Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/cls_api.c Packet classifier API.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Changes:
8 *
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10 */
11
12#include <linux/module.h>
13#include <linux/types.h>
14#include <linux/kernel.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/err.h>
18#include <linux/skbuff.h>
19#include <linux/init.h>
20#include <linux/kmod.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/jhash.h>
24#include <linux/rculist.h>
25#include <net/net_namespace.h>
26#include <net/sock.h>
27#include <net/netlink.h>
28#include <net/pkt_sched.h>
29#include <net/pkt_cls.h>
30#include <net/tc_act/tc_pedit.h>
31#include <net/tc_act/tc_mirred.h>
32#include <net/tc_act/tc_vlan.h>
33#include <net/tc_act/tc_tunnel_key.h>
34#include <net/tc_act/tc_csum.h>
35#include <net/tc_act/tc_gact.h>
36#include <net/tc_act/tc_police.h>
37#include <net/tc_act/tc_sample.h>
38#include <net/tc_act/tc_skbedit.h>
39#include <net/tc_act/tc_ct.h>
40#include <net/tc_act/tc_mpls.h>
41#include <net/tc_act/tc_gate.h>
42#include <net/flow_offload.h>
43#include <net/tc_wrapper.h>
44
45extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
46
47/* The list of all installed classifier types */
48static LIST_HEAD(tcf_proto_base);
49
50/* Protects list of registered TC modules. It is pure SMP lock. */
51static DEFINE_RWLOCK(cls_mod_lock);
52
53#ifdef CONFIG_NET_CLS_ACT
54DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
55EXPORT_SYMBOL(tc_skb_ext_tc);
56
57void tc_skb_ext_tc_enable(void)
58{
59 static_branch_inc(&tc_skb_ext_tc);
60}
61EXPORT_SYMBOL(tc_skb_ext_tc_enable);
62
63void tc_skb_ext_tc_disable(void)
64{
65 static_branch_dec(&tc_skb_ext_tc);
66}
67EXPORT_SYMBOL(tc_skb_ext_tc_disable);
68#endif
69
70static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
71{
72 return jhash_3words(tp->chain->index, tp->prio,
73 (__force __u32)tp->protocol, 0);
74}
75
76static void tcf_proto_signal_destroying(struct tcf_chain *chain,
77 struct tcf_proto *tp)
78{
79 struct tcf_block *block = chain->block;
80
81 mutex_lock(&block->proto_destroy_lock);
82 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
83 destroy_obj_hashfn(tp));
84 mutex_unlock(&block->proto_destroy_lock);
85}
86
87static bool tcf_proto_cmp(const struct tcf_proto *tp1,
88 const struct tcf_proto *tp2)
89{
90 return tp1->chain->index == tp2->chain->index &&
91 tp1->prio == tp2->prio &&
92 tp1->protocol == tp2->protocol;
93}
94
95static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
96 struct tcf_proto *tp)
97{
98 u32 hash = destroy_obj_hashfn(tp);
99 struct tcf_proto *iter;
100 bool found = false;
101
102 rcu_read_lock();
103 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
104 destroy_ht_node, hash) {
105 if (tcf_proto_cmp(tp, iter)) {
106 found = true;
107 break;
108 }
109 }
110 rcu_read_unlock();
111
112 return found;
113}
114
115static void
116tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
117{
118 struct tcf_block *block = chain->block;
119
120 mutex_lock(&block->proto_destroy_lock);
121 if (hash_hashed(&tp->destroy_ht_node))
122 hash_del_rcu(&tp->destroy_ht_node);
123 mutex_unlock(&block->proto_destroy_lock);
124}
125
126/* Find classifier type by string name */
127
128static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
129{
130 const struct tcf_proto_ops *t, *res = NULL;
131
132 if (kind) {
133 read_lock(&cls_mod_lock);
134 list_for_each_entry(t, &tcf_proto_base, head) {
135 if (strcmp(kind, t->kind) == 0) {
136 if (try_module_get(t->owner))
137 res = t;
138 break;
139 }
140 }
141 read_unlock(&cls_mod_lock);
142 }
143 return res;
144}
145
146static const struct tcf_proto_ops *
147tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
148 struct netlink_ext_ack *extack)
149{
150 const struct tcf_proto_ops *ops;
151
152 ops = __tcf_proto_lookup_ops(kind);
153 if (ops)
154 return ops;
155#ifdef CONFIG_MODULES
156 if (rtnl_held)
157 rtnl_unlock();
158 request_module("cls_%s", kind);
159 if (rtnl_held)
160 rtnl_lock();
161 ops = __tcf_proto_lookup_ops(kind);
162 /* We dropped the RTNL semaphore in order to perform
163 * the module load. So, even if we succeeded in loading
164 * the module we have to replay the request. We indicate
165 * this using -EAGAIN.
166 */
167 if (ops) {
168 module_put(ops->owner);
169 return ERR_PTR(-EAGAIN);
170 }
171#endif
172 NL_SET_ERR_MSG(extack, "TC classifier not found");
173 return ERR_PTR(-ENOENT);
174}
175
176/* Register(unregister) new classifier type */
177
178int register_tcf_proto_ops(struct tcf_proto_ops *ops)
179{
180 struct tcf_proto_ops *t;
181 int rc = -EEXIST;
182
183 write_lock(&cls_mod_lock);
184 list_for_each_entry(t, &tcf_proto_base, head)
185 if (!strcmp(ops->kind, t->kind))
186 goto out;
187
188 list_add_tail(&ops->head, &tcf_proto_base);
189 rc = 0;
190out:
191 write_unlock(&cls_mod_lock);
192 return rc;
193}
194EXPORT_SYMBOL(register_tcf_proto_ops);
195
196static struct workqueue_struct *tc_filter_wq;
197
198void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
199{
200 struct tcf_proto_ops *t;
201 int rc = -ENOENT;
202
203 /* Wait for outstanding call_rcu()s, if any, from a
204 * tcf_proto_ops's destroy() handler.
205 */
206 rcu_barrier();
207 flush_workqueue(tc_filter_wq);
208
209 write_lock(&cls_mod_lock);
210 list_for_each_entry(t, &tcf_proto_base, head) {
211 if (t == ops) {
212 list_del(&t->head);
213 rc = 0;
214 break;
215 }
216 }
217 write_unlock(&cls_mod_lock);
218
219 WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
220}
221EXPORT_SYMBOL(unregister_tcf_proto_ops);
222
223bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
224{
225 INIT_RCU_WORK(rwork, func);
226 return queue_rcu_work(tc_filter_wq, rwork);
227}
228EXPORT_SYMBOL(tcf_queue_work);
229
230/* Select new prio value from the range, managed by kernel. */
231
232static inline u32 tcf_auto_prio(struct tcf_proto *tp)
233{
234 u32 first = TC_H_MAKE(0xC0000000U, 0U);
235
236 if (tp)
237 first = tp->prio - 1;
238
239 return TC_H_MAJ(first);
240}
241
242static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
243{
244 if (kind)
245 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
246 memset(name, 0, IFNAMSIZ);
247 return false;
248}
249
250static bool tcf_proto_is_unlocked(const char *kind)
251{
252 const struct tcf_proto_ops *ops;
253 bool ret;
254
255 if (strlen(kind) == 0)
256 return false;
257
258 ops = tcf_proto_lookup_ops(kind, false, NULL);
259 /* On error return false to take rtnl lock. Proto lookup/create
260 * functions will perform lookup again and properly handle errors.
261 */
262 if (IS_ERR(ops))
263 return false;
264
265 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
266 module_put(ops->owner);
267 return ret;
268}
269
270static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
271 u32 prio, struct tcf_chain *chain,
272 bool rtnl_held,
273 struct netlink_ext_ack *extack)
274{
275 struct tcf_proto *tp;
276 int err;
277
278 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
279 if (!tp)
280 return ERR_PTR(-ENOBUFS);
281
282 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
283 if (IS_ERR(tp->ops)) {
284 err = PTR_ERR(tp->ops);
285 goto errout;
286 }
287 tp->classify = tp->ops->classify;
288 tp->protocol = protocol;
289 tp->prio = prio;
290 tp->chain = chain;
291 spin_lock_init(&tp->lock);
292 refcount_set(&tp->refcnt, 1);
293
294 err = tp->ops->init(tp);
295 if (err) {
296 module_put(tp->ops->owner);
297 goto errout;
298 }
299 return tp;
300
301errout:
302 kfree(tp);
303 return ERR_PTR(err);
304}
305
306static void tcf_proto_get(struct tcf_proto *tp)
307{
308 refcount_inc(&tp->refcnt);
309}
310
311static void tcf_chain_put(struct tcf_chain *chain);
312
313static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
314 bool sig_destroy, struct netlink_ext_ack *extack)
315{
316 tp->ops->destroy(tp, rtnl_held, extack);
317 if (sig_destroy)
318 tcf_proto_signal_destroyed(tp->chain, tp);
319 tcf_chain_put(tp->chain);
320 module_put(tp->ops->owner);
321 kfree_rcu(tp, rcu);
322}
323
324static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
325 struct netlink_ext_ack *extack)
326{
327 if (refcount_dec_and_test(&tp->refcnt))
328 tcf_proto_destroy(tp, rtnl_held, true, extack);
329}
330
331static bool tcf_proto_check_delete(struct tcf_proto *tp)
332{
333 if (tp->ops->delete_empty)
334 return tp->ops->delete_empty(tp);
335
336 tp->deleting = true;
337 return tp->deleting;
338}
339
340static void tcf_proto_mark_delete(struct tcf_proto *tp)
341{
342 spin_lock(&tp->lock);
343 tp->deleting = true;
344 spin_unlock(&tp->lock);
345}
346
347static bool tcf_proto_is_deleting(struct tcf_proto *tp)
348{
349 bool deleting;
350
351 spin_lock(&tp->lock);
352 deleting = tp->deleting;
353 spin_unlock(&tp->lock);
354
355 return deleting;
356}
357
358#define ASSERT_BLOCK_LOCKED(block) \
359 lockdep_assert_held(&(block)->lock)
360
361struct tcf_filter_chain_list_item {
362 struct list_head list;
363 tcf_chain_head_change_t *chain_head_change;
364 void *chain_head_change_priv;
365};
366
367static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
368 u32 chain_index)
369{
370 struct tcf_chain *chain;
371
372 ASSERT_BLOCK_LOCKED(block);
373
374 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
375 if (!chain)
376 return NULL;
377 list_add_tail_rcu(&chain->list, &block->chain_list);
378 mutex_init(&chain->filter_chain_lock);
379 chain->block = block;
380 chain->index = chain_index;
381 chain->refcnt = 1;
382 if (!chain->index)
383 block->chain0.chain = chain;
384 return chain;
385}
386
387static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
388 struct tcf_proto *tp_head)
389{
390 if (item->chain_head_change)
391 item->chain_head_change(tp_head, item->chain_head_change_priv);
392}
393
394static void tcf_chain0_head_change(struct tcf_chain *chain,
395 struct tcf_proto *tp_head)
396{
397 struct tcf_filter_chain_list_item *item;
398 struct tcf_block *block = chain->block;
399
400 if (chain->index)
401 return;
402
403 mutex_lock(&block->lock);
404 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
405 tcf_chain_head_change_item(item, tp_head);
406 mutex_unlock(&block->lock);
407}
408
409/* Returns true if block can be safely freed. */
410
411static bool tcf_chain_detach(struct tcf_chain *chain)
412{
413 struct tcf_block *block = chain->block;
414
415 ASSERT_BLOCK_LOCKED(block);
416
417 list_del_rcu(&chain->list);
418 if (!chain->index)
419 block->chain0.chain = NULL;
420
421 if (list_empty(&block->chain_list) &&
422 refcount_read(&block->refcnt) == 0)
423 return true;
424
425 return false;
426}
427
428static void tcf_block_destroy(struct tcf_block *block)
429{
430 mutex_destroy(&block->lock);
431 mutex_destroy(&block->proto_destroy_lock);
432 kfree_rcu(block, rcu);
433}
434
435static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
436{
437 struct tcf_block *block = chain->block;
438
439 mutex_destroy(&chain->filter_chain_lock);
440 kfree_rcu(chain, rcu);
441 if (free_block)
442 tcf_block_destroy(block);
443}
444
445static void tcf_chain_hold(struct tcf_chain *chain)
446{
447 ASSERT_BLOCK_LOCKED(chain->block);
448
449 ++chain->refcnt;
450}
451
452static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
453{
454 ASSERT_BLOCK_LOCKED(chain->block);
455
456 /* In case all the references are action references, this
457 * chain should not be shown to the user.
458 */
459 return chain->refcnt == chain->action_refcnt;
460}
461
462static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
463 u32 chain_index)
464{
465 struct tcf_chain *chain;
466
467 ASSERT_BLOCK_LOCKED(block);
468
469 list_for_each_entry(chain, &block->chain_list, list) {
470 if (chain->index == chain_index)
471 return chain;
472 }
473 return NULL;
474}
475
476#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
477static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
478 u32 chain_index)
479{
480 struct tcf_chain *chain;
481
482 list_for_each_entry_rcu(chain, &block->chain_list, list) {
483 if (chain->index == chain_index)
484 return chain;
485 }
486 return NULL;
487}
488#endif
489
490static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
491 u32 seq, u16 flags, int event, bool unicast);
492
493static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
494 u32 chain_index, bool create,
495 bool by_act)
496{
497 struct tcf_chain *chain = NULL;
498 bool is_first_reference;
499
500 mutex_lock(&block->lock);
501 chain = tcf_chain_lookup(block, chain_index);
502 if (chain) {
503 tcf_chain_hold(chain);
504 } else {
505 if (!create)
506 goto errout;
507 chain = tcf_chain_create(block, chain_index);
508 if (!chain)
509 goto errout;
510 }
511
512 if (by_act)
513 ++chain->action_refcnt;
514 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
515 mutex_unlock(&block->lock);
516
517 /* Send notification only in case we got the first
518 * non-action reference. Until then, the chain acts only as
519 * a placeholder for actions pointing to it and user ought
520 * not know about them.
521 */
522 if (is_first_reference && !by_act)
523 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
524 RTM_NEWCHAIN, false);
525
526 return chain;
527
528errout:
529 mutex_unlock(&block->lock);
530 return chain;
531}
532
533static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
534 bool create)
535{
536 return __tcf_chain_get(block, chain_index, create, false);
537}
538
539struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
540{
541 return __tcf_chain_get(block, chain_index, true, true);
542}
543EXPORT_SYMBOL(tcf_chain_get_by_act);
544
545static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
546 void *tmplt_priv);
547static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
548 void *tmplt_priv, u32 chain_index,
549 struct tcf_block *block, struct sk_buff *oskb,
550 u32 seq, u16 flags, bool unicast);
551
552static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
553 bool explicitly_created)
554{
555 struct tcf_block *block = chain->block;
556 const struct tcf_proto_ops *tmplt_ops;
557 bool free_block = false;
558 unsigned int refcnt;
559 void *tmplt_priv;
560
561 mutex_lock(&block->lock);
562 if (explicitly_created) {
563 if (!chain->explicitly_created) {
564 mutex_unlock(&block->lock);
565 return;
566 }
567 chain->explicitly_created = false;
568 }
569
570 if (by_act)
571 chain->action_refcnt--;
572
573 /* tc_chain_notify_delete can't be called while holding block lock.
574 * However, when block is unlocked chain can be changed concurrently, so
575 * save these to temporary variables.
576 */
577 refcnt = --chain->refcnt;
578 tmplt_ops = chain->tmplt_ops;
579 tmplt_priv = chain->tmplt_priv;
580
581 /* The last dropped non-action reference will trigger notification. */
582 if (refcnt - chain->action_refcnt == 0 && !by_act) {
583 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
584 block, NULL, 0, 0, false);
585 /* Last reference to chain, no need to lock. */
586 chain->flushing = false;
587 }
588
589 if (refcnt == 0)
590 free_block = tcf_chain_detach(chain);
591 mutex_unlock(&block->lock);
592
593 if (refcnt == 0) {
594 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
595 tcf_chain_destroy(chain, free_block);
596 }
597}
598
599static void tcf_chain_put(struct tcf_chain *chain)
600{
601 __tcf_chain_put(chain, false, false);
602}
603
604void tcf_chain_put_by_act(struct tcf_chain *chain)
605{
606 __tcf_chain_put(chain, true, false);
607}
608EXPORT_SYMBOL(tcf_chain_put_by_act);
609
610static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
611{
612 __tcf_chain_put(chain, false, true);
613}
614
615static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
616{
617 struct tcf_proto *tp, *tp_next;
618
619 mutex_lock(&chain->filter_chain_lock);
620 tp = tcf_chain_dereference(chain->filter_chain, chain);
621 while (tp) {
622 tp_next = rcu_dereference_protected(tp->next, 1);
623 tcf_proto_signal_destroying(chain, tp);
624 tp = tp_next;
625 }
626 tp = tcf_chain_dereference(chain->filter_chain, chain);
627 RCU_INIT_POINTER(chain->filter_chain, NULL);
628 tcf_chain0_head_change(chain, NULL);
629 chain->flushing = true;
630 mutex_unlock(&chain->filter_chain_lock);
631
632 while (tp) {
633 tp_next = rcu_dereference_protected(tp->next, 1);
634 tcf_proto_put(tp, rtnl_held, NULL);
635 tp = tp_next;
636 }
637}
638
639static int tcf_block_setup(struct tcf_block *block,
640 struct flow_block_offload *bo);
641
642static void tcf_block_offload_init(struct flow_block_offload *bo,
643 struct net_device *dev, struct Qdisc *sch,
644 enum flow_block_command command,
645 enum flow_block_binder_type binder_type,
646 struct flow_block *flow_block,
647 bool shared, struct netlink_ext_ack *extack)
648{
649 bo->net = dev_net(dev);
650 bo->command = command;
651 bo->binder_type = binder_type;
652 bo->block = flow_block;
653 bo->block_shared = shared;
654 bo->extack = extack;
655 bo->sch = sch;
656 bo->cb_list_head = &flow_block->cb_list;
657 INIT_LIST_HEAD(&bo->cb_list);
658}
659
660static void tcf_block_unbind(struct tcf_block *block,
661 struct flow_block_offload *bo);
662
663static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
664{
665 struct tcf_block *block = block_cb->indr.data;
666 struct net_device *dev = block_cb->indr.dev;
667 struct Qdisc *sch = block_cb->indr.sch;
668 struct netlink_ext_ack extack = {};
669 struct flow_block_offload bo = {};
670
671 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
672 block_cb->indr.binder_type,
673 &block->flow_block, tcf_block_shared(block),
674 &extack);
675 rtnl_lock();
676 down_write(&block->cb_lock);
677 list_del(&block_cb->driver_list);
678 list_move(&block_cb->list, &bo.cb_list);
679 tcf_block_unbind(block, &bo);
680 up_write(&block->cb_lock);
681 rtnl_unlock();
682}
683
684static bool tcf_block_offload_in_use(struct tcf_block *block)
685{
686 return atomic_read(&block->offloadcnt);
687}
688
689static int tcf_block_offload_cmd(struct tcf_block *block,
690 struct net_device *dev, struct Qdisc *sch,
691 struct tcf_block_ext_info *ei,
692 enum flow_block_command command,
693 struct netlink_ext_ack *extack)
694{
695 struct flow_block_offload bo = {};
696
697 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
698 &block->flow_block, tcf_block_shared(block),
699 extack);
700
701 if (dev->netdev_ops->ndo_setup_tc) {
702 int err;
703
704 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
705 if (err < 0) {
706 if (err != -EOPNOTSUPP)
707 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
708 return err;
709 }
710
711 return tcf_block_setup(block, &bo);
712 }
713
714 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
715 tc_block_indr_cleanup);
716 tcf_block_setup(block, &bo);
717
718 return -EOPNOTSUPP;
719}
720
721static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
722 struct tcf_block_ext_info *ei,
723 struct netlink_ext_ack *extack)
724{
725 struct net_device *dev = q->dev_queue->dev;
726 int err;
727
728 down_write(&block->cb_lock);
729
730 /* If tc offload feature is disabled and the block we try to bind
731 * to already has some offloaded filters, forbid to bind.
732 */
733 if (dev->netdev_ops->ndo_setup_tc &&
734 !tc_can_offload(dev) &&
735 tcf_block_offload_in_use(block)) {
736 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
737 err = -EOPNOTSUPP;
738 goto err_unlock;
739 }
740
741 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
742 if (err == -EOPNOTSUPP)
743 goto no_offload_dev_inc;
744 if (err)
745 goto err_unlock;
746
747 up_write(&block->cb_lock);
748 return 0;
749
750no_offload_dev_inc:
751 if (tcf_block_offload_in_use(block))
752 goto err_unlock;
753
754 err = 0;
755 block->nooffloaddevcnt++;
756err_unlock:
757 up_write(&block->cb_lock);
758 return err;
759}
760
761static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
762 struct tcf_block_ext_info *ei)
763{
764 struct net_device *dev = q->dev_queue->dev;
765 int err;
766
767 down_write(&block->cb_lock);
768 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
769 if (err == -EOPNOTSUPP)
770 goto no_offload_dev_dec;
771 up_write(&block->cb_lock);
772 return;
773
774no_offload_dev_dec:
775 WARN_ON(block->nooffloaddevcnt-- == 0);
776 up_write(&block->cb_lock);
777}
778
779static int
780tcf_chain0_head_change_cb_add(struct tcf_block *block,
781 struct tcf_block_ext_info *ei,
782 struct netlink_ext_ack *extack)
783{
784 struct tcf_filter_chain_list_item *item;
785 struct tcf_chain *chain0;
786
787 item = kmalloc(sizeof(*item), GFP_KERNEL);
788 if (!item) {
789 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
790 return -ENOMEM;
791 }
792 item->chain_head_change = ei->chain_head_change;
793 item->chain_head_change_priv = ei->chain_head_change_priv;
794
795 mutex_lock(&block->lock);
796 chain0 = block->chain0.chain;
797 if (chain0)
798 tcf_chain_hold(chain0);
799 else
800 list_add(&item->list, &block->chain0.filter_chain_list);
801 mutex_unlock(&block->lock);
802
803 if (chain0) {
804 struct tcf_proto *tp_head;
805
806 mutex_lock(&chain0->filter_chain_lock);
807
808 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
809 if (tp_head)
810 tcf_chain_head_change_item(item, tp_head);
811
812 mutex_lock(&block->lock);
813 list_add(&item->list, &block->chain0.filter_chain_list);
814 mutex_unlock(&block->lock);
815
816 mutex_unlock(&chain0->filter_chain_lock);
817 tcf_chain_put(chain0);
818 }
819
820 return 0;
821}
822
823static void
824tcf_chain0_head_change_cb_del(struct tcf_block *block,
825 struct tcf_block_ext_info *ei)
826{
827 struct tcf_filter_chain_list_item *item;
828
829 mutex_lock(&block->lock);
830 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
831 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
832 (item->chain_head_change == ei->chain_head_change &&
833 item->chain_head_change_priv == ei->chain_head_change_priv)) {
834 if (block->chain0.chain)
835 tcf_chain_head_change_item(item, NULL);
836 list_del(&item->list);
837 mutex_unlock(&block->lock);
838
839 kfree(item);
840 return;
841 }
842 }
843 mutex_unlock(&block->lock);
844 WARN_ON(1);
845}
846
847struct tcf_net {
848 spinlock_t idr_lock; /* Protects idr */
849 struct idr idr;
850};
851
852static unsigned int tcf_net_id;
853
854static int tcf_block_insert(struct tcf_block *block, struct net *net,
855 struct netlink_ext_ack *extack)
856{
857 struct tcf_net *tn = net_generic(net, tcf_net_id);
858 int err;
859
860 idr_preload(GFP_KERNEL);
861 spin_lock(&tn->idr_lock);
862 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
863 GFP_NOWAIT);
864 spin_unlock(&tn->idr_lock);
865 idr_preload_end();
866
867 return err;
868}
869
870static void tcf_block_remove(struct tcf_block *block, struct net *net)
871{
872 struct tcf_net *tn = net_generic(net, tcf_net_id);
873
874 spin_lock(&tn->idr_lock);
875 idr_remove(&tn->idr, block->index);
876 spin_unlock(&tn->idr_lock);
877}
878
879static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
880 u32 block_index,
881 struct netlink_ext_ack *extack)
882{
883 struct tcf_block *block;
884
885 block = kzalloc(sizeof(*block), GFP_KERNEL);
886 if (!block) {
887 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
888 return ERR_PTR(-ENOMEM);
889 }
890 mutex_init(&block->lock);
891 mutex_init(&block->proto_destroy_lock);
892 init_rwsem(&block->cb_lock);
893 flow_block_init(&block->flow_block);
894 INIT_LIST_HEAD(&block->chain_list);
895 INIT_LIST_HEAD(&block->owner_list);
896 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
897
898 refcount_set(&block->refcnt, 1);
899 block->net = net;
900 block->index = block_index;
901
902 /* Don't store q pointer for blocks which are shared */
903 if (!tcf_block_shared(block))
904 block->q = q;
905 return block;
906}
907
908static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
909{
910 struct tcf_net *tn = net_generic(net, tcf_net_id);
911
912 return idr_find(&tn->idr, block_index);
913}
914
915static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
916{
917 struct tcf_block *block;
918
919 rcu_read_lock();
920 block = tcf_block_lookup(net, block_index);
921 if (block && !refcount_inc_not_zero(&block->refcnt))
922 block = NULL;
923 rcu_read_unlock();
924
925 return block;
926}
927
928static struct tcf_chain *
929__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
930{
931 mutex_lock(&block->lock);
932 if (chain)
933 chain = list_is_last(&chain->list, &block->chain_list) ?
934 NULL : list_next_entry(chain, list);
935 else
936 chain = list_first_entry_or_null(&block->chain_list,
937 struct tcf_chain, list);
938
939 /* skip all action-only chains */
940 while (chain && tcf_chain_held_by_acts_only(chain))
941 chain = list_is_last(&chain->list, &block->chain_list) ?
942 NULL : list_next_entry(chain, list);
943
944 if (chain)
945 tcf_chain_hold(chain);
946 mutex_unlock(&block->lock);
947
948 return chain;
949}
950
951/* Function to be used by all clients that want to iterate over all chains on
952 * block. It properly obtains block->lock and takes reference to chain before
953 * returning it. Users of this function must be tolerant to concurrent chain
954 * insertion/deletion or ensure that no concurrent chain modification is
955 * possible. Note that all netlink dump callbacks cannot guarantee to provide
956 * consistent dump because rtnl lock is released each time skb is filled with
957 * data and sent to user-space.
958 */
959
960struct tcf_chain *
961tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
962{
963 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
964
965 if (chain)
966 tcf_chain_put(chain);
967
968 return chain_next;
969}
970EXPORT_SYMBOL(tcf_get_next_chain);
971
972static struct tcf_proto *
973__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
974{
975 u32 prio = 0;
976
977 ASSERT_RTNL();
978 mutex_lock(&chain->filter_chain_lock);
979
980 if (!tp) {
981 tp = tcf_chain_dereference(chain->filter_chain, chain);
982 } else if (tcf_proto_is_deleting(tp)) {
983 /* 'deleting' flag is set and chain->filter_chain_lock was
984 * unlocked, which means next pointer could be invalid. Restart
985 * search.
986 */
987 prio = tp->prio + 1;
988 tp = tcf_chain_dereference(chain->filter_chain, chain);
989
990 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
991 if (!tp->deleting && tp->prio >= prio)
992 break;
993 } else {
994 tp = tcf_chain_dereference(tp->next, chain);
995 }
996
997 if (tp)
998 tcf_proto_get(tp);
999
1000 mutex_unlock(&chain->filter_chain_lock);
1001
1002 return tp;
1003}
1004
1005/* Function to be used by all clients that want to iterate over all tp's on
1006 * chain. Users of this function must be tolerant to concurrent tp
1007 * insertion/deletion or ensure that no concurrent chain modification is
1008 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1009 * consistent dump because rtnl lock is released each time skb is filled with
1010 * data and sent to user-space.
1011 */
1012
1013struct tcf_proto *
1014tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1015{
1016 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1017
1018 if (tp)
1019 tcf_proto_put(tp, true, NULL);
1020
1021 return tp_next;
1022}
1023EXPORT_SYMBOL(tcf_get_next_proto);
1024
1025static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1026{
1027 struct tcf_chain *chain;
1028
1029 /* Last reference to block. At this point chains cannot be added or
1030 * removed concurrently.
1031 */
1032 for (chain = tcf_get_next_chain(block, NULL);
1033 chain;
1034 chain = tcf_get_next_chain(block, chain)) {
1035 tcf_chain_put_explicitly_created(chain);
1036 tcf_chain_flush(chain, rtnl_held);
1037 }
1038}
1039
1040/* Lookup Qdisc and increments its reference counter.
1041 * Set parent, if necessary.
1042 */
1043
1044static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1045 u32 *parent, int ifindex, bool rtnl_held,
1046 struct netlink_ext_ack *extack)
1047{
1048 const struct Qdisc_class_ops *cops;
1049 struct net_device *dev;
1050 int err = 0;
1051
1052 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1053 return 0;
1054
1055 rcu_read_lock();
1056
1057 /* Find link */
1058 dev = dev_get_by_index_rcu(net, ifindex);
1059 if (!dev) {
1060 rcu_read_unlock();
1061 return -ENODEV;
1062 }
1063
1064 /* Find qdisc */
1065 if (!*parent) {
1066 *q = rcu_dereference(dev->qdisc);
1067 *parent = (*q)->handle;
1068 } else {
1069 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1070 if (!*q) {
1071 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1072 err = -EINVAL;
1073 goto errout_rcu;
1074 }
1075 }
1076
1077 *q = qdisc_refcount_inc_nz(*q);
1078 if (!*q) {
1079 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1080 err = -EINVAL;
1081 goto errout_rcu;
1082 }
1083
1084 /* Is it classful? */
1085 cops = (*q)->ops->cl_ops;
1086 if (!cops) {
1087 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1088 err = -EINVAL;
1089 goto errout_qdisc;
1090 }
1091
1092 if (!cops->tcf_block) {
1093 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1094 err = -EOPNOTSUPP;
1095 goto errout_qdisc;
1096 }
1097
1098errout_rcu:
1099 /* At this point we know that qdisc is not noop_qdisc,
1100 * which means that qdisc holds a reference to net_device
1101 * and we hold a reference to qdisc, so it is safe to release
1102 * rcu read lock.
1103 */
1104 rcu_read_unlock();
1105 return err;
1106
1107errout_qdisc:
1108 rcu_read_unlock();
1109
1110 if (rtnl_held)
1111 qdisc_put(*q);
1112 else
1113 qdisc_put_unlocked(*q);
1114 *q = NULL;
1115
1116 return err;
1117}
1118
1119static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1120 int ifindex, struct netlink_ext_ack *extack)
1121{
1122 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1123 return 0;
1124
1125 /* Do we search for filter, attached to class? */
1126 if (TC_H_MIN(parent)) {
1127 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1128
1129 *cl = cops->find(q, parent);
1130 if (*cl == 0) {
1131 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1132 return -ENOENT;
1133 }
1134 }
1135
1136 return 0;
1137}
1138
1139static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1140 unsigned long cl, int ifindex,
1141 u32 block_index,
1142 struct netlink_ext_ack *extack)
1143{
1144 struct tcf_block *block;
1145
1146 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1147 block = tcf_block_refcnt_get(net, block_index);
1148 if (!block) {
1149 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1150 return ERR_PTR(-EINVAL);
1151 }
1152 } else {
1153 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1154
1155 block = cops->tcf_block(q, cl, extack);
1156 if (!block)
1157 return ERR_PTR(-EINVAL);
1158
1159 if (tcf_block_shared(block)) {
1160 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1161 return ERR_PTR(-EOPNOTSUPP);
1162 }
1163
1164 /* Always take reference to block in order to support execution
1165 * of rules update path of cls API without rtnl lock. Caller
1166 * must release block when it is finished using it. 'if' block
1167 * of this conditional obtain reference to block by calling
1168 * tcf_block_refcnt_get().
1169 */
1170 refcount_inc(&block->refcnt);
1171 }
1172
1173 return block;
1174}
1175
1176static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1177 struct tcf_block_ext_info *ei, bool rtnl_held)
1178{
1179 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1180 /* Flushing/putting all chains will cause the block to be
1181 * deallocated when last chain is freed. However, if chain_list
1182 * is empty, block has to be manually deallocated. After block
1183 * reference counter reached 0, it is no longer possible to
1184 * increment it or add new chains to block.
1185 */
1186 bool free_block = list_empty(&block->chain_list);
1187
1188 mutex_unlock(&block->lock);
1189 if (tcf_block_shared(block))
1190 tcf_block_remove(block, block->net);
1191
1192 if (q)
1193 tcf_block_offload_unbind(block, q, ei);
1194
1195 if (free_block)
1196 tcf_block_destroy(block);
1197 else
1198 tcf_block_flush_all_chains(block, rtnl_held);
1199 } else if (q) {
1200 tcf_block_offload_unbind(block, q, ei);
1201 }
1202}
1203
1204static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1205{
1206 __tcf_block_put(block, NULL, NULL, rtnl_held);
1207}
1208
1209/* Find tcf block.
1210 * Set q, parent, cl when appropriate.
1211 */
1212
1213static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1214 u32 *parent, unsigned long *cl,
1215 int ifindex, u32 block_index,
1216 struct netlink_ext_ack *extack)
1217{
1218 struct tcf_block *block;
1219 int err = 0;
1220
1221 ASSERT_RTNL();
1222
1223 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1224 if (err)
1225 goto errout;
1226
1227 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1228 if (err)
1229 goto errout_qdisc;
1230
1231 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1232 if (IS_ERR(block)) {
1233 err = PTR_ERR(block);
1234 goto errout_qdisc;
1235 }
1236
1237 return block;
1238
1239errout_qdisc:
1240 if (*q)
1241 qdisc_put(*q);
1242errout:
1243 *q = NULL;
1244 return ERR_PTR(err);
1245}
1246
1247static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1248 bool rtnl_held)
1249{
1250 if (!IS_ERR_OR_NULL(block))
1251 tcf_block_refcnt_put(block, rtnl_held);
1252
1253 if (q) {
1254 if (rtnl_held)
1255 qdisc_put(q);
1256 else
1257 qdisc_put_unlocked(q);
1258 }
1259}
1260
1261struct tcf_block_owner_item {
1262 struct list_head list;
1263 struct Qdisc *q;
1264 enum flow_block_binder_type binder_type;
1265};
1266
1267static void
1268tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1269 struct Qdisc *q,
1270 enum flow_block_binder_type binder_type)
1271{
1272 if (block->keep_dst &&
1273 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1274 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1275 netif_keep_dst(qdisc_dev(q));
1276}
1277
1278void tcf_block_netif_keep_dst(struct tcf_block *block)
1279{
1280 struct tcf_block_owner_item *item;
1281
1282 block->keep_dst = true;
1283 list_for_each_entry(item, &block->owner_list, list)
1284 tcf_block_owner_netif_keep_dst(block, item->q,
1285 item->binder_type);
1286}
1287EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1288
1289static int tcf_block_owner_add(struct tcf_block *block,
1290 struct Qdisc *q,
1291 enum flow_block_binder_type binder_type)
1292{
1293 struct tcf_block_owner_item *item;
1294
1295 item = kmalloc(sizeof(*item), GFP_KERNEL);
1296 if (!item)
1297 return -ENOMEM;
1298 item->q = q;
1299 item->binder_type = binder_type;
1300 list_add(&item->list, &block->owner_list);
1301 return 0;
1302}
1303
1304static void tcf_block_owner_del(struct tcf_block *block,
1305 struct Qdisc *q,
1306 enum flow_block_binder_type binder_type)
1307{
1308 struct tcf_block_owner_item *item;
1309
1310 list_for_each_entry(item, &block->owner_list, list) {
1311 if (item->q == q && item->binder_type == binder_type) {
1312 list_del(&item->list);
1313 kfree(item);
1314 return;
1315 }
1316 }
1317 WARN_ON(1);
1318}
1319
1320int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1321 struct tcf_block_ext_info *ei,
1322 struct netlink_ext_ack *extack)
1323{
1324 struct net *net = qdisc_net(q);
1325 struct tcf_block *block = NULL;
1326 int err;
1327
1328 if (ei->block_index)
1329 /* block_index not 0 means the shared block is requested */
1330 block = tcf_block_refcnt_get(net, ei->block_index);
1331
1332 if (!block) {
1333 block = tcf_block_create(net, q, ei->block_index, extack);
1334 if (IS_ERR(block))
1335 return PTR_ERR(block);
1336 if (tcf_block_shared(block)) {
1337 err = tcf_block_insert(block, net, extack);
1338 if (err)
1339 goto err_block_insert;
1340 }
1341 }
1342
1343 err = tcf_block_owner_add(block, q, ei->binder_type);
1344 if (err)
1345 goto err_block_owner_add;
1346
1347 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1348
1349 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1350 if (err)
1351 goto err_chain0_head_change_cb_add;
1352
1353 err = tcf_block_offload_bind(block, q, ei, extack);
1354 if (err)
1355 goto err_block_offload_bind;
1356
1357 *p_block = block;
1358 return 0;
1359
1360err_block_offload_bind:
1361 tcf_chain0_head_change_cb_del(block, ei);
1362err_chain0_head_change_cb_add:
1363 tcf_block_owner_del(block, q, ei->binder_type);
1364err_block_owner_add:
1365err_block_insert:
1366 tcf_block_refcnt_put(block, true);
1367 return err;
1368}
1369EXPORT_SYMBOL(tcf_block_get_ext);
1370
1371static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1372{
1373 struct tcf_proto __rcu **p_filter_chain = priv;
1374
1375 rcu_assign_pointer(*p_filter_chain, tp_head);
1376}
1377
1378int tcf_block_get(struct tcf_block **p_block,
1379 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1380 struct netlink_ext_ack *extack)
1381{
1382 struct tcf_block_ext_info ei = {
1383 .chain_head_change = tcf_chain_head_change_dflt,
1384 .chain_head_change_priv = p_filter_chain,
1385 };
1386
1387 WARN_ON(!p_filter_chain);
1388 return tcf_block_get_ext(p_block, q, &ei, extack);
1389}
1390EXPORT_SYMBOL(tcf_block_get);
1391
1392/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1393 * actions should be all removed after flushing.
1394 */
1395void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1396 struct tcf_block_ext_info *ei)
1397{
1398 if (!block)
1399 return;
1400 tcf_chain0_head_change_cb_del(block, ei);
1401 tcf_block_owner_del(block, q, ei->binder_type);
1402
1403 __tcf_block_put(block, q, ei, true);
1404}
1405EXPORT_SYMBOL(tcf_block_put_ext);
1406
1407void tcf_block_put(struct tcf_block *block)
1408{
1409 struct tcf_block_ext_info ei = {0, };
1410
1411 if (!block)
1412 return;
1413 tcf_block_put_ext(block, block->q, &ei);
1414}
1415
1416EXPORT_SYMBOL(tcf_block_put);
1417
1418static int
1419tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1420 void *cb_priv, bool add, bool offload_in_use,
1421 struct netlink_ext_ack *extack)
1422{
1423 struct tcf_chain *chain, *chain_prev;
1424 struct tcf_proto *tp, *tp_prev;
1425 int err;
1426
1427 lockdep_assert_held(&block->cb_lock);
1428
1429 for (chain = __tcf_get_next_chain(block, NULL);
1430 chain;
1431 chain_prev = chain,
1432 chain = __tcf_get_next_chain(block, chain),
1433 tcf_chain_put(chain_prev)) {
1434 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1435 tp_prev = tp,
1436 tp = __tcf_get_next_proto(chain, tp),
1437 tcf_proto_put(tp_prev, true, NULL)) {
1438 if (tp->ops->reoffload) {
1439 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1440 extack);
1441 if (err && add)
1442 goto err_playback_remove;
1443 } else if (add && offload_in_use) {
1444 err = -EOPNOTSUPP;
1445 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1446 goto err_playback_remove;
1447 }
1448 }
1449 }
1450
1451 return 0;
1452
1453err_playback_remove:
1454 tcf_proto_put(tp, true, NULL);
1455 tcf_chain_put(chain);
1456 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1457 extack);
1458 return err;
1459}
1460
1461static int tcf_block_bind(struct tcf_block *block,
1462 struct flow_block_offload *bo)
1463{
1464 struct flow_block_cb *block_cb, *next;
1465 int err, i = 0;
1466
1467 lockdep_assert_held(&block->cb_lock);
1468
1469 list_for_each_entry(block_cb, &bo->cb_list, list) {
1470 err = tcf_block_playback_offloads(block, block_cb->cb,
1471 block_cb->cb_priv, true,
1472 tcf_block_offload_in_use(block),
1473 bo->extack);
1474 if (err)
1475 goto err_unroll;
1476 if (!bo->unlocked_driver_cb)
1477 block->lockeddevcnt++;
1478
1479 i++;
1480 }
1481 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1482
1483 return 0;
1484
1485err_unroll:
1486 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1487 if (i-- > 0) {
1488 list_del(&block_cb->list);
1489 tcf_block_playback_offloads(block, block_cb->cb,
1490 block_cb->cb_priv, false,
1491 tcf_block_offload_in_use(block),
1492 NULL);
1493 if (!bo->unlocked_driver_cb)
1494 block->lockeddevcnt--;
1495 }
1496 flow_block_cb_free(block_cb);
1497 }
1498
1499 return err;
1500}
1501
1502static void tcf_block_unbind(struct tcf_block *block,
1503 struct flow_block_offload *bo)
1504{
1505 struct flow_block_cb *block_cb, *next;
1506
1507 lockdep_assert_held(&block->cb_lock);
1508
1509 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1510 tcf_block_playback_offloads(block, block_cb->cb,
1511 block_cb->cb_priv, false,
1512 tcf_block_offload_in_use(block),
1513 NULL);
1514 list_del(&block_cb->list);
1515 flow_block_cb_free(block_cb);
1516 if (!bo->unlocked_driver_cb)
1517 block->lockeddevcnt--;
1518 }
1519}
1520
1521static int tcf_block_setup(struct tcf_block *block,
1522 struct flow_block_offload *bo)
1523{
1524 int err;
1525
1526 switch (bo->command) {
1527 case FLOW_BLOCK_BIND:
1528 err = tcf_block_bind(block, bo);
1529 break;
1530 case FLOW_BLOCK_UNBIND:
1531 err = 0;
1532 tcf_block_unbind(block, bo);
1533 break;
1534 default:
1535 WARN_ON_ONCE(1);
1536 err = -EOPNOTSUPP;
1537 }
1538
1539 return err;
1540}
1541
1542/* Main classifier routine: scans classifier chain attached
1543 * to this qdisc, (optionally) tests for protocol and asks
1544 * specific classifiers.
1545 */
1546static inline int __tcf_classify(struct sk_buff *skb,
1547 const struct tcf_proto *tp,
1548 const struct tcf_proto *orig_tp,
1549 struct tcf_result *res,
1550 bool compat_mode,
1551 u32 *last_executed_chain)
1552{
1553#ifdef CONFIG_NET_CLS_ACT
1554 const int max_reclassify_loop = 16;
1555 const struct tcf_proto *first_tp;
1556 int limit = 0;
1557
1558reclassify:
1559#endif
1560 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1561 __be16 protocol = skb_protocol(skb, false);
1562 int err;
1563
1564 if (tp->protocol != protocol &&
1565 tp->protocol != htons(ETH_P_ALL))
1566 continue;
1567
1568 err = tc_classify(skb, tp, res);
1569#ifdef CONFIG_NET_CLS_ACT
1570 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1571 first_tp = orig_tp;
1572 *last_executed_chain = first_tp->chain->index;
1573 goto reset;
1574 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1575 first_tp = res->goto_tp;
1576 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1577 goto reset;
1578 }
1579#endif
1580 if (err >= 0)
1581 return err;
1582 }
1583
1584 return TC_ACT_UNSPEC; /* signal: continue lookup */
1585#ifdef CONFIG_NET_CLS_ACT
1586reset:
1587 if (unlikely(limit++ >= max_reclassify_loop)) {
1588 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1589 tp->chain->block->index,
1590 tp->prio & 0xffff,
1591 ntohs(tp->protocol));
1592 return TC_ACT_SHOT;
1593 }
1594
1595 tp = first_tp;
1596 goto reclassify;
1597#endif
1598}
1599
1600int tcf_classify(struct sk_buff *skb,
1601 const struct tcf_block *block,
1602 const struct tcf_proto *tp,
1603 struct tcf_result *res, bool compat_mode)
1604{
1605#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1606 u32 last_executed_chain = 0;
1607
1608 return __tcf_classify(skb, tp, tp, res, compat_mode,
1609 &last_executed_chain);
1610#else
1611 u32 last_executed_chain = tp ? tp->chain->index : 0;
1612 const struct tcf_proto *orig_tp = tp;
1613 struct tc_skb_ext *ext;
1614 int ret;
1615
1616 if (block) {
1617 ext = skb_ext_find(skb, TC_SKB_EXT);
1618
1619 if (ext && ext->chain) {
1620 struct tcf_chain *fchain;
1621
1622 fchain = tcf_chain_lookup_rcu(block, ext->chain);
1623 if (!fchain)
1624 return TC_ACT_SHOT;
1625
1626 /* Consume, so cloned/redirect skbs won't inherit ext */
1627 skb_ext_del(skb, TC_SKB_EXT);
1628
1629 tp = rcu_dereference_bh(fchain->filter_chain);
1630 last_executed_chain = fchain->index;
1631 }
1632 }
1633
1634 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1635 &last_executed_chain);
1636
1637 if (tc_skb_ext_tc_enabled()) {
1638 /* If we missed on some chain */
1639 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1640 struct tc_skb_cb *cb = tc_skb_cb(skb);
1641
1642 ext = tc_skb_ext_alloc(skb);
1643 if (WARN_ON_ONCE(!ext))
1644 return TC_ACT_SHOT;
1645 ext->chain = last_executed_chain;
1646 ext->mru = cb->mru;
1647 ext->post_ct = cb->post_ct;
1648 ext->post_ct_snat = cb->post_ct_snat;
1649 ext->post_ct_dnat = cb->post_ct_dnat;
1650 ext->zone = cb->zone;
1651 }
1652 }
1653
1654 return ret;
1655#endif
1656}
1657EXPORT_SYMBOL(tcf_classify);
1658
1659struct tcf_chain_info {
1660 struct tcf_proto __rcu **pprev;
1661 struct tcf_proto __rcu *next;
1662};
1663
1664static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1665 struct tcf_chain_info *chain_info)
1666{
1667 return tcf_chain_dereference(*chain_info->pprev, chain);
1668}
1669
1670static int tcf_chain_tp_insert(struct tcf_chain *chain,
1671 struct tcf_chain_info *chain_info,
1672 struct tcf_proto *tp)
1673{
1674 if (chain->flushing)
1675 return -EAGAIN;
1676
1677 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1678 if (*chain_info->pprev == chain->filter_chain)
1679 tcf_chain0_head_change(chain, tp);
1680 tcf_proto_get(tp);
1681 rcu_assign_pointer(*chain_info->pprev, tp);
1682
1683 return 0;
1684}
1685
1686static void tcf_chain_tp_remove(struct tcf_chain *chain,
1687 struct tcf_chain_info *chain_info,
1688 struct tcf_proto *tp)
1689{
1690 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1691
1692 tcf_proto_mark_delete(tp);
1693 if (tp == chain->filter_chain)
1694 tcf_chain0_head_change(chain, next);
1695 RCU_INIT_POINTER(*chain_info->pprev, next);
1696}
1697
1698static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1699 struct tcf_chain_info *chain_info,
1700 u32 protocol, u32 prio,
1701 bool prio_allocate);
1702
1703/* Try to insert new proto.
1704 * If proto with specified priority already exists, free new proto
1705 * and return existing one.
1706 */
1707
1708static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1709 struct tcf_proto *tp_new,
1710 u32 protocol, u32 prio,
1711 bool rtnl_held)
1712{
1713 struct tcf_chain_info chain_info;
1714 struct tcf_proto *tp;
1715 int err = 0;
1716
1717 mutex_lock(&chain->filter_chain_lock);
1718
1719 if (tcf_proto_exists_destroying(chain, tp_new)) {
1720 mutex_unlock(&chain->filter_chain_lock);
1721 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1722 return ERR_PTR(-EAGAIN);
1723 }
1724
1725 tp = tcf_chain_tp_find(chain, &chain_info,
1726 protocol, prio, false);
1727 if (!tp)
1728 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1729 mutex_unlock(&chain->filter_chain_lock);
1730
1731 if (tp) {
1732 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1733 tp_new = tp;
1734 } else if (err) {
1735 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1736 tp_new = ERR_PTR(err);
1737 }
1738
1739 return tp_new;
1740}
1741
1742static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1743 struct tcf_proto *tp, bool rtnl_held,
1744 struct netlink_ext_ack *extack)
1745{
1746 struct tcf_chain_info chain_info;
1747 struct tcf_proto *tp_iter;
1748 struct tcf_proto **pprev;
1749 struct tcf_proto *next;
1750
1751 mutex_lock(&chain->filter_chain_lock);
1752
1753 /* Atomically find and remove tp from chain. */
1754 for (pprev = &chain->filter_chain;
1755 (tp_iter = tcf_chain_dereference(*pprev, chain));
1756 pprev = &tp_iter->next) {
1757 if (tp_iter == tp) {
1758 chain_info.pprev = pprev;
1759 chain_info.next = tp_iter->next;
1760 WARN_ON(tp_iter->deleting);
1761 break;
1762 }
1763 }
1764 /* Verify that tp still exists and no new filters were inserted
1765 * concurrently.
1766 * Mark tp for deletion if it is empty.
1767 */
1768 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1769 mutex_unlock(&chain->filter_chain_lock);
1770 return;
1771 }
1772
1773 tcf_proto_signal_destroying(chain, tp);
1774 next = tcf_chain_dereference(chain_info.next, chain);
1775 if (tp == chain->filter_chain)
1776 tcf_chain0_head_change(chain, next);
1777 RCU_INIT_POINTER(*chain_info.pprev, next);
1778 mutex_unlock(&chain->filter_chain_lock);
1779
1780 tcf_proto_put(tp, rtnl_held, extack);
1781}
1782
1783static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1784 struct tcf_chain_info *chain_info,
1785 u32 protocol, u32 prio,
1786 bool prio_allocate)
1787{
1788 struct tcf_proto **pprev;
1789 struct tcf_proto *tp;
1790
1791 /* Check the chain for existence of proto-tcf with this priority */
1792 for (pprev = &chain->filter_chain;
1793 (tp = tcf_chain_dereference(*pprev, chain));
1794 pprev = &tp->next) {
1795 if (tp->prio >= prio) {
1796 if (tp->prio == prio) {
1797 if (prio_allocate ||
1798 (tp->protocol != protocol && protocol))
1799 return ERR_PTR(-EINVAL);
1800 } else {
1801 tp = NULL;
1802 }
1803 break;
1804 }
1805 }
1806 chain_info->pprev = pprev;
1807 if (tp) {
1808 chain_info->next = tp->next;
1809 tcf_proto_get(tp);
1810 } else {
1811 chain_info->next = NULL;
1812 }
1813 return tp;
1814}
1815
1816static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1817 struct tcf_proto *tp, struct tcf_block *block,
1818 struct Qdisc *q, u32 parent, void *fh,
1819 u32 portid, u32 seq, u16 flags, int event,
1820 bool terse_dump, bool rtnl_held)
1821{
1822 struct tcmsg *tcm;
1823 struct nlmsghdr *nlh;
1824 unsigned char *b = skb_tail_pointer(skb);
1825
1826 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1827 if (!nlh)
1828 goto out_nlmsg_trim;
1829 tcm = nlmsg_data(nlh);
1830 tcm->tcm_family = AF_UNSPEC;
1831 tcm->tcm__pad1 = 0;
1832 tcm->tcm__pad2 = 0;
1833 if (q) {
1834 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1835 tcm->tcm_parent = parent;
1836 } else {
1837 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1838 tcm->tcm_block_index = block->index;
1839 }
1840 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1841 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1842 goto nla_put_failure;
1843 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1844 goto nla_put_failure;
1845 if (!fh) {
1846 tcm->tcm_handle = 0;
1847 } else if (terse_dump) {
1848 if (tp->ops->terse_dump) {
1849 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1850 rtnl_held) < 0)
1851 goto nla_put_failure;
1852 } else {
1853 goto cls_op_not_supp;
1854 }
1855 } else {
1856 if (tp->ops->dump &&
1857 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1858 goto nla_put_failure;
1859 }
1860 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1861 return skb->len;
1862
1863out_nlmsg_trim:
1864nla_put_failure:
1865cls_op_not_supp:
1866 nlmsg_trim(skb, b);
1867 return -1;
1868}
1869
1870static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1871 struct nlmsghdr *n, struct tcf_proto *tp,
1872 struct tcf_block *block, struct Qdisc *q,
1873 u32 parent, void *fh, int event, bool unicast,
1874 bool rtnl_held)
1875{
1876 struct sk_buff *skb;
1877 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1878 int err = 0;
1879
1880 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1881 if (!skb)
1882 return -ENOBUFS;
1883
1884 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1885 n->nlmsg_seq, n->nlmsg_flags, event,
1886 false, rtnl_held) <= 0) {
1887 kfree_skb(skb);
1888 return -EINVAL;
1889 }
1890
1891 if (unicast)
1892 err = rtnl_unicast(skb, net, portid);
1893 else
1894 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1895 n->nlmsg_flags & NLM_F_ECHO);
1896 return err;
1897}
1898
1899static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1900 struct nlmsghdr *n, struct tcf_proto *tp,
1901 struct tcf_block *block, struct Qdisc *q,
1902 u32 parent, void *fh, bool unicast, bool *last,
1903 bool rtnl_held, struct netlink_ext_ack *extack)
1904{
1905 struct sk_buff *skb;
1906 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1907 int err;
1908
1909 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1910 if (!skb)
1911 return -ENOBUFS;
1912
1913 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1914 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1915 false, rtnl_held) <= 0) {
1916 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1917 kfree_skb(skb);
1918 return -EINVAL;
1919 }
1920
1921 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1922 if (err) {
1923 kfree_skb(skb);
1924 return err;
1925 }
1926
1927 if (unicast)
1928 err = rtnl_unicast(skb, net, portid);
1929 else
1930 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1931 n->nlmsg_flags & NLM_F_ECHO);
1932 if (err < 0)
1933 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1934
1935 return err;
1936}
1937
1938static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1939 struct tcf_block *block, struct Qdisc *q,
1940 u32 parent, struct nlmsghdr *n,
1941 struct tcf_chain *chain, int event)
1942{
1943 struct tcf_proto *tp;
1944
1945 for (tp = tcf_get_next_proto(chain, NULL);
1946 tp; tp = tcf_get_next_proto(chain, tp))
1947 tfilter_notify(net, oskb, n, tp, block,
1948 q, parent, NULL, event, false, true);
1949}
1950
1951static void tfilter_put(struct tcf_proto *tp, void *fh)
1952{
1953 if (tp->ops->put && fh)
1954 tp->ops->put(tp, fh);
1955}
1956
1957static bool is_qdisc_ingress(__u32 classid)
1958{
1959 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
1960}
1961
1962static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1963 struct netlink_ext_ack *extack)
1964{
1965 struct net *net = sock_net(skb->sk);
1966 struct nlattr *tca[TCA_MAX + 1];
1967 char name[IFNAMSIZ];
1968 struct tcmsg *t;
1969 u32 protocol;
1970 u32 prio;
1971 bool prio_allocate;
1972 u32 parent;
1973 u32 chain_index;
1974 struct Qdisc *q;
1975 struct tcf_chain_info chain_info;
1976 struct tcf_chain *chain;
1977 struct tcf_block *block;
1978 struct tcf_proto *tp;
1979 unsigned long cl;
1980 void *fh;
1981 int err;
1982 int tp_created;
1983 bool rtnl_held = false;
1984 u32 flags;
1985
1986replay:
1987 tp_created = 0;
1988
1989 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1990 rtm_tca_policy, extack);
1991 if (err < 0)
1992 return err;
1993
1994 t = nlmsg_data(n);
1995 protocol = TC_H_MIN(t->tcm_info);
1996 prio = TC_H_MAJ(t->tcm_info);
1997 prio_allocate = false;
1998 parent = t->tcm_parent;
1999 tp = NULL;
2000 cl = 0;
2001 block = NULL;
2002 q = NULL;
2003 chain = NULL;
2004 flags = 0;
2005
2006 if (prio == 0) {
2007 /* If no priority is provided by the user,
2008 * we allocate one.
2009 */
2010 if (n->nlmsg_flags & NLM_F_CREATE) {
2011 prio = TC_H_MAKE(0x80000000U, 0U);
2012 prio_allocate = true;
2013 } else {
2014 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2015 return -ENOENT;
2016 }
2017 }
2018
2019 /* Find head of filter chain. */
2020
2021 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2022 if (err)
2023 return err;
2024
2025 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2026 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2027 err = -EINVAL;
2028 goto errout;
2029 }
2030
2031 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2032 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2033 * type is not specified, classifier is not unlocked.
2034 */
2035 if (rtnl_held ||
2036 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2037 !tcf_proto_is_unlocked(name)) {
2038 rtnl_held = true;
2039 rtnl_lock();
2040 }
2041
2042 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2043 if (err)
2044 goto errout;
2045
2046 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2047 extack);
2048 if (IS_ERR(block)) {
2049 err = PTR_ERR(block);
2050 goto errout;
2051 }
2052 block->classid = parent;
2053
2054 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2055 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2056 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2057 err = -EINVAL;
2058 goto errout;
2059 }
2060 chain = tcf_chain_get(block, chain_index, true);
2061 if (!chain) {
2062 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2063 err = -ENOMEM;
2064 goto errout;
2065 }
2066
2067 mutex_lock(&chain->filter_chain_lock);
2068 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2069 prio, prio_allocate);
2070 if (IS_ERR(tp)) {
2071 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2072 err = PTR_ERR(tp);
2073 goto errout_locked;
2074 }
2075
2076 if (tp == NULL) {
2077 struct tcf_proto *tp_new = NULL;
2078
2079 if (chain->flushing) {
2080 err = -EAGAIN;
2081 goto errout_locked;
2082 }
2083
2084 /* Proto-tcf does not exist, create new one */
2085
2086 if (tca[TCA_KIND] == NULL || !protocol) {
2087 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2088 err = -EINVAL;
2089 goto errout_locked;
2090 }
2091
2092 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2093 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2094 err = -ENOENT;
2095 goto errout_locked;
2096 }
2097
2098 if (prio_allocate)
2099 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2100 &chain_info));
2101
2102 mutex_unlock(&chain->filter_chain_lock);
2103 tp_new = tcf_proto_create(name, protocol, prio, chain,
2104 rtnl_held, extack);
2105 if (IS_ERR(tp_new)) {
2106 err = PTR_ERR(tp_new);
2107 goto errout_tp;
2108 }
2109
2110 tp_created = 1;
2111 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2112 rtnl_held);
2113 if (IS_ERR(tp)) {
2114 err = PTR_ERR(tp);
2115 goto errout_tp;
2116 }
2117 } else {
2118 mutex_unlock(&chain->filter_chain_lock);
2119 }
2120
2121 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2122 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2123 err = -EINVAL;
2124 goto errout;
2125 }
2126
2127 fh = tp->ops->get(tp, t->tcm_handle);
2128
2129 if (!fh) {
2130 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2131 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2132 err = -ENOENT;
2133 goto errout;
2134 }
2135 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2136 tfilter_put(tp, fh);
2137 NL_SET_ERR_MSG(extack, "Filter already exists");
2138 err = -EEXIST;
2139 goto errout;
2140 }
2141
2142 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2143 tfilter_put(tp, fh);
2144 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2145 err = -EINVAL;
2146 goto errout;
2147 }
2148
2149 if (!(n->nlmsg_flags & NLM_F_CREATE))
2150 flags |= TCA_ACT_FLAGS_REPLACE;
2151 if (!rtnl_held)
2152 flags |= TCA_ACT_FLAGS_NO_RTNL;
2153 if (is_qdisc_ingress(parent))
2154 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2155 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2156 flags, extack);
2157 if (err == 0) {
2158 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2159 RTM_NEWTFILTER, false, rtnl_held);
2160 tfilter_put(tp, fh);
2161 /* q pointer is NULL for shared blocks */
2162 if (q)
2163 q->flags &= ~TCQ_F_CAN_BYPASS;
2164 }
2165
2166errout:
2167 if (err && tp_created)
2168 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2169errout_tp:
2170 if (chain) {
2171 if (tp && !IS_ERR(tp))
2172 tcf_proto_put(tp, rtnl_held, NULL);
2173 if (!tp_created)
2174 tcf_chain_put(chain);
2175 }
2176 tcf_block_release(q, block, rtnl_held);
2177
2178 if (rtnl_held)
2179 rtnl_unlock();
2180
2181 if (err == -EAGAIN) {
2182 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2183 * of target chain.
2184 */
2185 rtnl_held = true;
2186 /* Replay the request. */
2187 goto replay;
2188 }
2189 return err;
2190
2191errout_locked:
2192 mutex_unlock(&chain->filter_chain_lock);
2193 goto errout;
2194}
2195
2196static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2197 struct netlink_ext_ack *extack)
2198{
2199 struct net *net = sock_net(skb->sk);
2200 struct nlattr *tca[TCA_MAX + 1];
2201 char name[IFNAMSIZ];
2202 struct tcmsg *t;
2203 u32 protocol;
2204 u32 prio;
2205 u32 parent;
2206 u32 chain_index;
2207 struct Qdisc *q = NULL;
2208 struct tcf_chain_info chain_info;
2209 struct tcf_chain *chain = NULL;
2210 struct tcf_block *block = NULL;
2211 struct tcf_proto *tp = NULL;
2212 unsigned long cl = 0;
2213 void *fh = NULL;
2214 int err;
2215 bool rtnl_held = false;
2216
2217 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2218 rtm_tca_policy, extack);
2219 if (err < 0)
2220 return err;
2221
2222 t = nlmsg_data(n);
2223 protocol = TC_H_MIN(t->tcm_info);
2224 prio = TC_H_MAJ(t->tcm_info);
2225 parent = t->tcm_parent;
2226
2227 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2228 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2229 return -ENOENT;
2230 }
2231
2232 /* Find head of filter chain. */
2233
2234 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2235 if (err)
2236 return err;
2237
2238 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2239 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2240 err = -EINVAL;
2241 goto errout;
2242 }
2243 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2244 * found), qdisc is not unlocked, classifier type is not specified,
2245 * classifier is not unlocked.
2246 */
2247 if (!prio ||
2248 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2249 !tcf_proto_is_unlocked(name)) {
2250 rtnl_held = true;
2251 rtnl_lock();
2252 }
2253
2254 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2255 if (err)
2256 goto errout;
2257
2258 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2259 extack);
2260 if (IS_ERR(block)) {
2261 err = PTR_ERR(block);
2262 goto errout;
2263 }
2264
2265 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2266 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2267 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2268 err = -EINVAL;
2269 goto errout;
2270 }
2271 chain = tcf_chain_get(block, chain_index, false);
2272 if (!chain) {
2273 /* User requested flush on non-existent chain. Nothing to do,
2274 * so just return success.
2275 */
2276 if (prio == 0) {
2277 err = 0;
2278 goto errout;
2279 }
2280 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2281 err = -ENOENT;
2282 goto errout;
2283 }
2284
2285 if (prio == 0) {
2286 tfilter_notify_chain(net, skb, block, q, parent, n,
2287 chain, RTM_DELTFILTER);
2288 tcf_chain_flush(chain, rtnl_held);
2289 err = 0;
2290 goto errout;
2291 }
2292
2293 mutex_lock(&chain->filter_chain_lock);
2294 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2295 prio, false);
2296 if (!tp || IS_ERR(tp)) {
2297 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2298 err = tp ? PTR_ERR(tp) : -ENOENT;
2299 goto errout_locked;
2300 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2301 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2302 err = -EINVAL;
2303 goto errout_locked;
2304 } else if (t->tcm_handle == 0) {
2305 tcf_proto_signal_destroying(chain, tp);
2306 tcf_chain_tp_remove(chain, &chain_info, tp);
2307 mutex_unlock(&chain->filter_chain_lock);
2308
2309 tcf_proto_put(tp, rtnl_held, NULL);
2310 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2311 RTM_DELTFILTER, false, rtnl_held);
2312 err = 0;
2313 goto errout;
2314 }
2315 mutex_unlock(&chain->filter_chain_lock);
2316
2317 fh = tp->ops->get(tp, t->tcm_handle);
2318
2319 if (!fh) {
2320 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2321 err = -ENOENT;
2322 } else {
2323 bool last;
2324
2325 err = tfilter_del_notify(net, skb, n, tp, block,
2326 q, parent, fh, false, &last,
2327 rtnl_held, extack);
2328
2329 if (err)
2330 goto errout;
2331 if (last)
2332 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2333 }
2334
2335errout:
2336 if (chain) {
2337 if (tp && !IS_ERR(tp))
2338 tcf_proto_put(tp, rtnl_held, NULL);
2339 tcf_chain_put(chain);
2340 }
2341 tcf_block_release(q, block, rtnl_held);
2342
2343 if (rtnl_held)
2344 rtnl_unlock();
2345
2346 return err;
2347
2348errout_locked:
2349 mutex_unlock(&chain->filter_chain_lock);
2350 goto errout;
2351}
2352
2353static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2354 struct netlink_ext_ack *extack)
2355{
2356 struct net *net = sock_net(skb->sk);
2357 struct nlattr *tca[TCA_MAX + 1];
2358 char name[IFNAMSIZ];
2359 struct tcmsg *t;
2360 u32 protocol;
2361 u32 prio;
2362 u32 parent;
2363 u32 chain_index;
2364 struct Qdisc *q = NULL;
2365 struct tcf_chain_info chain_info;
2366 struct tcf_chain *chain = NULL;
2367 struct tcf_block *block = NULL;
2368 struct tcf_proto *tp = NULL;
2369 unsigned long cl = 0;
2370 void *fh = NULL;
2371 int err;
2372 bool rtnl_held = false;
2373
2374 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2375 rtm_tca_policy, extack);
2376 if (err < 0)
2377 return err;
2378
2379 t = nlmsg_data(n);
2380 protocol = TC_H_MIN(t->tcm_info);
2381 prio = TC_H_MAJ(t->tcm_info);
2382 parent = t->tcm_parent;
2383
2384 if (prio == 0) {
2385 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2386 return -ENOENT;
2387 }
2388
2389 /* Find head of filter chain. */
2390
2391 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2392 if (err)
2393 return err;
2394
2395 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2396 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2397 err = -EINVAL;
2398 goto errout;
2399 }
2400 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2401 * unlocked, classifier type is not specified, classifier is not
2402 * unlocked.
2403 */
2404 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2405 !tcf_proto_is_unlocked(name)) {
2406 rtnl_held = true;
2407 rtnl_lock();
2408 }
2409
2410 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2411 if (err)
2412 goto errout;
2413
2414 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2415 extack);
2416 if (IS_ERR(block)) {
2417 err = PTR_ERR(block);
2418 goto errout;
2419 }
2420
2421 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2422 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2423 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2424 err = -EINVAL;
2425 goto errout;
2426 }
2427 chain = tcf_chain_get(block, chain_index, false);
2428 if (!chain) {
2429 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2430 err = -EINVAL;
2431 goto errout;
2432 }
2433
2434 mutex_lock(&chain->filter_chain_lock);
2435 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2436 prio, false);
2437 mutex_unlock(&chain->filter_chain_lock);
2438 if (!tp || IS_ERR(tp)) {
2439 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2440 err = tp ? PTR_ERR(tp) : -ENOENT;
2441 goto errout;
2442 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2443 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2444 err = -EINVAL;
2445 goto errout;
2446 }
2447
2448 fh = tp->ops->get(tp, t->tcm_handle);
2449
2450 if (!fh) {
2451 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2452 err = -ENOENT;
2453 } else {
2454 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2455 fh, RTM_NEWTFILTER, true, rtnl_held);
2456 if (err < 0)
2457 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2458 }
2459
2460 tfilter_put(tp, fh);
2461errout:
2462 if (chain) {
2463 if (tp && !IS_ERR(tp))
2464 tcf_proto_put(tp, rtnl_held, NULL);
2465 tcf_chain_put(chain);
2466 }
2467 tcf_block_release(q, block, rtnl_held);
2468
2469 if (rtnl_held)
2470 rtnl_unlock();
2471
2472 return err;
2473}
2474
2475struct tcf_dump_args {
2476 struct tcf_walker w;
2477 struct sk_buff *skb;
2478 struct netlink_callback *cb;
2479 struct tcf_block *block;
2480 struct Qdisc *q;
2481 u32 parent;
2482 bool terse_dump;
2483};
2484
2485static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2486{
2487 struct tcf_dump_args *a = (void *)arg;
2488 struct net *net = sock_net(a->skb->sk);
2489
2490 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2491 n, NETLINK_CB(a->cb->skb).portid,
2492 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2493 RTM_NEWTFILTER, a->terse_dump, true);
2494}
2495
2496static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2497 struct sk_buff *skb, struct netlink_callback *cb,
2498 long index_start, long *p_index, bool terse)
2499{
2500 struct net *net = sock_net(skb->sk);
2501 struct tcf_block *block = chain->block;
2502 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2503 struct tcf_proto *tp, *tp_prev;
2504 struct tcf_dump_args arg;
2505
2506 for (tp = __tcf_get_next_proto(chain, NULL);
2507 tp;
2508 tp_prev = tp,
2509 tp = __tcf_get_next_proto(chain, tp),
2510 tcf_proto_put(tp_prev, true, NULL),
2511 (*p_index)++) {
2512 if (*p_index < index_start)
2513 continue;
2514 if (TC_H_MAJ(tcm->tcm_info) &&
2515 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2516 continue;
2517 if (TC_H_MIN(tcm->tcm_info) &&
2518 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2519 continue;
2520 if (*p_index > index_start)
2521 memset(&cb->args[1], 0,
2522 sizeof(cb->args) - sizeof(cb->args[0]));
2523 if (cb->args[1] == 0) {
2524 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2525 NETLINK_CB(cb->skb).portid,
2526 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2527 RTM_NEWTFILTER, false, true) <= 0)
2528 goto errout;
2529 cb->args[1] = 1;
2530 }
2531 if (!tp->ops->walk)
2532 continue;
2533 arg.w.fn = tcf_node_dump;
2534 arg.skb = skb;
2535 arg.cb = cb;
2536 arg.block = block;
2537 arg.q = q;
2538 arg.parent = parent;
2539 arg.w.stop = 0;
2540 arg.w.skip = cb->args[1] - 1;
2541 arg.w.count = 0;
2542 arg.w.cookie = cb->args[2];
2543 arg.terse_dump = terse;
2544 tp->ops->walk(tp, &arg.w, true);
2545 cb->args[2] = arg.w.cookie;
2546 cb->args[1] = arg.w.count + 1;
2547 if (arg.w.stop)
2548 goto errout;
2549 }
2550 return true;
2551
2552errout:
2553 tcf_proto_put(tp, true, NULL);
2554 return false;
2555}
2556
2557static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2558 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2559};
2560
2561/* called with RTNL */
2562static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2563{
2564 struct tcf_chain *chain, *chain_prev;
2565 struct net *net = sock_net(skb->sk);
2566 struct nlattr *tca[TCA_MAX + 1];
2567 struct Qdisc *q = NULL;
2568 struct tcf_block *block;
2569 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2570 bool terse_dump = false;
2571 long index_start;
2572 long index;
2573 u32 parent;
2574 int err;
2575
2576 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2577 return skb->len;
2578
2579 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2580 tcf_tfilter_dump_policy, cb->extack);
2581 if (err)
2582 return err;
2583
2584 if (tca[TCA_DUMP_FLAGS]) {
2585 struct nla_bitfield32 flags =
2586 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2587
2588 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2589 }
2590
2591 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2592 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2593 if (!block)
2594 goto out;
2595 /* If we work with block index, q is NULL and parent value
2596 * will never be used in the following code. The check
2597 * in tcf_fill_node prevents it. However, compiler does not
2598 * see that far, so set parent to zero to silence the warning
2599 * about parent being uninitialized.
2600 */
2601 parent = 0;
2602 } else {
2603 const struct Qdisc_class_ops *cops;
2604 struct net_device *dev;
2605 unsigned long cl = 0;
2606
2607 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2608 if (!dev)
2609 return skb->len;
2610
2611 parent = tcm->tcm_parent;
2612 if (!parent)
2613 q = rtnl_dereference(dev->qdisc);
2614 else
2615 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2616 if (!q)
2617 goto out;
2618 cops = q->ops->cl_ops;
2619 if (!cops)
2620 goto out;
2621 if (!cops->tcf_block)
2622 goto out;
2623 if (TC_H_MIN(tcm->tcm_parent)) {
2624 cl = cops->find(q, tcm->tcm_parent);
2625 if (cl == 0)
2626 goto out;
2627 }
2628 block = cops->tcf_block(q, cl, NULL);
2629 if (!block)
2630 goto out;
2631 parent = block->classid;
2632 if (tcf_block_shared(block))
2633 q = NULL;
2634 }
2635
2636 index_start = cb->args[0];
2637 index = 0;
2638
2639 for (chain = __tcf_get_next_chain(block, NULL);
2640 chain;
2641 chain_prev = chain,
2642 chain = __tcf_get_next_chain(block, chain),
2643 tcf_chain_put(chain_prev)) {
2644 if (tca[TCA_CHAIN] &&
2645 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2646 continue;
2647 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2648 index_start, &index, terse_dump)) {
2649 tcf_chain_put(chain);
2650 err = -EMSGSIZE;
2651 break;
2652 }
2653 }
2654
2655 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2656 tcf_block_refcnt_put(block, true);
2657 cb->args[0] = index;
2658
2659out:
2660 /* If we did no progress, the error (EMSGSIZE) is real */
2661 if (skb->len == 0 && err)
2662 return err;
2663 return skb->len;
2664}
2665
2666static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2667 void *tmplt_priv, u32 chain_index,
2668 struct net *net, struct sk_buff *skb,
2669 struct tcf_block *block,
2670 u32 portid, u32 seq, u16 flags, int event)
2671{
2672 unsigned char *b = skb_tail_pointer(skb);
2673 const struct tcf_proto_ops *ops;
2674 struct nlmsghdr *nlh;
2675 struct tcmsg *tcm;
2676 void *priv;
2677
2678 ops = tmplt_ops;
2679 priv = tmplt_priv;
2680
2681 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2682 if (!nlh)
2683 goto out_nlmsg_trim;
2684 tcm = nlmsg_data(nlh);
2685 tcm->tcm_family = AF_UNSPEC;
2686 tcm->tcm__pad1 = 0;
2687 tcm->tcm__pad2 = 0;
2688 tcm->tcm_handle = 0;
2689 if (block->q) {
2690 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2691 tcm->tcm_parent = block->q->handle;
2692 } else {
2693 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2694 tcm->tcm_block_index = block->index;
2695 }
2696
2697 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2698 goto nla_put_failure;
2699
2700 if (ops) {
2701 if (nla_put_string(skb, TCA_KIND, ops->kind))
2702 goto nla_put_failure;
2703 if (ops->tmplt_dump(skb, net, priv) < 0)
2704 goto nla_put_failure;
2705 }
2706
2707 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2708 return skb->len;
2709
2710out_nlmsg_trim:
2711nla_put_failure:
2712 nlmsg_trim(skb, b);
2713 return -EMSGSIZE;
2714}
2715
2716static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2717 u32 seq, u16 flags, int event, bool unicast)
2718{
2719 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2720 struct tcf_block *block = chain->block;
2721 struct net *net = block->net;
2722 struct sk_buff *skb;
2723 int err = 0;
2724
2725 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2726 if (!skb)
2727 return -ENOBUFS;
2728
2729 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2730 chain->index, net, skb, block, portid,
2731 seq, flags, event) <= 0) {
2732 kfree_skb(skb);
2733 return -EINVAL;
2734 }
2735
2736 if (unicast)
2737 err = rtnl_unicast(skb, net, portid);
2738 else
2739 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2740 flags & NLM_F_ECHO);
2741
2742 return err;
2743}
2744
2745static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2746 void *tmplt_priv, u32 chain_index,
2747 struct tcf_block *block, struct sk_buff *oskb,
2748 u32 seq, u16 flags, bool unicast)
2749{
2750 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2751 struct net *net = block->net;
2752 struct sk_buff *skb;
2753
2754 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2755 if (!skb)
2756 return -ENOBUFS;
2757
2758 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2759 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2760 kfree_skb(skb);
2761 return -EINVAL;
2762 }
2763
2764 if (unicast)
2765 return rtnl_unicast(skb, net, portid);
2766
2767 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2768}
2769
2770static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2771 struct nlattr **tca,
2772 struct netlink_ext_ack *extack)
2773{
2774 const struct tcf_proto_ops *ops;
2775 char name[IFNAMSIZ];
2776 void *tmplt_priv;
2777
2778 /* If kind is not set, user did not specify template. */
2779 if (!tca[TCA_KIND])
2780 return 0;
2781
2782 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2783 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2784 return -EINVAL;
2785 }
2786
2787 ops = tcf_proto_lookup_ops(name, true, extack);
2788 if (IS_ERR(ops))
2789 return PTR_ERR(ops);
2790 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2791 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2792 return -EOPNOTSUPP;
2793 }
2794
2795 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2796 if (IS_ERR(tmplt_priv)) {
2797 module_put(ops->owner);
2798 return PTR_ERR(tmplt_priv);
2799 }
2800 chain->tmplt_ops = ops;
2801 chain->tmplt_priv = tmplt_priv;
2802 return 0;
2803}
2804
2805static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2806 void *tmplt_priv)
2807{
2808 /* If template ops are set, no work to do for us. */
2809 if (!tmplt_ops)
2810 return;
2811
2812 tmplt_ops->tmplt_destroy(tmplt_priv);
2813 module_put(tmplt_ops->owner);
2814}
2815
2816/* Add/delete/get a chain */
2817
2818static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2819 struct netlink_ext_ack *extack)
2820{
2821 struct net *net = sock_net(skb->sk);
2822 struct nlattr *tca[TCA_MAX + 1];
2823 struct tcmsg *t;
2824 u32 parent;
2825 u32 chain_index;
2826 struct Qdisc *q;
2827 struct tcf_chain *chain;
2828 struct tcf_block *block;
2829 unsigned long cl;
2830 int err;
2831
2832replay:
2833 q = NULL;
2834 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2835 rtm_tca_policy, extack);
2836 if (err < 0)
2837 return err;
2838
2839 t = nlmsg_data(n);
2840 parent = t->tcm_parent;
2841 cl = 0;
2842
2843 block = tcf_block_find(net, &q, &parent, &cl,
2844 t->tcm_ifindex, t->tcm_block_index, extack);
2845 if (IS_ERR(block))
2846 return PTR_ERR(block);
2847
2848 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2849 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2850 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2851 err = -EINVAL;
2852 goto errout_block;
2853 }
2854
2855 mutex_lock(&block->lock);
2856 chain = tcf_chain_lookup(block, chain_index);
2857 if (n->nlmsg_type == RTM_NEWCHAIN) {
2858 if (chain) {
2859 if (tcf_chain_held_by_acts_only(chain)) {
2860 /* The chain exists only because there is
2861 * some action referencing it.
2862 */
2863 tcf_chain_hold(chain);
2864 } else {
2865 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2866 err = -EEXIST;
2867 goto errout_block_locked;
2868 }
2869 } else {
2870 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2871 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2872 err = -ENOENT;
2873 goto errout_block_locked;
2874 }
2875 chain = tcf_chain_create(block, chain_index);
2876 if (!chain) {
2877 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2878 err = -ENOMEM;
2879 goto errout_block_locked;
2880 }
2881 }
2882 } else {
2883 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2884 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2885 err = -EINVAL;
2886 goto errout_block_locked;
2887 }
2888 tcf_chain_hold(chain);
2889 }
2890
2891 if (n->nlmsg_type == RTM_NEWCHAIN) {
2892 /* Modifying chain requires holding parent block lock. In case
2893 * the chain was successfully added, take a reference to the
2894 * chain. This ensures that an empty chain does not disappear at
2895 * the end of this function.
2896 */
2897 tcf_chain_hold(chain);
2898 chain->explicitly_created = true;
2899 }
2900 mutex_unlock(&block->lock);
2901
2902 switch (n->nlmsg_type) {
2903 case RTM_NEWCHAIN:
2904 err = tc_chain_tmplt_add(chain, net, tca, extack);
2905 if (err) {
2906 tcf_chain_put_explicitly_created(chain);
2907 goto errout;
2908 }
2909
2910 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2911 RTM_NEWCHAIN, false);
2912 break;
2913 case RTM_DELCHAIN:
2914 tfilter_notify_chain(net, skb, block, q, parent, n,
2915 chain, RTM_DELTFILTER);
2916 /* Flush the chain first as the user requested chain removal. */
2917 tcf_chain_flush(chain, true);
2918 /* In case the chain was successfully deleted, put a reference
2919 * to the chain previously taken during addition.
2920 */
2921 tcf_chain_put_explicitly_created(chain);
2922 break;
2923 case RTM_GETCHAIN:
2924 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2925 n->nlmsg_flags, n->nlmsg_type, true);
2926 if (err < 0)
2927 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2928 break;
2929 default:
2930 err = -EOPNOTSUPP;
2931 NL_SET_ERR_MSG(extack, "Unsupported message type");
2932 goto errout;
2933 }
2934
2935errout:
2936 tcf_chain_put(chain);
2937errout_block:
2938 tcf_block_release(q, block, true);
2939 if (err == -EAGAIN)
2940 /* Replay the request. */
2941 goto replay;
2942 return err;
2943
2944errout_block_locked:
2945 mutex_unlock(&block->lock);
2946 goto errout_block;
2947}
2948
2949/* called with RTNL */
2950static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2951{
2952 struct net *net = sock_net(skb->sk);
2953 struct nlattr *tca[TCA_MAX + 1];
2954 struct Qdisc *q = NULL;
2955 struct tcf_block *block;
2956 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2957 struct tcf_chain *chain;
2958 long index_start;
2959 long index;
2960 int err;
2961
2962 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2963 return skb->len;
2964
2965 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2966 rtm_tca_policy, cb->extack);
2967 if (err)
2968 return err;
2969
2970 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2971 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2972 if (!block)
2973 goto out;
2974 } else {
2975 const struct Qdisc_class_ops *cops;
2976 struct net_device *dev;
2977 unsigned long cl = 0;
2978
2979 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2980 if (!dev)
2981 return skb->len;
2982
2983 if (!tcm->tcm_parent)
2984 q = rtnl_dereference(dev->qdisc);
2985 else
2986 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2987
2988 if (!q)
2989 goto out;
2990 cops = q->ops->cl_ops;
2991 if (!cops)
2992 goto out;
2993 if (!cops->tcf_block)
2994 goto out;
2995 if (TC_H_MIN(tcm->tcm_parent)) {
2996 cl = cops->find(q, tcm->tcm_parent);
2997 if (cl == 0)
2998 goto out;
2999 }
3000 block = cops->tcf_block(q, cl, NULL);
3001 if (!block)
3002 goto out;
3003 if (tcf_block_shared(block))
3004 q = NULL;
3005 }
3006
3007 index_start = cb->args[0];
3008 index = 0;
3009
3010 mutex_lock(&block->lock);
3011 list_for_each_entry(chain, &block->chain_list, list) {
3012 if ((tca[TCA_CHAIN] &&
3013 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3014 continue;
3015 if (index < index_start) {
3016 index++;
3017 continue;
3018 }
3019 if (tcf_chain_held_by_acts_only(chain))
3020 continue;
3021 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3022 chain->index, net, skb, block,
3023 NETLINK_CB(cb->skb).portid,
3024 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3025 RTM_NEWCHAIN);
3026 if (err <= 0)
3027 break;
3028 index++;
3029 }
3030 mutex_unlock(&block->lock);
3031
3032 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3033 tcf_block_refcnt_put(block, true);
3034 cb->args[0] = index;
3035
3036out:
3037 /* If we did no progress, the error (EMSGSIZE) is real */
3038 if (skb->len == 0 && err)
3039 return err;
3040 return skb->len;
3041}
3042
3043void tcf_exts_destroy(struct tcf_exts *exts)
3044{
3045#ifdef CONFIG_NET_CLS_ACT
3046 if (exts->actions) {
3047 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3048 kfree(exts->actions);
3049 }
3050 exts->nr_actions = 0;
3051#endif
3052}
3053EXPORT_SYMBOL(tcf_exts_destroy);
3054
3055int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3056 struct nlattr *rate_tlv, struct tcf_exts *exts,
3057 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3058{
3059#ifdef CONFIG_NET_CLS_ACT
3060 {
3061 int init_res[TCA_ACT_MAX_PRIO] = {};
3062 struct tc_action *act;
3063 size_t attr_size = 0;
3064
3065 if (exts->police && tb[exts->police]) {
3066 struct tc_action_ops *a_o;
3067
3068 a_o = tc_action_load_ops(tb[exts->police], true,
3069 !(flags & TCA_ACT_FLAGS_NO_RTNL),
3070 extack);
3071 if (IS_ERR(a_o))
3072 return PTR_ERR(a_o);
3073 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3074 act = tcf_action_init_1(net, tp, tb[exts->police],
3075 rate_tlv, a_o, init_res, flags,
3076 extack);
3077 module_put(a_o->owner);
3078 if (IS_ERR(act))
3079 return PTR_ERR(act);
3080
3081 act->type = exts->type = TCA_OLD_COMPAT;
3082 exts->actions[0] = act;
3083 exts->nr_actions = 1;
3084 tcf_idr_insert_many(exts->actions);
3085 } else if (exts->action && tb[exts->action]) {
3086 int err;
3087
3088 flags |= TCA_ACT_FLAGS_BIND;
3089 err = tcf_action_init(net, tp, tb[exts->action],
3090 rate_tlv, exts->actions, init_res,
3091 &attr_size, flags, fl_flags,
3092 extack);
3093 if (err < 0)
3094 return err;
3095 exts->nr_actions = err;
3096 }
3097 }
3098#else
3099 if ((exts->action && tb[exts->action]) ||
3100 (exts->police && tb[exts->police])) {
3101 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3102 return -EOPNOTSUPP;
3103 }
3104#endif
3105
3106 return 0;
3107}
3108EXPORT_SYMBOL(tcf_exts_validate_ex);
3109
3110int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3111 struct nlattr *rate_tlv, struct tcf_exts *exts,
3112 u32 flags, struct netlink_ext_ack *extack)
3113{
3114 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3115 flags, 0, extack);
3116}
3117EXPORT_SYMBOL(tcf_exts_validate);
3118
3119void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3120{
3121#ifdef CONFIG_NET_CLS_ACT
3122 struct tcf_exts old = *dst;
3123
3124 *dst = *src;
3125 tcf_exts_destroy(&old);
3126#endif
3127}
3128EXPORT_SYMBOL(tcf_exts_change);
3129
3130#ifdef CONFIG_NET_CLS_ACT
3131static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3132{
3133 if (exts->nr_actions == 0)
3134 return NULL;
3135 else
3136 return exts->actions[0];
3137}
3138#endif
3139
3140int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3141{
3142#ifdef CONFIG_NET_CLS_ACT
3143 struct nlattr *nest;
3144
3145 if (exts->action && tcf_exts_has_actions(exts)) {
3146 /*
3147 * again for backward compatible mode - we want
3148 * to work with both old and new modes of entering
3149 * tc data even if iproute2 was newer - jhs
3150 */
3151 if (exts->type != TCA_OLD_COMPAT) {
3152 nest = nla_nest_start_noflag(skb, exts->action);
3153 if (nest == NULL)
3154 goto nla_put_failure;
3155
3156 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3157 < 0)
3158 goto nla_put_failure;
3159 nla_nest_end(skb, nest);
3160 } else if (exts->police) {
3161 struct tc_action *act = tcf_exts_first_act(exts);
3162 nest = nla_nest_start_noflag(skb, exts->police);
3163 if (nest == NULL || !act)
3164 goto nla_put_failure;
3165 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3166 goto nla_put_failure;
3167 nla_nest_end(skb, nest);
3168 }
3169 }
3170 return 0;
3171
3172nla_put_failure:
3173 nla_nest_cancel(skb, nest);
3174 return -1;
3175#else
3176 return 0;
3177#endif
3178}
3179EXPORT_SYMBOL(tcf_exts_dump);
3180
3181int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3182{
3183#ifdef CONFIG_NET_CLS_ACT
3184 struct nlattr *nest;
3185
3186 if (!exts->action || !tcf_exts_has_actions(exts))
3187 return 0;
3188
3189 nest = nla_nest_start_noflag(skb, exts->action);
3190 if (!nest)
3191 goto nla_put_failure;
3192
3193 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3194 goto nla_put_failure;
3195 nla_nest_end(skb, nest);
3196 return 0;
3197
3198nla_put_failure:
3199 nla_nest_cancel(skb, nest);
3200 return -1;
3201#else
3202 return 0;
3203#endif
3204}
3205EXPORT_SYMBOL(tcf_exts_terse_dump);
3206
3207int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3208{
3209#ifdef CONFIG_NET_CLS_ACT
3210 struct tc_action *a = tcf_exts_first_act(exts);
3211 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3212 return -1;
3213#endif
3214 return 0;
3215}
3216EXPORT_SYMBOL(tcf_exts_dump_stats);
3217
3218static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3219{
3220 if (*flags & TCA_CLS_FLAGS_IN_HW)
3221 return;
3222 *flags |= TCA_CLS_FLAGS_IN_HW;
3223 atomic_inc(&block->offloadcnt);
3224}
3225
3226static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3227{
3228 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3229 return;
3230 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3231 atomic_dec(&block->offloadcnt);
3232}
3233
3234static void tc_cls_offload_cnt_update(struct tcf_block *block,
3235 struct tcf_proto *tp, u32 *cnt,
3236 u32 *flags, u32 diff, bool add)
3237{
3238 lockdep_assert_held(&block->cb_lock);
3239
3240 spin_lock(&tp->lock);
3241 if (add) {
3242 if (!*cnt)
3243 tcf_block_offload_inc(block, flags);
3244 *cnt += diff;
3245 } else {
3246 *cnt -= diff;
3247 if (!*cnt)
3248 tcf_block_offload_dec(block, flags);
3249 }
3250 spin_unlock(&tp->lock);
3251}
3252
3253static void
3254tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3255 u32 *cnt, u32 *flags)
3256{
3257 lockdep_assert_held(&block->cb_lock);
3258
3259 spin_lock(&tp->lock);
3260 tcf_block_offload_dec(block, flags);
3261 *cnt = 0;
3262 spin_unlock(&tp->lock);
3263}
3264
3265static int
3266__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3267 void *type_data, bool err_stop)
3268{
3269 struct flow_block_cb *block_cb;
3270 int ok_count = 0;
3271 int err;
3272
3273 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3274 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3275 if (err) {
3276 if (err_stop)
3277 return err;
3278 } else {
3279 ok_count++;
3280 }
3281 }
3282 return ok_count;
3283}
3284
3285int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3286 void *type_data, bool err_stop, bool rtnl_held)
3287{
3288 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3289 int ok_count;
3290
3291retry:
3292 if (take_rtnl)
3293 rtnl_lock();
3294 down_read(&block->cb_lock);
3295 /* Need to obtain rtnl lock if block is bound to devs that require it.
3296 * In block bind code cb_lock is obtained while holding rtnl, so we must
3297 * obtain the locks in same order here.
3298 */
3299 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3300 up_read(&block->cb_lock);
3301 take_rtnl = true;
3302 goto retry;
3303 }
3304
3305 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3306
3307 up_read(&block->cb_lock);
3308 if (take_rtnl)
3309 rtnl_unlock();
3310 return ok_count;
3311}
3312EXPORT_SYMBOL(tc_setup_cb_call);
3313
3314/* Non-destructive filter add. If filter that wasn't already in hardware is
3315 * successfully offloaded, increment block offloads counter. On failure,
3316 * previously offloaded filter is considered to be intact and offloads counter
3317 * is not decremented.
3318 */
3319
3320int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3321 enum tc_setup_type type, void *type_data, bool err_stop,
3322 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3323{
3324 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3325 int ok_count;
3326
3327retry:
3328 if (take_rtnl)
3329 rtnl_lock();
3330 down_read(&block->cb_lock);
3331 /* Need to obtain rtnl lock if block is bound to devs that require it.
3332 * In block bind code cb_lock is obtained while holding rtnl, so we must
3333 * obtain the locks in same order here.
3334 */
3335 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3336 up_read(&block->cb_lock);
3337 take_rtnl = true;
3338 goto retry;
3339 }
3340
3341 /* Make sure all netdevs sharing this block are offload-capable. */
3342 if (block->nooffloaddevcnt && err_stop) {
3343 ok_count = -EOPNOTSUPP;
3344 goto err_unlock;
3345 }
3346
3347 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3348 if (ok_count < 0)
3349 goto err_unlock;
3350
3351 if (tp->ops->hw_add)
3352 tp->ops->hw_add(tp, type_data);
3353 if (ok_count > 0)
3354 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3355 ok_count, true);
3356err_unlock:
3357 up_read(&block->cb_lock);
3358 if (take_rtnl)
3359 rtnl_unlock();
3360 return min(ok_count, 0);
3361}
3362EXPORT_SYMBOL(tc_setup_cb_add);
3363
3364/* Destructive filter replace. If filter that wasn't already in hardware is
3365 * successfully offloaded, increment block offload counter. On failure,
3366 * previously offloaded filter is considered to be destroyed and offload counter
3367 * is decremented.
3368 */
3369
3370int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3371 enum tc_setup_type type, void *type_data, bool err_stop,
3372 u32 *old_flags, unsigned int *old_in_hw_count,
3373 u32 *new_flags, unsigned int *new_in_hw_count,
3374 bool rtnl_held)
3375{
3376 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3377 int ok_count;
3378
3379retry:
3380 if (take_rtnl)
3381 rtnl_lock();
3382 down_read(&block->cb_lock);
3383 /* Need to obtain rtnl lock if block is bound to devs that require it.
3384 * In block bind code cb_lock is obtained while holding rtnl, so we must
3385 * obtain the locks in same order here.
3386 */
3387 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3388 up_read(&block->cb_lock);
3389 take_rtnl = true;
3390 goto retry;
3391 }
3392
3393 /* Make sure all netdevs sharing this block are offload-capable. */
3394 if (block->nooffloaddevcnt && err_stop) {
3395 ok_count = -EOPNOTSUPP;
3396 goto err_unlock;
3397 }
3398
3399 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3400 if (tp->ops->hw_del)
3401 tp->ops->hw_del(tp, type_data);
3402
3403 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3404 if (ok_count < 0)
3405 goto err_unlock;
3406
3407 if (tp->ops->hw_add)
3408 tp->ops->hw_add(tp, type_data);
3409 if (ok_count > 0)
3410 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3411 new_flags, ok_count, true);
3412err_unlock:
3413 up_read(&block->cb_lock);
3414 if (take_rtnl)
3415 rtnl_unlock();
3416 return min(ok_count, 0);
3417}
3418EXPORT_SYMBOL(tc_setup_cb_replace);
3419
3420/* Destroy filter and decrement block offload counter, if filter was previously
3421 * offloaded.
3422 */
3423
3424int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3425 enum tc_setup_type type, void *type_data, bool err_stop,
3426 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3427{
3428 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3429 int ok_count;
3430
3431retry:
3432 if (take_rtnl)
3433 rtnl_lock();
3434 down_read(&block->cb_lock);
3435 /* Need to obtain rtnl lock if block is bound to devs that require it.
3436 * In block bind code cb_lock is obtained while holding rtnl, so we must
3437 * obtain the locks in same order here.
3438 */
3439 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3440 up_read(&block->cb_lock);
3441 take_rtnl = true;
3442 goto retry;
3443 }
3444
3445 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3446
3447 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3448 if (tp->ops->hw_del)
3449 tp->ops->hw_del(tp, type_data);
3450
3451 up_read(&block->cb_lock);
3452 if (take_rtnl)
3453 rtnl_unlock();
3454 return min(ok_count, 0);
3455}
3456EXPORT_SYMBOL(tc_setup_cb_destroy);
3457
3458int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3459 bool add, flow_setup_cb_t *cb,
3460 enum tc_setup_type type, void *type_data,
3461 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3462{
3463 int err = cb(type, type_data, cb_priv);
3464
3465 if (err) {
3466 if (add && tc_skip_sw(*flags))
3467 return err;
3468 } else {
3469 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3470 add);
3471 }
3472
3473 return 0;
3474}
3475EXPORT_SYMBOL(tc_setup_cb_reoffload);
3476
3477static int tcf_act_get_cookie(struct flow_action_entry *entry,
3478 const struct tc_action *act)
3479{
3480 struct tc_cookie *cookie;
3481 int err = 0;
3482
3483 rcu_read_lock();
3484 cookie = rcu_dereference(act->act_cookie);
3485 if (cookie) {
3486 entry->cookie = flow_action_cookie_create(cookie->data,
3487 cookie->len,
3488 GFP_ATOMIC);
3489 if (!entry->cookie)
3490 err = -ENOMEM;
3491 }
3492 rcu_read_unlock();
3493 return err;
3494}
3495
3496static void tcf_act_put_cookie(struct flow_action_entry *entry)
3497{
3498 flow_action_cookie_destroy(entry->cookie);
3499}
3500
3501void tc_cleanup_offload_action(struct flow_action *flow_action)
3502{
3503 struct flow_action_entry *entry;
3504 int i;
3505
3506 flow_action_for_each(i, entry, flow_action) {
3507 tcf_act_put_cookie(entry);
3508 if (entry->destructor)
3509 entry->destructor(entry->destructor_priv);
3510 }
3511}
3512EXPORT_SYMBOL(tc_cleanup_offload_action);
3513
3514static int tc_setup_offload_act(struct tc_action *act,
3515 struct flow_action_entry *entry,
3516 u32 *index_inc,
3517 struct netlink_ext_ack *extack)
3518{
3519#ifdef CONFIG_NET_CLS_ACT
3520 if (act->ops->offload_act_setup) {
3521 return act->ops->offload_act_setup(act, entry, index_inc, true,
3522 extack);
3523 } else {
3524 NL_SET_ERR_MSG(extack, "Action does not support offload");
3525 return -EOPNOTSUPP;
3526 }
3527#else
3528 return 0;
3529#endif
3530}
3531
3532int tc_setup_action(struct flow_action *flow_action,
3533 struct tc_action *actions[],
3534 struct netlink_ext_ack *extack)
3535{
3536 int i, j, k, index, err = 0;
3537 struct tc_action *act;
3538
3539 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3540 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3541 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3542
3543 if (!actions)
3544 return 0;
3545
3546 j = 0;
3547 tcf_act_for_each_action(i, act, actions) {
3548 struct flow_action_entry *entry;
3549
3550 entry = &flow_action->entries[j];
3551 spin_lock_bh(&act->tcfa_lock);
3552 err = tcf_act_get_cookie(entry, act);
3553 if (err)
3554 goto err_out_locked;
3555
3556 index = 0;
3557 err = tc_setup_offload_act(act, entry, &index, extack);
3558 if (err)
3559 goto err_out_locked;
3560
3561 for (k = 0; k < index ; k++) {
3562 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3563 entry[k].hw_index = act->tcfa_index;
3564 }
3565
3566 j += index;
3567
3568 spin_unlock_bh(&act->tcfa_lock);
3569 }
3570
3571err_out:
3572 if (err)
3573 tc_cleanup_offload_action(flow_action);
3574
3575 return err;
3576err_out_locked:
3577 spin_unlock_bh(&act->tcfa_lock);
3578 goto err_out;
3579}
3580
3581int tc_setup_offload_action(struct flow_action *flow_action,
3582 const struct tcf_exts *exts,
3583 struct netlink_ext_ack *extack)
3584{
3585#ifdef CONFIG_NET_CLS_ACT
3586 if (!exts)
3587 return 0;
3588
3589 return tc_setup_action(flow_action, exts->actions, extack);
3590#else
3591 return 0;
3592#endif
3593}
3594EXPORT_SYMBOL(tc_setup_offload_action);
3595
3596unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3597{
3598 unsigned int num_acts = 0;
3599 struct tc_action *act;
3600 int i;
3601
3602 tcf_exts_for_each_action(i, act, exts) {
3603 if (is_tcf_pedit(act))
3604 num_acts += tcf_pedit_nkeys(act);
3605 else
3606 num_acts++;
3607 }
3608 return num_acts;
3609}
3610EXPORT_SYMBOL(tcf_exts_num_actions);
3611
3612#ifdef CONFIG_NET_CLS_ACT
3613static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3614 u32 *p_block_index,
3615 struct netlink_ext_ack *extack)
3616{
3617 *p_block_index = nla_get_u32(block_index_attr);
3618 if (!*p_block_index) {
3619 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3620 return -EINVAL;
3621 }
3622
3623 return 0;
3624}
3625
3626int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3627 enum flow_block_binder_type binder_type,
3628 struct nlattr *block_index_attr,
3629 struct netlink_ext_ack *extack)
3630{
3631 u32 block_index;
3632 int err;
3633
3634 if (!block_index_attr)
3635 return 0;
3636
3637 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3638 if (err)
3639 return err;
3640
3641 qe->info.binder_type = binder_type;
3642 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3643 qe->info.chain_head_change_priv = &qe->filter_chain;
3644 qe->info.block_index = block_index;
3645
3646 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3647}
3648EXPORT_SYMBOL(tcf_qevent_init);
3649
3650void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3651{
3652 if (qe->info.block_index)
3653 tcf_block_put_ext(qe->block, sch, &qe->info);
3654}
3655EXPORT_SYMBOL(tcf_qevent_destroy);
3656
3657int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3658 struct netlink_ext_ack *extack)
3659{
3660 u32 block_index;
3661 int err;
3662
3663 if (!block_index_attr)
3664 return 0;
3665
3666 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3667 if (err)
3668 return err;
3669
3670 /* Bounce newly-configured block or change in block. */
3671 if (block_index != qe->info.block_index) {
3672 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3673 return -EINVAL;
3674 }
3675
3676 return 0;
3677}
3678EXPORT_SYMBOL(tcf_qevent_validate_change);
3679
3680struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3681 struct sk_buff **to_free, int *ret)
3682{
3683 struct tcf_result cl_res;
3684 struct tcf_proto *fl;
3685
3686 if (!qe->info.block_index)
3687 return skb;
3688
3689 fl = rcu_dereference_bh(qe->filter_chain);
3690
3691 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3692 case TC_ACT_SHOT:
3693 qdisc_qstats_drop(sch);
3694 __qdisc_drop(skb, to_free);
3695 *ret = __NET_XMIT_BYPASS;
3696 return NULL;
3697 case TC_ACT_STOLEN:
3698 case TC_ACT_QUEUED:
3699 case TC_ACT_TRAP:
3700 __qdisc_drop(skb, to_free);
3701 *ret = __NET_XMIT_STOLEN;
3702 return NULL;
3703 case TC_ACT_REDIRECT:
3704 skb_do_redirect(skb);
3705 *ret = __NET_XMIT_STOLEN;
3706 return NULL;
3707 }
3708
3709 return skb;
3710}
3711EXPORT_SYMBOL(tcf_qevent_handle);
3712
3713int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3714{
3715 if (!qe->info.block_index)
3716 return 0;
3717 return nla_put_u32(skb, attr_name, qe->info.block_index);
3718}
3719EXPORT_SYMBOL(tcf_qevent_dump);
3720#endif
3721
3722static __net_init int tcf_net_init(struct net *net)
3723{
3724 struct tcf_net *tn = net_generic(net, tcf_net_id);
3725
3726 spin_lock_init(&tn->idr_lock);
3727 idr_init(&tn->idr);
3728 return 0;
3729}
3730
3731static void __net_exit tcf_net_exit(struct net *net)
3732{
3733 struct tcf_net *tn = net_generic(net, tcf_net_id);
3734
3735 idr_destroy(&tn->idr);
3736}
3737
3738static struct pernet_operations tcf_net_ops = {
3739 .init = tcf_net_init,
3740 .exit = tcf_net_exit,
3741 .id = &tcf_net_id,
3742 .size = sizeof(struct tcf_net),
3743};
3744
3745static int __init tc_filter_init(void)
3746{
3747 int err;
3748
3749 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3750 if (!tc_filter_wq)
3751 return -ENOMEM;
3752
3753 err = register_pernet_subsys(&tcf_net_ops);
3754 if (err)
3755 goto err_register_pernet_subsys;
3756
3757 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3758 RTNL_FLAG_DOIT_UNLOCKED);
3759 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3760 RTNL_FLAG_DOIT_UNLOCKED);
3761 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3762 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3763 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3764 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3765 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3766 tc_dump_chain, 0);
3767
3768 return 0;
3769
3770err_register_pernet_subsys:
3771 destroy_workqueue(tc_filter_wq);
3772 return err;
3773}
3774
3775subsys_initcall(tc_filter_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/cls_api.c Packet classifier API.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Changes:
8 *
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10 */
11
12#include <linux/module.h>
13#include <linux/types.h>
14#include <linux/kernel.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/err.h>
18#include <linux/skbuff.h>
19#include <linux/init.h>
20#include <linux/kmod.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/jhash.h>
24#include <linux/rculist.h>
25#include <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 < 0)
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(NET_CLS_ALIAS_PREFIX "%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 tp->usesw = !tp->ops->reoffload;
394 spin_lock_init(&tp->lock);
395 refcount_set(&tp->refcnt, 1);
396
397 err = tp->ops->init(tp);
398 if (err) {
399 module_put(tp->ops->owner);
400 goto errout;
401 }
402 return tp;
403
404errout:
405 kfree(tp);
406 return ERR_PTR(err);
407}
408
409static void tcf_proto_get(struct tcf_proto *tp)
410{
411 refcount_inc(&tp->refcnt);
412}
413
414static void tcf_proto_count_usesw(struct tcf_proto *tp, bool add)
415{
416#ifdef CONFIG_NET_CLS_ACT
417 struct tcf_block *block = tp->chain->block;
418 bool counted = false;
419
420 if (!add) {
421 if (tp->usesw && tp->counted) {
422 if (!atomic_dec_return(&block->useswcnt))
423 static_branch_dec(&tcf_sw_enabled_key);
424 tp->counted = false;
425 }
426 return;
427 }
428
429 spin_lock(&tp->lock);
430 if (tp->usesw && !tp->counted) {
431 counted = true;
432 tp->counted = true;
433 }
434 spin_unlock(&tp->lock);
435
436 if (counted && atomic_inc_return(&block->useswcnt) == 1)
437 static_branch_inc(&tcf_sw_enabled_key);
438#endif
439}
440
441static void tcf_chain_put(struct tcf_chain *chain);
442
443static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
444 bool sig_destroy, struct netlink_ext_ack *extack)
445{
446 tp->ops->destroy(tp, rtnl_held, extack);
447 tcf_proto_count_usesw(tp, false);
448 if (sig_destroy)
449 tcf_proto_signal_destroyed(tp->chain, tp);
450 tcf_chain_put(tp->chain);
451 module_put(tp->ops->owner);
452 kfree_rcu(tp, rcu);
453}
454
455static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
456 struct netlink_ext_ack *extack)
457{
458 if (refcount_dec_and_test(&tp->refcnt))
459 tcf_proto_destroy(tp, rtnl_held, true, extack);
460}
461
462static bool tcf_proto_check_delete(struct tcf_proto *tp)
463{
464 if (tp->ops->delete_empty)
465 return tp->ops->delete_empty(tp);
466
467 tp->deleting = true;
468 return tp->deleting;
469}
470
471static void tcf_proto_mark_delete(struct tcf_proto *tp)
472{
473 spin_lock(&tp->lock);
474 tp->deleting = true;
475 spin_unlock(&tp->lock);
476}
477
478static bool tcf_proto_is_deleting(struct tcf_proto *tp)
479{
480 bool deleting;
481
482 spin_lock(&tp->lock);
483 deleting = tp->deleting;
484 spin_unlock(&tp->lock);
485
486 return deleting;
487}
488
489#define ASSERT_BLOCK_LOCKED(block) \
490 lockdep_assert_held(&(block)->lock)
491
492struct tcf_filter_chain_list_item {
493 struct list_head list;
494 tcf_chain_head_change_t *chain_head_change;
495 void *chain_head_change_priv;
496};
497
498static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
499 u32 chain_index)
500{
501 struct tcf_chain *chain;
502
503 ASSERT_BLOCK_LOCKED(block);
504
505 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
506 if (!chain)
507 return NULL;
508 list_add_tail_rcu(&chain->list, &block->chain_list);
509 mutex_init(&chain->filter_chain_lock);
510 chain->block = block;
511 chain->index = chain_index;
512 chain->refcnt = 1;
513 if (!chain->index)
514 block->chain0.chain = chain;
515 return chain;
516}
517
518static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
519 struct tcf_proto *tp_head)
520{
521 if (item->chain_head_change)
522 item->chain_head_change(tp_head, item->chain_head_change_priv);
523}
524
525static void tcf_chain0_head_change(struct tcf_chain *chain,
526 struct tcf_proto *tp_head)
527{
528 struct tcf_filter_chain_list_item *item;
529 struct tcf_block *block = chain->block;
530
531 if (chain->index)
532 return;
533
534 mutex_lock(&block->lock);
535 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
536 tcf_chain_head_change_item(item, tp_head);
537 mutex_unlock(&block->lock);
538}
539
540/* Returns true if block can be safely freed. */
541
542static bool tcf_chain_detach(struct tcf_chain *chain)
543{
544 struct tcf_block *block = chain->block;
545
546 ASSERT_BLOCK_LOCKED(block);
547
548 list_del_rcu(&chain->list);
549 if (!chain->index)
550 block->chain0.chain = NULL;
551
552 if (list_empty(&block->chain_list) &&
553 refcount_read(&block->refcnt) == 0)
554 return true;
555
556 return false;
557}
558
559static void tcf_block_destroy(struct tcf_block *block)
560{
561 mutex_destroy(&block->lock);
562 mutex_destroy(&block->proto_destroy_lock);
563 xa_destroy(&block->ports);
564 kfree_rcu(block, rcu);
565}
566
567static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
568{
569 struct tcf_block *block = chain->block;
570
571 mutex_destroy(&chain->filter_chain_lock);
572 kfree_rcu(chain, rcu);
573 if (free_block)
574 tcf_block_destroy(block);
575}
576
577static void tcf_chain_hold(struct tcf_chain *chain)
578{
579 ASSERT_BLOCK_LOCKED(chain->block);
580
581 ++chain->refcnt;
582}
583
584static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
585{
586 ASSERT_BLOCK_LOCKED(chain->block);
587
588 /* In case all the references are action references, this
589 * chain should not be shown to the user.
590 */
591 return chain->refcnt == chain->action_refcnt;
592}
593
594static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
595 u32 chain_index)
596{
597 struct tcf_chain *chain;
598
599 ASSERT_BLOCK_LOCKED(block);
600
601 list_for_each_entry(chain, &block->chain_list, list) {
602 if (chain->index == chain_index)
603 return chain;
604 }
605 return NULL;
606}
607
608#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
609static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
610 u32 chain_index)
611{
612 struct tcf_chain *chain;
613
614 list_for_each_entry_rcu(chain, &block->chain_list, list) {
615 if (chain->index == chain_index)
616 return chain;
617 }
618 return NULL;
619}
620#endif
621
622static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
623 u32 seq, u16 flags, int event, bool unicast,
624 struct netlink_ext_ack *extack);
625
626static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
627 u32 chain_index, bool create,
628 bool by_act)
629{
630 struct tcf_chain *chain = NULL;
631 bool is_first_reference;
632
633 mutex_lock(&block->lock);
634 chain = tcf_chain_lookup(block, chain_index);
635 if (chain) {
636 tcf_chain_hold(chain);
637 } else {
638 if (!create)
639 goto errout;
640 chain = tcf_chain_create(block, chain_index);
641 if (!chain)
642 goto errout;
643 }
644
645 if (by_act)
646 ++chain->action_refcnt;
647 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
648 mutex_unlock(&block->lock);
649
650 /* Send notification only in case we got the first
651 * non-action reference. Until then, the chain acts only as
652 * a placeholder for actions pointing to it and user ought
653 * not know about them.
654 */
655 if (is_first_reference && !by_act)
656 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
657 RTM_NEWCHAIN, false, NULL);
658
659 return chain;
660
661errout:
662 mutex_unlock(&block->lock);
663 return chain;
664}
665
666static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
667 bool create)
668{
669 return __tcf_chain_get(block, chain_index, create, false);
670}
671
672struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
673{
674 return __tcf_chain_get(block, chain_index, true, true);
675}
676EXPORT_SYMBOL(tcf_chain_get_by_act);
677
678static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
679 void *tmplt_priv);
680static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
681 void *tmplt_priv, u32 chain_index,
682 struct tcf_block *block, struct sk_buff *oskb,
683 u32 seq, u16 flags);
684
685static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
686 bool explicitly_created)
687{
688 struct tcf_block *block = chain->block;
689 const struct tcf_proto_ops *tmplt_ops;
690 unsigned int refcnt, non_act_refcnt;
691 bool free_block = false;
692 void *tmplt_priv;
693
694 mutex_lock(&block->lock);
695 if (explicitly_created) {
696 if (!chain->explicitly_created) {
697 mutex_unlock(&block->lock);
698 return;
699 }
700 chain->explicitly_created = false;
701 }
702
703 if (by_act)
704 chain->action_refcnt--;
705
706 /* tc_chain_notify_delete can't be called while holding block lock.
707 * However, when block is unlocked chain can be changed concurrently, so
708 * save these to temporary variables.
709 */
710 refcnt = --chain->refcnt;
711 non_act_refcnt = refcnt - chain->action_refcnt;
712 tmplt_ops = chain->tmplt_ops;
713 tmplt_priv = chain->tmplt_priv;
714
715 if (non_act_refcnt == chain->explicitly_created && !by_act) {
716 if (non_act_refcnt == 0)
717 tc_chain_notify_delete(tmplt_ops, tmplt_priv,
718 chain->index, block, NULL, 0, 0);
719 /* Last reference to chain, no need to lock. */
720 chain->flushing = false;
721 }
722
723 if (refcnt == 0)
724 free_block = tcf_chain_detach(chain);
725 mutex_unlock(&block->lock);
726
727 if (refcnt == 0) {
728 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
729 tcf_chain_destroy(chain, free_block);
730 }
731}
732
733static void tcf_chain_put(struct tcf_chain *chain)
734{
735 __tcf_chain_put(chain, false, false);
736}
737
738void tcf_chain_put_by_act(struct tcf_chain *chain)
739{
740 __tcf_chain_put(chain, true, false);
741}
742EXPORT_SYMBOL(tcf_chain_put_by_act);
743
744static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
745{
746 __tcf_chain_put(chain, false, true);
747}
748
749static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
750{
751 struct tcf_proto *tp, *tp_next;
752
753 mutex_lock(&chain->filter_chain_lock);
754 tp = tcf_chain_dereference(chain->filter_chain, chain);
755 while (tp) {
756 tp_next = rcu_dereference_protected(tp->next, 1);
757 tcf_proto_signal_destroying(chain, tp);
758 tp = tp_next;
759 }
760 tp = tcf_chain_dereference(chain->filter_chain, chain);
761 RCU_INIT_POINTER(chain->filter_chain, NULL);
762 tcf_chain0_head_change(chain, NULL);
763 chain->flushing = true;
764 mutex_unlock(&chain->filter_chain_lock);
765
766 while (tp) {
767 tp_next = rcu_dereference_protected(tp->next, 1);
768 tcf_proto_put(tp, rtnl_held, NULL);
769 tp = tp_next;
770 }
771}
772
773static int tcf_block_setup(struct tcf_block *block,
774 struct flow_block_offload *bo);
775
776static void tcf_block_offload_init(struct flow_block_offload *bo,
777 struct net_device *dev, struct Qdisc *sch,
778 enum flow_block_command command,
779 enum flow_block_binder_type binder_type,
780 struct flow_block *flow_block,
781 bool shared, struct netlink_ext_ack *extack)
782{
783 bo->net = dev_net(dev);
784 bo->command = command;
785 bo->binder_type = binder_type;
786 bo->block = flow_block;
787 bo->block_shared = shared;
788 bo->extack = extack;
789 bo->sch = sch;
790 bo->cb_list_head = &flow_block->cb_list;
791 INIT_LIST_HEAD(&bo->cb_list);
792}
793
794static void tcf_block_unbind(struct tcf_block *block,
795 struct flow_block_offload *bo);
796
797static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
798{
799 struct tcf_block *block = block_cb->indr.data;
800 struct net_device *dev = block_cb->indr.dev;
801 struct Qdisc *sch = block_cb->indr.sch;
802 struct netlink_ext_ack extack = {};
803 struct flow_block_offload bo = {};
804
805 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
806 block_cb->indr.binder_type,
807 &block->flow_block, tcf_block_shared(block),
808 &extack);
809 rtnl_lock();
810 down_write(&block->cb_lock);
811 list_del(&block_cb->driver_list);
812 list_move(&block_cb->list, &bo.cb_list);
813 tcf_block_unbind(block, &bo);
814 up_write(&block->cb_lock);
815 rtnl_unlock();
816}
817
818static bool tcf_block_offload_in_use(struct tcf_block *block)
819{
820 return atomic_read(&block->offloadcnt);
821}
822
823static int tcf_block_offload_cmd(struct tcf_block *block,
824 struct net_device *dev, struct Qdisc *sch,
825 struct tcf_block_ext_info *ei,
826 enum flow_block_command command,
827 struct netlink_ext_ack *extack)
828{
829 struct flow_block_offload bo = {};
830
831 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
832 &block->flow_block, tcf_block_shared(block),
833 extack);
834
835 if (dev->netdev_ops->ndo_setup_tc) {
836 int err;
837
838 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
839 if (err < 0) {
840 if (err != -EOPNOTSUPP)
841 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
842 return err;
843 }
844
845 return tcf_block_setup(block, &bo);
846 }
847
848 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
849 tc_block_indr_cleanup);
850 tcf_block_setup(block, &bo);
851
852 return -EOPNOTSUPP;
853}
854
855static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
856 struct tcf_block_ext_info *ei,
857 struct netlink_ext_ack *extack)
858{
859 struct net_device *dev = q->dev_queue->dev;
860 int err;
861
862 down_write(&block->cb_lock);
863
864 /* If tc offload feature is disabled and the block we try to bind
865 * to already has some offloaded filters, forbid to bind.
866 */
867 if (dev->netdev_ops->ndo_setup_tc &&
868 !tc_can_offload(dev) &&
869 tcf_block_offload_in_use(block)) {
870 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
871 err = -EOPNOTSUPP;
872 goto err_unlock;
873 }
874
875 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
876 if (err == -EOPNOTSUPP)
877 goto no_offload_dev_inc;
878 if (err)
879 goto err_unlock;
880
881 up_write(&block->cb_lock);
882 return 0;
883
884no_offload_dev_inc:
885 if (tcf_block_offload_in_use(block))
886 goto err_unlock;
887
888 err = 0;
889 block->nooffloaddevcnt++;
890err_unlock:
891 up_write(&block->cb_lock);
892 return err;
893}
894
895static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
896 struct tcf_block_ext_info *ei)
897{
898 struct net_device *dev = q->dev_queue->dev;
899 int err;
900
901 down_write(&block->cb_lock);
902 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
903 if (err == -EOPNOTSUPP)
904 goto no_offload_dev_dec;
905 up_write(&block->cb_lock);
906 return;
907
908no_offload_dev_dec:
909 WARN_ON(block->nooffloaddevcnt-- == 0);
910 up_write(&block->cb_lock);
911}
912
913static int
914tcf_chain0_head_change_cb_add(struct tcf_block *block,
915 struct tcf_block_ext_info *ei,
916 struct netlink_ext_ack *extack)
917{
918 struct tcf_filter_chain_list_item *item;
919 struct tcf_chain *chain0;
920
921 item = kmalloc(sizeof(*item), GFP_KERNEL);
922 if (!item) {
923 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
924 return -ENOMEM;
925 }
926 item->chain_head_change = ei->chain_head_change;
927 item->chain_head_change_priv = ei->chain_head_change_priv;
928
929 mutex_lock(&block->lock);
930 chain0 = block->chain0.chain;
931 if (chain0)
932 tcf_chain_hold(chain0);
933 else
934 list_add(&item->list, &block->chain0.filter_chain_list);
935 mutex_unlock(&block->lock);
936
937 if (chain0) {
938 struct tcf_proto *tp_head;
939
940 mutex_lock(&chain0->filter_chain_lock);
941
942 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
943 if (tp_head)
944 tcf_chain_head_change_item(item, tp_head);
945
946 mutex_lock(&block->lock);
947 list_add(&item->list, &block->chain0.filter_chain_list);
948 mutex_unlock(&block->lock);
949
950 mutex_unlock(&chain0->filter_chain_lock);
951 tcf_chain_put(chain0);
952 }
953
954 return 0;
955}
956
957static void
958tcf_chain0_head_change_cb_del(struct tcf_block *block,
959 struct tcf_block_ext_info *ei)
960{
961 struct tcf_filter_chain_list_item *item;
962
963 mutex_lock(&block->lock);
964 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
965 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
966 (item->chain_head_change == ei->chain_head_change &&
967 item->chain_head_change_priv == ei->chain_head_change_priv)) {
968 if (block->chain0.chain)
969 tcf_chain_head_change_item(item, NULL);
970 list_del(&item->list);
971 mutex_unlock(&block->lock);
972
973 kfree(item);
974 return;
975 }
976 }
977 mutex_unlock(&block->lock);
978 WARN_ON(1);
979}
980
981struct tcf_net {
982 spinlock_t idr_lock; /* Protects idr */
983 struct idr idr;
984};
985
986static unsigned int tcf_net_id;
987
988static int tcf_block_insert(struct tcf_block *block, struct net *net,
989 struct netlink_ext_ack *extack)
990{
991 struct tcf_net *tn = net_generic(net, tcf_net_id);
992 int err;
993
994 idr_preload(GFP_KERNEL);
995 spin_lock(&tn->idr_lock);
996 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
997 GFP_NOWAIT);
998 spin_unlock(&tn->idr_lock);
999 idr_preload_end();
1000
1001 return err;
1002}
1003
1004static void tcf_block_remove(struct tcf_block *block, struct net *net)
1005{
1006 struct tcf_net *tn = net_generic(net, tcf_net_id);
1007
1008 spin_lock(&tn->idr_lock);
1009 idr_remove(&tn->idr, block->index);
1010 spin_unlock(&tn->idr_lock);
1011}
1012
1013static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
1014 u32 block_index,
1015 struct netlink_ext_ack *extack)
1016{
1017 struct tcf_block *block;
1018
1019 block = kzalloc(sizeof(*block), GFP_KERNEL);
1020 if (!block) {
1021 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
1022 return ERR_PTR(-ENOMEM);
1023 }
1024 mutex_init(&block->lock);
1025 mutex_init(&block->proto_destroy_lock);
1026 init_rwsem(&block->cb_lock);
1027 flow_block_init(&block->flow_block);
1028 INIT_LIST_HEAD(&block->chain_list);
1029 INIT_LIST_HEAD(&block->owner_list);
1030 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1031
1032 refcount_set(&block->refcnt, 1);
1033 block->net = net;
1034 block->index = block_index;
1035 xa_init(&block->ports);
1036
1037 /* Don't store q pointer for blocks which are shared */
1038 if (!tcf_block_shared(block))
1039 block->q = q;
1040 return block;
1041}
1042
1043struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1044{
1045 struct tcf_net *tn = net_generic(net, tcf_net_id);
1046
1047 return idr_find(&tn->idr, block_index);
1048}
1049EXPORT_SYMBOL(tcf_block_lookup);
1050
1051static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1052{
1053 struct tcf_block *block;
1054
1055 rcu_read_lock();
1056 block = tcf_block_lookup(net, block_index);
1057 if (block && !refcount_inc_not_zero(&block->refcnt))
1058 block = NULL;
1059 rcu_read_unlock();
1060
1061 return block;
1062}
1063
1064static struct tcf_chain *
1065__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1066{
1067 mutex_lock(&block->lock);
1068 if (chain)
1069 chain = list_is_last(&chain->list, &block->chain_list) ?
1070 NULL : list_next_entry(chain, list);
1071 else
1072 chain = list_first_entry_or_null(&block->chain_list,
1073 struct tcf_chain, list);
1074
1075 /* skip all action-only chains */
1076 while (chain && tcf_chain_held_by_acts_only(chain))
1077 chain = list_is_last(&chain->list, &block->chain_list) ?
1078 NULL : list_next_entry(chain, list);
1079
1080 if (chain)
1081 tcf_chain_hold(chain);
1082 mutex_unlock(&block->lock);
1083
1084 return chain;
1085}
1086
1087/* Function to be used by all clients that want to iterate over all chains on
1088 * block. It properly obtains block->lock and takes reference to chain before
1089 * returning it. Users of this function must be tolerant to concurrent chain
1090 * insertion/deletion or ensure that no concurrent chain modification is
1091 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1092 * consistent dump because rtnl lock is released each time skb is filled with
1093 * data and sent to user-space.
1094 */
1095
1096struct tcf_chain *
1097tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1098{
1099 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1100
1101 if (chain)
1102 tcf_chain_put(chain);
1103
1104 return chain_next;
1105}
1106EXPORT_SYMBOL(tcf_get_next_chain);
1107
1108static struct tcf_proto *
1109__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1110{
1111 u32 prio = 0;
1112
1113 ASSERT_RTNL();
1114 mutex_lock(&chain->filter_chain_lock);
1115
1116 if (!tp) {
1117 tp = tcf_chain_dereference(chain->filter_chain, chain);
1118 } else if (tcf_proto_is_deleting(tp)) {
1119 /* 'deleting' flag is set and chain->filter_chain_lock was
1120 * unlocked, which means next pointer could be invalid. Restart
1121 * search.
1122 */
1123 prio = tp->prio + 1;
1124 tp = tcf_chain_dereference(chain->filter_chain, chain);
1125
1126 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1127 if (!tp->deleting && tp->prio >= prio)
1128 break;
1129 } else {
1130 tp = tcf_chain_dereference(tp->next, chain);
1131 }
1132
1133 if (tp)
1134 tcf_proto_get(tp);
1135
1136 mutex_unlock(&chain->filter_chain_lock);
1137
1138 return tp;
1139}
1140
1141/* Function to be used by all clients that want to iterate over all tp's on
1142 * chain. Users of this function must be tolerant to concurrent tp
1143 * insertion/deletion or ensure that no concurrent chain modification is
1144 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1145 * consistent dump because rtnl lock is released each time skb is filled with
1146 * data and sent to user-space.
1147 */
1148
1149struct tcf_proto *
1150tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1151{
1152 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1153
1154 if (tp)
1155 tcf_proto_put(tp, true, NULL);
1156
1157 return tp_next;
1158}
1159EXPORT_SYMBOL(tcf_get_next_proto);
1160
1161static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1162{
1163 struct tcf_chain *chain;
1164
1165 /* Last reference to block. At this point chains cannot be added or
1166 * removed concurrently.
1167 */
1168 for (chain = tcf_get_next_chain(block, NULL);
1169 chain;
1170 chain = tcf_get_next_chain(block, chain)) {
1171 tcf_chain_put_explicitly_created(chain);
1172 tcf_chain_flush(chain, rtnl_held);
1173 }
1174}
1175
1176/* Lookup Qdisc and increments its reference counter.
1177 * Set parent, if necessary.
1178 */
1179
1180static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1181 u32 *parent, int ifindex, bool rtnl_held,
1182 struct netlink_ext_ack *extack)
1183{
1184 const struct Qdisc_class_ops *cops;
1185 struct net_device *dev;
1186 int err = 0;
1187
1188 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1189 return 0;
1190
1191 rcu_read_lock();
1192
1193 /* Find link */
1194 dev = dev_get_by_index_rcu(net, ifindex);
1195 if (!dev) {
1196 rcu_read_unlock();
1197 return -ENODEV;
1198 }
1199
1200 /* Find qdisc */
1201 if (!*parent) {
1202 *q = rcu_dereference(dev->qdisc);
1203 *parent = (*q)->handle;
1204 } else {
1205 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1206 if (!*q) {
1207 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1208 err = -EINVAL;
1209 goto errout_rcu;
1210 }
1211 }
1212
1213 *q = qdisc_refcount_inc_nz(*q);
1214 if (!*q) {
1215 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1216 err = -EINVAL;
1217 goto errout_rcu;
1218 }
1219
1220 /* Is it classful? */
1221 cops = (*q)->ops->cl_ops;
1222 if (!cops) {
1223 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1224 err = -EINVAL;
1225 goto errout_qdisc;
1226 }
1227
1228 if (!cops->tcf_block) {
1229 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1230 err = -EOPNOTSUPP;
1231 goto errout_qdisc;
1232 }
1233
1234errout_rcu:
1235 /* At this point we know that qdisc is not noop_qdisc,
1236 * which means that qdisc holds a reference to net_device
1237 * and we hold a reference to qdisc, so it is safe to release
1238 * rcu read lock.
1239 */
1240 rcu_read_unlock();
1241 return err;
1242
1243errout_qdisc:
1244 rcu_read_unlock();
1245
1246 if (rtnl_held)
1247 qdisc_put(*q);
1248 else
1249 qdisc_put_unlocked(*q);
1250 *q = NULL;
1251
1252 return err;
1253}
1254
1255static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1256 int ifindex, struct netlink_ext_ack *extack)
1257{
1258 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1259 return 0;
1260
1261 /* Do we search for filter, attached to class? */
1262 if (TC_H_MIN(parent)) {
1263 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1264
1265 *cl = cops->find(q, parent);
1266 if (*cl == 0) {
1267 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1268 return -ENOENT;
1269 }
1270 }
1271
1272 return 0;
1273}
1274
1275static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1276 unsigned long cl, int ifindex,
1277 u32 block_index,
1278 struct netlink_ext_ack *extack)
1279{
1280 struct tcf_block *block;
1281
1282 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1283 block = tcf_block_refcnt_get(net, block_index);
1284 if (!block) {
1285 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1286 return ERR_PTR(-EINVAL);
1287 }
1288 } else {
1289 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1290
1291 block = cops->tcf_block(q, cl, extack);
1292 if (!block)
1293 return ERR_PTR(-EINVAL);
1294
1295 if (tcf_block_shared(block)) {
1296 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1297 return ERR_PTR(-EOPNOTSUPP);
1298 }
1299
1300 /* Always take reference to block in order to support execution
1301 * of rules update path of cls API without rtnl lock. Caller
1302 * must release block when it is finished using it. 'if' block
1303 * of this conditional obtain reference to block by calling
1304 * tcf_block_refcnt_get().
1305 */
1306 refcount_inc(&block->refcnt);
1307 }
1308
1309 return block;
1310}
1311
1312static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1313 struct tcf_block_ext_info *ei, bool rtnl_held)
1314{
1315 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1316 /* Flushing/putting all chains will cause the block to be
1317 * deallocated when last chain is freed. However, if chain_list
1318 * is empty, block has to be manually deallocated. After block
1319 * reference counter reached 0, it is no longer possible to
1320 * increment it or add new chains to block.
1321 */
1322 bool free_block = list_empty(&block->chain_list);
1323
1324 mutex_unlock(&block->lock);
1325 if (tcf_block_shared(block))
1326 tcf_block_remove(block, block->net);
1327
1328 if (q)
1329 tcf_block_offload_unbind(block, q, ei);
1330
1331 if (free_block)
1332 tcf_block_destroy(block);
1333 else
1334 tcf_block_flush_all_chains(block, rtnl_held);
1335 } else if (q) {
1336 tcf_block_offload_unbind(block, q, ei);
1337 }
1338}
1339
1340static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1341{
1342 __tcf_block_put(block, NULL, NULL, rtnl_held);
1343}
1344
1345/* Find tcf block.
1346 * Set q, parent, cl when appropriate.
1347 */
1348
1349static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1350 u32 *parent, unsigned long *cl,
1351 int ifindex, u32 block_index,
1352 struct netlink_ext_ack *extack)
1353{
1354 struct tcf_block *block;
1355 int err = 0;
1356
1357 ASSERT_RTNL();
1358
1359 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1360 if (err)
1361 goto errout;
1362
1363 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1364 if (err)
1365 goto errout_qdisc;
1366
1367 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1368 if (IS_ERR(block)) {
1369 err = PTR_ERR(block);
1370 goto errout_qdisc;
1371 }
1372
1373 return block;
1374
1375errout_qdisc:
1376 if (*q)
1377 qdisc_put(*q);
1378errout:
1379 *q = NULL;
1380 return ERR_PTR(err);
1381}
1382
1383static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1384 bool rtnl_held)
1385{
1386 if (!IS_ERR_OR_NULL(block))
1387 tcf_block_refcnt_put(block, rtnl_held);
1388
1389 if (q) {
1390 if (rtnl_held)
1391 qdisc_put(q);
1392 else
1393 qdisc_put_unlocked(q);
1394 }
1395}
1396
1397struct tcf_block_owner_item {
1398 struct list_head list;
1399 struct Qdisc *q;
1400 enum flow_block_binder_type binder_type;
1401};
1402
1403static void
1404tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1405 struct Qdisc *q,
1406 enum flow_block_binder_type binder_type)
1407{
1408 if (block->keep_dst &&
1409 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1410 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1411 netif_keep_dst(qdisc_dev(q));
1412}
1413
1414void tcf_block_netif_keep_dst(struct tcf_block *block)
1415{
1416 struct tcf_block_owner_item *item;
1417
1418 block->keep_dst = true;
1419 list_for_each_entry(item, &block->owner_list, list)
1420 tcf_block_owner_netif_keep_dst(block, item->q,
1421 item->binder_type);
1422}
1423EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1424
1425static int tcf_block_owner_add(struct tcf_block *block,
1426 struct Qdisc *q,
1427 enum flow_block_binder_type binder_type)
1428{
1429 struct tcf_block_owner_item *item;
1430
1431 item = kmalloc(sizeof(*item), GFP_KERNEL);
1432 if (!item)
1433 return -ENOMEM;
1434 item->q = q;
1435 item->binder_type = binder_type;
1436 list_add(&item->list, &block->owner_list);
1437 return 0;
1438}
1439
1440static void tcf_block_owner_del(struct tcf_block *block,
1441 struct Qdisc *q,
1442 enum flow_block_binder_type binder_type)
1443{
1444 struct tcf_block_owner_item *item;
1445
1446 list_for_each_entry(item, &block->owner_list, list) {
1447 if (item->q == q && item->binder_type == binder_type) {
1448 list_del(&item->list);
1449 kfree(item);
1450 return;
1451 }
1452 }
1453 WARN_ON(1);
1454}
1455
1456static bool tcf_block_tracks_dev(struct tcf_block *block,
1457 struct tcf_block_ext_info *ei)
1458{
1459 return tcf_block_shared(block) &&
1460 (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1461 ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1462}
1463
1464int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1465 struct tcf_block_ext_info *ei,
1466 struct netlink_ext_ack *extack)
1467{
1468 struct net_device *dev = qdisc_dev(q);
1469 struct net *net = qdisc_net(q);
1470 struct tcf_block *block = NULL;
1471 int err;
1472
1473 if (ei->block_index)
1474 /* block_index not 0 means the shared block is requested */
1475 block = tcf_block_refcnt_get(net, ei->block_index);
1476
1477 if (!block) {
1478 block = tcf_block_create(net, q, ei->block_index, extack);
1479 if (IS_ERR(block))
1480 return PTR_ERR(block);
1481 if (tcf_block_shared(block)) {
1482 err = tcf_block_insert(block, net, extack);
1483 if (err)
1484 goto err_block_insert;
1485 }
1486 }
1487
1488 err = tcf_block_owner_add(block, q, ei->binder_type);
1489 if (err)
1490 goto err_block_owner_add;
1491
1492 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1493
1494 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1495 if (err)
1496 goto err_chain0_head_change_cb_add;
1497
1498 err = tcf_block_offload_bind(block, q, ei, extack);
1499 if (err)
1500 goto err_block_offload_bind;
1501
1502 if (tcf_block_tracks_dev(block, ei)) {
1503 err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1504 if (err) {
1505 NL_SET_ERR_MSG(extack, "block dev insert failed");
1506 goto err_dev_insert;
1507 }
1508 }
1509
1510 *p_block = block;
1511 return 0;
1512
1513err_dev_insert:
1514 tcf_block_offload_unbind(block, q, ei);
1515err_block_offload_bind:
1516 tcf_chain0_head_change_cb_del(block, ei);
1517err_chain0_head_change_cb_add:
1518 tcf_block_owner_del(block, q, ei->binder_type);
1519err_block_owner_add:
1520err_block_insert:
1521 tcf_block_refcnt_put(block, true);
1522 return err;
1523}
1524EXPORT_SYMBOL(tcf_block_get_ext);
1525
1526static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1527{
1528 struct tcf_proto __rcu **p_filter_chain = priv;
1529
1530 rcu_assign_pointer(*p_filter_chain, tp_head);
1531}
1532
1533int tcf_block_get(struct tcf_block **p_block,
1534 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1535 struct netlink_ext_ack *extack)
1536{
1537 struct tcf_block_ext_info ei = {
1538 .chain_head_change = tcf_chain_head_change_dflt,
1539 .chain_head_change_priv = p_filter_chain,
1540 };
1541
1542 WARN_ON(!p_filter_chain);
1543 return tcf_block_get_ext(p_block, q, &ei, extack);
1544}
1545EXPORT_SYMBOL(tcf_block_get);
1546
1547/* XXX: Standalone actions are not allowed to jump to any chain, and bound
1548 * actions should be all removed after flushing.
1549 */
1550void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1551 struct tcf_block_ext_info *ei)
1552{
1553 struct net_device *dev = qdisc_dev(q);
1554
1555 if (!block)
1556 return;
1557 if (tcf_block_tracks_dev(block, ei))
1558 xa_erase(&block->ports, dev->ifindex);
1559 tcf_chain0_head_change_cb_del(block, ei);
1560 tcf_block_owner_del(block, q, ei->binder_type);
1561
1562 __tcf_block_put(block, q, ei, true);
1563}
1564EXPORT_SYMBOL(tcf_block_put_ext);
1565
1566void tcf_block_put(struct tcf_block *block)
1567{
1568 struct tcf_block_ext_info ei = {0, };
1569
1570 if (!block)
1571 return;
1572 tcf_block_put_ext(block, block->q, &ei);
1573}
1574
1575EXPORT_SYMBOL(tcf_block_put);
1576
1577static int
1578tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1579 void *cb_priv, bool add, bool offload_in_use,
1580 struct netlink_ext_ack *extack)
1581{
1582 struct tcf_chain *chain, *chain_prev;
1583 struct tcf_proto *tp, *tp_prev;
1584 int err;
1585
1586 lockdep_assert_held(&block->cb_lock);
1587
1588 for (chain = __tcf_get_next_chain(block, NULL);
1589 chain;
1590 chain_prev = chain,
1591 chain = __tcf_get_next_chain(block, chain),
1592 tcf_chain_put(chain_prev)) {
1593 if (chain->tmplt_ops && add)
1594 chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1595 cb_priv);
1596 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1597 tp_prev = tp,
1598 tp = __tcf_get_next_proto(chain, tp),
1599 tcf_proto_put(tp_prev, true, NULL)) {
1600 if (tp->ops->reoffload) {
1601 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1602 extack);
1603 if (err && add)
1604 goto err_playback_remove;
1605 } else if (add && offload_in_use) {
1606 err = -EOPNOTSUPP;
1607 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1608 goto err_playback_remove;
1609 }
1610 }
1611 if (chain->tmplt_ops && !add)
1612 chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1613 cb_priv);
1614 }
1615
1616 return 0;
1617
1618err_playback_remove:
1619 tcf_proto_put(tp, true, NULL);
1620 tcf_chain_put(chain);
1621 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1622 extack);
1623 return err;
1624}
1625
1626static int tcf_block_bind(struct tcf_block *block,
1627 struct flow_block_offload *bo)
1628{
1629 struct flow_block_cb *block_cb, *next;
1630 int err, i = 0;
1631
1632 lockdep_assert_held(&block->cb_lock);
1633
1634 list_for_each_entry(block_cb, &bo->cb_list, list) {
1635 err = tcf_block_playback_offloads(block, block_cb->cb,
1636 block_cb->cb_priv, true,
1637 tcf_block_offload_in_use(block),
1638 bo->extack);
1639 if (err)
1640 goto err_unroll;
1641 if (!bo->unlocked_driver_cb)
1642 block->lockeddevcnt++;
1643
1644 i++;
1645 }
1646 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1647
1648 return 0;
1649
1650err_unroll:
1651 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1652 list_del(&block_cb->driver_list);
1653 if (i-- > 0) {
1654 list_del(&block_cb->list);
1655 tcf_block_playback_offloads(block, block_cb->cb,
1656 block_cb->cb_priv, false,
1657 tcf_block_offload_in_use(block),
1658 NULL);
1659 if (!bo->unlocked_driver_cb)
1660 block->lockeddevcnt--;
1661 }
1662 flow_block_cb_free(block_cb);
1663 }
1664
1665 return err;
1666}
1667
1668static void tcf_block_unbind(struct tcf_block *block,
1669 struct flow_block_offload *bo)
1670{
1671 struct flow_block_cb *block_cb, *next;
1672
1673 lockdep_assert_held(&block->cb_lock);
1674
1675 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1676 tcf_block_playback_offloads(block, block_cb->cb,
1677 block_cb->cb_priv, false,
1678 tcf_block_offload_in_use(block),
1679 NULL);
1680 list_del(&block_cb->list);
1681 flow_block_cb_free(block_cb);
1682 if (!bo->unlocked_driver_cb)
1683 block->lockeddevcnt--;
1684 }
1685}
1686
1687static int tcf_block_setup(struct tcf_block *block,
1688 struct flow_block_offload *bo)
1689{
1690 int err;
1691
1692 switch (bo->command) {
1693 case FLOW_BLOCK_BIND:
1694 err = tcf_block_bind(block, bo);
1695 break;
1696 case FLOW_BLOCK_UNBIND:
1697 err = 0;
1698 tcf_block_unbind(block, bo);
1699 break;
1700 default:
1701 WARN_ON_ONCE(1);
1702 err = -EOPNOTSUPP;
1703 }
1704
1705 return err;
1706}
1707
1708/* Main classifier routine: scans classifier chain attached
1709 * to this qdisc, (optionally) tests for protocol and asks
1710 * specific classifiers.
1711 */
1712static inline int __tcf_classify(struct sk_buff *skb,
1713 const struct tcf_proto *tp,
1714 const struct tcf_proto *orig_tp,
1715 struct tcf_result *res,
1716 bool compat_mode,
1717 struct tcf_exts_miss_cookie_node *n,
1718 int act_index,
1719 u32 *last_executed_chain)
1720{
1721#ifdef CONFIG_NET_CLS_ACT
1722 const int max_reclassify_loop = 16;
1723 const struct tcf_proto *first_tp;
1724 int limit = 0;
1725
1726reclassify:
1727#endif
1728 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1729 __be16 protocol = skb_protocol(skb, false);
1730 int err = 0;
1731
1732 if (n) {
1733 struct tcf_exts *exts;
1734
1735 if (n->tp_prio != tp->prio)
1736 continue;
1737
1738 /* We re-lookup the tp and chain based on index instead
1739 * of having hard refs and locks to them, so do a sanity
1740 * check if any of tp,chain,exts was replaced by the
1741 * time we got here with a cookie from hardware.
1742 */
1743 if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1744 !tp->ops->get_exts)) {
1745 tcf_set_drop_reason(skb,
1746 SKB_DROP_REASON_TC_COOKIE_ERROR);
1747 return TC_ACT_SHOT;
1748 }
1749
1750 exts = tp->ops->get_exts(tp, n->handle);
1751 if (unlikely(!exts || n->exts != exts)) {
1752 tcf_set_drop_reason(skb,
1753 SKB_DROP_REASON_TC_COOKIE_ERROR);
1754 return TC_ACT_SHOT;
1755 }
1756
1757 n = NULL;
1758 err = tcf_exts_exec_ex(skb, exts, act_index, res);
1759 } else {
1760 if (tp->protocol != protocol &&
1761 tp->protocol != htons(ETH_P_ALL))
1762 continue;
1763
1764 err = tc_classify(skb, tp, res);
1765 }
1766#ifdef CONFIG_NET_CLS_ACT
1767 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1768 first_tp = orig_tp;
1769 *last_executed_chain = first_tp->chain->index;
1770 goto reset;
1771 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1772 first_tp = res->goto_tp;
1773 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1774 goto reset;
1775 }
1776#endif
1777 if (err >= 0)
1778 return err;
1779 }
1780
1781 if (unlikely(n)) {
1782 tcf_set_drop_reason(skb,
1783 SKB_DROP_REASON_TC_COOKIE_ERROR);
1784 return TC_ACT_SHOT;
1785 }
1786
1787 return TC_ACT_UNSPEC; /* signal: continue lookup */
1788#ifdef CONFIG_NET_CLS_ACT
1789reset:
1790 if (unlikely(limit++ >= max_reclassify_loop)) {
1791 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1792 tp->chain->block->index,
1793 tp->prio & 0xffff,
1794 ntohs(tp->protocol));
1795 tcf_set_drop_reason(skb,
1796 SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1797 return TC_ACT_SHOT;
1798 }
1799
1800 tp = first_tp;
1801 goto reclassify;
1802#endif
1803}
1804
1805int tcf_classify(struct sk_buff *skb,
1806 const struct tcf_block *block,
1807 const struct tcf_proto *tp,
1808 struct tcf_result *res, bool compat_mode)
1809{
1810#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1811 u32 last_executed_chain = 0;
1812
1813 return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1814 &last_executed_chain);
1815#else
1816 u32 last_executed_chain = tp ? tp->chain->index : 0;
1817 struct tcf_exts_miss_cookie_node *n = NULL;
1818 const struct tcf_proto *orig_tp = tp;
1819 struct tc_skb_ext *ext;
1820 int act_index = 0;
1821 int ret;
1822
1823 if (block) {
1824 ext = skb_ext_find(skb, TC_SKB_EXT);
1825
1826 if (ext && (ext->chain || ext->act_miss)) {
1827 struct tcf_chain *fchain;
1828 u32 chain;
1829
1830 if (ext->act_miss) {
1831 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1832 &act_index);
1833 if (!n) {
1834 tcf_set_drop_reason(skb,
1835 SKB_DROP_REASON_TC_COOKIE_ERROR);
1836 return TC_ACT_SHOT;
1837 }
1838
1839 chain = n->chain_index;
1840 } else {
1841 chain = ext->chain;
1842 }
1843
1844 fchain = tcf_chain_lookup_rcu(block, chain);
1845 if (!fchain) {
1846 tcf_set_drop_reason(skb,
1847 SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1848
1849 return TC_ACT_SHOT;
1850 }
1851
1852 /* Consume, so cloned/redirect skbs won't inherit ext */
1853 skb_ext_del(skb, TC_SKB_EXT);
1854
1855 tp = rcu_dereference_bh(fchain->filter_chain);
1856 last_executed_chain = fchain->index;
1857 }
1858 }
1859
1860 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1861 &last_executed_chain);
1862
1863 if (tc_skb_ext_tc_enabled()) {
1864 /* If we missed on some chain */
1865 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1866 struct tc_skb_cb *cb = tc_skb_cb(skb);
1867
1868 ext = tc_skb_ext_alloc(skb);
1869 if (WARN_ON_ONCE(!ext)) {
1870 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1871 return TC_ACT_SHOT;
1872 }
1873 ext->chain = last_executed_chain;
1874 ext->mru = cb->mru;
1875 ext->post_ct = cb->post_ct;
1876 ext->post_ct_snat = cb->post_ct_snat;
1877 ext->post_ct_dnat = cb->post_ct_dnat;
1878 ext->zone = cb->zone;
1879 }
1880 }
1881
1882 return ret;
1883#endif
1884}
1885EXPORT_SYMBOL(tcf_classify);
1886
1887struct tcf_chain_info {
1888 struct tcf_proto __rcu **pprev;
1889 struct tcf_proto __rcu *next;
1890};
1891
1892static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1893 struct tcf_chain_info *chain_info)
1894{
1895 return tcf_chain_dereference(*chain_info->pprev, chain);
1896}
1897
1898static int tcf_chain_tp_insert(struct tcf_chain *chain,
1899 struct tcf_chain_info *chain_info,
1900 struct tcf_proto *tp)
1901{
1902 if (chain->flushing)
1903 return -EAGAIN;
1904
1905 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1906 if (*chain_info->pprev == chain->filter_chain)
1907 tcf_chain0_head_change(chain, tp);
1908 tcf_proto_get(tp);
1909 rcu_assign_pointer(*chain_info->pprev, tp);
1910
1911 return 0;
1912}
1913
1914static void tcf_chain_tp_remove(struct tcf_chain *chain,
1915 struct tcf_chain_info *chain_info,
1916 struct tcf_proto *tp)
1917{
1918 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1919
1920 tcf_proto_mark_delete(tp);
1921 if (tp == chain->filter_chain)
1922 tcf_chain0_head_change(chain, next);
1923 RCU_INIT_POINTER(*chain_info->pprev, next);
1924}
1925
1926static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1927 struct tcf_chain_info *chain_info,
1928 u32 protocol, u32 prio,
1929 bool prio_allocate,
1930 struct netlink_ext_ack *extack);
1931
1932/* Try to insert new proto.
1933 * If proto with specified priority already exists, free new proto
1934 * and return existing one.
1935 */
1936
1937static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1938 struct tcf_proto *tp_new,
1939 u32 protocol, u32 prio,
1940 bool rtnl_held)
1941{
1942 struct tcf_chain_info chain_info;
1943 struct tcf_proto *tp;
1944 int err = 0;
1945
1946 mutex_lock(&chain->filter_chain_lock);
1947
1948 if (tcf_proto_exists_destroying(chain, tp_new)) {
1949 mutex_unlock(&chain->filter_chain_lock);
1950 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1951 return ERR_PTR(-EAGAIN);
1952 }
1953
1954 tp = tcf_chain_tp_find(chain, &chain_info, protocol, prio, false, NULL);
1955 if (!tp)
1956 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1957 mutex_unlock(&chain->filter_chain_lock);
1958
1959 if (tp) {
1960 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1961 tp_new = tp;
1962 } else if (err) {
1963 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1964 tp_new = ERR_PTR(err);
1965 }
1966
1967 return tp_new;
1968}
1969
1970static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1971 struct tcf_proto *tp, bool rtnl_held,
1972 struct netlink_ext_ack *extack)
1973{
1974 struct tcf_chain_info chain_info;
1975 struct tcf_proto *tp_iter;
1976 struct tcf_proto **pprev;
1977 struct tcf_proto *next;
1978
1979 mutex_lock(&chain->filter_chain_lock);
1980
1981 /* Atomically find and remove tp from chain. */
1982 for (pprev = &chain->filter_chain;
1983 (tp_iter = tcf_chain_dereference(*pprev, chain));
1984 pprev = &tp_iter->next) {
1985 if (tp_iter == tp) {
1986 chain_info.pprev = pprev;
1987 chain_info.next = tp_iter->next;
1988 WARN_ON(tp_iter->deleting);
1989 break;
1990 }
1991 }
1992 /* Verify that tp still exists and no new filters were inserted
1993 * concurrently.
1994 * Mark tp for deletion if it is empty.
1995 */
1996 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1997 mutex_unlock(&chain->filter_chain_lock);
1998 return;
1999 }
2000
2001 tcf_proto_signal_destroying(chain, tp);
2002 next = tcf_chain_dereference(chain_info.next, chain);
2003 if (tp == chain->filter_chain)
2004 tcf_chain0_head_change(chain, next);
2005 RCU_INIT_POINTER(*chain_info.pprev, next);
2006 mutex_unlock(&chain->filter_chain_lock);
2007
2008 tcf_proto_put(tp, rtnl_held, extack);
2009}
2010
2011static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
2012 struct tcf_chain_info *chain_info,
2013 u32 protocol, u32 prio,
2014 bool prio_allocate,
2015 struct netlink_ext_ack *extack)
2016{
2017 struct tcf_proto **pprev;
2018 struct tcf_proto *tp;
2019
2020 /* Check the chain for existence of proto-tcf with this priority */
2021 for (pprev = &chain->filter_chain;
2022 (tp = tcf_chain_dereference(*pprev, chain));
2023 pprev = &tp->next) {
2024 if (tp->prio >= prio) {
2025 if (tp->prio == prio) {
2026 if (prio_allocate) {
2027 NL_SET_ERR_MSG(extack, "Lowest ID from auto-alloc range already in use");
2028 return ERR_PTR(-ENOSPC);
2029 }
2030 if (tp->protocol != protocol && protocol) {
2031 NL_SET_ERR_MSG(extack, "Protocol mismatch for filter with specified priority");
2032 return ERR_PTR(-EINVAL);
2033 }
2034 } else {
2035 tp = NULL;
2036 }
2037 break;
2038 }
2039 }
2040 chain_info->pprev = pprev;
2041 if (tp) {
2042 chain_info->next = tp->next;
2043 tcf_proto_get(tp);
2044 } else {
2045 chain_info->next = NULL;
2046 }
2047 return tp;
2048}
2049
2050static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2051 struct tcf_proto *tp, struct tcf_block *block,
2052 struct Qdisc *q, u32 parent, void *fh,
2053 u32 portid, u32 seq, u16 flags, int event,
2054 bool terse_dump, bool rtnl_held,
2055 struct netlink_ext_ack *extack)
2056{
2057 struct tcmsg *tcm;
2058 struct nlmsghdr *nlh;
2059 unsigned char *b = skb_tail_pointer(skb);
2060
2061 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2062 if (!nlh)
2063 goto out_nlmsg_trim;
2064 tcm = nlmsg_data(nlh);
2065 tcm->tcm_family = AF_UNSPEC;
2066 tcm->tcm__pad1 = 0;
2067 tcm->tcm__pad2 = 0;
2068 if (q) {
2069 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2070 tcm->tcm_parent = parent;
2071 } else {
2072 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2073 tcm->tcm_block_index = block->index;
2074 }
2075 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2076 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2077 goto nla_put_failure;
2078 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2079 goto nla_put_failure;
2080 if (!fh) {
2081 tcm->tcm_handle = 0;
2082 } else if (terse_dump) {
2083 if (tp->ops->terse_dump) {
2084 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2085 rtnl_held) < 0)
2086 goto nla_put_failure;
2087 } else {
2088 goto cls_op_not_supp;
2089 }
2090 } else {
2091 if (tp->ops->dump &&
2092 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2093 goto nla_put_failure;
2094 }
2095
2096 if (extack && extack->_msg &&
2097 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2098 goto nla_put_failure;
2099
2100 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2101
2102 return skb->len;
2103
2104out_nlmsg_trim:
2105nla_put_failure:
2106cls_op_not_supp:
2107 nlmsg_trim(skb, b);
2108 return -1;
2109}
2110
2111static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2112 struct nlmsghdr *n, struct tcf_proto *tp,
2113 struct tcf_block *block, struct Qdisc *q,
2114 u32 parent, void *fh, int event, bool unicast,
2115 bool rtnl_held, struct netlink_ext_ack *extack)
2116{
2117 struct sk_buff *skb;
2118 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2119 int err = 0;
2120
2121 if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2122 return 0;
2123
2124 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2125 if (!skb)
2126 return -ENOBUFS;
2127
2128 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2129 n->nlmsg_seq, n->nlmsg_flags, event,
2130 false, rtnl_held, extack) <= 0) {
2131 kfree_skb(skb);
2132 return -EINVAL;
2133 }
2134
2135 if (unicast)
2136 err = rtnl_unicast(skb, net, portid);
2137 else
2138 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2139 n->nlmsg_flags & NLM_F_ECHO);
2140 return err;
2141}
2142
2143static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2144 struct nlmsghdr *n, struct tcf_proto *tp,
2145 struct tcf_block *block, struct Qdisc *q,
2146 u32 parent, void *fh, bool *last, bool rtnl_held,
2147 struct netlink_ext_ack *extack)
2148{
2149 struct sk_buff *skb;
2150 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2151 int err;
2152
2153 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2154 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2155
2156 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2157 if (!skb)
2158 return -ENOBUFS;
2159
2160 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2161 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2162 false, rtnl_held, extack) <= 0) {
2163 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2164 kfree_skb(skb);
2165 return -EINVAL;
2166 }
2167
2168 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2169 if (err) {
2170 kfree_skb(skb);
2171 return err;
2172 }
2173
2174 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2175 n->nlmsg_flags & NLM_F_ECHO);
2176 if (err < 0)
2177 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2178
2179 return err;
2180}
2181
2182static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2183 struct tcf_block *block, struct Qdisc *q,
2184 u32 parent, struct nlmsghdr *n,
2185 struct tcf_chain *chain, int event,
2186 struct netlink_ext_ack *extack)
2187{
2188 struct tcf_proto *tp;
2189
2190 for (tp = tcf_get_next_proto(chain, NULL);
2191 tp; tp = tcf_get_next_proto(chain, tp))
2192 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2193 event, false, true, extack);
2194}
2195
2196static void tfilter_put(struct tcf_proto *tp, void *fh)
2197{
2198 if (tp->ops->put && fh)
2199 tp->ops->put(tp, fh);
2200}
2201
2202static bool is_qdisc_ingress(__u32 classid)
2203{
2204 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2205}
2206
2207static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2208 struct netlink_ext_ack *extack)
2209{
2210 struct net *net = sock_net(skb->sk);
2211 struct nlattr *tca[TCA_MAX + 1];
2212 char name[IFNAMSIZ];
2213 struct tcmsg *t;
2214 u32 protocol;
2215 u32 prio;
2216 bool prio_allocate;
2217 u32 parent;
2218 u32 chain_index;
2219 struct Qdisc *q;
2220 struct tcf_chain_info chain_info;
2221 struct tcf_chain *chain;
2222 struct tcf_block *block;
2223 struct tcf_proto *tp;
2224 unsigned long cl;
2225 void *fh;
2226 int err;
2227 int tp_created;
2228 bool rtnl_held = false;
2229 u32 flags;
2230
2231replay:
2232 tp_created = 0;
2233
2234 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2235 rtm_tca_policy, extack);
2236 if (err < 0)
2237 return err;
2238
2239 t = nlmsg_data(n);
2240 protocol = TC_H_MIN(t->tcm_info);
2241 prio = TC_H_MAJ(t->tcm_info);
2242 prio_allocate = false;
2243 parent = t->tcm_parent;
2244 tp = NULL;
2245 cl = 0;
2246 block = NULL;
2247 q = NULL;
2248 chain = NULL;
2249 flags = 0;
2250
2251 if (prio == 0) {
2252 /* If no priority is provided by the user,
2253 * we allocate one.
2254 */
2255 if (n->nlmsg_flags & NLM_F_CREATE) {
2256 prio = TC_H_MAKE(0x80000000U, 0U);
2257 prio_allocate = true;
2258 } else {
2259 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2260 return -ENOENT;
2261 }
2262 }
2263
2264 /* Find head of filter chain. */
2265
2266 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2267 if (err)
2268 return err;
2269
2270 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2271 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2272 err = -EINVAL;
2273 goto errout;
2274 }
2275
2276 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2277 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2278 * type is not specified, classifier is not unlocked.
2279 */
2280 if (rtnl_held ||
2281 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2282 !tcf_proto_is_unlocked(name)) {
2283 rtnl_held = true;
2284 rtnl_lock();
2285 }
2286
2287 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2288 if (err)
2289 goto errout;
2290
2291 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2292 extack);
2293 if (IS_ERR(block)) {
2294 err = PTR_ERR(block);
2295 goto errout;
2296 }
2297 block->classid = parent;
2298
2299 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2300 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2301 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2302 err = -EINVAL;
2303 goto errout;
2304 }
2305 chain = tcf_chain_get(block, chain_index, true);
2306 if (!chain) {
2307 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2308 err = -ENOMEM;
2309 goto errout;
2310 }
2311
2312 mutex_lock(&chain->filter_chain_lock);
2313 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2314 prio, prio_allocate, extack);
2315 if (IS_ERR(tp)) {
2316 err = PTR_ERR(tp);
2317 goto errout_locked;
2318 }
2319
2320 if (tp == NULL) {
2321 struct tcf_proto *tp_new = NULL;
2322
2323 if (chain->flushing) {
2324 err = -EAGAIN;
2325 goto errout_locked;
2326 }
2327
2328 /* Proto-tcf does not exist, create new one */
2329
2330 if (tca[TCA_KIND] == NULL || !protocol) {
2331 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2332 err = -EINVAL;
2333 goto errout_locked;
2334 }
2335
2336 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2337 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2338 err = -ENOENT;
2339 goto errout_locked;
2340 }
2341
2342 if (prio_allocate)
2343 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2344 &chain_info));
2345
2346 mutex_unlock(&chain->filter_chain_lock);
2347 tp_new = tcf_proto_create(name, protocol, prio, chain,
2348 rtnl_held, extack);
2349 if (IS_ERR(tp_new)) {
2350 err = PTR_ERR(tp_new);
2351 goto errout_tp;
2352 }
2353
2354 tp_created = 1;
2355 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2356 rtnl_held);
2357 if (IS_ERR(tp)) {
2358 err = PTR_ERR(tp);
2359 goto errout_tp;
2360 }
2361 } else {
2362 mutex_unlock(&chain->filter_chain_lock);
2363 }
2364
2365 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2366 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2367 err = -EINVAL;
2368 goto errout;
2369 }
2370
2371 fh = tp->ops->get(tp, t->tcm_handle);
2372
2373 if (!fh) {
2374 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2375 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2376 err = -ENOENT;
2377 goto errout;
2378 }
2379 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2380 tfilter_put(tp, fh);
2381 NL_SET_ERR_MSG(extack, "Filter already exists");
2382 err = -EEXIST;
2383 goto errout;
2384 }
2385
2386 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2387 tfilter_put(tp, fh);
2388 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2389 err = -EINVAL;
2390 goto errout;
2391 }
2392
2393 if (!(n->nlmsg_flags & NLM_F_CREATE))
2394 flags |= TCA_ACT_FLAGS_REPLACE;
2395 if (!rtnl_held)
2396 flags |= TCA_ACT_FLAGS_NO_RTNL;
2397 if (is_qdisc_ingress(parent))
2398 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2399 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2400 flags, extack);
2401 if (err == 0) {
2402 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2403 RTM_NEWTFILTER, false, rtnl_held, extack);
2404 tfilter_put(tp, fh);
2405 tcf_proto_count_usesw(tp, true);
2406 /* q pointer is NULL for shared blocks */
2407 if (q)
2408 q->flags &= ~TCQ_F_CAN_BYPASS;
2409 }
2410
2411errout:
2412 if (err && tp_created)
2413 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2414errout_tp:
2415 if (chain) {
2416 if (tp && !IS_ERR(tp))
2417 tcf_proto_put(tp, rtnl_held, NULL);
2418 if (!tp_created)
2419 tcf_chain_put(chain);
2420 }
2421 tcf_block_release(q, block, rtnl_held);
2422
2423 if (rtnl_held)
2424 rtnl_unlock();
2425
2426 if (err == -EAGAIN) {
2427 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2428 * of target chain.
2429 */
2430 rtnl_held = true;
2431 /* Replay the request. */
2432 goto replay;
2433 }
2434 return err;
2435
2436errout_locked:
2437 mutex_unlock(&chain->filter_chain_lock);
2438 goto errout;
2439}
2440
2441static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2442 struct netlink_ext_ack *extack)
2443{
2444 struct net *net = sock_net(skb->sk);
2445 struct nlattr *tca[TCA_MAX + 1];
2446 char name[IFNAMSIZ];
2447 struct tcmsg *t;
2448 u32 protocol;
2449 u32 prio;
2450 u32 parent;
2451 u32 chain_index;
2452 struct Qdisc *q = NULL;
2453 struct tcf_chain_info chain_info;
2454 struct tcf_chain *chain = NULL;
2455 struct tcf_block *block = NULL;
2456 struct tcf_proto *tp = NULL;
2457 unsigned long cl = 0;
2458 void *fh = NULL;
2459 int err;
2460 bool rtnl_held = false;
2461
2462 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2463 rtm_tca_policy, extack);
2464 if (err < 0)
2465 return err;
2466
2467 t = nlmsg_data(n);
2468 protocol = TC_H_MIN(t->tcm_info);
2469 prio = TC_H_MAJ(t->tcm_info);
2470 parent = t->tcm_parent;
2471
2472 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2473 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2474 return -ENOENT;
2475 }
2476
2477 /* Find head of filter chain. */
2478
2479 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2480 if (err)
2481 return err;
2482
2483 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2484 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2485 err = -EINVAL;
2486 goto errout;
2487 }
2488 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2489 * found), qdisc is not unlocked, classifier type is not specified,
2490 * classifier is not unlocked.
2491 */
2492 if (!prio ||
2493 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2494 !tcf_proto_is_unlocked(name)) {
2495 rtnl_held = true;
2496 rtnl_lock();
2497 }
2498
2499 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2500 if (err)
2501 goto errout;
2502
2503 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2504 extack);
2505 if (IS_ERR(block)) {
2506 err = PTR_ERR(block);
2507 goto errout;
2508 }
2509
2510 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2511 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2512 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2513 err = -EINVAL;
2514 goto errout;
2515 }
2516 chain = tcf_chain_get(block, chain_index, false);
2517 if (!chain) {
2518 /* User requested flush on non-existent chain. Nothing to do,
2519 * so just return success.
2520 */
2521 if (prio == 0) {
2522 err = 0;
2523 goto errout;
2524 }
2525 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2526 err = -ENOENT;
2527 goto errout;
2528 }
2529
2530 if (prio == 0) {
2531 tfilter_notify_chain(net, skb, block, q, parent, n,
2532 chain, RTM_DELTFILTER, extack);
2533 tcf_chain_flush(chain, rtnl_held);
2534 err = 0;
2535 goto errout;
2536 }
2537
2538 mutex_lock(&chain->filter_chain_lock);
2539 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2540 prio, false, extack);
2541 if (!tp) {
2542 err = -ENOENT;
2543 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2544 goto errout_locked;
2545 } else if (IS_ERR(tp)) {
2546 err = PTR_ERR(tp);
2547 goto errout_locked;
2548 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2549 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2550 err = -EINVAL;
2551 goto errout_locked;
2552 } else if (t->tcm_handle == 0) {
2553 tcf_proto_signal_destroying(chain, tp);
2554 tcf_chain_tp_remove(chain, &chain_info, tp);
2555 mutex_unlock(&chain->filter_chain_lock);
2556
2557 tcf_proto_put(tp, rtnl_held, NULL);
2558 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2559 RTM_DELTFILTER, false, rtnl_held, extack);
2560 err = 0;
2561 goto errout;
2562 }
2563 mutex_unlock(&chain->filter_chain_lock);
2564
2565 fh = tp->ops->get(tp, t->tcm_handle);
2566
2567 if (!fh) {
2568 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2569 err = -ENOENT;
2570 } else {
2571 bool last;
2572
2573 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2574 &last, rtnl_held, extack);
2575
2576 if (err)
2577 goto errout;
2578 if (last)
2579 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2580 }
2581
2582errout:
2583 if (chain) {
2584 if (tp && !IS_ERR(tp))
2585 tcf_proto_put(tp, rtnl_held, NULL);
2586 tcf_chain_put(chain);
2587 }
2588 tcf_block_release(q, block, rtnl_held);
2589
2590 if (rtnl_held)
2591 rtnl_unlock();
2592
2593 return err;
2594
2595errout_locked:
2596 mutex_unlock(&chain->filter_chain_lock);
2597 goto errout;
2598}
2599
2600static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2601 struct netlink_ext_ack *extack)
2602{
2603 struct net *net = sock_net(skb->sk);
2604 struct nlattr *tca[TCA_MAX + 1];
2605 char name[IFNAMSIZ];
2606 struct tcmsg *t;
2607 u32 protocol;
2608 u32 prio;
2609 u32 parent;
2610 u32 chain_index;
2611 struct Qdisc *q = NULL;
2612 struct tcf_chain_info chain_info;
2613 struct tcf_chain *chain = NULL;
2614 struct tcf_block *block = NULL;
2615 struct tcf_proto *tp = NULL;
2616 unsigned long cl = 0;
2617 void *fh = NULL;
2618 int err;
2619 bool rtnl_held = false;
2620
2621 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2622 rtm_tca_policy, extack);
2623 if (err < 0)
2624 return err;
2625
2626 t = nlmsg_data(n);
2627 protocol = TC_H_MIN(t->tcm_info);
2628 prio = TC_H_MAJ(t->tcm_info);
2629 parent = t->tcm_parent;
2630
2631 if (prio == 0) {
2632 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2633 return -ENOENT;
2634 }
2635
2636 /* Find head of filter chain. */
2637
2638 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2639 if (err)
2640 return err;
2641
2642 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2643 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2644 err = -EINVAL;
2645 goto errout;
2646 }
2647 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2648 * unlocked, classifier type is not specified, classifier is not
2649 * unlocked.
2650 */
2651 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2652 !tcf_proto_is_unlocked(name)) {
2653 rtnl_held = true;
2654 rtnl_lock();
2655 }
2656
2657 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2658 if (err)
2659 goto errout;
2660
2661 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2662 extack);
2663 if (IS_ERR(block)) {
2664 err = PTR_ERR(block);
2665 goto errout;
2666 }
2667
2668 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2669 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2670 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2671 err = -EINVAL;
2672 goto errout;
2673 }
2674 chain = tcf_chain_get(block, chain_index, false);
2675 if (!chain) {
2676 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2677 err = -EINVAL;
2678 goto errout;
2679 }
2680
2681 mutex_lock(&chain->filter_chain_lock);
2682 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2683 prio, false, extack);
2684 mutex_unlock(&chain->filter_chain_lock);
2685 if (!tp) {
2686 err = -ENOENT;
2687 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2688 goto errout;
2689 } else if (IS_ERR(tp)) {
2690 err = PTR_ERR(tp);
2691 goto errout;
2692 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2693 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2694 err = -EINVAL;
2695 goto errout;
2696 }
2697
2698 fh = tp->ops->get(tp, t->tcm_handle);
2699
2700 if (!fh) {
2701 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2702 err = -ENOENT;
2703 } else {
2704 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2705 fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2706 if (err < 0)
2707 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2708 }
2709
2710 tfilter_put(tp, fh);
2711errout:
2712 if (chain) {
2713 if (tp && !IS_ERR(tp))
2714 tcf_proto_put(tp, rtnl_held, NULL);
2715 tcf_chain_put(chain);
2716 }
2717 tcf_block_release(q, block, rtnl_held);
2718
2719 if (rtnl_held)
2720 rtnl_unlock();
2721
2722 return err;
2723}
2724
2725struct tcf_dump_args {
2726 struct tcf_walker w;
2727 struct sk_buff *skb;
2728 struct netlink_callback *cb;
2729 struct tcf_block *block;
2730 struct Qdisc *q;
2731 u32 parent;
2732 bool terse_dump;
2733};
2734
2735static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2736{
2737 struct tcf_dump_args *a = (void *)arg;
2738 struct net *net = sock_net(a->skb->sk);
2739
2740 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2741 n, NETLINK_CB(a->cb->skb).portid,
2742 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2743 RTM_NEWTFILTER, a->terse_dump, true, NULL);
2744}
2745
2746static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2747 struct sk_buff *skb, struct netlink_callback *cb,
2748 long index_start, long *p_index, bool terse)
2749{
2750 struct net *net = sock_net(skb->sk);
2751 struct tcf_block *block = chain->block;
2752 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2753 struct tcf_proto *tp, *tp_prev;
2754 struct tcf_dump_args arg;
2755
2756 for (tp = __tcf_get_next_proto(chain, NULL);
2757 tp;
2758 tp_prev = tp,
2759 tp = __tcf_get_next_proto(chain, tp),
2760 tcf_proto_put(tp_prev, true, NULL),
2761 (*p_index)++) {
2762 if (*p_index < index_start)
2763 continue;
2764 if (TC_H_MAJ(tcm->tcm_info) &&
2765 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2766 continue;
2767 if (TC_H_MIN(tcm->tcm_info) &&
2768 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2769 continue;
2770 if (*p_index > index_start)
2771 memset(&cb->args[1], 0,
2772 sizeof(cb->args) - sizeof(cb->args[0]));
2773 if (cb->args[1] == 0) {
2774 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2775 NETLINK_CB(cb->skb).portid,
2776 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2777 RTM_NEWTFILTER, false, true, NULL) <= 0)
2778 goto errout;
2779 cb->args[1] = 1;
2780 }
2781 if (!tp->ops->walk)
2782 continue;
2783 arg.w.fn = tcf_node_dump;
2784 arg.skb = skb;
2785 arg.cb = cb;
2786 arg.block = block;
2787 arg.q = q;
2788 arg.parent = parent;
2789 arg.w.stop = 0;
2790 arg.w.skip = cb->args[1] - 1;
2791 arg.w.count = 0;
2792 arg.w.cookie = cb->args[2];
2793 arg.terse_dump = terse;
2794 tp->ops->walk(tp, &arg.w, true);
2795 cb->args[2] = arg.w.cookie;
2796 cb->args[1] = arg.w.count + 1;
2797 if (arg.w.stop)
2798 goto errout;
2799 }
2800 return true;
2801
2802errout:
2803 tcf_proto_put(tp, true, NULL);
2804 return false;
2805}
2806
2807static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2808 [TCA_CHAIN] = { .type = NLA_U32 },
2809 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2810};
2811
2812/* called with RTNL */
2813static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2814{
2815 struct tcf_chain *chain, *chain_prev;
2816 struct net *net = sock_net(skb->sk);
2817 struct nlattr *tca[TCA_MAX + 1];
2818 struct Qdisc *q = NULL;
2819 struct tcf_block *block;
2820 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2821 bool terse_dump = false;
2822 long index_start;
2823 long index;
2824 u32 parent;
2825 int err;
2826
2827 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2828 return skb->len;
2829
2830 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2831 tcf_tfilter_dump_policy, cb->extack);
2832 if (err)
2833 return err;
2834
2835 if (tca[TCA_DUMP_FLAGS]) {
2836 struct nla_bitfield32 flags =
2837 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2838
2839 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2840 }
2841
2842 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2843 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2844 if (!block)
2845 goto out;
2846 /* If we work with block index, q is NULL and parent value
2847 * will never be used in the following code. The check
2848 * in tcf_fill_node prevents it. However, compiler does not
2849 * see that far, so set parent to zero to silence the warning
2850 * about parent being uninitialized.
2851 */
2852 parent = 0;
2853 } else {
2854 const struct Qdisc_class_ops *cops;
2855 struct net_device *dev;
2856 unsigned long cl = 0;
2857
2858 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2859 if (!dev)
2860 return skb->len;
2861
2862 parent = tcm->tcm_parent;
2863 if (!parent)
2864 q = rtnl_dereference(dev->qdisc);
2865 else
2866 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2867 if (!q)
2868 goto out;
2869 cops = q->ops->cl_ops;
2870 if (!cops)
2871 goto out;
2872 if (!cops->tcf_block)
2873 goto out;
2874 if (TC_H_MIN(tcm->tcm_parent)) {
2875 cl = cops->find(q, tcm->tcm_parent);
2876 if (cl == 0)
2877 goto out;
2878 }
2879 block = cops->tcf_block(q, cl, NULL);
2880 if (!block)
2881 goto out;
2882 parent = block->classid;
2883 if (tcf_block_shared(block))
2884 q = NULL;
2885 }
2886
2887 index_start = cb->args[0];
2888 index = 0;
2889
2890 for (chain = __tcf_get_next_chain(block, NULL);
2891 chain;
2892 chain_prev = chain,
2893 chain = __tcf_get_next_chain(block, chain),
2894 tcf_chain_put(chain_prev)) {
2895 if (tca[TCA_CHAIN] &&
2896 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2897 continue;
2898 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2899 index_start, &index, terse_dump)) {
2900 tcf_chain_put(chain);
2901 err = -EMSGSIZE;
2902 break;
2903 }
2904 }
2905
2906 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2907 tcf_block_refcnt_put(block, true);
2908 cb->args[0] = index;
2909
2910out:
2911 /* If we did no progress, the error (EMSGSIZE) is real */
2912 if (skb->len == 0 && err)
2913 return err;
2914 return skb->len;
2915}
2916
2917static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2918 void *tmplt_priv, u32 chain_index,
2919 struct net *net, struct sk_buff *skb,
2920 struct tcf_block *block,
2921 u32 portid, u32 seq, u16 flags, int event,
2922 struct netlink_ext_ack *extack)
2923{
2924 unsigned char *b = skb_tail_pointer(skb);
2925 const struct tcf_proto_ops *ops;
2926 struct nlmsghdr *nlh;
2927 struct tcmsg *tcm;
2928 void *priv;
2929
2930 ops = tmplt_ops;
2931 priv = tmplt_priv;
2932
2933 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2934 if (!nlh)
2935 goto out_nlmsg_trim;
2936 tcm = nlmsg_data(nlh);
2937 tcm->tcm_family = AF_UNSPEC;
2938 tcm->tcm__pad1 = 0;
2939 tcm->tcm__pad2 = 0;
2940 tcm->tcm_handle = 0;
2941 if (block->q) {
2942 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2943 tcm->tcm_parent = block->q->handle;
2944 } else {
2945 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2946 tcm->tcm_block_index = block->index;
2947 }
2948
2949 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2950 goto nla_put_failure;
2951
2952 if (ops) {
2953 if (nla_put_string(skb, TCA_KIND, ops->kind))
2954 goto nla_put_failure;
2955 if (ops->tmplt_dump(skb, net, priv) < 0)
2956 goto nla_put_failure;
2957 }
2958
2959 if (extack && extack->_msg &&
2960 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2961 goto out_nlmsg_trim;
2962
2963 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2964
2965 return skb->len;
2966
2967out_nlmsg_trim:
2968nla_put_failure:
2969 nlmsg_trim(skb, b);
2970 return -EMSGSIZE;
2971}
2972
2973static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2974 u32 seq, u16 flags, int event, bool unicast,
2975 struct netlink_ext_ack *extack)
2976{
2977 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2978 struct tcf_block *block = chain->block;
2979 struct net *net = block->net;
2980 struct sk_buff *skb;
2981 int err = 0;
2982
2983 if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2984 return 0;
2985
2986 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2987 if (!skb)
2988 return -ENOBUFS;
2989
2990 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2991 chain->index, net, skb, block, portid,
2992 seq, flags, event, extack) <= 0) {
2993 kfree_skb(skb);
2994 return -EINVAL;
2995 }
2996
2997 if (unicast)
2998 err = rtnl_unicast(skb, net, portid);
2999 else
3000 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
3001 flags & NLM_F_ECHO);
3002
3003 return err;
3004}
3005
3006static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
3007 void *tmplt_priv, u32 chain_index,
3008 struct tcf_block *block, struct sk_buff *oskb,
3009 u32 seq, u16 flags)
3010{
3011 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3012 struct net *net = block->net;
3013 struct sk_buff *skb;
3014
3015 if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
3016 return 0;
3017
3018 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3019 if (!skb)
3020 return -ENOBUFS;
3021
3022 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
3023 block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
3024 kfree_skb(skb);
3025 return -EINVAL;
3026 }
3027
3028 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
3029}
3030
3031static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
3032 struct nlattr **tca,
3033 struct netlink_ext_ack *extack)
3034{
3035 const struct tcf_proto_ops *ops;
3036 char name[IFNAMSIZ];
3037 void *tmplt_priv;
3038
3039 /* If kind is not set, user did not specify template. */
3040 if (!tca[TCA_KIND])
3041 return 0;
3042
3043 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3044 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3045 return -EINVAL;
3046 }
3047
3048 ops = tcf_proto_lookup_ops(name, true, extack);
3049 if (IS_ERR(ops))
3050 return PTR_ERR(ops);
3051 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3052 !ops->tmplt_reoffload) {
3053 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3054 module_put(ops->owner);
3055 return -EOPNOTSUPP;
3056 }
3057
3058 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3059 if (IS_ERR(tmplt_priv)) {
3060 module_put(ops->owner);
3061 return PTR_ERR(tmplt_priv);
3062 }
3063 chain->tmplt_ops = ops;
3064 chain->tmplt_priv = tmplt_priv;
3065 return 0;
3066}
3067
3068static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3069 void *tmplt_priv)
3070{
3071 /* If template ops are set, no work to do for us. */
3072 if (!tmplt_ops)
3073 return;
3074
3075 tmplt_ops->tmplt_destroy(tmplt_priv);
3076 module_put(tmplt_ops->owner);
3077}
3078
3079/* Add/delete/get a chain */
3080
3081static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3082 struct netlink_ext_ack *extack)
3083{
3084 struct net *net = sock_net(skb->sk);
3085 struct nlattr *tca[TCA_MAX + 1];
3086 struct tcmsg *t;
3087 u32 parent;
3088 u32 chain_index;
3089 struct Qdisc *q;
3090 struct tcf_chain *chain;
3091 struct tcf_block *block;
3092 unsigned long cl;
3093 int err;
3094
3095replay:
3096 q = NULL;
3097 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3098 rtm_tca_policy, extack);
3099 if (err < 0)
3100 return err;
3101
3102 t = nlmsg_data(n);
3103 parent = t->tcm_parent;
3104 cl = 0;
3105
3106 block = tcf_block_find(net, &q, &parent, &cl,
3107 t->tcm_ifindex, t->tcm_block_index, extack);
3108 if (IS_ERR(block))
3109 return PTR_ERR(block);
3110
3111 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
3112 if (chain_index > TC_ACT_EXT_VAL_MASK) {
3113 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3114 err = -EINVAL;
3115 goto errout_block;
3116 }
3117
3118 mutex_lock(&block->lock);
3119 chain = tcf_chain_lookup(block, chain_index);
3120 if (n->nlmsg_type == RTM_NEWCHAIN) {
3121 if (chain) {
3122 if (tcf_chain_held_by_acts_only(chain)) {
3123 /* The chain exists only because there is
3124 * some action referencing it.
3125 */
3126 tcf_chain_hold(chain);
3127 } else {
3128 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3129 err = -EEXIST;
3130 goto errout_block_locked;
3131 }
3132 } else {
3133 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3134 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3135 err = -ENOENT;
3136 goto errout_block_locked;
3137 }
3138 chain = tcf_chain_create(block, chain_index);
3139 if (!chain) {
3140 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3141 err = -ENOMEM;
3142 goto errout_block_locked;
3143 }
3144 }
3145 } else {
3146 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3147 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3148 err = -EINVAL;
3149 goto errout_block_locked;
3150 }
3151 tcf_chain_hold(chain);
3152 }
3153
3154 if (n->nlmsg_type == RTM_NEWCHAIN) {
3155 /* Modifying chain requires holding parent block lock. In case
3156 * the chain was successfully added, take a reference to the
3157 * chain. This ensures that an empty chain does not disappear at
3158 * the end of this function.
3159 */
3160 tcf_chain_hold(chain);
3161 chain->explicitly_created = true;
3162 }
3163 mutex_unlock(&block->lock);
3164
3165 switch (n->nlmsg_type) {
3166 case RTM_NEWCHAIN:
3167 err = tc_chain_tmplt_add(chain, net, tca, extack);
3168 if (err) {
3169 tcf_chain_put_explicitly_created(chain);
3170 goto errout;
3171 }
3172
3173 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3174 RTM_NEWCHAIN, false, extack);
3175 break;
3176 case RTM_DELCHAIN:
3177 tfilter_notify_chain(net, skb, block, q, parent, n,
3178 chain, RTM_DELTFILTER, extack);
3179 /* Flush the chain first as the user requested chain removal. */
3180 tcf_chain_flush(chain, true);
3181 /* In case the chain was successfully deleted, put a reference
3182 * to the chain previously taken during addition.
3183 */
3184 tcf_chain_put_explicitly_created(chain);
3185 break;
3186 case RTM_GETCHAIN:
3187 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3188 n->nlmsg_flags, n->nlmsg_type, true, extack);
3189 if (err < 0)
3190 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3191 break;
3192 default:
3193 err = -EOPNOTSUPP;
3194 NL_SET_ERR_MSG(extack, "Unsupported message type");
3195 goto errout;
3196 }
3197
3198errout:
3199 tcf_chain_put(chain);
3200errout_block:
3201 tcf_block_release(q, block, true);
3202 if (err == -EAGAIN)
3203 /* Replay the request. */
3204 goto replay;
3205 return err;
3206
3207errout_block_locked:
3208 mutex_unlock(&block->lock);
3209 goto errout_block;
3210}
3211
3212/* called with RTNL */
3213static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3214{
3215 struct net *net = sock_net(skb->sk);
3216 struct nlattr *tca[TCA_MAX + 1];
3217 struct Qdisc *q = NULL;
3218 struct tcf_block *block;
3219 struct tcmsg *tcm = nlmsg_data(cb->nlh);
3220 struct tcf_chain *chain;
3221 long index_start;
3222 long index;
3223 int err;
3224
3225 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3226 return skb->len;
3227
3228 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3229 rtm_tca_policy, cb->extack);
3230 if (err)
3231 return err;
3232
3233 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3234 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3235 if (!block)
3236 goto out;
3237 } else {
3238 const struct Qdisc_class_ops *cops;
3239 struct net_device *dev;
3240 unsigned long cl = 0;
3241
3242 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3243 if (!dev)
3244 return skb->len;
3245
3246 if (!tcm->tcm_parent)
3247 q = rtnl_dereference(dev->qdisc);
3248 else
3249 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3250
3251 if (!q)
3252 goto out;
3253 cops = q->ops->cl_ops;
3254 if (!cops)
3255 goto out;
3256 if (!cops->tcf_block)
3257 goto out;
3258 if (TC_H_MIN(tcm->tcm_parent)) {
3259 cl = cops->find(q, tcm->tcm_parent);
3260 if (cl == 0)
3261 goto out;
3262 }
3263 block = cops->tcf_block(q, cl, NULL);
3264 if (!block)
3265 goto out;
3266 if (tcf_block_shared(block))
3267 q = NULL;
3268 }
3269
3270 index_start = cb->args[0];
3271 index = 0;
3272
3273 mutex_lock(&block->lock);
3274 list_for_each_entry(chain, &block->chain_list, list) {
3275 if ((tca[TCA_CHAIN] &&
3276 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3277 continue;
3278 if (index < index_start) {
3279 index++;
3280 continue;
3281 }
3282 if (tcf_chain_held_by_acts_only(chain))
3283 continue;
3284 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3285 chain->index, net, skb, block,
3286 NETLINK_CB(cb->skb).portid,
3287 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3288 RTM_NEWCHAIN, NULL);
3289 if (err <= 0)
3290 break;
3291 index++;
3292 }
3293 mutex_unlock(&block->lock);
3294
3295 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3296 tcf_block_refcnt_put(block, true);
3297 cb->args[0] = index;
3298
3299out:
3300 /* If we did no progress, the error (EMSGSIZE) is real */
3301 if (skb->len == 0 && err)
3302 return err;
3303 return skb->len;
3304}
3305
3306int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3307 int police, struct tcf_proto *tp, u32 handle,
3308 bool use_action_miss)
3309{
3310 int err = 0;
3311
3312#ifdef CONFIG_NET_CLS_ACT
3313 exts->type = 0;
3314 exts->nr_actions = 0;
3315 exts->miss_cookie_node = NULL;
3316 /* Note: we do not own yet a reference on net.
3317 * This reference might be taken later from tcf_exts_get_net().
3318 */
3319 exts->net = net;
3320 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3321 GFP_KERNEL);
3322 if (!exts->actions)
3323 return -ENOMEM;
3324#endif
3325
3326 exts->action = action;
3327 exts->police = police;
3328
3329 if (!use_action_miss)
3330 return 0;
3331
3332 err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3333 if (err)
3334 goto err_miss_alloc;
3335
3336 return 0;
3337
3338err_miss_alloc:
3339 tcf_exts_destroy(exts);
3340#ifdef CONFIG_NET_CLS_ACT
3341 exts->actions = NULL;
3342#endif
3343 return err;
3344}
3345EXPORT_SYMBOL(tcf_exts_init_ex);
3346
3347void tcf_exts_destroy(struct tcf_exts *exts)
3348{
3349 tcf_exts_miss_cookie_base_destroy(exts);
3350
3351#ifdef CONFIG_NET_CLS_ACT
3352 if (exts->actions) {
3353 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3354 kfree(exts->actions);
3355 }
3356 exts->nr_actions = 0;
3357#endif
3358}
3359EXPORT_SYMBOL(tcf_exts_destroy);
3360
3361int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3362 struct nlattr *rate_tlv, struct tcf_exts *exts,
3363 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3364{
3365#ifdef CONFIG_NET_CLS_ACT
3366 {
3367 int init_res[TCA_ACT_MAX_PRIO] = {};
3368 struct tc_action *act;
3369 size_t attr_size = 0;
3370
3371 if (exts->police && tb[exts->police]) {
3372 struct tc_action_ops *a_o;
3373
3374 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3375 a_o = tc_action_load_ops(tb[exts->police], flags,
3376 extack);
3377 if (IS_ERR(a_o))
3378 return PTR_ERR(a_o);
3379 act = tcf_action_init_1(net, tp, tb[exts->police],
3380 rate_tlv, a_o, init_res, flags,
3381 extack);
3382 module_put(a_o->owner);
3383 if (IS_ERR(act))
3384 return PTR_ERR(act);
3385
3386 act->type = exts->type = TCA_OLD_COMPAT;
3387 exts->actions[0] = act;
3388 exts->nr_actions = 1;
3389 tcf_idr_insert_many(exts->actions, init_res);
3390 } else if (exts->action && tb[exts->action]) {
3391 int err;
3392
3393 flags |= TCA_ACT_FLAGS_BIND;
3394 err = tcf_action_init(net, tp, tb[exts->action],
3395 rate_tlv, exts->actions, init_res,
3396 &attr_size, flags, fl_flags,
3397 extack);
3398 if (err < 0)
3399 return err;
3400 exts->nr_actions = err;
3401 }
3402 }
3403#else
3404 if ((exts->action && tb[exts->action]) ||
3405 (exts->police && tb[exts->police])) {
3406 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3407 return -EOPNOTSUPP;
3408 }
3409#endif
3410
3411 return 0;
3412}
3413EXPORT_SYMBOL(tcf_exts_validate_ex);
3414
3415int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3416 struct nlattr *rate_tlv, struct tcf_exts *exts,
3417 u32 flags, struct netlink_ext_ack *extack)
3418{
3419 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3420 flags, 0, extack);
3421}
3422EXPORT_SYMBOL(tcf_exts_validate);
3423
3424void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3425{
3426#ifdef CONFIG_NET_CLS_ACT
3427 struct tcf_exts old = *dst;
3428
3429 *dst = *src;
3430 tcf_exts_destroy(&old);
3431#endif
3432}
3433EXPORT_SYMBOL(tcf_exts_change);
3434
3435#ifdef CONFIG_NET_CLS_ACT
3436static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3437{
3438 if (exts->nr_actions == 0)
3439 return NULL;
3440 else
3441 return exts->actions[0];
3442}
3443#endif
3444
3445int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3446{
3447#ifdef CONFIG_NET_CLS_ACT
3448 struct nlattr *nest;
3449
3450 if (exts->action && tcf_exts_has_actions(exts)) {
3451 /*
3452 * again for backward compatible mode - we want
3453 * to work with both old and new modes of entering
3454 * tc data even if iproute2 was newer - jhs
3455 */
3456 if (exts->type != TCA_OLD_COMPAT) {
3457 nest = nla_nest_start_noflag(skb, exts->action);
3458 if (nest == NULL)
3459 goto nla_put_failure;
3460
3461 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3462 < 0)
3463 goto nla_put_failure;
3464 nla_nest_end(skb, nest);
3465 } else if (exts->police) {
3466 struct tc_action *act = tcf_exts_first_act(exts);
3467 nest = nla_nest_start_noflag(skb, exts->police);
3468 if (nest == NULL || !act)
3469 goto nla_put_failure;
3470 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3471 goto nla_put_failure;
3472 nla_nest_end(skb, nest);
3473 }
3474 }
3475 return 0;
3476
3477nla_put_failure:
3478 nla_nest_cancel(skb, nest);
3479 return -1;
3480#else
3481 return 0;
3482#endif
3483}
3484EXPORT_SYMBOL(tcf_exts_dump);
3485
3486int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3487{
3488#ifdef CONFIG_NET_CLS_ACT
3489 struct nlattr *nest;
3490
3491 if (!exts->action || !tcf_exts_has_actions(exts))
3492 return 0;
3493
3494 nest = nla_nest_start_noflag(skb, exts->action);
3495 if (!nest)
3496 goto nla_put_failure;
3497
3498 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3499 goto nla_put_failure;
3500 nla_nest_end(skb, nest);
3501 return 0;
3502
3503nla_put_failure:
3504 nla_nest_cancel(skb, nest);
3505 return -1;
3506#else
3507 return 0;
3508#endif
3509}
3510EXPORT_SYMBOL(tcf_exts_terse_dump);
3511
3512int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3513{
3514#ifdef CONFIG_NET_CLS_ACT
3515 struct tc_action *a = tcf_exts_first_act(exts);
3516 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3517 return -1;
3518#endif
3519 return 0;
3520}
3521EXPORT_SYMBOL(tcf_exts_dump_stats);
3522
3523static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3524{
3525 if (*flags & TCA_CLS_FLAGS_IN_HW)
3526 return;
3527 *flags |= TCA_CLS_FLAGS_IN_HW;
3528 atomic_inc(&block->offloadcnt);
3529}
3530
3531static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3532{
3533 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3534 return;
3535 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3536 atomic_dec(&block->offloadcnt);
3537}
3538
3539static void tc_cls_offload_cnt_update(struct tcf_block *block,
3540 struct tcf_proto *tp, u32 *cnt,
3541 u32 *flags, u32 diff, bool add)
3542{
3543 lockdep_assert_held(&block->cb_lock);
3544
3545 spin_lock(&tp->lock);
3546 if (add) {
3547 if (!*cnt)
3548 tcf_block_offload_inc(block, flags);
3549 *cnt += diff;
3550 } else {
3551 *cnt -= diff;
3552 if (!*cnt)
3553 tcf_block_offload_dec(block, flags);
3554 }
3555 spin_unlock(&tp->lock);
3556}
3557
3558static void
3559tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3560 u32 *cnt, u32 *flags)
3561{
3562 lockdep_assert_held(&block->cb_lock);
3563
3564 spin_lock(&tp->lock);
3565 tcf_block_offload_dec(block, flags);
3566 *cnt = 0;
3567 spin_unlock(&tp->lock);
3568}
3569
3570static int
3571__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3572 void *type_data, bool err_stop)
3573{
3574 struct flow_block_cb *block_cb;
3575 int ok_count = 0;
3576 int err;
3577
3578 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3579 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3580 if (err) {
3581 if (err_stop)
3582 return err;
3583 } else {
3584 ok_count++;
3585 }
3586 }
3587 return ok_count;
3588}
3589
3590int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3591 void *type_data, bool err_stop, bool rtnl_held)
3592{
3593 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3594 int ok_count;
3595
3596retry:
3597 if (take_rtnl)
3598 rtnl_lock();
3599 down_read(&block->cb_lock);
3600 /* Need to obtain rtnl lock if block is bound to devs that require it.
3601 * In block bind code cb_lock is obtained while holding rtnl, so we must
3602 * obtain the locks in same order here.
3603 */
3604 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3605 up_read(&block->cb_lock);
3606 take_rtnl = true;
3607 goto retry;
3608 }
3609
3610 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3611
3612 up_read(&block->cb_lock);
3613 if (take_rtnl)
3614 rtnl_unlock();
3615 return ok_count;
3616}
3617EXPORT_SYMBOL(tc_setup_cb_call);
3618
3619/* Non-destructive filter add. If filter that wasn't already in hardware is
3620 * successfully offloaded, increment block offloads counter. On failure,
3621 * previously offloaded filter is considered to be intact and offloads counter
3622 * is not decremented.
3623 */
3624
3625int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3626 enum tc_setup_type type, void *type_data, bool err_stop,
3627 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3628{
3629 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3630 int ok_count;
3631
3632retry:
3633 if (take_rtnl)
3634 rtnl_lock();
3635 down_read(&block->cb_lock);
3636 /* Need to obtain rtnl lock if block is bound to devs that require it.
3637 * In block bind code cb_lock is obtained while holding rtnl, so we must
3638 * obtain the locks in same order here.
3639 */
3640 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3641 up_read(&block->cb_lock);
3642 take_rtnl = true;
3643 goto retry;
3644 }
3645
3646 /* Make sure all netdevs sharing this block are offload-capable. */
3647 if (block->nooffloaddevcnt && err_stop) {
3648 ok_count = -EOPNOTSUPP;
3649 goto err_unlock;
3650 }
3651
3652 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3653 if (ok_count < 0)
3654 goto err_unlock;
3655
3656 if (tp->ops->hw_add)
3657 tp->ops->hw_add(tp, type_data);
3658 if (ok_count > 0)
3659 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3660 ok_count, true);
3661err_unlock:
3662 up_read(&block->cb_lock);
3663 if (take_rtnl)
3664 rtnl_unlock();
3665 return min(ok_count, 0);
3666}
3667EXPORT_SYMBOL(tc_setup_cb_add);
3668
3669/* Destructive filter replace. If filter that wasn't already in hardware is
3670 * successfully offloaded, increment block offload counter. On failure,
3671 * previously offloaded filter is considered to be destroyed and offload counter
3672 * is decremented.
3673 */
3674
3675int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3676 enum tc_setup_type type, void *type_data, bool err_stop,
3677 u32 *old_flags, unsigned int *old_in_hw_count,
3678 u32 *new_flags, unsigned int *new_in_hw_count,
3679 bool rtnl_held)
3680{
3681 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3682 int ok_count;
3683
3684retry:
3685 if (take_rtnl)
3686 rtnl_lock();
3687 down_read(&block->cb_lock);
3688 /* Need to obtain rtnl lock if block is bound to devs that require it.
3689 * In block bind code cb_lock is obtained while holding rtnl, so we must
3690 * obtain the locks in same order here.
3691 */
3692 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3693 up_read(&block->cb_lock);
3694 take_rtnl = true;
3695 goto retry;
3696 }
3697
3698 /* Make sure all netdevs sharing this block are offload-capable. */
3699 if (block->nooffloaddevcnt && err_stop) {
3700 ok_count = -EOPNOTSUPP;
3701 goto err_unlock;
3702 }
3703
3704 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3705 if (tp->ops->hw_del)
3706 tp->ops->hw_del(tp, type_data);
3707
3708 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3709 if (ok_count < 0)
3710 goto err_unlock;
3711
3712 if (tp->ops->hw_add)
3713 tp->ops->hw_add(tp, type_data);
3714 if (ok_count > 0)
3715 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3716 new_flags, ok_count, true);
3717err_unlock:
3718 up_read(&block->cb_lock);
3719 if (take_rtnl)
3720 rtnl_unlock();
3721 return min(ok_count, 0);
3722}
3723EXPORT_SYMBOL(tc_setup_cb_replace);
3724
3725/* Destroy filter and decrement block offload counter, if filter was previously
3726 * offloaded.
3727 */
3728
3729int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3730 enum tc_setup_type type, void *type_data, bool err_stop,
3731 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3732{
3733 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3734 int ok_count;
3735
3736retry:
3737 if (take_rtnl)
3738 rtnl_lock();
3739 down_read(&block->cb_lock);
3740 /* Need to obtain rtnl lock if block is bound to devs that require it.
3741 * In block bind code cb_lock is obtained while holding rtnl, so we must
3742 * obtain the locks in same order here.
3743 */
3744 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3745 up_read(&block->cb_lock);
3746 take_rtnl = true;
3747 goto retry;
3748 }
3749
3750 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3751
3752 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3753 if (tp->ops->hw_del)
3754 tp->ops->hw_del(tp, type_data);
3755
3756 up_read(&block->cb_lock);
3757 if (take_rtnl)
3758 rtnl_unlock();
3759 return min(ok_count, 0);
3760}
3761EXPORT_SYMBOL(tc_setup_cb_destroy);
3762
3763int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3764 bool add, flow_setup_cb_t *cb,
3765 enum tc_setup_type type, void *type_data,
3766 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3767{
3768 int err = cb(type, type_data, cb_priv);
3769
3770 if (err) {
3771 if (add && tc_skip_sw(*flags))
3772 return err;
3773 } else {
3774 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3775 add);
3776 }
3777
3778 return 0;
3779}
3780EXPORT_SYMBOL(tc_setup_cb_reoffload);
3781
3782static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3783 const struct tc_action *act)
3784{
3785 struct tc_cookie *user_cookie;
3786 int err = 0;
3787
3788 rcu_read_lock();
3789 user_cookie = rcu_dereference(act->user_cookie);
3790 if (user_cookie) {
3791 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3792 user_cookie->len,
3793 GFP_ATOMIC);
3794 if (!entry->user_cookie)
3795 err = -ENOMEM;
3796 }
3797 rcu_read_unlock();
3798 return err;
3799}
3800
3801static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3802{
3803 flow_action_cookie_destroy(entry->user_cookie);
3804}
3805
3806void tc_cleanup_offload_action(struct flow_action *flow_action)
3807{
3808 struct flow_action_entry *entry;
3809 int i;
3810
3811 flow_action_for_each(i, entry, flow_action) {
3812 tcf_act_put_user_cookie(entry);
3813 if (entry->destructor)
3814 entry->destructor(entry->destructor_priv);
3815 }
3816}
3817EXPORT_SYMBOL(tc_cleanup_offload_action);
3818
3819static int tc_setup_offload_act(struct tc_action *act,
3820 struct flow_action_entry *entry,
3821 u32 *index_inc,
3822 struct netlink_ext_ack *extack)
3823{
3824#ifdef CONFIG_NET_CLS_ACT
3825 if (act->ops->offload_act_setup) {
3826 return act->ops->offload_act_setup(act, entry, index_inc, true,
3827 extack);
3828 } else {
3829 NL_SET_ERR_MSG(extack, "Action does not support offload");
3830 return -EOPNOTSUPP;
3831 }
3832#else
3833 return 0;
3834#endif
3835}
3836
3837int tc_setup_action(struct flow_action *flow_action,
3838 struct tc_action *actions[],
3839 u32 miss_cookie_base,
3840 struct netlink_ext_ack *extack)
3841{
3842 int i, j, k, index, err = 0;
3843 struct tc_action *act;
3844
3845 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3846 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3847 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3848
3849 if (!actions)
3850 return 0;
3851
3852 j = 0;
3853 tcf_act_for_each_action(i, act, actions) {
3854 struct flow_action_entry *entry;
3855
3856 entry = &flow_action->entries[j];
3857 spin_lock_bh(&act->tcfa_lock);
3858 err = tcf_act_get_user_cookie(entry, act);
3859 if (err)
3860 goto err_out_locked;
3861
3862 index = 0;
3863 err = tc_setup_offload_act(act, entry, &index, extack);
3864 if (err)
3865 goto err_out_locked;
3866
3867 for (k = 0; k < index ; k++) {
3868 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3869 entry[k].hw_index = act->tcfa_index;
3870 entry[k].cookie = (unsigned long)act;
3871 entry[k].miss_cookie =
3872 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3873 }
3874
3875 j += index;
3876
3877 spin_unlock_bh(&act->tcfa_lock);
3878 }
3879
3880err_out:
3881 if (err)
3882 tc_cleanup_offload_action(flow_action);
3883
3884 return err;
3885err_out_locked:
3886 spin_unlock_bh(&act->tcfa_lock);
3887 goto err_out;
3888}
3889
3890int tc_setup_offload_action(struct flow_action *flow_action,
3891 const struct tcf_exts *exts,
3892 struct netlink_ext_ack *extack)
3893{
3894#ifdef CONFIG_NET_CLS_ACT
3895 u32 miss_cookie_base;
3896
3897 if (!exts)
3898 return 0;
3899
3900 miss_cookie_base = exts->miss_cookie_node ?
3901 exts->miss_cookie_node->miss_cookie_base : 0;
3902 return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3903 extack);
3904#else
3905 return 0;
3906#endif
3907}
3908EXPORT_SYMBOL(tc_setup_offload_action);
3909
3910unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3911{
3912 unsigned int num_acts = 0;
3913 struct tc_action *act;
3914 int i;
3915
3916 tcf_exts_for_each_action(i, act, exts) {
3917 if (is_tcf_pedit(act))
3918 num_acts += tcf_pedit_nkeys(act);
3919 else
3920 num_acts++;
3921 }
3922 return num_acts;
3923}
3924EXPORT_SYMBOL(tcf_exts_num_actions);
3925
3926#ifdef CONFIG_NET_CLS_ACT
3927static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3928 u32 *p_block_index,
3929 struct netlink_ext_ack *extack)
3930{
3931 *p_block_index = nla_get_u32(block_index_attr);
3932 if (!*p_block_index) {
3933 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3934 return -EINVAL;
3935 }
3936
3937 return 0;
3938}
3939
3940int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3941 enum flow_block_binder_type binder_type,
3942 struct nlattr *block_index_attr,
3943 struct netlink_ext_ack *extack)
3944{
3945 u32 block_index;
3946 int err;
3947
3948 if (!block_index_attr)
3949 return 0;
3950
3951 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3952 if (err)
3953 return err;
3954
3955 qe->info.binder_type = binder_type;
3956 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3957 qe->info.chain_head_change_priv = &qe->filter_chain;
3958 qe->info.block_index = block_index;
3959
3960 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3961}
3962EXPORT_SYMBOL(tcf_qevent_init);
3963
3964void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3965{
3966 if (qe->info.block_index)
3967 tcf_block_put_ext(qe->block, sch, &qe->info);
3968}
3969EXPORT_SYMBOL(tcf_qevent_destroy);
3970
3971int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3972 struct netlink_ext_ack *extack)
3973{
3974 u32 block_index;
3975 int err;
3976
3977 if (!block_index_attr)
3978 return 0;
3979
3980 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3981 if (err)
3982 return err;
3983
3984 /* Bounce newly-configured block or change in block. */
3985 if (block_index != qe->info.block_index) {
3986 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3987 return -EINVAL;
3988 }
3989
3990 return 0;
3991}
3992EXPORT_SYMBOL(tcf_qevent_validate_change);
3993
3994struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3995 struct sk_buff **to_free, int *ret)
3996{
3997 struct tcf_result cl_res;
3998 struct tcf_proto *fl;
3999
4000 if (!qe->info.block_index)
4001 return skb;
4002
4003 fl = rcu_dereference_bh(qe->filter_chain);
4004
4005 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
4006 case TC_ACT_SHOT:
4007 qdisc_qstats_drop(sch);
4008 __qdisc_drop(skb, to_free);
4009 *ret = __NET_XMIT_BYPASS;
4010 return NULL;
4011 case TC_ACT_STOLEN:
4012 case TC_ACT_QUEUED:
4013 case TC_ACT_TRAP:
4014 __qdisc_drop(skb, to_free);
4015 *ret = __NET_XMIT_STOLEN;
4016 return NULL;
4017 case TC_ACT_REDIRECT:
4018 skb_do_redirect(skb);
4019 *ret = __NET_XMIT_STOLEN;
4020 return NULL;
4021 }
4022
4023 return skb;
4024}
4025EXPORT_SYMBOL(tcf_qevent_handle);
4026
4027int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
4028{
4029 if (!qe->info.block_index)
4030 return 0;
4031 return nla_put_u32(skb, attr_name, qe->info.block_index);
4032}
4033EXPORT_SYMBOL(tcf_qevent_dump);
4034#endif
4035
4036static __net_init int tcf_net_init(struct net *net)
4037{
4038 struct tcf_net *tn = net_generic(net, tcf_net_id);
4039
4040 spin_lock_init(&tn->idr_lock);
4041 idr_init(&tn->idr);
4042 return 0;
4043}
4044
4045static void __net_exit tcf_net_exit(struct net *net)
4046{
4047 struct tcf_net *tn = net_generic(net, tcf_net_id);
4048
4049 idr_destroy(&tn->idr);
4050}
4051
4052static struct pernet_operations tcf_net_ops = {
4053 .init = tcf_net_init,
4054 .exit = tcf_net_exit,
4055 .id = &tcf_net_id,
4056 .size = sizeof(struct tcf_net),
4057};
4058
4059static const struct rtnl_msg_handler tc_filter_rtnl_msg_handlers[] __initconst = {
4060 {.msgtype = RTM_NEWTFILTER, .doit = tc_new_tfilter,
4061 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4062 {.msgtype = RTM_DELTFILTER, .doit = tc_del_tfilter,
4063 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4064 {.msgtype = RTM_GETTFILTER, .doit = tc_get_tfilter,
4065 .dumpit = tc_dump_tfilter, .flags = RTNL_FLAG_DOIT_UNLOCKED},
4066 {.msgtype = RTM_NEWCHAIN, .doit = tc_ctl_chain},
4067 {.msgtype = RTM_DELCHAIN, .doit = tc_ctl_chain},
4068 {.msgtype = RTM_GETCHAIN, .doit = tc_ctl_chain,
4069 .dumpit = tc_dump_chain},
4070};
4071
4072static int __init tc_filter_init(void)
4073{
4074 int err;
4075
4076 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4077 if (!tc_filter_wq)
4078 return -ENOMEM;
4079
4080 err = register_pernet_subsys(&tcf_net_ops);
4081 if (err)
4082 goto err_register_pernet_subsys;
4083
4084 xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4085 rtnl_register_many(tc_filter_rtnl_msg_handlers);
4086
4087 return 0;
4088
4089err_register_pernet_subsys:
4090 destroy_workqueue(tc_filter_wq);
4091 return err;
4092}
4093
4094subsys_initcall(tc_filter_init);