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/rhashtable.h>
24#include <linux/jhash.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/flow_offload.h>
42
43extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
44
45/* The list of all installed classifier types */
46static LIST_HEAD(tcf_proto_base);
47
48/* Protects list of registered TC modules. It is pure SMP lock. */
49static DEFINE_RWLOCK(cls_mod_lock);
50
51static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
52{
53 return jhash_3words(tp->chain->index, tp->prio,
54 (__force __u32)tp->protocol, 0);
55}
56
57static void tcf_proto_signal_destroying(struct tcf_chain *chain,
58 struct tcf_proto *tp)
59{
60 struct tcf_block *block = chain->block;
61
62 mutex_lock(&block->proto_destroy_lock);
63 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
64 destroy_obj_hashfn(tp));
65 mutex_unlock(&block->proto_destroy_lock);
66}
67
68static bool tcf_proto_cmp(const struct tcf_proto *tp1,
69 const struct tcf_proto *tp2)
70{
71 return tp1->chain->index == tp2->chain->index &&
72 tp1->prio == tp2->prio &&
73 tp1->protocol == tp2->protocol;
74}
75
76static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
77 struct tcf_proto *tp)
78{
79 u32 hash = destroy_obj_hashfn(tp);
80 struct tcf_proto *iter;
81 bool found = false;
82
83 rcu_read_lock();
84 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
85 destroy_ht_node, hash) {
86 if (tcf_proto_cmp(tp, iter)) {
87 found = true;
88 break;
89 }
90 }
91 rcu_read_unlock();
92
93 return found;
94}
95
96static void
97tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
98{
99 struct tcf_block *block = chain->block;
100
101 mutex_lock(&block->proto_destroy_lock);
102 if (hash_hashed(&tp->destroy_ht_node))
103 hash_del_rcu(&tp->destroy_ht_node);
104 mutex_unlock(&block->proto_destroy_lock);
105}
106
107/* Find classifier type by string name */
108
109static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
110{
111 const struct tcf_proto_ops *t, *res = NULL;
112
113 if (kind) {
114 read_lock(&cls_mod_lock);
115 list_for_each_entry(t, &tcf_proto_base, head) {
116 if (strcmp(kind, t->kind) == 0) {
117 if (try_module_get(t->owner))
118 res = t;
119 break;
120 }
121 }
122 read_unlock(&cls_mod_lock);
123 }
124 return res;
125}
126
127static const struct tcf_proto_ops *
128tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
129 struct netlink_ext_ack *extack)
130{
131 const struct tcf_proto_ops *ops;
132
133 ops = __tcf_proto_lookup_ops(kind);
134 if (ops)
135 return ops;
136#ifdef CONFIG_MODULES
137 if (rtnl_held)
138 rtnl_unlock();
139 request_module("cls_%s", kind);
140 if (rtnl_held)
141 rtnl_lock();
142 ops = __tcf_proto_lookup_ops(kind);
143 /* We dropped the RTNL semaphore in order to perform
144 * the module load. So, even if we succeeded in loading
145 * the module we have to replay the request. We indicate
146 * this using -EAGAIN.
147 */
148 if (ops) {
149 module_put(ops->owner);
150 return ERR_PTR(-EAGAIN);
151 }
152#endif
153 NL_SET_ERR_MSG(extack, "TC classifier not found");
154 return ERR_PTR(-ENOENT);
155}
156
157/* Register(unregister) new classifier type */
158
159int register_tcf_proto_ops(struct tcf_proto_ops *ops)
160{
161 struct tcf_proto_ops *t;
162 int rc = -EEXIST;
163
164 write_lock(&cls_mod_lock);
165 list_for_each_entry(t, &tcf_proto_base, head)
166 if (!strcmp(ops->kind, t->kind))
167 goto out;
168
169 list_add_tail(&ops->head, &tcf_proto_base);
170 rc = 0;
171out:
172 write_unlock(&cls_mod_lock);
173 return rc;
174}
175EXPORT_SYMBOL(register_tcf_proto_ops);
176
177static struct workqueue_struct *tc_filter_wq;
178
179int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
180{
181 struct tcf_proto_ops *t;
182 int rc = -ENOENT;
183
184 /* Wait for outstanding call_rcu()s, if any, from a
185 * tcf_proto_ops's destroy() handler.
186 */
187 rcu_barrier();
188 flush_workqueue(tc_filter_wq);
189
190 write_lock(&cls_mod_lock);
191 list_for_each_entry(t, &tcf_proto_base, head) {
192 if (t == ops) {
193 list_del(&t->head);
194 rc = 0;
195 break;
196 }
197 }
198 write_unlock(&cls_mod_lock);
199 return rc;
200}
201EXPORT_SYMBOL(unregister_tcf_proto_ops);
202
203bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
204{
205 INIT_RCU_WORK(rwork, func);
206 return queue_rcu_work(tc_filter_wq, rwork);
207}
208EXPORT_SYMBOL(tcf_queue_work);
209
210/* Select new prio value from the range, managed by kernel. */
211
212static inline u32 tcf_auto_prio(struct tcf_proto *tp)
213{
214 u32 first = TC_H_MAKE(0xC0000000U, 0U);
215
216 if (tp)
217 first = tp->prio - 1;
218
219 return TC_H_MAJ(first);
220}
221
222static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
223{
224 if (kind)
225 return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
226 memset(name, 0, IFNAMSIZ);
227 return false;
228}
229
230static bool tcf_proto_is_unlocked(const char *kind)
231{
232 const struct tcf_proto_ops *ops;
233 bool ret;
234
235 if (strlen(kind) == 0)
236 return false;
237
238 ops = tcf_proto_lookup_ops(kind, false, NULL);
239 /* On error return false to take rtnl lock. Proto lookup/create
240 * functions will perform lookup again and properly handle errors.
241 */
242 if (IS_ERR(ops))
243 return false;
244
245 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
246 module_put(ops->owner);
247 return ret;
248}
249
250static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
251 u32 prio, struct tcf_chain *chain,
252 bool rtnl_held,
253 struct netlink_ext_ack *extack)
254{
255 struct tcf_proto *tp;
256 int err;
257
258 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
259 if (!tp)
260 return ERR_PTR(-ENOBUFS);
261
262 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
263 if (IS_ERR(tp->ops)) {
264 err = PTR_ERR(tp->ops);
265 goto errout;
266 }
267 tp->classify = tp->ops->classify;
268 tp->protocol = protocol;
269 tp->prio = prio;
270 tp->chain = chain;
271 spin_lock_init(&tp->lock);
272 refcount_set(&tp->refcnt, 1);
273
274 err = tp->ops->init(tp);
275 if (err) {
276 module_put(tp->ops->owner);
277 goto errout;
278 }
279 return tp;
280
281errout:
282 kfree(tp);
283 return ERR_PTR(err);
284}
285
286static void tcf_proto_get(struct tcf_proto *tp)
287{
288 refcount_inc(&tp->refcnt);
289}
290
291static void tcf_chain_put(struct tcf_chain *chain);
292
293static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
294 bool sig_destroy, struct netlink_ext_ack *extack)
295{
296 tp->ops->destroy(tp, rtnl_held, extack);
297 if (sig_destroy)
298 tcf_proto_signal_destroyed(tp->chain, tp);
299 tcf_chain_put(tp->chain);
300 module_put(tp->ops->owner);
301 kfree_rcu(tp, rcu);
302}
303
304static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
305 struct netlink_ext_ack *extack)
306{
307 if (refcount_dec_and_test(&tp->refcnt))
308 tcf_proto_destroy(tp, rtnl_held, true, extack);
309}
310
311static int walker_check_empty(struct tcf_proto *tp, void *fh,
312 struct tcf_walker *arg)
313{
314 if (fh) {
315 arg->nonempty = true;
316 return -1;
317 }
318 return 0;
319}
320
321static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
322{
323 struct tcf_walker walker = { .fn = walker_check_empty, };
324
325 if (tp->ops->walk) {
326 tp->ops->walk(tp, &walker, rtnl_held);
327 return !walker.nonempty;
328 }
329 return true;
330}
331
332static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
333{
334 spin_lock(&tp->lock);
335 if (tcf_proto_is_empty(tp, rtnl_held))
336 tp->deleting = true;
337 spin_unlock(&tp->lock);
338 return tp->deleting;
339}
340
341static void tcf_proto_mark_delete(struct tcf_proto *tp)
342{
343 spin_lock(&tp->lock);
344 tp->deleting = true;
345 spin_unlock(&tp->lock);
346}
347
348static bool tcf_proto_is_deleting(struct tcf_proto *tp)
349{
350 bool deleting;
351
352 spin_lock(&tp->lock);
353 deleting = tp->deleting;
354 spin_unlock(&tp->lock);
355
356 return deleting;
357}
358
359#define ASSERT_BLOCK_LOCKED(block) \
360 lockdep_assert_held(&(block)->lock)
361
362struct tcf_filter_chain_list_item {
363 struct list_head list;
364 tcf_chain_head_change_t *chain_head_change;
365 void *chain_head_change_priv;
366};
367
368static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
369 u32 chain_index)
370{
371 struct tcf_chain *chain;
372
373 ASSERT_BLOCK_LOCKED(block);
374
375 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
376 if (!chain)
377 return NULL;
378 list_add_tail(&chain->list, &block->chain_list);
379 mutex_init(&chain->filter_chain_lock);
380 chain->block = block;
381 chain->index = chain_index;
382 chain->refcnt = 1;
383 if (!chain->index)
384 block->chain0.chain = chain;
385 return chain;
386}
387
388static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
389 struct tcf_proto *tp_head)
390{
391 if (item->chain_head_change)
392 item->chain_head_change(tp_head, item->chain_head_change_priv);
393}
394
395static void tcf_chain0_head_change(struct tcf_chain *chain,
396 struct tcf_proto *tp_head)
397{
398 struct tcf_filter_chain_list_item *item;
399 struct tcf_block *block = chain->block;
400
401 if (chain->index)
402 return;
403
404 mutex_lock(&block->lock);
405 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
406 tcf_chain_head_change_item(item, tp_head);
407 mutex_unlock(&block->lock);
408}
409
410/* Returns true if block can be safely freed. */
411
412static bool tcf_chain_detach(struct tcf_chain *chain)
413{
414 struct tcf_block *block = chain->block;
415
416 ASSERT_BLOCK_LOCKED(block);
417
418 list_del(&chain->list);
419 if (!chain->index)
420 block->chain0.chain = NULL;
421
422 if (list_empty(&block->chain_list) &&
423 refcount_read(&block->refcnt) == 0)
424 return true;
425
426 return false;
427}
428
429static void tcf_block_destroy(struct tcf_block *block)
430{
431 mutex_destroy(&block->lock);
432 mutex_destroy(&block->proto_destroy_lock);
433 kfree_rcu(block, rcu);
434}
435
436static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
437{
438 struct tcf_block *block = chain->block;
439
440 mutex_destroy(&chain->filter_chain_lock);
441 kfree_rcu(chain, rcu);
442 if (free_block)
443 tcf_block_destroy(block);
444}
445
446static void tcf_chain_hold(struct tcf_chain *chain)
447{
448 ASSERT_BLOCK_LOCKED(chain->block);
449
450 ++chain->refcnt;
451}
452
453static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
454{
455 ASSERT_BLOCK_LOCKED(chain->block);
456
457 /* In case all the references are action references, this
458 * chain should not be shown to the user.
459 */
460 return chain->refcnt == chain->action_refcnt;
461}
462
463static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
464 u32 chain_index)
465{
466 struct tcf_chain *chain;
467
468 ASSERT_BLOCK_LOCKED(block);
469
470 list_for_each_entry(chain, &block->chain_list, list) {
471 if (chain->index == chain_index)
472 return chain;
473 }
474 return NULL;
475}
476
477static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
478 u32 seq, u16 flags, int event, bool unicast);
479
480static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
481 u32 chain_index, bool create,
482 bool by_act)
483{
484 struct tcf_chain *chain = NULL;
485 bool is_first_reference;
486
487 mutex_lock(&block->lock);
488 chain = tcf_chain_lookup(block, chain_index);
489 if (chain) {
490 tcf_chain_hold(chain);
491 } else {
492 if (!create)
493 goto errout;
494 chain = tcf_chain_create(block, chain_index);
495 if (!chain)
496 goto errout;
497 }
498
499 if (by_act)
500 ++chain->action_refcnt;
501 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
502 mutex_unlock(&block->lock);
503
504 /* Send notification only in case we got the first
505 * non-action reference. Until then, the chain acts only as
506 * a placeholder for actions pointing to it and user ought
507 * not know about them.
508 */
509 if (is_first_reference && !by_act)
510 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
511 RTM_NEWCHAIN, false);
512
513 return chain;
514
515errout:
516 mutex_unlock(&block->lock);
517 return chain;
518}
519
520static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
521 bool create)
522{
523 return __tcf_chain_get(block, chain_index, create, false);
524}
525
526struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
527{
528 return __tcf_chain_get(block, chain_index, true, true);
529}
530EXPORT_SYMBOL(tcf_chain_get_by_act);
531
532static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
533 void *tmplt_priv);
534static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
535 void *tmplt_priv, u32 chain_index,
536 struct tcf_block *block, struct sk_buff *oskb,
537 u32 seq, u16 flags, bool unicast);
538
539static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
540 bool explicitly_created)
541{
542 struct tcf_block *block = chain->block;
543 const struct tcf_proto_ops *tmplt_ops;
544 bool free_block = false;
545 unsigned int refcnt;
546 void *tmplt_priv;
547
548 mutex_lock(&block->lock);
549 if (explicitly_created) {
550 if (!chain->explicitly_created) {
551 mutex_unlock(&block->lock);
552 return;
553 }
554 chain->explicitly_created = false;
555 }
556
557 if (by_act)
558 chain->action_refcnt--;
559
560 /* tc_chain_notify_delete can't be called while holding block lock.
561 * However, when block is unlocked chain can be changed concurrently, so
562 * save these to temporary variables.
563 */
564 refcnt = --chain->refcnt;
565 tmplt_ops = chain->tmplt_ops;
566 tmplt_priv = chain->tmplt_priv;
567
568 /* The last dropped non-action reference will trigger notification. */
569 if (refcnt - chain->action_refcnt == 0 && !by_act) {
570 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
571 block, NULL, 0, 0, false);
572 /* Last reference to chain, no need to lock. */
573 chain->flushing = false;
574 }
575
576 if (refcnt == 0)
577 free_block = tcf_chain_detach(chain);
578 mutex_unlock(&block->lock);
579
580 if (refcnt == 0) {
581 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
582 tcf_chain_destroy(chain, free_block);
583 }
584}
585
586static void tcf_chain_put(struct tcf_chain *chain)
587{
588 __tcf_chain_put(chain, false, false);
589}
590
591void tcf_chain_put_by_act(struct tcf_chain *chain)
592{
593 __tcf_chain_put(chain, true, false);
594}
595EXPORT_SYMBOL(tcf_chain_put_by_act);
596
597static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
598{
599 __tcf_chain_put(chain, false, true);
600}
601
602static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
603{
604 struct tcf_proto *tp, *tp_next;
605
606 mutex_lock(&chain->filter_chain_lock);
607 tp = tcf_chain_dereference(chain->filter_chain, chain);
608 while (tp) {
609 tp_next = rcu_dereference_protected(tp->next, 1);
610 tcf_proto_signal_destroying(chain, tp);
611 tp = tp_next;
612 }
613 tp = tcf_chain_dereference(chain->filter_chain, chain);
614 RCU_INIT_POINTER(chain->filter_chain, NULL);
615 tcf_chain0_head_change(chain, NULL);
616 chain->flushing = true;
617 mutex_unlock(&chain->filter_chain_lock);
618
619 while (tp) {
620 tp_next = rcu_dereference_protected(tp->next, 1);
621 tcf_proto_put(tp, rtnl_held, NULL);
622 tp = tp_next;
623 }
624}
625
626static int tcf_block_setup(struct tcf_block *block,
627 struct flow_block_offload *bo);
628
629static void tc_indr_block_ing_cmd(struct net_device *dev,
630 struct tcf_block *block,
631 flow_indr_block_bind_cb_t *cb,
632 void *cb_priv,
633 enum flow_block_command command)
634{
635 struct flow_block_offload bo = {
636 .command = command,
637 .binder_type = FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
638 .net = dev_net(dev),
639 .block_shared = tcf_block_non_null_shared(block),
640 };
641 INIT_LIST_HEAD(&bo.cb_list);
642
643 if (!block)
644 return;
645
646 bo.block = &block->flow_block;
647
648 down_write(&block->cb_lock);
649 cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
650
651 tcf_block_setup(block, &bo);
652 up_write(&block->cb_lock);
653}
654
655static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
656{
657 const struct Qdisc_class_ops *cops;
658 struct Qdisc *qdisc;
659
660 if (!dev_ingress_queue(dev))
661 return NULL;
662
663 qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
664 if (!qdisc)
665 return NULL;
666
667 cops = qdisc->ops->cl_ops;
668 if (!cops)
669 return NULL;
670
671 if (!cops->tcf_block)
672 return NULL;
673
674 return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
675}
676
677static void tc_indr_block_get_and_ing_cmd(struct net_device *dev,
678 flow_indr_block_bind_cb_t *cb,
679 void *cb_priv,
680 enum flow_block_command command)
681{
682 struct tcf_block *block = tc_dev_ingress_block(dev);
683
684 tc_indr_block_ing_cmd(dev, block, cb, cb_priv, command);
685}
686
687static void tc_indr_block_call(struct tcf_block *block,
688 struct net_device *dev,
689 struct tcf_block_ext_info *ei,
690 enum flow_block_command command,
691 struct netlink_ext_ack *extack)
692{
693 struct flow_block_offload bo = {
694 .command = command,
695 .binder_type = ei->binder_type,
696 .net = dev_net(dev),
697 .block = &block->flow_block,
698 .block_shared = tcf_block_shared(block),
699 .extack = extack,
700 };
701 INIT_LIST_HEAD(&bo.cb_list);
702
703 flow_indr_block_call(dev, &bo, command);
704 tcf_block_setup(block, &bo);
705}
706
707static bool tcf_block_offload_in_use(struct tcf_block *block)
708{
709 return atomic_read(&block->offloadcnt);
710}
711
712static int tcf_block_offload_cmd(struct tcf_block *block,
713 struct net_device *dev,
714 struct tcf_block_ext_info *ei,
715 enum flow_block_command command,
716 struct netlink_ext_ack *extack)
717{
718 struct flow_block_offload bo = {};
719 int err;
720
721 bo.net = dev_net(dev);
722 bo.command = command;
723 bo.binder_type = ei->binder_type;
724 bo.block = &block->flow_block;
725 bo.block_shared = tcf_block_shared(block);
726 bo.extack = extack;
727 INIT_LIST_HEAD(&bo.cb_list);
728
729 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
730 if (err < 0)
731 return err;
732
733 return tcf_block_setup(block, &bo);
734}
735
736static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
737 struct tcf_block_ext_info *ei,
738 struct netlink_ext_ack *extack)
739{
740 struct net_device *dev = q->dev_queue->dev;
741 int err;
742
743 down_write(&block->cb_lock);
744 if (!dev->netdev_ops->ndo_setup_tc)
745 goto no_offload_dev_inc;
746
747 /* If tc offload feature is disabled and the block we try to bind
748 * to already has some offloaded filters, forbid to bind.
749 */
750 if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
751 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
752 err = -EOPNOTSUPP;
753 goto err_unlock;
754 }
755
756 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
757 if (err == -EOPNOTSUPP)
758 goto no_offload_dev_inc;
759 if (err)
760 goto err_unlock;
761
762 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
763 up_write(&block->cb_lock);
764 return 0;
765
766no_offload_dev_inc:
767 if (tcf_block_offload_in_use(block)) {
768 err = -EOPNOTSUPP;
769 goto err_unlock;
770 }
771 err = 0;
772 block->nooffloaddevcnt++;
773 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
774err_unlock:
775 up_write(&block->cb_lock);
776 return err;
777}
778
779static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
780 struct tcf_block_ext_info *ei)
781{
782 struct net_device *dev = q->dev_queue->dev;
783 int err;
784
785 down_write(&block->cb_lock);
786 tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
787
788 if (!dev->netdev_ops->ndo_setup_tc)
789 goto no_offload_dev_dec;
790 err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
791 if (err == -EOPNOTSUPP)
792 goto no_offload_dev_dec;
793 up_write(&block->cb_lock);
794 return;
795
796no_offload_dev_dec:
797 WARN_ON(block->nooffloaddevcnt-- == 0);
798 up_write(&block->cb_lock);
799}
800
801static int
802tcf_chain0_head_change_cb_add(struct tcf_block *block,
803 struct tcf_block_ext_info *ei,
804 struct netlink_ext_ack *extack)
805{
806 struct tcf_filter_chain_list_item *item;
807 struct tcf_chain *chain0;
808
809 item = kmalloc(sizeof(*item), GFP_KERNEL);
810 if (!item) {
811 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
812 return -ENOMEM;
813 }
814 item->chain_head_change = ei->chain_head_change;
815 item->chain_head_change_priv = ei->chain_head_change_priv;
816
817 mutex_lock(&block->lock);
818 chain0 = block->chain0.chain;
819 if (chain0)
820 tcf_chain_hold(chain0);
821 else
822 list_add(&item->list, &block->chain0.filter_chain_list);
823 mutex_unlock(&block->lock);
824
825 if (chain0) {
826 struct tcf_proto *tp_head;
827
828 mutex_lock(&chain0->filter_chain_lock);
829
830 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
831 if (tp_head)
832 tcf_chain_head_change_item(item, tp_head);
833
834 mutex_lock(&block->lock);
835 list_add(&item->list, &block->chain0.filter_chain_list);
836 mutex_unlock(&block->lock);
837
838 mutex_unlock(&chain0->filter_chain_lock);
839 tcf_chain_put(chain0);
840 }
841
842 return 0;
843}
844
845static void
846tcf_chain0_head_change_cb_del(struct tcf_block *block,
847 struct tcf_block_ext_info *ei)
848{
849 struct tcf_filter_chain_list_item *item;
850
851 mutex_lock(&block->lock);
852 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
853 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
854 (item->chain_head_change == ei->chain_head_change &&
855 item->chain_head_change_priv == ei->chain_head_change_priv)) {
856 if (block->chain0.chain)
857 tcf_chain_head_change_item(item, NULL);
858 list_del(&item->list);
859 mutex_unlock(&block->lock);
860
861 kfree(item);
862 return;
863 }
864 }
865 mutex_unlock(&block->lock);
866 WARN_ON(1);
867}
868
869struct tcf_net {
870 spinlock_t idr_lock; /* Protects idr */
871 struct idr idr;
872};
873
874static unsigned int tcf_net_id;
875
876static int tcf_block_insert(struct tcf_block *block, struct net *net,
877 struct netlink_ext_ack *extack)
878{
879 struct tcf_net *tn = net_generic(net, tcf_net_id);
880 int err;
881
882 idr_preload(GFP_KERNEL);
883 spin_lock(&tn->idr_lock);
884 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
885 GFP_NOWAIT);
886 spin_unlock(&tn->idr_lock);
887 idr_preload_end();
888
889 return err;
890}
891
892static void tcf_block_remove(struct tcf_block *block, struct net *net)
893{
894 struct tcf_net *tn = net_generic(net, tcf_net_id);
895
896 spin_lock(&tn->idr_lock);
897 idr_remove(&tn->idr, block->index);
898 spin_unlock(&tn->idr_lock);
899}
900
901static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
902 u32 block_index,
903 struct netlink_ext_ack *extack)
904{
905 struct tcf_block *block;
906
907 block = kzalloc(sizeof(*block), GFP_KERNEL);
908 if (!block) {
909 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
910 return ERR_PTR(-ENOMEM);
911 }
912 mutex_init(&block->lock);
913 mutex_init(&block->proto_destroy_lock);
914 init_rwsem(&block->cb_lock);
915 flow_block_init(&block->flow_block);
916 INIT_LIST_HEAD(&block->chain_list);
917 INIT_LIST_HEAD(&block->owner_list);
918 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
919
920 refcount_set(&block->refcnt, 1);
921 block->net = net;
922 block->index = block_index;
923
924 /* Don't store q pointer for blocks which are shared */
925 if (!tcf_block_shared(block))
926 block->q = q;
927 return block;
928}
929
930static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
931{
932 struct tcf_net *tn = net_generic(net, tcf_net_id);
933
934 return idr_find(&tn->idr, block_index);
935}
936
937static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
938{
939 struct tcf_block *block;
940
941 rcu_read_lock();
942 block = tcf_block_lookup(net, block_index);
943 if (block && !refcount_inc_not_zero(&block->refcnt))
944 block = NULL;
945 rcu_read_unlock();
946
947 return block;
948}
949
950static struct tcf_chain *
951__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
952{
953 mutex_lock(&block->lock);
954 if (chain)
955 chain = list_is_last(&chain->list, &block->chain_list) ?
956 NULL : list_next_entry(chain, list);
957 else
958 chain = list_first_entry_or_null(&block->chain_list,
959 struct tcf_chain, list);
960
961 /* skip all action-only chains */
962 while (chain && tcf_chain_held_by_acts_only(chain))
963 chain = list_is_last(&chain->list, &block->chain_list) ?
964 NULL : list_next_entry(chain, list);
965
966 if (chain)
967 tcf_chain_hold(chain);
968 mutex_unlock(&block->lock);
969
970 return chain;
971}
972
973/* Function to be used by all clients that want to iterate over all chains on
974 * block. It properly obtains block->lock and takes reference to chain before
975 * returning it. Users of this function must be tolerant to concurrent chain
976 * insertion/deletion or ensure that no concurrent chain modification is
977 * possible. Note that all netlink dump callbacks cannot guarantee to provide
978 * consistent dump because rtnl lock is released each time skb is filled with
979 * data and sent to user-space.
980 */
981
982struct tcf_chain *
983tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
984{
985 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
986
987 if (chain)
988 tcf_chain_put(chain);
989
990 return chain_next;
991}
992EXPORT_SYMBOL(tcf_get_next_chain);
993
994static struct tcf_proto *
995__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
996{
997 u32 prio = 0;
998
999 ASSERT_RTNL();
1000 mutex_lock(&chain->filter_chain_lock);
1001
1002 if (!tp) {
1003 tp = tcf_chain_dereference(chain->filter_chain, chain);
1004 } else if (tcf_proto_is_deleting(tp)) {
1005 /* 'deleting' flag is set and chain->filter_chain_lock was
1006 * unlocked, which means next pointer could be invalid. Restart
1007 * search.
1008 */
1009 prio = tp->prio + 1;
1010 tp = tcf_chain_dereference(chain->filter_chain, chain);
1011
1012 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1013 if (!tp->deleting && tp->prio >= prio)
1014 break;
1015 } else {
1016 tp = tcf_chain_dereference(tp->next, chain);
1017 }
1018
1019 if (tp)
1020 tcf_proto_get(tp);
1021
1022 mutex_unlock(&chain->filter_chain_lock);
1023
1024 return tp;
1025}
1026
1027/* Function to be used by all clients that want to iterate over all tp's on
1028 * chain. Users of this function must be tolerant to concurrent tp
1029 * insertion/deletion or ensure that no concurrent chain modification is
1030 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1031 * consistent dump because rtnl lock is released each time skb is filled with
1032 * data and sent to user-space.
1033 */
1034
1035struct tcf_proto *
1036tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
1037 bool rtnl_held)
1038{
1039 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1040
1041 if (tp)
1042 tcf_proto_put(tp, rtnl_held, NULL);
1043
1044 return tp_next;
1045}
1046EXPORT_SYMBOL(tcf_get_next_proto);
1047
1048static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1049{
1050 struct tcf_chain *chain;
1051
1052 /* Last reference to block. At this point chains cannot be added or
1053 * removed concurrently.
1054 */
1055 for (chain = tcf_get_next_chain(block, NULL);
1056 chain;
1057 chain = tcf_get_next_chain(block, chain)) {
1058 tcf_chain_put_explicitly_created(chain);
1059 tcf_chain_flush(chain, rtnl_held);
1060 }
1061}
1062
1063/* Lookup Qdisc and increments its reference counter.
1064 * Set parent, if necessary.
1065 */
1066
1067static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1068 u32 *parent, int ifindex, bool rtnl_held,
1069 struct netlink_ext_ack *extack)
1070{
1071 const struct Qdisc_class_ops *cops;
1072 struct net_device *dev;
1073 int err = 0;
1074
1075 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1076 return 0;
1077
1078 rcu_read_lock();
1079
1080 /* Find link */
1081 dev = dev_get_by_index_rcu(net, ifindex);
1082 if (!dev) {
1083 rcu_read_unlock();
1084 return -ENODEV;
1085 }
1086
1087 /* Find qdisc */
1088 if (!*parent) {
1089 *q = dev->qdisc;
1090 *parent = (*q)->handle;
1091 } else {
1092 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1093 if (!*q) {
1094 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1095 err = -EINVAL;
1096 goto errout_rcu;
1097 }
1098 }
1099
1100 *q = qdisc_refcount_inc_nz(*q);
1101 if (!*q) {
1102 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1103 err = -EINVAL;
1104 goto errout_rcu;
1105 }
1106
1107 /* Is it classful? */
1108 cops = (*q)->ops->cl_ops;
1109 if (!cops) {
1110 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1111 err = -EINVAL;
1112 goto errout_qdisc;
1113 }
1114
1115 if (!cops->tcf_block) {
1116 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1117 err = -EOPNOTSUPP;
1118 goto errout_qdisc;
1119 }
1120
1121errout_rcu:
1122 /* At this point we know that qdisc is not noop_qdisc,
1123 * which means that qdisc holds a reference to net_device
1124 * and we hold a reference to qdisc, so it is safe to release
1125 * rcu read lock.
1126 */
1127 rcu_read_unlock();
1128 return err;
1129
1130errout_qdisc:
1131 rcu_read_unlock();
1132
1133 if (rtnl_held)
1134 qdisc_put(*q);
1135 else
1136 qdisc_put_unlocked(*q);
1137 *q = NULL;
1138
1139 return err;
1140}
1141
1142static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1143 int ifindex, struct netlink_ext_ack *extack)
1144{
1145 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1146 return 0;
1147
1148 /* Do we search for filter, attached to class? */
1149 if (TC_H_MIN(parent)) {
1150 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1151
1152 *cl = cops->find(q, parent);
1153 if (*cl == 0) {
1154 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1155 return -ENOENT;
1156 }
1157 }
1158
1159 return 0;
1160}
1161
1162static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1163 unsigned long cl, int ifindex,
1164 u32 block_index,
1165 struct netlink_ext_ack *extack)
1166{
1167 struct tcf_block *block;
1168
1169 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1170 block = tcf_block_refcnt_get(net, block_index);
1171 if (!block) {
1172 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1173 return ERR_PTR(-EINVAL);
1174 }
1175 } else {
1176 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1177
1178 block = cops->tcf_block(q, cl, extack);
1179 if (!block)
1180 return ERR_PTR(-EINVAL);
1181
1182 if (tcf_block_shared(block)) {
1183 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1184 return ERR_PTR(-EOPNOTSUPP);
1185 }
1186
1187 /* Always take reference to block in order to support execution
1188 * of rules update path of cls API without rtnl lock. Caller
1189 * must release block when it is finished using it. 'if' block
1190 * of this conditional obtain reference to block by calling
1191 * tcf_block_refcnt_get().
1192 */
1193 refcount_inc(&block->refcnt);
1194 }
1195
1196 return block;
1197}
1198
1199static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1200 struct tcf_block_ext_info *ei, bool rtnl_held)
1201{
1202 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1203 /* Flushing/putting all chains will cause the block to be
1204 * deallocated when last chain is freed. However, if chain_list
1205 * is empty, block has to be manually deallocated. After block
1206 * reference counter reached 0, it is no longer possible to
1207 * increment it or add new chains to block.
1208 */
1209 bool free_block = list_empty(&block->chain_list);
1210
1211 mutex_unlock(&block->lock);
1212 if (tcf_block_shared(block))
1213 tcf_block_remove(block, block->net);
1214
1215 if (q)
1216 tcf_block_offload_unbind(block, q, ei);
1217
1218 if (free_block)
1219 tcf_block_destroy(block);
1220 else
1221 tcf_block_flush_all_chains(block, rtnl_held);
1222 } else if (q) {
1223 tcf_block_offload_unbind(block, q, ei);
1224 }
1225}
1226
1227static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1228{
1229 __tcf_block_put(block, NULL, NULL, rtnl_held);
1230}
1231
1232/* Find tcf block.
1233 * Set q, parent, cl when appropriate.
1234 */
1235
1236static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1237 u32 *parent, unsigned long *cl,
1238 int ifindex, u32 block_index,
1239 struct netlink_ext_ack *extack)
1240{
1241 struct tcf_block *block;
1242 int err = 0;
1243
1244 ASSERT_RTNL();
1245
1246 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1247 if (err)
1248 goto errout;
1249
1250 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1251 if (err)
1252 goto errout_qdisc;
1253
1254 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1255 if (IS_ERR(block)) {
1256 err = PTR_ERR(block);
1257 goto errout_qdisc;
1258 }
1259
1260 return block;
1261
1262errout_qdisc:
1263 if (*q)
1264 qdisc_put(*q);
1265errout:
1266 *q = NULL;
1267 return ERR_PTR(err);
1268}
1269
1270static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1271 bool rtnl_held)
1272{
1273 if (!IS_ERR_OR_NULL(block))
1274 tcf_block_refcnt_put(block, rtnl_held);
1275
1276 if (q) {
1277 if (rtnl_held)
1278 qdisc_put(q);
1279 else
1280 qdisc_put_unlocked(q);
1281 }
1282}
1283
1284struct tcf_block_owner_item {
1285 struct list_head list;
1286 struct Qdisc *q;
1287 enum flow_block_binder_type binder_type;
1288};
1289
1290static void
1291tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1292 struct Qdisc *q,
1293 enum flow_block_binder_type binder_type)
1294{
1295 if (block->keep_dst &&
1296 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1297 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1298 netif_keep_dst(qdisc_dev(q));
1299}
1300
1301void tcf_block_netif_keep_dst(struct tcf_block *block)
1302{
1303 struct tcf_block_owner_item *item;
1304
1305 block->keep_dst = true;
1306 list_for_each_entry(item, &block->owner_list, list)
1307 tcf_block_owner_netif_keep_dst(block, item->q,
1308 item->binder_type);
1309}
1310EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1311
1312static int tcf_block_owner_add(struct tcf_block *block,
1313 struct Qdisc *q,
1314 enum flow_block_binder_type binder_type)
1315{
1316 struct tcf_block_owner_item *item;
1317
1318 item = kmalloc(sizeof(*item), GFP_KERNEL);
1319 if (!item)
1320 return -ENOMEM;
1321 item->q = q;
1322 item->binder_type = binder_type;
1323 list_add(&item->list, &block->owner_list);
1324 return 0;
1325}
1326
1327static void tcf_block_owner_del(struct tcf_block *block,
1328 struct Qdisc *q,
1329 enum flow_block_binder_type binder_type)
1330{
1331 struct tcf_block_owner_item *item;
1332
1333 list_for_each_entry(item, &block->owner_list, list) {
1334 if (item->q == q && item->binder_type == binder_type) {
1335 list_del(&item->list);
1336 kfree(item);
1337 return;
1338 }
1339 }
1340 WARN_ON(1);
1341}
1342
1343int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1344 struct tcf_block_ext_info *ei,
1345 struct netlink_ext_ack *extack)
1346{
1347 struct net *net = qdisc_net(q);
1348 struct tcf_block *block = NULL;
1349 int err;
1350
1351 if (ei->block_index)
1352 /* block_index not 0 means the shared block is requested */
1353 block = tcf_block_refcnt_get(net, ei->block_index);
1354
1355 if (!block) {
1356 block = tcf_block_create(net, q, ei->block_index, extack);
1357 if (IS_ERR(block))
1358 return PTR_ERR(block);
1359 if (tcf_block_shared(block)) {
1360 err = tcf_block_insert(block, net, extack);
1361 if (err)
1362 goto err_block_insert;
1363 }
1364 }
1365
1366 err = tcf_block_owner_add(block, q, ei->binder_type);
1367 if (err)
1368 goto err_block_owner_add;
1369
1370 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1371
1372 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1373 if (err)
1374 goto err_chain0_head_change_cb_add;
1375
1376 err = tcf_block_offload_bind(block, q, ei, extack);
1377 if (err)
1378 goto err_block_offload_bind;
1379
1380 *p_block = block;
1381 return 0;
1382
1383err_block_offload_bind:
1384 tcf_chain0_head_change_cb_del(block, ei);
1385err_chain0_head_change_cb_add:
1386 tcf_block_owner_del(block, q, ei->binder_type);
1387err_block_owner_add:
1388err_block_insert:
1389 tcf_block_refcnt_put(block, true);
1390 return err;
1391}
1392EXPORT_SYMBOL(tcf_block_get_ext);
1393
1394static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1395{
1396 struct tcf_proto __rcu **p_filter_chain = priv;
1397
1398 rcu_assign_pointer(*p_filter_chain, tp_head);
1399}
1400
1401int tcf_block_get(struct tcf_block **p_block,
1402 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1403 struct netlink_ext_ack *extack)
1404{
1405 struct tcf_block_ext_info ei = {
1406 .chain_head_change = tcf_chain_head_change_dflt,
1407 .chain_head_change_priv = p_filter_chain,
1408 };
1409
1410 WARN_ON(!p_filter_chain);
1411 return tcf_block_get_ext(p_block, q, &ei, extack);
1412}
1413EXPORT_SYMBOL(tcf_block_get);
1414
1415/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1416 * actions should be all removed after flushing.
1417 */
1418void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1419 struct tcf_block_ext_info *ei)
1420{
1421 if (!block)
1422 return;
1423 tcf_chain0_head_change_cb_del(block, ei);
1424 tcf_block_owner_del(block, q, ei->binder_type);
1425
1426 __tcf_block_put(block, q, ei, true);
1427}
1428EXPORT_SYMBOL(tcf_block_put_ext);
1429
1430void tcf_block_put(struct tcf_block *block)
1431{
1432 struct tcf_block_ext_info ei = {0, };
1433
1434 if (!block)
1435 return;
1436 tcf_block_put_ext(block, block->q, &ei);
1437}
1438
1439EXPORT_SYMBOL(tcf_block_put);
1440
1441static int
1442tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1443 void *cb_priv, bool add, bool offload_in_use,
1444 struct netlink_ext_ack *extack)
1445{
1446 struct tcf_chain *chain, *chain_prev;
1447 struct tcf_proto *tp, *tp_prev;
1448 int err;
1449
1450 lockdep_assert_held(&block->cb_lock);
1451
1452 for (chain = __tcf_get_next_chain(block, NULL);
1453 chain;
1454 chain_prev = chain,
1455 chain = __tcf_get_next_chain(block, chain),
1456 tcf_chain_put(chain_prev)) {
1457 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1458 tp_prev = tp,
1459 tp = __tcf_get_next_proto(chain, tp),
1460 tcf_proto_put(tp_prev, true, NULL)) {
1461 if (tp->ops->reoffload) {
1462 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1463 extack);
1464 if (err && add)
1465 goto err_playback_remove;
1466 } else if (add && offload_in_use) {
1467 err = -EOPNOTSUPP;
1468 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1469 goto err_playback_remove;
1470 }
1471 }
1472 }
1473
1474 return 0;
1475
1476err_playback_remove:
1477 tcf_proto_put(tp, true, NULL);
1478 tcf_chain_put(chain);
1479 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1480 extack);
1481 return err;
1482}
1483
1484static int tcf_block_bind(struct tcf_block *block,
1485 struct flow_block_offload *bo)
1486{
1487 struct flow_block_cb *block_cb, *next;
1488 int err, i = 0;
1489
1490 lockdep_assert_held(&block->cb_lock);
1491
1492 list_for_each_entry(block_cb, &bo->cb_list, list) {
1493 err = tcf_block_playback_offloads(block, block_cb->cb,
1494 block_cb->cb_priv, true,
1495 tcf_block_offload_in_use(block),
1496 bo->extack);
1497 if (err)
1498 goto err_unroll;
1499 if (!bo->unlocked_driver_cb)
1500 block->lockeddevcnt++;
1501
1502 i++;
1503 }
1504 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1505
1506 return 0;
1507
1508err_unroll:
1509 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1510 if (i-- > 0) {
1511 list_del(&block_cb->list);
1512 tcf_block_playback_offloads(block, block_cb->cb,
1513 block_cb->cb_priv, false,
1514 tcf_block_offload_in_use(block),
1515 NULL);
1516 if (!bo->unlocked_driver_cb)
1517 block->lockeddevcnt--;
1518 }
1519 flow_block_cb_free(block_cb);
1520 }
1521
1522 return err;
1523}
1524
1525static void tcf_block_unbind(struct tcf_block *block,
1526 struct flow_block_offload *bo)
1527{
1528 struct flow_block_cb *block_cb, *next;
1529
1530 lockdep_assert_held(&block->cb_lock);
1531
1532 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1533 tcf_block_playback_offloads(block, block_cb->cb,
1534 block_cb->cb_priv, false,
1535 tcf_block_offload_in_use(block),
1536 NULL);
1537 list_del(&block_cb->list);
1538 flow_block_cb_free(block_cb);
1539 if (!bo->unlocked_driver_cb)
1540 block->lockeddevcnt--;
1541 }
1542}
1543
1544static int tcf_block_setup(struct tcf_block *block,
1545 struct flow_block_offload *bo)
1546{
1547 int err;
1548
1549 switch (bo->command) {
1550 case FLOW_BLOCK_BIND:
1551 err = tcf_block_bind(block, bo);
1552 break;
1553 case FLOW_BLOCK_UNBIND:
1554 err = 0;
1555 tcf_block_unbind(block, bo);
1556 break;
1557 default:
1558 WARN_ON_ONCE(1);
1559 err = -EOPNOTSUPP;
1560 }
1561
1562 return err;
1563}
1564
1565/* Main classifier routine: scans classifier chain attached
1566 * to this qdisc, (optionally) tests for protocol and asks
1567 * specific classifiers.
1568 */
1569int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1570 struct tcf_result *res, bool compat_mode)
1571{
1572#ifdef CONFIG_NET_CLS_ACT
1573 const int max_reclassify_loop = 4;
1574 const struct tcf_proto *orig_tp = tp;
1575 const struct tcf_proto *first_tp;
1576 int limit = 0;
1577
1578reclassify:
1579#endif
1580 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1581 __be16 protocol = tc_skb_protocol(skb);
1582 int err;
1583
1584 if (tp->protocol != protocol &&
1585 tp->protocol != htons(ETH_P_ALL))
1586 continue;
1587
1588 err = tp->classify(skb, tp, res);
1589#ifdef CONFIG_NET_CLS_ACT
1590 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1591 first_tp = orig_tp;
1592 goto reset;
1593 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1594 first_tp = res->goto_tp;
1595
1596#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1597 {
1598 struct tc_skb_ext *ext;
1599
1600 ext = skb_ext_add(skb, TC_SKB_EXT);
1601 if (WARN_ON_ONCE(!ext))
1602 return TC_ACT_SHOT;
1603
1604 ext->chain = err & TC_ACT_EXT_VAL_MASK;
1605 }
1606#endif
1607 goto reset;
1608 }
1609#endif
1610 if (err >= 0)
1611 return err;
1612 }
1613
1614 return TC_ACT_UNSPEC; /* signal: continue lookup */
1615#ifdef CONFIG_NET_CLS_ACT
1616reset:
1617 if (unlikely(limit++ >= max_reclassify_loop)) {
1618 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1619 tp->chain->block->index,
1620 tp->prio & 0xffff,
1621 ntohs(tp->protocol));
1622 return TC_ACT_SHOT;
1623 }
1624
1625 tp = first_tp;
1626 goto reclassify;
1627#endif
1628}
1629EXPORT_SYMBOL(tcf_classify);
1630
1631struct tcf_chain_info {
1632 struct tcf_proto __rcu **pprev;
1633 struct tcf_proto __rcu *next;
1634};
1635
1636static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1637 struct tcf_chain_info *chain_info)
1638{
1639 return tcf_chain_dereference(*chain_info->pprev, chain);
1640}
1641
1642static int tcf_chain_tp_insert(struct tcf_chain *chain,
1643 struct tcf_chain_info *chain_info,
1644 struct tcf_proto *tp)
1645{
1646 if (chain->flushing)
1647 return -EAGAIN;
1648
1649 if (*chain_info->pprev == chain->filter_chain)
1650 tcf_chain0_head_change(chain, tp);
1651 tcf_proto_get(tp);
1652 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1653 rcu_assign_pointer(*chain_info->pprev, tp);
1654
1655 return 0;
1656}
1657
1658static void tcf_chain_tp_remove(struct tcf_chain *chain,
1659 struct tcf_chain_info *chain_info,
1660 struct tcf_proto *tp)
1661{
1662 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1663
1664 tcf_proto_mark_delete(tp);
1665 if (tp == chain->filter_chain)
1666 tcf_chain0_head_change(chain, next);
1667 RCU_INIT_POINTER(*chain_info->pprev, next);
1668}
1669
1670static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1671 struct tcf_chain_info *chain_info,
1672 u32 protocol, u32 prio,
1673 bool prio_allocate);
1674
1675/* Try to insert new proto.
1676 * If proto with specified priority already exists, free new proto
1677 * and return existing one.
1678 */
1679
1680static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1681 struct tcf_proto *tp_new,
1682 u32 protocol, u32 prio,
1683 bool rtnl_held)
1684{
1685 struct tcf_chain_info chain_info;
1686 struct tcf_proto *tp;
1687 int err = 0;
1688
1689 mutex_lock(&chain->filter_chain_lock);
1690
1691 if (tcf_proto_exists_destroying(chain, tp_new)) {
1692 mutex_unlock(&chain->filter_chain_lock);
1693 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1694 return ERR_PTR(-EAGAIN);
1695 }
1696
1697 tp = tcf_chain_tp_find(chain, &chain_info,
1698 protocol, prio, false);
1699 if (!tp)
1700 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1701 mutex_unlock(&chain->filter_chain_lock);
1702
1703 if (tp) {
1704 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1705 tp_new = tp;
1706 } else if (err) {
1707 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1708 tp_new = ERR_PTR(err);
1709 }
1710
1711 return tp_new;
1712}
1713
1714static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1715 struct tcf_proto *tp, bool rtnl_held,
1716 struct netlink_ext_ack *extack)
1717{
1718 struct tcf_chain_info chain_info;
1719 struct tcf_proto *tp_iter;
1720 struct tcf_proto **pprev;
1721 struct tcf_proto *next;
1722
1723 mutex_lock(&chain->filter_chain_lock);
1724
1725 /* Atomically find and remove tp from chain. */
1726 for (pprev = &chain->filter_chain;
1727 (tp_iter = tcf_chain_dereference(*pprev, chain));
1728 pprev = &tp_iter->next) {
1729 if (tp_iter == tp) {
1730 chain_info.pprev = pprev;
1731 chain_info.next = tp_iter->next;
1732 WARN_ON(tp_iter->deleting);
1733 break;
1734 }
1735 }
1736 /* Verify that tp still exists and no new filters were inserted
1737 * concurrently.
1738 * Mark tp for deletion if it is empty.
1739 */
1740 if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1741 mutex_unlock(&chain->filter_chain_lock);
1742 return;
1743 }
1744
1745 tcf_proto_signal_destroying(chain, tp);
1746 next = tcf_chain_dereference(chain_info.next, chain);
1747 if (tp == chain->filter_chain)
1748 tcf_chain0_head_change(chain, next);
1749 RCU_INIT_POINTER(*chain_info.pprev, next);
1750 mutex_unlock(&chain->filter_chain_lock);
1751
1752 tcf_proto_put(tp, rtnl_held, extack);
1753}
1754
1755static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1756 struct tcf_chain_info *chain_info,
1757 u32 protocol, u32 prio,
1758 bool prio_allocate)
1759{
1760 struct tcf_proto **pprev;
1761 struct tcf_proto *tp;
1762
1763 /* Check the chain for existence of proto-tcf with this priority */
1764 for (pprev = &chain->filter_chain;
1765 (tp = tcf_chain_dereference(*pprev, chain));
1766 pprev = &tp->next) {
1767 if (tp->prio >= prio) {
1768 if (tp->prio == prio) {
1769 if (prio_allocate ||
1770 (tp->protocol != protocol && protocol))
1771 return ERR_PTR(-EINVAL);
1772 } else {
1773 tp = NULL;
1774 }
1775 break;
1776 }
1777 }
1778 chain_info->pprev = pprev;
1779 if (tp) {
1780 chain_info->next = tp->next;
1781 tcf_proto_get(tp);
1782 } else {
1783 chain_info->next = NULL;
1784 }
1785 return tp;
1786}
1787
1788static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1789 struct tcf_proto *tp, struct tcf_block *block,
1790 struct Qdisc *q, u32 parent, void *fh,
1791 u32 portid, u32 seq, u16 flags, int event,
1792 bool rtnl_held)
1793{
1794 struct tcmsg *tcm;
1795 struct nlmsghdr *nlh;
1796 unsigned char *b = skb_tail_pointer(skb);
1797
1798 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1799 if (!nlh)
1800 goto out_nlmsg_trim;
1801 tcm = nlmsg_data(nlh);
1802 tcm->tcm_family = AF_UNSPEC;
1803 tcm->tcm__pad1 = 0;
1804 tcm->tcm__pad2 = 0;
1805 if (q) {
1806 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1807 tcm->tcm_parent = parent;
1808 } else {
1809 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1810 tcm->tcm_block_index = block->index;
1811 }
1812 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1813 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1814 goto nla_put_failure;
1815 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1816 goto nla_put_failure;
1817 if (!fh) {
1818 tcm->tcm_handle = 0;
1819 } else {
1820 if (tp->ops->dump &&
1821 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1822 goto nla_put_failure;
1823 }
1824 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1825 return skb->len;
1826
1827out_nlmsg_trim:
1828nla_put_failure:
1829 nlmsg_trim(skb, b);
1830 return -1;
1831}
1832
1833static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1834 struct nlmsghdr *n, struct tcf_proto *tp,
1835 struct tcf_block *block, struct Qdisc *q,
1836 u32 parent, void *fh, int event, bool unicast,
1837 bool rtnl_held)
1838{
1839 struct sk_buff *skb;
1840 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1841 int err = 0;
1842
1843 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1844 if (!skb)
1845 return -ENOBUFS;
1846
1847 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1848 n->nlmsg_seq, n->nlmsg_flags, event,
1849 rtnl_held) <= 0) {
1850 kfree_skb(skb);
1851 return -EINVAL;
1852 }
1853
1854 if (unicast)
1855 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1856 else
1857 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1858 n->nlmsg_flags & NLM_F_ECHO);
1859
1860 if (err > 0)
1861 err = 0;
1862 return err;
1863}
1864
1865static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1866 struct nlmsghdr *n, struct tcf_proto *tp,
1867 struct tcf_block *block, struct Qdisc *q,
1868 u32 parent, void *fh, bool unicast, bool *last,
1869 bool rtnl_held, struct netlink_ext_ack *extack)
1870{
1871 struct sk_buff *skb;
1872 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1873 int err;
1874
1875 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1876 if (!skb)
1877 return -ENOBUFS;
1878
1879 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1880 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1881 rtnl_held) <= 0) {
1882 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1883 kfree_skb(skb);
1884 return -EINVAL;
1885 }
1886
1887 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1888 if (err) {
1889 kfree_skb(skb);
1890 return err;
1891 }
1892
1893 if (unicast)
1894 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1895 else
1896 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1897 n->nlmsg_flags & NLM_F_ECHO);
1898 if (err < 0)
1899 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1900
1901 if (err > 0)
1902 err = 0;
1903 return err;
1904}
1905
1906static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1907 struct tcf_block *block, struct Qdisc *q,
1908 u32 parent, struct nlmsghdr *n,
1909 struct tcf_chain *chain, int event,
1910 bool rtnl_held)
1911{
1912 struct tcf_proto *tp;
1913
1914 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1915 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1916 tfilter_notify(net, oskb, n, tp, block,
1917 q, parent, NULL, event, false, rtnl_held);
1918}
1919
1920static void tfilter_put(struct tcf_proto *tp, void *fh)
1921{
1922 if (tp->ops->put && fh)
1923 tp->ops->put(tp, fh);
1924}
1925
1926static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1927 struct netlink_ext_ack *extack)
1928{
1929 struct net *net = sock_net(skb->sk);
1930 struct nlattr *tca[TCA_MAX + 1];
1931 char name[IFNAMSIZ];
1932 struct tcmsg *t;
1933 u32 protocol;
1934 u32 prio;
1935 bool prio_allocate;
1936 u32 parent;
1937 u32 chain_index;
1938 struct Qdisc *q = NULL;
1939 struct tcf_chain_info chain_info;
1940 struct tcf_chain *chain = NULL;
1941 struct tcf_block *block;
1942 struct tcf_proto *tp;
1943 unsigned long cl;
1944 void *fh;
1945 int err;
1946 int tp_created;
1947 bool rtnl_held = false;
1948
1949 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1950 return -EPERM;
1951
1952replay:
1953 tp_created = 0;
1954
1955 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1956 rtm_tca_policy, extack);
1957 if (err < 0)
1958 return err;
1959
1960 t = nlmsg_data(n);
1961 protocol = TC_H_MIN(t->tcm_info);
1962 prio = TC_H_MAJ(t->tcm_info);
1963 prio_allocate = false;
1964 parent = t->tcm_parent;
1965 tp = NULL;
1966 cl = 0;
1967 block = NULL;
1968
1969 if (prio == 0) {
1970 /* If no priority is provided by the user,
1971 * we allocate one.
1972 */
1973 if (n->nlmsg_flags & NLM_F_CREATE) {
1974 prio = TC_H_MAKE(0x80000000U, 0U);
1975 prio_allocate = true;
1976 } else {
1977 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1978 return -ENOENT;
1979 }
1980 }
1981
1982 /* Find head of filter chain. */
1983
1984 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
1985 if (err)
1986 return err;
1987
1988 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
1989 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
1990 err = -EINVAL;
1991 goto errout;
1992 }
1993
1994 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
1995 * block is shared (no qdisc found), qdisc is not unlocked, classifier
1996 * type is not specified, classifier is not unlocked.
1997 */
1998 if (rtnl_held ||
1999 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2000 !tcf_proto_is_unlocked(name)) {
2001 rtnl_held = true;
2002 rtnl_lock();
2003 }
2004
2005 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2006 if (err)
2007 goto errout;
2008
2009 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2010 extack);
2011 if (IS_ERR(block)) {
2012 err = PTR_ERR(block);
2013 goto errout;
2014 }
2015
2016 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2017 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2018 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2019 err = -EINVAL;
2020 goto errout;
2021 }
2022 chain = tcf_chain_get(block, chain_index, true);
2023 if (!chain) {
2024 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2025 err = -ENOMEM;
2026 goto errout;
2027 }
2028
2029 mutex_lock(&chain->filter_chain_lock);
2030 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2031 prio, prio_allocate);
2032 if (IS_ERR(tp)) {
2033 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2034 err = PTR_ERR(tp);
2035 goto errout_locked;
2036 }
2037
2038 if (tp == NULL) {
2039 struct tcf_proto *tp_new = NULL;
2040
2041 if (chain->flushing) {
2042 err = -EAGAIN;
2043 goto errout_locked;
2044 }
2045
2046 /* Proto-tcf does not exist, create new one */
2047
2048 if (tca[TCA_KIND] == NULL || !protocol) {
2049 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2050 err = -EINVAL;
2051 goto errout_locked;
2052 }
2053
2054 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2055 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2056 err = -ENOENT;
2057 goto errout_locked;
2058 }
2059
2060 if (prio_allocate)
2061 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2062 &chain_info));
2063
2064 mutex_unlock(&chain->filter_chain_lock);
2065 tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
2066 protocol, prio, chain, rtnl_held,
2067 extack);
2068 if (IS_ERR(tp_new)) {
2069 err = PTR_ERR(tp_new);
2070 goto errout_tp;
2071 }
2072
2073 tp_created = 1;
2074 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2075 rtnl_held);
2076 if (IS_ERR(tp)) {
2077 err = PTR_ERR(tp);
2078 goto errout_tp;
2079 }
2080 } else {
2081 mutex_unlock(&chain->filter_chain_lock);
2082 }
2083
2084 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2085 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2086 err = -EINVAL;
2087 goto errout;
2088 }
2089
2090 fh = tp->ops->get(tp, t->tcm_handle);
2091
2092 if (!fh) {
2093 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2094 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2095 err = -ENOENT;
2096 goto errout;
2097 }
2098 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2099 tfilter_put(tp, fh);
2100 NL_SET_ERR_MSG(extack, "Filter already exists");
2101 err = -EEXIST;
2102 goto errout;
2103 }
2104
2105 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2106 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2107 err = -EINVAL;
2108 goto errout;
2109 }
2110
2111 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2112 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2113 rtnl_held, extack);
2114 if (err == 0) {
2115 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2116 RTM_NEWTFILTER, false, rtnl_held);
2117 tfilter_put(tp, fh);
2118 /* q pointer is NULL for shared blocks */
2119 if (q)
2120 q->flags &= ~TCQ_F_CAN_BYPASS;
2121 }
2122
2123errout:
2124 if (err && tp_created)
2125 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2126errout_tp:
2127 if (chain) {
2128 if (tp && !IS_ERR(tp))
2129 tcf_proto_put(tp, rtnl_held, NULL);
2130 if (!tp_created)
2131 tcf_chain_put(chain);
2132 }
2133 tcf_block_release(q, block, rtnl_held);
2134
2135 if (rtnl_held)
2136 rtnl_unlock();
2137
2138 if (err == -EAGAIN) {
2139 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2140 * of target chain.
2141 */
2142 rtnl_held = true;
2143 /* Replay the request. */
2144 goto replay;
2145 }
2146 return err;
2147
2148errout_locked:
2149 mutex_unlock(&chain->filter_chain_lock);
2150 goto errout;
2151}
2152
2153static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2154 struct netlink_ext_ack *extack)
2155{
2156 struct net *net = sock_net(skb->sk);
2157 struct nlattr *tca[TCA_MAX + 1];
2158 char name[IFNAMSIZ];
2159 struct tcmsg *t;
2160 u32 protocol;
2161 u32 prio;
2162 u32 parent;
2163 u32 chain_index;
2164 struct Qdisc *q = NULL;
2165 struct tcf_chain_info chain_info;
2166 struct tcf_chain *chain = NULL;
2167 struct tcf_block *block = NULL;
2168 struct tcf_proto *tp = NULL;
2169 unsigned long cl = 0;
2170 void *fh = NULL;
2171 int err;
2172 bool rtnl_held = false;
2173
2174 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2175 return -EPERM;
2176
2177 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2178 rtm_tca_policy, extack);
2179 if (err < 0)
2180 return err;
2181
2182 t = nlmsg_data(n);
2183 protocol = TC_H_MIN(t->tcm_info);
2184 prio = TC_H_MAJ(t->tcm_info);
2185 parent = t->tcm_parent;
2186
2187 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2188 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2189 return -ENOENT;
2190 }
2191
2192 /* Find head of filter chain. */
2193
2194 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2195 if (err)
2196 return err;
2197
2198 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2199 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2200 err = -EINVAL;
2201 goto errout;
2202 }
2203 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2204 * found), qdisc is not unlocked, classifier type is not specified,
2205 * classifier is not unlocked.
2206 */
2207 if (!prio ||
2208 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2209 !tcf_proto_is_unlocked(name)) {
2210 rtnl_held = true;
2211 rtnl_lock();
2212 }
2213
2214 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2215 if (err)
2216 goto errout;
2217
2218 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2219 extack);
2220 if (IS_ERR(block)) {
2221 err = PTR_ERR(block);
2222 goto errout;
2223 }
2224
2225 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2226 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2227 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2228 err = -EINVAL;
2229 goto errout;
2230 }
2231 chain = tcf_chain_get(block, chain_index, false);
2232 if (!chain) {
2233 /* User requested flush on non-existent chain. Nothing to do,
2234 * so just return success.
2235 */
2236 if (prio == 0) {
2237 err = 0;
2238 goto errout;
2239 }
2240 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2241 err = -ENOENT;
2242 goto errout;
2243 }
2244
2245 if (prio == 0) {
2246 tfilter_notify_chain(net, skb, block, q, parent, n,
2247 chain, RTM_DELTFILTER, rtnl_held);
2248 tcf_chain_flush(chain, rtnl_held);
2249 err = 0;
2250 goto errout;
2251 }
2252
2253 mutex_lock(&chain->filter_chain_lock);
2254 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2255 prio, false);
2256 if (!tp || IS_ERR(tp)) {
2257 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2258 err = tp ? PTR_ERR(tp) : -ENOENT;
2259 goto errout_locked;
2260 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2261 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2262 err = -EINVAL;
2263 goto errout_locked;
2264 } else if (t->tcm_handle == 0) {
2265 tcf_proto_signal_destroying(chain, tp);
2266 tcf_chain_tp_remove(chain, &chain_info, tp);
2267 mutex_unlock(&chain->filter_chain_lock);
2268
2269 tcf_proto_put(tp, rtnl_held, NULL);
2270 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2271 RTM_DELTFILTER, false, rtnl_held);
2272 err = 0;
2273 goto errout;
2274 }
2275 mutex_unlock(&chain->filter_chain_lock);
2276
2277 fh = tp->ops->get(tp, t->tcm_handle);
2278
2279 if (!fh) {
2280 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2281 err = -ENOENT;
2282 } else {
2283 bool last;
2284
2285 err = tfilter_del_notify(net, skb, n, tp, block,
2286 q, parent, fh, false, &last,
2287 rtnl_held, extack);
2288
2289 if (err)
2290 goto errout;
2291 if (last)
2292 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2293 }
2294
2295errout:
2296 if (chain) {
2297 if (tp && !IS_ERR(tp))
2298 tcf_proto_put(tp, rtnl_held, NULL);
2299 tcf_chain_put(chain);
2300 }
2301 tcf_block_release(q, block, rtnl_held);
2302
2303 if (rtnl_held)
2304 rtnl_unlock();
2305
2306 return err;
2307
2308errout_locked:
2309 mutex_unlock(&chain->filter_chain_lock);
2310 goto errout;
2311}
2312
2313static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2314 struct netlink_ext_ack *extack)
2315{
2316 struct net *net = sock_net(skb->sk);
2317 struct nlattr *tca[TCA_MAX + 1];
2318 char name[IFNAMSIZ];
2319 struct tcmsg *t;
2320 u32 protocol;
2321 u32 prio;
2322 u32 parent;
2323 u32 chain_index;
2324 struct Qdisc *q = NULL;
2325 struct tcf_chain_info chain_info;
2326 struct tcf_chain *chain = NULL;
2327 struct tcf_block *block = NULL;
2328 struct tcf_proto *tp = NULL;
2329 unsigned long cl = 0;
2330 void *fh = NULL;
2331 int err;
2332 bool rtnl_held = false;
2333
2334 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2335 rtm_tca_policy, extack);
2336 if (err < 0)
2337 return err;
2338
2339 t = nlmsg_data(n);
2340 protocol = TC_H_MIN(t->tcm_info);
2341 prio = TC_H_MAJ(t->tcm_info);
2342 parent = t->tcm_parent;
2343
2344 if (prio == 0) {
2345 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2346 return -ENOENT;
2347 }
2348
2349 /* Find head of filter chain. */
2350
2351 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2352 if (err)
2353 return err;
2354
2355 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2356 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2357 err = -EINVAL;
2358 goto errout;
2359 }
2360 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2361 * unlocked, classifier type is not specified, classifier is not
2362 * unlocked.
2363 */
2364 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2365 !tcf_proto_is_unlocked(name)) {
2366 rtnl_held = true;
2367 rtnl_lock();
2368 }
2369
2370 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2371 if (err)
2372 goto errout;
2373
2374 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2375 extack);
2376 if (IS_ERR(block)) {
2377 err = PTR_ERR(block);
2378 goto errout;
2379 }
2380
2381 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2382 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2383 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2384 err = -EINVAL;
2385 goto errout;
2386 }
2387 chain = tcf_chain_get(block, chain_index, false);
2388 if (!chain) {
2389 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2390 err = -EINVAL;
2391 goto errout;
2392 }
2393
2394 mutex_lock(&chain->filter_chain_lock);
2395 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2396 prio, false);
2397 mutex_unlock(&chain->filter_chain_lock);
2398 if (!tp || IS_ERR(tp)) {
2399 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2400 err = tp ? PTR_ERR(tp) : -ENOENT;
2401 goto errout;
2402 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2403 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2404 err = -EINVAL;
2405 goto errout;
2406 }
2407
2408 fh = tp->ops->get(tp, t->tcm_handle);
2409
2410 if (!fh) {
2411 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2412 err = -ENOENT;
2413 } else {
2414 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2415 fh, RTM_NEWTFILTER, true, rtnl_held);
2416 if (err < 0)
2417 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2418 }
2419
2420 tfilter_put(tp, fh);
2421errout:
2422 if (chain) {
2423 if (tp && !IS_ERR(tp))
2424 tcf_proto_put(tp, rtnl_held, NULL);
2425 tcf_chain_put(chain);
2426 }
2427 tcf_block_release(q, block, rtnl_held);
2428
2429 if (rtnl_held)
2430 rtnl_unlock();
2431
2432 return err;
2433}
2434
2435struct tcf_dump_args {
2436 struct tcf_walker w;
2437 struct sk_buff *skb;
2438 struct netlink_callback *cb;
2439 struct tcf_block *block;
2440 struct Qdisc *q;
2441 u32 parent;
2442};
2443
2444static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2445{
2446 struct tcf_dump_args *a = (void *)arg;
2447 struct net *net = sock_net(a->skb->sk);
2448
2449 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2450 n, NETLINK_CB(a->cb->skb).portid,
2451 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2452 RTM_NEWTFILTER, true);
2453}
2454
2455static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2456 struct sk_buff *skb, struct netlink_callback *cb,
2457 long index_start, long *p_index)
2458{
2459 struct net *net = sock_net(skb->sk);
2460 struct tcf_block *block = chain->block;
2461 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2462 struct tcf_proto *tp, *tp_prev;
2463 struct tcf_dump_args arg;
2464
2465 for (tp = __tcf_get_next_proto(chain, NULL);
2466 tp;
2467 tp_prev = tp,
2468 tp = __tcf_get_next_proto(chain, tp),
2469 tcf_proto_put(tp_prev, true, NULL),
2470 (*p_index)++) {
2471 if (*p_index < index_start)
2472 continue;
2473 if (TC_H_MAJ(tcm->tcm_info) &&
2474 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2475 continue;
2476 if (TC_H_MIN(tcm->tcm_info) &&
2477 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2478 continue;
2479 if (*p_index > index_start)
2480 memset(&cb->args[1], 0,
2481 sizeof(cb->args) - sizeof(cb->args[0]));
2482 if (cb->args[1] == 0) {
2483 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2484 NETLINK_CB(cb->skb).portid,
2485 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2486 RTM_NEWTFILTER, true) <= 0)
2487 goto errout;
2488 cb->args[1] = 1;
2489 }
2490 if (!tp->ops->walk)
2491 continue;
2492 arg.w.fn = tcf_node_dump;
2493 arg.skb = skb;
2494 arg.cb = cb;
2495 arg.block = block;
2496 arg.q = q;
2497 arg.parent = parent;
2498 arg.w.stop = 0;
2499 arg.w.skip = cb->args[1] - 1;
2500 arg.w.count = 0;
2501 arg.w.cookie = cb->args[2];
2502 tp->ops->walk(tp, &arg.w, true);
2503 cb->args[2] = arg.w.cookie;
2504 cb->args[1] = arg.w.count + 1;
2505 if (arg.w.stop)
2506 goto errout;
2507 }
2508 return true;
2509
2510errout:
2511 tcf_proto_put(tp, true, NULL);
2512 return false;
2513}
2514
2515/* called with RTNL */
2516static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2517{
2518 struct tcf_chain *chain, *chain_prev;
2519 struct net *net = sock_net(skb->sk);
2520 struct nlattr *tca[TCA_MAX + 1];
2521 struct Qdisc *q = NULL;
2522 struct tcf_block *block;
2523 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2524 long index_start;
2525 long index;
2526 u32 parent;
2527 int err;
2528
2529 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2530 return skb->len;
2531
2532 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2533 NULL, cb->extack);
2534 if (err)
2535 return err;
2536
2537 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2538 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2539 if (!block)
2540 goto out;
2541 /* If we work with block index, q is NULL and parent value
2542 * will never be used in the following code. The check
2543 * in tcf_fill_node prevents it. However, compiler does not
2544 * see that far, so set parent to zero to silence the warning
2545 * about parent being uninitialized.
2546 */
2547 parent = 0;
2548 } else {
2549 const struct Qdisc_class_ops *cops;
2550 struct net_device *dev;
2551 unsigned long cl = 0;
2552
2553 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2554 if (!dev)
2555 return skb->len;
2556
2557 parent = tcm->tcm_parent;
2558 if (!parent) {
2559 q = dev->qdisc;
2560 parent = q->handle;
2561 } else {
2562 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2563 }
2564 if (!q)
2565 goto out;
2566 cops = q->ops->cl_ops;
2567 if (!cops)
2568 goto out;
2569 if (!cops->tcf_block)
2570 goto out;
2571 if (TC_H_MIN(tcm->tcm_parent)) {
2572 cl = cops->find(q, tcm->tcm_parent);
2573 if (cl == 0)
2574 goto out;
2575 }
2576 block = cops->tcf_block(q, cl, NULL);
2577 if (!block)
2578 goto out;
2579 if (tcf_block_shared(block))
2580 q = NULL;
2581 }
2582
2583 index_start = cb->args[0];
2584 index = 0;
2585
2586 for (chain = __tcf_get_next_chain(block, NULL);
2587 chain;
2588 chain_prev = chain,
2589 chain = __tcf_get_next_chain(block, chain),
2590 tcf_chain_put(chain_prev)) {
2591 if (tca[TCA_CHAIN] &&
2592 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2593 continue;
2594 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2595 index_start, &index)) {
2596 tcf_chain_put(chain);
2597 err = -EMSGSIZE;
2598 break;
2599 }
2600 }
2601
2602 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2603 tcf_block_refcnt_put(block, true);
2604 cb->args[0] = index;
2605
2606out:
2607 /* If we did no progress, the error (EMSGSIZE) is real */
2608 if (skb->len == 0 && err)
2609 return err;
2610 return skb->len;
2611}
2612
2613static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2614 void *tmplt_priv, u32 chain_index,
2615 struct net *net, struct sk_buff *skb,
2616 struct tcf_block *block,
2617 u32 portid, u32 seq, u16 flags, int event)
2618{
2619 unsigned char *b = skb_tail_pointer(skb);
2620 const struct tcf_proto_ops *ops;
2621 struct nlmsghdr *nlh;
2622 struct tcmsg *tcm;
2623 void *priv;
2624
2625 ops = tmplt_ops;
2626 priv = tmplt_priv;
2627
2628 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2629 if (!nlh)
2630 goto out_nlmsg_trim;
2631 tcm = nlmsg_data(nlh);
2632 tcm->tcm_family = AF_UNSPEC;
2633 tcm->tcm__pad1 = 0;
2634 tcm->tcm__pad2 = 0;
2635 tcm->tcm_handle = 0;
2636 if (block->q) {
2637 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2638 tcm->tcm_parent = block->q->handle;
2639 } else {
2640 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2641 tcm->tcm_block_index = block->index;
2642 }
2643
2644 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2645 goto nla_put_failure;
2646
2647 if (ops) {
2648 if (nla_put_string(skb, TCA_KIND, ops->kind))
2649 goto nla_put_failure;
2650 if (ops->tmplt_dump(skb, net, priv) < 0)
2651 goto nla_put_failure;
2652 }
2653
2654 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2655 return skb->len;
2656
2657out_nlmsg_trim:
2658nla_put_failure:
2659 nlmsg_trim(skb, b);
2660 return -EMSGSIZE;
2661}
2662
2663static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2664 u32 seq, u16 flags, int event, bool unicast)
2665{
2666 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2667 struct tcf_block *block = chain->block;
2668 struct net *net = block->net;
2669 struct sk_buff *skb;
2670 int err = 0;
2671
2672 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2673 if (!skb)
2674 return -ENOBUFS;
2675
2676 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2677 chain->index, net, skb, block, portid,
2678 seq, flags, event) <= 0) {
2679 kfree_skb(skb);
2680 return -EINVAL;
2681 }
2682
2683 if (unicast)
2684 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2685 else
2686 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2687 flags & NLM_F_ECHO);
2688
2689 if (err > 0)
2690 err = 0;
2691 return err;
2692}
2693
2694static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2695 void *tmplt_priv, u32 chain_index,
2696 struct tcf_block *block, struct sk_buff *oskb,
2697 u32 seq, u16 flags, bool unicast)
2698{
2699 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2700 struct net *net = block->net;
2701 struct sk_buff *skb;
2702
2703 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2704 if (!skb)
2705 return -ENOBUFS;
2706
2707 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2708 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2709 kfree_skb(skb);
2710 return -EINVAL;
2711 }
2712
2713 if (unicast)
2714 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2715
2716 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2717}
2718
2719static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2720 struct nlattr **tca,
2721 struct netlink_ext_ack *extack)
2722{
2723 const struct tcf_proto_ops *ops;
2724 void *tmplt_priv;
2725
2726 /* If kind is not set, user did not specify template. */
2727 if (!tca[TCA_KIND])
2728 return 0;
2729
2730 ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), true, extack);
2731 if (IS_ERR(ops))
2732 return PTR_ERR(ops);
2733 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2734 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2735 return -EOPNOTSUPP;
2736 }
2737
2738 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2739 if (IS_ERR(tmplt_priv)) {
2740 module_put(ops->owner);
2741 return PTR_ERR(tmplt_priv);
2742 }
2743 chain->tmplt_ops = ops;
2744 chain->tmplt_priv = tmplt_priv;
2745 return 0;
2746}
2747
2748static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2749 void *tmplt_priv)
2750{
2751 /* If template ops are set, no work to do for us. */
2752 if (!tmplt_ops)
2753 return;
2754
2755 tmplt_ops->tmplt_destroy(tmplt_priv);
2756 module_put(tmplt_ops->owner);
2757}
2758
2759/* Add/delete/get a chain */
2760
2761static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2762 struct netlink_ext_ack *extack)
2763{
2764 struct net *net = sock_net(skb->sk);
2765 struct nlattr *tca[TCA_MAX + 1];
2766 struct tcmsg *t;
2767 u32 parent;
2768 u32 chain_index;
2769 struct Qdisc *q = NULL;
2770 struct tcf_chain *chain = NULL;
2771 struct tcf_block *block;
2772 unsigned long cl;
2773 int err;
2774
2775 if (n->nlmsg_type != RTM_GETCHAIN &&
2776 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2777 return -EPERM;
2778
2779replay:
2780 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2781 rtm_tca_policy, extack);
2782 if (err < 0)
2783 return err;
2784
2785 t = nlmsg_data(n);
2786 parent = t->tcm_parent;
2787 cl = 0;
2788
2789 block = tcf_block_find(net, &q, &parent, &cl,
2790 t->tcm_ifindex, t->tcm_block_index, extack);
2791 if (IS_ERR(block))
2792 return PTR_ERR(block);
2793
2794 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2795 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2796 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2797 err = -EINVAL;
2798 goto errout_block;
2799 }
2800
2801 mutex_lock(&block->lock);
2802 chain = tcf_chain_lookup(block, chain_index);
2803 if (n->nlmsg_type == RTM_NEWCHAIN) {
2804 if (chain) {
2805 if (tcf_chain_held_by_acts_only(chain)) {
2806 /* The chain exists only because there is
2807 * some action referencing it.
2808 */
2809 tcf_chain_hold(chain);
2810 } else {
2811 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2812 err = -EEXIST;
2813 goto errout_block_locked;
2814 }
2815 } else {
2816 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2817 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2818 err = -ENOENT;
2819 goto errout_block_locked;
2820 }
2821 chain = tcf_chain_create(block, chain_index);
2822 if (!chain) {
2823 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2824 err = -ENOMEM;
2825 goto errout_block_locked;
2826 }
2827 }
2828 } else {
2829 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2830 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2831 err = -EINVAL;
2832 goto errout_block_locked;
2833 }
2834 tcf_chain_hold(chain);
2835 }
2836
2837 if (n->nlmsg_type == RTM_NEWCHAIN) {
2838 /* Modifying chain requires holding parent block lock. In case
2839 * the chain was successfully added, take a reference to the
2840 * chain. This ensures that an empty chain does not disappear at
2841 * the end of this function.
2842 */
2843 tcf_chain_hold(chain);
2844 chain->explicitly_created = true;
2845 }
2846 mutex_unlock(&block->lock);
2847
2848 switch (n->nlmsg_type) {
2849 case RTM_NEWCHAIN:
2850 err = tc_chain_tmplt_add(chain, net, tca, extack);
2851 if (err) {
2852 tcf_chain_put_explicitly_created(chain);
2853 goto errout;
2854 }
2855
2856 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2857 RTM_NEWCHAIN, false);
2858 break;
2859 case RTM_DELCHAIN:
2860 tfilter_notify_chain(net, skb, block, q, parent, n,
2861 chain, RTM_DELTFILTER, true);
2862 /* Flush the chain first as the user requested chain removal. */
2863 tcf_chain_flush(chain, true);
2864 /* In case the chain was successfully deleted, put a reference
2865 * to the chain previously taken during addition.
2866 */
2867 tcf_chain_put_explicitly_created(chain);
2868 break;
2869 case RTM_GETCHAIN:
2870 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2871 n->nlmsg_seq, n->nlmsg_type, true);
2872 if (err < 0)
2873 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2874 break;
2875 default:
2876 err = -EOPNOTSUPP;
2877 NL_SET_ERR_MSG(extack, "Unsupported message type");
2878 goto errout;
2879 }
2880
2881errout:
2882 tcf_chain_put(chain);
2883errout_block:
2884 tcf_block_release(q, block, true);
2885 if (err == -EAGAIN)
2886 /* Replay the request. */
2887 goto replay;
2888 return err;
2889
2890errout_block_locked:
2891 mutex_unlock(&block->lock);
2892 goto errout_block;
2893}
2894
2895/* called with RTNL */
2896static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2897{
2898 struct net *net = sock_net(skb->sk);
2899 struct nlattr *tca[TCA_MAX + 1];
2900 struct Qdisc *q = NULL;
2901 struct tcf_block *block;
2902 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2903 struct tcf_chain *chain;
2904 long index_start;
2905 long index;
2906 u32 parent;
2907 int err;
2908
2909 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2910 return skb->len;
2911
2912 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2913 rtm_tca_policy, cb->extack);
2914 if (err)
2915 return err;
2916
2917 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2918 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2919 if (!block)
2920 goto out;
2921 /* If we work with block index, q is NULL and parent value
2922 * will never be used in the following code. The check
2923 * in tcf_fill_node prevents it. However, compiler does not
2924 * see that far, so set parent to zero to silence the warning
2925 * about parent being uninitialized.
2926 */
2927 parent = 0;
2928 } else {
2929 const struct Qdisc_class_ops *cops;
2930 struct net_device *dev;
2931 unsigned long cl = 0;
2932
2933 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2934 if (!dev)
2935 return skb->len;
2936
2937 parent = tcm->tcm_parent;
2938 if (!parent) {
2939 q = dev->qdisc;
2940 parent = q->handle;
2941 } else {
2942 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2943 }
2944 if (!q)
2945 goto out;
2946 cops = q->ops->cl_ops;
2947 if (!cops)
2948 goto out;
2949 if (!cops->tcf_block)
2950 goto out;
2951 if (TC_H_MIN(tcm->tcm_parent)) {
2952 cl = cops->find(q, tcm->tcm_parent);
2953 if (cl == 0)
2954 goto out;
2955 }
2956 block = cops->tcf_block(q, cl, NULL);
2957 if (!block)
2958 goto out;
2959 if (tcf_block_shared(block))
2960 q = NULL;
2961 }
2962
2963 index_start = cb->args[0];
2964 index = 0;
2965
2966 mutex_lock(&block->lock);
2967 list_for_each_entry(chain, &block->chain_list, list) {
2968 if ((tca[TCA_CHAIN] &&
2969 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2970 continue;
2971 if (index < index_start) {
2972 index++;
2973 continue;
2974 }
2975 if (tcf_chain_held_by_acts_only(chain))
2976 continue;
2977 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2978 chain->index, net, skb, block,
2979 NETLINK_CB(cb->skb).portid,
2980 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2981 RTM_NEWCHAIN);
2982 if (err <= 0)
2983 break;
2984 index++;
2985 }
2986 mutex_unlock(&block->lock);
2987
2988 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2989 tcf_block_refcnt_put(block, true);
2990 cb->args[0] = index;
2991
2992out:
2993 /* If we did no progress, the error (EMSGSIZE) is real */
2994 if (skb->len == 0 && err)
2995 return err;
2996 return skb->len;
2997}
2998
2999void tcf_exts_destroy(struct tcf_exts *exts)
3000{
3001#ifdef CONFIG_NET_CLS_ACT
3002 if (exts->actions) {
3003 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3004 kfree(exts->actions);
3005 }
3006 exts->nr_actions = 0;
3007#endif
3008}
3009EXPORT_SYMBOL(tcf_exts_destroy);
3010
3011int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3012 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
3013 bool rtnl_held, struct netlink_ext_ack *extack)
3014{
3015#ifdef CONFIG_NET_CLS_ACT
3016 {
3017 struct tc_action *act;
3018 size_t attr_size = 0;
3019
3020 if (exts->police && tb[exts->police]) {
3021 act = tcf_action_init_1(net, tp, tb[exts->police],
3022 rate_tlv, "police", ovr,
3023 TCA_ACT_BIND, rtnl_held,
3024 extack);
3025 if (IS_ERR(act))
3026 return PTR_ERR(act);
3027
3028 act->type = exts->type = TCA_OLD_COMPAT;
3029 exts->actions[0] = act;
3030 exts->nr_actions = 1;
3031 } else if (exts->action && tb[exts->action]) {
3032 int err;
3033
3034 err = tcf_action_init(net, tp, tb[exts->action],
3035 rate_tlv, NULL, ovr, TCA_ACT_BIND,
3036 exts->actions, &attr_size,
3037 rtnl_held, extack);
3038 if (err < 0)
3039 return err;
3040 exts->nr_actions = err;
3041 }
3042 }
3043#else
3044 if ((exts->action && tb[exts->action]) ||
3045 (exts->police && tb[exts->police])) {
3046 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3047 return -EOPNOTSUPP;
3048 }
3049#endif
3050
3051 return 0;
3052}
3053EXPORT_SYMBOL(tcf_exts_validate);
3054
3055void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3056{
3057#ifdef CONFIG_NET_CLS_ACT
3058 struct tcf_exts old = *dst;
3059
3060 *dst = *src;
3061 tcf_exts_destroy(&old);
3062#endif
3063}
3064EXPORT_SYMBOL(tcf_exts_change);
3065
3066#ifdef CONFIG_NET_CLS_ACT
3067static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3068{
3069 if (exts->nr_actions == 0)
3070 return NULL;
3071 else
3072 return exts->actions[0];
3073}
3074#endif
3075
3076int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3077{
3078#ifdef CONFIG_NET_CLS_ACT
3079 struct nlattr *nest;
3080
3081 if (exts->action && tcf_exts_has_actions(exts)) {
3082 /*
3083 * again for backward compatible mode - we want
3084 * to work with both old and new modes of entering
3085 * tc data even if iproute2 was newer - jhs
3086 */
3087 if (exts->type != TCA_OLD_COMPAT) {
3088 nest = nla_nest_start_noflag(skb, exts->action);
3089 if (nest == NULL)
3090 goto nla_put_failure;
3091
3092 if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3093 goto nla_put_failure;
3094 nla_nest_end(skb, nest);
3095 } else if (exts->police) {
3096 struct tc_action *act = tcf_exts_first_act(exts);
3097 nest = nla_nest_start_noflag(skb, exts->police);
3098 if (nest == NULL || !act)
3099 goto nla_put_failure;
3100 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3101 goto nla_put_failure;
3102 nla_nest_end(skb, nest);
3103 }
3104 }
3105 return 0;
3106
3107nla_put_failure:
3108 nla_nest_cancel(skb, nest);
3109 return -1;
3110#else
3111 return 0;
3112#endif
3113}
3114EXPORT_SYMBOL(tcf_exts_dump);
3115
3116
3117int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3118{
3119#ifdef CONFIG_NET_CLS_ACT
3120 struct tc_action *a = tcf_exts_first_act(exts);
3121 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3122 return -1;
3123#endif
3124 return 0;
3125}
3126EXPORT_SYMBOL(tcf_exts_dump_stats);
3127
3128static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3129{
3130 if (*flags & TCA_CLS_FLAGS_IN_HW)
3131 return;
3132 *flags |= TCA_CLS_FLAGS_IN_HW;
3133 atomic_inc(&block->offloadcnt);
3134}
3135
3136static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3137{
3138 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3139 return;
3140 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3141 atomic_dec(&block->offloadcnt);
3142}
3143
3144static void tc_cls_offload_cnt_update(struct tcf_block *block,
3145 struct tcf_proto *tp, u32 *cnt,
3146 u32 *flags, u32 diff, bool add)
3147{
3148 lockdep_assert_held(&block->cb_lock);
3149
3150 spin_lock(&tp->lock);
3151 if (add) {
3152 if (!*cnt)
3153 tcf_block_offload_inc(block, flags);
3154 *cnt += diff;
3155 } else {
3156 *cnt -= diff;
3157 if (!*cnt)
3158 tcf_block_offload_dec(block, flags);
3159 }
3160 spin_unlock(&tp->lock);
3161}
3162
3163static void
3164tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3165 u32 *cnt, u32 *flags)
3166{
3167 lockdep_assert_held(&block->cb_lock);
3168
3169 spin_lock(&tp->lock);
3170 tcf_block_offload_dec(block, flags);
3171 *cnt = 0;
3172 spin_unlock(&tp->lock);
3173}
3174
3175static int
3176__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3177 void *type_data, bool err_stop)
3178{
3179 struct flow_block_cb *block_cb;
3180 int ok_count = 0;
3181 int err;
3182
3183 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3184 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3185 if (err) {
3186 if (err_stop)
3187 return err;
3188 } else {
3189 ok_count++;
3190 }
3191 }
3192 return ok_count;
3193}
3194
3195int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3196 void *type_data, bool err_stop, bool rtnl_held)
3197{
3198 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3199 int ok_count;
3200
3201retry:
3202 if (take_rtnl)
3203 rtnl_lock();
3204 down_read(&block->cb_lock);
3205 /* Need to obtain rtnl lock if block is bound to devs that require it.
3206 * In block bind code cb_lock is obtained while holding rtnl, so we must
3207 * obtain the locks in same order here.
3208 */
3209 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3210 up_read(&block->cb_lock);
3211 take_rtnl = true;
3212 goto retry;
3213 }
3214
3215 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3216
3217 up_read(&block->cb_lock);
3218 if (take_rtnl)
3219 rtnl_unlock();
3220 return ok_count;
3221}
3222EXPORT_SYMBOL(tc_setup_cb_call);
3223
3224/* Non-destructive filter add. If filter that wasn't already in hardware is
3225 * successfully offloaded, increment block offloads counter. On failure,
3226 * previously offloaded filter is considered to be intact and offloads counter
3227 * is not decremented.
3228 */
3229
3230int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3231 enum tc_setup_type type, void *type_data, bool err_stop,
3232 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3233{
3234 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3235 int ok_count;
3236
3237retry:
3238 if (take_rtnl)
3239 rtnl_lock();
3240 down_read(&block->cb_lock);
3241 /* Need to obtain rtnl lock if block is bound to devs that require it.
3242 * In block bind code cb_lock is obtained while holding rtnl, so we must
3243 * obtain the locks in same order here.
3244 */
3245 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3246 up_read(&block->cb_lock);
3247 take_rtnl = true;
3248 goto retry;
3249 }
3250
3251 /* Make sure all netdevs sharing this block are offload-capable. */
3252 if (block->nooffloaddevcnt && err_stop) {
3253 ok_count = -EOPNOTSUPP;
3254 goto err_unlock;
3255 }
3256
3257 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3258 if (ok_count < 0)
3259 goto err_unlock;
3260
3261 if (tp->ops->hw_add)
3262 tp->ops->hw_add(tp, type_data);
3263 if (ok_count > 0)
3264 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3265 ok_count, true);
3266err_unlock:
3267 up_read(&block->cb_lock);
3268 if (take_rtnl)
3269 rtnl_unlock();
3270 return ok_count < 0 ? ok_count : 0;
3271}
3272EXPORT_SYMBOL(tc_setup_cb_add);
3273
3274/* Destructive filter replace. If filter that wasn't already in hardware is
3275 * successfully offloaded, increment block offload counter. On failure,
3276 * previously offloaded filter is considered to be destroyed and offload counter
3277 * is decremented.
3278 */
3279
3280int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3281 enum tc_setup_type type, void *type_data, bool err_stop,
3282 u32 *old_flags, unsigned int *old_in_hw_count,
3283 u32 *new_flags, unsigned int *new_in_hw_count,
3284 bool rtnl_held)
3285{
3286 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3287 int ok_count;
3288
3289retry:
3290 if (take_rtnl)
3291 rtnl_lock();
3292 down_read(&block->cb_lock);
3293 /* Need to obtain rtnl lock if block is bound to devs that require it.
3294 * In block bind code cb_lock is obtained while holding rtnl, so we must
3295 * obtain the locks in same order here.
3296 */
3297 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3298 up_read(&block->cb_lock);
3299 take_rtnl = true;
3300 goto retry;
3301 }
3302
3303 /* Make sure all netdevs sharing this block are offload-capable. */
3304 if (block->nooffloaddevcnt && err_stop) {
3305 ok_count = -EOPNOTSUPP;
3306 goto err_unlock;
3307 }
3308
3309 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3310 if (tp->ops->hw_del)
3311 tp->ops->hw_del(tp, type_data);
3312
3313 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3314 if (ok_count < 0)
3315 goto err_unlock;
3316
3317 if (tp->ops->hw_add)
3318 tp->ops->hw_add(tp, type_data);
3319 if (ok_count > 0)
3320 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3321 new_flags, ok_count, true);
3322err_unlock:
3323 up_read(&block->cb_lock);
3324 if (take_rtnl)
3325 rtnl_unlock();
3326 return ok_count < 0 ? ok_count : 0;
3327}
3328EXPORT_SYMBOL(tc_setup_cb_replace);
3329
3330/* Destroy filter and decrement block offload counter, if filter was previously
3331 * offloaded.
3332 */
3333
3334int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3335 enum tc_setup_type type, void *type_data, bool err_stop,
3336 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3337{
3338 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3339 int ok_count;
3340
3341retry:
3342 if (take_rtnl)
3343 rtnl_lock();
3344 down_read(&block->cb_lock);
3345 /* Need to obtain rtnl lock if block is bound to devs that require it.
3346 * In block bind code cb_lock is obtained while holding rtnl, so we must
3347 * obtain the locks in same order here.
3348 */
3349 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3350 up_read(&block->cb_lock);
3351 take_rtnl = true;
3352 goto retry;
3353 }
3354
3355 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3356
3357 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3358 if (tp->ops->hw_del)
3359 tp->ops->hw_del(tp, type_data);
3360
3361 up_read(&block->cb_lock);
3362 if (take_rtnl)
3363 rtnl_unlock();
3364 return ok_count < 0 ? ok_count : 0;
3365}
3366EXPORT_SYMBOL(tc_setup_cb_destroy);
3367
3368int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3369 bool add, flow_setup_cb_t *cb,
3370 enum tc_setup_type type, void *type_data,
3371 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3372{
3373 int err = cb(type, type_data, cb_priv);
3374
3375 if (err) {
3376 if (add && tc_skip_sw(*flags))
3377 return err;
3378 } else {
3379 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3380 add);
3381 }
3382
3383 return 0;
3384}
3385EXPORT_SYMBOL(tc_setup_cb_reoffload);
3386
3387void tc_cleanup_flow_action(struct flow_action *flow_action)
3388{
3389 struct flow_action_entry *entry;
3390 int i;
3391
3392 flow_action_for_each(i, entry, flow_action)
3393 if (entry->destructor)
3394 entry->destructor(entry->destructor_priv);
3395}
3396EXPORT_SYMBOL(tc_cleanup_flow_action);
3397
3398static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3399 const struct tc_action *act)
3400{
3401#ifdef CONFIG_NET_CLS_ACT
3402 entry->dev = act->ops->get_dev(act, &entry->destructor);
3403 if (!entry->dev)
3404 return;
3405 entry->destructor_priv = entry->dev;
3406#endif
3407}
3408
3409static void tcf_tunnel_encap_put_tunnel(void *priv)
3410{
3411 struct ip_tunnel_info *tunnel = priv;
3412
3413 kfree(tunnel);
3414}
3415
3416static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3417 const struct tc_action *act)
3418{
3419 entry->tunnel = tcf_tunnel_info_copy(act);
3420 if (!entry->tunnel)
3421 return -ENOMEM;
3422 entry->destructor = tcf_tunnel_encap_put_tunnel;
3423 entry->destructor_priv = entry->tunnel;
3424 return 0;
3425}
3426
3427static void tcf_sample_get_group(struct flow_action_entry *entry,
3428 const struct tc_action *act)
3429{
3430#ifdef CONFIG_NET_CLS_ACT
3431 entry->sample.psample_group =
3432 act->ops->get_psample_group(act, &entry->destructor);
3433 entry->destructor_priv = entry->sample.psample_group;
3434#endif
3435}
3436
3437int tc_setup_flow_action(struct flow_action *flow_action,
3438 const struct tcf_exts *exts, bool rtnl_held)
3439{
3440 const struct tc_action *act;
3441 int i, j, k, err = 0;
3442
3443 if (!exts)
3444 return 0;
3445
3446 if (!rtnl_held)
3447 rtnl_lock();
3448
3449 j = 0;
3450 tcf_exts_for_each_action(i, act, exts) {
3451 struct flow_action_entry *entry;
3452
3453 entry = &flow_action->entries[j];
3454 if (is_tcf_gact_ok(act)) {
3455 entry->id = FLOW_ACTION_ACCEPT;
3456 } else if (is_tcf_gact_shot(act)) {
3457 entry->id = FLOW_ACTION_DROP;
3458 } else if (is_tcf_gact_trap(act)) {
3459 entry->id = FLOW_ACTION_TRAP;
3460 } else if (is_tcf_gact_goto_chain(act)) {
3461 entry->id = FLOW_ACTION_GOTO;
3462 entry->chain_index = tcf_gact_goto_chain_index(act);
3463 } else if (is_tcf_mirred_egress_redirect(act)) {
3464 entry->id = FLOW_ACTION_REDIRECT;
3465 tcf_mirred_get_dev(entry, act);
3466 } else if (is_tcf_mirred_egress_mirror(act)) {
3467 entry->id = FLOW_ACTION_MIRRED;
3468 tcf_mirred_get_dev(entry, act);
3469 } else if (is_tcf_mirred_ingress_redirect(act)) {
3470 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3471 tcf_mirred_get_dev(entry, act);
3472 } else if (is_tcf_mirred_ingress_mirror(act)) {
3473 entry->id = FLOW_ACTION_MIRRED_INGRESS;
3474 tcf_mirred_get_dev(entry, act);
3475 } else if (is_tcf_vlan(act)) {
3476 switch (tcf_vlan_action(act)) {
3477 case TCA_VLAN_ACT_PUSH:
3478 entry->id = FLOW_ACTION_VLAN_PUSH;
3479 entry->vlan.vid = tcf_vlan_push_vid(act);
3480 entry->vlan.proto = tcf_vlan_push_proto(act);
3481 entry->vlan.prio = tcf_vlan_push_prio(act);
3482 break;
3483 case TCA_VLAN_ACT_POP:
3484 entry->id = FLOW_ACTION_VLAN_POP;
3485 break;
3486 case TCA_VLAN_ACT_MODIFY:
3487 entry->id = FLOW_ACTION_VLAN_MANGLE;
3488 entry->vlan.vid = tcf_vlan_push_vid(act);
3489 entry->vlan.proto = tcf_vlan_push_proto(act);
3490 entry->vlan.prio = tcf_vlan_push_prio(act);
3491 break;
3492 default:
3493 err = -EOPNOTSUPP;
3494 goto err_out;
3495 }
3496 } else if (is_tcf_tunnel_set(act)) {
3497 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3498 err = tcf_tunnel_encap_get_tunnel(entry, act);
3499 if (err)
3500 goto err_out;
3501 } else if (is_tcf_tunnel_release(act)) {
3502 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3503 } else if (is_tcf_pedit(act)) {
3504 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3505 switch (tcf_pedit_cmd(act, k)) {
3506 case TCA_PEDIT_KEY_EX_CMD_SET:
3507 entry->id = FLOW_ACTION_MANGLE;
3508 break;
3509 case TCA_PEDIT_KEY_EX_CMD_ADD:
3510 entry->id = FLOW_ACTION_ADD;
3511 break;
3512 default:
3513 err = -EOPNOTSUPP;
3514 goto err_out;
3515 }
3516 entry->mangle.htype = tcf_pedit_htype(act, k);
3517 entry->mangle.mask = tcf_pedit_mask(act, k);
3518 entry->mangle.val = tcf_pedit_val(act, k);
3519 entry->mangle.offset = tcf_pedit_offset(act, k);
3520 entry = &flow_action->entries[++j];
3521 }
3522 } else if (is_tcf_csum(act)) {
3523 entry->id = FLOW_ACTION_CSUM;
3524 entry->csum_flags = tcf_csum_update_flags(act);
3525 } else if (is_tcf_skbedit_mark(act)) {
3526 entry->id = FLOW_ACTION_MARK;
3527 entry->mark = tcf_skbedit_mark(act);
3528 } else if (is_tcf_sample(act)) {
3529 entry->id = FLOW_ACTION_SAMPLE;
3530 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3531 entry->sample.truncate = tcf_sample_truncate(act);
3532 entry->sample.rate = tcf_sample_rate(act);
3533 tcf_sample_get_group(entry, act);
3534 } else if (is_tcf_police(act)) {
3535 entry->id = FLOW_ACTION_POLICE;
3536 entry->police.burst = tcf_police_tcfp_burst(act);
3537 entry->police.rate_bytes_ps =
3538 tcf_police_rate_bytes_ps(act);
3539 } else if (is_tcf_ct(act)) {
3540 entry->id = FLOW_ACTION_CT;
3541 entry->ct.action = tcf_ct_action(act);
3542 entry->ct.zone = tcf_ct_zone(act);
3543 } else if (is_tcf_mpls(act)) {
3544 switch (tcf_mpls_action(act)) {
3545 case TCA_MPLS_ACT_PUSH:
3546 entry->id = FLOW_ACTION_MPLS_PUSH;
3547 entry->mpls_push.proto = tcf_mpls_proto(act);
3548 entry->mpls_push.label = tcf_mpls_label(act);
3549 entry->mpls_push.tc = tcf_mpls_tc(act);
3550 entry->mpls_push.bos = tcf_mpls_bos(act);
3551 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3552 break;
3553 case TCA_MPLS_ACT_POP:
3554 entry->id = FLOW_ACTION_MPLS_POP;
3555 entry->mpls_pop.proto = tcf_mpls_proto(act);
3556 break;
3557 case TCA_MPLS_ACT_MODIFY:
3558 entry->id = FLOW_ACTION_MPLS_MANGLE;
3559 entry->mpls_mangle.label = tcf_mpls_label(act);
3560 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3561 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3562 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3563 break;
3564 default:
3565 goto err_out;
3566 }
3567 } else if (is_tcf_skbedit_ptype(act)) {
3568 entry->id = FLOW_ACTION_PTYPE;
3569 entry->ptype = tcf_skbedit_ptype(act);
3570 } else {
3571 err = -EOPNOTSUPP;
3572 goto err_out;
3573 }
3574
3575 if (!is_tcf_pedit(act))
3576 j++;
3577 }
3578
3579err_out:
3580 if (!rtnl_held)
3581 rtnl_unlock();
3582
3583 if (err)
3584 tc_cleanup_flow_action(flow_action);
3585
3586 return err;
3587}
3588EXPORT_SYMBOL(tc_setup_flow_action);
3589
3590unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3591{
3592 unsigned int num_acts = 0;
3593 struct tc_action *act;
3594 int i;
3595
3596 tcf_exts_for_each_action(i, act, exts) {
3597 if (is_tcf_pedit(act))
3598 num_acts += tcf_pedit_nkeys(act);
3599 else
3600 num_acts++;
3601 }
3602 return num_acts;
3603}
3604EXPORT_SYMBOL(tcf_exts_num_actions);
3605
3606static __net_init int tcf_net_init(struct net *net)
3607{
3608 struct tcf_net *tn = net_generic(net, tcf_net_id);
3609
3610 spin_lock_init(&tn->idr_lock);
3611 idr_init(&tn->idr);
3612 return 0;
3613}
3614
3615static void __net_exit tcf_net_exit(struct net *net)
3616{
3617 struct tcf_net *tn = net_generic(net, tcf_net_id);
3618
3619 idr_destroy(&tn->idr);
3620}
3621
3622static struct pernet_operations tcf_net_ops = {
3623 .init = tcf_net_init,
3624 .exit = tcf_net_exit,
3625 .id = &tcf_net_id,
3626 .size = sizeof(struct tcf_net),
3627};
3628
3629static struct flow_indr_block_ing_entry block_ing_entry = {
3630 .cb = tc_indr_block_get_and_ing_cmd,
3631 .list = LIST_HEAD_INIT(block_ing_entry.list),
3632};
3633
3634static int __init tc_filter_init(void)
3635{
3636 int err;
3637
3638 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3639 if (!tc_filter_wq)
3640 return -ENOMEM;
3641
3642 err = register_pernet_subsys(&tcf_net_ops);
3643 if (err)
3644 goto err_register_pernet_subsys;
3645
3646 flow_indr_add_block_ing_cb(&block_ing_entry);
3647
3648 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3649 RTNL_FLAG_DOIT_UNLOCKED);
3650 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3651 RTNL_FLAG_DOIT_UNLOCKED);
3652 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3653 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3654 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3655 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3656 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3657 tc_dump_chain, 0);
3658
3659 return 0;
3660
3661err_register_pernet_subsys:
3662 destroy_workqueue(tc_filter_wq);
3663 return err;
3664}
3665
3666subsys_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 <linux/rhashtable.h>
26#include <net/net_namespace.h>
27#include <net/sock.h>
28#include <net/netlink.h>
29#include <net/pkt_sched.h>
30#include <net/pkt_cls.h>
31#include <net/tc_act/tc_pedit.h>
32#include <net/tc_act/tc_mirred.h>
33#include <net/tc_act/tc_vlan.h>
34#include <net/tc_act/tc_tunnel_key.h>
35#include <net/tc_act/tc_csum.h>
36#include <net/tc_act/tc_gact.h>
37#include <net/tc_act/tc_police.h>
38#include <net/tc_act/tc_sample.h>
39#include <net/tc_act/tc_skbedit.h>
40#include <net/tc_act/tc_ct.h>
41#include <net/tc_act/tc_mpls.h>
42#include <net/tc_act/tc_gate.h>
43#include <net/flow_offload.h>
44#include <net/tc_wrapper.h>
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 struct xarray tcf_exts_miss_cookies_xa;
53struct tcf_exts_miss_cookie_node {
54 const struct tcf_chain *chain;
55 const struct tcf_proto *tp;
56 const struct tcf_exts *exts;
57 u32 chain_index;
58 u32 tp_prio;
59 u32 handle;
60 u32 miss_cookie_base;
61 struct rcu_head rcu;
62};
63
64/* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
65 * action index in the exts tc actions array.
66 */
67union tcf_exts_miss_cookie {
68 struct {
69 u32 miss_cookie_base;
70 u32 act_index;
71 };
72 u64 miss_cookie;
73};
74
75#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
76static int
77tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
78 u32 handle)
79{
80 struct tcf_exts_miss_cookie_node *n;
81 static u32 next;
82 int err;
83
84 if (WARN_ON(!handle || !tp->ops->get_exts))
85 return -EINVAL;
86
87 n = kzalloc(sizeof(*n), GFP_KERNEL);
88 if (!n)
89 return -ENOMEM;
90
91 n->chain_index = tp->chain->index;
92 n->chain = tp->chain;
93 n->tp_prio = tp->prio;
94 n->tp = tp;
95 n->exts = exts;
96 n->handle = handle;
97
98 err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
99 n, xa_limit_32b, &next, GFP_KERNEL);
100 if (err)
101 goto err_xa_alloc;
102
103 exts->miss_cookie_node = n;
104 return 0;
105
106err_xa_alloc:
107 kfree(n);
108 return err;
109}
110
111static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
112{
113 struct tcf_exts_miss_cookie_node *n;
114
115 if (!exts->miss_cookie_node)
116 return;
117
118 n = exts->miss_cookie_node;
119 xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
120 kfree_rcu(n, rcu);
121}
122
123static struct tcf_exts_miss_cookie_node *
124tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
125{
126 union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
127
128 *act_index = mc.act_index;
129 return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
130}
131#else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
132static int
133tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
134 u32 handle)
135{
136 return 0;
137}
138
139static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
140{
141}
142#endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
143
144static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
145{
146 union tcf_exts_miss_cookie mc = { .act_index = act_index, };
147
148 if (!miss_cookie_base)
149 return 0;
150
151 mc.miss_cookie_base = miss_cookie_base;
152 return mc.miss_cookie;
153}
154
155#ifdef CONFIG_NET_CLS_ACT
156DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
157EXPORT_SYMBOL(tc_skb_ext_tc);
158
159void tc_skb_ext_tc_enable(void)
160{
161 static_branch_inc(&tc_skb_ext_tc);
162}
163EXPORT_SYMBOL(tc_skb_ext_tc_enable);
164
165void tc_skb_ext_tc_disable(void)
166{
167 static_branch_dec(&tc_skb_ext_tc);
168}
169EXPORT_SYMBOL(tc_skb_ext_tc_disable);
170#endif
171
172static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
173{
174 return jhash_3words(tp->chain->index, tp->prio,
175 (__force __u32)tp->protocol, 0);
176}
177
178static void tcf_proto_signal_destroying(struct tcf_chain *chain,
179 struct tcf_proto *tp)
180{
181 struct tcf_block *block = chain->block;
182
183 mutex_lock(&block->proto_destroy_lock);
184 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
185 destroy_obj_hashfn(tp));
186 mutex_unlock(&block->proto_destroy_lock);
187}
188
189static bool tcf_proto_cmp(const struct tcf_proto *tp1,
190 const struct tcf_proto *tp2)
191{
192 return tp1->chain->index == tp2->chain->index &&
193 tp1->prio == tp2->prio &&
194 tp1->protocol == tp2->protocol;
195}
196
197static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
198 struct tcf_proto *tp)
199{
200 u32 hash = destroy_obj_hashfn(tp);
201 struct tcf_proto *iter;
202 bool found = false;
203
204 rcu_read_lock();
205 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
206 destroy_ht_node, hash) {
207 if (tcf_proto_cmp(tp, iter)) {
208 found = true;
209 break;
210 }
211 }
212 rcu_read_unlock();
213
214 return found;
215}
216
217static void
218tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
219{
220 struct tcf_block *block = chain->block;
221
222 mutex_lock(&block->proto_destroy_lock);
223 if (hash_hashed(&tp->destroy_ht_node))
224 hash_del_rcu(&tp->destroy_ht_node);
225 mutex_unlock(&block->proto_destroy_lock);
226}
227
228/* Find classifier type by string name */
229
230static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
231{
232 const struct tcf_proto_ops *t, *res = NULL;
233
234 if (kind) {
235 read_lock(&cls_mod_lock);
236 list_for_each_entry(t, &tcf_proto_base, head) {
237 if (strcmp(kind, t->kind) == 0) {
238 if (try_module_get(t->owner))
239 res = t;
240 break;
241 }
242 }
243 read_unlock(&cls_mod_lock);
244 }
245 return res;
246}
247
248static const struct tcf_proto_ops *
249tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
250 struct netlink_ext_ack *extack)
251{
252 const struct tcf_proto_ops *ops;
253
254 ops = __tcf_proto_lookup_ops(kind);
255 if (ops)
256 return ops;
257#ifdef CONFIG_MODULES
258 if (rtnl_held)
259 rtnl_unlock();
260 request_module("cls_%s", kind);
261 if (rtnl_held)
262 rtnl_lock();
263 ops = __tcf_proto_lookup_ops(kind);
264 /* We dropped the RTNL semaphore in order to perform
265 * the module load. So, even if we succeeded in loading
266 * the module we have to replay the request. We indicate
267 * this using -EAGAIN.
268 */
269 if (ops) {
270 module_put(ops->owner);
271 return ERR_PTR(-EAGAIN);
272 }
273#endif
274 NL_SET_ERR_MSG(extack, "TC classifier not found");
275 return ERR_PTR(-ENOENT);
276}
277
278/* Register(unregister) new classifier type */
279
280int register_tcf_proto_ops(struct tcf_proto_ops *ops)
281{
282 struct tcf_proto_ops *t;
283 int rc = -EEXIST;
284
285 write_lock(&cls_mod_lock);
286 list_for_each_entry(t, &tcf_proto_base, head)
287 if (!strcmp(ops->kind, t->kind))
288 goto out;
289
290 list_add_tail(&ops->head, &tcf_proto_base);
291 rc = 0;
292out:
293 write_unlock(&cls_mod_lock);
294 return rc;
295}
296EXPORT_SYMBOL(register_tcf_proto_ops);
297
298static struct workqueue_struct *tc_filter_wq;
299
300void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
301{
302 struct tcf_proto_ops *t;
303 int rc = -ENOENT;
304
305 /* Wait for outstanding call_rcu()s, if any, from a
306 * tcf_proto_ops's destroy() handler.
307 */
308 rcu_barrier();
309 flush_workqueue(tc_filter_wq);
310
311 write_lock(&cls_mod_lock);
312 list_for_each_entry(t, &tcf_proto_base, head) {
313 if (t == ops) {
314 list_del(&t->head);
315 rc = 0;
316 break;
317 }
318 }
319 write_unlock(&cls_mod_lock);
320
321 WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
322}
323EXPORT_SYMBOL(unregister_tcf_proto_ops);
324
325bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
326{
327 INIT_RCU_WORK(rwork, func);
328 return queue_rcu_work(tc_filter_wq, rwork);
329}
330EXPORT_SYMBOL(tcf_queue_work);
331
332/* Select new prio value from the range, managed by kernel. */
333
334static inline u32 tcf_auto_prio(struct tcf_proto *tp)
335{
336 u32 first = TC_H_MAKE(0xC0000000U, 0U);
337
338 if (tp)
339 first = tp->prio - 1;
340
341 return TC_H_MAJ(first);
342}
343
344static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
345{
346 if (kind)
347 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
348 memset(name, 0, IFNAMSIZ);
349 return false;
350}
351
352static bool tcf_proto_is_unlocked(const char *kind)
353{
354 const struct tcf_proto_ops *ops;
355 bool ret;
356
357 if (strlen(kind) == 0)
358 return false;
359
360 ops = tcf_proto_lookup_ops(kind, false, NULL);
361 /* On error return false to take rtnl lock. Proto lookup/create
362 * functions will perform lookup again and properly handle errors.
363 */
364 if (IS_ERR(ops))
365 return false;
366
367 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
368 module_put(ops->owner);
369 return ret;
370}
371
372static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
373 u32 prio, struct tcf_chain *chain,
374 bool rtnl_held,
375 struct netlink_ext_ack *extack)
376{
377 struct tcf_proto *tp;
378 int err;
379
380 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
381 if (!tp)
382 return ERR_PTR(-ENOBUFS);
383
384 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
385 if (IS_ERR(tp->ops)) {
386 err = PTR_ERR(tp->ops);
387 goto errout;
388 }
389 tp->classify = tp->ops->classify;
390 tp->protocol = protocol;
391 tp->prio = prio;
392 tp->chain = chain;
393 spin_lock_init(&tp->lock);
394 refcount_set(&tp->refcnt, 1);
395
396 err = tp->ops->init(tp);
397 if (err) {
398 module_put(tp->ops->owner);
399 goto errout;
400 }
401 return tp;
402
403errout:
404 kfree(tp);
405 return ERR_PTR(err);
406}
407
408static void tcf_proto_get(struct tcf_proto *tp)
409{
410 refcount_inc(&tp->refcnt);
411}
412
413static void tcf_chain_put(struct tcf_chain *chain);
414
415static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
416 bool sig_destroy, struct netlink_ext_ack *extack)
417{
418 tp->ops->destroy(tp, rtnl_held, extack);
419 if (sig_destroy)
420 tcf_proto_signal_destroyed(tp->chain, tp);
421 tcf_chain_put(tp->chain);
422 module_put(tp->ops->owner);
423 kfree_rcu(tp, rcu);
424}
425
426static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
427 struct netlink_ext_ack *extack)
428{
429 if (refcount_dec_and_test(&tp->refcnt))
430 tcf_proto_destroy(tp, rtnl_held, true, extack);
431}
432
433static bool tcf_proto_check_delete(struct tcf_proto *tp)
434{
435 if (tp->ops->delete_empty)
436 return tp->ops->delete_empty(tp);
437
438 tp->deleting = true;
439 return tp->deleting;
440}
441
442static void tcf_proto_mark_delete(struct tcf_proto *tp)
443{
444 spin_lock(&tp->lock);
445 tp->deleting = true;
446 spin_unlock(&tp->lock);
447}
448
449static bool tcf_proto_is_deleting(struct tcf_proto *tp)
450{
451 bool deleting;
452
453 spin_lock(&tp->lock);
454 deleting = tp->deleting;
455 spin_unlock(&tp->lock);
456
457 return deleting;
458}
459
460#define ASSERT_BLOCK_LOCKED(block) \
461 lockdep_assert_held(&(block)->lock)
462
463struct tcf_filter_chain_list_item {
464 struct list_head list;
465 tcf_chain_head_change_t *chain_head_change;
466 void *chain_head_change_priv;
467};
468
469static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
470 u32 chain_index)
471{
472 struct tcf_chain *chain;
473
474 ASSERT_BLOCK_LOCKED(block);
475
476 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
477 if (!chain)
478 return NULL;
479 list_add_tail_rcu(&chain->list, &block->chain_list);
480 mutex_init(&chain->filter_chain_lock);
481 chain->block = block;
482 chain->index = chain_index;
483 chain->refcnt = 1;
484 if (!chain->index)
485 block->chain0.chain = chain;
486 return chain;
487}
488
489static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
490 struct tcf_proto *tp_head)
491{
492 if (item->chain_head_change)
493 item->chain_head_change(tp_head, item->chain_head_change_priv);
494}
495
496static void tcf_chain0_head_change(struct tcf_chain *chain,
497 struct tcf_proto *tp_head)
498{
499 struct tcf_filter_chain_list_item *item;
500 struct tcf_block *block = chain->block;
501
502 if (chain->index)
503 return;
504
505 mutex_lock(&block->lock);
506 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
507 tcf_chain_head_change_item(item, tp_head);
508 mutex_unlock(&block->lock);
509}
510
511/* Returns true if block can be safely freed. */
512
513static bool tcf_chain_detach(struct tcf_chain *chain)
514{
515 struct tcf_block *block = chain->block;
516
517 ASSERT_BLOCK_LOCKED(block);
518
519 list_del_rcu(&chain->list);
520 if (!chain->index)
521 block->chain0.chain = NULL;
522
523 if (list_empty(&block->chain_list) &&
524 refcount_read(&block->refcnt) == 0)
525 return true;
526
527 return false;
528}
529
530static void tcf_block_destroy(struct tcf_block *block)
531{
532 mutex_destroy(&block->lock);
533 mutex_destroy(&block->proto_destroy_lock);
534 xa_destroy(&block->ports);
535 kfree_rcu(block, rcu);
536}
537
538static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
539{
540 struct tcf_block *block = chain->block;
541
542 mutex_destroy(&chain->filter_chain_lock);
543 kfree_rcu(chain, rcu);
544 if (free_block)
545 tcf_block_destroy(block);
546}
547
548static void tcf_chain_hold(struct tcf_chain *chain)
549{
550 ASSERT_BLOCK_LOCKED(chain->block);
551
552 ++chain->refcnt;
553}
554
555static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
556{
557 ASSERT_BLOCK_LOCKED(chain->block);
558
559 /* In case all the references are action references, this
560 * chain should not be shown to the user.
561 */
562 return chain->refcnt == chain->action_refcnt;
563}
564
565static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
566 u32 chain_index)
567{
568 struct tcf_chain *chain;
569
570 ASSERT_BLOCK_LOCKED(block);
571
572 list_for_each_entry(chain, &block->chain_list, list) {
573 if (chain->index == chain_index)
574 return chain;
575 }
576 return NULL;
577}
578
579#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
580static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
581 u32 chain_index)
582{
583 struct tcf_chain *chain;
584
585 list_for_each_entry_rcu(chain, &block->chain_list, list) {
586 if (chain->index == chain_index)
587 return chain;
588 }
589 return NULL;
590}
591#endif
592
593static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
594 u32 seq, u16 flags, int event, bool unicast,
595 struct netlink_ext_ack *extack);
596
597static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
598 u32 chain_index, bool create,
599 bool by_act)
600{
601 struct tcf_chain *chain = NULL;
602 bool is_first_reference;
603
604 mutex_lock(&block->lock);
605 chain = tcf_chain_lookup(block, chain_index);
606 if (chain) {
607 tcf_chain_hold(chain);
608 } else {
609 if (!create)
610 goto errout;
611 chain = tcf_chain_create(block, chain_index);
612 if (!chain)
613 goto errout;
614 }
615
616 if (by_act)
617 ++chain->action_refcnt;
618 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
619 mutex_unlock(&block->lock);
620
621 /* Send notification only in case we got the first
622 * non-action reference. Until then, the chain acts only as
623 * a placeholder for actions pointing to it and user ought
624 * not know about them.
625 */
626 if (is_first_reference && !by_act)
627 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
628 RTM_NEWCHAIN, false, NULL);
629
630 return chain;
631
632errout:
633 mutex_unlock(&block->lock);
634 return chain;
635}
636
637static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
638 bool create)
639{
640 return __tcf_chain_get(block, chain_index, create, false);
641}
642
643struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
644{
645 return __tcf_chain_get(block, chain_index, true, true);
646}
647EXPORT_SYMBOL(tcf_chain_get_by_act);
648
649static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
650 void *tmplt_priv);
651static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
652 void *tmplt_priv, u32 chain_index,
653 struct tcf_block *block, struct sk_buff *oskb,
654 u32 seq, u16 flags);
655
656static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
657 bool explicitly_created)
658{
659 struct tcf_block *block = chain->block;
660 const struct tcf_proto_ops *tmplt_ops;
661 unsigned int refcnt, non_act_refcnt;
662 bool free_block = false;
663 void *tmplt_priv;
664
665 mutex_lock(&block->lock);
666 if (explicitly_created) {
667 if (!chain->explicitly_created) {
668 mutex_unlock(&block->lock);
669 return;
670 }
671 chain->explicitly_created = false;
672 }
673
674 if (by_act)
675 chain->action_refcnt--;
676
677 /* tc_chain_notify_delete can't be called while holding block lock.
678 * However, when block is unlocked chain can be changed concurrently, so
679 * save these to temporary variables.
680 */
681 refcnt = --chain->refcnt;
682 non_act_refcnt = refcnt - chain->action_refcnt;
683 tmplt_ops = chain->tmplt_ops;
684 tmplt_priv = chain->tmplt_priv;
685
686 if (non_act_refcnt == chain->explicitly_created && !by_act) {
687 if (non_act_refcnt == 0)
688 tc_chain_notify_delete(tmplt_ops, tmplt_priv,
689 chain->index, block, NULL, 0, 0);
690 /* Last reference to chain, no need to lock. */
691 chain->flushing = false;
692 }
693
694 if (refcnt == 0)
695 free_block = tcf_chain_detach(chain);
696 mutex_unlock(&block->lock);
697
698 if (refcnt == 0) {
699 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
700 tcf_chain_destroy(chain, free_block);
701 }
702}
703
704static void tcf_chain_put(struct tcf_chain *chain)
705{
706 __tcf_chain_put(chain, false, false);
707}
708
709void tcf_chain_put_by_act(struct tcf_chain *chain)
710{
711 __tcf_chain_put(chain, true, false);
712}
713EXPORT_SYMBOL(tcf_chain_put_by_act);
714
715static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
716{
717 __tcf_chain_put(chain, false, true);
718}
719
720static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
721{
722 struct tcf_proto *tp, *tp_next;
723
724 mutex_lock(&chain->filter_chain_lock);
725 tp = tcf_chain_dereference(chain->filter_chain, chain);
726 while (tp) {
727 tp_next = rcu_dereference_protected(tp->next, 1);
728 tcf_proto_signal_destroying(chain, tp);
729 tp = tp_next;
730 }
731 tp = tcf_chain_dereference(chain->filter_chain, chain);
732 RCU_INIT_POINTER(chain->filter_chain, NULL);
733 tcf_chain0_head_change(chain, NULL);
734 chain->flushing = true;
735 mutex_unlock(&chain->filter_chain_lock);
736
737 while (tp) {
738 tp_next = rcu_dereference_protected(tp->next, 1);
739 tcf_proto_put(tp, rtnl_held, NULL);
740 tp = tp_next;
741 }
742}
743
744static int tcf_block_setup(struct tcf_block *block,
745 struct flow_block_offload *bo);
746
747static void tcf_block_offload_init(struct flow_block_offload *bo,
748 struct net_device *dev, struct Qdisc *sch,
749 enum flow_block_command command,
750 enum flow_block_binder_type binder_type,
751 struct flow_block *flow_block,
752 bool shared, struct netlink_ext_ack *extack)
753{
754 bo->net = dev_net(dev);
755 bo->command = command;
756 bo->binder_type = binder_type;
757 bo->block = flow_block;
758 bo->block_shared = shared;
759 bo->extack = extack;
760 bo->sch = sch;
761 bo->cb_list_head = &flow_block->cb_list;
762 INIT_LIST_HEAD(&bo->cb_list);
763}
764
765static void tcf_block_unbind(struct tcf_block *block,
766 struct flow_block_offload *bo);
767
768static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
769{
770 struct tcf_block *block = block_cb->indr.data;
771 struct net_device *dev = block_cb->indr.dev;
772 struct Qdisc *sch = block_cb->indr.sch;
773 struct netlink_ext_ack extack = {};
774 struct flow_block_offload bo = {};
775
776 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
777 block_cb->indr.binder_type,
778 &block->flow_block, tcf_block_shared(block),
779 &extack);
780 rtnl_lock();
781 down_write(&block->cb_lock);
782 list_del(&block_cb->driver_list);
783 list_move(&block_cb->list, &bo.cb_list);
784 tcf_block_unbind(block, &bo);
785 up_write(&block->cb_lock);
786 rtnl_unlock();
787}
788
789static bool tcf_block_offload_in_use(struct tcf_block *block)
790{
791 return atomic_read(&block->offloadcnt);
792}
793
794static int tcf_block_offload_cmd(struct tcf_block *block,
795 struct net_device *dev, struct Qdisc *sch,
796 struct tcf_block_ext_info *ei,
797 enum flow_block_command command,
798 struct netlink_ext_ack *extack)
799{
800 struct flow_block_offload bo = {};
801
802 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
803 &block->flow_block, tcf_block_shared(block),
804 extack);
805
806 if (dev->netdev_ops->ndo_setup_tc) {
807 int err;
808
809 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
810 if (err < 0) {
811 if (err != -EOPNOTSUPP)
812 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
813 return err;
814 }
815
816 return tcf_block_setup(block, &bo);
817 }
818
819 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
820 tc_block_indr_cleanup);
821 tcf_block_setup(block, &bo);
822
823 return -EOPNOTSUPP;
824}
825
826static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
827 struct tcf_block_ext_info *ei,
828 struct netlink_ext_ack *extack)
829{
830 struct net_device *dev = q->dev_queue->dev;
831 int err;
832
833 down_write(&block->cb_lock);
834
835 /* If tc offload feature is disabled and the block we try to bind
836 * to already has some offloaded filters, forbid to bind.
837 */
838 if (dev->netdev_ops->ndo_setup_tc &&
839 !tc_can_offload(dev) &&
840 tcf_block_offload_in_use(block)) {
841 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
842 err = -EOPNOTSUPP;
843 goto err_unlock;
844 }
845
846 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
847 if (err == -EOPNOTSUPP)
848 goto no_offload_dev_inc;
849 if (err)
850 goto err_unlock;
851
852 up_write(&block->cb_lock);
853 return 0;
854
855no_offload_dev_inc:
856 if (tcf_block_offload_in_use(block))
857 goto err_unlock;
858
859 err = 0;
860 block->nooffloaddevcnt++;
861err_unlock:
862 up_write(&block->cb_lock);
863 return err;
864}
865
866static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
867 struct tcf_block_ext_info *ei)
868{
869 struct net_device *dev = q->dev_queue->dev;
870 int err;
871
872 down_write(&block->cb_lock);
873 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
874 if (err == -EOPNOTSUPP)
875 goto no_offload_dev_dec;
876 up_write(&block->cb_lock);
877 return;
878
879no_offload_dev_dec:
880 WARN_ON(block->nooffloaddevcnt-- == 0);
881 up_write(&block->cb_lock);
882}
883
884static int
885tcf_chain0_head_change_cb_add(struct tcf_block *block,
886 struct tcf_block_ext_info *ei,
887 struct netlink_ext_ack *extack)
888{
889 struct tcf_filter_chain_list_item *item;
890 struct tcf_chain *chain0;
891
892 item = kmalloc(sizeof(*item), GFP_KERNEL);
893 if (!item) {
894 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
895 return -ENOMEM;
896 }
897 item->chain_head_change = ei->chain_head_change;
898 item->chain_head_change_priv = ei->chain_head_change_priv;
899
900 mutex_lock(&block->lock);
901 chain0 = block->chain0.chain;
902 if (chain0)
903 tcf_chain_hold(chain0);
904 else
905 list_add(&item->list, &block->chain0.filter_chain_list);
906 mutex_unlock(&block->lock);
907
908 if (chain0) {
909 struct tcf_proto *tp_head;
910
911 mutex_lock(&chain0->filter_chain_lock);
912
913 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
914 if (tp_head)
915 tcf_chain_head_change_item(item, tp_head);
916
917 mutex_lock(&block->lock);
918 list_add(&item->list, &block->chain0.filter_chain_list);
919 mutex_unlock(&block->lock);
920
921 mutex_unlock(&chain0->filter_chain_lock);
922 tcf_chain_put(chain0);
923 }
924
925 return 0;
926}
927
928static void
929tcf_chain0_head_change_cb_del(struct tcf_block *block,
930 struct tcf_block_ext_info *ei)
931{
932 struct tcf_filter_chain_list_item *item;
933
934 mutex_lock(&block->lock);
935 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
936 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
937 (item->chain_head_change == ei->chain_head_change &&
938 item->chain_head_change_priv == ei->chain_head_change_priv)) {
939 if (block->chain0.chain)
940 tcf_chain_head_change_item(item, NULL);
941 list_del(&item->list);
942 mutex_unlock(&block->lock);
943
944 kfree(item);
945 return;
946 }
947 }
948 mutex_unlock(&block->lock);
949 WARN_ON(1);
950}
951
952struct tcf_net {
953 spinlock_t idr_lock; /* Protects idr */
954 struct idr idr;
955};
956
957static unsigned int tcf_net_id;
958
959static int tcf_block_insert(struct tcf_block *block, struct net *net,
960 struct netlink_ext_ack *extack)
961{
962 struct tcf_net *tn = net_generic(net, tcf_net_id);
963 int err;
964
965 idr_preload(GFP_KERNEL);
966 spin_lock(&tn->idr_lock);
967 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
968 GFP_NOWAIT);
969 spin_unlock(&tn->idr_lock);
970 idr_preload_end();
971
972 return err;
973}
974
975static void tcf_block_remove(struct tcf_block *block, struct net *net)
976{
977 struct tcf_net *tn = net_generic(net, tcf_net_id);
978
979 spin_lock(&tn->idr_lock);
980 idr_remove(&tn->idr, block->index);
981 spin_unlock(&tn->idr_lock);
982}
983
984static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
985 u32 block_index,
986 struct netlink_ext_ack *extack)
987{
988 struct tcf_block *block;
989
990 block = kzalloc(sizeof(*block), GFP_KERNEL);
991 if (!block) {
992 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
993 return ERR_PTR(-ENOMEM);
994 }
995 mutex_init(&block->lock);
996 mutex_init(&block->proto_destroy_lock);
997 init_rwsem(&block->cb_lock);
998 flow_block_init(&block->flow_block);
999 INIT_LIST_HEAD(&block->chain_list);
1000 INIT_LIST_HEAD(&block->owner_list);
1001 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1002
1003 refcount_set(&block->refcnt, 1);
1004 block->net = net;
1005 block->index = block_index;
1006 xa_init(&block->ports);
1007
1008 /* Don't store q pointer for blocks which are shared */
1009 if (!tcf_block_shared(block))
1010 block->q = q;
1011 return block;
1012}
1013
1014struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1015{
1016 struct tcf_net *tn = net_generic(net, tcf_net_id);
1017
1018 return idr_find(&tn->idr, block_index);
1019}
1020EXPORT_SYMBOL(tcf_block_lookup);
1021
1022static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1023{
1024 struct tcf_block *block;
1025
1026 rcu_read_lock();
1027 block = tcf_block_lookup(net, block_index);
1028 if (block && !refcount_inc_not_zero(&block->refcnt))
1029 block = NULL;
1030 rcu_read_unlock();
1031
1032 return block;
1033}
1034
1035static struct tcf_chain *
1036__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1037{
1038 mutex_lock(&block->lock);
1039 if (chain)
1040 chain = list_is_last(&chain->list, &block->chain_list) ?
1041 NULL : list_next_entry(chain, list);
1042 else
1043 chain = list_first_entry_or_null(&block->chain_list,
1044 struct tcf_chain, list);
1045
1046 /* skip all action-only chains */
1047 while (chain && tcf_chain_held_by_acts_only(chain))
1048 chain = list_is_last(&chain->list, &block->chain_list) ?
1049 NULL : list_next_entry(chain, list);
1050
1051 if (chain)
1052 tcf_chain_hold(chain);
1053 mutex_unlock(&block->lock);
1054
1055 return chain;
1056}
1057
1058/* Function to be used by all clients that want to iterate over all chains on
1059 * block. It properly obtains block->lock and takes reference to chain before
1060 * returning it. Users of this function must be tolerant to concurrent chain
1061 * insertion/deletion or ensure that no concurrent chain modification is
1062 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1063 * consistent dump because rtnl lock is released each time skb is filled with
1064 * data and sent to user-space.
1065 */
1066
1067struct tcf_chain *
1068tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1069{
1070 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1071
1072 if (chain)
1073 tcf_chain_put(chain);
1074
1075 return chain_next;
1076}
1077EXPORT_SYMBOL(tcf_get_next_chain);
1078
1079static struct tcf_proto *
1080__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1081{
1082 u32 prio = 0;
1083
1084 ASSERT_RTNL();
1085 mutex_lock(&chain->filter_chain_lock);
1086
1087 if (!tp) {
1088 tp = tcf_chain_dereference(chain->filter_chain, chain);
1089 } else if (tcf_proto_is_deleting(tp)) {
1090 /* 'deleting' flag is set and chain->filter_chain_lock was
1091 * unlocked, which means next pointer could be invalid. Restart
1092 * search.
1093 */
1094 prio = tp->prio + 1;
1095 tp = tcf_chain_dereference(chain->filter_chain, chain);
1096
1097 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1098 if (!tp->deleting && tp->prio >= prio)
1099 break;
1100 } else {
1101 tp = tcf_chain_dereference(tp->next, chain);
1102 }
1103
1104 if (tp)
1105 tcf_proto_get(tp);
1106
1107 mutex_unlock(&chain->filter_chain_lock);
1108
1109 return tp;
1110}
1111
1112/* Function to be used by all clients that want to iterate over all tp's on
1113 * chain. Users of this function must be tolerant to concurrent tp
1114 * insertion/deletion or ensure that no concurrent chain modification is
1115 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1116 * consistent dump because rtnl lock is released each time skb is filled with
1117 * data and sent to user-space.
1118 */
1119
1120struct tcf_proto *
1121tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1122{
1123 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1124
1125 if (tp)
1126 tcf_proto_put(tp, true, NULL);
1127
1128 return tp_next;
1129}
1130EXPORT_SYMBOL(tcf_get_next_proto);
1131
1132static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1133{
1134 struct tcf_chain *chain;
1135
1136 /* Last reference to block. At this point chains cannot be added or
1137 * removed concurrently.
1138 */
1139 for (chain = tcf_get_next_chain(block, NULL);
1140 chain;
1141 chain = tcf_get_next_chain(block, chain)) {
1142 tcf_chain_put_explicitly_created(chain);
1143 tcf_chain_flush(chain, rtnl_held);
1144 }
1145}
1146
1147/* Lookup Qdisc and increments its reference counter.
1148 * Set parent, if necessary.
1149 */
1150
1151static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1152 u32 *parent, int ifindex, bool rtnl_held,
1153 struct netlink_ext_ack *extack)
1154{
1155 const struct Qdisc_class_ops *cops;
1156 struct net_device *dev;
1157 int err = 0;
1158
1159 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1160 return 0;
1161
1162 rcu_read_lock();
1163
1164 /* Find link */
1165 dev = dev_get_by_index_rcu(net, ifindex);
1166 if (!dev) {
1167 rcu_read_unlock();
1168 return -ENODEV;
1169 }
1170
1171 /* Find qdisc */
1172 if (!*parent) {
1173 *q = rcu_dereference(dev->qdisc);
1174 *parent = (*q)->handle;
1175 } else {
1176 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1177 if (!*q) {
1178 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1179 err = -EINVAL;
1180 goto errout_rcu;
1181 }
1182 }
1183
1184 *q = qdisc_refcount_inc_nz(*q);
1185 if (!*q) {
1186 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1187 err = -EINVAL;
1188 goto errout_rcu;
1189 }
1190
1191 /* Is it classful? */
1192 cops = (*q)->ops->cl_ops;
1193 if (!cops) {
1194 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1195 err = -EINVAL;
1196 goto errout_qdisc;
1197 }
1198
1199 if (!cops->tcf_block) {
1200 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1201 err = -EOPNOTSUPP;
1202 goto errout_qdisc;
1203 }
1204
1205errout_rcu:
1206 /* At this point we know that qdisc is not noop_qdisc,
1207 * which means that qdisc holds a reference to net_device
1208 * and we hold a reference to qdisc, so it is safe to release
1209 * rcu read lock.
1210 */
1211 rcu_read_unlock();
1212 return err;
1213
1214errout_qdisc:
1215 rcu_read_unlock();
1216
1217 if (rtnl_held)
1218 qdisc_put(*q);
1219 else
1220 qdisc_put_unlocked(*q);
1221 *q = NULL;
1222
1223 return err;
1224}
1225
1226static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1227 int ifindex, struct netlink_ext_ack *extack)
1228{
1229 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1230 return 0;
1231
1232 /* Do we search for filter, attached to class? */
1233 if (TC_H_MIN(parent)) {
1234 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1235
1236 *cl = cops->find(q, parent);
1237 if (*cl == 0) {
1238 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1239 return -ENOENT;
1240 }
1241 }
1242
1243 return 0;
1244}
1245
1246static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1247 unsigned long cl, int ifindex,
1248 u32 block_index,
1249 struct netlink_ext_ack *extack)
1250{
1251 struct tcf_block *block;
1252
1253 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1254 block = tcf_block_refcnt_get(net, block_index);
1255 if (!block) {
1256 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1257 return ERR_PTR(-EINVAL);
1258 }
1259 } else {
1260 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1261
1262 block = cops->tcf_block(q, cl, extack);
1263 if (!block)
1264 return ERR_PTR(-EINVAL);
1265
1266 if (tcf_block_shared(block)) {
1267 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1268 return ERR_PTR(-EOPNOTSUPP);
1269 }
1270
1271 /* Always take reference to block in order to support execution
1272 * of rules update path of cls API without rtnl lock. Caller
1273 * must release block when it is finished using it. 'if' block
1274 * of this conditional obtain reference to block by calling
1275 * tcf_block_refcnt_get().
1276 */
1277 refcount_inc(&block->refcnt);
1278 }
1279
1280 return block;
1281}
1282
1283static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1284 struct tcf_block_ext_info *ei, bool rtnl_held)
1285{
1286 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1287 /* Flushing/putting all chains will cause the block to be
1288 * deallocated when last chain is freed. However, if chain_list
1289 * is empty, block has to be manually deallocated. After block
1290 * reference counter reached 0, it is no longer possible to
1291 * increment it or add new chains to block.
1292 */
1293 bool free_block = list_empty(&block->chain_list);
1294
1295 mutex_unlock(&block->lock);
1296 if (tcf_block_shared(block))
1297 tcf_block_remove(block, block->net);
1298
1299 if (q)
1300 tcf_block_offload_unbind(block, q, ei);
1301
1302 if (free_block)
1303 tcf_block_destroy(block);
1304 else
1305 tcf_block_flush_all_chains(block, rtnl_held);
1306 } else if (q) {
1307 tcf_block_offload_unbind(block, q, ei);
1308 }
1309}
1310
1311static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1312{
1313 __tcf_block_put(block, NULL, NULL, rtnl_held);
1314}
1315
1316/* Find tcf block.
1317 * Set q, parent, cl when appropriate.
1318 */
1319
1320static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1321 u32 *parent, unsigned long *cl,
1322 int ifindex, u32 block_index,
1323 struct netlink_ext_ack *extack)
1324{
1325 struct tcf_block *block;
1326 int err = 0;
1327
1328 ASSERT_RTNL();
1329
1330 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1331 if (err)
1332 goto errout;
1333
1334 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1335 if (err)
1336 goto errout_qdisc;
1337
1338 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1339 if (IS_ERR(block)) {
1340 err = PTR_ERR(block);
1341 goto errout_qdisc;
1342 }
1343
1344 return block;
1345
1346errout_qdisc:
1347 if (*q)
1348 qdisc_put(*q);
1349errout:
1350 *q = NULL;
1351 return ERR_PTR(err);
1352}
1353
1354static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1355 bool rtnl_held)
1356{
1357 if (!IS_ERR_OR_NULL(block))
1358 tcf_block_refcnt_put(block, rtnl_held);
1359
1360 if (q) {
1361 if (rtnl_held)
1362 qdisc_put(q);
1363 else
1364 qdisc_put_unlocked(q);
1365 }
1366}
1367
1368struct tcf_block_owner_item {
1369 struct list_head list;
1370 struct Qdisc *q;
1371 enum flow_block_binder_type binder_type;
1372};
1373
1374static void
1375tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1376 struct Qdisc *q,
1377 enum flow_block_binder_type binder_type)
1378{
1379 if (block->keep_dst &&
1380 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1381 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1382 netif_keep_dst(qdisc_dev(q));
1383}
1384
1385void tcf_block_netif_keep_dst(struct tcf_block *block)
1386{
1387 struct tcf_block_owner_item *item;
1388
1389 block->keep_dst = true;
1390 list_for_each_entry(item, &block->owner_list, list)
1391 tcf_block_owner_netif_keep_dst(block, item->q,
1392 item->binder_type);
1393}
1394EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1395
1396static int tcf_block_owner_add(struct tcf_block *block,
1397 struct Qdisc *q,
1398 enum flow_block_binder_type binder_type)
1399{
1400 struct tcf_block_owner_item *item;
1401
1402 item = kmalloc(sizeof(*item), GFP_KERNEL);
1403 if (!item)
1404 return -ENOMEM;
1405 item->q = q;
1406 item->binder_type = binder_type;
1407 list_add(&item->list, &block->owner_list);
1408 return 0;
1409}
1410
1411static void tcf_block_owner_del(struct tcf_block *block,
1412 struct Qdisc *q,
1413 enum flow_block_binder_type binder_type)
1414{
1415 struct tcf_block_owner_item *item;
1416
1417 list_for_each_entry(item, &block->owner_list, list) {
1418 if (item->q == q && item->binder_type == binder_type) {
1419 list_del(&item->list);
1420 kfree(item);
1421 return;
1422 }
1423 }
1424 WARN_ON(1);
1425}
1426
1427static bool tcf_block_tracks_dev(struct tcf_block *block,
1428 struct tcf_block_ext_info *ei)
1429{
1430 return tcf_block_shared(block) &&
1431 (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1432 ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1433}
1434
1435int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1436 struct tcf_block_ext_info *ei,
1437 struct netlink_ext_ack *extack)
1438{
1439 struct net_device *dev = qdisc_dev(q);
1440 struct net *net = qdisc_net(q);
1441 struct tcf_block *block = NULL;
1442 int err;
1443
1444 if (ei->block_index)
1445 /* block_index not 0 means the shared block is requested */
1446 block = tcf_block_refcnt_get(net, ei->block_index);
1447
1448 if (!block) {
1449 block = tcf_block_create(net, q, ei->block_index, extack);
1450 if (IS_ERR(block))
1451 return PTR_ERR(block);
1452 if (tcf_block_shared(block)) {
1453 err = tcf_block_insert(block, net, extack);
1454 if (err)
1455 goto err_block_insert;
1456 }
1457 }
1458
1459 err = tcf_block_owner_add(block, q, ei->binder_type);
1460 if (err)
1461 goto err_block_owner_add;
1462
1463 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1464
1465 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1466 if (err)
1467 goto err_chain0_head_change_cb_add;
1468
1469 err = tcf_block_offload_bind(block, q, ei, extack);
1470 if (err)
1471 goto err_block_offload_bind;
1472
1473 if (tcf_block_tracks_dev(block, ei)) {
1474 err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1475 if (err) {
1476 NL_SET_ERR_MSG(extack, "block dev insert failed");
1477 goto err_dev_insert;
1478 }
1479 }
1480
1481 *p_block = block;
1482 return 0;
1483
1484err_dev_insert:
1485err_block_offload_bind:
1486 tcf_chain0_head_change_cb_del(block, ei);
1487err_chain0_head_change_cb_add:
1488 tcf_block_owner_del(block, q, ei->binder_type);
1489err_block_owner_add:
1490err_block_insert:
1491 tcf_block_refcnt_put(block, true);
1492 return err;
1493}
1494EXPORT_SYMBOL(tcf_block_get_ext);
1495
1496static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1497{
1498 struct tcf_proto __rcu **p_filter_chain = priv;
1499
1500 rcu_assign_pointer(*p_filter_chain, tp_head);
1501}
1502
1503int tcf_block_get(struct tcf_block **p_block,
1504 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1505 struct netlink_ext_ack *extack)
1506{
1507 struct tcf_block_ext_info ei = {
1508 .chain_head_change = tcf_chain_head_change_dflt,
1509 .chain_head_change_priv = p_filter_chain,
1510 };
1511
1512 WARN_ON(!p_filter_chain);
1513 return tcf_block_get_ext(p_block, q, &ei, extack);
1514}
1515EXPORT_SYMBOL(tcf_block_get);
1516
1517/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1518 * actions should be all removed after flushing.
1519 */
1520void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1521 struct tcf_block_ext_info *ei)
1522{
1523 struct net_device *dev = qdisc_dev(q);
1524
1525 if (!block)
1526 return;
1527 if (tcf_block_tracks_dev(block, ei))
1528 xa_erase(&block->ports, dev->ifindex);
1529 tcf_chain0_head_change_cb_del(block, ei);
1530 tcf_block_owner_del(block, q, ei->binder_type);
1531
1532 __tcf_block_put(block, q, ei, true);
1533}
1534EXPORT_SYMBOL(tcf_block_put_ext);
1535
1536void tcf_block_put(struct tcf_block *block)
1537{
1538 struct tcf_block_ext_info ei = {0, };
1539
1540 if (!block)
1541 return;
1542 tcf_block_put_ext(block, block->q, &ei);
1543}
1544
1545EXPORT_SYMBOL(tcf_block_put);
1546
1547static int
1548tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1549 void *cb_priv, bool add, bool offload_in_use,
1550 struct netlink_ext_ack *extack)
1551{
1552 struct tcf_chain *chain, *chain_prev;
1553 struct tcf_proto *tp, *tp_prev;
1554 int err;
1555
1556 lockdep_assert_held(&block->cb_lock);
1557
1558 for (chain = __tcf_get_next_chain(block, NULL);
1559 chain;
1560 chain_prev = chain,
1561 chain = __tcf_get_next_chain(block, chain),
1562 tcf_chain_put(chain_prev)) {
1563 if (chain->tmplt_ops && add)
1564 chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1565 cb_priv);
1566 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1567 tp_prev = tp,
1568 tp = __tcf_get_next_proto(chain, tp),
1569 tcf_proto_put(tp_prev, true, NULL)) {
1570 if (tp->ops->reoffload) {
1571 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1572 extack);
1573 if (err && add)
1574 goto err_playback_remove;
1575 } else if (add && offload_in_use) {
1576 err = -EOPNOTSUPP;
1577 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1578 goto err_playback_remove;
1579 }
1580 }
1581 if (chain->tmplt_ops && !add)
1582 chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1583 cb_priv);
1584 }
1585
1586 return 0;
1587
1588err_playback_remove:
1589 tcf_proto_put(tp, true, NULL);
1590 tcf_chain_put(chain);
1591 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1592 extack);
1593 return err;
1594}
1595
1596static int tcf_block_bind(struct tcf_block *block,
1597 struct flow_block_offload *bo)
1598{
1599 struct flow_block_cb *block_cb, *next;
1600 int err, i = 0;
1601
1602 lockdep_assert_held(&block->cb_lock);
1603
1604 list_for_each_entry(block_cb, &bo->cb_list, list) {
1605 err = tcf_block_playback_offloads(block, block_cb->cb,
1606 block_cb->cb_priv, true,
1607 tcf_block_offload_in_use(block),
1608 bo->extack);
1609 if (err)
1610 goto err_unroll;
1611 if (!bo->unlocked_driver_cb)
1612 block->lockeddevcnt++;
1613
1614 i++;
1615 }
1616 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1617
1618 return 0;
1619
1620err_unroll:
1621 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1622 list_del(&block_cb->driver_list);
1623 if (i-- > 0) {
1624 list_del(&block_cb->list);
1625 tcf_block_playback_offloads(block, block_cb->cb,
1626 block_cb->cb_priv, false,
1627 tcf_block_offload_in_use(block),
1628 NULL);
1629 if (!bo->unlocked_driver_cb)
1630 block->lockeddevcnt--;
1631 }
1632 flow_block_cb_free(block_cb);
1633 }
1634
1635 return err;
1636}
1637
1638static void tcf_block_unbind(struct tcf_block *block,
1639 struct flow_block_offload *bo)
1640{
1641 struct flow_block_cb *block_cb, *next;
1642
1643 lockdep_assert_held(&block->cb_lock);
1644
1645 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1646 tcf_block_playback_offloads(block, block_cb->cb,
1647 block_cb->cb_priv, false,
1648 tcf_block_offload_in_use(block),
1649 NULL);
1650 list_del(&block_cb->list);
1651 flow_block_cb_free(block_cb);
1652 if (!bo->unlocked_driver_cb)
1653 block->lockeddevcnt--;
1654 }
1655}
1656
1657static int tcf_block_setup(struct tcf_block *block,
1658 struct flow_block_offload *bo)
1659{
1660 int err;
1661
1662 switch (bo->command) {
1663 case FLOW_BLOCK_BIND:
1664 err = tcf_block_bind(block, bo);
1665 break;
1666 case FLOW_BLOCK_UNBIND:
1667 err = 0;
1668 tcf_block_unbind(block, bo);
1669 break;
1670 default:
1671 WARN_ON_ONCE(1);
1672 err = -EOPNOTSUPP;
1673 }
1674
1675 return err;
1676}
1677
1678/* Main classifier routine: scans classifier chain attached
1679 * to this qdisc, (optionally) tests for protocol and asks
1680 * specific classifiers.
1681 */
1682static inline int __tcf_classify(struct sk_buff *skb,
1683 const struct tcf_proto *tp,
1684 const struct tcf_proto *orig_tp,
1685 struct tcf_result *res,
1686 bool compat_mode,
1687 struct tcf_exts_miss_cookie_node *n,
1688 int act_index,
1689 u32 *last_executed_chain)
1690{
1691#ifdef CONFIG_NET_CLS_ACT
1692 const int max_reclassify_loop = 16;
1693 const struct tcf_proto *first_tp;
1694 int limit = 0;
1695
1696reclassify:
1697#endif
1698 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1699 __be16 protocol = skb_protocol(skb, false);
1700 int err = 0;
1701
1702 if (n) {
1703 struct tcf_exts *exts;
1704
1705 if (n->tp_prio != tp->prio)
1706 continue;
1707
1708 /* We re-lookup the tp and chain based on index instead
1709 * of having hard refs and locks to them, so do a sanity
1710 * check if any of tp,chain,exts was replaced by the
1711 * time we got here with a cookie from hardware.
1712 */
1713 if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1714 !tp->ops->get_exts)) {
1715 tcf_set_drop_reason(skb,
1716 SKB_DROP_REASON_TC_COOKIE_ERROR);
1717 return TC_ACT_SHOT;
1718 }
1719
1720 exts = tp->ops->get_exts(tp, n->handle);
1721 if (unlikely(!exts || n->exts != exts)) {
1722 tcf_set_drop_reason(skb,
1723 SKB_DROP_REASON_TC_COOKIE_ERROR);
1724 return TC_ACT_SHOT;
1725 }
1726
1727 n = NULL;
1728 err = tcf_exts_exec_ex(skb, exts, act_index, res);
1729 } else {
1730 if (tp->protocol != protocol &&
1731 tp->protocol != htons(ETH_P_ALL))
1732 continue;
1733
1734 err = tc_classify(skb, tp, res);
1735 }
1736#ifdef CONFIG_NET_CLS_ACT
1737 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1738 first_tp = orig_tp;
1739 *last_executed_chain = first_tp->chain->index;
1740 goto reset;
1741 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1742 first_tp = res->goto_tp;
1743 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1744 goto reset;
1745 }
1746#endif
1747 if (err >= 0)
1748 return err;
1749 }
1750
1751 if (unlikely(n)) {
1752 tcf_set_drop_reason(skb,
1753 SKB_DROP_REASON_TC_COOKIE_ERROR);
1754 return TC_ACT_SHOT;
1755 }
1756
1757 return TC_ACT_UNSPEC; /* signal: continue lookup */
1758#ifdef CONFIG_NET_CLS_ACT
1759reset:
1760 if (unlikely(limit++ >= max_reclassify_loop)) {
1761 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1762 tp->chain->block->index,
1763 tp->prio & 0xffff,
1764 ntohs(tp->protocol));
1765 tcf_set_drop_reason(skb,
1766 SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1767 return TC_ACT_SHOT;
1768 }
1769
1770 tp = first_tp;
1771 goto reclassify;
1772#endif
1773}
1774
1775int tcf_classify(struct sk_buff *skb,
1776 const struct tcf_block *block,
1777 const struct tcf_proto *tp,
1778 struct tcf_result *res, bool compat_mode)
1779{
1780#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1781 u32 last_executed_chain = 0;
1782
1783 return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1784 &last_executed_chain);
1785#else
1786 u32 last_executed_chain = tp ? tp->chain->index : 0;
1787 struct tcf_exts_miss_cookie_node *n = NULL;
1788 const struct tcf_proto *orig_tp = tp;
1789 struct tc_skb_ext *ext;
1790 int act_index = 0;
1791 int ret;
1792
1793 if (block) {
1794 ext = skb_ext_find(skb, TC_SKB_EXT);
1795
1796 if (ext && (ext->chain || ext->act_miss)) {
1797 struct tcf_chain *fchain;
1798 u32 chain;
1799
1800 if (ext->act_miss) {
1801 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1802 &act_index);
1803 if (!n) {
1804 tcf_set_drop_reason(skb,
1805 SKB_DROP_REASON_TC_COOKIE_ERROR);
1806 return TC_ACT_SHOT;
1807 }
1808
1809 chain = n->chain_index;
1810 } else {
1811 chain = ext->chain;
1812 }
1813
1814 fchain = tcf_chain_lookup_rcu(block, chain);
1815 if (!fchain) {
1816 tcf_set_drop_reason(skb,
1817 SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1818
1819 return TC_ACT_SHOT;
1820 }
1821
1822 /* Consume, so cloned/redirect skbs won't inherit ext */
1823 skb_ext_del(skb, TC_SKB_EXT);
1824
1825 tp = rcu_dereference_bh(fchain->filter_chain);
1826 last_executed_chain = fchain->index;
1827 }
1828 }
1829
1830 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1831 &last_executed_chain);
1832
1833 if (tc_skb_ext_tc_enabled()) {
1834 /* If we missed on some chain */
1835 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1836 struct tc_skb_cb *cb = tc_skb_cb(skb);
1837
1838 ext = tc_skb_ext_alloc(skb);
1839 if (WARN_ON_ONCE(!ext)) {
1840 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1841 return TC_ACT_SHOT;
1842 }
1843 ext->chain = last_executed_chain;
1844 ext->mru = cb->mru;
1845 ext->post_ct = cb->post_ct;
1846 ext->post_ct_snat = cb->post_ct_snat;
1847 ext->post_ct_dnat = cb->post_ct_dnat;
1848 ext->zone = cb->zone;
1849 }
1850 }
1851
1852 return ret;
1853#endif
1854}
1855EXPORT_SYMBOL(tcf_classify);
1856
1857struct tcf_chain_info {
1858 struct tcf_proto __rcu **pprev;
1859 struct tcf_proto __rcu *next;
1860};
1861
1862static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1863 struct tcf_chain_info *chain_info)
1864{
1865 return tcf_chain_dereference(*chain_info->pprev, chain);
1866}
1867
1868static int tcf_chain_tp_insert(struct tcf_chain *chain,
1869 struct tcf_chain_info *chain_info,
1870 struct tcf_proto *tp)
1871{
1872 if (chain->flushing)
1873 return -EAGAIN;
1874
1875 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1876 if (*chain_info->pprev == chain->filter_chain)
1877 tcf_chain0_head_change(chain, tp);
1878 tcf_proto_get(tp);
1879 rcu_assign_pointer(*chain_info->pprev, tp);
1880
1881 return 0;
1882}
1883
1884static void tcf_chain_tp_remove(struct tcf_chain *chain,
1885 struct tcf_chain_info *chain_info,
1886 struct tcf_proto *tp)
1887{
1888 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1889
1890 tcf_proto_mark_delete(tp);
1891 if (tp == chain->filter_chain)
1892 tcf_chain0_head_change(chain, next);
1893 RCU_INIT_POINTER(*chain_info->pprev, next);
1894}
1895
1896static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1897 struct tcf_chain_info *chain_info,
1898 u32 protocol, u32 prio,
1899 bool prio_allocate);
1900
1901/* Try to insert new proto.
1902 * If proto with specified priority already exists, free new proto
1903 * and return existing one.
1904 */
1905
1906static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1907 struct tcf_proto *tp_new,
1908 u32 protocol, u32 prio,
1909 bool rtnl_held)
1910{
1911 struct tcf_chain_info chain_info;
1912 struct tcf_proto *tp;
1913 int err = 0;
1914
1915 mutex_lock(&chain->filter_chain_lock);
1916
1917 if (tcf_proto_exists_destroying(chain, tp_new)) {
1918 mutex_unlock(&chain->filter_chain_lock);
1919 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1920 return ERR_PTR(-EAGAIN);
1921 }
1922
1923 tp = tcf_chain_tp_find(chain, &chain_info,
1924 protocol, prio, false);
1925 if (!tp)
1926 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1927 mutex_unlock(&chain->filter_chain_lock);
1928
1929 if (tp) {
1930 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1931 tp_new = tp;
1932 } else if (err) {
1933 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1934 tp_new = ERR_PTR(err);
1935 }
1936
1937 return tp_new;
1938}
1939
1940static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1941 struct tcf_proto *tp, bool rtnl_held,
1942 struct netlink_ext_ack *extack)
1943{
1944 struct tcf_chain_info chain_info;
1945 struct tcf_proto *tp_iter;
1946 struct tcf_proto **pprev;
1947 struct tcf_proto *next;
1948
1949 mutex_lock(&chain->filter_chain_lock);
1950
1951 /* Atomically find and remove tp from chain. */
1952 for (pprev = &chain->filter_chain;
1953 (tp_iter = tcf_chain_dereference(*pprev, chain));
1954 pprev = &tp_iter->next) {
1955 if (tp_iter == tp) {
1956 chain_info.pprev = pprev;
1957 chain_info.next = tp_iter->next;
1958 WARN_ON(tp_iter->deleting);
1959 break;
1960 }
1961 }
1962 /* Verify that tp still exists and no new filters were inserted
1963 * concurrently.
1964 * Mark tp for deletion if it is empty.
1965 */
1966 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1967 mutex_unlock(&chain->filter_chain_lock);
1968 return;
1969 }
1970
1971 tcf_proto_signal_destroying(chain, tp);
1972 next = tcf_chain_dereference(chain_info.next, chain);
1973 if (tp == chain->filter_chain)
1974 tcf_chain0_head_change(chain, next);
1975 RCU_INIT_POINTER(*chain_info.pprev, next);
1976 mutex_unlock(&chain->filter_chain_lock);
1977
1978 tcf_proto_put(tp, rtnl_held, extack);
1979}
1980
1981static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1982 struct tcf_chain_info *chain_info,
1983 u32 protocol, u32 prio,
1984 bool prio_allocate)
1985{
1986 struct tcf_proto **pprev;
1987 struct tcf_proto *tp;
1988
1989 /* Check the chain for existence of proto-tcf with this priority */
1990 for (pprev = &chain->filter_chain;
1991 (tp = tcf_chain_dereference(*pprev, chain));
1992 pprev = &tp->next) {
1993 if (tp->prio >= prio) {
1994 if (tp->prio == prio) {
1995 if (prio_allocate ||
1996 (tp->protocol != protocol && protocol))
1997 return ERR_PTR(-EINVAL);
1998 } else {
1999 tp = NULL;
2000 }
2001 break;
2002 }
2003 }
2004 chain_info->pprev = pprev;
2005 if (tp) {
2006 chain_info->next = tp->next;
2007 tcf_proto_get(tp);
2008 } else {
2009 chain_info->next = NULL;
2010 }
2011 return tp;
2012}
2013
2014static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2015 struct tcf_proto *tp, struct tcf_block *block,
2016 struct Qdisc *q, u32 parent, void *fh,
2017 u32 portid, u32 seq, u16 flags, int event,
2018 bool terse_dump, bool rtnl_held,
2019 struct netlink_ext_ack *extack)
2020{
2021 struct tcmsg *tcm;
2022 struct nlmsghdr *nlh;
2023 unsigned char *b = skb_tail_pointer(skb);
2024
2025 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2026 if (!nlh)
2027 goto out_nlmsg_trim;
2028 tcm = nlmsg_data(nlh);
2029 tcm->tcm_family = AF_UNSPEC;
2030 tcm->tcm__pad1 = 0;
2031 tcm->tcm__pad2 = 0;
2032 if (q) {
2033 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2034 tcm->tcm_parent = parent;
2035 } else {
2036 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2037 tcm->tcm_block_index = block->index;
2038 }
2039 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2040 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2041 goto nla_put_failure;
2042 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2043 goto nla_put_failure;
2044 if (!fh) {
2045 tcm->tcm_handle = 0;
2046 } else if (terse_dump) {
2047 if (tp->ops->terse_dump) {
2048 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2049 rtnl_held) < 0)
2050 goto nla_put_failure;
2051 } else {
2052 goto cls_op_not_supp;
2053 }
2054 } else {
2055 if (tp->ops->dump &&
2056 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2057 goto nla_put_failure;
2058 }
2059
2060 if (extack && extack->_msg &&
2061 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2062 goto nla_put_failure;
2063
2064 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2065
2066 return skb->len;
2067
2068out_nlmsg_trim:
2069nla_put_failure:
2070cls_op_not_supp:
2071 nlmsg_trim(skb, b);
2072 return -1;
2073}
2074
2075static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2076 struct nlmsghdr *n, struct tcf_proto *tp,
2077 struct tcf_block *block, struct Qdisc *q,
2078 u32 parent, void *fh, int event, bool unicast,
2079 bool rtnl_held, struct netlink_ext_ack *extack)
2080{
2081 struct sk_buff *skb;
2082 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2083 int err = 0;
2084
2085 if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2086 return 0;
2087
2088 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2089 if (!skb)
2090 return -ENOBUFS;
2091
2092 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2093 n->nlmsg_seq, n->nlmsg_flags, event,
2094 false, rtnl_held, extack) <= 0) {
2095 kfree_skb(skb);
2096 return -EINVAL;
2097 }
2098
2099 if (unicast)
2100 err = rtnl_unicast(skb, net, portid);
2101 else
2102 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2103 n->nlmsg_flags & NLM_F_ECHO);
2104 return err;
2105}
2106
2107static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2108 struct nlmsghdr *n, struct tcf_proto *tp,
2109 struct tcf_block *block, struct Qdisc *q,
2110 u32 parent, void *fh, bool *last, bool rtnl_held,
2111 struct netlink_ext_ack *extack)
2112{
2113 struct sk_buff *skb;
2114 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2115 int err;
2116
2117 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2118 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2119
2120 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2121 if (!skb)
2122 return -ENOBUFS;
2123
2124 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2125 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2126 false, rtnl_held, extack) <= 0) {
2127 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2128 kfree_skb(skb);
2129 return -EINVAL;
2130 }
2131
2132 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2133 if (err) {
2134 kfree_skb(skb);
2135 return err;
2136 }
2137
2138 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2139 n->nlmsg_flags & NLM_F_ECHO);
2140 if (err < 0)
2141 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2142
2143 return err;
2144}
2145
2146static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2147 struct tcf_block *block, struct Qdisc *q,
2148 u32 parent, struct nlmsghdr *n,
2149 struct tcf_chain *chain, int event,
2150 struct netlink_ext_ack *extack)
2151{
2152 struct tcf_proto *tp;
2153
2154 for (tp = tcf_get_next_proto(chain, NULL);
2155 tp; tp = tcf_get_next_proto(chain, tp))
2156 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2157 event, false, true, extack);
2158}
2159
2160static void tfilter_put(struct tcf_proto *tp, void *fh)
2161{
2162 if (tp->ops->put && fh)
2163 tp->ops->put(tp, fh);
2164}
2165
2166static bool is_qdisc_ingress(__u32 classid)
2167{
2168 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2169}
2170
2171static int tc_new_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 bool prio_allocate;
2181 u32 parent;
2182 u32 chain_index;
2183 struct Qdisc *q;
2184 struct tcf_chain_info chain_info;
2185 struct tcf_chain *chain;
2186 struct tcf_block *block;
2187 struct tcf_proto *tp;
2188 unsigned long cl;
2189 void *fh;
2190 int err;
2191 int tp_created;
2192 bool rtnl_held = false;
2193 u32 flags;
2194
2195replay:
2196 tp_created = 0;
2197
2198 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2199 rtm_tca_policy, extack);
2200 if (err < 0)
2201 return err;
2202
2203 t = nlmsg_data(n);
2204 protocol = TC_H_MIN(t->tcm_info);
2205 prio = TC_H_MAJ(t->tcm_info);
2206 prio_allocate = false;
2207 parent = t->tcm_parent;
2208 tp = NULL;
2209 cl = 0;
2210 block = NULL;
2211 q = NULL;
2212 chain = NULL;
2213 flags = 0;
2214
2215 if (prio == 0) {
2216 /* If no priority is provided by the user,
2217 * we allocate one.
2218 */
2219 if (n->nlmsg_flags & NLM_F_CREATE) {
2220 prio = TC_H_MAKE(0x80000000U, 0U);
2221 prio_allocate = true;
2222 } else {
2223 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2224 return -ENOENT;
2225 }
2226 }
2227
2228 /* Find head of filter chain. */
2229
2230 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2231 if (err)
2232 return err;
2233
2234 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2235 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2236 err = -EINVAL;
2237 goto errout;
2238 }
2239
2240 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2241 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2242 * type is not specified, classifier is not unlocked.
2243 */
2244 if (rtnl_held ||
2245 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2246 !tcf_proto_is_unlocked(name)) {
2247 rtnl_held = true;
2248 rtnl_lock();
2249 }
2250
2251 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2252 if (err)
2253 goto errout;
2254
2255 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2256 extack);
2257 if (IS_ERR(block)) {
2258 err = PTR_ERR(block);
2259 goto errout;
2260 }
2261 block->classid = parent;
2262
2263 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2264 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2265 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2266 err = -EINVAL;
2267 goto errout;
2268 }
2269 chain = tcf_chain_get(block, chain_index, true);
2270 if (!chain) {
2271 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2272 err = -ENOMEM;
2273 goto errout;
2274 }
2275
2276 mutex_lock(&chain->filter_chain_lock);
2277 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2278 prio, prio_allocate);
2279 if (IS_ERR(tp)) {
2280 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2281 err = PTR_ERR(tp);
2282 goto errout_locked;
2283 }
2284
2285 if (tp == NULL) {
2286 struct tcf_proto *tp_new = NULL;
2287
2288 if (chain->flushing) {
2289 err = -EAGAIN;
2290 goto errout_locked;
2291 }
2292
2293 /* Proto-tcf does not exist, create new one */
2294
2295 if (tca[TCA_KIND] == NULL || !protocol) {
2296 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2297 err = -EINVAL;
2298 goto errout_locked;
2299 }
2300
2301 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2302 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2303 err = -ENOENT;
2304 goto errout_locked;
2305 }
2306
2307 if (prio_allocate)
2308 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2309 &chain_info));
2310
2311 mutex_unlock(&chain->filter_chain_lock);
2312 tp_new = tcf_proto_create(name, protocol, prio, chain,
2313 rtnl_held, extack);
2314 if (IS_ERR(tp_new)) {
2315 err = PTR_ERR(tp_new);
2316 goto errout_tp;
2317 }
2318
2319 tp_created = 1;
2320 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2321 rtnl_held);
2322 if (IS_ERR(tp)) {
2323 err = PTR_ERR(tp);
2324 goto errout_tp;
2325 }
2326 } else {
2327 mutex_unlock(&chain->filter_chain_lock);
2328 }
2329
2330 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2331 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2332 err = -EINVAL;
2333 goto errout;
2334 }
2335
2336 fh = tp->ops->get(tp, t->tcm_handle);
2337
2338 if (!fh) {
2339 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2340 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2341 err = -ENOENT;
2342 goto errout;
2343 }
2344 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2345 tfilter_put(tp, fh);
2346 NL_SET_ERR_MSG(extack, "Filter already exists");
2347 err = -EEXIST;
2348 goto errout;
2349 }
2350
2351 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2352 tfilter_put(tp, fh);
2353 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2354 err = -EINVAL;
2355 goto errout;
2356 }
2357
2358 if (!(n->nlmsg_flags & NLM_F_CREATE))
2359 flags |= TCA_ACT_FLAGS_REPLACE;
2360 if (!rtnl_held)
2361 flags |= TCA_ACT_FLAGS_NO_RTNL;
2362 if (is_qdisc_ingress(parent))
2363 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2364 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2365 flags, extack);
2366 if (err == 0) {
2367 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2368 RTM_NEWTFILTER, false, rtnl_held, extack);
2369 tfilter_put(tp, fh);
2370 /* q pointer is NULL for shared blocks */
2371 if (q)
2372 q->flags &= ~TCQ_F_CAN_BYPASS;
2373 }
2374
2375errout:
2376 if (err && tp_created)
2377 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2378errout_tp:
2379 if (chain) {
2380 if (tp && !IS_ERR(tp))
2381 tcf_proto_put(tp, rtnl_held, NULL);
2382 if (!tp_created)
2383 tcf_chain_put(chain);
2384 }
2385 tcf_block_release(q, block, rtnl_held);
2386
2387 if (rtnl_held)
2388 rtnl_unlock();
2389
2390 if (err == -EAGAIN) {
2391 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2392 * of target chain.
2393 */
2394 rtnl_held = true;
2395 /* Replay the request. */
2396 goto replay;
2397 }
2398 return err;
2399
2400errout_locked:
2401 mutex_unlock(&chain->filter_chain_lock);
2402 goto errout;
2403}
2404
2405static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2406 struct netlink_ext_ack *extack)
2407{
2408 struct net *net = sock_net(skb->sk);
2409 struct nlattr *tca[TCA_MAX + 1];
2410 char name[IFNAMSIZ];
2411 struct tcmsg *t;
2412 u32 protocol;
2413 u32 prio;
2414 u32 parent;
2415 u32 chain_index;
2416 struct Qdisc *q = NULL;
2417 struct tcf_chain_info chain_info;
2418 struct tcf_chain *chain = NULL;
2419 struct tcf_block *block = NULL;
2420 struct tcf_proto *tp = NULL;
2421 unsigned long cl = 0;
2422 void *fh = NULL;
2423 int err;
2424 bool rtnl_held = false;
2425
2426 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2427 rtm_tca_policy, extack);
2428 if (err < 0)
2429 return err;
2430
2431 t = nlmsg_data(n);
2432 protocol = TC_H_MIN(t->tcm_info);
2433 prio = TC_H_MAJ(t->tcm_info);
2434 parent = t->tcm_parent;
2435
2436 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2437 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2438 return -ENOENT;
2439 }
2440
2441 /* Find head of filter chain. */
2442
2443 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2444 if (err)
2445 return err;
2446
2447 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2448 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2449 err = -EINVAL;
2450 goto errout;
2451 }
2452 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2453 * found), qdisc is not unlocked, classifier type is not specified,
2454 * classifier is not unlocked.
2455 */
2456 if (!prio ||
2457 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2458 !tcf_proto_is_unlocked(name)) {
2459 rtnl_held = true;
2460 rtnl_lock();
2461 }
2462
2463 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2464 if (err)
2465 goto errout;
2466
2467 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2468 extack);
2469 if (IS_ERR(block)) {
2470 err = PTR_ERR(block);
2471 goto errout;
2472 }
2473
2474 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2475 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2476 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2477 err = -EINVAL;
2478 goto errout;
2479 }
2480 chain = tcf_chain_get(block, chain_index, false);
2481 if (!chain) {
2482 /* User requested flush on non-existent chain. Nothing to do,
2483 * so just return success.
2484 */
2485 if (prio == 0) {
2486 err = 0;
2487 goto errout;
2488 }
2489 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2490 err = -ENOENT;
2491 goto errout;
2492 }
2493
2494 if (prio == 0) {
2495 tfilter_notify_chain(net, skb, block, q, parent, n,
2496 chain, RTM_DELTFILTER, extack);
2497 tcf_chain_flush(chain, rtnl_held);
2498 err = 0;
2499 goto errout;
2500 }
2501
2502 mutex_lock(&chain->filter_chain_lock);
2503 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2504 prio, false);
2505 if (!tp || IS_ERR(tp)) {
2506 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2507 err = tp ? PTR_ERR(tp) : -ENOENT;
2508 goto errout_locked;
2509 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2510 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2511 err = -EINVAL;
2512 goto errout_locked;
2513 } else if (t->tcm_handle == 0) {
2514 tcf_proto_signal_destroying(chain, tp);
2515 tcf_chain_tp_remove(chain, &chain_info, tp);
2516 mutex_unlock(&chain->filter_chain_lock);
2517
2518 tcf_proto_put(tp, rtnl_held, NULL);
2519 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2520 RTM_DELTFILTER, false, rtnl_held, extack);
2521 err = 0;
2522 goto errout;
2523 }
2524 mutex_unlock(&chain->filter_chain_lock);
2525
2526 fh = tp->ops->get(tp, t->tcm_handle);
2527
2528 if (!fh) {
2529 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2530 err = -ENOENT;
2531 } else {
2532 bool last;
2533
2534 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2535 &last, rtnl_held, extack);
2536
2537 if (err)
2538 goto errout;
2539 if (last)
2540 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2541 }
2542
2543errout:
2544 if (chain) {
2545 if (tp && !IS_ERR(tp))
2546 tcf_proto_put(tp, rtnl_held, NULL);
2547 tcf_chain_put(chain);
2548 }
2549 tcf_block_release(q, block, rtnl_held);
2550
2551 if (rtnl_held)
2552 rtnl_unlock();
2553
2554 return err;
2555
2556errout_locked:
2557 mutex_unlock(&chain->filter_chain_lock);
2558 goto errout;
2559}
2560
2561static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2562 struct netlink_ext_ack *extack)
2563{
2564 struct net *net = sock_net(skb->sk);
2565 struct nlattr *tca[TCA_MAX + 1];
2566 char name[IFNAMSIZ];
2567 struct tcmsg *t;
2568 u32 protocol;
2569 u32 prio;
2570 u32 parent;
2571 u32 chain_index;
2572 struct Qdisc *q = NULL;
2573 struct tcf_chain_info chain_info;
2574 struct tcf_chain *chain = NULL;
2575 struct tcf_block *block = NULL;
2576 struct tcf_proto *tp = NULL;
2577 unsigned long cl = 0;
2578 void *fh = NULL;
2579 int err;
2580 bool rtnl_held = false;
2581
2582 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2583 rtm_tca_policy, extack);
2584 if (err < 0)
2585 return err;
2586
2587 t = nlmsg_data(n);
2588 protocol = TC_H_MIN(t->tcm_info);
2589 prio = TC_H_MAJ(t->tcm_info);
2590 parent = t->tcm_parent;
2591
2592 if (prio == 0) {
2593 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2594 return -ENOENT;
2595 }
2596
2597 /* Find head of filter chain. */
2598
2599 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2600 if (err)
2601 return err;
2602
2603 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2604 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2605 err = -EINVAL;
2606 goto errout;
2607 }
2608 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2609 * unlocked, classifier type is not specified, classifier is not
2610 * unlocked.
2611 */
2612 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2613 !tcf_proto_is_unlocked(name)) {
2614 rtnl_held = true;
2615 rtnl_lock();
2616 }
2617
2618 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2619 if (err)
2620 goto errout;
2621
2622 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2623 extack);
2624 if (IS_ERR(block)) {
2625 err = PTR_ERR(block);
2626 goto errout;
2627 }
2628
2629 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2630 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2631 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2632 err = -EINVAL;
2633 goto errout;
2634 }
2635 chain = tcf_chain_get(block, chain_index, false);
2636 if (!chain) {
2637 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2638 err = -EINVAL;
2639 goto errout;
2640 }
2641
2642 mutex_lock(&chain->filter_chain_lock);
2643 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2644 prio, false);
2645 mutex_unlock(&chain->filter_chain_lock);
2646 if (!tp || IS_ERR(tp)) {
2647 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2648 err = tp ? PTR_ERR(tp) : -ENOENT;
2649 goto errout;
2650 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2651 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2652 err = -EINVAL;
2653 goto errout;
2654 }
2655
2656 fh = tp->ops->get(tp, t->tcm_handle);
2657
2658 if (!fh) {
2659 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2660 err = -ENOENT;
2661 } else {
2662 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2663 fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2664 if (err < 0)
2665 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2666 }
2667
2668 tfilter_put(tp, fh);
2669errout:
2670 if (chain) {
2671 if (tp && !IS_ERR(tp))
2672 tcf_proto_put(tp, rtnl_held, NULL);
2673 tcf_chain_put(chain);
2674 }
2675 tcf_block_release(q, block, rtnl_held);
2676
2677 if (rtnl_held)
2678 rtnl_unlock();
2679
2680 return err;
2681}
2682
2683struct tcf_dump_args {
2684 struct tcf_walker w;
2685 struct sk_buff *skb;
2686 struct netlink_callback *cb;
2687 struct tcf_block *block;
2688 struct Qdisc *q;
2689 u32 parent;
2690 bool terse_dump;
2691};
2692
2693static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2694{
2695 struct tcf_dump_args *a = (void *)arg;
2696 struct net *net = sock_net(a->skb->sk);
2697
2698 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2699 n, NETLINK_CB(a->cb->skb).portid,
2700 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2701 RTM_NEWTFILTER, a->terse_dump, true, NULL);
2702}
2703
2704static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2705 struct sk_buff *skb, struct netlink_callback *cb,
2706 long index_start, long *p_index, bool terse)
2707{
2708 struct net *net = sock_net(skb->sk);
2709 struct tcf_block *block = chain->block;
2710 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2711 struct tcf_proto *tp, *tp_prev;
2712 struct tcf_dump_args arg;
2713
2714 for (tp = __tcf_get_next_proto(chain, NULL);
2715 tp;
2716 tp_prev = tp,
2717 tp = __tcf_get_next_proto(chain, tp),
2718 tcf_proto_put(tp_prev, true, NULL),
2719 (*p_index)++) {
2720 if (*p_index < index_start)
2721 continue;
2722 if (TC_H_MAJ(tcm->tcm_info) &&
2723 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2724 continue;
2725 if (TC_H_MIN(tcm->tcm_info) &&
2726 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2727 continue;
2728 if (*p_index > index_start)
2729 memset(&cb->args[1], 0,
2730 sizeof(cb->args) - sizeof(cb->args[0]));
2731 if (cb->args[1] == 0) {
2732 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2733 NETLINK_CB(cb->skb).portid,
2734 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2735 RTM_NEWTFILTER, false, true, NULL) <= 0)
2736 goto errout;
2737 cb->args[1] = 1;
2738 }
2739 if (!tp->ops->walk)
2740 continue;
2741 arg.w.fn = tcf_node_dump;
2742 arg.skb = skb;
2743 arg.cb = cb;
2744 arg.block = block;
2745 arg.q = q;
2746 arg.parent = parent;
2747 arg.w.stop = 0;
2748 arg.w.skip = cb->args[1] - 1;
2749 arg.w.count = 0;
2750 arg.w.cookie = cb->args[2];
2751 arg.terse_dump = terse;
2752 tp->ops->walk(tp, &arg.w, true);
2753 cb->args[2] = arg.w.cookie;
2754 cb->args[1] = arg.w.count + 1;
2755 if (arg.w.stop)
2756 goto errout;
2757 }
2758 return true;
2759
2760errout:
2761 tcf_proto_put(tp, true, NULL);
2762 return false;
2763}
2764
2765static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2766 [TCA_CHAIN] = { .type = NLA_U32 },
2767 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2768};
2769
2770/* called with RTNL */
2771static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2772{
2773 struct tcf_chain *chain, *chain_prev;
2774 struct net *net = sock_net(skb->sk);
2775 struct nlattr *tca[TCA_MAX + 1];
2776 struct Qdisc *q = NULL;
2777 struct tcf_block *block;
2778 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2779 bool terse_dump = false;
2780 long index_start;
2781 long index;
2782 u32 parent;
2783 int err;
2784
2785 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2786 return skb->len;
2787
2788 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2789 tcf_tfilter_dump_policy, cb->extack);
2790 if (err)
2791 return err;
2792
2793 if (tca[TCA_DUMP_FLAGS]) {
2794 struct nla_bitfield32 flags =
2795 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2796
2797 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2798 }
2799
2800 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2801 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2802 if (!block)
2803 goto out;
2804 /* If we work with block index, q is NULL and parent value
2805 * will never be used in the following code. The check
2806 * in tcf_fill_node prevents it. However, compiler does not
2807 * see that far, so set parent to zero to silence the warning
2808 * about parent being uninitialized.
2809 */
2810 parent = 0;
2811 } else {
2812 const struct Qdisc_class_ops *cops;
2813 struct net_device *dev;
2814 unsigned long cl = 0;
2815
2816 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2817 if (!dev)
2818 return skb->len;
2819
2820 parent = tcm->tcm_parent;
2821 if (!parent)
2822 q = rtnl_dereference(dev->qdisc);
2823 else
2824 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2825 if (!q)
2826 goto out;
2827 cops = q->ops->cl_ops;
2828 if (!cops)
2829 goto out;
2830 if (!cops->tcf_block)
2831 goto out;
2832 if (TC_H_MIN(tcm->tcm_parent)) {
2833 cl = cops->find(q, tcm->tcm_parent);
2834 if (cl == 0)
2835 goto out;
2836 }
2837 block = cops->tcf_block(q, cl, NULL);
2838 if (!block)
2839 goto out;
2840 parent = block->classid;
2841 if (tcf_block_shared(block))
2842 q = NULL;
2843 }
2844
2845 index_start = cb->args[0];
2846 index = 0;
2847
2848 for (chain = __tcf_get_next_chain(block, NULL);
2849 chain;
2850 chain_prev = chain,
2851 chain = __tcf_get_next_chain(block, chain),
2852 tcf_chain_put(chain_prev)) {
2853 if (tca[TCA_CHAIN] &&
2854 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2855 continue;
2856 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2857 index_start, &index, terse_dump)) {
2858 tcf_chain_put(chain);
2859 err = -EMSGSIZE;
2860 break;
2861 }
2862 }
2863
2864 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2865 tcf_block_refcnt_put(block, true);
2866 cb->args[0] = index;
2867
2868out:
2869 /* If we did no progress, the error (EMSGSIZE) is real */
2870 if (skb->len == 0 && err)
2871 return err;
2872 return skb->len;
2873}
2874
2875static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2876 void *tmplt_priv, u32 chain_index,
2877 struct net *net, struct sk_buff *skb,
2878 struct tcf_block *block,
2879 u32 portid, u32 seq, u16 flags, int event,
2880 struct netlink_ext_ack *extack)
2881{
2882 unsigned char *b = skb_tail_pointer(skb);
2883 const struct tcf_proto_ops *ops;
2884 struct nlmsghdr *nlh;
2885 struct tcmsg *tcm;
2886 void *priv;
2887
2888 ops = tmplt_ops;
2889 priv = tmplt_priv;
2890
2891 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2892 if (!nlh)
2893 goto out_nlmsg_trim;
2894 tcm = nlmsg_data(nlh);
2895 tcm->tcm_family = AF_UNSPEC;
2896 tcm->tcm__pad1 = 0;
2897 tcm->tcm__pad2 = 0;
2898 tcm->tcm_handle = 0;
2899 if (block->q) {
2900 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2901 tcm->tcm_parent = block->q->handle;
2902 } else {
2903 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2904 tcm->tcm_block_index = block->index;
2905 }
2906
2907 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2908 goto nla_put_failure;
2909
2910 if (ops) {
2911 if (nla_put_string(skb, TCA_KIND, ops->kind))
2912 goto nla_put_failure;
2913 if (ops->tmplt_dump(skb, net, priv) < 0)
2914 goto nla_put_failure;
2915 }
2916
2917 if (extack && extack->_msg &&
2918 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2919 goto out_nlmsg_trim;
2920
2921 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2922
2923 return skb->len;
2924
2925out_nlmsg_trim:
2926nla_put_failure:
2927 nlmsg_trim(skb, b);
2928 return -EMSGSIZE;
2929}
2930
2931static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2932 u32 seq, u16 flags, int event, bool unicast,
2933 struct netlink_ext_ack *extack)
2934{
2935 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2936 struct tcf_block *block = chain->block;
2937 struct net *net = block->net;
2938 struct sk_buff *skb;
2939 int err = 0;
2940
2941 if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2942 return 0;
2943
2944 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2945 if (!skb)
2946 return -ENOBUFS;
2947
2948 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2949 chain->index, net, skb, block, portid,
2950 seq, flags, event, extack) <= 0) {
2951 kfree_skb(skb);
2952 return -EINVAL;
2953 }
2954
2955 if (unicast)
2956 err = rtnl_unicast(skb, net, portid);
2957 else
2958 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2959 flags & NLM_F_ECHO);
2960
2961 return err;
2962}
2963
2964static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2965 void *tmplt_priv, u32 chain_index,
2966 struct tcf_block *block, struct sk_buff *oskb,
2967 u32 seq, u16 flags)
2968{
2969 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2970 struct net *net = block->net;
2971 struct sk_buff *skb;
2972
2973 if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
2974 return 0;
2975
2976 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2977 if (!skb)
2978 return -ENOBUFS;
2979
2980 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2981 block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
2982 kfree_skb(skb);
2983 return -EINVAL;
2984 }
2985
2986 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2987}
2988
2989static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2990 struct nlattr **tca,
2991 struct netlink_ext_ack *extack)
2992{
2993 const struct tcf_proto_ops *ops;
2994 char name[IFNAMSIZ];
2995 void *tmplt_priv;
2996
2997 /* If kind is not set, user did not specify template. */
2998 if (!tca[TCA_KIND])
2999 return 0;
3000
3001 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3002 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3003 return -EINVAL;
3004 }
3005
3006 ops = tcf_proto_lookup_ops(name, true, extack);
3007 if (IS_ERR(ops))
3008 return PTR_ERR(ops);
3009 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3010 !ops->tmplt_reoffload) {
3011 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3012 module_put(ops->owner);
3013 return -EOPNOTSUPP;
3014 }
3015
3016 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3017 if (IS_ERR(tmplt_priv)) {
3018 module_put(ops->owner);
3019 return PTR_ERR(tmplt_priv);
3020 }
3021 chain->tmplt_ops = ops;
3022 chain->tmplt_priv = tmplt_priv;
3023 return 0;
3024}
3025
3026static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3027 void *tmplt_priv)
3028{
3029 /* If template ops are set, no work to do for us. */
3030 if (!tmplt_ops)
3031 return;
3032
3033 tmplt_ops->tmplt_destroy(tmplt_priv);
3034 module_put(tmplt_ops->owner);
3035}
3036
3037/* Add/delete/get a chain */
3038
3039static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3040 struct netlink_ext_ack *extack)
3041{
3042 struct net *net = sock_net(skb->sk);
3043 struct nlattr *tca[TCA_MAX + 1];
3044 struct tcmsg *t;
3045 u32 parent;
3046 u32 chain_index;
3047 struct Qdisc *q;
3048 struct tcf_chain *chain;
3049 struct tcf_block *block;
3050 unsigned long cl;
3051 int err;
3052
3053replay:
3054 q = NULL;
3055 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3056 rtm_tca_policy, extack);
3057 if (err < 0)
3058 return err;
3059
3060 t = nlmsg_data(n);
3061 parent = t->tcm_parent;
3062 cl = 0;
3063
3064 block = tcf_block_find(net, &q, &parent, &cl,
3065 t->tcm_ifindex, t->tcm_block_index, extack);
3066 if (IS_ERR(block))
3067 return PTR_ERR(block);
3068
3069 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3070 if (chain_index > TC_ACT_EXT_VAL_MASK) {
3071 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3072 err = -EINVAL;
3073 goto errout_block;
3074 }
3075
3076 mutex_lock(&block->lock);
3077 chain = tcf_chain_lookup(block, chain_index);
3078 if (n->nlmsg_type == RTM_NEWCHAIN) {
3079 if (chain) {
3080 if (tcf_chain_held_by_acts_only(chain)) {
3081 /* The chain exists only because there is
3082 * some action referencing it.
3083 */
3084 tcf_chain_hold(chain);
3085 } else {
3086 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3087 err = -EEXIST;
3088 goto errout_block_locked;
3089 }
3090 } else {
3091 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3092 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3093 err = -ENOENT;
3094 goto errout_block_locked;
3095 }
3096 chain = tcf_chain_create(block, chain_index);
3097 if (!chain) {
3098 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3099 err = -ENOMEM;
3100 goto errout_block_locked;
3101 }
3102 }
3103 } else {
3104 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3105 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3106 err = -EINVAL;
3107 goto errout_block_locked;
3108 }
3109 tcf_chain_hold(chain);
3110 }
3111
3112 if (n->nlmsg_type == RTM_NEWCHAIN) {
3113 /* Modifying chain requires holding parent block lock. In case
3114 * the chain was successfully added, take a reference to the
3115 * chain. This ensures that an empty chain does not disappear at
3116 * the end of this function.
3117 */
3118 tcf_chain_hold(chain);
3119 chain->explicitly_created = true;
3120 }
3121 mutex_unlock(&block->lock);
3122
3123 switch (n->nlmsg_type) {
3124 case RTM_NEWCHAIN:
3125 err = tc_chain_tmplt_add(chain, net, tca, extack);
3126 if (err) {
3127 tcf_chain_put_explicitly_created(chain);
3128 goto errout;
3129 }
3130
3131 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3132 RTM_NEWCHAIN, false, extack);
3133 break;
3134 case RTM_DELCHAIN:
3135 tfilter_notify_chain(net, skb, block, q, parent, n,
3136 chain, RTM_DELTFILTER, extack);
3137 /* Flush the chain first as the user requested chain removal. */
3138 tcf_chain_flush(chain, true);
3139 /* In case the chain was successfully deleted, put a reference
3140 * to the chain previously taken during addition.
3141 */
3142 tcf_chain_put_explicitly_created(chain);
3143 break;
3144 case RTM_GETCHAIN:
3145 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3146 n->nlmsg_flags, n->nlmsg_type, true, extack);
3147 if (err < 0)
3148 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3149 break;
3150 default:
3151 err = -EOPNOTSUPP;
3152 NL_SET_ERR_MSG(extack, "Unsupported message type");
3153 goto errout;
3154 }
3155
3156errout:
3157 tcf_chain_put(chain);
3158errout_block:
3159 tcf_block_release(q, block, true);
3160 if (err == -EAGAIN)
3161 /* Replay the request. */
3162 goto replay;
3163 return err;
3164
3165errout_block_locked:
3166 mutex_unlock(&block->lock);
3167 goto errout_block;
3168}
3169
3170/* called with RTNL */
3171static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3172{
3173 struct net *net = sock_net(skb->sk);
3174 struct nlattr *tca[TCA_MAX + 1];
3175 struct Qdisc *q = NULL;
3176 struct tcf_block *block;
3177 struct tcmsg *tcm = nlmsg_data(cb->nlh);
3178 struct tcf_chain *chain;
3179 long index_start;
3180 long index;
3181 int err;
3182
3183 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3184 return skb->len;
3185
3186 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3187 rtm_tca_policy, cb->extack);
3188 if (err)
3189 return err;
3190
3191 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3192 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3193 if (!block)
3194 goto out;
3195 } else {
3196 const struct Qdisc_class_ops *cops;
3197 struct net_device *dev;
3198 unsigned long cl = 0;
3199
3200 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3201 if (!dev)
3202 return skb->len;
3203
3204 if (!tcm->tcm_parent)
3205 q = rtnl_dereference(dev->qdisc);
3206 else
3207 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3208
3209 if (!q)
3210 goto out;
3211 cops = q->ops->cl_ops;
3212 if (!cops)
3213 goto out;
3214 if (!cops->tcf_block)
3215 goto out;
3216 if (TC_H_MIN(tcm->tcm_parent)) {
3217 cl = cops->find(q, tcm->tcm_parent);
3218 if (cl == 0)
3219 goto out;
3220 }
3221 block = cops->tcf_block(q, cl, NULL);
3222 if (!block)
3223 goto out;
3224 if (tcf_block_shared(block))
3225 q = NULL;
3226 }
3227
3228 index_start = cb->args[0];
3229 index = 0;
3230
3231 mutex_lock(&block->lock);
3232 list_for_each_entry(chain, &block->chain_list, list) {
3233 if ((tca[TCA_CHAIN] &&
3234 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3235 continue;
3236 if (index < index_start) {
3237 index++;
3238 continue;
3239 }
3240 if (tcf_chain_held_by_acts_only(chain))
3241 continue;
3242 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3243 chain->index, net, skb, block,
3244 NETLINK_CB(cb->skb).portid,
3245 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3246 RTM_NEWCHAIN, NULL);
3247 if (err <= 0)
3248 break;
3249 index++;
3250 }
3251 mutex_unlock(&block->lock);
3252
3253 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3254 tcf_block_refcnt_put(block, true);
3255 cb->args[0] = index;
3256
3257out:
3258 /* If we did no progress, the error (EMSGSIZE) is real */
3259 if (skb->len == 0 && err)
3260 return err;
3261 return skb->len;
3262}
3263
3264int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3265 int police, struct tcf_proto *tp, u32 handle,
3266 bool use_action_miss)
3267{
3268 int err = 0;
3269
3270#ifdef CONFIG_NET_CLS_ACT
3271 exts->type = 0;
3272 exts->nr_actions = 0;
3273 exts->miss_cookie_node = NULL;
3274 /* Note: we do not own yet a reference on net.
3275 * This reference might be taken later from tcf_exts_get_net().
3276 */
3277 exts->net = net;
3278 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3279 GFP_KERNEL);
3280 if (!exts->actions)
3281 return -ENOMEM;
3282#endif
3283
3284 exts->action = action;
3285 exts->police = police;
3286
3287 if (!use_action_miss)
3288 return 0;
3289
3290 err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3291 if (err)
3292 goto err_miss_alloc;
3293
3294 return 0;
3295
3296err_miss_alloc:
3297 tcf_exts_destroy(exts);
3298#ifdef CONFIG_NET_CLS_ACT
3299 exts->actions = NULL;
3300#endif
3301 return err;
3302}
3303EXPORT_SYMBOL(tcf_exts_init_ex);
3304
3305void tcf_exts_destroy(struct tcf_exts *exts)
3306{
3307 tcf_exts_miss_cookie_base_destroy(exts);
3308
3309#ifdef CONFIG_NET_CLS_ACT
3310 if (exts->actions) {
3311 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3312 kfree(exts->actions);
3313 }
3314 exts->nr_actions = 0;
3315#endif
3316}
3317EXPORT_SYMBOL(tcf_exts_destroy);
3318
3319int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3320 struct nlattr *rate_tlv, struct tcf_exts *exts,
3321 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3322{
3323#ifdef CONFIG_NET_CLS_ACT
3324 {
3325 int init_res[TCA_ACT_MAX_PRIO] = {};
3326 struct tc_action *act;
3327 size_t attr_size = 0;
3328
3329 if (exts->police && tb[exts->police]) {
3330 struct tc_action_ops *a_o;
3331
3332 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3333 a_o = tc_action_load_ops(tb[exts->police], flags,
3334 extack);
3335 if (IS_ERR(a_o))
3336 return PTR_ERR(a_o);
3337 act = tcf_action_init_1(net, tp, tb[exts->police],
3338 rate_tlv, a_o, init_res, flags,
3339 extack);
3340 module_put(a_o->owner);
3341 if (IS_ERR(act))
3342 return PTR_ERR(act);
3343
3344 act->type = exts->type = TCA_OLD_COMPAT;
3345 exts->actions[0] = act;
3346 exts->nr_actions = 1;
3347 tcf_idr_insert_many(exts->actions, init_res);
3348 } else if (exts->action && tb[exts->action]) {
3349 int err;
3350
3351 flags |= TCA_ACT_FLAGS_BIND;
3352 err = tcf_action_init(net, tp, tb[exts->action],
3353 rate_tlv, exts->actions, init_res,
3354 &attr_size, flags, fl_flags,
3355 extack);
3356 if (err < 0)
3357 return err;
3358 exts->nr_actions = err;
3359 }
3360 }
3361#else
3362 if ((exts->action && tb[exts->action]) ||
3363 (exts->police && tb[exts->police])) {
3364 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3365 return -EOPNOTSUPP;
3366 }
3367#endif
3368
3369 return 0;
3370}
3371EXPORT_SYMBOL(tcf_exts_validate_ex);
3372
3373int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3374 struct nlattr *rate_tlv, struct tcf_exts *exts,
3375 u32 flags, struct netlink_ext_ack *extack)
3376{
3377 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3378 flags, 0, extack);
3379}
3380EXPORT_SYMBOL(tcf_exts_validate);
3381
3382void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3383{
3384#ifdef CONFIG_NET_CLS_ACT
3385 struct tcf_exts old = *dst;
3386
3387 *dst = *src;
3388 tcf_exts_destroy(&old);
3389#endif
3390}
3391EXPORT_SYMBOL(tcf_exts_change);
3392
3393#ifdef CONFIG_NET_CLS_ACT
3394static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3395{
3396 if (exts->nr_actions == 0)
3397 return NULL;
3398 else
3399 return exts->actions[0];
3400}
3401#endif
3402
3403int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3404{
3405#ifdef CONFIG_NET_CLS_ACT
3406 struct nlattr *nest;
3407
3408 if (exts->action && tcf_exts_has_actions(exts)) {
3409 /*
3410 * again for backward compatible mode - we want
3411 * to work with both old and new modes of entering
3412 * tc data even if iproute2 was newer - jhs
3413 */
3414 if (exts->type != TCA_OLD_COMPAT) {
3415 nest = nla_nest_start_noflag(skb, exts->action);
3416 if (nest == NULL)
3417 goto nla_put_failure;
3418
3419 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3420 < 0)
3421 goto nla_put_failure;
3422 nla_nest_end(skb, nest);
3423 } else if (exts->police) {
3424 struct tc_action *act = tcf_exts_first_act(exts);
3425 nest = nla_nest_start_noflag(skb, exts->police);
3426 if (nest == NULL || !act)
3427 goto nla_put_failure;
3428 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3429 goto nla_put_failure;
3430 nla_nest_end(skb, nest);
3431 }
3432 }
3433 return 0;
3434
3435nla_put_failure:
3436 nla_nest_cancel(skb, nest);
3437 return -1;
3438#else
3439 return 0;
3440#endif
3441}
3442EXPORT_SYMBOL(tcf_exts_dump);
3443
3444int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3445{
3446#ifdef CONFIG_NET_CLS_ACT
3447 struct nlattr *nest;
3448
3449 if (!exts->action || !tcf_exts_has_actions(exts))
3450 return 0;
3451
3452 nest = nla_nest_start_noflag(skb, exts->action);
3453 if (!nest)
3454 goto nla_put_failure;
3455
3456 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3457 goto nla_put_failure;
3458 nla_nest_end(skb, nest);
3459 return 0;
3460
3461nla_put_failure:
3462 nla_nest_cancel(skb, nest);
3463 return -1;
3464#else
3465 return 0;
3466#endif
3467}
3468EXPORT_SYMBOL(tcf_exts_terse_dump);
3469
3470int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3471{
3472#ifdef CONFIG_NET_CLS_ACT
3473 struct tc_action *a = tcf_exts_first_act(exts);
3474 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3475 return -1;
3476#endif
3477 return 0;
3478}
3479EXPORT_SYMBOL(tcf_exts_dump_stats);
3480
3481static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3482{
3483 if (*flags & TCA_CLS_FLAGS_IN_HW)
3484 return;
3485 *flags |= TCA_CLS_FLAGS_IN_HW;
3486 atomic_inc(&block->offloadcnt);
3487}
3488
3489static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3490{
3491 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3492 return;
3493 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3494 atomic_dec(&block->offloadcnt);
3495}
3496
3497static void tc_cls_offload_cnt_update(struct tcf_block *block,
3498 struct tcf_proto *tp, u32 *cnt,
3499 u32 *flags, u32 diff, bool add)
3500{
3501 lockdep_assert_held(&block->cb_lock);
3502
3503 spin_lock(&tp->lock);
3504 if (add) {
3505 if (!*cnt)
3506 tcf_block_offload_inc(block, flags);
3507 *cnt += diff;
3508 } else {
3509 *cnt -= diff;
3510 if (!*cnt)
3511 tcf_block_offload_dec(block, flags);
3512 }
3513 spin_unlock(&tp->lock);
3514}
3515
3516static void
3517tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3518 u32 *cnt, u32 *flags)
3519{
3520 lockdep_assert_held(&block->cb_lock);
3521
3522 spin_lock(&tp->lock);
3523 tcf_block_offload_dec(block, flags);
3524 *cnt = 0;
3525 spin_unlock(&tp->lock);
3526}
3527
3528static int
3529__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3530 void *type_data, bool err_stop)
3531{
3532 struct flow_block_cb *block_cb;
3533 int ok_count = 0;
3534 int err;
3535
3536 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3537 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3538 if (err) {
3539 if (err_stop)
3540 return err;
3541 } else {
3542 ok_count++;
3543 }
3544 }
3545 return ok_count;
3546}
3547
3548int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3549 void *type_data, bool err_stop, bool rtnl_held)
3550{
3551 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3552 int ok_count;
3553
3554retry:
3555 if (take_rtnl)
3556 rtnl_lock();
3557 down_read(&block->cb_lock);
3558 /* Need to obtain rtnl lock if block is bound to devs that require it.
3559 * In block bind code cb_lock is obtained while holding rtnl, so we must
3560 * obtain the locks in same order here.
3561 */
3562 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3563 up_read(&block->cb_lock);
3564 take_rtnl = true;
3565 goto retry;
3566 }
3567
3568 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3569
3570 up_read(&block->cb_lock);
3571 if (take_rtnl)
3572 rtnl_unlock();
3573 return ok_count;
3574}
3575EXPORT_SYMBOL(tc_setup_cb_call);
3576
3577/* Non-destructive filter add. If filter that wasn't already in hardware is
3578 * successfully offloaded, increment block offloads counter. On failure,
3579 * previously offloaded filter is considered to be intact and offloads counter
3580 * is not decremented.
3581 */
3582
3583int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3584 enum tc_setup_type type, void *type_data, bool err_stop,
3585 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3586{
3587 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3588 int ok_count;
3589
3590retry:
3591 if (take_rtnl)
3592 rtnl_lock();
3593 down_read(&block->cb_lock);
3594 /* Need to obtain rtnl lock if block is bound to devs that require it.
3595 * In block bind code cb_lock is obtained while holding rtnl, so we must
3596 * obtain the locks in same order here.
3597 */
3598 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3599 up_read(&block->cb_lock);
3600 take_rtnl = true;
3601 goto retry;
3602 }
3603
3604 /* Make sure all netdevs sharing this block are offload-capable. */
3605 if (block->nooffloaddevcnt && err_stop) {
3606 ok_count = -EOPNOTSUPP;
3607 goto err_unlock;
3608 }
3609
3610 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3611 if (ok_count < 0)
3612 goto err_unlock;
3613
3614 if (tp->ops->hw_add)
3615 tp->ops->hw_add(tp, type_data);
3616 if (ok_count > 0)
3617 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3618 ok_count, true);
3619err_unlock:
3620 up_read(&block->cb_lock);
3621 if (take_rtnl)
3622 rtnl_unlock();
3623 return min(ok_count, 0);
3624}
3625EXPORT_SYMBOL(tc_setup_cb_add);
3626
3627/* Destructive filter replace. If filter that wasn't already in hardware is
3628 * successfully offloaded, increment block offload counter. On failure,
3629 * previously offloaded filter is considered to be destroyed and offload counter
3630 * is decremented.
3631 */
3632
3633int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3634 enum tc_setup_type type, void *type_data, bool err_stop,
3635 u32 *old_flags, unsigned int *old_in_hw_count,
3636 u32 *new_flags, unsigned int *new_in_hw_count,
3637 bool rtnl_held)
3638{
3639 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3640 int ok_count;
3641
3642retry:
3643 if (take_rtnl)
3644 rtnl_lock();
3645 down_read(&block->cb_lock);
3646 /* Need to obtain rtnl lock if block is bound to devs that require it.
3647 * In block bind code cb_lock is obtained while holding rtnl, so we must
3648 * obtain the locks in same order here.
3649 */
3650 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3651 up_read(&block->cb_lock);
3652 take_rtnl = true;
3653 goto retry;
3654 }
3655
3656 /* Make sure all netdevs sharing this block are offload-capable. */
3657 if (block->nooffloaddevcnt && err_stop) {
3658 ok_count = -EOPNOTSUPP;
3659 goto err_unlock;
3660 }
3661
3662 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3663 if (tp->ops->hw_del)
3664 tp->ops->hw_del(tp, type_data);
3665
3666 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3667 if (ok_count < 0)
3668 goto err_unlock;
3669
3670 if (tp->ops->hw_add)
3671 tp->ops->hw_add(tp, type_data);
3672 if (ok_count > 0)
3673 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3674 new_flags, ok_count, true);
3675err_unlock:
3676 up_read(&block->cb_lock);
3677 if (take_rtnl)
3678 rtnl_unlock();
3679 return min(ok_count, 0);
3680}
3681EXPORT_SYMBOL(tc_setup_cb_replace);
3682
3683/* Destroy filter and decrement block offload counter, if filter was previously
3684 * offloaded.
3685 */
3686
3687int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3688 enum tc_setup_type type, void *type_data, bool err_stop,
3689 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3690{
3691 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3692 int ok_count;
3693
3694retry:
3695 if (take_rtnl)
3696 rtnl_lock();
3697 down_read(&block->cb_lock);
3698 /* Need to obtain rtnl lock if block is bound to devs that require it.
3699 * In block bind code cb_lock is obtained while holding rtnl, so we must
3700 * obtain the locks in same order here.
3701 */
3702 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3703 up_read(&block->cb_lock);
3704 take_rtnl = true;
3705 goto retry;
3706 }
3707
3708 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3709
3710 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3711 if (tp->ops->hw_del)
3712 tp->ops->hw_del(tp, type_data);
3713
3714 up_read(&block->cb_lock);
3715 if (take_rtnl)
3716 rtnl_unlock();
3717 return min(ok_count, 0);
3718}
3719EXPORT_SYMBOL(tc_setup_cb_destroy);
3720
3721int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3722 bool add, flow_setup_cb_t *cb,
3723 enum tc_setup_type type, void *type_data,
3724 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3725{
3726 int err = cb(type, type_data, cb_priv);
3727
3728 if (err) {
3729 if (add && tc_skip_sw(*flags))
3730 return err;
3731 } else {
3732 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3733 add);
3734 }
3735
3736 return 0;
3737}
3738EXPORT_SYMBOL(tc_setup_cb_reoffload);
3739
3740static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3741 const struct tc_action *act)
3742{
3743 struct tc_cookie *user_cookie;
3744 int err = 0;
3745
3746 rcu_read_lock();
3747 user_cookie = rcu_dereference(act->user_cookie);
3748 if (user_cookie) {
3749 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3750 user_cookie->len,
3751 GFP_ATOMIC);
3752 if (!entry->user_cookie)
3753 err = -ENOMEM;
3754 }
3755 rcu_read_unlock();
3756 return err;
3757}
3758
3759static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3760{
3761 flow_action_cookie_destroy(entry->user_cookie);
3762}
3763
3764void tc_cleanup_offload_action(struct flow_action *flow_action)
3765{
3766 struct flow_action_entry *entry;
3767 int i;
3768
3769 flow_action_for_each(i, entry, flow_action) {
3770 tcf_act_put_user_cookie(entry);
3771 if (entry->destructor)
3772 entry->destructor(entry->destructor_priv);
3773 }
3774}
3775EXPORT_SYMBOL(tc_cleanup_offload_action);
3776
3777static int tc_setup_offload_act(struct tc_action *act,
3778 struct flow_action_entry *entry,
3779 u32 *index_inc,
3780 struct netlink_ext_ack *extack)
3781{
3782#ifdef CONFIG_NET_CLS_ACT
3783 if (act->ops->offload_act_setup) {
3784 return act->ops->offload_act_setup(act, entry, index_inc, true,
3785 extack);
3786 } else {
3787 NL_SET_ERR_MSG(extack, "Action does not support offload");
3788 return -EOPNOTSUPP;
3789 }
3790#else
3791 return 0;
3792#endif
3793}
3794
3795int tc_setup_action(struct flow_action *flow_action,
3796 struct tc_action *actions[],
3797 u32 miss_cookie_base,
3798 struct netlink_ext_ack *extack)
3799{
3800 int i, j, k, index, err = 0;
3801 struct tc_action *act;
3802
3803 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3804 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3805 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3806
3807 if (!actions)
3808 return 0;
3809
3810 j = 0;
3811 tcf_act_for_each_action(i, act, actions) {
3812 struct flow_action_entry *entry;
3813
3814 entry = &flow_action->entries[j];
3815 spin_lock_bh(&act->tcfa_lock);
3816 err = tcf_act_get_user_cookie(entry, act);
3817 if (err)
3818 goto err_out_locked;
3819
3820 index = 0;
3821 err = tc_setup_offload_act(act, entry, &index, extack);
3822 if (err)
3823 goto err_out_locked;
3824
3825 for (k = 0; k < index ; k++) {
3826 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3827 entry[k].hw_index = act->tcfa_index;
3828 entry[k].cookie = (unsigned long)act;
3829 entry[k].miss_cookie =
3830 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3831 }
3832
3833 j += index;
3834
3835 spin_unlock_bh(&act->tcfa_lock);
3836 }
3837
3838err_out:
3839 if (err)
3840 tc_cleanup_offload_action(flow_action);
3841
3842 return err;
3843err_out_locked:
3844 spin_unlock_bh(&act->tcfa_lock);
3845 goto err_out;
3846}
3847
3848int tc_setup_offload_action(struct flow_action *flow_action,
3849 const struct tcf_exts *exts,
3850 struct netlink_ext_ack *extack)
3851{
3852#ifdef CONFIG_NET_CLS_ACT
3853 u32 miss_cookie_base;
3854
3855 if (!exts)
3856 return 0;
3857
3858 miss_cookie_base = exts->miss_cookie_node ?
3859 exts->miss_cookie_node->miss_cookie_base : 0;
3860 return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3861 extack);
3862#else
3863 return 0;
3864#endif
3865}
3866EXPORT_SYMBOL(tc_setup_offload_action);
3867
3868unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3869{
3870 unsigned int num_acts = 0;
3871 struct tc_action *act;
3872 int i;
3873
3874 tcf_exts_for_each_action(i, act, exts) {
3875 if (is_tcf_pedit(act))
3876 num_acts += tcf_pedit_nkeys(act);
3877 else
3878 num_acts++;
3879 }
3880 return num_acts;
3881}
3882EXPORT_SYMBOL(tcf_exts_num_actions);
3883
3884#ifdef CONFIG_NET_CLS_ACT
3885static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3886 u32 *p_block_index,
3887 struct netlink_ext_ack *extack)
3888{
3889 *p_block_index = nla_get_u32(block_index_attr);
3890 if (!*p_block_index) {
3891 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3892 return -EINVAL;
3893 }
3894
3895 return 0;
3896}
3897
3898int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3899 enum flow_block_binder_type binder_type,
3900 struct nlattr *block_index_attr,
3901 struct netlink_ext_ack *extack)
3902{
3903 u32 block_index;
3904 int err;
3905
3906 if (!block_index_attr)
3907 return 0;
3908
3909 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3910 if (err)
3911 return err;
3912
3913 qe->info.binder_type = binder_type;
3914 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3915 qe->info.chain_head_change_priv = &qe->filter_chain;
3916 qe->info.block_index = block_index;
3917
3918 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3919}
3920EXPORT_SYMBOL(tcf_qevent_init);
3921
3922void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3923{
3924 if (qe->info.block_index)
3925 tcf_block_put_ext(qe->block, sch, &qe->info);
3926}
3927EXPORT_SYMBOL(tcf_qevent_destroy);
3928
3929int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3930 struct netlink_ext_ack *extack)
3931{
3932 u32 block_index;
3933 int err;
3934
3935 if (!block_index_attr)
3936 return 0;
3937
3938 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3939 if (err)
3940 return err;
3941
3942 /* Bounce newly-configured block or change in block. */
3943 if (block_index != qe->info.block_index) {
3944 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3945 return -EINVAL;
3946 }
3947
3948 return 0;
3949}
3950EXPORT_SYMBOL(tcf_qevent_validate_change);
3951
3952struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3953 struct sk_buff **to_free, int *ret)
3954{
3955 struct tcf_result cl_res;
3956 struct tcf_proto *fl;
3957
3958 if (!qe->info.block_index)
3959 return skb;
3960
3961 fl = rcu_dereference_bh(qe->filter_chain);
3962
3963 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3964 case TC_ACT_SHOT:
3965 qdisc_qstats_drop(sch);
3966 __qdisc_drop(skb, to_free);
3967 *ret = __NET_XMIT_BYPASS;
3968 return NULL;
3969 case TC_ACT_STOLEN:
3970 case TC_ACT_QUEUED:
3971 case TC_ACT_TRAP:
3972 __qdisc_drop(skb, to_free);
3973 *ret = __NET_XMIT_STOLEN;
3974 return NULL;
3975 case TC_ACT_REDIRECT:
3976 skb_do_redirect(skb);
3977 *ret = __NET_XMIT_STOLEN;
3978 return NULL;
3979 }
3980
3981 return skb;
3982}
3983EXPORT_SYMBOL(tcf_qevent_handle);
3984
3985int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3986{
3987 if (!qe->info.block_index)
3988 return 0;
3989 return nla_put_u32(skb, attr_name, qe->info.block_index);
3990}
3991EXPORT_SYMBOL(tcf_qevent_dump);
3992#endif
3993
3994static __net_init int tcf_net_init(struct net *net)
3995{
3996 struct tcf_net *tn = net_generic(net, tcf_net_id);
3997
3998 spin_lock_init(&tn->idr_lock);
3999 idr_init(&tn->idr);
4000 return 0;
4001}
4002
4003static void __net_exit tcf_net_exit(struct net *net)
4004{
4005 struct tcf_net *tn = net_generic(net, tcf_net_id);
4006
4007 idr_destroy(&tn->idr);
4008}
4009
4010static struct pernet_operations tcf_net_ops = {
4011 .init = tcf_net_init,
4012 .exit = tcf_net_exit,
4013 .id = &tcf_net_id,
4014 .size = sizeof(struct tcf_net),
4015};
4016
4017static int __init tc_filter_init(void)
4018{
4019 int err;
4020
4021 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4022 if (!tc_filter_wq)
4023 return -ENOMEM;
4024
4025 err = register_pernet_subsys(&tcf_net_ops);
4026 if (err)
4027 goto err_register_pernet_subsys;
4028
4029 xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4030
4031 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
4032 RTNL_FLAG_DOIT_UNLOCKED);
4033 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
4034 RTNL_FLAG_DOIT_UNLOCKED);
4035 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
4036 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
4037 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
4038 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
4039 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
4040 tc_dump_chain, 0);
4041
4042 return 0;
4043
4044err_register_pernet_subsys:
4045 destroy_workqueue(tc_filter_wq);
4046 return err;
4047}
4048
4049subsys_initcall(tc_filter_init);