid);
1711 }
1712 
1713 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1714 					  struct nlmsghdr *n, u32 portid,
1715 					  struct netlink_ext_ack *extack)
1716 {
1717 	struct nlattr *tb[TCA_ACT_MAX + 1];
1718 	const struct tc_action_ops *ops;
1719 	struct tc_action *a;
1720 	int index;
1721 	int err;
1722 
1723 	err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1724 					  tcf_action_policy, extack);
1725 	if (err < 0)
1726 		goto err_out;
1727 
1728 	err = -EINVAL;
1729 	if (tb[TCA_ACT_INDEX] == NULL ||
1730 	    nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1731 		NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1732 		goto err_out;
1733 	}
1734 	index = nla_get_u32(tb[TCA_ACT_INDEX]);
1735 
1736 	err = -EINVAL;
1737 	ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1738 	if (!ops) { /* could happen in batch of actions */
1739 		NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1740 		goto err_out;
1741 	}
1742 	err = -ENOENT;
1743 	if (__tcf_idr_search(net, ops, &a, index) == 0) {
1744 		NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1745 		goto err_mod;
1746 	}
1747 
1748 	module_put(ops->owner);
1749 	return a;
1750 
1751 err_mod:
1752 	module_put(ops->owner);
1753 err_out:
1754 	return ERR_PTR(err);
1755 }
1756 
1757 static int tca_action_flush(struct net *net, struct nlattr *nla,
1758 			    struct nlmsghdr *n, u32 portid,
1759 			    struct netlink_ext_ack *extack)
1760 {
1761 	struct sk_buff *skb;
1762 	unsigned char *b;
1763 	struct nlmsghdr *nlh;
1764 	struct tcamsg *t;
1765 	struct netlink_callback dcb;
1766 	struct nlattr *nest;
1767 	struct nlattr *tb[TCA_ACT_MAX + 1];
1768 	const struct tc_action_ops *ops;
1769 	struct nlattr *kind;
1770 	int err = -ENOMEM;
1771 
1772 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1773 	if (!skb)
1774 		return err;
1775 
1776 	b = skb_tail_pointer(skb);
1777 
1778 	err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1779 					  tcf_action_policy, extack);
1780 	if (err < 0)
1781 		goto err_out;
1782 
1783 	err = -EINVAL;
1784 	kind = tb[TCA_ACT_KIND];
1785 	ops = tc_lookup_action(kind);
1786 	if (!ops) { /*some idjot trying to flush unknown action */
1787 		NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1788 		goto err_out;
1789 	}
1790 
1791 	nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1792 			sizeof(*t), 0);
1793 	if (!nlh) {
1794 		NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1795 		goto out_module_put;
1796 	}
1797 	t = nlmsg_data(nlh);
1798 	t->tca_family = AF_UNSPEC;
1799 	t->tca__pad1 = 0;
1800 	t->tca__pad2 = 0;
1801 
1802 	nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1803 	if (!nest) {
1804 		NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1805 		goto out_module_put;
1806 	}
1807 
1808 	err = __tcf_generic_walker(net, skb, &dcb, RTM_DELACTION, ops, extack);
1809 	if (err <= 0) {
1810 		nla_nest_cancel(skb, nest);
1811 		goto out_module_put;
1812 	}
1813 
1814 	nla_nest_end(skb, nest);
1815 
1816 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1817 	nlh->nlmsg_flags |= NLM_F_ROOT;
1818 	module_put(ops->owner);
1819 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1820 			     n->nlmsg_flags & NLM_F_ECHO);
1821 	if (err < 0)
1822 		NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1823 
1824 	return err;
1825 
1826 out_module_put:
1827 	module_put(ops->owner);
1828 err_out:
1829 	kfree_skb(skb);
1830 	return err;
1831 }
1832 
1833 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1834 {
1835 	struct tc_action *a;
1836 	int i;
1837 
1838 	tcf_act_for_each_action(i, a, actions) {
1839 		const struct tc_action_ops *ops = a->ops;
1840 		/* Actions can be deleted concurrently so we must save their
1841 		 * type and id to search again after reference is released.
1842 		 */
1843 		struct tcf_idrinfo *idrinfo = a->idrinfo;
1844 		u32 act_index = a->tcfa_index;
1845 
1846 		actions[i] = NULL;
1847 		if (tcf_action_put(a)) {
1848 			/* last reference, action was deleted concurrently */
1849 			module_put(ops->owner);
1850 		} else {
1851 			int ret;
1852 
1853 			/* now do the delete */
1854 			ret = tcf_idr_delete_index(idrinfo, act_index);
1855 			if (ret < 0)
1856 				return ret;
1857 		}
1858 	}
1859 	return 0;
1860 }
1861 
1862 static struct sk_buff *tcf_reoffload_del_notify_msg(struct net *net,
1863 						    struct tc_action *action)
1864 {
1865 	struct tc_action *actions[TCA_ACT_MAX_PRIO] = {
1866 		[0] = action,
1867 	};
1868 	struct sk_buff *skb;
1869 	size_t attr_size;
1870 
1871 	attr_size = tcf_action_full_attrs_size(tcf_action_fill_size(action));
1872 
1873 	skb = alloc_skb(max(attr_size, NLMSG_GOODSIZE), GFP_KERNEL);
1874 	if (!skb)
1875 		return ERR_PTR(-ENOBUFS);
1876 
1877 	if (tca_get_fill(skb, actions, 0, 0, 0, RTM_DELACTION, 0, 1, NULL) <= 0) {
1878 		kfree_skb(skb);
1879 		return ERR_PTR(-EINVAL);
1880 	}
1881 
1882 	return skb;
1883 }
1884 
1885 static int tcf_reoffload_del_notify(struct net *net, struct tc_action *action)
1886 {
1887 	const struct tc_action_ops *ops = action->ops;
1888 	struct sk_buff *skb = NULL;
1889 	int ret;
1890 
1891 	if (rtnl_notify_needed(net, 0, RTNLGRP_TC)) {
1892 		skb = tcf_reoffload_del_notify_msg(net, action);
1893 		/* The action has already lost its hardware instance and is
1894 		 * skip_sw, so it must be released whether or not the
1895 		 * notification can be built.  Drop the notification rather
1896 		 * than leave an action behind that processes no packets.
1897 		 */
1898 		if (IS_ERR(skb))
1899 			skb = NULL;
1900 	}
1901 
1902 	ret = tcf_idr_release_unsafe(action);
1903 	if (ret == ACT_P_DELETED) {
1904 		module_put(ops->owner);
1905 		ret = rtnetlink_maybe_send(skb, net, 0, RTNLGRP_TC, 0);
1906 	} else {
1907 		kfree_skb(skb);
1908 	}
1909 
1910 	return ret;
1911 }
1912 
1913 int tcf_action_reoffload_cb(flow_indr_block_bind_cb_t *cb,
1914 			    void *cb_priv, bool add)
1915 {
1916 	struct tc_act_pernet_id *id_ptr;
1917 	struct tcf_idrinfo *idrinfo;
1918 	struct tc_action_net *tn;
1919 	struct tc_action *p;
1920 	unsigned int act_id;
1921 	unsigned long tmp;
1922 	unsigned long id;
1923 	struct idr *idr;
1924 	struct net *net;
1925 	int ret;
1926 
1927 	if (!cb)
1928 		return -EINVAL;
1929 
1930 	down_read(&net_rwsem);
1931 	mutex_lock(&act_id_mutex);
1932 
1933 	for_each_net(net) {
1934 		list_for_each_entry(id_ptr, &act_pernet_id_list, list) {
1935 			act_id = id_ptr->id;
1936 			tn = net_generic(net, act_id);
1937 			if (!tn)
1938 				continue;
1939 			idrinfo = tn->idrinfo;
1940 			if (!idrinfo)
1941 				continue;
1942 
1943 			mutex_lock(&idrinfo->lock);
1944 			idr = &idrinfo->action_idr;
1945 			idr_for_each_entry_ul(idr, p, tmp, id) {
1946 				if (IS_ERR(p) || tc_act_bind(p->tcfa_flags))
1947 					continue;
1948 				if (add) {
1949 					tcf_action_offload_add_ex(p, NULL, cb,
1950 								  cb_priv);
1951 					continue;
1952 				}
1953 
1954 				/* cb unregister to update hw count */
1955 				ret = tcf_action_offload_del_ex(p, cb, cb_priv);
1956 				if (ret < 0)
1957 					continue;
1958 				if (tc_act_skip_sw(p->tcfa_flags) &&
1959 				    !tc_act_in_hw(p))
1960 					tcf_reoffload_del_notify(net, p);
1961 			}
1962 			mutex_unlock(&idrinfo->lock);
1963 		}
1964 	}
1965 	mutex_unlock(&act_id_mutex);
1966 	up_read(&net_rwsem);
1967 
1968 	return 0;
1969 }
1970 
1971 static struct sk_buff *tcf_del_notify_msg(struct net *net, struct nlmsghdr *n,
1972 					  struct tc_action *actions[],
1973 					  u32 portid, size_t attr_size,
1974 					  struct netlink_ext_ack *extack)
1975 {
1976 	struct sk_buff *skb;
1977 
1978 	skb = alloc_skb(max(attr_size, NLMSG_GOODSIZE), GFP_KERNEL);
1979 	if (!skb)
1980 		return ERR_PTR(-ENOBUFS);
1981 
1982 	if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1983 			 0, 2, extack) <= 0) {
1984 		NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1985 		kfree_skb(skb);
1986 		return ERR_PTR(-EINVAL);
1987 	}
1988 
1989 	return skb;
1990 }
1991 
1992 static int tcf_del_notify(struct net *net, struct nlmsghdr *n,
1993 			  struct tc_action *actions[], u32 portid,
1994 			  size_t attr_size, struct netlink_ext_ack *extack)
1995 {
1996 	struct sk_buff *skb;
1997 	int ret;
1998 
1999 	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) {
2000 		skb = NULL;
2001 	} else {
2002 		skb = tcf_del_notify_msg(net, n, actions, portid, attr_size,
2003 					 extack);
2004 		if (IS_ERR(skb))
2005 			return PTR_ERR(skb);
2006 	}
2007 
2008 	/* now do the delete */
2009 	ret = tcf_action_delete(net, actions);
2010 	if (ret < 0) {
2011 		NL_SET_ERR_MSG(extack, "Failed to delete TC action");
2012 		kfree_skb(skb);
2013 		return ret;
2014 	}
2015 
2016 	return rtnetlink_maybe_send(skb, net, portid, RTNLGRP_TC,
2017 				    n->nlmsg_flags & NLM_F_ECHO);
2018 }
2019 
2020 static int
2021 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
2022 	      u32 portid, int event, struct netlink_ext_ack *extack)
2023 {
2024 	int i, ret;
2025 	struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
2026 	struct tc_action *act;
2027 	size_t attr_size = 0;
2028 	struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
2029 
2030 	ret = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
2031 					  extack);
2032 	if (ret < 0)
2033 		return ret;
2034 
2035 	if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
2036 		if (tb[1])
2037 			return tca_action_flush(net, tb[1], n, portid, extack);
2038 
2039 		NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
2040 		return -EINVAL;
2041 	}
2042 
2043 	for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
2044 		act = tcf_action_get_1(net, tb[i], n, portid, extack);
2045 		if (IS_ERR(act)) {
2046 			ret = PTR_ERR(act);
2047 			goto err;
2048 		}
2049 		attr_size += tcf_action_fill_size(act);
2050 		actions[i - 1] = act;
2051 	}
2052 
2053 	attr_size = tcf_action_full_attrs_size(attr_size);
2054 
2055 	if (event == RTM_GETACTION)
2056 		ret = tcf_get_notify(net, portid, n, actions, attr_size, event,
2057 				     extack);
2058 	else { /* delete */
2059 		ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
2060 		if (ret)
2061 			goto err;
2062 		return 0;
2063 	}
2064 err:
2065 	tcf_action_put_many(actions);
2066 	return ret;
2067 }
2068 
2069 static struct sk_buff *tcf_add_notify_msg(struct net *net, struct nlmsghdr *n,
2070 					  struct tc_action *actions[],
2071 					  u32 portid, size_t attr_size,
2072 					  struct netlink_ext_ack *extack)
2073 {
2074 	struct sk_buff *skb;
2075 
2076 	skb = alloc_skb(max(attr_size, NLMSG_GOODSIZE), GFP_KERNEL);
2077 	if (!skb)
2078 		return ERR_PTR(-ENOBUFS);
2079 
2080 	if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
2081 			 RTM_NEWACTION, 0, 0, extack) <= 0) {
2082 		NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
2083 		kfree_skb(skb);
2084 		return ERR_PTR(-EINVAL);
2085 	}
2086 
2087 	return skb;
2088 }
2089 
2090 static int tcf_add_notify(struct net *net, struct nlmsghdr *n,
2091 			  struct tc_action *actions[], u32 portid,
2092 			  size_t attr_size, struct netlink_ext_ack *extack)
2093 {
2094 	struct sk_buff *skb;
2095 
2096 	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) {
2097 		skb = NULL;
2098 	} else {
2099 		skb = tcf_add_notify_msg(net, n, actions, portid, attr_size,
2100 					 extack);
2101 		if (IS_ERR(skb))
2102 			return PTR_ERR(skb);
2103 	}
2104 
2105 	return rtnetlink_maybe_send(skb, net, portid, RTNLGRP_TC,
2106 				    n->nlmsg_flags & NLM_F_ECHO);
2107 }
2108 
2109 static int tcf_action_add(struct net *net, struct nlattr *nla,
2110 			  struct nlmsghdr *n, u32 portid, u32 flags,
2111 			  struct netlink_ext_ack *extack)
2112 {
2113 	size_t attr_size = 0;
2114 	int loop, ret;
2115 	struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
2116 	int init_res[TCA_ACT_MAX_PRIO] = {};
2117 
2118 	for (loop = 0; loop < 10; loop++) {
2119 		ret = tcf_action_init(net, NULL, nla, NULL, actions, init_res,
2120 				      &attr_size, flags, 0, extack);
2121 		if (ret != -EAGAIN)
2122 			break;
2123 	}
2124 
2125 	if (ret < 0)
2126 		return ret;
2127 
2128 	ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
2129 
2130 	/* only put bound actions */
2131 	tca_put_bound_many(actions, init_res);
2132 
2133 	return ret;
2134 }
2135 
2136 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
2137 	[TCA_ROOT_FLAGS] = NLA_POLICY_BITFIELD32(TCA_ACT_FLAG_LARGE_DUMP_ON |
2138 						 TCA_ACT_FLAG_TERSE_DUMP),
2139 	[TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
2140 };
2141 
2142 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
2143 			 struct netlink_ext_ack *extack)
2144 {
2145 	struct net *net = sock_net(skb->sk);
2146 	struct nlattr *tca[TCA_ROOT_MAX + 1];
2147 	u32 portid = NETLINK_CB(skb).portid;
2148 	u32 flags = 0;
2149 	int ret = 0;
2150 
2151 	if ((n->nlmsg_type != RTM_GETACTION) &&
2152 	    !netlink_capable(skb, CAP_NET_ADMIN))
2153 		return -EPERM;
2154 
2155 	ret = nlmsg_parse_deprecated(n, sizeof(struct tcamsg), tca,
2156 				     TCA_ROOT_MAX, NULL, extack);
2157 	if (ret < 0)
2158 		return ret;
2159 
2160 	if (tca[TCA_ACT_TAB] == NULL) {
2161 		NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
2162 		return -EINVAL;
2163 	}
2164 
2165 	/* n->nlmsg_flags & NLM_F_CREATE */
2166 	switch (n->nlmsg_type) {
2167 	case RTM_NEWACTION:
2168 		/* we are going to assume all other flags
2169 		 * imply create only if it doesn't exist
2170 		 * Note that CREATE | EXCL implies that
2171 		 * but since we want avoid ambiguity (eg when flags
2172 		 * is zero) then just set this
2173 		 */
2174 		if (n->nlmsg_flags & NLM_F_REPLACE)
2175 			flags = TCA_ACT_FLAGS_REPLACE;
2176 		ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, flags,
2177 				     extack);
2178 		break;
2179 	case RTM_DELACTION:
2180 		ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
2181 				    portid, RTM_DELACTION, extack);
2182 		break;
2183 	case RTM_GETACTION:
2184 		ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
2185 				    portid, RTM_GETACTION, extack);
2186 		break;
2187 	default:
2188 		BUG();
2189 	}
2190 
2191 	return ret;
2192 }
2193 
2194 static struct nlattr *find_dump_kind(struct nlattr **nla)
2195 {
2196 	struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
2197 	struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
2198 	struct nlattr *kind;
2199 
2200 	tb1 = nla[TCA_ACT_TAB];
2201 	if (tb1 == NULL)
2202 		return NULL;
2203 
2204 	if (nla_parse_deprecated(tb, TCA_ACT_MAX_PRIO, nla_data(tb1), NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
2205 		return NULL;
2206 
2207 	if (tb[1] == NULL)
2208 		return NULL;
2209 	if (nla_parse_nested_deprecated(tb2, TCA_ACT_MAX, tb[1], tcf_action_policy, NULL) < 0)
2210 		return NULL;
2211 	kind = tb2[TCA_ACT_KIND];
2212 
2213 	return kind;
2214 }
2215 
2216 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
2217 {
2218 	struct net *net = sock_net(skb->sk);
2219 	struct nlmsghdr *nlh;
2220 	unsigned char *b = skb_tail_pointer(skb);
2221 	struct nlattr *nest;
2222 	struct tc_action_ops *a_o;
2223 	int ret = 0;
2224 	struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
2225 	struct nlattr *tb[TCA_ROOT_MAX + 1];
2226 	struct nlattr *count_attr = NULL;
2227 	unsigned long jiffy_since = 0;
2228 	struct nlattr *kind = NULL;
2229 	struct nla_bitfield32 bf;
2230 	u32 msecs_since = 0;
2231 	u32 act_count = 0;
2232 
2233 	ret = nlmsg_parse_deprecated(cb->nlh, sizeof(struct tcamsg), tb,
2234 				     TCA_ROOT_MAX, tcaa_policy, cb->extack);
2235 	if (ret < 0)
2236 		return ret;
2237 
2238 	kind = find_dump_kind(tb);
2239 	if (kind == NULL) {
2240 		pr_info("tc_dump_action: action bad kind\n");
2241 		return 0;
2242 	}
2243 
2244 	a_o = tc_lookup_action(kind);
2245 	if (a_o == NULL)
2246 		return 0;
2247 
2248 	cb->args[2] = 0;
2249 	if (tb[TCA_ROOT_FLAGS]) {
2250 		bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
2251 		cb->args[2] = bf.value;
2252 	}
2253 
2254 	if (tb[TCA_ROOT_TIME_DELTA]) {
2255 		msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
2256 	}
2257 
2258 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
2259 			cb->nlh->nlmsg_type, sizeof(*t), 0);
2260 	if (!nlh)
2261 		goto out_module_put;
2262 
2263 	if (msecs_since)
2264 		jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
2265 
2266 	t = nlmsg_data(nlh);
2267 	t->tca_family = AF_UNSPEC;
2268 	t->tca__pad1 = 0;
2269 	t->tca__pad2 = 0;
2270 	cb->args[3] = jiffy_since;
2271 	count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
2272 	if (!count_attr)
2273 		goto out_module_put;
2274 
2275 	nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
2276 	if (nest == NULL)
2277 		goto out_module_put;
2278 
2279 	ret = __tcf_generic_walker(net, skb, cb, RTM_GETACTION, a_o, NULL);
2280 	if (ret < 0)
2281 		goto out_module_put;
2282 
2283 	if (ret > 0) {
2284 		nla_nest_end(skb, nest);
2285 		ret = skb->len;
2286 		act_count = cb->args[1];
2287 		memcpy(nla_data(count_attr), &act_count, sizeof(u32));
2288 		cb->args[1] = 0;
2289 	} else
2290 		nlmsg_trim(skb, b);
2291 
2292 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2293 	if (NETLINK_CB(cb->skb).portid && ret)
2294 		nlh->nlmsg_flags |= NLM_F_MULTI;
2295 	module_put(a_o->owner);
2296 	return skb->len;
2297 
2298 out_module_put:
2299 	module_put(a_o->owner);
2300 	nlmsg_trim(skb, b);
2301 	return skb->len;
2302 }
2303 
2304 static const struct rtnl_msg_handler tc_action_rtnl_msg_handlers[] __initconst = {
2305 	{.msgtype = RTM_NEWACTION, .doit = tc_ctl_action},
2306 	{.msgtype = RTM_DELACTION, .doit = tc_ctl_action},
2307 	{.msgtype = RTM_GETACTION, .doit = tc_ctl_action,
2308 	 .dumpit = tc_dump_action},
2309 };
2310 
2311 static int __init tc_action_init(void)
2312 {
2313 	rtnl_register_many(tc_action_rtnl_msg_handlers);
2314 	return 0;
2315 }
2316 
2317 subsys_initcall(tc_action_init);
2318