Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/cls_api.c Packet classifier API.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Changes:
8 *
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10 */
11
12#include <linux/module.h>
13#include <linux/types.h>
14#include <linux/kernel.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/err.h>
18#include <linux/skbuff.h>
19#include <linux/init.h>
20#include <linux/kmod.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/jhash.h>
24#include <linux/rculist.h>
25#include <net/net_namespace.h>
26#include <net/sock.h>
27#include <net/netlink.h>
28#include <net/pkt_sched.h>
29#include <net/pkt_cls.h>
30#include <net/tc_act/tc_pedit.h>
31#include <net/tc_act/tc_mirred.h>
32#include <net/tc_act/tc_vlan.h>
33#include <net/tc_act/tc_tunnel_key.h>
34#include <net/tc_act/tc_csum.h>
35#include <net/tc_act/tc_gact.h>
36#include <net/tc_act/tc_police.h>
37#include <net/tc_act/tc_sample.h>
38#include <net/tc_act/tc_skbedit.h>
39#include <net/tc_act/tc_ct.h>
40#include <net/tc_act/tc_mpls.h>
41#include <net/tc_act/tc_gate.h>
42#include <net/flow_offload.h>
43#include <net/tc_wrapper.h>
44
45extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
46
47/* The list of all installed classifier types */
48static LIST_HEAD(tcf_proto_base);
49
50/* Protects list of registered TC modules. It is pure SMP lock. */
51static DEFINE_RWLOCK(cls_mod_lock);
52
53#ifdef CONFIG_NET_CLS_ACT
54DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
55EXPORT_SYMBOL(tc_skb_ext_tc);
56
57void tc_skb_ext_tc_enable(void)
58{
59 static_branch_inc(&tc_skb_ext_tc);
60}
61EXPORT_SYMBOL(tc_skb_ext_tc_enable);
62
63void tc_skb_ext_tc_disable(void)
64{
65 static_branch_dec(&tc_skb_ext_tc);
66}
67EXPORT_SYMBOL(tc_skb_ext_tc_disable);
68#endif
69
70static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
71{
72 return jhash_3words(tp->chain->index, tp->prio,
73 (__force __u32)tp->protocol, 0);
74}
75
76static void tcf_proto_signal_destroying(struct tcf_chain *chain,
77 struct tcf_proto *tp)
78{
79 struct tcf_block *block = chain->block;
80
81 mutex_lock(&block->proto_destroy_lock);
82 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
83 destroy_obj_hashfn(tp));
84 mutex_unlock(&block->proto_destroy_lock);
85}
86
87static bool tcf_proto_cmp(const struct tcf_proto *tp1,
88 const struct tcf_proto *tp2)
89{
90 return tp1->chain->index == tp2->chain->index &&
91 tp1->prio == tp2->prio &&
92 tp1->protocol == tp2->protocol;
93}
94
95static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
96 struct tcf_proto *tp)
97{
98 u32 hash = destroy_obj_hashfn(tp);
99 struct tcf_proto *iter;
100 bool found = false;
101
102 rcu_read_lock();
103 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
104 destroy_ht_node, hash) {
105 if (tcf_proto_cmp(tp, iter)) {
106 found = true;
107 break;
108 }
109 }
110 rcu_read_unlock();
111
112 return found;
113}
114
115static void
116tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
117{
118 struct tcf_block *block = chain->block;
119
120 mutex_lock(&block->proto_destroy_lock);
121 if (hash_hashed(&tp->destroy_ht_node))
122 hash_del_rcu(&tp->destroy_ht_node);
123 mutex_unlock(&block->proto_destroy_lock);
124}
125
126/* Find classifier type by string name */
127
128static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
129{
130 const struct tcf_proto_ops *t, *res = NULL;
131
132 if (kind) {
133 read_lock(&cls_mod_lock);
134 list_for_each_entry(t, &tcf_proto_base, head) {
135 if (strcmp(kind, t->kind) == 0) {
136 if (try_module_get(t->owner))
137 res = t;
138 break;
139 }
140 }
141 read_unlock(&cls_mod_lock);
142 }
143 return res;
144}
145
146static const struct tcf_proto_ops *
147tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
148 struct netlink_ext_ack *extack)
149{
150 const struct tcf_proto_ops *ops;
151
152 ops = __tcf_proto_lookup_ops(kind);
153 if (ops)
154 return ops;
155#ifdef CONFIG_MODULES
156 if (rtnl_held)
157 rtnl_unlock();
158 request_module("cls_%s", kind);
159 if (rtnl_held)
160 rtnl_lock();
161 ops = __tcf_proto_lookup_ops(kind);
162 /* We dropped the RTNL semaphore in order to perform
163 * the module load. So, even if we succeeded in loading
164 * the module we have to replay the request. We indicate
165 * this using -EAGAIN.
166 */
167 if (ops) {
168 module_put(ops->owner);
169 return ERR_PTR(-EAGAIN);
170 }
171#endif
172 NL_SET_ERR_MSG(extack, "TC classifier not found");
173 return ERR_PTR(-ENOENT);
174}
175
176/* Register(unregister) new classifier type */
177
178int register_tcf_proto_ops(struct tcf_proto_ops *ops)
179{
180 struct tcf_proto_ops *t;
181 int rc = -EEXIST;
182
183 write_lock(&cls_mod_lock);
184 list_for_each_entry(t, &tcf_proto_base, head)
185 if (!strcmp(ops->kind, t->kind))
186 goto out;
187
188 list_add_tail(&ops->head, &tcf_proto_base);
189 rc = 0;
190out:
191 write_unlock(&cls_mod_lock);
192 return rc;
193}
194EXPORT_SYMBOL(register_tcf_proto_ops);
195
196static struct workqueue_struct *tc_filter_wq;
197
198void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
199{
200 struct tcf_proto_ops *t;
201 int rc = -ENOENT;
202
203 /* Wait for outstanding call_rcu()s, if any, from a
204 * tcf_proto_ops's destroy() handler.
205 */
206 rcu_barrier();
207 flush_workqueue(tc_filter_wq);
208
209 write_lock(&cls_mod_lock);
210 list_for_each_entry(t, &tcf_proto_base, head) {
211 if (t == ops) {
212 list_del(&t->head);
213 rc = 0;
214 break;
215 }
216 }
217 write_unlock(&cls_mod_lock);
218
219 WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
220}
221EXPORT_SYMBOL(unregister_tcf_proto_ops);
222
223bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
224{
225 INIT_RCU_WORK(rwork, func);
226 return queue_rcu_work(tc_filter_wq, rwork);
227}
228EXPORT_SYMBOL(tcf_queue_work);
229
230/* Select new prio value from the range, managed by kernel. */
231
232static inline u32 tcf_auto_prio(struct tcf_proto *tp)
233{
234 u32 first = TC_H_MAKE(0xC0000000U, 0U);
235
236 if (tp)
237 first = tp->prio - 1;
238
239 return TC_H_MAJ(first);
240}
241
242static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
243{
244 if (kind)
245 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
246 memset(name, 0, IFNAMSIZ);
247 return false;
248}
249
250static bool tcf_proto_is_unlocked(const char *kind)
251{
252 const struct tcf_proto_ops *ops;
253 bool ret;
254
255 if (strlen(kind) == 0)
256 return false;
257
258 ops = tcf_proto_lookup_ops(kind, false, NULL);
259 /* On error return false to take rtnl lock. Proto lookup/create
260 * functions will perform lookup again and properly handle errors.
261 */
262 if (IS_ERR(ops))
263 return false;
264
265 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
266 module_put(ops->owner);
267 return ret;
268}
269
270static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
271 u32 prio, struct tcf_chain *chain,
272 bool rtnl_held,
273 struct netlink_ext_ack *extack)
274{
275 struct tcf_proto *tp;
276 int err;
277
278 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
279 if (!tp)
280 return ERR_PTR(-ENOBUFS);
281
282 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
283 if (IS_ERR(tp->ops)) {
284 err = PTR_ERR(tp->ops);
285 goto errout;
286 }
287 tp->classify = tp->ops->classify;
288 tp->protocol = protocol;
289 tp->prio = prio;
290 tp->chain = chain;
291 spin_lock_init(&tp->lock);
292 refcount_set(&tp->refcnt, 1);
293
294 err = tp->ops->init(tp);
295 if (err) {
296 module_put(tp->ops->owner);
297 goto errout;
298 }
299 return tp;
300
301errout:
302 kfree(tp);
303 return ERR_PTR(err);
304}
305
306static void tcf_proto_get(struct tcf_proto *tp)
307{
308 refcount_inc(&tp->refcnt);
309}
310
311static void tcf_chain_put(struct tcf_chain *chain);
312
313static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
314 bool sig_destroy, struct netlink_ext_ack *extack)
315{
316 tp->ops->destroy(tp, rtnl_held, extack);
317 if (sig_destroy)
318 tcf_proto_signal_destroyed(tp->chain, tp);
319 tcf_chain_put(tp->chain);
320 module_put(tp->ops->owner);
321 kfree_rcu(tp, rcu);
322}
323
324static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
325 struct netlink_ext_ack *extack)
326{
327 if (refcount_dec_and_test(&tp->refcnt))
328 tcf_proto_destroy(tp, rtnl_held, true, extack);
329}
330
331static bool tcf_proto_check_delete(struct tcf_proto *tp)
332{
333 if (tp->ops->delete_empty)
334 return tp->ops->delete_empty(tp);
335
336 tp->deleting = true;
337 return tp->deleting;
338}
339
340static void tcf_proto_mark_delete(struct tcf_proto *tp)
341{
342 spin_lock(&tp->lock);
343 tp->deleting = true;
344 spin_unlock(&tp->lock);
345}
346
347static bool tcf_proto_is_deleting(struct tcf_proto *tp)
348{
349 bool deleting;
350
351 spin_lock(&tp->lock);
352 deleting = tp->deleting;
353 spin_unlock(&tp->lock);
354
355 return deleting;
356}
357
358#define ASSERT_BLOCK_LOCKED(block) \
359 lockdep_assert_held(&(block)->lock)
360
361struct tcf_filter_chain_list_item {
362 struct list_head list;
363 tcf_chain_head_change_t *chain_head_change;
364 void *chain_head_change_priv;
365};
366
367static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
368 u32 chain_index)
369{
370 struct tcf_chain *chain;
371
372 ASSERT_BLOCK_LOCKED(block);
373
374 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
375 if (!chain)
376 return NULL;
377 list_add_tail_rcu(&chain->list, &block->chain_list);
378 mutex_init(&chain->filter_chain_lock);
379 chain->block = block;
380 chain->index = chain_index;
381 chain->refcnt = 1;
382 if (!chain->index)
383 block->chain0.chain = chain;
384 return chain;
385}
386
387static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
388 struct tcf_proto *tp_head)
389{
390 if (item->chain_head_change)
391 item->chain_head_change(tp_head, item->chain_head_change_priv);
392}
393
394static void tcf_chain0_head_change(struct tcf_chain *chain,
395 struct tcf_proto *tp_head)
396{
397 struct tcf_filter_chain_list_item *item;
398 struct tcf_block *block = chain->block;
399
400 if (chain->index)
401 return;
402
403 mutex_lock(&block->lock);
404 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
405 tcf_chain_head_change_item(item, tp_head);
406 mutex_unlock(&block->lock);
407}
408
409/* Returns true if block can be safely freed. */
410
411static bool tcf_chain_detach(struct tcf_chain *chain)
412{
413 struct tcf_block *block = chain->block;
414
415 ASSERT_BLOCK_LOCKED(block);
416
417 list_del_rcu(&chain->list);
418 if (!chain->index)
419 block->chain0.chain = NULL;
420
421 if (list_empty(&block->chain_list) &&
422 refcount_read(&block->refcnt) == 0)
423 return true;
424
425 return false;
426}
427
428static void tcf_block_destroy(struct tcf_block *block)
429{
430 mutex_destroy(&block->lock);
431 mutex_destroy(&block->proto_destroy_lock);
432 kfree_rcu(block, rcu);
433}
434
435static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
436{
437 struct tcf_block *block = chain->block;
438
439 mutex_destroy(&chain->filter_chain_lock);
440 kfree_rcu(chain, rcu);
441 if (free_block)
442 tcf_block_destroy(block);
443}
444
445static void tcf_chain_hold(struct tcf_chain *chain)
446{
447 ASSERT_BLOCK_LOCKED(chain->block);
448
449 ++chain->refcnt;
450}
451
452static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
453{
454 ASSERT_BLOCK_LOCKED(chain->block);
455
456 /* In case all the references are action references, this
457 * chain should not be shown to the user.
458 */
459 return chain->refcnt == chain->action_refcnt;
460}
461
462static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
463 u32 chain_index)
464{
465 struct tcf_chain *chain;
466
467 ASSERT_BLOCK_LOCKED(block);
468
469 list_for_each_entry(chain, &block->chain_list, list) {
470 if (chain->index == chain_index)
471 return chain;
472 }
473 return NULL;
474}
475
476#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
477static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
478 u32 chain_index)
479{
480 struct tcf_chain *chain;
481
482 list_for_each_entry_rcu(chain, &block->chain_list, list) {
483 if (chain->index == chain_index)
484 return chain;
485 }
486 return NULL;
487}
488#endif
489
490static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
491 u32 seq, u16 flags, int event, bool unicast);
492
493static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
494 u32 chain_index, bool create,
495 bool by_act)
496{
497 struct tcf_chain *chain = NULL;
498 bool is_first_reference;
499
500 mutex_lock(&block->lock);
501 chain = tcf_chain_lookup(block, chain_index);
502 if (chain) {
503 tcf_chain_hold(chain);
504 } else {
505 if (!create)
506 goto errout;
507 chain = tcf_chain_create(block, chain_index);
508 if (!chain)
509 goto errout;
510 }
511
512 if (by_act)
513 ++chain->action_refcnt;
514 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
515 mutex_unlock(&block->lock);
516
517 /* Send notification only in case we got the first
518 * non-action reference. Until then, the chain acts only as
519 * a placeholder for actions pointing to it and user ought
520 * not know about them.
521 */
522 if (is_first_reference && !by_act)
523 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
524 RTM_NEWCHAIN, false);
525
526 return chain;
527
528errout:
529 mutex_unlock(&block->lock);
530 return chain;
531}
532
533static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
534 bool create)
535{
536 return __tcf_chain_get(block, chain_index, create, false);
537}
538
539struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
540{
541 return __tcf_chain_get(block, chain_index, true, true);
542}
543EXPORT_SYMBOL(tcf_chain_get_by_act);
544
545static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
546 void *tmplt_priv);
547static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
548 void *tmplt_priv, u32 chain_index,
549 struct tcf_block *block, struct sk_buff *oskb,
550 u32 seq, u16 flags, bool unicast);
551
552static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
553 bool explicitly_created)
554{
555 struct tcf_block *block = chain->block;
556 const struct tcf_proto_ops *tmplt_ops;
557 bool free_block = false;
558 unsigned int refcnt;
559 void *tmplt_priv;
560
561 mutex_lock(&block->lock);
562 if (explicitly_created) {
563 if (!chain->explicitly_created) {
564 mutex_unlock(&block->lock);
565 return;
566 }
567 chain->explicitly_created = false;
568 }
569
570 if (by_act)
571 chain->action_refcnt--;
572
573 /* tc_chain_notify_delete can't be called while holding block lock.
574 * However, when block is unlocked chain can be changed concurrently, so
575 * save these to temporary variables.
576 */
577 refcnt = --chain->refcnt;
578 tmplt_ops = chain->tmplt_ops;
579 tmplt_priv = chain->tmplt_priv;
580
581 /* The last dropped non-action reference will trigger notification. */
582 if (refcnt - chain->action_refcnt == 0 && !by_act) {
583 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
584 block, NULL, 0, 0, false);
585 /* Last reference to chain, no need to lock. */
586 chain->flushing = false;
587 }
588
589 if (refcnt == 0)
590 free_block = tcf_chain_detach(chain);
591 mutex_unlock(&block->lock);
592
593 if (refcnt == 0) {
594 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
595 tcf_chain_destroy(chain, free_block);
596 }
597}
598
599static void tcf_chain_put(struct tcf_chain *chain)
600{
601 __tcf_chain_put(chain, false, false);
602}
603
604void tcf_chain_put_by_act(struct tcf_chain *chain)
605{
606 __tcf_chain_put(chain, true, false);
607}
608EXPORT_SYMBOL(tcf_chain_put_by_act);
609
610static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
611{
612 __tcf_chain_put(chain, false, true);
613}
614
615static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
616{
617 struct tcf_proto *tp, *tp_next;
618
619 mutex_lock(&chain->filter_chain_lock);
620 tp = tcf_chain_dereference(chain->filter_chain, chain);
621 while (tp) {
622 tp_next = rcu_dereference_protected(tp->next, 1);
623 tcf_proto_signal_destroying(chain, tp);
624 tp = tp_next;
625 }
626 tp = tcf_chain_dereference(chain->filter_chain, chain);
627 RCU_INIT_POINTER(chain->filter_chain, NULL);
628 tcf_chain0_head_change(chain, NULL);
629 chain->flushing = true;
630 mutex_unlock(&chain->filter_chain_lock);
631
632 while (tp) {
633 tp_next = rcu_dereference_protected(tp->next, 1);
634 tcf_proto_put(tp, rtnl_held, NULL);
635 tp = tp_next;
636 }
637}
638
639static int tcf_block_setup(struct tcf_block *block,
640 struct flow_block_offload *bo);
641
642static void tcf_block_offload_init(struct flow_block_offload *bo,
643 struct net_device *dev, struct Qdisc *sch,
644 enum flow_block_command command,
645 enum flow_block_binder_type binder_type,
646 struct flow_block *flow_block,
647 bool shared, struct netlink_ext_ack *extack)
648{
649 bo->net = dev_net(dev);
650 bo->command = command;
651 bo->binder_type = binder_type;
652 bo->block = flow_block;
653 bo->block_shared = shared;
654 bo->extack = extack;
655 bo->sch = sch;
656 bo->cb_list_head = &flow_block->cb_list;
657 INIT_LIST_HEAD(&bo->cb_list);
658}
659
660static void tcf_block_unbind(struct tcf_block *block,
661 struct flow_block_offload *bo);
662
663static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
664{
665 struct tcf_block *block = block_cb->indr.data;
666 struct net_device *dev = block_cb->indr.dev;
667 struct Qdisc *sch = block_cb->indr.sch;
668 struct netlink_ext_ack extack = {};
669 struct flow_block_offload bo = {};
670
671 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
672 block_cb->indr.binder_type,
673 &block->flow_block, tcf_block_shared(block),
674 &extack);
675 rtnl_lock();
676 down_write(&block->cb_lock);
677 list_del(&block_cb->driver_list);
678 list_move(&block_cb->list, &bo.cb_list);
679 tcf_block_unbind(block, &bo);
680 up_write(&block->cb_lock);
681 rtnl_unlock();
682}
683
684static bool tcf_block_offload_in_use(struct tcf_block *block)
685{
686 return atomic_read(&block->offloadcnt);
687}
688
689static int tcf_block_offload_cmd(struct tcf_block *block,
690 struct net_device *dev, struct Qdisc *sch,
691 struct tcf_block_ext_info *ei,
692 enum flow_block_command command,
693 struct netlink_ext_ack *extack)
694{
695 struct flow_block_offload bo = {};
696
697 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
698 &block->flow_block, tcf_block_shared(block),
699 extack);
700
701 if (dev->netdev_ops->ndo_setup_tc) {
702 int err;
703
704 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
705 if (err < 0) {
706 if (err != -EOPNOTSUPP)
707 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
708 return err;
709 }
710
711 return tcf_block_setup(block, &bo);
712 }
713
714 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
715 tc_block_indr_cleanup);
716 tcf_block_setup(block, &bo);
717
718 return -EOPNOTSUPP;
719}
720
721static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
722 struct tcf_block_ext_info *ei,
723 struct netlink_ext_ack *extack)
724{
725 struct net_device *dev = q->dev_queue->dev;
726 int err;
727
728 down_write(&block->cb_lock);
729
730 /* If tc offload feature is disabled and the block we try to bind
731 * to already has some offloaded filters, forbid to bind.
732 */
733 if (dev->netdev_ops->ndo_setup_tc &&
734 !tc_can_offload(dev) &&
735 tcf_block_offload_in_use(block)) {
736 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
737 err = -EOPNOTSUPP;
738 goto err_unlock;
739 }
740
741 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
742 if (err == -EOPNOTSUPP)
743 goto no_offload_dev_inc;
744 if (err)
745 goto err_unlock;
746
747 up_write(&block->cb_lock);
748 return 0;
749
750no_offload_dev_inc:
751 if (tcf_block_offload_in_use(block))
752 goto err_unlock;
753
754 err = 0;
755 block->nooffloaddevcnt++;
756err_unlock:
757 up_write(&block->cb_lock);
758 return err;
759}
760
761static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
762 struct tcf_block_ext_info *ei)
763{
764 struct net_device *dev = q->dev_queue->dev;
765 int err;
766
767 down_write(&block->cb_lock);
768 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
769 if (err == -EOPNOTSUPP)
770 goto no_offload_dev_dec;
771 up_write(&block->cb_lock);
772 return;
773
774no_offload_dev_dec:
775 WARN_ON(block->nooffloaddevcnt-- == 0);
776 up_write(&block->cb_lock);
777}
778
779static int
780tcf_chain0_head_change_cb_add(struct tcf_block *block,
781 struct tcf_block_ext_info *ei,
782 struct netlink_ext_ack *extack)
783{
784 struct tcf_filter_chain_list_item *item;
785 struct tcf_chain *chain0;
786
787 item = kmalloc(sizeof(*item), GFP_KERNEL);
788 if (!item) {
789 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
790 return -ENOMEM;
791 }
792 item->chain_head_change = ei->chain_head_change;
793 item->chain_head_change_priv = ei->chain_head_change_priv;
794
795 mutex_lock(&block->lock);
796 chain0 = block->chain0.chain;
797 if (chain0)
798 tcf_chain_hold(chain0);
799 else
800 list_add(&item->list, &block->chain0.filter_chain_list);
801 mutex_unlock(&block->lock);
802
803 if (chain0) {
804 struct tcf_proto *tp_head;
805
806 mutex_lock(&chain0->filter_chain_lock);
807
808 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
809 if (tp_head)
810 tcf_chain_head_change_item(item, tp_head);
811
812 mutex_lock(&block->lock);
813 list_add(&item->list, &block->chain0.filter_chain_list);
814 mutex_unlock(&block->lock);
815
816 mutex_unlock(&chain0->filter_chain_lock);
817 tcf_chain_put(chain0);
818 }
819
820 return 0;
821}
822
823static void
824tcf_chain0_head_change_cb_del(struct tcf_block *block,
825 struct tcf_block_ext_info *ei)
826{
827 struct tcf_filter_chain_list_item *item;
828
829 mutex_lock(&block->lock);
830 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
831 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
832 (item->chain_head_change == ei->chain_head_change &&
833 item->chain_head_change_priv == ei->chain_head_change_priv)) {
834 if (block->chain0.chain)
835 tcf_chain_head_change_item(item, NULL);
836 list_del(&item->list);
837 mutex_unlock(&block->lock);
838
839 kfree(item);
840 return;
841 }
842 }
843 mutex_unlock(&block->lock);
844 WARN_ON(1);
845}
846
847struct tcf_net {
848 spinlock_t idr_lock; /* Protects idr */
849 struct idr idr;
850};
851
852static unsigned int tcf_net_id;
853
854static int tcf_block_insert(struct tcf_block *block, struct net *net,
855 struct netlink_ext_ack *extack)
856{
857 struct tcf_net *tn = net_generic(net, tcf_net_id);
858 int err;
859
860 idr_preload(GFP_KERNEL);
861 spin_lock(&tn->idr_lock);
862 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
863 GFP_NOWAIT);
864 spin_unlock(&tn->idr_lock);
865 idr_preload_end();
866
867 return err;
868}
869
870static void tcf_block_remove(struct tcf_block *block, struct net *net)
871{
872 struct tcf_net *tn = net_generic(net, tcf_net_id);
873
874 spin_lock(&tn->idr_lock);
875 idr_remove(&tn->idr, block->index);
876 spin_unlock(&tn->idr_lock);
877}
878
879static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
880 u32 block_index,
881 struct netlink_ext_ack *extack)
882{
883 struct tcf_block *block;
884
885 block = kzalloc(sizeof(*block), GFP_KERNEL);
886 if (!block) {
887 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
888 return ERR_PTR(-ENOMEM);
889 }
890 mutex_init(&block->lock);
891 mutex_init(&block->proto_destroy_lock);
892 init_rwsem(&block->cb_lock);
893 flow_block_init(&block->flow_block);
894 INIT_LIST_HEAD(&block->chain_list);
895 INIT_LIST_HEAD(&block->owner_list);
896 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
897
898 refcount_set(&block->refcnt, 1);
899 block->net = net;
900 block->index = block_index;
901
902 /* Don't store q pointer for blocks which are shared */
903 if (!tcf_block_shared(block))
904 block->q = q;
905 return block;
906}
907
908static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
909{
910 struct tcf_net *tn = net_generic(net, tcf_net_id);
911
912 return idr_find(&tn->idr, block_index);
913}
914
915static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
916{
917 struct tcf_block *block;
918
919 rcu_read_lock();
920 block = tcf_block_lookup(net, block_index);
921 if (block && !refcount_inc_not_zero(&block->refcnt))
922 block = NULL;
923 rcu_read_unlock();
924
925 return block;
926}
927
928static struct tcf_chain *
929__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
930{
931 mutex_lock(&block->lock);
932 if (chain)
933 chain = list_is_last(&chain->list, &block->chain_list) ?
934 NULL : list_next_entry(chain, list);
935 else
936 chain = list_first_entry_or_null(&block->chain_list,
937 struct tcf_chain, list);
938
939 /* skip all action-only chains */
940 while (chain && tcf_chain_held_by_acts_only(chain))
941 chain = list_is_last(&chain->list, &block->chain_list) ?
942 NULL : list_next_entry(chain, list);
943
944 if (chain)
945 tcf_chain_hold(chain);
946 mutex_unlock(&block->lock);
947
948 return chain;
949}
950
951/* Function to be used by all clients that want to iterate over all chains on
952 * block. It properly obtains block->lock and takes reference to chain before
953 * returning it. Users of this function must be tolerant to concurrent chain
954 * insertion/deletion or ensure that no concurrent chain modification is
955 * possible. Note that all netlink dump callbacks cannot guarantee to provide
956 * consistent dump because rtnl lock is released each time skb is filled with
957 * data and sent to user-space.
958 */
959
960struct tcf_chain *
961tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
962{
963 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
964
965 if (chain)
966 tcf_chain_put(chain);
967
968 return chain_next;
969}
970EXPORT_SYMBOL(tcf_get_next_chain);
971
972static struct tcf_proto *
973__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
974{
975 u32 prio = 0;
976
977 ASSERT_RTNL();
978 mutex_lock(&chain->filter_chain_lock);
979
980 if (!tp) {
981 tp = tcf_chain_dereference(chain->filter_chain, chain);
982 } else if (tcf_proto_is_deleting(tp)) {
983 /* 'deleting' flag is set and chain->filter_chain_lock was
984 * unlocked, which means next pointer could be invalid. Restart
985 * search.
986 */
987 prio = tp->prio + 1;
988 tp = tcf_chain_dereference(chain->filter_chain, chain);
989
990 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
991 if (!tp->deleting && tp->prio >= prio)
992 break;
993 } else {
994 tp = tcf_chain_dereference(tp->next, chain);
995 }
996
997 if (tp)
998 tcf_proto_get(tp);
999
1000 mutex_unlock(&chain->filter_chain_lock);
1001
1002 return tp;
1003}
1004
1005/* Function to be used by all clients that want to iterate over all tp's on
1006 * chain. Users of this function must be tolerant to concurrent tp
1007 * insertion/deletion or ensure that no concurrent chain modification is
1008 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1009 * consistent dump because rtnl lock is released each time skb is filled with
1010 * data and sent to user-space.
1011 */
1012
1013struct tcf_proto *
1014tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1015{
1016 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1017
1018 if (tp)
1019 tcf_proto_put(tp, true, NULL);
1020
1021 return tp_next;
1022}
1023EXPORT_SYMBOL(tcf_get_next_proto);
1024
1025static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1026{
1027 struct tcf_chain *chain;
1028
1029 /* Last reference to block. At this point chains cannot be added or
1030 * removed concurrently.
1031 */
1032 for (chain = tcf_get_next_chain(block, NULL);
1033 chain;
1034 chain = tcf_get_next_chain(block, chain)) {
1035 tcf_chain_put_explicitly_created(chain);
1036 tcf_chain_flush(chain, rtnl_held);
1037 }
1038}
1039
1040/* Lookup Qdisc and increments its reference counter.
1041 * Set parent, if necessary.
1042 */
1043
1044static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1045 u32 *parent, int ifindex, bool rtnl_held,
1046 struct netlink_ext_ack *extack)
1047{
1048 const struct Qdisc_class_ops *cops;
1049 struct net_device *dev;
1050 int err = 0;
1051
1052 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1053 return 0;
1054
1055 rcu_read_lock();
1056
1057 /* Find link */
1058 dev = dev_get_by_index_rcu(net, ifindex);
1059 if (!dev) {
1060 rcu_read_unlock();
1061 return -ENODEV;
1062 }
1063
1064 /* Find qdisc */
1065 if (!*parent) {
1066 *q = rcu_dereference(dev->qdisc);
1067 *parent = (*q)->handle;
1068 } else {
1069 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1070 if (!*q) {
1071 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1072 err = -EINVAL;
1073 goto errout_rcu;
1074 }
1075 }
1076
1077 *q = qdisc_refcount_inc_nz(*q);
1078 if (!*q) {
1079 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1080 err = -EINVAL;
1081 goto errout_rcu;
1082 }
1083
1084 /* Is it classful? */
1085 cops = (*q)->ops->cl_ops;
1086 if (!cops) {
1087 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1088 err = -EINVAL;
1089 goto errout_qdisc;
1090 }
1091
1092 if (!cops->tcf_block) {
1093 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1094 err = -EOPNOTSUPP;
1095 goto errout_qdisc;
1096 }
1097
1098errout_rcu:
1099 /* At this point we know that qdisc is not noop_qdisc,
1100 * which means that qdisc holds a reference to net_device
1101 * and we hold a reference to qdisc, so it is safe to release
1102 * rcu read lock.
1103 */
1104 rcu_read_unlock();
1105 return err;
1106
1107errout_qdisc:
1108 rcu_read_unlock();
1109
1110 if (rtnl_held)
1111 qdisc_put(*q);
1112 else
1113 qdisc_put_unlocked(*q);
1114 *q = NULL;
1115
1116 return err;
1117}
1118
1119static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1120 int ifindex, struct netlink_ext_ack *extack)
1121{
1122 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1123 return 0;
1124
1125 /* Do we search for filter, attached to class? */
1126 if (TC_H_MIN(parent)) {
1127 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1128
1129 *cl = cops->find(q, parent);
1130 if (*cl == 0) {
1131 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1132 return -ENOENT;
1133 }
1134 }
1135
1136 return 0;
1137}
1138
1139static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1140 unsigned long cl, int ifindex,
1141 u32 block_index,
1142 struct netlink_ext_ack *extack)
1143{
1144 struct tcf_block *block;
1145
1146 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1147 block = tcf_block_refcnt_get(net, block_index);
1148 if (!block) {
1149 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1150 return ERR_PTR(-EINVAL);
1151 }
1152 } else {
1153 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1154
1155 block = cops->tcf_block(q, cl, extack);
1156 if (!block)
1157 return ERR_PTR(-EINVAL);
1158
1159 if (tcf_block_shared(block)) {
1160 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1161 return ERR_PTR(-EOPNOTSUPP);
1162 }
1163
1164 /* Always take reference to block in order to support execution
1165 * of rules update path of cls API without rtnl lock. Caller
1166 * must release block when it is finished using it. 'if' block
1167 * of this conditional obtain reference to block by calling
1168 * tcf_block_refcnt_get().
1169 */
1170 refcount_inc(&block->refcnt);
1171 }
1172
1173 return block;
1174}
1175
1176static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1177 struct tcf_block_ext_info *ei, bool rtnl_held)
1178{
1179 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1180 /* Flushing/putting all chains will cause the block to be
1181 * deallocated when last chain is freed. However, if chain_list
1182 * is empty, block has to be manually deallocated. After block
1183 * reference counter reached 0, it is no longer possible to
1184 * increment it or add new chains to block.
1185 */
1186 bool free_block = list_empty(&block->chain_list);
1187
1188 mutex_unlock(&block->lock);
1189 if (tcf_block_shared(block))
1190 tcf_block_remove(block, block->net);
1191
1192 if (q)
1193 tcf_block_offload_unbind(block, q, ei);
1194
1195 if (free_block)
1196 tcf_block_destroy(block);
1197 else
1198 tcf_block_flush_all_chains(block, rtnl_held);
1199 } else if (q) {
1200 tcf_block_offload_unbind(block, q, ei);
1201 }
1202}
1203
1204static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1205{
1206 __tcf_block_put(block, NULL, NULL, rtnl_held);
1207}
1208
1209/* Find tcf block.
1210 * Set q, parent, cl when appropriate.
1211 */
1212
1213static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1214 u32 *parent, unsigned long *cl,
1215 int ifindex, u32 block_index,
1216 struct netlink_ext_ack *extack)
1217{
1218 struct tcf_block *block;
1219 int err = 0;
1220
1221 ASSERT_RTNL();
1222
1223 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1224 if (err)
1225 goto errout;
1226
1227 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1228 if (err)
1229 goto errout_qdisc;
1230
1231 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1232 if (IS_ERR(block)) {
1233 err = PTR_ERR(block);
1234 goto errout_qdisc;
1235 }
1236
1237 return block;
1238
1239errout_qdisc:
1240 if (*q)
1241 qdisc_put(*q);
1242errout:
1243 *q = NULL;
1244 return ERR_PTR(err);
1245}
1246
1247static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1248 bool rtnl_held)
1249{
1250 if (!IS_ERR_OR_NULL(block))
1251 tcf_block_refcnt_put(block, rtnl_held);
1252
1253 if (q) {
1254 if (rtnl_held)
1255 qdisc_put(q);
1256 else
1257 qdisc_put_unlocked(q);
1258 }
1259}
1260
1261struct tcf_block_owner_item {
1262 struct list_head list;
1263 struct Qdisc *q;
1264 enum flow_block_binder_type binder_type;
1265};
1266
1267static void
1268tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1269 struct Qdisc *q,
1270 enum flow_block_binder_type binder_type)
1271{
1272 if (block->keep_dst &&
1273 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1274 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1275 netif_keep_dst(qdisc_dev(q));
1276}
1277
1278void tcf_block_netif_keep_dst(struct tcf_block *block)
1279{
1280 struct tcf_block_owner_item *item;
1281
1282 block->keep_dst = true;
1283 list_for_each_entry(item, &block->owner_list, list)
1284 tcf_block_owner_netif_keep_dst(block, item->q,
1285 item->binder_type);
1286}
1287EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1288
1289static int tcf_block_owner_add(struct tcf_block *block,
1290 struct Qdisc *q,
1291 enum flow_block_binder_type binder_type)
1292{
1293 struct tcf_block_owner_item *item;
1294
1295 item = kmalloc(sizeof(*item), GFP_KERNEL);
1296 if (!item)
1297 return -ENOMEM;
1298 item->q = q;
1299 item->binder_type = binder_type;
1300 list_add(&item->list, &block->owner_list);
1301 return 0;
1302}
1303
1304static void tcf_block_owner_del(struct tcf_block *block,
1305 struct Qdisc *q,
1306 enum flow_block_binder_type binder_type)
1307{
1308 struct tcf_block_owner_item *item;
1309
1310 list_for_each_entry(item, &block->owner_list, list) {
1311 if (item->q == q && item->binder_type == binder_type) {
1312 list_del(&item->list);
1313 kfree(item);
1314 return;
1315 }
1316 }
1317 WARN_ON(1);
1318}
1319
1320int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1321 struct tcf_block_ext_info *ei,
1322 struct netlink_ext_ack *extack)
1323{
1324 struct net *net = qdisc_net(q);
1325 struct tcf_block *block = NULL;
1326 int err;
1327
1328 if (ei->block_index)
1329 /* block_index not 0 means the shared block is requested */
1330 block = tcf_block_refcnt_get(net, ei->block_index);
1331
1332 if (!block) {
1333 block = tcf_block_create(net, q, ei->block_index, extack);
1334 if (IS_ERR(block))
1335 return PTR_ERR(block);
1336 if (tcf_block_shared(block)) {
1337 err = tcf_block_insert(block, net, extack);
1338 if (err)
1339 goto err_block_insert;
1340 }
1341 }
1342
1343 err = tcf_block_owner_add(block, q, ei->binder_type);
1344 if (err)
1345 goto err_block_owner_add;
1346
1347 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1348
1349 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1350 if (err)
1351 goto err_chain0_head_change_cb_add;
1352
1353 err = tcf_block_offload_bind(block, q, ei, extack);
1354 if (err)
1355 goto err_block_offload_bind;
1356
1357 *p_block = block;
1358 return 0;
1359
1360err_block_offload_bind:
1361 tcf_chain0_head_change_cb_del(block, ei);
1362err_chain0_head_change_cb_add:
1363 tcf_block_owner_del(block, q, ei->binder_type);
1364err_block_owner_add:
1365err_block_insert:
1366 tcf_block_refcnt_put(block, true);
1367 return err;
1368}
1369EXPORT_SYMBOL(tcf_block_get_ext);
1370
1371static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1372{
1373 struct tcf_proto __rcu **p_filter_chain = priv;
1374
1375 rcu_assign_pointer(*p_filter_chain, tp_head);
1376}
1377
1378int tcf_block_get(struct tcf_block **p_block,
1379 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1380 struct netlink_ext_ack *extack)
1381{
1382 struct tcf_block_ext_info ei = {
1383 .chain_head_change = tcf_chain_head_change_dflt,
1384 .chain_head_change_priv = p_filter_chain,
1385 };
1386
1387 WARN_ON(!p_filter_chain);
1388 return tcf_block_get_ext(p_block, q, &ei, extack);
1389}
1390EXPORT_SYMBOL(tcf_block_get);
1391
1392/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1393 * actions should be all removed after flushing.
1394 */
1395void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1396 struct tcf_block_ext_info *ei)
1397{
1398 if (!block)
1399 return;
1400 tcf_chain0_head_change_cb_del(block, ei);
1401 tcf_block_owner_del(block, q, ei->binder_type);
1402
1403 __tcf_block_put(block, q, ei, true);
1404}
1405EXPORT_SYMBOL(tcf_block_put_ext);
1406
1407void tcf_block_put(struct tcf_block *block)
1408{
1409 struct tcf_block_ext_info ei = {0, };
1410
1411 if (!block)
1412 return;
1413 tcf_block_put_ext(block, block->q, &ei);
1414}
1415
1416EXPORT_SYMBOL(tcf_block_put);
1417
1418static int
1419tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1420 void *cb_priv, bool add, bool offload_in_use,
1421 struct netlink_ext_ack *extack)
1422{
1423 struct tcf_chain *chain, *chain_prev;
1424 struct tcf_proto *tp, *tp_prev;
1425 int err;
1426
1427 lockdep_assert_held(&block->cb_lock);
1428
1429 for (chain = __tcf_get_next_chain(block, NULL);
1430 chain;
1431 chain_prev = chain,
1432 chain = __tcf_get_next_chain(block, chain),
1433 tcf_chain_put(chain_prev)) {
1434 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1435 tp_prev = tp,
1436 tp = __tcf_get_next_proto(chain, tp),
1437 tcf_proto_put(tp_prev, true, NULL)) {
1438 if (tp->ops->reoffload) {
1439 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1440 extack);
1441 if (err && add)
1442 goto err_playback_remove;
1443 } else if (add && offload_in_use) {
1444 err = -EOPNOTSUPP;
1445 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1446 goto err_playback_remove;
1447 }
1448 }
1449 }
1450
1451 return 0;
1452
1453err_playback_remove:
1454 tcf_proto_put(tp, true, NULL);
1455 tcf_chain_put(chain);
1456 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1457 extack);
1458 return err;
1459}
1460
1461static int tcf_block_bind(struct tcf_block *block,
1462 struct flow_block_offload *bo)
1463{
1464 struct flow_block_cb *block_cb, *next;
1465 int err, i = 0;
1466
1467 lockdep_assert_held(&block->cb_lock);
1468
1469 list_for_each_entry(block_cb, &bo->cb_list, list) {
1470 err = tcf_block_playback_offloads(block, block_cb->cb,
1471 block_cb->cb_priv, true,
1472 tcf_block_offload_in_use(block),
1473 bo->extack);
1474 if (err)
1475 goto err_unroll;
1476 if (!bo->unlocked_driver_cb)
1477 block->lockeddevcnt++;
1478
1479 i++;
1480 }
1481 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1482
1483 return 0;
1484
1485err_unroll:
1486 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1487 if (i-- > 0) {
1488 list_del(&block_cb->list);
1489 tcf_block_playback_offloads(block, block_cb->cb,
1490 block_cb->cb_priv, false,
1491 tcf_block_offload_in_use(block),
1492 NULL);
1493 if (!bo->unlocked_driver_cb)
1494 block->lockeddevcnt--;
1495 }
1496 flow_block_cb_free(block_cb);
1497 }
1498
1499 return err;
1500}
1501
1502static void tcf_block_unbind(struct tcf_block *block,
1503 struct flow_block_offload *bo)
1504{
1505 struct flow_block_cb *block_cb, *next;
1506
1507 lockdep_assert_held(&block->cb_lock);
1508
1509 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1510 tcf_block_playback_offloads(block, block_cb->cb,
1511 block_cb->cb_priv, false,
1512 tcf_block_offload_in_use(block),
1513 NULL);
1514 list_del(&block_cb->list);
1515 flow_block_cb_free(block_cb);
1516 if (!bo->unlocked_driver_cb)
1517 block->lockeddevcnt--;
1518 }
1519}
1520
1521static int tcf_block_setup(struct tcf_block *block,
1522 struct flow_block_offload *bo)
1523{
1524 int err;
1525
1526 switch (bo->command) {
1527 case FLOW_BLOCK_BIND:
1528 err = tcf_block_bind(block, bo);
1529 break;
1530 case FLOW_BLOCK_UNBIND:
1531 err = 0;
1532 tcf_block_unbind(block, bo);
1533 break;
1534 default:
1535 WARN_ON_ONCE(1);
1536 err = -EOPNOTSUPP;
1537 }
1538
1539 return err;
1540}
1541
1542/* Main classifier routine: scans classifier chain attached
1543 * to this qdisc, (optionally) tests for protocol and asks
1544 * specific classifiers.
1545 */
1546static inline int __tcf_classify(struct sk_buff *skb,
1547 const struct tcf_proto *tp,
1548 const struct tcf_proto *orig_tp,
1549 struct tcf_result *res,
1550 bool compat_mode,
1551 u32 *last_executed_chain)
1552{
1553#ifdef CONFIG_NET_CLS_ACT
1554 const int max_reclassify_loop = 16;
1555 const struct tcf_proto *first_tp;
1556 int limit = 0;
1557
1558reclassify:
1559#endif
1560 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1561 __be16 protocol = skb_protocol(skb, false);
1562 int err;
1563
1564 if (tp->protocol != protocol &&
1565 tp->protocol != htons(ETH_P_ALL))
1566 continue;
1567
1568 err = tc_classify(skb, tp, res);
1569#ifdef CONFIG_NET_CLS_ACT
1570 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1571 first_tp = orig_tp;
1572 *last_executed_chain = first_tp->chain->index;
1573 goto reset;
1574 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1575 first_tp = res->goto_tp;
1576 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1577 goto reset;
1578 }
1579#endif
1580 if (err >= 0)
1581 return err;
1582 }
1583
1584 return TC_ACT_UNSPEC; /* signal: continue lookup */
1585#ifdef CONFIG_NET_CLS_ACT
1586reset:
1587 if (unlikely(limit++ >= max_reclassify_loop)) {
1588 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1589 tp->chain->block->index,
1590 tp->prio & 0xffff,
1591 ntohs(tp->protocol));
1592 return TC_ACT_SHOT;
1593 }
1594
1595 tp = first_tp;
1596 goto reclassify;
1597#endif
1598}
1599
1600int tcf_classify(struct sk_buff *skb,
1601 const struct tcf_block *block,
1602 const struct tcf_proto *tp,
1603 struct tcf_result *res, bool compat_mode)
1604{
1605#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1606 u32 last_executed_chain = 0;
1607
1608 return __tcf_classify(skb, tp, tp, res, compat_mode,
1609 &last_executed_chain);
1610#else
1611 u32 last_executed_chain = tp ? tp->chain->index : 0;
1612 const struct tcf_proto *orig_tp = tp;
1613 struct tc_skb_ext *ext;
1614 int ret;
1615
1616 if (block) {
1617 ext = skb_ext_find(skb, TC_SKB_EXT);
1618
1619 if (ext && ext->chain) {
1620 struct tcf_chain *fchain;
1621
1622 fchain = tcf_chain_lookup_rcu(block, ext->chain);
1623 if (!fchain)
1624 return TC_ACT_SHOT;
1625
1626 /* Consume, so cloned/redirect skbs won't inherit ext */
1627 skb_ext_del(skb, TC_SKB_EXT);
1628
1629 tp = rcu_dereference_bh(fchain->filter_chain);
1630 last_executed_chain = fchain->index;
1631 }
1632 }
1633
1634 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1635 &last_executed_chain);
1636
1637 if (tc_skb_ext_tc_enabled()) {
1638 /* If we missed on some chain */
1639 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1640 struct tc_skb_cb *cb = tc_skb_cb(skb);
1641
1642 ext = tc_skb_ext_alloc(skb);
1643 if (WARN_ON_ONCE(!ext))
1644 return TC_ACT_SHOT;
1645 ext->chain = last_executed_chain;
1646 ext->mru = cb->mru;
1647 ext->post_ct = cb->post_ct;
1648 ext->post_ct_snat = cb->post_ct_snat;
1649 ext->post_ct_dnat = cb->post_ct_dnat;
1650 ext->zone = cb->zone;
1651 }
1652 }
1653
1654 return ret;
1655#endif
1656}
1657EXPORT_SYMBOL(tcf_classify);
1658
1659struct tcf_chain_info {
1660 struct tcf_proto __rcu **pprev;
1661 struct tcf_proto __rcu *next;
1662};
1663
1664static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1665 struct tcf_chain_info *chain_info)
1666{
1667 return tcf_chain_dereference(*chain_info->pprev, chain);
1668}
1669
1670static int tcf_chain_tp_insert(struct tcf_chain *chain,
1671 struct tcf_chain_info *chain_info,
1672 struct tcf_proto *tp)
1673{
1674 if (chain->flushing)
1675 return -EAGAIN;
1676
1677 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1678 if (*chain_info->pprev == chain->filter_chain)
1679 tcf_chain0_head_change(chain, tp);
1680 tcf_proto_get(tp);
1681 rcu_assign_pointer(*chain_info->pprev, tp);
1682
1683 return 0;
1684}
1685
1686static void tcf_chain_tp_remove(struct tcf_chain *chain,
1687 struct tcf_chain_info *chain_info,
1688 struct tcf_proto *tp)
1689{
1690 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1691
1692 tcf_proto_mark_delete(tp);
1693 if (tp == chain->filter_chain)
1694 tcf_chain0_head_change(chain, next);
1695 RCU_INIT_POINTER(*chain_info->pprev, next);
1696}
1697
1698static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1699 struct tcf_chain_info *chain_info,
1700 u32 protocol, u32 prio,
1701 bool prio_allocate);
1702
1703/* Try to insert new proto.
1704 * If proto with specified priority already exists, free new proto
1705 * and return existing one.
1706 */
1707
1708static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1709 struct tcf_proto *tp_new,
1710 u32 protocol, u32 prio,
1711 bool rtnl_held)
1712{
1713 struct tcf_chain_info chain_info;
1714 struct tcf_proto *tp;
1715 int err = 0;
1716
1717 mutex_lock(&chain->filter_chain_lock);
1718
1719 if (tcf_proto_exists_destroying(chain, tp_new)) {
1720 mutex_unlock(&chain->filter_chain_lock);
1721 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1722 return ERR_PTR(-EAGAIN);
1723 }
1724
1725 tp = tcf_chain_tp_find(chain, &chain_info,
1726 protocol, prio, false);
1727 if (!tp)
1728 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1729 mutex_unlock(&chain->filter_chain_lock);
1730
1731 if (tp) {
1732 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1733 tp_new = tp;
1734 } else if (err) {
1735 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1736 tp_new = ERR_PTR(err);
1737 }
1738
1739 return tp_new;
1740}
1741
1742static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1743 struct tcf_proto *tp, bool rtnl_held,
1744 struct netlink_ext_ack *extack)
1745{
1746 struct tcf_chain_info chain_info;
1747 struct tcf_proto *tp_iter;
1748 struct tcf_proto **pprev;
1749 struct tcf_proto *next;
1750
1751 mutex_lock(&chain->filter_chain_lock);
1752
1753 /* Atomically find and remove tp from chain. */
1754 for (pprev = &chain->filter_chain;
1755 (tp_iter = tcf_chain_dereference(*pprev, chain));
1756 pprev = &tp_iter->next) {
1757 if (tp_iter == tp) {
1758 chain_info.pprev = pprev;
1759 chain_info.next = tp_iter->next;
1760 WARN_ON(tp_iter->deleting);
1761 break;
1762 }
1763 }
1764 /* Verify that tp still exists and no new filters were inserted
1765 * concurrently.
1766 * Mark tp for deletion if it is empty.
1767 */
1768 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1769 mutex_unlock(&chain->filter_chain_lock);
1770 return;
1771 }
1772
1773 tcf_proto_signal_destroying(chain, tp);
1774 next = tcf_chain_dereference(chain_info.next, chain);
1775 if (tp == chain->filter_chain)
1776 tcf_chain0_head_change(chain, next);
1777 RCU_INIT_POINTER(*chain_info.pprev, next);
1778 mutex_unlock(&chain->filter_chain_lock);
1779
1780 tcf_proto_put(tp, rtnl_held, extack);
1781}
1782
1783static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1784 struct tcf_chain_info *chain_info,
1785 u32 protocol, u32 prio,
1786 bool prio_allocate)
1787{
1788 struct tcf_proto **pprev;
1789 struct tcf_proto *tp;
1790
1791 /* Check the chain for existence of proto-tcf with this priority */
1792 for (pprev = &chain->filter_chain;
1793 (tp = tcf_chain_dereference(*pprev, chain));
1794 pprev = &tp->next) {
1795 if (tp->prio >= prio) {
1796 if (tp->prio == prio) {
1797 if (prio_allocate ||
1798 (tp->protocol != protocol && protocol))
1799 return ERR_PTR(-EINVAL);
1800 } else {
1801 tp = NULL;
1802 }
1803 break;
1804 }
1805 }
1806 chain_info->pprev = pprev;
1807 if (tp) {
1808 chain_info->next = tp->next;
1809 tcf_proto_get(tp);
1810 } else {
1811 chain_info->next = NULL;
1812 }
1813 return tp;
1814}
1815
1816static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1817 struct tcf_proto *tp, struct tcf_block *block,
1818 struct Qdisc *q, u32 parent, void *fh,
1819 u32 portid, u32 seq, u16 flags, int event,
1820 bool terse_dump, bool rtnl_held)
1821{
1822 struct tcmsg *tcm;
1823 struct nlmsghdr *nlh;
1824 unsigned char *b = skb_tail_pointer(skb);
1825
1826 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1827 if (!nlh)
1828 goto out_nlmsg_trim;
1829 tcm = nlmsg_data(nlh);
1830 tcm->tcm_family = AF_UNSPEC;
1831 tcm->tcm__pad1 = 0;
1832 tcm->tcm__pad2 = 0;
1833 if (q) {
1834 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1835 tcm->tcm_parent = parent;
1836 } else {
1837 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1838 tcm->tcm_block_index = block->index;
1839 }
1840 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1841 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1842 goto nla_put_failure;
1843 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1844 goto nla_put_failure;
1845 if (!fh) {
1846 tcm->tcm_handle = 0;
1847 } else if (terse_dump) {
1848 if (tp->ops->terse_dump) {
1849 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1850 rtnl_held) < 0)
1851 goto nla_put_failure;
1852 } else {
1853 goto cls_op_not_supp;
1854 }
1855 } else {
1856 if (tp->ops->dump &&
1857 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1858 goto nla_put_failure;
1859 }
1860 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1861 return skb->len;
1862
1863out_nlmsg_trim:
1864nla_put_failure:
1865cls_op_not_supp:
1866 nlmsg_trim(skb, b);
1867 return -1;
1868}
1869
1870static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1871 struct nlmsghdr *n, struct tcf_proto *tp,
1872 struct tcf_block *block, struct Qdisc *q,
1873 u32 parent, void *fh, int event, bool unicast,
1874 bool rtnl_held)
1875{
1876 struct sk_buff *skb;
1877 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1878 int err = 0;
1879
1880 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1881 if (!skb)
1882 return -ENOBUFS;
1883
1884 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1885 n->nlmsg_seq, n->nlmsg_flags, event,
1886 false, rtnl_held) <= 0) {
1887 kfree_skb(skb);
1888 return -EINVAL;
1889 }
1890
1891 if (unicast)
1892 err = rtnl_unicast(skb, net, portid);
1893 else
1894 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1895 n->nlmsg_flags & NLM_F_ECHO);
1896 return err;
1897}
1898
1899static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1900 struct nlmsghdr *n, struct tcf_proto *tp,
1901 struct tcf_block *block, struct Qdisc *q,
1902 u32 parent, void *fh, bool unicast, bool *last,
1903 bool rtnl_held, struct netlink_ext_ack *extack)
1904{
1905 struct sk_buff *skb;
1906 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1907 int err;
1908
1909 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1910 if (!skb)
1911 return -ENOBUFS;
1912
1913 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1914 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1915 false, rtnl_held) <= 0) {
1916 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1917 kfree_skb(skb);
1918 return -EINVAL;
1919 }
1920
1921 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1922 if (err) {
1923 kfree_skb(skb);
1924 return err;
1925 }
1926
1927 if (unicast)
1928 err = rtnl_unicast(skb, net, portid);
1929 else
1930 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1931 n->nlmsg_flags & NLM_F_ECHO);
1932 if (err < 0)
1933 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1934
1935 return err;
1936}
1937
1938static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1939 struct tcf_block *block, struct Qdisc *q,
1940 u32 parent, struct nlmsghdr *n,
1941 struct tcf_chain *chain, int event)
1942{
1943 struct tcf_proto *tp;
1944
1945 for (tp = tcf_get_next_proto(chain, NULL);
1946 tp; tp = tcf_get_next_proto(chain, tp))
1947 tfilter_notify(net, oskb, n, tp, block,
1948 q, parent, NULL, event, false, true);
1949}
1950
1951static void tfilter_put(struct tcf_proto *tp, void *fh)
1952{
1953 if (tp->ops->put && fh)
1954 tp->ops->put(tp, fh);
1955}
1956
1957static bool is_qdisc_ingress(__u32 classid)
1958{
1959 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
1960}
1961
1962static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1963 struct netlink_ext_ack *extack)
1964{
1965 struct net *net = sock_net(skb->sk);
1966 struct nlattr *tca[TCA_MAX + 1];
1967 char name[IFNAMSIZ];
1968 struct tcmsg *t;
1969 u32 protocol;
1970 u32 prio;
1971 bool prio_allocate;
1972 u32 parent;
1973 u32 chain_index;
1974 struct Qdisc *q;
1975 struct tcf_chain_info chain_info;
1976 struct tcf_chain *chain;
1977 struct tcf_block *block;
1978 struct tcf_proto *tp;
1979 unsigned long cl;
1980 void *fh;
1981 int err;
1982 int tp_created;
1983 bool rtnl_held = false;
1984 u32 flags;
1985
1986replay:
1987 tp_created = 0;
1988
1989 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1990 rtm_tca_policy, extack);
1991 if (err < 0)
1992 return err;
1993
1994 t = nlmsg_data(n);
1995 protocol = TC_H_MIN(t->tcm_info);
1996 prio = TC_H_MAJ(t->tcm_info);
1997 prio_allocate = false;
1998 parent = t->tcm_parent;
1999 tp = NULL;
2000 cl = 0;
2001 block = NULL;
2002 q = NULL;
2003 chain = NULL;
2004 flags = 0;
2005
2006 if (prio == 0) {
2007 /* If no priority is provided by the user,
2008 * we allocate one.
2009 */
2010 if (n->nlmsg_flags & NLM_F_CREATE) {
2011 prio = TC_H_MAKE(0x80000000U, 0U);
2012 prio_allocate = true;
2013 } else {
2014 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2015 return -ENOENT;
2016 }
2017 }
2018
2019 /* Find head of filter chain. */
2020
2021 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2022 if (err)
2023 return err;
2024
2025 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2026 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2027 err = -EINVAL;
2028 goto errout;
2029 }
2030
2031 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2032 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2033 * type is not specified, classifier is not unlocked.
2034 */
2035 if (rtnl_held ||
2036 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2037 !tcf_proto_is_unlocked(name)) {
2038 rtnl_held = true;
2039 rtnl_lock();
2040 }
2041
2042 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2043 if (err)
2044 goto errout;
2045
2046 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2047 extack);
2048 if (IS_ERR(block)) {
2049 err = PTR_ERR(block);
2050 goto errout;
2051 }
2052 block->classid = parent;
2053
2054 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2055 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2056 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2057 err = -EINVAL;
2058 goto errout;
2059 }
2060 chain = tcf_chain_get(block, chain_index, true);
2061 if (!chain) {
2062 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2063 err = -ENOMEM;
2064 goto errout;
2065 }
2066
2067 mutex_lock(&chain->filter_chain_lock);
2068 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2069 prio, prio_allocate);
2070 if (IS_ERR(tp)) {
2071 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2072 err = PTR_ERR(tp);
2073 goto errout_locked;
2074 }
2075
2076 if (tp == NULL) {
2077 struct tcf_proto *tp_new = NULL;
2078
2079 if (chain->flushing) {
2080 err = -EAGAIN;
2081 goto errout_locked;
2082 }
2083
2084 /* Proto-tcf does not exist, create new one */
2085
2086 if (tca[TCA_KIND] == NULL || !protocol) {
2087 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2088 err = -EINVAL;
2089 goto errout_locked;
2090 }
2091
2092 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2093 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2094 err = -ENOENT;
2095 goto errout_locked;
2096 }
2097
2098 if (prio_allocate)
2099 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2100 &chain_info));
2101
2102 mutex_unlock(&chain->filter_chain_lock);
2103 tp_new = tcf_proto_create(name, protocol, prio, chain,
2104 rtnl_held, extack);
2105 if (IS_ERR(tp_new)) {
2106 err = PTR_ERR(tp_new);
2107 goto errout_tp;
2108 }
2109
2110 tp_created = 1;
2111 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2112 rtnl_held);
2113 if (IS_ERR(tp)) {
2114 err = PTR_ERR(tp);
2115 goto errout_tp;
2116 }
2117 } else {
2118 mutex_unlock(&chain->filter_chain_lock);
2119 }
2120
2121 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2122 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2123 err = -EINVAL;
2124 goto errout;
2125 }
2126
2127 fh = tp->ops->get(tp, t->tcm_handle);
2128
2129 if (!fh) {
2130 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2131 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2132 err = -ENOENT;
2133 goto errout;
2134 }
2135 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2136 tfilter_put(tp, fh);
2137 NL_SET_ERR_MSG(extack, "Filter already exists");
2138 err = -EEXIST;
2139 goto errout;
2140 }
2141
2142 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2143 tfilter_put(tp, fh);
2144 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2145 err = -EINVAL;
2146 goto errout;
2147 }
2148
2149 if (!(n->nlmsg_flags & NLM_F_CREATE))
2150 flags |= TCA_ACT_FLAGS_REPLACE;
2151 if (!rtnl_held)
2152 flags |= TCA_ACT_FLAGS_NO_RTNL;
2153 if (is_qdisc_ingress(parent))
2154 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2155 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2156 flags, extack);
2157 if (err == 0) {
2158 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2159 RTM_NEWTFILTER, false, rtnl_held);
2160 tfilter_put(tp, fh);
2161 /* q pointer is NULL for shared blocks */
2162 if (q)
2163 q->flags &= ~TCQ_F_CAN_BYPASS;
2164 }
2165
2166errout:
2167 if (err && tp_created)
2168 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2169errout_tp:
2170 if (chain) {
2171 if (tp && !IS_ERR(tp))
2172 tcf_proto_put(tp, rtnl_held, NULL);
2173 if (!tp_created)
2174 tcf_chain_put(chain);
2175 }
2176 tcf_block_release(q, block, rtnl_held);
2177
2178 if (rtnl_held)
2179 rtnl_unlock();
2180
2181 if (err == -EAGAIN) {
2182 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2183 * of target chain.
2184 */
2185 rtnl_held = true;
2186 /* Replay the request. */
2187 goto replay;
2188 }
2189 return err;
2190
2191errout_locked:
2192 mutex_unlock(&chain->filter_chain_lock);
2193 goto errout;
2194}
2195
2196static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2197 struct netlink_ext_ack *extack)
2198{
2199 struct net *net = sock_net(skb->sk);
2200 struct nlattr *tca[TCA_MAX + 1];
2201 char name[IFNAMSIZ];
2202 struct tcmsg *t;
2203 u32 protocol;
2204 u32 prio;
2205 u32 parent;
2206 u32 chain_index;
2207 struct Qdisc *q = NULL;
2208 struct tcf_chain_info chain_info;
2209 struct tcf_chain *chain = NULL;
2210 struct tcf_block *block = NULL;
2211 struct tcf_proto *tp = NULL;
2212 unsigned long cl = 0;
2213 void *fh = NULL;
2214 int err;
2215 bool rtnl_held = false;
2216
2217 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2218 rtm_tca_policy, extack);
2219 if (err < 0)
2220 return err;
2221
2222 t = nlmsg_data(n);
2223 protocol = TC_H_MIN(t->tcm_info);
2224 prio = TC_H_MAJ(t->tcm_info);
2225 parent = t->tcm_parent;
2226
2227 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2228 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2229 return -ENOENT;
2230 }
2231
2232 /* Find head of filter chain. */
2233
2234 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2235 if (err)
2236 return err;
2237
2238 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2239 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2240 err = -EINVAL;
2241 goto errout;
2242 }
2243 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2244 * found), qdisc is not unlocked, classifier type is not specified,
2245 * classifier is not unlocked.
2246 */
2247 if (!prio ||
2248 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2249 !tcf_proto_is_unlocked(name)) {
2250 rtnl_held = true;
2251 rtnl_lock();
2252 }
2253
2254 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2255 if (err)
2256 goto errout;
2257
2258 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2259 extack);
2260 if (IS_ERR(block)) {
2261 err = PTR_ERR(block);
2262 goto errout;
2263 }
2264
2265 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2266 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2267 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2268 err = -EINVAL;
2269 goto errout;
2270 }
2271 chain = tcf_chain_get(block, chain_index, false);
2272 if (!chain) {
2273 /* User requested flush on non-existent chain. Nothing to do,
2274 * so just return success.
2275 */
2276 if (prio == 0) {
2277 err = 0;
2278 goto errout;
2279 }
2280 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2281 err = -ENOENT;
2282 goto errout;
2283 }
2284
2285 if (prio == 0) {
2286 tfilter_notify_chain(net, skb, block, q, parent, n,
2287 chain, RTM_DELTFILTER);
2288 tcf_chain_flush(chain, rtnl_held);
2289 err = 0;
2290 goto errout;
2291 }
2292
2293 mutex_lock(&chain->filter_chain_lock);
2294 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2295 prio, false);
2296 if (!tp || IS_ERR(tp)) {
2297 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2298 err = tp ? PTR_ERR(tp) : -ENOENT;
2299 goto errout_locked;
2300 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2301 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2302 err = -EINVAL;
2303 goto errout_locked;
2304 } else if (t->tcm_handle == 0) {
2305 tcf_proto_signal_destroying(chain, tp);
2306 tcf_chain_tp_remove(chain, &chain_info, tp);
2307 mutex_unlock(&chain->filter_chain_lock);
2308
2309 tcf_proto_put(tp, rtnl_held, NULL);
2310 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2311 RTM_DELTFILTER, false, rtnl_held);
2312 err = 0;
2313 goto errout;
2314 }
2315 mutex_unlock(&chain->filter_chain_lock);
2316
2317 fh = tp->ops->get(tp, t->tcm_handle);
2318
2319 if (!fh) {
2320 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2321 err = -ENOENT;
2322 } else {
2323 bool last;
2324
2325 err = tfilter_del_notify(net, skb, n, tp, block,
2326 q, parent, fh, false, &last,
2327 rtnl_held, extack);
2328
2329 if (err)
2330 goto errout;
2331 if (last)
2332 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2333 }
2334
2335errout:
2336 if (chain) {
2337 if (tp && !IS_ERR(tp))
2338 tcf_proto_put(tp, rtnl_held, NULL);
2339 tcf_chain_put(chain);
2340 }
2341 tcf_block_release(q, block, rtnl_held);
2342
2343 if (rtnl_held)
2344 rtnl_unlock();
2345
2346 return err;
2347
2348errout_locked:
2349 mutex_unlock(&chain->filter_chain_lock);
2350 goto errout;
2351}
2352
2353static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2354 struct netlink_ext_ack *extack)
2355{
2356 struct net *net = sock_net(skb->sk);
2357 struct nlattr *tca[TCA_MAX + 1];
2358 char name[IFNAMSIZ];
2359 struct tcmsg *t;
2360 u32 protocol;
2361 u32 prio;
2362 u32 parent;
2363 u32 chain_index;
2364 struct Qdisc *q = NULL;
2365 struct tcf_chain_info chain_info;
2366 struct tcf_chain *chain = NULL;
2367 struct tcf_block *block = NULL;
2368 struct tcf_proto *tp = NULL;
2369 unsigned long cl = 0;
2370 void *fh = NULL;
2371 int err;
2372 bool rtnl_held = false;
2373
2374 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2375 rtm_tca_policy, extack);
2376 if (err < 0)
2377 return err;
2378
2379 t = nlmsg_data(n);
2380 protocol = TC_H_MIN(t->tcm_info);
2381 prio = TC_H_MAJ(t->tcm_info);
2382 parent = t->tcm_parent;
2383
2384 if (prio == 0) {
2385 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2386 return -ENOENT;
2387 }
2388
2389 /* Find head of filter chain. */
2390
2391 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2392 if (err)
2393 return err;
2394
2395 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2396 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2397 err = -EINVAL;
2398 goto errout;
2399 }
2400 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2401 * unlocked, classifier type is not specified, classifier is not
2402 * unlocked.
2403 */
2404 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2405 !tcf_proto_is_unlocked(name)) {
2406 rtnl_held = true;
2407 rtnl_lock();
2408 }
2409
2410 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2411 if (err)
2412 goto errout;
2413
2414 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2415 extack);
2416 if (IS_ERR(block)) {
2417 err = PTR_ERR(block);
2418 goto errout;
2419 }
2420
2421 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2422 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2423 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2424 err = -EINVAL;
2425 goto errout;
2426 }
2427 chain = tcf_chain_get(block, chain_index, false);
2428 if (!chain) {
2429 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2430 err = -EINVAL;
2431 goto errout;
2432 }
2433
2434 mutex_lock(&chain->filter_chain_lock);
2435 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2436 prio, false);
2437 mutex_unlock(&chain->filter_chain_lock);
2438 if (!tp || IS_ERR(tp)) {
2439 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2440 err = tp ? PTR_ERR(tp) : -ENOENT;
2441 goto errout;
2442 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2443 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2444 err = -EINVAL;
2445 goto errout;
2446 }
2447
2448 fh = tp->ops->get(tp, t->tcm_handle);
2449
2450 if (!fh) {
2451 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2452 err = -ENOENT;
2453 } else {
2454 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2455 fh, RTM_NEWTFILTER, true, rtnl_held);
2456 if (err < 0)
2457 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2458 }
2459
2460 tfilter_put(tp, fh);
2461errout:
2462 if (chain) {
2463 if (tp && !IS_ERR(tp))
2464 tcf_proto_put(tp, rtnl_held, NULL);
2465 tcf_chain_put(chain);
2466 }
2467 tcf_block_release(q, block, rtnl_held);
2468
2469 if (rtnl_held)
2470 rtnl_unlock();
2471
2472 return err;
2473}
2474
2475struct tcf_dump_args {
2476 struct tcf_walker w;
2477 struct sk_buff *skb;
2478 struct netlink_callback *cb;
2479 struct tcf_block *block;
2480 struct Qdisc *q;
2481 u32 parent;
2482 bool terse_dump;
2483};
2484
2485static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2486{
2487 struct tcf_dump_args *a = (void *)arg;
2488 struct net *net = sock_net(a->skb->sk);
2489
2490 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2491 n, NETLINK_CB(a->cb->skb).portid,
2492 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2493 RTM_NEWTFILTER, a->terse_dump, true);
2494}
2495
2496static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2497 struct sk_buff *skb, struct netlink_callback *cb,
2498 long index_start, long *p_index, bool terse)
2499{
2500 struct net *net = sock_net(skb->sk);
2501 struct tcf_block *block = chain->block;
2502 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2503 struct tcf_proto *tp, *tp_prev;
2504 struct tcf_dump_args arg;
2505
2506 for (tp = __tcf_get_next_proto(chain, NULL);
2507 tp;
2508 tp_prev = tp,
2509 tp = __tcf_get_next_proto(chain, tp),
2510 tcf_proto_put(tp_prev, true, NULL),
2511 (*p_index)++) {
2512 if (*p_index < index_start)
2513 continue;
2514 if (TC_H_MAJ(tcm->tcm_info) &&
2515 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2516 continue;
2517 if (TC_H_MIN(tcm->tcm_info) &&
2518 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2519 continue;
2520 if (*p_index > index_start)
2521 memset(&cb->args[1], 0,
2522 sizeof(cb->args) - sizeof(cb->args[0]));
2523 if (cb->args[1] == 0) {
2524 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2525 NETLINK_CB(cb->skb).portid,
2526 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2527 RTM_NEWTFILTER, false, true) <= 0)
2528 goto errout;
2529 cb->args[1] = 1;
2530 }
2531 if (!tp->ops->walk)
2532 continue;
2533 arg.w.fn = tcf_node_dump;
2534 arg.skb = skb;
2535 arg.cb = cb;
2536 arg.block = block;
2537 arg.q = q;
2538 arg.parent = parent;
2539 arg.w.stop = 0;
2540 arg.w.skip = cb->args[1] - 1;
2541 arg.w.count = 0;
2542 arg.w.cookie = cb->args[2];
2543 arg.terse_dump = terse;
2544 tp->ops->walk(tp, &arg.w, true);
2545 cb->args[2] = arg.w.cookie;
2546 cb->args[1] = arg.w.count + 1;
2547 if (arg.w.stop)
2548 goto errout;
2549 }
2550 return true;
2551
2552errout:
2553 tcf_proto_put(tp, true, NULL);
2554 return false;
2555}
2556
2557static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2558 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2559};
2560
2561/* called with RTNL */
2562static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2563{
2564 struct tcf_chain *chain, *chain_prev;
2565 struct net *net = sock_net(skb->sk);
2566 struct nlattr *tca[TCA_MAX + 1];
2567 struct Qdisc *q = NULL;
2568 struct tcf_block *block;
2569 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2570 bool terse_dump = false;
2571 long index_start;
2572 long index;
2573 u32 parent;
2574 int err;
2575
2576 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2577 return skb->len;
2578
2579 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2580 tcf_tfilter_dump_policy, cb->extack);
2581 if (err)
2582 return err;
2583
2584 if (tca[TCA_DUMP_FLAGS]) {
2585 struct nla_bitfield32 flags =
2586 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2587
2588 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2589 }
2590
2591 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2592 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2593 if (!block)
2594 goto out;
2595 /* If we work with block index, q is NULL and parent value
2596 * will never be used in the following code. The check
2597 * in tcf_fill_node prevents it. However, compiler does not
2598 * see that far, so set parent to zero to silence the warning
2599 * about parent being uninitialized.
2600 */
2601 parent = 0;
2602 } else {
2603 const struct Qdisc_class_ops *cops;
2604 struct net_device *dev;
2605 unsigned long cl = 0;
2606
2607 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2608 if (!dev)
2609 return skb->len;
2610
2611 parent = tcm->tcm_parent;
2612 if (!parent)
2613 q = rtnl_dereference(dev->qdisc);
2614 else
2615 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2616 if (!q)
2617 goto out;
2618 cops = q->ops->cl_ops;
2619 if (!cops)
2620 goto out;
2621 if (!cops->tcf_block)
2622 goto out;
2623 if (TC_H_MIN(tcm->tcm_parent)) {
2624 cl = cops->find(q, tcm->tcm_parent);
2625 if (cl == 0)
2626 goto out;
2627 }
2628 block = cops->tcf_block(q, cl, NULL);
2629 if (!block)
2630 goto out;
2631 parent = block->classid;
2632 if (tcf_block_shared(block))
2633 q = NULL;
2634 }
2635
2636 index_start = cb->args[0];
2637 index = 0;
2638
2639 for (chain = __tcf_get_next_chain(block, NULL);
2640 chain;
2641 chain_prev = chain,
2642 chain = __tcf_get_next_chain(block, chain),
2643 tcf_chain_put(chain_prev)) {
2644 if (tca[TCA_CHAIN] &&
2645 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2646 continue;
2647 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2648 index_start, &index, terse_dump)) {
2649 tcf_chain_put(chain);
2650 err = -EMSGSIZE;
2651 break;
2652 }
2653 }
2654
2655 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2656 tcf_block_refcnt_put(block, true);
2657 cb->args[0] = index;
2658
2659out:
2660 /* If we did no progress, the error (EMSGSIZE) is real */
2661 if (skb->len == 0 && err)
2662 return err;
2663 return skb->len;
2664}
2665
2666static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2667 void *tmplt_priv, u32 chain_index,
2668 struct net *net, struct sk_buff *skb,
2669 struct tcf_block *block,
2670 u32 portid, u32 seq, u16 flags, int event)
2671{
2672 unsigned char *b = skb_tail_pointer(skb);
2673 const struct tcf_proto_ops *ops;
2674 struct nlmsghdr *nlh;
2675 struct tcmsg *tcm;
2676 void *priv;
2677
2678 ops = tmplt_ops;
2679 priv = tmplt_priv;
2680
2681 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2682 if (!nlh)
2683 goto out_nlmsg_trim;
2684 tcm = nlmsg_data(nlh);
2685 tcm->tcm_family = AF_UNSPEC;
2686 tcm->tcm__pad1 = 0;
2687 tcm->tcm__pad2 = 0;
2688 tcm->tcm_handle = 0;
2689 if (block->q) {
2690 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2691 tcm->tcm_parent = block->q->handle;
2692 } else {
2693 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2694 tcm->tcm_block_index = block->index;
2695 }
2696
2697 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2698 goto nla_put_failure;
2699
2700 if (ops) {
2701 if (nla_put_string(skb, TCA_KIND, ops->kind))
2702 goto nla_put_failure;
2703 if (ops->tmplt_dump(skb, net, priv) < 0)
2704 goto nla_put_failure;
2705 }
2706
2707 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2708 return skb->len;
2709
2710out_nlmsg_trim:
2711nla_put_failure:
2712 nlmsg_trim(skb, b);
2713 return -EMSGSIZE;
2714}
2715
2716static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2717 u32 seq, u16 flags, int event, bool unicast)
2718{
2719 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2720 struct tcf_block *block = chain->block;
2721 struct net *net = block->net;
2722 struct sk_buff *skb;
2723 int err = 0;
2724
2725 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2726 if (!skb)
2727 return -ENOBUFS;
2728
2729 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2730 chain->index, net, skb, block, portid,
2731 seq, flags, event) <= 0) {
2732 kfree_skb(skb);
2733 return -EINVAL;
2734 }
2735
2736 if (unicast)
2737 err = rtnl_unicast(skb, net, portid);
2738 else
2739 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2740 flags & NLM_F_ECHO);
2741
2742 return err;
2743}
2744
2745static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2746 void *tmplt_priv, u32 chain_index,
2747 struct tcf_block *block, struct sk_buff *oskb,
2748 u32 seq, u16 flags, bool unicast)
2749{
2750 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2751 struct net *net = block->net;
2752 struct sk_buff *skb;
2753
2754 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2755 if (!skb)
2756 return -ENOBUFS;
2757
2758 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2759 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2760 kfree_skb(skb);
2761 return -EINVAL;
2762 }
2763
2764 if (unicast)
2765 return rtnl_unicast(skb, net, portid);
2766
2767 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2768}
2769
2770static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2771 struct nlattr **tca,
2772 struct netlink_ext_ack *extack)
2773{
2774 const struct tcf_proto_ops *ops;
2775 char name[IFNAMSIZ];
2776 void *tmplt_priv;
2777
2778 /* If kind is not set, user did not specify template. */
2779 if (!tca[TCA_KIND])
2780 return 0;
2781
2782 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2783 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2784 return -EINVAL;
2785 }
2786
2787 ops = tcf_proto_lookup_ops(name, true, extack);
2788 if (IS_ERR(ops))
2789 return PTR_ERR(ops);
2790 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2791 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2792 return -EOPNOTSUPP;
2793 }
2794
2795 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2796 if (IS_ERR(tmplt_priv)) {
2797 module_put(ops->owner);
2798 return PTR_ERR(tmplt_priv);
2799 }
2800 chain->tmplt_ops = ops;
2801 chain->tmplt_priv = tmplt_priv;
2802 return 0;
2803}
2804
2805static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2806 void *tmplt_priv)
2807{
2808 /* If template ops are set, no work to do for us. */
2809 if (!tmplt_ops)
2810 return;
2811
2812 tmplt_ops->tmplt_destroy(tmplt_priv);
2813 module_put(tmplt_ops->owner);
2814}
2815
2816/* Add/delete/get a chain */
2817
2818static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2819 struct netlink_ext_ack *extack)
2820{
2821 struct net *net = sock_net(skb->sk);
2822 struct nlattr *tca[TCA_MAX + 1];
2823 struct tcmsg *t;
2824 u32 parent;
2825 u32 chain_index;
2826 struct Qdisc *q;
2827 struct tcf_chain *chain;
2828 struct tcf_block *block;
2829 unsigned long cl;
2830 int err;
2831
2832replay:
2833 q = NULL;
2834 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2835 rtm_tca_policy, extack);
2836 if (err < 0)
2837 return err;
2838
2839 t = nlmsg_data(n);
2840 parent = t->tcm_parent;
2841 cl = 0;
2842
2843 block = tcf_block_find(net, &q, &parent, &cl,
2844 t->tcm_ifindex, t->tcm_block_index, extack);
2845 if (IS_ERR(block))
2846 return PTR_ERR(block);
2847
2848 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2849 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2850 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2851 err = -EINVAL;
2852 goto errout_block;
2853 }
2854
2855 mutex_lock(&block->lock);
2856 chain = tcf_chain_lookup(block, chain_index);
2857 if (n->nlmsg_type == RTM_NEWCHAIN) {
2858 if (chain) {
2859 if (tcf_chain_held_by_acts_only(chain)) {
2860 /* The chain exists only because there is
2861 * some action referencing it.
2862 */
2863 tcf_chain_hold(chain);
2864 } else {
2865 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2866 err = -EEXIST;
2867 goto errout_block_locked;
2868 }
2869 } else {
2870 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2871 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2872 err = -ENOENT;
2873 goto errout_block_locked;
2874 }
2875 chain = tcf_chain_create(block, chain_index);
2876 if (!chain) {
2877 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2878 err = -ENOMEM;
2879 goto errout_block_locked;
2880 }
2881 }
2882 } else {
2883 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2884 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2885 err = -EINVAL;
2886 goto errout_block_locked;
2887 }
2888 tcf_chain_hold(chain);
2889 }
2890
2891 if (n->nlmsg_type == RTM_NEWCHAIN) {
2892 /* Modifying chain requires holding parent block lock. In case
2893 * the chain was successfully added, take a reference to the
2894 * chain. This ensures that an empty chain does not disappear at
2895 * the end of this function.
2896 */
2897 tcf_chain_hold(chain);
2898 chain->explicitly_created = true;
2899 }
2900 mutex_unlock(&block->lock);
2901
2902 switch (n->nlmsg_type) {
2903 case RTM_NEWCHAIN:
2904 err = tc_chain_tmplt_add(chain, net, tca, extack);
2905 if (err) {
2906 tcf_chain_put_explicitly_created(chain);
2907 goto errout;
2908 }
2909
2910 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2911 RTM_NEWCHAIN, false);
2912 break;
2913 case RTM_DELCHAIN:
2914 tfilter_notify_chain(net, skb, block, q, parent, n,
2915 chain, RTM_DELTFILTER);
2916 /* Flush the chain first as the user requested chain removal. */
2917 tcf_chain_flush(chain, true);
2918 /* In case the chain was successfully deleted, put a reference
2919 * to the chain previously taken during addition.
2920 */
2921 tcf_chain_put_explicitly_created(chain);
2922 break;
2923 case RTM_GETCHAIN:
2924 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2925 n->nlmsg_flags, n->nlmsg_type, true);
2926 if (err < 0)
2927 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2928 break;
2929 default:
2930 err = -EOPNOTSUPP;
2931 NL_SET_ERR_MSG(extack, "Unsupported message type");
2932 goto errout;
2933 }
2934
2935errout:
2936 tcf_chain_put(chain);
2937errout_block:
2938 tcf_block_release(q, block, true);
2939 if (err == -EAGAIN)
2940 /* Replay the request. */
2941 goto replay;
2942 return err;
2943
2944errout_block_locked:
2945 mutex_unlock(&block->lock);
2946 goto errout_block;
2947}
2948
2949/* called with RTNL */
2950static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2951{
2952 struct net *net = sock_net(skb->sk);
2953 struct nlattr *tca[TCA_MAX + 1];
2954 struct Qdisc *q = NULL;
2955 struct tcf_block *block;
2956 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2957 struct tcf_chain *chain;
2958 long index_start;
2959 long index;
2960 int err;
2961
2962 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2963 return skb->len;
2964
2965 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2966 rtm_tca_policy, cb->extack);
2967 if (err)
2968 return err;
2969
2970 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2971 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2972 if (!block)
2973 goto out;
2974 } else {
2975 const struct Qdisc_class_ops *cops;
2976 struct net_device *dev;
2977 unsigned long cl = 0;
2978
2979 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2980 if (!dev)
2981 return skb->len;
2982
2983 if (!tcm->tcm_parent)
2984 q = rtnl_dereference(dev->qdisc);
2985 else
2986 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2987
2988 if (!q)
2989 goto out;
2990 cops = q->ops->cl_ops;
2991 if (!cops)
2992 goto out;
2993 if (!cops->tcf_block)
2994 goto out;
2995 if (TC_H_MIN(tcm->tcm_parent)) {
2996 cl = cops->find(q, tcm->tcm_parent);
2997 if (cl == 0)
2998 goto out;
2999 }
3000 block = cops->tcf_block(q, cl, NULL);
3001 if (!block)
3002 goto out;
3003 if (tcf_block_shared(block))
3004 q = NULL;
3005 }
3006
3007 index_start = cb->args[0];
3008 index = 0;
3009
3010 mutex_lock(&block->lock);
3011 list_for_each_entry(chain, &block->chain_list, list) {
3012 if ((tca[TCA_CHAIN] &&
3013 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3014 continue;
3015 if (index < index_start) {
3016 index++;
3017 continue;
3018 }
3019 if (tcf_chain_held_by_acts_only(chain))
3020 continue;
3021 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3022 chain->index, net, skb, block,
3023 NETLINK_CB(cb->skb).portid,
3024 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3025 RTM_NEWCHAIN);
3026 if (err <= 0)
3027 break;
3028 index++;
3029 }
3030 mutex_unlock(&block->lock);
3031
3032 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3033 tcf_block_refcnt_put(block, true);
3034 cb->args[0] = index;
3035
3036out:
3037 /* If we did no progress, the error (EMSGSIZE) is real */
3038 if (skb->len == 0 && err)
3039 return err;
3040 return skb->len;
3041}
3042
3043void tcf_exts_destroy(struct tcf_exts *exts)
3044{
3045#ifdef CONFIG_NET_CLS_ACT
3046 if (exts->actions) {
3047 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3048 kfree(exts->actions);
3049 }
3050 exts->nr_actions = 0;
3051#endif
3052}
3053EXPORT_SYMBOL(tcf_exts_destroy);
3054
3055int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3056 struct nlattr *rate_tlv, struct tcf_exts *exts,
3057 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3058{
3059#ifdef CONFIG_NET_CLS_ACT
3060 {
3061 int init_res[TCA_ACT_MAX_PRIO] = {};
3062 struct tc_action *act;
3063 size_t attr_size = 0;
3064
3065 if (exts->police && tb[exts->police]) {
3066 struct tc_action_ops *a_o;
3067
3068 a_o = tc_action_load_ops(tb[exts->police], true,
3069 !(flags & TCA_ACT_FLAGS_NO_RTNL),
3070 extack);
3071 if (IS_ERR(a_o))
3072 return PTR_ERR(a_o);
3073 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3074 act = tcf_action_init_1(net, tp, tb[exts->police],
3075 rate_tlv, a_o, init_res, flags,
3076 extack);
3077 module_put(a_o->owner);
3078 if (IS_ERR(act))
3079 return PTR_ERR(act);
3080
3081 act->type = exts->type = TCA_OLD_COMPAT;
3082 exts->actions[0] = act;
3083 exts->nr_actions = 1;
3084 tcf_idr_insert_many(exts->actions);
3085 } else if (exts->action && tb[exts->action]) {
3086 int err;
3087
3088 flags |= TCA_ACT_FLAGS_BIND;
3089 err = tcf_action_init(net, tp, tb[exts->action],
3090 rate_tlv, exts->actions, init_res,
3091 &attr_size, flags, fl_flags,
3092 extack);
3093 if (err < 0)
3094 return err;
3095 exts->nr_actions = err;
3096 }
3097 }
3098#else
3099 if ((exts->action && tb[exts->action]) ||
3100 (exts->police && tb[exts->police])) {
3101 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3102 return -EOPNOTSUPP;
3103 }
3104#endif
3105
3106 return 0;
3107}
3108EXPORT_SYMBOL(tcf_exts_validate_ex);
3109
3110int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3111 struct nlattr *rate_tlv, struct tcf_exts *exts,
3112 u32 flags, struct netlink_ext_ack *extack)
3113{
3114 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3115 flags, 0, extack);
3116}
3117EXPORT_SYMBOL(tcf_exts_validate);
3118
3119void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3120{
3121#ifdef CONFIG_NET_CLS_ACT
3122 struct tcf_exts old = *dst;
3123
3124 *dst = *src;
3125 tcf_exts_destroy(&old);
3126#endif
3127}
3128EXPORT_SYMBOL(tcf_exts_change);
3129
3130#ifdef CONFIG_NET_CLS_ACT
3131static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3132{
3133 if (exts->nr_actions == 0)
3134 return NULL;
3135 else
3136 return exts->actions[0];
3137}
3138#endif
3139
3140int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3141{
3142#ifdef CONFIG_NET_CLS_ACT
3143 struct nlattr *nest;
3144
3145 if (exts->action && tcf_exts_has_actions(exts)) {
3146 /*
3147 * again for backward compatible mode - we want
3148 * to work with both old and new modes of entering
3149 * tc data even if iproute2 was newer - jhs
3150 */
3151 if (exts->type != TCA_OLD_COMPAT) {
3152 nest = nla_nest_start_noflag(skb, exts->action);
3153 if (nest == NULL)
3154 goto nla_put_failure;
3155
3156 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3157 < 0)
3158 goto nla_put_failure;
3159 nla_nest_end(skb, nest);
3160 } else if (exts->police) {
3161 struct tc_action *act = tcf_exts_first_act(exts);
3162 nest = nla_nest_start_noflag(skb, exts->police);
3163 if (nest == NULL || !act)
3164 goto nla_put_failure;
3165 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3166 goto nla_put_failure;
3167 nla_nest_end(skb, nest);
3168 }
3169 }
3170 return 0;
3171
3172nla_put_failure:
3173 nla_nest_cancel(skb, nest);
3174 return -1;
3175#else
3176 return 0;
3177#endif
3178}
3179EXPORT_SYMBOL(tcf_exts_dump);
3180
3181int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3182{
3183#ifdef CONFIG_NET_CLS_ACT
3184 struct nlattr *nest;
3185
3186 if (!exts->action || !tcf_exts_has_actions(exts))
3187 return 0;
3188
3189 nest = nla_nest_start_noflag(skb, exts->action);
3190 if (!nest)
3191 goto nla_put_failure;
3192
3193 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3194 goto nla_put_failure;
3195 nla_nest_end(skb, nest);
3196 return 0;
3197
3198nla_put_failure:
3199 nla_nest_cancel(skb, nest);
3200 return -1;
3201#else
3202 return 0;
3203#endif
3204}
3205EXPORT_SYMBOL(tcf_exts_terse_dump);
3206
3207int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3208{
3209#ifdef CONFIG_NET_CLS_ACT
3210 struct tc_action *a = tcf_exts_first_act(exts);
3211 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3212 return -1;
3213#endif
3214 return 0;
3215}
3216EXPORT_SYMBOL(tcf_exts_dump_stats);
3217
3218static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3219{
3220 if (*flags & TCA_CLS_FLAGS_IN_HW)
3221 return;
3222 *flags |= TCA_CLS_FLAGS_IN_HW;
3223 atomic_inc(&block->offloadcnt);
3224}
3225
3226static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3227{
3228 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3229 return;
3230 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3231 atomic_dec(&block->offloadcnt);
3232}
3233
3234static void tc_cls_offload_cnt_update(struct tcf_block *block,
3235 struct tcf_proto *tp, u32 *cnt,
3236 u32 *flags, u32 diff, bool add)
3237{
3238 lockdep_assert_held(&block->cb_lock);
3239
3240 spin_lock(&tp->lock);
3241 if (add) {
3242 if (!*cnt)
3243 tcf_block_offload_inc(block, flags);
3244 *cnt += diff;
3245 } else {
3246 *cnt -= diff;
3247 if (!*cnt)
3248 tcf_block_offload_dec(block, flags);
3249 }
3250 spin_unlock(&tp->lock);
3251}
3252
3253static void
3254tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3255 u32 *cnt, u32 *flags)
3256{
3257 lockdep_assert_held(&block->cb_lock);
3258
3259 spin_lock(&tp->lock);
3260 tcf_block_offload_dec(block, flags);
3261 *cnt = 0;
3262 spin_unlock(&tp->lock);
3263}
3264
3265static int
3266__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3267 void *type_data, bool err_stop)
3268{
3269 struct flow_block_cb *block_cb;
3270 int ok_count = 0;
3271 int err;
3272
3273 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3274 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3275 if (err) {
3276 if (err_stop)
3277 return err;
3278 } else {
3279 ok_count++;
3280 }
3281 }
3282 return ok_count;
3283}
3284
3285int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3286 void *type_data, bool err_stop, bool rtnl_held)
3287{
3288 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3289 int ok_count;
3290
3291retry:
3292 if (take_rtnl)
3293 rtnl_lock();
3294 down_read(&block->cb_lock);
3295 /* Need to obtain rtnl lock if block is bound to devs that require it.
3296 * In block bind code cb_lock is obtained while holding rtnl, so we must
3297 * obtain the locks in same order here.
3298 */
3299 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3300 up_read(&block->cb_lock);
3301 take_rtnl = true;
3302 goto retry;
3303 }
3304
3305 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3306
3307 up_read(&block->cb_lock);
3308 if (take_rtnl)
3309 rtnl_unlock();
3310 return ok_count;
3311}
3312EXPORT_SYMBOL(tc_setup_cb_call);
3313
3314/* Non-destructive filter add. If filter that wasn't already in hardware is
3315 * successfully offloaded, increment block offloads counter. On failure,
3316 * previously offloaded filter is considered to be intact and offloads counter
3317 * is not decremented.
3318 */
3319
3320int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3321 enum tc_setup_type type, void *type_data, bool err_stop,
3322 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3323{
3324 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3325 int ok_count;
3326
3327retry:
3328 if (take_rtnl)
3329 rtnl_lock();
3330 down_read(&block->cb_lock);
3331 /* Need to obtain rtnl lock if block is bound to devs that require it.
3332 * In block bind code cb_lock is obtained while holding rtnl, so we must
3333 * obtain the locks in same order here.
3334 */
3335 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3336 up_read(&block->cb_lock);
3337 take_rtnl = true;
3338 goto retry;
3339 }
3340
3341 /* Make sure all netdevs sharing this block are offload-capable. */
3342 if (block->nooffloaddevcnt && err_stop) {
3343 ok_count = -EOPNOTSUPP;
3344 goto err_unlock;
3345 }
3346
3347 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3348 if (ok_count < 0)
3349 goto err_unlock;
3350
3351 if (tp->ops->hw_add)
3352 tp->ops->hw_add(tp, type_data);
3353 if (ok_count > 0)
3354 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3355 ok_count, true);
3356err_unlock:
3357 up_read(&block->cb_lock);
3358 if (take_rtnl)
3359 rtnl_unlock();
3360 return min(ok_count, 0);
3361}
3362EXPORT_SYMBOL(tc_setup_cb_add);
3363
3364/* Destructive filter replace. If filter that wasn't already in hardware is
3365 * successfully offloaded, increment block offload counter. On failure,
3366 * previously offloaded filter is considered to be destroyed and offload counter
3367 * is decremented.
3368 */
3369
3370int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3371 enum tc_setup_type type, void *type_data, bool err_stop,
3372 u32 *old_flags, unsigned int *old_in_hw_count,
3373 u32 *new_flags, unsigned int *new_in_hw_count,
3374 bool rtnl_held)
3375{
3376 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3377 int ok_count;
3378
3379retry:
3380 if (take_rtnl)
3381 rtnl_lock();
3382 down_read(&block->cb_lock);
3383 /* Need to obtain rtnl lock if block is bound to devs that require it.
3384 * In block bind code cb_lock is obtained while holding rtnl, so we must
3385 * obtain the locks in same order here.
3386 */
3387 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3388 up_read(&block->cb_lock);
3389 take_rtnl = true;
3390 goto retry;
3391 }
3392
3393 /* Make sure all netdevs sharing this block are offload-capable. */
3394 if (block->nooffloaddevcnt && err_stop) {
3395 ok_count = -EOPNOTSUPP;
3396 goto err_unlock;
3397 }
3398
3399 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3400 if (tp->ops->hw_del)
3401 tp->ops->hw_del(tp, type_data);
3402
3403 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3404 if (ok_count < 0)
3405 goto err_unlock;
3406
3407 if (tp->ops->hw_add)
3408 tp->ops->hw_add(tp, type_data);
3409 if (ok_count > 0)
3410 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3411 new_flags, ok_count, true);
3412err_unlock:
3413 up_read(&block->cb_lock);
3414 if (take_rtnl)
3415 rtnl_unlock();
3416 return min(ok_count, 0);
3417}
3418EXPORT_SYMBOL(tc_setup_cb_replace);
3419
3420/* Destroy filter and decrement block offload counter, if filter was previously
3421 * offloaded.
3422 */
3423
3424int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3425 enum tc_setup_type type, void *type_data, bool err_stop,
3426 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3427{
3428 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3429 int ok_count;
3430
3431retry:
3432 if (take_rtnl)
3433 rtnl_lock();
3434 down_read(&block->cb_lock);
3435 /* Need to obtain rtnl lock if block is bound to devs that require it.
3436 * In block bind code cb_lock is obtained while holding rtnl, so we must
3437 * obtain the locks in same order here.
3438 */
3439 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3440 up_read(&block->cb_lock);
3441 take_rtnl = true;
3442 goto retry;
3443 }
3444
3445 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3446
3447 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3448 if (tp->ops->hw_del)
3449 tp->ops->hw_del(tp, type_data);
3450
3451 up_read(&block->cb_lock);
3452 if (take_rtnl)
3453 rtnl_unlock();
3454 return min(ok_count, 0);
3455}
3456EXPORT_SYMBOL(tc_setup_cb_destroy);
3457
3458int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3459 bool add, flow_setup_cb_t *cb,
3460 enum tc_setup_type type, void *type_data,
3461 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3462{
3463 int err = cb(type, type_data, cb_priv);
3464
3465 if (err) {
3466 if (add && tc_skip_sw(*flags))
3467 return err;
3468 } else {
3469 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3470 add);
3471 }
3472
3473 return 0;
3474}
3475EXPORT_SYMBOL(tc_setup_cb_reoffload);
3476
3477static int tcf_act_get_cookie(struct flow_action_entry *entry,
3478 const struct tc_action *act)
3479{
3480 struct tc_cookie *cookie;
3481 int err = 0;
3482
3483 rcu_read_lock();
3484 cookie = rcu_dereference(act->act_cookie);
3485 if (cookie) {
3486 entry->cookie = flow_action_cookie_create(cookie->data,
3487 cookie->len,
3488 GFP_ATOMIC);
3489 if (!entry->cookie)
3490 err = -ENOMEM;
3491 }
3492 rcu_read_unlock();
3493 return err;
3494}
3495
3496static void tcf_act_put_cookie(struct flow_action_entry *entry)
3497{
3498 flow_action_cookie_destroy(entry->cookie);
3499}
3500
3501void tc_cleanup_offload_action(struct flow_action *flow_action)
3502{
3503 struct flow_action_entry *entry;
3504 int i;
3505
3506 flow_action_for_each(i, entry, flow_action) {
3507 tcf_act_put_cookie(entry);
3508 if (entry->destructor)
3509 entry->destructor(entry->destructor_priv);
3510 }
3511}
3512EXPORT_SYMBOL(tc_cleanup_offload_action);
3513
3514static int tc_setup_offload_act(struct tc_action *act,
3515 struct flow_action_entry *entry,
3516 u32 *index_inc,
3517 struct netlink_ext_ack *extack)
3518{
3519#ifdef CONFIG_NET_CLS_ACT
3520 if (act->ops->offload_act_setup) {
3521 return act->ops->offload_act_setup(act, entry, index_inc, true,
3522 extack);
3523 } else {
3524 NL_SET_ERR_MSG(extack, "Action does not support offload");
3525 return -EOPNOTSUPP;
3526 }
3527#else
3528 return 0;
3529#endif
3530}
3531
3532int tc_setup_action(struct flow_action *flow_action,
3533 struct tc_action *actions[],
3534 struct netlink_ext_ack *extack)
3535{
3536 int i, j, k, index, err = 0;
3537 struct tc_action *act;
3538
3539 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3540 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3541 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3542
3543 if (!actions)
3544 return 0;
3545
3546 j = 0;
3547 tcf_act_for_each_action(i, act, actions) {
3548 struct flow_action_entry *entry;
3549
3550 entry = &flow_action->entries[j];
3551 spin_lock_bh(&act->tcfa_lock);
3552 err = tcf_act_get_cookie(entry, act);
3553 if (err)
3554 goto err_out_locked;
3555
3556 index = 0;
3557 err = tc_setup_offload_act(act, entry, &index, extack);
3558 if (err)
3559 goto err_out_locked;
3560
3561 for (k = 0; k < index ; k++) {
3562 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3563 entry[k].hw_index = act->tcfa_index;
3564 }
3565
3566 j += index;
3567
3568 spin_unlock_bh(&act->tcfa_lock);
3569 }
3570
3571err_out:
3572 if (err)
3573 tc_cleanup_offload_action(flow_action);
3574
3575 return err;
3576err_out_locked:
3577 spin_unlock_bh(&act->tcfa_lock);
3578 goto err_out;
3579}
3580
3581int tc_setup_offload_action(struct flow_action *flow_action,
3582 const struct tcf_exts *exts,
3583 struct netlink_ext_ack *extack)
3584{
3585#ifdef CONFIG_NET_CLS_ACT
3586 if (!exts)
3587 return 0;
3588
3589 return tc_setup_action(flow_action, exts->actions, extack);
3590#else
3591 return 0;
3592#endif
3593}
3594EXPORT_SYMBOL(tc_setup_offload_action);
3595
3596unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3597{
3598 unsigned int num_acts = 0;
3599 struct tc_action *act;
3600 int i;
3601
3602 tcf_exts_for_each_action(i, act, exts) {
3603 if (is_tcf_pedit(act))
3604 num_acts += tcf_pedit_nkeys(act);
3605 else
3606 num_acts++;
3607 }
3608 return num_acts;
3609}
3610EXPORT_SYMBOL(tcf_exts_num_actions);
3611
3612#ifdef CONFIG_NET_CLS_ACT
3613static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3614 u32 *p_block_index,
3615 struct netlink_ext_ack *extack)
3616{
3617 *p_block_index = nla_get_u32(block_index_attr);
3618 if (!*p_block_index) {
3619 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3620 return -EINVAL;
3621 }
3622
3623 return 0;
3624}
3625
3626int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3627 enum flow_block_binder_type binder_type,
3628 struct nlattr *block_index_attr,
3629 struct netlink_ext_ack *extack)
3630{
3631 u32 block_index;
3632 int err;
3633
3634 if (!block_index_attr)
3635 return 0;
3636
3637 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3638 if (err)
3639 return err;
3640
3641 qe->info.binder_type = binder_type;
3642 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3643 qe->info.chain_head_change_priv = &qe->filter_chain;
3644 qe->info.block_index = block_index;
3645
3646 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3647}
3648EXPORT_SYMBOL(tcf_qevent_init);
3649
3650void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3651{
3652 if (qe->info.block_index)
3653 tcf_block_put_ext(qe->block, sch, &qe->info);
3654}
3655EXPORT_SYMBOL(tcf_qevent_destroy);
3656
3657int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3658 struct netlink_ext_ack *extack)
3659{
3660 u32 block_index;
3661 int err;
3662
3663 if (!block_index_attr)
3664 return 0;
3665
3666 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3667 if (err)
3668 return err;
3669
3670 /* Bounce newly-configured block or change in block. */
3671 if (block_index != qe->info.block_index) {
3672 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3673 return -EINVAL;
3674 }
3675
3676 return 0;
3677}
3678EXPORT_SYMBOL(tcf_qevent_validate_change);
3679
3680struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3681 struct sk_buff **to_free, int *ret)
3682{
3683 struct tcf_result cl_res;
3684 struct tcf_proto *fl;
3685
3686 if (!qe->info.block_index)
3687 return skb;
3688
3689 fl = rcu_dereference_bh(qe->filter_chain);
3690
3691 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3692 case TC_ACT_SHOT:
3693 qdisc_qstats_drop(sch);
3694 __qdisc_drop(skb, to_free);
3695 *ret = __NET_XMIT_BYPASS;
3696 return NULL;
3697 case TC_ACT_STOLEN:
3698 case TC_ACT_QUEUED:
3699 case TC_ACT_TRAP:
3700 __qdisc_drop(skb, to_free);
3701 *ret = __NET_XMIT_STOLEN;
3702 return NULL;
3703 case TC_ACT_REDIRECT:
3704 skb_do_redirect(skb);
3705 *ret = __NET_XMIT_STOLEN;
3706 return NULL;
3707 }
3708
3709 return skb;
3710}
3711EXPORT_SYMBOL(tcf_qevent_handle);
3712
3713int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3714{
3715 if (!qe->info.block_index)
3716 return 0;
3717 return nla_put_u32(skb, attr_name, qe->info.block_index);
3718}
3719EXPORT_SYMBOL(tcf_qevent_dump);
3720#endif
3721
3722static __net_init int tcf_net_init(struct net *net)
3723{
3724 struct tcf_net *tn = net_generic(net, tcf_net_id);
3725
3726 spin_lock_init(&tn->idr_lock);
3727 idr_init(&tn->idr);
3728 return 0;
3729}
3730
3731static void __net_exit tcf_net_exit(struct net *net)
3732{
3733 struct tcf_net *tn = net_generic(net, tcf_net_id);
3734
3735 idr_destroy(&tn->idr);
3736}
3737
3738static struct pernet_operations tcf_net_ops = {
3739 .init = tcf_net_init,
3740 .exit = tcf_net_exit,
3741 .id = &tcf_net_id,
3742 .size = sizeof(struct tcf_net),
3743};
3744
3745static int __init tc_filter_init(void)
3746{
3747 int err;
3748
3749 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3750 if (!tc_filter_wq)
3751 return -ENOMEM;
3752
3753 err = register_pernet_subsys(&tcf_net_ops);
3754 if (err)
3755 goto err_register_pernet_subsys;
3756
3757 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3758 RTNL_FLAG_DOIT_UNLOCKED);
3759 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3760 RTNL_FLAG_DOIT_UNLOCKED);
3761 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3762 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3763 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3764 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3765 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3766 tc_dump_chain, 0);
3767
3768 return 0;
3769
3770err_register_pernet_subsys:
3771 destroy_workqueue(tc_filter_wq);
3772 return err;
3773}
3774
3775subsys_initcall(tc_filter_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/cls_api.c Packet classifier API.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Changes:
8 *
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10 */
11
12#include <linux/module.h>
13#include <linux/types.h>
14#include <linux/kernel.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/err.h>
18#include <linux/skbuff.h>
19#include <linux/init.h>
20#include <linux/kmod.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/jhash.h>
24#include <linux/rculist.h>
25#include <net/net_namespace.h>
26#include <net/sock.h>
27#include <net/netlink.h>
28#include <net/pkt_sched.h>
29#include <net/pkt_cls.h>
30#include <net/tc_act/tc_pedit.h>
31#include <net/tc_act/tc_mirred.h>
32#include <net/tc_act/tc_vlan.h>
33#include <net/tc_act/tc_tunnel_key.h>
34#include <net/tc_act/tc_csum.h>
35#include <net/tc_act/tc_gact.h>
36#include <net/tc_act/tc_police.h>
37#include <net/tc_act/tc_sample.h>
38#include <net/tc_act/tc_skbedit.h>
39#include <net/tc_act/tc_ct.h>
40#include <net/tc_act/tc_mpls.h>
41#include <net/tc_act/tc_gate.h>
42#include <net/flow_offload.h>
43
44extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
45
46/* The list of all installed classifier types */
47static LIST_HEAD(tcf_proto_base);
48
49/* Protects list of registered TC modules. It is pure SMP lock. */
50static DEFINE_RWLOCK(cls_mod_lock);
51
52static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
53{
54 return jhash_3words(tp->chain->index, tp->prio,
55 (__force __u32)tp->protocol, 0);
56}
57
58static void tcf_proto_signal_destroying(struct tcf_chain *chain,
59 struct tcf_proto *tp)
60{
61 struct tcf_block *block = chain->block;
62
63 mutex_lock(&block->proto_destroy_lock);
64 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
65 destroy_obj_hashfn(tp));
66 mutex_unlock(&block->proto_destroy_lock);
67}
68
69static bool tcf_proto_cmp(const struct tcf_proto *tp1,
70 const struct tcf_proto *tp2)
71{
72 return tp1->chain->index == tp2->chain->index &&
73 tp1->prio == tp2->prio &&
74 tp1->protocol == tp2->protocol;
75}
76
77static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
78 struct tcf_proto *tp)
79{
80 u32 hash = destroy_obj_hashfn(tp);
81 struct tcf_proto *iter;
82 bool found = false;
83
84 rcu_read_lock();
85 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
86 destroy_ht_node, hash) {
87 if (tcf_proto_cmp(tp, iter)) {
88 found = true;
89 break;
90 }
91 }
92 rcu_read_unlock();
93
94 return found;
95}
96
97static void
98tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
99{
100 struct tcf_block *block = chain->block;
101
102 mutex_lock(&block->proto_destroy_lock);
103 if (hash_hashed(&tp->destroy_ht_node))
104 hash_del_rcu(&tp->destroy_ht_node);
105 mutex_unlock(&block->proto_destroy_lock);
106}
107
108/* Find classifier type by string name */
109
110static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
111{
112 const struct tcf_proto_ops *t, *res = NULL;
113
114 if (kind) {
115 read_lock(&cls_mod_lock);
116 list_for_each_entry(t, &tcf_proto_base, head) {
117 if (strcmp(kind, t->kind) == 0) {
118 if (try_module_get(t->owner))
119 res = t;
120 break;
121 }
122 }
123 read_unlock(&cls_mod_lock);
124 }
125 return res;
126}
127
128static const struct tcf_proto_ops *
129tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
130 struct netlink_ext_ack *extack)
131{
132 const struct tcf_proto_ops *ops;
133
134 ops = __tcf_proto_lookup_ops(kind);
135 if (ops)
136 return ops;
137#ifdef CONFIG_MODULES
138 if (rtnl_held)
139 rtnl_unlock();
140 request_module("cls_%s", kind);
141 if (rtnl_held)
142 rtnl_lock();
143 ops = __tcf_proto_lookup_ops(kind);
144 /* We dropped the RTNL semaphore in order to perform
145 * the module load. So, even if we succeeded in loading
146 * the module we have to replay the request. We indicate
147 * this using -EAGAIN.
148 */
149 if (ops) {
150 module_put(ops->owner);
151 return ERR_PTR(-EAGAIN);
152 }
153#endif
154 NL_SET_ERR_MSG(extack, "TC classifier not found");
155 return ERR_PTR(-ENOENT);
156}
157
158/* Register(unregister) new classifier type */
159
160int register_tcf_proto_ops(struct tcf_proto_ops *ops)
161{
162 struct tcf_proto_ops *t;
163 int rc = -EEXIST;
164
165 write_lock(&cls_mod_lock);
166 list_for_each_entry(t, &tcf_proto_base, head)
167 if (!strcmp(ops->kind, t->kind))
168 goto out;
169
170 list_add_tail(&ops->head, &tcf_proto_base);
171 rc = 0;
172out:
173 write_unlock(&cls_mod_lock);
174 return rc;
175}
176EXPORT_SYMBOL(register_tcf_proto_ops);
177
178static struct workqueue_struct *tc_filter_wq;
179
180int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
181{
182 struct tcf_proto_ops *t;
183 int rc = -ENOENT;
184
185 /* Wait for outstanding call_rcu()s, if any, from a
186 * tcf_proto_ops's destroy() handler.
187 */
188 rcu_barrier();
189 flush_workqueue(tc_filter_wq);
190
191 write_lock(&cls_mod_lock);
192 list_for_each_entry(t, &tcf_proto_base, head) {
193 if (t == ops) {
194 list_del(&t->head);
195 rc = 0;
196 break;
197 }
198 }
199 write_unlock(&cls_mod_lock);
200 return rc;
201}
202EXPORT_SYMBOL(unregister_tcf_proto_ops);
203
204bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
205{
206 INIT_RCU_WORK(rwork, func);
207 return queue_rcu_work(tc_filter_wq, rwork);
208}
209EXPORT_SYMBOL(tcf_queue_work);
210
211/* Select new prio value from the range, managed by kernel. */
212
213static inline u32 tcf_auto_prio(struct tcf_proto *tp)
214{
215 u32 first = TC_H_MAKE(0xC0000000U, 0U);
216
217 if (tp)
218 first = tp->prio - 1;
219
220 return TC_H_MAJ(first);
221}
222
223static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
224{
225 if (kind)
226 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
227 memset(name, 0, IFNAMSIZ);
228 return false;
229}
230
231static bool tcf_proto_is_unlocked(const char *kind)
232{
233 const struct tcf_proto_ops *ops;
234 bool ret;
235
236 if (strlen(kind) == 0)
237 return false;
238
239 ops = tcf_proto_lookup_ops(kind, false, NULL);
240 /* On error return false to take rtnl lock. Proto lookup/create
241 * functions will perform lookup again and properly handle errors.
242 */
243 if (IS_ERR(ops))
244 return false;
245
246 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
247 module_put(ops->owner);
248 return ret;
249}
250
251static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
252 u32 prio, struct tcf_chain *chain,
253 bool rtnl_held,
254 struct netlink_ext_ack *extack)
255{
256 struct tcf_proto *tp;
257 int err;
258
259 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
260 if (!tp)
261 return ERR_PTR(-ENOBUFS);
262
263 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
264 if (IS_ERR(tp->ops)) {
265 err = PTR_ERR(tp->ops);
266 goto errout;
267 }
268 tp->classify = tp->ops->classify;
269 tp->protocol = protocol;
270 tp->prio = prio;
271 tp->chain = chain;
272 spin_lock_init(&tp->lock);
273 refcount_set(&tp->refcnt, 1);
274
275 err = tp->ops->init(tp);
276 if (err) {
277 module_put(tp->ops->owner);
278 goto errout;
279 }
280 return tp;
281
282errout:
283 kfree(tp);
284 return ERR_PTR(err);
285}
286
287static void tcf_proto_get(struct tcf_proto *tp)
288{
289 refcount_inc(&tp->refcnt);
290}
291
292static void tcf_chain_put(struct tcf_chain *chain);
293
294static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
295 bool sig_destroy, struct netlink_ext_ack *extack)
296{
297 tp->ops->destroy(tp, rtnl_held, extack);
298 if (sig_destroy)
299 tcf_proto_signal_destroyed(tp->chain, tp);
300 tcf_chain_put(tp->chain);
301 module_put(tp->ops->owner);
302 kfree_rcu(tp, rcu);
303}
304
305static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
306 struct netlink_ext_ack *extack)
307{
308 if (refcount_dec_and_test(&tp->refcnt))
309 tcf_proto_destroy(tp, rtnl_held, true, extack);
310}
311
312static bool tcf_proto_check_delete(struct tcf_proto *tp)
313{
314 if (tp->ops->delete_empty)
315 return tp->ops->delete_empty(tp);
316
317 tp->deleting = true;
318 return tp->deleting;
319}
320
321static void tcf_proto_mark_delete(struct tcf_proto *tp)
322{
323 spin_lock(&tp->lock);
324 tp->deleting = true;
325 spin_unlock(&tp->lock);
326}
327
328static bool tcf_proto_is_deleting(struct tcf_proto *tp)
329{
330 bool deleting;
331
332 spin_lock(&tp->lock);
333 deleting = tp->deleting;
334 spin_unlock(&tp->lock);
335
336 return deleting;
337}
338
339#define ASSERT_BLOCK_LOCKED(block) \
340 lockdep_assert_held(&(block)->lock)
341
342struct tcf_filter_chain_list_item {
343 struct list_head list;
344 tcf_chain_head_change_t *chain_head_change;
345 void *chain_head_change_priv;
346};
347
348static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
349 u32 chain_index)
350{
351 struct tcf_chain *chain;
352
353 ASSERT_BLOCK_LOCKED(block);
354
355 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
356 if (!chain)
357 return NULL;
358 list_add_tail_rcu(&chain->list, &block->chain_list);
359 mutex_init(&chain->filter_chain_lock);
360 chain->block = block;
361 chain->index = chain_index;
362 chain->refcnt = 1;
363 if (!chain->index)
364 block->chain0.chain = chain;
365 return chain;
366}
367
368static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
369 struct tcf_proto *tp_head)
370{
371 if (item->chain_head_change)
372 item->chain_head_change(tp_head, item->chain_head_change_priv);
373}
374
375static void tcf_chain0_head_change(struct tcf_chain *chain,
376 struct tcf_proto *tp_head)
377{
378 struct tcf_filter_chain_list_item *item;
379 struct tcf_block *block = chain->block;
380
381 if (chain->index)
382 return;
383
384 mutex_lock(&block->lock);
385 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
386 tcf_chain_head_change_item(item, tp_head);
387 mutex_unlock(&block->lock);
388}
389
390/* Returns true if block can be safely freed. */
391
392static bool tcf_chain_detach(struct tcf_chain *chain)
393{
394 struct tcf_block *block = chain->block;
395
396 ASSERT_BLOCK_LOCKED(block);
397
398 list_del_rcu(&chain->list);
399 if (!chain->index)
400 block->chain0.chain = NULL;
401
402 if (list_empty(&block->chain_list) &&
403 refcount_read(&block->refcnt) == 0)
404 return true;
405
406 return false;
407}
408
409static void tcf_block_destroy(struct tcf_block *block)
410{
411 mutex_destroy(&block->lock);
412 mutex_destroy(&block->proto_destroy_lock);
413 kfree_rcu(block, rcu);
414}
415
416static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
417{
418 struct tcf_block *block = chain->block;
419
420 mutex_destroy(&chain->filter_chain_lock);
421 kfree_rcu(chain, rcu);
422 if (free_block)
423 tcf_block_destroy(block);
424}
425
426static void tcf_chain_hold(struct tcf_chain *chain)
427{
428 ASSERT_BLOCK_LOCKED(chain->block);
429
430 ++chain->refcnt;
431}
432
433static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
434{
435 ASSERT_BLOCK_LOCKED(chain->block);
436
437 /* In case all the references are action references, this
438 * chain should not be shown to the user.
439 */
440 return chain->refcnt == chain->action_refcnt;
441}
442
443static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
444 u32 chain_index)
445{
446 struct tcf_chain *chain;
447
448 ASSERT_BLOCK_LOCKED(block);
449
450 list_for_each_entry(chain, &block->chain_list, list) {
451 if (chain->index == chain_index)
452 return chain;
453 }
454 return NULL;
455}
456
457#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
458static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
459 u32 chain_index)
460{
461 struct tcf_chain *chain;
462
463 list_for_each_entry_rcu(chain, &block->chain_list, list) {
464 if (chain->index == chain_index)
465 return chain;
466 }
467 return NULL;
468}
469#endif
470
471static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
472 u32 seq, u16 flags, int event, bool unicast);
473
474static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
475 u32 chain_index, bool create,
476 bool by_act)
477{
478 struct tcf_chain *chain = NULL;
479 bool is_first_reference;
480
481 mutex_lock(&block->lock);
482 chain = tcf_chain_lookup(block, chain_index);
483 if (chain) {
484 tcf_chain_hold(chain);
485 } else {
486 if (!create)
487 goto errout;
488 chain = tcf_chain_create(block, chain_index);
489 if (!chain)
490 goto errout;
491 }
492
493 if (by_act)
494 ++chain->action_refcnt;
495 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
496 mutex_unlock(&block->lock);
497
498 /* Send notification only in case we got the first
499 * non-action reference. Until then, the chain acts only as
500 * a placeholder for actions pointing to it and user ought
501 * not know about them.
502 */
503 if (is_first_reference && !by_act)
504 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
505 RTM_NEWCHAIN, false);
506
507 return chain;
508
509errout:
510 mutex_unlock(&block->lock);
511 return chain;
512}
513
514static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
515 bool create)
516{
517 return __tcf_chain_get(block, chain_index, create, false);
518}
519
520struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
521{
522 return __tcf_chain_get(block, chain_index, true, true);
523}
524EXPORT_SYMBOL(tcf_chain_get_by_act);
525
526static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
527 void *tmplt_priv);
528static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
529 void *tmplt_priv, u32 chain_index,
530 struct tcf_block *block, struct sk_buff *oskb,
531 u32 seq, u16 flags, bool unicast);
532
533static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
534 bool explicitly_created)
535{
536 struct tcf_block *block = chain->block;
537 const struct tcf_proto_ops *tmplt_ops;
538 bool free_block = false;
539 unsigned int refcnt;
540 void *tmplt_priv;
541
542 mutex_lock(&block->lock);
543 if (explicitly_created) {
544 if (!chain->explicitly_created) {
545 mutex_unlock(&block->lock);
546 return;
547 }
548 chain->explicitly_created = false;
549 }
550
551 if (by_act)
552 chain->action_refcnt--;
553
554 /* tc_chain_notify_delete can't be called while holding block lock.
555 * However, when block is unlocked chain can be changed concurrently, so
556 * save these to temporary variables.
557 */
558 refcnt = --chain->refcnt;
559 tmplt_ops = chain->tmplt_ops;
560 tmplt_priv = chain->tmplt_priv;
561
562 /* The last dropped non-action reference will trigger notification. */
563 if (refcnt - chain->action_refcnt == 0 && !by_act) {
564 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
565 block, NULL, 0, 0, false);
566 /* Last reference to chain, no need to lock. */
567 chain->flushing = false;
568 }
569
570 if (refcnt == 0)
571 free_block = tcf_chain_detach(chain);
572 mutex_unlock(&block->lock);
573
574 if (refcnt == 0) {
575 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
576 tcf_chain_destroy(chain, free_block);
577 }
578}
579
580static void tcf_chain_put(struct tcf_chain *chain)
581{
582 __tcf_chain_put(chain, false, false);
583}
584
585void tcf_chain_put_by_act(struct tcf_chain *chain)
586{
587 __tcf_chain_put(chain, true, false);
588}
589EXPORT_SYMBOL(tcf_chain_put_by_act);
590
591static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
592{
593 __tcf_chain_put(chain, false, true);
594}
595
596static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
597{
598 struct tcf_proto *tp, *tp_next;
599
600 mutex_lock(&chain->filter_chain_lock);
601 tp = tcf_chain_dereference(chain->filter_chain, chain);
602 while (tp) {
603 tp_next = rcu_dereference_protected(tp->next, 1);
604 tcf_proto_signal_destroying(chain, tp);
605 tp = tp_next;
606 }
607 tp = tcf_chain_dereference(chain->filter_chain, chain);
608 RCU_INIT_POINTER(chain->filter_chain, NULL);
609 tcf_chain0_head_change(chain, NULL);
610 chain->flushing = true;
611 mutex_unlock(&chain->filter_chain_lock);
612
613 while (tp) {
614 tp_next = rcu_dereference_protected(tp->next, 1);
615 tcf_proto_put(tp, rtnl_held, NULL);
616 tp = tp_next;
617 }
618}
619
620static int tcf_block_setup(struct tcf_block *block,
621 struct flow_block_offload *bo);
622
623static void tcf_block_offload_init(struct flow_block_offload *bo,
624 struct net_device *dev, struct Qdisc *sch,
625 enum flow_block_command command,
626 enum flow_block_binder_type binder_type,
627 struct flow_block *flow_block,
628 bool shared, struct netlink_ext_ack *extack)
629{
630 bo->net = dev_net(dev);
631 bo->command = command;
632 bo->binder_type = binder_type;
633 bo->block = flow_block;
634 bo->block_shared = shared;
635 bo->extack = extack;
636 bo->sch = sch;
637 bo->cb_list_head = &flow_block->cb_list;
638 INIT_LIST_HEAD(&bo->cb_list);
639}
640
641static void tcf_block_unbind(struct tcf_block *block,
642 struct flow_block_offload *bo);
643
644static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
645{
646 struct tcf_block *block = block_cb->indr.data;
647 struct net_device *dev = block_cb->indr.dev;
648 struct Qdisc *sch = block_cb->indr.sch;
649 struct netlink_ext_ack extack = {};
650 struct flow_block_offload bo = {};
651
652 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
653 block_cb->indr.binder_type,
654 &block->flow_block, tcf_block_shared(block),
655 &extack);
656 rtnl_lock();
657 down_write(&block->cb_lock);
658 list_del(&block_cb->driver_list);
659 list_move(&block_cb->list, &bo.cb_list);
660 tcf_block_unbind(block, &bo);
661 up_write(&block->cb_lock);
662 rtnl_unlock();
663}
664
665static bool tcf_block_offload_in_use(struct tcf_block *block)
666{
667 return atomic_read(&block->offloadcnt);
668}
669
670static int tcf_block_offload_cmd(struct tcf_block *block,
671 struct net_device *dev, struct Qdisc *sch,
672 struct tcf_block_ext_info *ei,
673 enum flow_block_command command,
674 struct netlink_ext_ack *extack)
675{
676 struct flow_block_offload bo = {};
677
678 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
679 &block->flow_block, tcf_block_shared(block),
680 extack);
681
682 if (dev->netdev_ops->ndo_setup_tc) {
683 int err;
684
685 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
686 if (err < 0) {
687 if (err != -EOPNOTSUPP)
688 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
689 return err;
690 }
691
692 return tcf_block_setup(block, &bo);
693 }
694
695 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
696 tc_block_indr_cleanup);
697 tcf_block_setup(block, &bo);
698
699 return -EOPNOTSUPP;
700}
701
702static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
703 struct tcf_block_ext_info *ei,
704 struct netlink_ext_ack *extack)
705{
706 struct net_device *dev = q->dev_queue->dev;
707 int err;
708
709 down_write(&block->cb_lock);
710
711 /* If tc offload feature is disabled and the block we try to bind
712 * to already has some offloaded filters, forbid to bind.
713 */
714 if (dev->netdev_ops->ndo_setup_tc &&
715 !tc_can_offload(dev) &&
716 tcf_block_offload_in_use(block)) {
717 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
718 err = -EOPNOTSUPP;
719 goto err_unlock;
720 }
721
722 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
723 if (err == -EOPNOTSUPP)
724 goto no_offload_dev_inc;
725 if (err)
726 goto err_unlock;
727
728 up_write(&block->cb_lock);
729 return 0;
730
731no_offload_dev_inc:
732 if (tcf_block_offload_in_use(block))
733 goto err_unlock;
734
735 err = 0;
736 block->nooffloaddevcnt++;
737err_unlock:
738 up_write(&block->cb_lock);
739 return err;
740}
741
742static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
743 struct tcf_block_ext_info *ei)
744{
745 struct net_device *dev = q->dev_queue->dev;
746 int err;
747
748 down_write(&block->cb_lock);
749 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
750 if (err == -EOPNOTSUPP)
751 goto no_offload_dev_dec;
752 up_write(&block->cb_lock);
753 return;
754
755no_offload_dev_dec:
756 WARN_ON(block->nooffloaddevcnt-- == 0);
757 up_write(&block->cb_lock);
758}
759
760static int
761tcf_chain0_head_change_cb_add(struct tcf_block *block,
762 struct tcf_block_ext_info *ei,
763 struct netlink_ext_ack *extack)
764{
765 struct tcf_filter_chain_list_item *item;
766 struct tcf_chain *chain0;
767
768 item = kmalloc(sizeof(*item), GFP_KERNEL);
769 if (!item) {
770 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
771 return -ENOMEM;
772 }
773 item->chain_head_change = ei->chain_head_change;
774 item->chain_head_change_priv = ei->chain_head_change_priv;
775
776 mutex_lock(&block->lock);
777 chain0 = block->chain0.chain;
778 if (chain0)
779 tcf_chain_hold(chain0);
780 else
781 list_add(&item->list, &block->chain0.filter_chain_list);
782 mutex_unlock(&block->lock);
783
784 if (chain0) {
785 struct tcf_proto *tp_head;
786
787 mutex_lock(&chain0->filter_chain_lock);
788
789 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
790 if (tp_head)
791 tcf_chain_head_change_item(item, tp_head);
792
793 mutex_lock(&block->lock);
794 list_add(&item->list, &block->chain0.filter_chain_list);
795 mutex_unlock(&block->lock);
796
797 mutex_unlock(&chain0->filter_chain_lock);
798 tcf_chain_put(chain0);
799 }
800
801 return 0;
802}
803
804static void
805tcf_chain0_head_change_cb_del(struct tcf_block *block,
806 struct tcf_block_ext_info *ei)
807{
808 struct tcf_filter_chain_list_item *item;
809
810 mutex_lock(&block->lock);
811 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
812 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
813 (item->chain_head_change == ei->chain_head_change &&
814 item->chain_head_change_priv == ei->chain_head_change_priv)) {
815 if (block->chain0.chain)
816 tcf_chain_head_change_item(item, NULL);
817 list_del(&item->list);
818 mutex_unlock(&block->lock);
819
820 kfree(item);
821 return;
822 }
823 }
824 mutex_unlock(&block->lock);
825 WARN_ON(1);
826}
827
828struct tcf_net {
829 spinlock_t idr_lock; /* Protects idr */
830 struct idr idr;
831};
832
833static unsigned int tcf_net_id;
834
835static int tcf_block_insert(struct tcf_block *block, struct net *net,
836 struct netlink_ext_ack *extack)
837{
838 struct tcf_net *tn = net_generic(net, tcf_net_id);
839 int err;
840
841 idr_preload(GFP_KERNEL);
842 spin_lock(&tn->idr_lock);
843 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
844 GFP_NOWAIT);
845 spin_unlock(&tn->idr_lock);
846 idr_preload_end();
847
848 return err;
849}
850
851static void tcf_block_remove(struct tcf_block *block, struct net *net)
852{
853 struct tcf_net *tn = net_generic(net, tcf_net_id);
854
855 spin_lock(&tn->idr_lock);
856 idr_remove(&tn->idr, block->index);
857 spin_unlock(&tn->idr_lock);
858}
859
860static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
861 u32 block_index,
862 struct netlink_ext_ack *extack)
863{
864 struct tcf_block *block;
865
866 block = kzalloc(sizeof(*block), GFP_KERNEL);
867 if (!block) {
868 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
869 return ERR_PTR(-ENOMEM);
870 }
871 mutex_init(&block->lock);
872 mutex_init(&block->proto_destroy_lock);
873 init_rwsem(&block->cb_lock);
874 flow_block_init(&block->flow_block);
875 INIT_LIST_HEAD(&block->chain_list);
876 INIT_LIST_HEAD(&block->owner_list);
877 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
878
879 refcount_set(&block->refcnt, 1);
880 block->net = net;
881 block->index = block_index;
882
883 /* Don't store q pointer for blocks which are shared */
884 if (!tcf_block_shared(block))
885 block->q = q;
886 return block;
887}
888
889static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
890{
891 struct tcf_net *tn = net_generic(net, tcf_net_id);
892
893 return idr_find(&tn->idr, block_index);
894}
895
896static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
897{
898 struct tcf_block *block;
899
900 rcu_read_lock();
901 block = tcf_block_lookup(net, block_index);
902 if (block && !refcount_inc_not_zero(&block->refcnt))
903 block = NULL;
904 rcu_read_unlock();
905
906 return block;
907}
908
909static struct tcf_chain *
910__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
911{
912 mutex_lock(&block->lock);
913 if (chain)
914 chain = list_is_last(&chain->list, &block->chain_list) ?
915 NULL : list_next_entry(chain, list);
916 else
917 chain = list_first_entry_or_null(&block->chain_list,
918 struct tcf_chain, list);
919
920 /* skip all action-only chains */
921 while (chain && tcf_chain_held_by_acts_only(chain))
922 chain = list_is_last(&chain->list, &block->chain_list) ?
923 NULL : list_next_entry(chain, list);
924
925 if (chain)
926 tcf_chain_hold(chain);
927 mutex_unlock(&block->lock);
928
929 return chain;
930}
931
932/* Function to be used by all clients that want to iterate over all chains on
933 * block. It properly obtains block->lock and takes reference to chain before
934 * returning it. Users of this function must be tolerant to concurrent chain
935 * insertion/deletion or ensure that no concurrent chain modification is
936 * possible. Note that all netlink dump callbacks cannot guarantee to provide
937 * consistent dump because rtnl lock is released each time skb is filled with
938 * data and sent to user-space.
939 */
940
941struct tcf_chain *
942tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
943{
944 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
945
946 if (chain)
947 tcf_chain_put(chain);
948
949 return chain_next;
950}
951EXPORT_SYMBOL(tcf_get_next_chain);
952
953static struct tcf_proto *
954__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
955{
956 u32 prio = 0;
957
958 ASSERT_RTNL();
959 mutex_lock(&chain->filter_chain_lock);
960
961 if (!tp) {
962 tp = tcf_chain_dereference(chain->filter_chain, chain);
963 } else if (tcf_proto_is_deleting(tp)) {
964 /* 'deleting' flag is set and chain->filter_chain_lock was
965 * unlocked, which means next pointer could be invalid. Restart
966 * search.
967 */
968 prio = tp->prio + 1;
969 tp = tcf_chain_dereference(chain->filter_chain, chain);
970
971 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
972 if (!tp->deleting && tp->prio >= prio)
973 break;
974 } else {
975 tp = tcf_chain_dereference(tp->next, chain);
976 }
977
978 if (tp)
979 tcf_proto_get(tp);
980
981 mutex_unlock(&chain->filter_chain_lock);
982
983 return tp;
984}
985
986/* Function to be used by all clients that want to iterate over all tp's on
987 * chain. Users of this function must be tolerant to concurrent tp
988 * insertion/deletion or ensure that no concurrent chain modification is
989 * possible. Note that all netlink dump callbacks cannot guarantee to provide
990 * consistent dump because rtnl lock is released each time skb is filled with
991 * data and sent to user-space.
992 */
993
994struct tcf_proto *
995tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
996{
997 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
998
999 if (tp)
1000 tcf_proto_put(tp, true, NULL);
1001
1002 return tp_next;
1003}
1004EXPORT_SYMBOL(tcf_get_next_proto);
1005
1006static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1007{
1008 struct tcf_chain *chain;
1009
1010 /* Last reference to block. At this point chains cannot be added or
1011 * removed concurrently.
1012 */
1013 for (chain = tcf_get_next_chain(block, NULL);
1014 chain;
1015 chain = tcf_get_next_chain(block, chain)) {
1016 tcf_chain_put_explicitly_created(chain);
1017 tcf_chain_flush(chain, rtnl_held);
1018 }
1019}
1020
1021/* Lookup Qdisc and increments its reference counter.
1022 * Set parent, if necessary.
1023 */
1024
1025static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1026 u32 *parent, int ifindex, bool rtnl_held,
1027 struct netlink_ext_ack *extack)
1028{
1029 const struct Qdisc_class_ops *cops;
1030 struct net_device *dev;
1031 int err = 0;
1032
1033 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1034 return 0;
1035
1036 rcu_read_lock();
1037
1038 /* Find link */
1039 dev = dev_get_by_index_rcu(net, ifindex);
1040 if (!dev) {
1041 rcu_read_unlock();
1042 return -ENODEV;
1043 }
1044
1045 /* Find qdisc */
1046 if (!*parent) {
1047 *q = dev->qdisc;
1048 *parent = (*q)->handle;
1049 } else {
1050 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1051 if (!*q) {
1052 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1053 err = -EINVAL;
1054 goto errout_rcu;
1055 }
1056 }
1057
1058 *q = qdisc_refcount_inc_nz(*q);
1059 if (!*q) {
1060 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1061 err = -EINVAL;
1062 goto errout_rcu;
1063 }
1064
1065 /* Is it classful? */
1066 cops = (*q)->ops->cl_ops;
1067 if (!cops) {
1068 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1069 err = -EINVAL;
1070 goto errout_qdisc;
1071 }
1072
1073 if (!cops->tcf_block) {
1074 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1075 err = -EOPNOTSUPP;
1076 goto errout_qdisc;
1077 }
1078
1079errout_rcu:
1080 /* At this point we know that qdisc is not noop_qdisc,
1081 * which means that qdisc holds a reference to net_device
1082 * and we hold a reference to qdisc, so it is safe to release
1083 * rcu read lock.
1084 */
1085 rcu_read_unlock();
1086 return err;
1087
1088errout_qdisc:
1089 rcu_read_unlock();
1090
1091 if (rtnl_held)
1092 qdisc_put(*q);
1093 else
1094 qdisc_put_unlocked(*q);
1095 *q = NULL;
1096
1097 return err;
1098}
1099
1100static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1101 int ifindex, struct netlink_ext_ack *extack)
1102{
1103 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1104 return 0;
1105
1106 /* Do we search for filter, attached to class? */
1107 if (TC_H_MIN(parent)) {
1108 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1109
1110 *cl = cops->find(q, parent);
1111 if (*cl == 0) {
1112 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1113 return -ENOENT;
1114 }
1115 }
1116
1117 return 0;
1118}
1119
1120static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1121 unsigned long cl, int ifindex,
1122 u32 block_index,
1123 struct netlink_ext_ack *extack)
1124{
1125 struct tcf_block *block;
1126
1127 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1128 block = tcf_block_refcnt_get(net, block_index);
1129 if (!block) {
1130 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1131 return ERR_PTR(-EINVAL);
1132 }
1133 } else {
1134 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1135
1136 block = cops->tcf_block(q, cl, extack);
1137 if (!block)
1138 return ERR_PTR(-EINVAL);
1139
1140 if (tcf_block_shared(block)) {
1141 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1142 return ERR_PTR(-EOPNOTSUPP);
1143 }
1144
1145 /* Always take reference to block in order to support execution
1146 * of rules update path of cls API without rtnl lock. Caller
1147 * must release block when it is finished using it. 'if' block
1148 * of this conditional obtain reference to block by calling
1149 * tcf_block_refcnt_get().
1150 */
1151 refcount_inc(&block->refcnt);
1152 }
1153
1154 return block;
1155}
1156
1157static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1158 struct tcf_block_ext_info *ei, bool rtnl_held)
1159{
1160 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1161 /* Flushing/putting all chains will cause the block to be
1162 * deallocated when last chain is freed. However, if chain_list
1163 * is empty, block has to be manually deallocated. After block
1164 * reference counter reached 0, it is no longer possible to
1165 * increment it or add new chains to block.
1166 */
1167 bool free_block = list_empty(&block->chain_list);
1168
1169 mutex_unlock(&block->lock);
1170 if (tcf_block_shared(block))
1171 tcf_block_remove(block, block->net);
1172
1173 if (q)
1174 tcf_block_offload_unbind(block, q, ei);
1175
1176 if (free_block)
1177 tcf_block_destroy(block);
1178 else
1179 tcf_block_flush_all_chains(block, rtnl_held);
1180 } else if (q) {
1181 tcf_block_offload_unbind(block, q, ei);
1182 }
1183}
1184
1185static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1186{
1187 __tcf_block_put(block, NULL, NULL, rtnl_held);
1188}
1189
1190/* Find tcf block.
1191 * Set q, parent, cl when appropriate.
1192 */
1193
1194static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1195 u32 *parent, unsigned long *cl,
1196 int ifindex, u32 block_index,
1197 struct netlink_ext_ack *extack)
1198{
1199 struct tcf_block *block;
1200 int err = 0;
1201
1202 ASSERT_RTNL();
1203
1204 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1205 if (err)
1206 goto errout;
1207
1208 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1209 if (err)
1210 goto errout_qdisc;
1211
1212 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1213 if (IS_ERR(block)) {
1214 err = PTR_ERR(block);
1215 goto errout_qdisc;
1216 }
1217
1218 return block;
1219
1220errout_qdisc:
1221 if (*q)
1222 qdisc_put(*q);
1223errout:
1224 *q = NULL;
1225 return ERR_PTR(err);
1226}
1227
1228static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1229 bool rtnl_held)
1230{
1231 if (!IS_ERR_OR_NULL(block))
1232 tcf_block_refcnt_put(block, rtnl_held);
1233
1234 if (q) {
1235 if (rtnl_held)
1236 qdisc_put(q);
1237 else
1238 qdisc_put_unlocked(q);
1239 }
1240}
1241
1242struct tcf_block_owner_item {
1243 struct list_head list;
1244 struct Qdisc *q;
1245 enum flow_block_binder_type binder_type;
1246};
1247
1248static void
1249tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1250 struct Qdisc *q,
1251 enum flow_block_binder_type binder_type)
1252{
1253 if (block->keep_dst &&
1254 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1255 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1256 netif_keep_dst(qdisc_dev(q));
1257}
1258
1259void tcf_block_netif_keep_dst(struct tcf_block *block)
1260{
1261 struct tcf_block_owner_item *item;
1262
1263 block->keep_dst = true;
1264 list_for_each_entry(item, &block->owner_list, list)
1265 tcf_block_owner_netif_keep_dst(block, item->q,
1266 item->binder_type);
1267}
1268EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1269
1270static int tcf_block_owner_add(struct tcf_block *block,
1271 struct Qdisc *q,
1272 enum flow_block_binder_type binder_type)
1273{
1274 struct tcf_block_owner_item *item;
1275
1276 item = kmalloc(sizeof(*item), GFP_KERNEL);
1277 if (!item)
1278 return -ENOMEM;
1279 item->q = q;
1280 item->binder_type = binder_type;
1281 list_add(&item->list, &block->owner_list);
1282 return 0;
1283}
1284
1285static void tcf_block_owner_del(struct tcf_block *block,
1286 struct Qdisc *q,
1287 enum flow_block_binder_type binder_type)
1288{
1289 struct tcf_block_owner_item *item;
1290
1291 list_for_each_entry(item, &block->owner_list, list) {
1292 if (item->q == q && item->binder_type == binder_type) {
1293 list_del(&item->list);
1294 kfree(item);
1295 return;
1296 }
1297 }
1298 WARN_ON(1);
1299}
1300
1301int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1302 struct tcf_block_ext_info *ei,
1303 struct netlink_ext_ack *extack)
1304{
1305 struct net *net = qdisc_net(q);
1306 struct tcf_block *block = NULL;
1307 int err;
1308
1309 if (ei->block_index)
1310 /* block_index not 0 means the shared block is requested */
1311 block = tcf_block_refcnt_get(net, ei->block_index);
1312
1313 if (!block) {
1314 block = tcf_block_create(net, q, ei->block_index, extack);
1315 if (IS_ERR(block))
1316 return PTR_ERR(block);
1317 if (tcf_block_shared(block)) {
1318 err = tcf_block_insert(block, net, extack);
1319 if (err)
1320 goto err_block_insert;
1321 }
1322 }
1323
1324 err = tcf_block_owner_add(block, q, ei->binder_type);
1325 if (err)
1326 goto err_block_owner_add;
1327
1328 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1329
1330 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1331 if (err)
1332 goto err_chain0_head_change_cb_add;
1333
1334 err = tcf_block_offload_bind(block, q, ei, extack);
1335 if (err)
1336 goto err_block_offload_bind;
1337
1338 *p_block = block;
1339 return 0;
1340
1341err_block_offload_bind:
1342 tcf_chain0_head_change_cb_del(block, ei);
1343err_chain0_head_change_cb_add:
1344 tcf_block_owner_del(block, q, ei->binder_type);
1345err_block_owner_add:
1346err_block_insert:
1347 tcf_block_refcnt_put(block, true);
1348 return err;
1349}
1350EXPORT_SYMBOL(tcf_block_get_ext);
1351
1352static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1353{
1354 struct tcf_proto __rcu **p_filter_chain = priv;
1355
1356 rcu_assign_pointer(*p_filter_chain, tp_head);
1357}
1358
1359int tcf_block_get(struct tcf_block **p_block,
1360 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1361 struct netlink_ext_ack *extack)
1362{
1363 struct tcf_block_ext_info ei = {
1364 .chain_head_change = tcf_chain_head_change_dflt,
1365 .chain_head_change_priv = p_filter_chain,
1366 };
1367
1368 WARN_ON(!p_filter_chain);
1369 return tcf_block_get_ext(p_block, q, &ei, extack);
1370}
1371EXPORT_SYMBOL(tcf_block_get);
1372
1373/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1374 * actions should be all removed after flushing.
1375 */
1376void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1377 struct tcf_block_ext_info *ei)
1378{
1379 if (!block)
1380 return;
1381 tcf_chain0_head_change_cb_del(block, ei);
1382 tcf_block_owner_del(block, q, ei->binder_type);
1383
1384 __tcf_block_put(block, q, ei, true);
1385}
1386EXPORT_SYMBOL(tcf_block_put_ext);
1387
1388void tcf_block_put(struct tcf_block *block)
1389{
1390 struct tcf_block_ext_info ei = {0, };
1391
1392 if (!block)
1393 return;
1394 tcf_block_put_ext(block, block->q, &ei);
1395}
1396
1397EXPORT_SYMBOL(tcf_block_put);
1398
1399static int
1400tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1401 void *cb_priv, bool add, bool offload_in_use,
1402 struct netlink_ext_ack *extack)
1403{
1404 struct tcf_chain *chain, *chain_prev;
1405 struct tcf_proto *tp, *tp_prev;
1406 int err;
1407
1408 lockdep_assert_held(&block->cb_lock);
1409
1410 for (chain = __tcf_get_next_chain(block, NULL);
1411 chain;
1412 chain_prev = chain,
1413 chain = __tcf_get_next_chain(block, chain),
1414 tcf_chain_put(chain_prev)) {
1415 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1416 tp_prev = tp,
1417 tp = __tcf_get_next_proto(chain, tp),
1418 tcf_proto_put(tp_prev, true, NULL)) {
1419 if (tp->ops->reoffload) {
1420 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1421 extack);
1422 if (err && add)
1423 goto err_playback_remove;
1424 } else if (add && offload_in_use) {
1425 err = -EOPNOTSUPP;
1426 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1427 goto err_playback_remove;
1428 }
1429 }
1430 }
1431
1432 return 0;
1433
1434err_playback_remove:
1435 tcf_proto_put(tp, true, NULL);
1436 tcf_chain_put(chain);
1437 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1438 extack);
1439 return err;
1440}
1441
1442static int tcf_block_bind(struct tcf_block *block,
1443 struct flow_block_offload *bo)
1444{
1445 struct flow_block_cb *block_cb, *next;
1446 int err, i = 0;
1447
1448 lockdep_assert_held(&block->cb_lock);
1449
1450 list_for_each_entry(block_cb, &bo->cb_list, list) {
1451 err = tcf_block_playback_offloads(block, block_cb->cb,
1452 block_cb->cb_priv, true,
1453 tcf_block_offload_in_use(block),
1454 bo->extack);
1455 if (err)
1456 goto err_unroll;
1457 if (!bo->unlocked_driver_cb)
1458 block->lockeddevcnt++;
1459
1460 i++;
1461 }
1462 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1463
1464 return 0;
1465
1466err_unroll:
1467 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1468 if (i-- > 0) {
1469 list_del(&block_cb->list);
1470 tcf_block_playback_offloads(block, block_cb->cb,
1471 block_cb->cb_priv, false,
1472 tcf_block_offload_in_use(block),
1473 NULL);
1474 if (!bo->unlocked_driver_cb)
1475 block->lockeddevcnt--;
1476 }
1477 flow_block_cb_free(block_cb);
1478 }
1479
1480 return err;
1481}
1482
1483static void tcf_block_unbind(struct tcf_block *block,
1484 struct flow_block_offload *bo)
1485{
1486 struct flow_block_cb *block_cb, *next;
1487
1488 lockdep_assert_held(&block->cb_lock);
1489
1490 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1491 tcf_block_playback_offloads(block, block_cb->cb,
1492 block_cb->cb_priv, false,
1493 tcf_block_offload_in_use(block),
1494 NULL);
1495 list_del(&block_cb->list);
1496 flow_block_cb_free(block_cb);
1497 if (!bo->unlocked_driver_cb)
1498 block->lockeddevcnt--;
1499 }
1500}
1501
1502static int tcf_block_setup(struct tcf_block *block,
1503 struct flow_block_offload *bo)
1504{
1505 int err;
1506
1507 switch (bo->command) {
1508 case FLOW_BLOCK_BIND:
1509 err = tcf_block_bind(block, bo);
1510 break;
1511 case FLOW_BLOCK_UNBIND:
1512 err = 0;
1513 tcf_block_unbind(block, bo);
1514 break;
1515 default:
1516 WARN_ON_ONCE(1);
1517 err = -EOPNOTSUPP;
1518 }
1519
1520 return err;
1521}
1522
1523/* Main classifier routine: scans classifier chain attached
1524 * to this qdisc, (optionally) tests for protocol and asks
1525 * specific classifiers.
1526 */
1527static inline int __tcf_classify(struct sk_buff *skb,
1528 const struct tcf_proto *tp,
1529 const struct tcf_proto *orig_tp,
1530 struct tcf_result *res,
1531 bool compat_mode,
1532 u32 *last_executed_chain)
1533{
1534#ifdef CONFIG_NET_CLS_ACT
1535 const int max_reclassify_loop = 16;
1536 const struct tcf_proto *first_tp;
1537 int limit = 0;
1538
1539reclassify:
1540#endif
1541 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1542 __be16 protocol = skb_protocol(skb, false);
1543 int err;
1544
1545 if (tp->protocol != protocol &&
1546 tp->protocol != htons(ETH_P_ALL))
1547 continue;
1548
1549 err = tp->classify(skb, tp, res);
1550#ifdef CONFIG_NET_CLS_ACT
1551 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1552 first_tp = orig_tp;
1553 *last_executed_chain = first_tp->chain->index;
1554 goto reset;
1555 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1556 first_tp = res->goto_tp;
1557 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1558 goto reset;
1559 }
1560#endif
1561 if (err >= 0)
1562 return err;
1563 }
1564
1565 return TC_ACT_UNSPEC; /* signal: continue lookup */
1566#ifdef CONFIG_NET_CLS_ACT
1567reset:
1568 if (unlikely(limit++ >= max_reclassify_loop)) {
1569 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1570 tp->chain->block->index,
1571 tp->prio & 0xffff,
1572 ntohs(tp->protocol));
1573 return TC_ACT_SHOT;
1574 }
1575
1576 tp = first_tp;
1577 goto reclassify;
1578#endif
1579}
1580
1581int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1582 struct tcf_result *res, bool compat_mode)
1583{
1584 u32 last_executed_chain = 0;
1585
1586 return __tcf_classify(skb, tp, tp, res, compat_mode,
1587 &last_executed_chain);
1588}
1589EXPORT_SYMBOL(tcf_classify);
1590
1591int tcf_classify_ingress(struct sk_buff *skb,
1592 const struct tcf_block *ingress_block,
1593 const struct tcf_proto *tp,
1594 struct tcf_result *res, bool compat_mode)
1595{
1596#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1597 u32 last_executed_chain = 0;
1598
1599 return __tcf_classify(skb, tp, tp, res, compat_mode,
1600 &last_executed_chain);
1601#else
1602 u32 last_executed_chain = tp ? tp->chain->index : 0;
1603 const struct tcf_proto *orig_tp = tp;
1604 struct tc_skb_ext *ext;
1605 int ret;
1606
1607 ext = skb_ext_find(skb, TC_SKB_EXT);
1608
1609 if (ext && ext->chain) {
1610 struct tcf_chain *fchain;
1611
1612 fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain);
1613 if (!fchain)
1614 return TC_ACT_SHOT;
1615
1616 /* Consume, so cloned/redirect skbs won't inherit ext */
1617 skb_ext_del(skb, TC_SKB_EXT);
1618
1619 tp = rcu_dereference_bh(fchain->filter_chain);
1620 last_executed_chain = fchain->index;
1621 }
1622
1623 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1624 &last_executed_chain);
1625
1626 /* If we missed on some chain */
1627 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1628 ext = tc_skb_ext_alloc(skb);
1629 if (WARN_ON_ONCE(!ext))
1630 return TC_ACT_SHOT;
1631 ext->chain = last_executed_chain;
1632 ext->mru = qdisc_skb_cb(skb)->mru;
1633 ext->post_ct = qdisc_skb_cb(skb)->post_ct;
1634 }
1635
1636 return ret;
1637#endif
1638}
1639EXPORT_SYMBOL(tcf_classify_ingress);
1640
1641struct tcf_chain_info {
1642 struct tcf_proto __rcu **pprev;
1643 struct tcf_proto __rcu *next;
1644};
1645
1646static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1647 struct tcf_chain_info *chain_info)
1648{
1649 return tcf_chain_dereference(*chain_info->pprev, chain);
1650}
1651
1652static int tcf_chain_tp_insert(struct tcf_chain *chain,
1653 struct tcf_chain_info *chain_info,
1654 struct tcf_proto *tp)
1655{
1656 if (chain->flushing)
1657 return -EAGAIN;
1658
1659 if (*chain_info->pprev == chain->filter_chain)
1660 tcf_chain0_head_change(chain, tp);
1661 tcf_proto_get(tp);
1662 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1663 rcu_assign_pointer(*chain_info->pprev, tp);
1664
1665 return 0;
1666}
1667
1668static void tcf_chain_tp_remove(struct tcf_chain *chain,
1669 struct tcf_chain_info *chain_info,
1670 struct tcf_proto *tp)
1671{
1672 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1673
1674 tcf_proto_mark_delete(tp);
1675 if (tp == chain->filter_chain)
1676 tcf_chain0_head_change(chain, next);
1677 RCU_INIT_POINTER(*chain_info->pprev, next);
1678}
1679
1680static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1681 struct tcf_chain_info *chain_info,
1682 u32 protocol, u32 prio,
1683 bool prio_allocate);
1684
1685/* Try to insert new proto.
1686 * If proto with specified priority already exists, free new proto
1687 * and return existing one.
1688 */
1689
1690static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1691 struct tcf_proto *tp_new,
1692 u32 protocol, u32 prio,
1693 bool rtnl_held)
1694{
1695 struct tcf_chain_info chain_info;
1696 struct tcf_proto *tp;
1697 int err = 0;
1698
1699 mutex_lock(&chain->filter_chain_lock);
1700
1701 if (tcf_proto_exists_destroying(chain, tp_new)) {
1702 mutex_unlock(&chain->filter_chain_lock);
1703 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1704 return ERR_PTR(-EAGAIN);
1705 }
1706
1707 tp = tcf_chain_tp_find(chain, &chain_info,
1708 protocol, prio, false);
1709 if (!tp)
1710 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1711 mutex_unlock(&chain->filter_chain_lock);
1712
1713 if (tp) {
1714 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1715 tp_new = tp;
1716 } else if (err) {
1717 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1718 tp_new = ERR_PTR(err);
1719 }
1720
1721 return tp_new;
1722}
1723
1724static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1725 struct tcf_proto *tp, bool rtnl_held,
1726 struct netlink_ext_ack *extack)
1727{
1728 struct tcf_chain_info chain_info;
1729 struct tcf_proto *tp_iter;
1730 struct tcf_proto **pprev;
1731 struct tcf_proto *next;
1732
1733 mutex_lock(&chain->filter_chain_lock);
1734
1735 /* Atomically find and remove tp from chain. */
1736 for (pprev = &chain->filter_chain;
1737 (tp_iter = tcf_chain_dereference(*pprev, chain));
1738 pprev = &tp_iter->next) {
1739 if (tp_iter == tp) {
1740 chain_info.pprev = pprev;
1741 chain_info.next = tp_iter->next;
1742 WARN_ON(tp_iter->deleting);
1743 break;
1744 }
1745 }
1746 /* Verify that tp still exists and no new filters were inserted
1747 * concurrently.
1748 * Mark tp for deletion if it is empty.
1749 */
1750 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1751 mutex_unlock(&chain->filter_chain_lock);
1752 return;
1753 }
1754
1755 tcf_proto_signal_destroying(chain, tp);
1756 next = tcf_chain_dereference(chain_info.next, chain);
1757 if (tp == chain->filter_chain)
1758 tcf_chain0_head_change(chain, next);
1759 RCU_INIT_POINTER(*chain_info.pprev, next);
1760 mutex_unlock(&chain->filter_chain_lock);
1761
1762 tcf_proto_put(tp, rtnl_held, extack);
1763}
1764
1765static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1766 struct tcf_chain_info *chain_info,
1767 u32 protocol, u32 prio,
1768 bool prio_allocate)
1769{
1770 struct tcf_proto **pprev;
1771 struct tcf_proto *tp;
1772
1773 /* Check the chain for existence of proto-tcf with this priority */
1774 for (pprev = &chain->filter_chain;
1775 (tp = tcf_chain_dereference(*pprev, chain));
1776 pprev = &tp->next) {
1777 if (tp->prio >= prio) {
1778 if (tp->prio == prio) {
1779 if (prio_allocate ||
1780 (tp->protocol != protocol && protocol))
1781 return ERR_PTR(-EINVAL);
1782 } else {
1783 tp = NULL;
1784 }
1785 break;
1786 }
1787 }
1788 chain_info->pprev = pprev;
1789 if (tp) {
1790 chain_info->next = tp->next;
1791 tcf_proto_get(tp);
1792 } else {
1793 chain_info->next = NULL;
1794 }
1795 return tp;
1796}
1797
1798static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1799 struct tcf_proto *tp, struct tcf_block *block,
1800 struct Qdisc *q, u32 parent, void *fh,
1801 u32 portid, u32 seq, u16 flags, int event,
1802 bool terse_dump, bool rtnl_held)
1803{
1804 struct tcmsg *tcm;
1805 struct nlmsghdr *nlh;
1806 unsigned char *b = skb_tail_pointer(skb);
1807
1808 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1809 if (!nlh)
1810 goto out_nlmsg_trim;
1811 tcm = nlmsg_data(nlh);
1812 tcm->tcm_family = AF_UNSPEC;
1813 tcm->tcm__pad1 = 0;
1814 tcm->tcm__pad2 = 0;
1815 if (q) {
1816 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1817 tcm->tcm_parent = parent;
1818 } else {
1819 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1820 tcm->tcm_block_index = block->index;
1821 }
1822 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1823 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1824 goto nla_put_failure;
1825 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1826 goto nla_put_failure;
1827 if (!fh) {
1828 tcm->tcm_handle = 0;
1829 } else if (terse_dump) {
1830 if (tp->ops->terse_dump) {
1831 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1832 rtnl_held) < 0)
1833 goto nla_put_failure;
1834 } else {
1835 goto cls_op_not_supp;
1836 }
1837 } else {
1838 if (tp->ops->dump &&
1839 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1840 goto nla_put_failure;
1841 }
1842 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1843 return skb->len;
1844
1845out_nlmsg_trim:
1846nla_put_failure:
1847cls_op_not_supp:
1848 nlmsg_trim(skb, b);
1849 return -1;
1850}
1851
1852static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1853 struct nlmsghdr *n, struct tcf_proto *tp,
1854 struct tcf_block *block, struct Qdisc *q,
1855 u32 parent, void *fh, int event, bool unicast,
1856 bool rtnl_held)
1857{
1858 struct sk_buff *skb;
1859 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1860 int err = 0;
1861
1862 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1863 if (!skb)
1864 return -ENOBUFS;
1865
1866 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1867 n->nlmsg_seq, n->nlmsg_flags, event,
1868 false, rtnl_held) <= 0) {
1869 kfree_skb(skb);
1870 return -EINVAL;
1871 }
1872
1873 if (unicast)
1874 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1875 else
1876 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1877 n->nlmsg_flags & NLM_F_ECHO);
1878
1879 if (err > 0)
1880 err = 0;
1881 return err;
1882}
1883
1884static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1885 struct nlmsghdr *n, struct tcf_proto *tp,
1886 struct tcf_block *block, struct Qdisc *q,
1887 u32 parent, void *fh, bool unicast, bool *last,
1888 bool rtnl_held, struct netlink_ext_ack *extack)
1889{
1890 struct sk_buff *skb;
1891 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1892 int err;
1893
1894 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1895 if (!skb)
1896 return -ENOBUFS;
1897
1898 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1899 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1900 false, rtnl_held) <= 0) {
1901 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1902 kfree_skb(skb);
1903 return -EINVAL;
1904 }
1905
1906 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1907 if (err) {
1908 kfree_skb(skb);
1909 return err;
1910 }
1911
1912 if (unicast)
1913 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1914 else
1915 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1916 n->nlmsg_flags & NLM_F_ECHO);
1917 if (err < 0)
1918 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1919
1920 if (err > 0)
1921 err = 0;
1922 return err;
1923}
1924
1925static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1926 struct tcf_block *block, struct Qdisc *q,
1927 u32 parent, struct nlmsghdr *n,
1928 struct tcf_chain *chain, int event)
1929{
1930 struct tcf_proto *tp;
1931
1932 for (tp = tcf_get_next_proto(chain, NULL);
1933 tp; tp = tcf_get_next_proto(chain, tp))
1934 tfilter_notify(net, oskb, n, tp, block,
1935 q, parent, NULL, event, false, true);
1936}
1937
1938static void tfilter_put(struct tcf_proto *tp, void *fh)
1939{
1940 if (tp->ops->put && fh)
1941 tp->ops->put(tp, fh);
1942}
1943
1944static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1945 struct netlink_ext_ack *extack)
1946{
1947 struct net *net = sock_net(skb->sk);
1948 struct nlattr *tca[TCA_MAX + 1];
1949 char name[IFNAMSIZ];
1950 struct tcmsg *t;
1951 u32 protocol;
1952 u32 prio;
1953 bool prio_allocate;
1954 u32 parent;
1955 u32 chain_index;
1956 struct Qdisc *q = NULL;
1957 struct tcf_chain_info chain_info;
1958 struct tcf_chain *chain = NULL;
1959 struct tcf_block *block;
1960 struct tcf_proto *tp;
1961 unsigned long cl;
1962 void *fh;
1963 int err;
1964 int tp_created;
1965 bool rtnl_held = false;
1966
1967 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1968 return -EPERM;
1969
1970replay:
1971 tp_created = 0;
1972
1973 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1974 rtm_tca_policy, extack);
1975 if (err < 0)
1976 return err;
1977
1978 t = nlmsg_data(n);
1979 protocol = TC_H_MIN(t->tcm_info);
1980 prio = TC_H_MAJ(t->tcm_info);
1981 prio_allocate = false;
1982 parent = t->tcm_parent;
1983 tp = NULL;
1984 cl = 0;
1985 block = NULL;
1986
1987 if (prio == 0) {
1988 /* If no priority is provided by the user,
1989 * we allocate one.
1990 */
1991 if (n->nlmsg_flags & NLM_F_CREATE) {
1992 prio = TC_H_MAKE(0x80000000U, 0U);
1993 prio_allocate = true;
1994 } else {
1995 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1996 return -ENOENT;
1997 }
1998 }
1999
2000 /* Find head of filter chain. */
2001
2002 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2003 if (err)
2004 return err;
2005
2006 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2007 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2008 err = -EINVAL;
2009 goto errout;
2010 }
2011
2012 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2013 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2014 * type is not specified, classifier is not unlocked.
2015 */
2016 if (rtnl_held ||
2017 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2018 !tcf_proto_is_unlocked(name)) {
2019 rtnl_held = true;
2020 rtnl_lock();
2021 }
2022
2023 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2024 if (err)
2025 goto errout;
2026
2027 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2028 extack);
2029 if (IS_ERR(block)) {
2030 err = PTR_ERR(block);
2031 goto errout;
2032 }
2033 block->classid = parent;
2034
2035 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2036 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2037 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2038 err = -EINVAL;
2039 goto errout;
2040 }
2041 chain = tcf_chain_get(block, chain_index, true);
2042 if (!chain) {
2043 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2044 err = -ENOMEM;
2045 goto errout;
2046 }
2047
2048 mutex_lock(&chain->filter_chain_lock);
2049 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2050 prio, prio_allocate);
2051 if (IS_ERR(tp)) {
2052 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2053 err = PTR_ERR(tp);
2054 goto errout_locked;
2055 }
2056
2057 if (tp == NULL) {
2058 struct tcf_proto *tp_new = NULL;
2059
2060 if (chain->flushing) {
2061 err = -EAGAIN;
2062 goto errout_locked;
2063 }
2064
2065 /* Proto-tcf does not exist, create new one */
2066
2067 if (tca[TCA_KIND] == NULL || !protocol) {
2068 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2069 err = -EINVAL;
2070 goto errout_locked;
2071 }
2072
2073 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2074 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2075 err = -ENOENT;
2076 goto errout_locked;
2077 }
2078
2079 if (prio_allocate)
2080 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2081 &chain_info));
2082
2083 mutex_unlock(&chain->filter_chain_lock);
2084 tp_new = tcf_proto_create(name, protocol, prio, chain,
2085 rtnl_held, extack);
2086 if (IS_ERR(tp_new)) {
2087 err = PTR_ERR(tp_new);
2088 goto errout_tp;
2089 }
2090
2091 tp_created = 1;
2092 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2093 rtnl_held);
2094 if (IS_ERR(tp)) {
2095 err = PTR_ERR(tp);
2096 goto errout_tp;
2097 }
2098 } else {
2099 mutex_unlock(&chain->filter_chain_lock);
2100 }
2101
2102 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2103 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2104 err = -EINVAL;
2105 goto errout;
2106 }
2107
2108 fh = tp->ops->get(tp, t->tcm_handle);
2109
2110 if (!fh) {
2111 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2112 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2113 err = -ENOENT;
2114 goto errout;
2115 }
2116 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2117 tfilter_put(tp, fh);
2118 NL_SET_ERR_MSG(extack, "Filter already exists");
2119 err = -EEXIST;
2120 goto errout;
2121 }
2122
2123 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2124 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2125 err = -EINVAL;
2126 goto errout;
2127 }
2128
2129 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2130 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2131 rtnl_held, extack);
2132 if (err == 0) {
2133 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2134 RTM_NEWTFILTER, false, rtnl_held);
2135 tfilter_put(tp, fh);
2136 /* q pointer is NULL for shared blocks */
2137 if (q)
2138 q->flags &= ~TCQ_F_CAN_BYPASS;
2139 }
2140
2141errout:
2142 if (err && tp_created)
2143 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2144errout_tp:
2145 if (chain) {
2146 if (tp && !IS_ERR(tp))
2147 tcf_proto_put(tp, rtnl_held, NULL);
2148 if (!tp_created)
2149 tcf_chain_put(chain);
2150 }
2151 tcf_block_release(q, block, rtnl_held);
2152
2153 if (rtnl_held)
2154 rtnl_unlock();
2155
2156 if (err == -EAGAIN) {
2157 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2158 * of target chain.
2159 */
2160 rtnl_held = true;
2161 /* Replay the request. */
2162 goto replay;
2163 }
2164 return err;
2165
2166errout_locked:
2167 mutex_unlock(&chain->filter_chain_lock);
2168 goto errout;
2169}
2170
2171static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2172 struct netlink_ext_ack *extack)
2173{
2174 struct net *net = sock_net(skb->sk);
2175 struct nlattr *tca[TCA_MAX + 1];
2176 char name[IFNAMSIZ];
2177 struct tcmsg *t;
2178 u32 protocol;
2179 u32 prio;
2180 u32 parent;
2181 u32 chain_index;
2182 struct Qdisc *q = NULL;
2183 struct tcf_chain_info chain_info;
2184 struct tcf_chain *chain = NULL;
2185 struct tcf_block *block = NULL;
2186 struct tcf_proto *tp = NULL;
2187 unsigned long cl = 0;
2188 void *fh = NULL;
2189 int err;
2190 bool rtnl_held = false;
2191
2192 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2193 return -EPERM;
2194
2195 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2196 rtm_tca_policy, extack);
2197 if (err < 0)
2198 return err;
2199
2200 t = nlmsg_data(n);
2201 protocol = TC_H_MIN(t->tcm_info);
2202 prio = TC_H_MAJ(t->tcm_info);
2203 parent = t->tcm_parent;
2204
2205 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2206 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2207 return -ENOENT;
2208 }
2209
2210 /* Find head of filter chain. */
2211
2212 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2213 if (err)
2214 return err;
2215
2216 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2217 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2218 err = -EINVAL;
2219 goto errout;
2220 }
2221 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2222 * found), qdisc is not unlocked, classifier type is not specified,
2223 * classifier is not unlocked.
2224 */
2225 if (!prio ||
2226 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2227 !tcf_proto_is_unlocked(name)) {
2228 rtnl_held = true;
2229 rtnl_lock();
2230 }
2231
2232 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2233 if (err)
2234 goto errout;
2235
2236 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2237 extack);
2238 if (IS_ERR(block)) {
2239 err = PTR_ERR(block);
2240 goto errout;
2241 }
2242
2243 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2244 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2245 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2246 err = -EINVAL;
2247 goto errout;
2248 }
2249 chain = tcf_chain_get(block, chain_index, false);
2250 if (!chain) {
2251 /* User requested flush on non-existent chain. Nothing to do,
2252 * so just return success.
2253 */
2254 if (prio == 0) {
2255 err = 0;
2256 goto errout;
2257 }
2258 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2259 err = -ENOENT;
2260 goto errout;
2261 }
2262
2263 if (prio == 0) {
2264 tfilter_notify_chain(net, skb, block, q, parent, n,
2265 chain, RTM_DELTFILTER);
2266 tcf_chain_flush(chain, rtnl_held);
2267 err = 0;
2268 goto errout;
2269 }
2270
2271 mutex_lock(&chain->filter_chain_lock);
2272 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2273 prio, false);
2274 if (!tp || IS_ERR(tp)) {
2275 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2276 err = tp ? PTR_ERR(tp) : -ENOENT;
2277 goto errout_locked;
2278 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2279 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2280 err = -EINVAL;
2281 goto errout_locked;
2282 } else if (t->tcm_handle == 0) {
2283 tcf_proto_signal_destroying(chain, tp);
2284 tcf_chain_tp_remove(chain, &chain_info, tp);
2285 mutex_unlock(&chain->filter_chain_lock);
2286
2287 tcf_proto_put(tp, rtnl_held, NULL);
2288 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2289 RTM_DELTFILTER, false, rtnl_held);
2290 err = 0;
2291 goto errout;
2292 }
2293 mutex_unlock(&chain->filter_chain_lock);
2294
2295 fh = tp->ops->get(tp, t->tcm_handle);
2296
2297 if (!fh) {
2298 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2299 err = -ENOENT;
2300 } else {
2301 bool last;
2302
2303 err = tfilter_del_notify(net, skb, n, tp, block,
2304 q, parent, fh, false, &last,
2305 rtnl_held, extack);
2306
2307 if (err)
2308 goto errout;
2309 if (last)
2310 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2311 }
2312
2313errout:
2314 if (chain) {
2315 if (tp && !IS_ERR(tp))
2316 tcf_proto_put(tp, rtnl_held, NULL);
2317 tcf_chain_put(chain);
2318 }
2319 tcf_block_release(q, block, rtnl_held);
2320
2321 if (rtnl_held)
2322 rtnl_unlock();
2323
2324 return err;
2325
2326errout_locked:
2327 mutex_unlock(&chain->filter_chain_lock);
2328 goto errout;
2329}
2330
2331static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2332 struct netlink_ext_ack *extack)
2333{
2334 struct net *net = sock_net(skb->sk);
2335 struct nlattr *tca[TCA_MAX + 1];
2336 char name[IFNAMSIZ];
2337 struct tcmsg *t;
2338 u32 protocol;
2339 u32 prio;
2340 u32 parent;
2341 u32 chain_index;
2342 struct Qdisc *q = NULL;
2343 struct tcf_chain_info chain_info;
2344 struct tcf_chain *chain = NULL;
2345 struct tcf_block *block = NULL;
2346 struct tcf_proto *tp = NULL;
2347 unsigned long cl = 0;
2348 void *fh = NULL;
2349 int err;
2350 bool rtnl_held = false;
2351
2352 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2353 rtm_tca_policy, extack);
2354 if (err < 0)
2355 return err;
2356
2357 t = nlmsg_data(n);
2358 protocol = TC_H_MIN(t->tcm_info);
2359 prio = TC_H_MAJ(t->tcm_info);
2360 parent = t->tcm_parent;
2361
2362 if (prio == 0) {
2363 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2364 return -ENOENT;
2365 }
2366
2367 /* Find head of filter chain. */
2368
2369 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2370 if (err)
2371 return err;
2372
2373 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2374 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2375 err = -EINVAL;
2376 goto errout;
2377 }
2378 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2379 * unlocked, classifier type is not specified, classifier is not
2380 * unlocked.
2381 */
2382 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2383 !tcf_proto_is_unlocked(name)) {
2384 rtnl_held = true;
2385 rtnl_lock();
2386 }
2387
2388 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2389 if (err)
2390 goto errout;
2391
2392 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2393 extack);
2394 if (IS_ERR(block)) {
2395 err = PTR_ERR(block);
2396 goto errout;
2397 }
2398
2399 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2400 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2401 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2402 err = -EINVAL;
2403 goto errout;
2404 }
2405 chain = tcf_chain_get(block, chain_index, false);
2406 if (!chain) {
2407 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2408 err = -EINVAL;
2409 goto errout;
2410 }
2411
2412 mutex_lock(&chain->filter_chain_lock);
2413 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2414 prio, false);
2415 mutex_unlock(&chain->filter_chain_lock);
2416 if (!tp || IS_ERR(tp)) {
2417 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2418 err = tp ? PTR_ERR(tp) : -ENOENT;
2419 goto errout;
2420 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2421 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2422 err = -EINVAL;
2423 goto errout;
2424 }
2425
2426 fh = tp->ops->get(tp, t->tcm_handle);
2427
2428 if (!fh) {
2429 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2430 err = -ENOENT;
2431 } else {
2432 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2433 fh, RTM_NEWTFILTER, true, rtnl_held);
2434 if (err < 0)
2435 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2436 }
2437
2438 tfilter_put(tp, fh);
2439errout:
2440 if (chain) {
2441 if (tp && !IS_ERR(tp))
2442 tcf_proto_put(tp, rtnl_held, NULL);
2443 tcf_chain_put(chain);
2444 }
2445 tcf_block_release(q, block, rtnl_held);
2446
2447 if (rtnl_held)
2448 rtnl_unlock();
2449
2450 return err;
2451}
2452
2453struct tcf_dump_args {
2454 struct tcf_walker w;
2455 struct sk_buff *skb;
2456 struct netlink_callback *cb;
2457 struct tcf_block *block;
2458 struct Qdisc *q;
2459 u32 parent;
2460 bool terse_dump;
2461};
2462
2463static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2464{
2465 struct tcf_dump_args *a = (void *)arg;
2466 struct net *net = sock_net(a->skb->sk);
2467
2468 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2469 n, NETLINK_CB(a->cb->skb).portid,
2470 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2471 RTM_NEWTFILTER, a->terse_dump, true);
2472}
2473
2474static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2475 struct sk_buff *skb, struct netlink_callback *cb,
2476 long index_start, long *p_index, bool terse)
2477{
2478 struct net *net = sock_net(skb->sk);
2479 struct tcf_block *block = chain->block;
2480 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2481 struct tcf_proto *tp, *tp_prev;
2482 struct tcf_dump_args arg;
2483
2484 for (tp = __tcf_get_next_proto(chain, NULL);
2485 tp;
2486 tp_prev = tp,
2487 tp = __tcf_get_next_proto(chain, tp),
2488 tcf_proto_put(tp_prev, true, NULL),
2489 (*p_index)++) {
2490 if (*p_index < index_start)
2491 continue;
2492 if (TC_H_MAJ(tcm->tcm_info) &&
2493 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2494 continue;
2495 if (TC_H_MIN(tcm->tcm_info) &&
2496 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2497 continue;
2498 if (*p_index > index_start)
2499 memset(&cb->args[1], 0,
2500 sizeof(cb->args) - sizeof(cb->args[0]));
2501 if (cb->args[1] == 0) {
2502 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2503 NETLINK_CB(cb->skb).portid,
2504 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2505 RTM_NEWTFILTER, false, true) <= 0)
2506 goto errout;
2507 cb->args[1] = 1;
2508 }
2509 if (!tp->ops->walk)
2510 continue;
2511 arg.w.fn = tcf_node_dump;
2512 arg.skb = skb;
2513 arg.cb = cb;
2514 arg.block = block;
2515 arg.q = q;
2516 arg.parent = parent;
2517 arg.w.stop = 0;
2518 arg.w.skip = cb->args[1] - 1;
2519 arg.w.count = 0;
2520 arg.w.cookie = cb->args[2];
2521 arg.terse_dump = terse;
2522 tp->ops->walk(tp, &arg.w, true);
2523 cb->args[2] = arg.w.cookie;
2524 cb->args[1] = arg.w.count + 1;
2525 if (arg.w.stop)
2526 goto errout;
2527 }
2528 return true;
2529
2530errout:
2531 tcf_proto_put(tp, true, NULL);
2532 return false;
2533}
2534
2535static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2536 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2537};
2538
2539/* called with RTNL */
2540static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2541{
2542 struct tcf_chain *chain, *chain_prev;
2543 struct net *net = sock_net(skb->sk);
2544 struct nlattr *tca[TCA_MAX + 1];
2545 struct Qdisc *q = NULL;
2546 struct tcf_block *block;
2547 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2548 bool terse_dump = false;
2549 long index_start;
2550 long index;
2551 u32 parent;
2552 int err;
2553
2554 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2555 return skb->len;
2556
2557 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2558 tcf_tfilter_dump_policy, cb->extack);
2559 if (err)
2560 return err;
2561
2562 if (tca[TCA_DUMP_FLAGS]) {
2563 struct nla_bitfield32 flags =
2564 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2565
2566 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2567 }
2568
2569 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2570 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2571 if (!block)
2572 goto out;
2573 /* If we work with block index, q is NULL and parent value
2574 * will never be used in the following code. The check
2575 * in tcf_fill_node prevents it. However, compiler does not
2576 * see that far, so set parent to zero to silence the warning
2577 * about parent being uninitialized.
2578 */
2579 parent = 0;
2580 } else {
2581 const struct Qdisc_class_ops *cops;
2582 struct net_device *dev;
2583 unsigned long cl = 0;
2584
2585 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2586 if (!dev)
2587 return skb->len;
2588
2589 parent = tcm->tcm_parent;
2590 if (!parent)
2591 q = dev->qdisc;
2592 else
2593 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2594 if (!q)
2595 goto out;
2596 cops = q->ops->cl_ops;
2597 if (!cops)
2598 goto out;
2599 if (!cops->tcf_block)
2600 goto out;
2601 if (TC_H_MIN(tcm->tcm_parent)) {
2602 cl = cops->find(q, tcm->tcm_parent);
2603 if (cl == 0)
2604 goto out;
2605 }
2606 block = cops->tcf_block(q, cl, NULL);
2607 if (!block)
2608 goto out;
2609 parent = block->classid;
2610 if (tcf_block_shared(block))
2611 q = NULL;
2612 }
2613
2614 index_start = cb->args[0];
2615 index = 0;
2616
2617 for (chain = __tcf_get_next_chain(block, NULL);
2618 chain;
2619 chain_prev = chain,
2620 chain = __tcf_get_next_chain(block, chain),
2621 tcf_chain_put(chain_prev)) {
2622 if (tca[TCA_CHAIN] &&
2623 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2624 continue;
2625 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2626 index_start, &index, terse_dump)) {
2627 tcf_chain_put(chain);
2628 err = -EMSGSIZE;
2629 break;
2630 }
2631 }
2632
2633 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2634 tcf_block_refcnt_put(block, true);
2635 cb->args[0] = index;
2636
2637out:
2638 /* If we did no progress, the error (EMSGSIZE) is real */
2639 if (skb->len == 0 && err)
2640 return err;
2641 return skb->len;
2642}
2643
2644static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2645 void *tmplt_priv, u32 chain_index,
2646 struct net *net, struct sk_buff *skb,
2647 struct tcf_block *block,
2648 u32 portid, u32 seq, u16 flags, int event)
2649{
2650 unsigned char *b = skb_tail_pointer(skb);
2651 const struct tcf_proto_ops *ops;
2652 struct nlmsghdr *nlh;
2653 struct tcmsg *tcm;
2654 void *priv;
2655
2656 ops = tmplt_ops;
2657 priv = tmplt_priv;
2658
2659 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2660 if (!nlh)
2661 goto out_nlmsg_trim;
2662 tcm = nlmsg_data(nlh);
2663 tcm->tcm_family = AF_UNSPEC;
2664 tcm->tcm__pad1 = 0;
2665 tcm->tcm__pad2 = 0;
2666 tcm->tcm_handle = 0;
2667 if (block->q) {
2668 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2669 tcm->tcm_parent = block->q->handle;
2670 } else {
2671 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2672 tcm->tcm_block_index = block->index;
2673 }
2674
2675 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2676 goto nla_put_failure;
2677
2678 if (ops) {
2679 if (nla_put_string(skb, TCA_KIND, ops->kind))
2680 goto nla_put_failure;
2681 if (ops->tmplt_dump(skb, net, priv) < 0)
2682 goto nla_put_failure;
2683 }
2684
2685 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2686 return skb->len;
2687
2688out_nlmsg_trim:
2689nla_put_failure:
2690 nlmsg_trim(skb, b);
2691 return -EMSGSIZE;
2692}
2693
2694static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2695 u32 seq, u16 flags, int event, bool unicast)
2696{
2697 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2698 struct tcf_block *block = chain->block;
2699 struct net *net = block->net;
2700 struct sk_buff *skb;
2701 int err = 0;
2702
2703 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2704 if (!skb)
2705 return -ENOBUFS;
2706
2707 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2708 chain->index, net, skb, block, portid,
2709 seq, flags, event) <= 0) {
2710 kfree_skb(skb);
2711 return -EINVAL;
2712 }
2713
2714 if (unicast)
2715 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2716 else
2717 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2718 flags & NLM_F_ECHO);
2719
2720 if (err > 0)
2721 err = 0;
2722 return err;
2723}
2724
2725static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2726 void *tmplt_priv, u32 chain_index,
2727 struct tcf_block *block, struct sk_buff *oskb,
2728 u32 seq, u16 flags, bool unicast)
2729{
2730 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2731 struct net *net = block->net;
2732 struct sk_buff *skb;
2733
2734 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2735 if (!skb)
2736 return -ENOBUFS;
2737
2738 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2739 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2740 kfree_skb(skb);
2741 return -EINVAL;
2742 }
2743
2744 if (unicast)
2745 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2746
2747 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2748}
2749
2750static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2751 struct nlattr **tca,
2752 struct netlink_ext_ack *extack)
2753{
2754 const struct tcf_proto_ops *ops;
2755 char name[IFNAMSIZ];
2756 void *tmplt_priv;
2757
2758 /* If kind is not set, user did not specify template. */
2759 if (!tca[TCA_KIND])
2760 return 0;
2761
2762 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2763 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2764 return -EINVAL;
2765 }
2766
2767 ops = tcf_proto_lookup_ops(name, true, extack);
2768 if (IS_ERR(ops))
2769 return PTR_ERR(ops);
2770 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2771 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2772 return -EOPNOTSUPP;
2773 }
2774
2775 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2776 if (IS_ERR(tmplt_priv)) {
2777 module_put(ops->owner);
2778 return PTR_ERR(tmplt_priv);
2779 }
2780 chain->tmplt_ops = ops;
2781 chain->tmplt_priv = tmplt_priv;
2782 return 0;
2783}
2784
2785static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2786 void *tmplt_priv)
2787{
2788 /* If template ops are set, no work to do for us. */
2789 if (!tmplt_ops)
2790 return;
2791
2792 tmplt_ops->tmplt_destroy(tmplt_priv);
2793 module_put(tmplt_ops->owner);
2794}
2795
2796/* Add/delete/get a chain */
2797
2798static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2799 struct netlink_ext_ack *extack)
2800{
2801 struct net *net = sock_net(skb->sk);
2802 struct nlattr *tca[TCA_MAX + 1];
2803 struct tcmsg *t;
2804 u32 parent;
2805 u32 chain_index;
2806 struct Qdisc *q = NULL;
2807 struct tcf_chain *chain = NULL;
2808 struct tcf_block *block;
2809 unsigned long cl;
2810 int err;
2811
2812 if (n->nlmsg_type != RTM_GETCHAIN &&
2813 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2814 return -EPERM;
2815
2816replay:
2817 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2818 rtm_tca_policy, extack);
2819 if (err < 0)
2820 return err;
2821
2822 t = nlmsg_data(n);
2823 parent = t->tcm_parent;
2824 cl = 0;
2825
2826 block = tcf_block_find(net, &q, &parent, &cl,
2827 t->tcm_ifindex, t->tcm_block_index, extack);
2828 if (IS_ERR(block))
2829 return PTR_ERR(block);
2830
2831 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2832 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2833 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2834 err = -EINVAL;
2835 goto errout_block;
2836 }
2837
2838 mutex_lock(&block->lock);
2839 chain = tcf_chain_lookup(block, chain_index);
2840 if (n->nlmsg_type == RTM_NEWCHAIN) {
2841 if (chain) {
2842 if (tcf_chain_held_by_acts_only(chain)) {
2843 /* The chain exists only because there is
2844 * some action referencing it.
2845 */
2846 tcf_chain_hold(chain);
2847 } else {
2848 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2849 err = -EEXIST;
2850 goto errout_block_locked;
2851 }
2852 } else {
2853 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2854 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2855 err = -ENOENT;
2856 goto errout_block_locked;
2857 }
2858 chain = tcf_chain_create(block, chain_index);
2859 if (!chain) {
2860 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2861 err = -ENOMEM;
2862 goto errout_block_locked;
2863 }
2864 }
2865 } else {
2866 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2867 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2868 err = -EINVAL;
2869 goto errout_block_locked;
2870 }
2871 tcf_chain_hold(chain);
2872 }
2873
2874 if (n->nlmsg_type == RTM_NEWCHAIN) {
2875 /* Modifying chain requires holding parent block lock. In case
2876 * the chain was successfully added, take a reference to the
2877 * chain. This ensures that an empty chain does not disappear at
2878 * the end of this function.
2879 */
2880 tcf_chain_hold(chain);
2881 chain->explicitly_created = true;
2882 }
2883 mutex_unlock(&block->lock);
2884
2885 switch (n->nlmsg_type) {
2886 case RTM_NEWCHAIN:
2887 err = tc_chain_tmplt_add(chain, net, tca, extack);
2888 if (err) {
2889 tcf_chain_put_explicitly_created(chain);
2890 goto errout;
2891 }
2892
2893 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2894 RTM_NEWCHAIN, false);
2895 break;
2896 case RTM_DELCHAIN:
2897 tfilter_notify_chain(net, skb, block, q, parent, n,
2898 chain, RTM_DELTFILTER);
2899 /* Flush the chain first as the user requested chain removal. */
2900 tcf_chain_flush(chain, true);
2901 /* In case the chain was successfully deleted, put a reference
2902 * to the chain previously taken during addition.
2903 */
2904 tcf_chain_put_explicitly_created(chain);
2905 break;
2906 case RTM_GETCHAIN:
2907 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2908 n->nlmsg_flags, n->nlmsg_type, true);
2909 if (err < 0)
2910 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2911 break;
2912 default:
2913 err = -EOPNOTSUPP;
2914 NL_SET_ERR_MSG(extack, "Unsupported message type");
2915 goto errout;
2916 }
2917
2918errout:
2919 tcf_chain_put(chain);
2920errout_block:
2921 tcf_block_release(q, block, true);
2922 if (err == -EAGAIN)
2923 /* Replay the request. */
2924 goto replay;
2925 return err;
2926
2927errout_block_locked:
2928 mutex_unlock(&block->lock);
2929 goto errout_block;
2930}
2931
2932/* called with RTNL */
2933static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2934{
2935 struct net *net = sock_net(skb->sk);
2936 struct nlattr *tca[TCA_MAX + 1];
2937 struct Qdisc *q = NULL;
2938 struct tcf_block *block;
2939 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2940 struct tcf_chain *chain;
2941 long index_start;
2942 long index;
2943 int err;
2944
2945 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2946 return skb->len;
2947
2948 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2949 rtm_tca_policy, cb->extack);
2950 if (err)
2951 return err;
2952
2953 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2954 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2955 if (!block)
2956 goto out;
2957 } else {
2958 const struct Qdisc_class_ops *cops;
2959 struct net_device *dev;
2960 unsigned long cl = 0;
2961
2962 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2963 if (!dev)
2964 return skb->len;
2965
2966 if (!tcm->tcm_parent)
2967 q = dev->qdisc;
2968 else
2969 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2970
2971 if (!q)
2972 goto out;
2973 cops = q->ops->cl_ops;
2974 if (!cops)
2975 goto out;
2976 if (!cops->tcf_block)
2977 goto out;
2978 if (TC_H_MIN(tcm->tcm_parent)) {
2979 cl = cops->find(q, tcm->tcm_parent);
2980 if (cl == 0)
2981 goto out;
2982 }
2983 block = cops->tcf_block(q, cl, NULL);
2984 if (!block)
2985 goto out;
2986 if (tcf_block_shared(block))
2987 q = NULL;
2988 }
2989
2990 index_start = cb->args[0];
2991 index = 0;
2992
2993 mutex_lock(&block->lock);
2994 list_for_each_entry(chain, &block->chain_list, list) {
2995 if ((tca[TCA_CHAIN] &&
2996 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2997 continue;
2998 if (index < index_start) {
2999 index++;
3000 continue;
3001 }
3002 if (tcf_chain_held_by_acts_only(chain))
3003 continue;
3004 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3005 chain->index, net, skb, block,
3006 NETLINK_CB(cb->skb).portid,
3007 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3008 RTM_NEWCHAIN);
3009 if (err <= 0)
3010 break;
3011 index++;
3012 }
3013 mutex_unlock(&block->lock);
3014
3015 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3016 tcf_block_refcnt_put(block, true);
3017 cb->args[0] = index;
3018
3019out:
3020 /* If we did no progress, the error (EMSGSIZE) is real */
3021 if (skb->len == 0 && err)
3022 return err;
3023 return skb->len;
3024}
3025
3026void tcf_exts_destroy(struct tcf_exts *exts)
3027{
3028#ifdef CONFIG_NET_CLS_ACT
3029 if (exts->actions) {
3030 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3031 kfree(exts->actions);
3032 }
3033 exts->nr_actions = 0;
3034#endif
3035}
3036EXPORT_SYMBOL(tcf_exts_destroy);
3037
3038int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3039 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3040 bool rtnl_held, struct netlink_ext_ack *extack)
3041{
3042#ifdef CONFIG_NET_CLS_ACT
3043 {
3044 int init_res[TCA_ACT_MAX_PRIO] = {};
3045 struct tc_action *act;
3046 size_t attr_size = 0;
3047
3048 if (exts->police && tb[exts->police]) {
3049 struct tc_action_ops *a_o;
3050
3051 a_o = tc_action_load_ops("police", tb[exts->police], rtnl_held, extack);
3052 if (IS_ERR(a_o))
3053 return PTR_ERR(a_o);
3054 act = tcf_action_init_1(net, tp, tb[exts->police],
3055 rate_tlv, "police", ovr,
3056 TCA_ACT_BIND, a_o, init_res,
3057 rtnl_held, extack);
3058 module_put(a_o->owner);
3059 if (IS_ERR(act))
3060 return PTR_ERR(act);
3061
3062 act->type = exts->type = TCA_OLD_COMPAT;
3063 exts->actions[0] = act;
3064 exts->nr_actions = 1;
3065 tcf_idr_insert_many(exts->actions);
3066 } else if (exts->action && tb[exts->action]) {
3067 int err;
3068
3069 err = tcf_action_init(net, tp, tb[exts->action],
3070 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3071 exts->actions, init_res,
3072 &attr_size, rtnl_held, extack);
3073 if (err < 0)
3074 return err;
3075 exts->nr_actions = err;
3076 }
3077 }
3078#else
3079 if ((exts->action && tb[exts->action]) ||
3080 (exts->police && tb[exts->police])) {
3081 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3082 return -EOPNOTSUPP;
3083 }
3084#endif
3085
3086 return 0;
3087}
3088EXPORT_SYMBOL(tcf_exts_validate);
3089
3090void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3091{
3092#ifdef CONFIG_NET_CLS_ACT
3093 struct tcf_exts old = *dst;
3094
3095 *dst = *src;
3096 tcf_exts_destroy(&old);
3097#endif
3098}
3099EXPORT_SYMBOL(tcf_exts_change);
3100
3101#ifdef CONFIG_NET_CLS_ACT
3102static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3103{
3104 if (exts->nr_actions == 0)
3105 return NULL;
3106 else
3107 return exts->actions[0];
3108}
3109#endif
3110
3111int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3112{
3113#ifdef CONFIG_NET_CLS_ACT
3114 struct nlattr *nest;
3115
3116 if (exts->action && tcf_exts_has_actions(exts)) {
3117 /*
3118 * again for backward compatible mode - we want
3119 * to work with both old and new modes of entering
3120 * tc data even if iproute2 was newer - jhs
3121 */
3122 if (exts->type != TCA_OLD_COMPAT) {
3123 nest = nla_nest_start_noflag(skb, exts->action);
3124 if (nest == NULL)
3125 goto nla_put_failure;
3126
3127 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3128 < 0)
3129 goto nla_put_failure;
3130 nla_nest_end(skb, nest);
3131 } else if (exts->police) {
3132 struct tc_action *act = tcf_exts_first_act(exts);
3133 nest = nla_nest_start_noflag(skb, exts->police);
3134 if (nest == NULL || !act)
3135 goto nla_put_failure;
3136 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3137 goto nla_put_failure;
3138 nla_nest_end(skb, nest);
3139 }
3140 }
3141 return 0;
3142
3143nla_put_failure:
3144 nla_nest_cancel(skb, nest);
3145 return -1;
3146#else
3147 return 0;
3148#endif
3149}
3150EXPORT_SYMBOL(tcf_exts_dump);
3151
3152int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3153{
3154#ifdef CONFIG_NET_CLS_ACT
3155 struct nlattr *nest;
3156
3157 if (!exts->action || !tcf_exts_has_actions(exts))
3158 return 0;
3159
3160 nest = nla_nest_start_noflag(skb, exts->action);
3161 if (!nest)
3162 goto nla_put_failure;
3163
3164 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3165 goto nla_put_failure;
3166 nla_nest_end(skb, nest);
3167 return 0;
3168
3169nla_put_failure:
3170 nla_nest_cancel(skb, nest);
3171 return -1;
3172#else
3173 return 0;
3174#endif
3175}
3176EXPORT_SYMBOL(tcf_exts_terse_dump);
3177
3178int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3179{
3180#ifdef CONFIG_NET_CLS_ACT
3181 struct tc_action *a = tcf_exts_first_act(exts);
3182 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3183 return -1;
3184#endif
3185 return 0;
3186}
3187EXPORT_SYMBOL(tcf_exts_dump_stats);
3188
3189static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3190{
3191 if (*flags & TCA_CLS_FLAGS_IN_HW)
3192 return;
3193 *flags |= TCA_CLS_FLAGS_IN_HW;
3194 atomic_inc(&block->offloadcnt);
3195}
3196
3197static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3198{
3199 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3200 return;
3201 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3202 atomic_dec(&block->offloadcnt);
3203}
3204
3205static void tc_cls_offload_cnt_update(struct tcf_block *block,
3206 struct tcf_proto *tp, u32 *cnt,
3207 u32 *flags, u32 diff, bool add)
3208{
3209 lockdep_assert_held(&block->cb_lock);
3210
3211 spin_lock(&tp->lock);
3212 if (add) {
3213 if (!*cnt)
3214 tcf_block_offload_inc(block, flags);
3215 *cnt += diff;
3216 } else {
3217 *cnt -= diff;
3218 if (!*cnt)
3219 tcf_block_offload_dec(block, flags);
3220 }
3221 spin_unlock(&tp->lock);
3222}
3223
3224static void
3225tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3226 u32 *cnt, u32 *flags)
3227{
3228 lockdep_assert_held(&block->cb_lock);
3229
3230 spin_lock(&tp->lock);
3231 tcf_block_offload_dec(block, flags);
3232 *cnt = 0;
3233 spin_unlock(&tp->lock);
3234}
3235
3236static int
3237__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3238 void *type_data, bool err_stop)
3239{
3240 struct flow_block_cb *block_cb;
3241 int ok_count = 0;
3242 int err;
3243
3244 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3245 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3246 if (err) {
3247 if (err_stop)
3248 return err;
3249 } else {
3250 ok_count++;
3251 }
3252 }
3253 return ok_count;
3254}
3255
3256int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3257 void *type_data, bool err_stop, bool rtnl_held)
3258{
3259 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3260 int ok_count;
3261
3262retry:
3263 if (take_rtnl)
3264 rtnl_lock();
3265 down_read(&block->cb_lock);
3266 /* Need to obtain rtnl lock if block is bound to devs that require it.
3267 * In block bind code cb_lock is obtained while holding rtnl, so we must
3268 * obtain the locks in same order here.
3269 */
3270 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3271 up_read(&block->cb_lock);
3272 take_rtnl = true;
3273 goto retry;
3274 }
3275
3276 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3277
3278 up_read(&block->cb_lock);
3279 if (take_rtnl)
3280 rtnl_unlock();
3281 return ok_count;
3282}
3283EXPORT_SYMBOL(tc_setup_cb_call);
3284
3285/* Non-destructive filter add. If filter that wasn't already in hardware is
3286 * successfully offloaded, increment block offloads counter. On failure,
3287 * previously offloaded filter is considered to be intact and offloads counter
3288 * is not decremented.
3289 */
3290
3291int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3292 enum tc_setup_type type, void *type_data, bool err_stop,
3293 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3294{
3295 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3296 int ok_count;
3297
3298retry:
3299 if (take_rtnl)
3300 rtnl_lock();
3301 down_read(&block->cb_lock);
3302 /* Need to obtain rtnl lock if block is bound to devs that require it.
3303 * In block bind code cb_lock is obtained while holding rtnl, so we must
3304 * obtain the locks in same order here.
3305 */
3306 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3307 up_read(&block->cb_lock);
3308 take_rtnl = true;
3309 goto retry;
3310 }
3311
3312 /* Make sure all netdevs sharing this block are offload-capable. */
3313 if (block->nooffloaddevcnt && err_stop) {
3314 ok_count = -EOPNOTSUPP;
3315 goto err_unlock;
3316 }
3317
3318 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3319 if (ok_count < 0)
3320 goto err_unlock;
3321
3322 if (tp->ops->hw_add)
3323 tp->ops->hw_add(tp, type_data);
3324 if (ok_count > 0)
3325 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3326 ok_count, true);
3327err_unlock:
3328 up_read(&block->cb_lock);
3329 if (take_rtnl)
3330 rtnl_unlock();
3331 return ok_count < 0 ? ok_count : 0;
3332}
3333EXPORT_SYMBOL(tc_setup_cb_add);
3334
3335/* Destructive filter replace. If filter that wasn't already in hardware is
3336 * successfully offloaded, increment block offload counter. On failure,
3337 * previously offloaded filter is considered to be destroyed and offload counter
3338 * is decremented.
3339 */
3340
3341int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3342 enum tc_setup_type type, void *type_data, bool err_stop,
3343 u32 *old_flags, unsigned int *old_in_hw_count,
3344 u32 *new_flags, unsigned int *new_in_hw_count,
3345 bool rtnl_held)
3346{
3347 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3348 int ok_count;
3349
3350retry:
3351 if (take_rtnl)
3352 rtnl_lock();
3353 down_read(&block->cb_lock);
3354 /* Need to obtain rtnl lock if block is bound to devs that require it.
3355 * In block bind code cb_lock is obtained while holding rtnl, so we must
3356 * obtain the locks in same order here.
3357 */
3358 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3359 up_read(&block->cb_lock);
3360 take_rtnl = true;
3361 goto retry;
3362 }
3363
3364 /* Make sure all netdevs sharing this block are offload-capable. */
3365 if (block->nooffloaddevcnt && err_stop) {
3366 ok_count = -EOPNOTSUPP;
3367 goto err_unlock;
3368 }
3369
3370 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3371 if (tp->ops->hw_del)
3372 tp->ops->hw_del(tp, type_data);
3373
3374 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3375 if (ok_count < 0)
3376 goto err_unlock;
3377
3378 if (tp->ops->hw_add)
3379 tp->ops->hw_add(tp, type_data);
3380 if (ok_count > 0)
3381 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3382 new_flags, ok_count, true);
3383err_unlock:
3384 up_read(&block->cb_lock);
3385 if (take_rtnl)
3386 rtnl_unlock();
3387 return ok_count < 0 ? ok_count : 0;
3388}
3389EXPORT_SYMBOL(tc_setup_cb_replace);
3390
3391/* Destroy filter and decrement block offload counter, if filter was previously
3392 * offloaded.
3393 */
3394
3395int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3396 enum tc_setup_type type, void *type_data, bool err_stop,
3397 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3398{
3399 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3400 int ok_count;
3401
3402retry:
3403 if (take_rtnl)
3404 rtnl_lock();
3405 down_read(&block->cb_lock);
3406 /* Need to obtain rtnl lock if block is bound to devs that require it.
3407 * In block bind code cb_lock is obtained while holding rtnl, so we must
3408 * obtain the locks in same order here.
3409 */
3410 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3411 up_read(&block->cb_lock);
3412 take_rtnl = true;
3413 goto retry;
3414 }
3415
3416 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3417
3418 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3419 if (tp->ops->hw_del)
3420 tp->ops->hw_del(tp, type_data);
3421
3422 up_read(&block->cb_lock);
3423 if (take_rtnl)
3424 rtnl_unlock();
3425 return ok_count < 0 ? ok_count : 0;
3426}
3427EXPORT_SYMBOL(tc_setup_cb_destroy);
3428
3429int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3430 bool add, flow_setup_cb_t *cb,
3431 enum tc_setup_type type, void *type_data,
3432 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3433{
3434 int err = cb(type, type_data, cb_priv);
3435
3436 if (err) {
3437 if (add && tc_skip_sw(*flags))
3438 return err;
3439 } else {
3440 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3441 add);
3442 }
3443
3444 return 0;
3445}
3446EXPORT_SYMBOL(tc_setup_cb_reoffload);
3447
3448static int tcf_act_get_cookie(struct flow_action_entry *entry,
3449 const struct tc_action *act)
3450{
3451 struct tc_cookie *cookie;
3452 int err = 0;
3453
3454 rcu_read_lock();
3455 cookie = rcu_dereference(act->act_cookie);
3456 if (cookie) {
3457 entry->cookie = flow_action_cookie_create(cookie->data,
3458 cookie->len,
3459 GFP_ATOMIC);
3460 if (!entry->cookie)
3461 err = -ENOMEM;
3462 }
3463 rcu_read_unlock();
3464 return err;
3465}
3466
3467static void tcf_act_put_cookie(struct flow_action_entry *entry)
3468{
3469 flow_action_cookie_destroy(entry->cookie);
3470}
3471
3472void tc_cleanup_flow_action(struct flow_action *flow_action)
3473{
3474 struct flow_action_entry *entry;
3475 int i;
3476
3477 flow_action_for_each(i, entry, flow_action) {
3478 tcf_act_put_cookie(entry);
3479 if (entry->destructor)
3480 entry->destructor(entry->destructor_priv);
3481 }
3482}
3483EXPORT_SYMBOL(tc_cleanup_flow_action);
3484
3485static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3486 const struct tc_action *act)
3487{
3488#ifdef CONFIG_NET_CLS_ACT
3489 entry->dev = act->ops->get_dev(act, &entry->destructor);
3490 if (!entry->dev)
3491 return;
3492 entry->destructor_priv = entry->dev;
3493#endif
3494}
3495
3496static void tcf_tunnel_encap_put_tunnel(void *priv)
3497{
3498 struct ip_tunnel_info *tunnel = priv;
3499
3500 kfree(tunnel);
3501}
3502
3503static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3504 const struct tc_action *act)
3505{
3506 entry->tunnel = tcf_tunnel_info_copy(act);
3507 if (!entry->tunnel)
3508 return -ENOMEM;
3509 entry->destructor = tcf_tunnel_encap_put_tunnel;
3510 entry->destructor_priv = entry->tunnel;
3511 return 0;
3512}
3513
3514static void tcf_sample_get_group(struct flow_action_entry *entry,
3515 const struct tc_action *act)
3516{
3517#ifdef CONFIG_NET_CLS_ACT
3518 entry->sample.psample_group =
3519 act->ops->get_psample_group(act, &entry->destructor);
3520 entry->destructor_priv = entry->sample.psample_group;
3521#endif
3522}
3523
3524static void tcf_gate_entry_destructor(void *priv)
3525{
3526 struct action_gate_entry *oe = priv;
3527
3528 kfree(oe);
3529}
3530
3531static int tcf_gate_get_entries(struct flow_action_entry *entry,
3532 const struct tc_action *act)
3533{
3534 entry->gate.entries = tcf_gate_get_list(act);
3535
3536 if (!entry->gate.entries)
3537 return -EINVAL;
3538
3539 entry->destructor = tcf_gate_entry_destructor;
3540 entry->destructor_priv = entry->gate.entries;
3541
3542 return 0;
3543}
3544
3545static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats)
3546{
3547 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY))
3548 return FLOW_ACTION_HW_STATS_DONT_CARE;
3549 else if (!hw_stats)
3550 return FLOW_ACTION_HW_STATS_DISABLED;
3551
3552 return hw_stats;
3553}
3554
3555int tc_setup_flow_action(struct flow_action *flow_action,
3556 const struct tcf_exts *exts)
3557{
3558 struct tc_action *act;
3559 int i, j, k, err = 0;
3560
3561 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3562 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3563 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3564
3565 if (!exts)
3566 return 0;
3567
3568 j = 0;
3569 tcf_exts_for_each_action(i, act, exts) {
3570 struct flow_action_entry *entry;
3571
3572 entry = &flow_action->entries[j];
3573 spin_lock_bh(&act->tcfa_lock);
3574 err = tcf_act_get_cookie(entry, act);
3575 if (err)
3576 goto err_out_locked;
3577
3578 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3579
3580 if (is_tcf_gact_ok(act)) {
3581 entry->id = FLOW_ACTION_ACCEPT;
3582 } else if (is_tcf_gact_shot(act)) {
3583 entry->id = FLOW_ACTION_DROP;
3584 } else if (is_tcf_gact_trap(act)) {
3585 entry->id = FLOW_ACTION_TRAP;
3586 } else if (is_tcf_gact_goto_chain(act)) {
3587 entry->id = FLOW_ACTION_GOTO;
3588 entry->chain_index = tcf_gact_goto_chain_index(act);
3589 } else if (is_tcf_mirred_egress_redirect(act)) {
3590 entry->id = FLOW_ACTION_REDIRECT;
3591 tcf_mirred_get_dev(entry, act);
3592 } else if (is_tcf_mirred_egress_mirror(act)) {
3593 entry->id = FLOW_ACTION_MIRRED;
3594 tcf_mirred_get_dev(entry, act);
3595 } else if (is_tcf_mirred_ingress_redirect(act)) {
3596 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3597 tcf_mirred_get_dev(entry, act);
3598 } else if (is_tcf_mirred_ingress_mirror(act)) {
3599 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3600 tcf_mirred_get_dev(entry, act);
3601 } else if (is_tcf_vlan(act)) {
3602 switch (tcf_vlan_action(act)) {
3603 case TCA_VLAN_ACT_PUSH:
3604 entry->id = FLOW_ACTION_VLAN_PUSH;
3605 entry->vlan.vid = tcf_vlan_push_vid(act);
3606 entry->vlan.proto = tcf_vlan_push_proto(act);
3607 entry->vlan.prio = tcf_vlan_push_prio(act);
3608 break;
3609 case TCA_VLAN_ACT_POP:
3610 entry->id = FLOW_ACTION_VLAN_POP;
3611 break;
3612 case TCA_VLAN_ACT_MODIFY:
3613 entry->id = FLOW_ACTION_VLAN_MANGLE;
3614 entry->vlan.vid = tcf_vlan_push_vid(act);
3615 entry->vlan.proto = tcf_vlan_push_proto(act);
3616 entry->vlan.prio = tcf_vlan_push_prio(act);
3617 break;
3618 default:
3619 err = -EOPNOTSUPP;
3620 goto err_out_locked;
3621 }
3622 } else if (is_tcf_tunnel_set(act)) {
3623 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3624 err = tcf_tunnel_encap_get_tunnel(entry, act);
3625 if (err)
3626 goto err_out_locked;
3627 } else if (is_tcf_tunnel_release(act)) {
3628 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3629 } else if (is_tcf_pedit(act)) {
3630 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3631 switch (tcf_pedit_cmd(act, k)) {
3632 case TCA_PEDIT_KEY_EX_CMD_SET:
3633 entry->id = FLOW_ACTION_MANGLE;
3634 break;
3635 case TCA_PEDIT_KEY_EX_CMD_ADD:
3636 entry->id = FLOW_ACTION_ADD;
3637 break;
3638 default:
3639 err = -EOPNOTSUPP;
3640 goto err_out_locked;
3641 }
3642 entry->mangle.htype = tcf_pedit_htype(act, k);
3643 entry->mangle.mask = tcf_pedit_mask(act, k);
3644 entry->mangle.val = tcf_pedit_val(act, k);
3645 entry->mangle.offset = tcf_pedit_offset(act, k);
3646 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
3647 entry = &flow_action->entries[++j];
3648 }
3649 } else if (is_tcf_csum(act)) {
3650 entry->id = FLOW_ACTION_CSUM;
3651 entry->csum_flags = tcf_csum_update_flags(act);
3652 } else if (is_tcf_skbedit_mark(act)) {
3653 entry->id = FLOW_ACTION_MARK;
3654 entry->mark = tcf_skbedit_mark(act);
3655 } else if (is_tcf_sample(act)) {
3656 entry->id = FLOW_ACTION_SAMPLE;
3657 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3658 entry->sample.truncate = tcf_sample_truncate(act);
3659 entry->sample.rate = tcf_sample_rate(act);
3660 tcf_sample_get_group(entry, act);
3661 } else if (is_tcf_police(act)) {
3662 entry->id = FLOW_ACTION_POLICE;
3663 entry->police.burst = tcf_police_burst(act);
3664 entry->police.rate_bytes_ps =
3665 tcf_police_rate_bytes_ps(act);
3666 entry->police.burst_pkt = tcf_police_burst_pkt(act);
3667 entry->police.rate_pkt_ps =
3668 tcf_police_rate_pkt_ps(act);
3669 entry->police.mtu = tcf_police_tcfp_mtu(act);
3670 entry->police.index = act->tcfa_index;
3671 } else if (is_tcf_ct(act)) {
3672 entry->id = FLOW_ACTION_CT;
3673 entry->ct.action = tcf_ct_action(act);
3674 entry->ct.zone = tcf_ct_zone(act);
3675 entry->ct.flow_table = tcf_ct_ft(act);
3676 } else if (is_tcf_mpls(act)) {
3677 switch (tcf_mpls_action(act)) {
3678 case TCA_MPLS_ACT_PUSH:
3679 entry->id = FLOW_ACTION_MPLS_PUSH;
3680 entry->mpls_push.proto = tcf_mpls_proto(act);
3681 entry->mpls_push.label = tcf_mpls_label(act);
3682 entry->mpls_push.tc = tcf_mpls_tc(act);
3683 entry->mpls_push.bos = tcf_mpls_bos(act);
3684 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3685 break;
3686 case TCA_MPLS_ACT_POP:
3687 entry->id = FLOW_ACTION_MPLS_POP;
3688 entry->mpls_pop.proto = tcf_mpls_proto(act);
3689 break;
3690 case TCA_MPLS_ACT_MODIFY:
3691 entry->id = FLOW_ACTION_MPLS_MANGLE;
3692 entry->mpls_mangle.label = tcf_mpls_label(act);
3693 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3694 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3695 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3696 break;
3697 default:
3698 goto err_out_locked;
3699 }
3700 } else if (is_tcf_skbedit_ptype(act)) {
3701 entry->id = FLOW_ACTION_PTYPE;
3702 entry->ptype = tcf_skbedit_ptype(act);
3703 } else if (is_tcf_skbedit_priority(act)) {
3704 entry->id = FLOW_ACTION_PRIORITY;
3705 entry->priority = tcf_skbedit_priority(act);
3706 } else if (is_tcf_gate(act)) {
3707 entry->id = FLOW_ACTION_GATE;
3708 entry->gate.index = tcf_gate_index(act);
3709 entry->gate.prio = tcf_gate_prio(act);
3710 entry->gate.basetime = tcf_gate_basetime(act);
3711 entry->gate.cycletime = tcf_gate_cycletime(act);
3712 entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
3713 entry->gate.num_entries = tcf_gate_num_entries(act);
3714 err = tcf_gate_get_entries(entry, act);
3715 if (err)
3716 goto err_out_locked;
3717 } else {
3718 err = -EOPNOTSUPP;
3719 goto err_out_locked;
3720 }
3721 spin_unlock_bh(&act->tcfa_lock);
3722
3723 if (!is_tcf_pedit(act))
3724 j++;
3725 }
3726
3727err_out:
3728 if (err)
3729 tc_cleanup_flow_action(flow_action);
3730
3731 return err;
3732err_out_locked:
3733 spin_unlock_bh(&act->tcfa_lock);
3734 goto err_out;
3735}
3736EXPORT_SYMBOL(tc_setup_flow_action);
3737
3738unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3739{
3740 unsigned int num_acts = 0;
3741 struct tc_action *act;
3742 int i;
3743
3744 tcf_exts_for_each_action(i, act, exts) {
3745 if (is_tcf_pedit(act))
3746 num_acts += tcf_pedit_nkeys(act);
3747 else
3748 num_acts++;
3749 }
3750 return num_acts;
3751}
3752EXPORT_SYMBOL(tcf_exts_num_actions);
3753
3754#ifdef CONFIG_NET_CLS_ACT
3755static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3756 u32 *p_block_index,
3757 struct netlink_ext_ack *extack)
3758{
3759 *p_block_index = nla_get_u32(block_index_attr);
3760 if (!*p_block_index) {
3761 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3762 return -EINVAL;
3763 }
3764
3765 return 0;
3766}
3767
3768int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3769 enum flow_block_binder_type binder_type,
3770 struct nlattr *block_index_attr,
3771 struct netlink_ext_ack *extack)
3772{
3773 u32 block_index;
3774 int err;
3775
3776 if (!block_index_attr)
3777 return 0;
3778
3779 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3780 if (err)
3781 return err;
3782
3783 if (!block_index)
3784 return 0;
3785
3786 qe->info.binder_type = binder_type;
3787 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3788 qe->info.chain_head_change_priv = &qe->filter_chain;
3789 qe->info.block_index = block_index;
3790
3791 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3792}
3793EXPORT_SYMBOL(tcf_qevent_init);
3794
3795void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3796{
3797 if (qe->info.block_index)
3798 tcf_block_put_ext(qe->block, sch, &qe->info);
3799}
3800EXPORT_SYMBOL(tcf_qevent_destroy);
3801
3802int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3803 struct netlink_ext_ack *extack)
3804{
3805 u32 block_index;
3806 int err;
3807
3808 if (!block_index_attr)
3809 return 0;
3810
3811 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3812 if (err)
3813 return err;
3814
3815 /* Bounce newly-configured block or change in block. */
3816 if (block_index != qe->info.block_index) {
3817 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3818 return -EINVAL;
3819 }
3820
3821 return 0;
3822}
3823EXPORT_SYMBOL(tcf_qevent_validate_change);
3824
3825struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3826 struct sk_buff **to_free, int *ret)
3827{
3828 struct tcf_result cl_res;
3829 struct tcf_proto *fl;
3830
3831 if (!qe->info.block_index)
3832 return skb;
3833
3834 fl = rcu_dereference_bh(qe->filter_chain);
3835
3836 switch (tcf_classify(skb, fl, &cl_res, false)) {
3837 case TC_ACT_SHOT:
3838 qdisc_qstats_drop(sch);
3839 __qdisc_drop(skb, to_free);
3840 *ret = __NET_XMIT_BYPASS;
3841 return NULL;
3842 case TC_ACT_STOLEN:
3843 case TC_ACT_QUEUED:
3844 case TC_ACT_TRAP:
3845 __qdisc_drop(skb, to_free);
3846 *ret = __NET_XMIT_STOLEN;
3847 return NULL;
3848 case TC_ACT_REDIRECT:
3849 skb_do_redirect(skb);
3850 *ret = __NET_XMIT_STOLEN;
3851 return NULL;
3852 }
3853
3854 return skb;
3855}
3856EXPORT_SYMBOL(tcf_qevent_handle);
3857
3858int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3859{
3860 if (!qe->info.block_index)
3861 return 0;
3862 return nla_put_u32(skb, attr_name, qe->info.block_index);
3863}
3864EXPORT_SYMBOL(tcf_qevent_dump);
3865#endif
3866
3867static __net_init int tcf_net_init(struct net *net)
3868{
3869 struct tcf_net *tn = net_generic(net, tcf_net_id);
3870
3871 spin_lock_init(&tn->idr_lock);
3872 idr_init(&tn->idr);
3873 return 0;
3874}
3875
3876static void __net_exit tcf_net_exit(struct net *net)
3877{
3878 struct tcf_net *tn = net_generic(net, tcf_net_id);
3879
3880 idr_destroy(&tn->idr);
3881}
3882
3883static struct pernet_operations tcf_net_ops = {
3884 .init = tcf_net_init,
3885 .exit = tcf_net_exit,
3886 .id = &tcf_net_id,
3887 .size = sizeof(struct tcf_net),
3888};
3889
3890static int __init tc_filter_init(void)
3891{
3892 int err;
3893
3894 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3895 if (!tc_filter_wq)
3896 return -ENOMEM;
3897
3898 err = register_pernet_subsys(&tcf_net_ops);
3899 if (err)
3900 goto err_register_pernet_subsys;
3901
3902 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3903 RTNL_FLAG_DOIT_UNLOCKED);
3904 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3905 RTNL_FLAG_DOIT_UNLOCKED);
3906 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3907 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3908 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3909 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3910 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3911 tc_dump_chain, 0);
3912
3913 return 0;
3914
3915err_register_pernet_subsys:
3916 destroy_workqueue(tc_filter_wq);
3917 return err;
3918}
3919
3920subsys_initcall(tc_filter_init);