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