Merge tag 'rproc-v6.14' of git://git.kernel.org/pub/scm/linux/kernel/git/remoteproc...
[linux.git] / net / sched / cls_api.c
blob8e47e5355be613d6eccea098d0d83291f8f2d33c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * net/sched/cls_api.c Packet classifier API.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
7 * Changes:
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/err.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/jhash.h>
24 #include <linux/rculist.h>
25 #include <linux/rhashtable.h>
26 #include <net/net_namespace.h>
27 #include <net/sock.h>
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/pkt_cls.h>
31 #include <net/tc_act/tc_pedit.h>
32 #include <net/tc_act/tc_mirred.h>
33 #include <net/tc_act/tc_vlan.h>
34 #include <net/tc_act/tc_tunnel_key.h>
35 #include <net/tc_act/tc_csum.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_police.h>
38 #include <net/tc_act/tc_sample.h>
39 #include <net/tc_act/tc_skbedit.h>
40 #include <net/tc_act/tc_ct.h>
41 #include <net/tc_act/tc_mpls.h>
42 #include <net/tc_act/tc_gate.h>
43 #include <net/flow_offload.h>
44 #include <net/tc_wrapper.h>
46 /* The list of all installed classifier types */
47 static LIST_HEAD(tcf_proto_base);
49 /* Protects list of registered TC modules. It is pure SMP lock. */
50 static DEFINE_RWLOCK(cls_mod_lock);
52 static struct xarray tcf_exts_miss_cookies_xa;
53 struct tcf_exts_miss_cookie_node {
54 const struct tcf_chain *chain;
55 const struct tcf_proto *tp;
56 const struct tcf_exts *exts;
57 u32 chain_index;
58 u32 tp_prio;
59 u32 handle;
60 u32 miss_cookie_base;
61 struct rcu_head rcu;
64 /* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
65 * action index in the exts tc actions array.
67 union tcf_exts_miss_cookie {
68 struct {
69 u32 miss_cookie_base;
70 u32 act_index;
72 u64 miss_cookie;
75 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
76 static int
77 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
78 u32 handle)
80 struct tcf_exts_miss_cookie_node *n;
81 static u32 next;
82 int err;
84 if (WARN_ON(!handle || !tp->ops->get_exts))
85 return -EINVAL;
87 n = kzalloc(sizeof(*n), GFP_KERNEL);
88 if (!n)
89 return -ENOMEM;
91 n->chain_index = tp->chain->index;
92 n->chain = tp->chain;
93 n->tp_prio = tp->prio;
94 n->tp = tp;
95 n->exts = exts;
96 n->handle = handle;
98 err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
99 n, xa_limit_32b, &next, GFP_KERNEL);
100 if (err)
101 goto err_xa_alloc;
103 exts->miss_cookie_node = n;
104 return 0;
106 err_xa_alloc:
107 kfree(n);
108 return err;
111 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
113 struct tcf_exts_miss_cookie_node *n;
115 if (!exts->miss_cookie_node)
116 return;
118 n = exts->miss_cookie_node;
119 xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
120 kfree_rcu(n, rcu);
123 static struct tcf_exts_miss_cookie_node *
124 tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
126 union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
128 *act_index = mc.act_index;
129 return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
131 #else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
132 static int
133 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
134 u32 handle)
136 return 0;
139 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
142 #endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
144 static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
146 union tcf_exts_miss_cookie mc = { .act_index = act_index, };
148 if (!miss_cookie_base)
149 return 0;
151 mc.miss_cookie_base = miss_cookie_base;
152 return mc.miss_cookie;
155 #ifdef CONFIG_NET_CLS_ACT
156 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
157 EXPORT_SYMBOL(tc_skb_ext_tc);
159 void tc_skb_ext_tc_enable(void)
161 static_branch_inc(&tc_skb_ext_tc);
163 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
165 void tc_skb_ext_tc_disable(void)
167 static_branch_dec(&tc_skb_ext_tc);
169 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
170 #endif
172 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
174 return jhash_3words(tp->chain->index, tp->prio,
175 (__force __u32)tp->protocol, 0);
178 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
179 struct tcf_proto *tp)
181 struct tcf_block *block = chain->block;
183 mutex_lock(&block->proto_destroy_lock);
184 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
185 destroy_obj_hashfn(tp));
186 mutex_unlock(&block->proto_destroy_lock);
189 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
190 const struct tcf_proto *tp2)
192 return tp1->chain->index == tp2->chain->index &&
193 tp1->prio == tp2->prio &&
194 tp1->protocol == tp2->protocol;
197 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
198 struct tcf_proto *tp)
200 u32 hash = destroy_obj_hashfn(tp);
201 struct tcf_proto *iter;
202 bool found = false;
204 rcu_read_lock();
205 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
206 destroy_ht_node, hash) {
207 if (tcf_proto_cmp(tp, iter)) {
208 found = true;
209 break;
212 rcu_read_unlock();
214 return found;
217 static void
218 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
220 struct tcf_block *block = chain->block;
222 mutex_lock(&block->proto_destroy_lock);
223 if (hash_hashed(&tp->destroy_ht_node))
224 hash_del_rcu(&tp->destroy_ht_node);
225 mutex_unlock(&block->proto_destroy_lock);
228 /* Find classifier type by string name */
230 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
232 const struct tcf_proto_ops *t, *res = NULL;
234 if (kind) {
235 read_lock(&cls_mod_lock);
236 list_for_each_entry(t, &tcf_proto_base, head) {
237 if (strcmp(kind, t->kind) == 0) {
238 if (try_module_get(t->owner))
239 res = t;
240 break;
243 read_unlock(&cls_mod_lock);
245 return res;
248 static const struct tcf_proto_ops *
249 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
250 struct netlink_ext_ack *extack)
252 const struct tcf_proto_ops *ops;
254 ops = __tcf_proto_lookup_ops(kind);
255 if (ops)
256 return ops;
257 #ifdef CONFIG_MODULES
258 if (rtnl_held)
259 rtnl_unlock();
260 request_module(NET_CLS_ALIAS_PREFIX "%s", kind);
261 if (rtnl_held)
262 rtnl_lock();
263 ops = __tcf_proto_lookup_ops(kind);
264 /* We dropped the RTNL semaphore in order to perform
265 * the module load. So, even if we succeeded in loading
266 * the module we have to replay the request. We indicate
267 * this using -EAGAIN.
269 if (ops) {
270 module_put(ops->owner);
271 return ERR_PTR(-EAGAIN);
273 #endif
274 NL_SET_ERR_MSG(extack, "TC classifier not found");
275 return ERR_PTR(-ENOENT);
278 /* Register(unregister) new classifier type */
280 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
282 struct tcf_proto_ops *t;
283 int rc = -EEXIST;
285 write_lock(&cls_mod_lock);
286 list_for_each_entry(t, &tcf_proto_base, head)
287 if (!strcmp(ops->kind, t->kind))
288 goto out;
290 list_add_tail(&ops->head, &tcf_proto_base);
291 rc = 0;
292 out:
293 write_unlock(&cls_mod_lock);
294 return rc;
296 EXPORT_SYMBOL(register_tcf_proto_ops);
298 static struct workqueue_struct *tc_filter_wq;
300 void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
302 struct tcf_proto_ops *t;
303 int rc = -ENOENT;
305 /* Wait for outstanding call_rcu()s, if any, from a
306 * tcf_proto_ops's destroy() handler.
308 rcu_barrier();
309 flush_workqueue(tc_filter_wq);
311 write_lock(&cls_mod_lock);
312 list_for_each_entry(t, &tcf_proto_base, head) {
313 if (t == ops) {
314 list_del(&t->head);
315 rc = 0;
316 break;
319 write_unlock(&cls_mod_lock);
321 WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
323 EXPORT_SYMBOL(unregister_tcf_proto_ops);
325 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
327 INIT_RCU_WORK(rwork, func);
328 return queue_rcu_work(tc_filter_wq, rwork);
330 EXPORT_SYMBOL(tcf_queue_work);
332 /* Select new prio value from the range, managed by kernel. */
334 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
336 u32 first = TC_H_MAKE(0xC0000000U, 0U);
338 if (tp)
339 first = tp->prio - 1;
341 return TC_H_MAJ(first);
344 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
346 if (kind)
347 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
348 memset(name, 0, IFNAMSIZ);
349 return false;
352 static bool tcf_proto_is_unlocked(const char *kind)
354 const struct tcf_proto_ops *ops;
355 bool ret;
357 if (strlen(kind) == 0)
358 return false;
360 ops = tcf_proto_lookup_ops(kind, false, NULL);
361 /* On error return false to take rtnl lock. Proto lookup/create
362 * functions will perform lookup again and properly handle errors.
364 if (IS_ERR(ops))
365 return false;
367 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
368 module_put(ops->owner);
369 return ret;
372 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
373 u32 prio, struct tcf_chain *chain,
374 bool rtnl_held,
375 struct netlink_ext_ack *extack)
377 struct tcf_proto *tp;
378 int err;
380 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
381 if (!tp)
382 return ERR_PTR(-ENOBUFS);
384 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
385 if (IS_ERR(tp->ops)) {
386 err = PTR_ERR(tp->ops);
387 goto errout;
389 tp->classify = tp->ops->classify;
390 tp->protocol = protocol;
391 tp->prio = prio;
392 tp->chain = chain;
393 tp->usesw = !tp->ops->reoffload;
394 spin_lock_init(&tp->lock);
395 refcount_set(&tp->refcnt, 1);
397 err = tp->ops->init(tp);
398 if (err) {
399 module_put(tp->ops->owner);
400 goto errout;
402 return tp;
404 errout:
405 kfree(tp);
406 return ERR_PTR(err);
409 static void tcf_proto_get(struct tcf_proto *tp)
411 refcount_inc(&tp->refcnt);
414 static void tcf_proto_count_usesw(struct tcf_proto *tp, bool add)
416 #ifdef CONFIG_NET_CLS_ACT
417 struct tcf_block *block = tp->chain->block;
418 bool counted = false;
420 if (!add) {
421 if (tp->usesw && tp->counted) {
422 if (!atomic_dec_return(&block->useswcnt))
423 static_branch_dec(&tcf_sw_enabled_key);
424 tp->counted = false;
426 return;
429 spin_lock(&tp->lock);
430 if (tp->usesw && !tp->counted) {
431 counted = true;
432 tp->counted = true;
434 spin_unlock(&tp->lock);
436 if (counted && atomic_inc_return(&block->useswcnt) == 1)
437 static_branch_inc(&tcf_sw_enabled_key);
438 #endif
441 static void tcf_chain_put(struct tcf_chain *chain);
443 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
444 bool sig_destroy, struct netlink_ext_ack *extack)
446 tp->ops->destroy(tp, rtnl_held, extack);
447 tcf_proto_count_usesw(tp, false);
448 if (sig_destroy)
449 tcf_proto_signal_destroyed(tp->chain, tp);
450 tcf_chain_put(tp->chain);
451 module_put(tp->ops->owner);
452 kfree_rcu(tp, rcu);
455 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
456 struct netlink_ext_ack *extack)
458 if (refcount_dec_and_test(&tp->refcnt))
459 tcf_proto_destroy(tp, rtnl_held, true, extack);
462 static bool tcf_proto_check_delete(struct tcf_proto *tp)
464 if (tp->ops->delete_empty)
465 return tp->ops->delete_empty(tp);
467 tp->deleting = true;
468 return tp->deleting;
471 static void tcf_proto_mark_delete(struct tcf_proto *tp)
473 spin_lock(&tp->lock);
474 tp->deleting = true;
475 spin_unlock(&tp->lock);
478 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
480 bool deleting;
482 spin_lock(&tp->lock);
483 deleting = tp->deleting;
484 spin_unlock(&tp->lock);
486 return deleting;
489 #define ASSERT_BLOCK_LOCKED(block) \
490 lockdep_assert_held(&(block)->lock)
492 struct tcf_filter_chain_list_item {
493 struct list_head list;
494 tcf_chain_head_change_t *chain_head_change;
495 void *chain_head_change_priv;
498 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
499 u32 chain_index)
501 struct tcf_chain *chain;
503 ASSERT_BLOCK_LOCKED(block);
505 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
506 if (!chain)
507 return NULL;
508 list_add_tail_rcu(&chain->list, &block->chain_list);
509 mutex_init(&chain->filter_chain_lock);
510 chain->block = block;
511 chain->index = chain_index;
512 chain->refcnt = 1;
513 if (!chain->index)
514 block->chain0.chain = chain;
515 return chain;
518 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
519 struct tcf_proto *tp_head)
521 if (item->chain_head_change)
522 item->chain_head_change(tp_head, item->chain_head_change_priv);
525 static void tcf_chain0_head_change(struct tcf_chain *chain,
526 struct tcf_proto *tp_head)
528 struct tcf_filter_chain_list_item *item;
529 struct tcf_block *block = chain->block;
531 if (chain->index)
532 return;
534 mutex_lock(&block->lock);
535 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
536 tcf_chain_head_change_item(item, tp_head);
537 mutex_unlock(&block->lock);
540 /* Returns true if block can be safely freed. */
542 static bool tcf_chain_detach(struct tcf_chain *chain)
544 struct tcf_block *block = chain->block;
546 ASSERT_BLOCK_LOCKED(block);
548 list_del_rcu(&chain->list);
549 if (!chain->index)
550 block->chain0.chain = NULL;
552 if (list_empty(&block->chain_list) &&
553 refcount_read(&block->refcnt) == 0)
554 return true;
556 return false;
559 static void tcf_block_destroy(struct tcf_block *block)
561 mutex_destroy(&block->lock);
562 mutex_destroy(&block->proto_destroy_lock);
563 xa_destroy(&block->ports);
564 kfree_rcu(block, rcu);
567 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
569 struct tcf_block *block = chain->block;
571 mutex_destroy(&chain->filter_chain_lock);
572 kfree_rcu(chain, rcu);
573 if (free_block)
574 tcf_block_destroy(block);
577 static void tcf_chain_hold(struct tcf_chain *chain)
579 ASSERT_BLOCK_LOCKED(chain->block);
581 ++chain->refcnt;
584 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
586 ASSERT_BLOCK_LOCKED(chain->block);
588 /* In case all the references are action references, this
589 * chain should not be shown to the user.
591 return chain->refcnt == chain->action_refcnt;
594 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
595 u32 chain_index)
597 struct tcf_chain *chain;
599 ASSERT_BLOCK_LOCKED(block);
601 list_for_each_entry(chain, &block->chain_list, list) {
602 if (chain->index == chain_index)
603 return chain;
605 return NULL;
608 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
609 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
610 u32 chain_index)
612 struct tcf_chain *chain;
614 list_for_each_entry_rcu(chain, &block->chain_list, list) {
615 if (chain->index == chain_index)
616 return chain;
618 return NULL;
620 #endif
622 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
623 u32 seq, u16 flags, int event, bool unicast,
624 struct netlink_ext_ack *extack);
626 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
627 u32 chain_index, bool create,
628 bool by_act)
630 struct tcf_chain *chain = NULL;
631 bool is_first_reference;
633 mutex_lock(&block->lock);
634 chain = tcf_chain_lookup(block, chain_index);
635 if (chain) {
636 tcf_chain_hold(chain);
637 } else {
638 if (!create)
639 goto errout;
640 chain = tcf_chain_create(block, chain_index);
641 if (!chain)
642 goto errout;
645 if (by_act)
646 ++chain->action_refcnt;
647 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
648 mutex_unlock(&block->lock);
650 /* Send notification only in case we got the first
651 * non-action reference. Until then, the chain acts only as
652 * a placeholder for actions pointing to it and user ought
653 * not know about them.
655 if (is_first_reference && !by_act)
656 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
657 RTM_NEWCHAIN, false, NULL);
659 return chain;
661 errout:
662 mutex_unlock(&block->lock);
663 return chain;
666 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
667 bool create)
669 return __tcf_chain_get(block, chain_index, create, false);
672 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
674 return __tcf_chain_get(block, chain_index, true, true);
676 EXPORT_SYMBOL(tcf_chain_get_by_act);
678 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
679 void *tmplt_priv);
680 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
681 void *tmplt_priv, u32 chain_index,
682 struct tcf_block *block, struct sk_buff *oskb,
683 u32 seq, u16 flags);
685 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
686 bool explicitly_created)
688 struct tcf_block *block = chain->block;
689 const struct tcf_proto_ops *tmplt_ops;
690 unsigned int refcnt, non_act_refcnt;
691 bool free_block = false;
692 void *tmplt_priv;
694 mutex_lock(&block->lock);
695 if (explicitly_created) {
696 if (!chain->explicitly_created) {
697 mutex_unlock(&block->lock);
698 return;
700 chain->explicitly_created = false;
703 if (by_act)
704 chain->action_refcnt--;
706 /* tc_chain_notify_delete can't be called while holding block lock.
707 * However, when block is unlocked chain can be changed concurrently, so
708 * save these to temporary variables.
710 refcnt = --chain->refcnt;
711 non_act_refcnt = refcnt - chain->action_refcnt;
712 tmplt_ops = chain->tmplt_ops;
713 tmplt_priv = chain->tmplt_priv;
715 if (non_act_refcnt == chain->explicitly_created && !by_act) {
716 if (non_act_refcnt == 0)
717 tc_chain_notify_delete(tmplt_ops, tmplt_priv,
718 chain->index, block, NULL, 0, 0);
719 /* Last reference to chain, no need to lock. */
720 chain->flushing = false;
723 if (refcnt == 0)
724 free_block = tcf_chain_detach(chain);
725 mutex_unlock(&block->lock);
727 if (refcnt == 0) {
728 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
729 tcf_chain_destroy(chain, free_block);
733 static void tcf_chain_put(struct tcf_chain *chain)
735 __tcf_chain_put(chain, false, false);
738 void tcf_chain_put_by_act(struct tcf_chain *chain)
740 __tcf_chain_put(chain, true, false);
742 EXPORT_SYMBOL(tcf_chain_put_by_act);
744 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
746 __tcf_chain_put(chain, false, true);
749 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
751 struct tcf_proto *tp, *tp_next;
753 mutex_lock(&chain->filter_chain_lock);
754 tp = tcf_chain_dereference(chain->filter_chain, chain);
755 while (tp) {
756 tp_next = rcu_dereference_protected(tp->next, 1);
757 tcf_proto_signal_destroying(chain, tp);
758 tp = tp_next;
760 tp = tcf_chain_dereference(chain->filter_chain, chain);
761 RCU_INIT_POINTER(chain->filter_chain, NULL);
762 tcf_chain0_head_change(chain, NULL);
763 chain->flushing = true;
764 mutex_unlock(&chain->filter_chain_lock);
766 while (tp) {
767 tp_next = rcu_dereference_protected(tp->next, 1);
768 tcf_proto_put(tp, rtnl_held, NULL);
769 tp = tp_next;
773 static int tcf_block_setup(struct tcf_block *block,
774 struct flow_block_offload *bo);
776 static void tcf_block_offload_init(struct flow_block_offload *bo,
777 struct net_device *dev, struct Qdisc *sch,
778 enum flow_block_command command,
779 enum flow_block_binder_type binder_type,
780 struct flow_block *flow_block,
781 bool shared, struct netlink_ext_ack *extack)
783 bo->net = dev_net(dev);
784 bo->command = command;
785 bo->binder_type = binder_type;
786 bo->block = flow_block;
787 bo->block_shared = shared;
788 bo->extack = extack;
789 bo->sch = sch;
790 bo->cb_list_head = &flow_block->cb_list;
791 INIT_LIST_HEAD(&bo->cb_list);
794 static void tcf_block_unbind(struct tcf_block *block,
795 struct flow_block_offload *bo);
797 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
799 struct tcf_block *block = block_cb->indr.data;
800 struct net_device *dev = block_cb->indr.dev;
801 struct Qdisc *sch = block_cb->indr.sch;
802 struct netlink_ext_ack extack = {};
803 struct flow_block_offload bo = {};
805 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
806 block_cb->indr.binder_type,
807 &block->flow_block, tcf_block_shared(block),
808 &extack);
809 rtnl_lock();
810 down_write(&block->cb_lock);
811 list_del(&block_cb->driver_list);
812 list_move(&block_cb->list, &bo.cb_list);
813 tcf_block_unbind(block, &bo);
814 up_write(&block->cb_lock);
815 rtnl_unlock();
818 static bool tcf_block_offload_in_use(struct tcf_block *block)
820 return atomic_read(&block->offloadcnt);
823 static int tcf_block_offload_cmd(struct tcf_block *block,
824 struct net_device *dev, struct Qdisc *sch,
825 struct tcf_block_ext_info *ei,
826 enum flow_block_command command,
827 struct netlink_ext_ack *extack)
829 struct flow_block_offload bo = {};
831 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
832 &block->flow_block, tcf_block_shared(block),
833 extack);
835 if (dev->netdev_ops->ndo_setup_tc) {
836 int err;
838 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
839 if (err < 0) {
840 if (err != -EOPNOTSUPP)
841 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
842 return err;
845 return tcf_block_setup(block, &bo);
848 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
849 tc_block_indr_cleanup);
850 tcf_block_setup(block, &bo);
852 return -EOPNOTSUPP;
855 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
856 struct tcf_block_ext_info *ei,
857 struct netlink_ext_ack *extack)
859 struct net_device *dev = q->dev_queue->dev;
860 int err;
862 down_write(&block->cb_lock);
864 /* If tc offload feature is disabled and the block we try to bind
865 * to already has some offloaded filters, forbid to bind.
867 if (dev->netdev_ops->ndo_setup_tc &&
868 !tc_can_offload(dev) &&
869 tcf_block_offload_in_use(block)) {
870 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
871 err = -EOPNOTSUPP;
872 goto err_unlock;
875 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
876 if (err == -EOPNOTSUPP)
877 goto no_offload_dev_inc;
878 if (err)
879 goto err_unlock;
881 up_write(&block->cb_lock);
882 return 0;
884 no_offload_dev_inc:
885 if (tcf_block_offload_in_use(block))
886 goto err_unlock;
888 err = 0;
889 block->nooffloaddevcnt++;
890 err_unlock:
891 up_write(&block->cb_lock);
892 return err;
895 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
896 struct tcf_block_ext_info *ei)
898 struct net_device *dev = q->dev_queue->dev;
899 int err;
901 down_write(&block->cb_lock);
902 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
903 if (err == -EOPNOTSUPP)
904 goto no_offload_dev_dec;
905 up_write(&block->cb_lock);
906 return;
908 no_offload_dev_dec:
909 WARN_ON(block->nooffloaddevcnt-- == 0);
910 up_write(&block->cb_lock);
913 static int
914 tcf_chain0_head_change_cb_add(struct tcf_block *block,
915 struct tcf_block_ext_info *ei,
916 struct netlink_ext_ack *extack)
918 struct tcf_filter_chain_list_item *item;
919 struct tcf_chain *chain0;
921 item = kmalloc(sizeof(*item), GFP_KERNEL);
922 if (!item) {
923 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
924 return -ENOMEM;
926 item->chain_head_change = ei->chain_head_change;
927 item->chain_head_change_priv = ei->chain_head_change_priv;
929 mutex_lock(&block->lock);
930 chain0 = block->chain0.chain;
931 if (chain0)
932 tcf_chain_hold(chain0);
933 else
934 list_add(&item->list, &block->chain0.filter_chain_list);
935 mutex_unlock(&block->lock);
937 if (chain0) {
938 struct tcf_proto *tp_head;
940 mutex_lock(&chain0->filter_chain_lock);
942 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
943 if (tp_head)
944 tcf_chain_head_change_item(item, tp_head);
946 mutex_lock(&block->lock);
947 list_add(&item->list, &block->chain0.filter_chain_list);
948 mutex_unlock(&block->lock);
950 mutex_unlock(&chain0->filter_chain_lock);
951 tcf_chain_put(chain0);
954 return 0;
957 static void
958 tcf_chain0_head_change_cb_del(struct tcf_block *block,
959 struct tcf_block_ext_info *ei)
961 struct tcf_filter_chain_list_item *item;
963 mutex_lock(&block->lock);
964 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
965 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
966 (item->chain_head_change == ei->chain_head_change &&
967 item->chain_head_change_priv == ei->chain_head_change_priv)) {
968 if (block->chain0.chain)
969 tcf_chain_head_change_item(item, NULL);
970 list_del(&item->list);
971 mutex_unlock(&block->lock);
973 kfree(item);
974 return;
977 mutex_unlock(&block->lock);
978 WARN_ON(1);
981 struct tcf_net {
982 spinlock_t idr_lock; /* Protects idr */
983 struct idr idr;
986 static unsigned int tcf_net_id;
988 static int tcf_block_insert(struct tcf_block *block, struct net *net,
989 struct netlink_ext_ack *extack)
991 struct tcf_net *tn = net_generic(net, tcf_net_id);
992 int err;
994 idr_preload(GFP_KERNEL);
995 spin_lock(&tn->idr_lock);
996 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
997 GFP_NOWAIT);
998 spin_unlock(&tn->idr_lock);
999 idr_preload_end();
1001 return err;
1004 static void tcf_block_remove(struct tcf_block *block, struct net *net)
1006 struct tcf_net *tn = net_generic(net, tcf_net_id);
1008 spin_lock(&tn->idr_lock);
1009 idr_remove(&tn->idr, block->index);
1010 spin_unlock(&tn->idr_lock);
1013 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
1014 u32 block_index,
1015 struct netlink_ext_ack *extack)
1017 struct tcf_block *block;
1019 block = kzalloc(sizeof(*block), GFP_KERNEL);
1020 if (!block) {
1021 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
1022 return ERR_PTR(-ENOMEM);
1024 mutex_init(&block->lock);
1025 mutex_init(&block->proto_destroy_lock);
1026 init_rwsem(&block->cb_lock);
1027 flow_block_init(&block->flow_block);
1028 INIT_LIST_HEAD(&block->chain_list);
1029 INIT_LIST_HEAD(&block->owner_list);
1030 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1032 refcount_set(&block->refcnt, 1);
1033 block->net = net;
1034 block->index = block_index;
1035 xa_init(&block->ports);
1037 /* Don't store q pointer for blocks which are shared */
1038 if (!tcf_block_shared(block))
1039 block->q = q;
1040 return block;
1043 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1045 struct tcf_net *tn = net_generic(net, tcf_net_id);
1047 return idr_find(&tn->idr, block_index);
1049 EXPORT_SYMBOL(tcf_block_lookup);
1051 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1053 struct tcf_block *block;
1055 rcu_read_lock();
1056 block = tcf_block_lookup(net, block_index);
1057 if (block && !refcount_inc_not_zero(&block->refcnt))
1058 block = NULL;
1059 rcu_read_unlock();
1061 return block;
1064 static struct tcf_chain *
1065 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1067 mutex_lock(&block->lock);
1068 if (chain)
1069 chain = list_is_last(&chain->list, &block->chain_list) ?
1070 NULL : list_next_entry(chain, list);
1071 else
1072 chain = list_first_entry_or_null(&block->chain_list,
1073 struct tcf_chain, list);
1075 /* skip all action-only chains */
1076 while (chain && tcf_chain_held_by_acts_only(chain))
1077 chain = list_is_last(&chain->list, &block->chain_list) ?
1078 NULL : list_next_entry(chain, list);
1080 if (chain)
1081 tcf_chain_hold(chain);
1082 mutex_unlock(&block->lock);
1084 return chain;
1087 /* Function to be used by all clients that want to iterate over all chains on
1088 * block. It properly obtains block->lock and takes reference to chain before
1089 * returning it. Users of this function must be tolerant to concurrent chain
1090 * insertion/deletion or ensure that no concurrent chain modification is
1091 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1092 * consistent dump because rtnl lock is released each time skb is filled with
1093 * data and sent to user-space.
1096 struct tcf_chain *
1097 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1099 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1101 if (chain)
1102 tcf_chain_put(chain);
1104 return chain_next;
1106 EXPORT_SYMBOL(tcf_get_next_chain);
1108 static struct tcf_proto *
1109 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1111 u32 prio = 0;
1113 ASSERT_RTNL();
1114 mutex_lock(&chain->filter_chain_lock);
1116 if (!tp) {
1117 tp = tcf_chain_dereference(chain->filter_chain, chain);
1118 } else if (tcf_proto_is_deleting(tp)) {
1119 /* 'deleting' flag is set and chain->filter_chain_lock was
1120 * unlocked, which means next pointer could be invalid. Restart
1121 * search.
1123 prio = tp->prio + 1;
1124 tp = tcf_chain_dereference(chain->filter_chain, chain);
1126 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1127 if (!tp->deleting && tp->prio >= prio)
1128 break;
1129 } else {
1130 tp = tcf_chain_dereference(tp->next, chain);
1133 if (tp)
1134 tcf_proto_get(tp);
1136 mutex_unlock(&chain->filter_chain_lock);
1138 return tp;
1141 /* Function to be used by all clients that want to iterate over all tp's on
1142 * chain. Users of this function must be tolerant to concurrent tp
1143 * insertion/deletion or ensure that no concurrent chain modification is
1144 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1145 * consistent dump because rtnl lock is released each time skb is filled with
1146 * data and sent to user-space.
1149 struct tcf_proto *
1150 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1152 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1154 if (tp)
1155 tcf_proto_put(tp, true, NULL);
1157 return tp_next;
1159 EXPORT_SYMBOL(tcf_get_next_proto);
1161 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1163 struct tcf_chain *chain;
1165 /* Last reference to block. At this point chains cannot be added or
1166 * removed concurrently.
1168 for (chain = tcf_get_next_chain(block, NULL);
1169 chain;
1170 chain = tcf_get_next_chain(block, chain)) {
1171 tcf_chain_put_explicitly_created(chain);
1172 tcf_chain_flush(chain, rtnl_held);
1176 /* Lookup Qdisc and increments its reference counter.
1177 * Set parent, if necessary.
1180 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1181 u32 *parent, int ifindex, bool rtnl_held,
1182 struct netlink_ext_ack *extack)
1184 const struct Qdisc_class_ops *cops;
1185 struct net_device *dev;
1186 int err = 0;
1188 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1189 return 0;
1191 rcu_read_lock();
1193 /* Find link */
1194 dev = dev_get_by_index_rcu(net, ifindex);
1195 if (!dev) {
1196 rcu_read_unlock();
1197 return -ENODEV;
1200 /* Find qdisc */
1201 if (!*parent) {
1202 *q = rcu_dereference(dev->qdisc);
1203 *parent = (*q)->handle;
1204 } else {
1205 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1206 if (!*q) {
1207 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1208 err = -EINVAL;
1209 goto errout_rcu;
1213 *q = qdisc_refcount_inc_nz(*q);
1214 if (!*q) {
1215 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1216 err = -EINVAL;
1217 goto errout_rcu;
1220 /* Is it classful? */
1221 cops = (*q)->ops->cl_ops;
1222 if (!cops) {
1223 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1224 err = -EINVAL;
1225 goto errout_qdisc;
1228 if (!cops->tcf_block) {
1229 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1230 err = -EOPNOTSUPP;
1231 goto errout_qdisc;
1234 errout_rcu:
1235 /* At this point we know that qdisc is not noop_qdisc,
1236 * which means that qdisc holds a reference to net_device
1237 * and we hold a reference to qdisc, so it is safe to release
1238 * rcu read lock.
1240 rcu_read_unlock();
1241 return err;
1243 errout_qdisc:
1244 rcu_read_unlock();
1246 if (rtnl_held)
1247 qdisc_put(*q);
1248 else
1249 qdisc_put_unlocked(*q);
1250 *q = NULL;
1252 return err;
1255 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1256 int ifindex, struct netlink_ext_ack *extack)
1258 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1259 return 0;
1261 /* Do we search for filter, attached to class? */
1262 if (TC_H_MIN(parent)) {
1263 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1265 *cl = cops->find(q, parent);
1266 if (*cl == 0) {
1267 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1268 return -ENOENT;
1272 return 0;
1275 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1276 unsigned long cl, int ifindex,
1277 u32 block_index,
1278 struct netlink_ext_ack *extack)
1280 struct tcf_block *block;
1282 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1283 block = tcf_block_refcnt_get(net, block_index);
1284 if (!block) {
1285 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1286 return ERR_PTR(-EINVAL);
1288 } else {
1289 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1291 block = cops->tcf_block(q, cl, extack);
1292 if (!block)
1293 return ERR_PTR(-EINVAL);
1295 if (tcf_block_shared(block)) {
1296 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1297 return ERR_PTR(-EOPNOTSUPP);
1300 /* Always take reference to block in order to support execution
1301 * of rules update path of cls API without rtnl lock. Caller
1302 * must release block when it is finished using it. 'if' block
1303 * of this conditional obtain reference to block by calling
1304 * tcf_block_refcnt_get().
1306 refcount_inc(&block->refcnt);
1309 return block;
1312 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1313 struct tcf_block_ext_info *ei, bool rtnl_held)
1315 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1316 /* Flushing/putting all chains will cause the block to be
1317 * deallocated when last chain is freed. However, if chain_list
1318 * is empty, block has to be manually deallocated. After block
1319 * reference counter reached 0, it is no longer possible to
1320 * increment it or add new chains to block.
1322 bool free_block = list_empty(&block->chain_list);
1324 mutex_unlock(&block->lock);
1325 if (tcf_block_shared(block))
1326 tcf_block_remove(block, block->net);
1328 if (q)
1329 tcf_block_offload_unbind(block, q, ei);
1331 if (free_block)
1332 tcf_block_destroy(block);
1333 else
1334 tcf_block_flush_all_chains(block, rtnl_held);
1335 } else if (q) {
1336 tcf_block_offload_unbind(block, q, ei);
1340 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1342 __tcf_block_put(block, NULL, NULL, rtnl_held);
1345 /* Find tcf block.
1346 * Set q, parent, cl when appropriate.
1349 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1350 u32 *parent, unsigned long *cl,
1351 int ifindex, u32 block_index,
1352 struct netlink_ext_ack *extack)
1354 struct tcf_block *block;
1355 int err = 0;
1357 ASSERT_RTNL();
1359 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1360 if (err)
1361 goto errout;
1363 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1364 if (err)
1365 goto errout_qdisc;
1367 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1368 if (IS_ERR(block)) {
1369 err = PTR_ERR(block);
1370 goto errout_qdisc;
1373 return block;
1375 errout_qdisc:
1376 if (*q)
1377 qdisc_put(*q);
1378 errout:
1379 *q = NULL;
1380 return ERR_PTR(err);
1383 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1384 bool rtnl_held)
1386 if (!IS_ERR_OR_NULL(block))
1387 tcf_block_refcnt_put(block, rtnl_held);
1389 if (q) {
1390 if (rtnl_held)
1391 qdisc_put(q);
1392 else
1393 qdisc_put_unlocked(q);
1397 struct tcf_block_owner_item {
1398 struct list_head list;
1399 struct Qdisc *q;
1400 enum flow_block_binder_type binder_type;
1403 static void
1404 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1405 struct Qdisc *q,
1406 enum flow_block_binder_type binder_type)
1408 if (block->keep_dst &&
1409 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1410 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1411 netif_keep_dst(qdisc_dev(q));
1414 void tcf_block_netif_keep_dst(struct tcf_block *block)
1416 struct tcf_block_owner_item *item;
1418 block->keep_dst = true;
1419 list_for_each_entry(item, &block->owner_list, list)
1420 tcf_block_owner_netif_keep_dst(block, item->q,
1421 item->binder_type);
1423 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1425 static int tcf_block_owner_add(struct tcf_block *block,
1426 struct Qdisc *q,
1427 enum flow_block_binder_type binder_type)
1429 struct tcf_block_owner_item *item;
1431 item = kmalloc(sizeof(*item), GFP_KERNEL);
1432 if (!item)
1433 return -ENOMEM;
1434 item->q = q;
1435 item->binder_type = binder_type;
1436 list_add(&item->list, &block->owner_list);
1437 return 0;
1440 static void tcf_block_owner_del(struct tcf_block *block,
1441 struct Qdisc *q,
1442 enum flow_block_binder_type binder_type)
1444 struct tcf_block_owner_item *item;
1446 list_for_each_entry(item, &block->owner_list, list) {
1447 if (item->q == q && item->binder_type == binder_type) {
1448 list_del(&item->list);
1449 kfree(item);
1450 return;
1453 WARN_ON(1);
1456 static bool tcf_block_tracks_dev(struct tcf_block *block,
1457 struct tcf_block_ext_info *ei)
1459 return tcf_block_shared(block) &&
1460 (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1461 ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1464 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1465 struct tcf_block_ext_info *ei,
1466 struct netlink_ext_ack *extack)
1468 struct net_device *dev = qdisc_dev(q);
1469 struct net *net = qdisc_net(q);
1470 struct tcf_block *block = NULL;
1471 int err;
1473 if (ei->block_index)
1474 /* block_index not 0 means the shared block is requested */
1475 block = tcf_block_refcnt_get(net, ei->block_index);
1477 if (!block) {
1478 block = tcf_block_create(net, q, ei->block_index, extack);
1479 if (IS_ERR(block))
1480 return PTR_ERR(block);
1481 if (tcf_block_shared(block)) {
1482 err = tcf_block_insert(block, net, extack);
1483 if (err)
1484 goto err_block_insert;
1488 err = tcf_block_owner_add(block, q, ei->binder_type);
1489 if (err)
1490 goto err_block_owner_add;
1492 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1494 err = tcf_chain0_head_change_cb_add(block, ei, extack);
1495 if (err)
1496 goto err_chain0_head_change_cb_add;
1498 err = tcf_block_offload_bind(block, q, ei, extack);
1499 if (err)
1500 goto err_block_offload_bind;
1502 if (tcf_block_tracks_dev(block, ei)) {
1503 err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1504 if (err) {
1505 NL_SET_ERR_MSG(extack, "block dev insert failed");
1506 goto err_dev_insert;
1510 *p_block = block;
1511 return 0;
1513 err_dev_insert:
1514 tcf_block_offload_unbind(block, q, ei);
1515 err_block_offload_bind:
1516 tcf_chain0_head_change_cb_del(block, ei);
1517 err_chain0_head_change_cb_add:
1518 tcf_block_owner_del(block, q, ei->binder_type);
1519 err_block_owner_add:
1520 err_block_insert:
1521 tcf_block_refcnt_put(block, true);
1522 return err;
1524 EXPORT_SYMBOL(tcf_block_get_ext);
1526 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1528 struct tcf_proto __rcu **p_filter_chain = priv;
1530 rcu_assign_pointer(*p_filter_chain, tp_head);
1533 int tcf_block_get(struct tcf_block **p_block,
1534 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1535 struct netlink_ext_ack *extack)
1537 struct tcf_block_ext_info ei = {
1538 .chain_head_change = tcf_chain_head_change_dflt,
1539 .chain_head_change_priv = p_filter_chain,
1542 WARN_ON(!p_filter_chain);
1543 return tcf_block_get_ext(p_block, q, &ei, extack);
1545 EXPORT_SYMBOL(tcf_block_get);
1547 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1548 * actions should be all removed after flushing.
1550 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1551 struct tcf_block_ext_info *ei)
1553 struct net_device *dev = qdisc_dev(q);
1555 if (!block)
1556 return;
1557 if (tcf_block_tracks_dev(block, ei))
1558 xa_erase(&block->ports, dev->ifindex);
1559 tcf_chain0_head_change_cb_del(block, ei);
1560 tcf_block_owner_del(block, q, ei->binder_type);
1562 __tcf_block_put(block, q, ei, true);
1564 EXPORT_SYMBOL(tcf_block_put_ext);
1566 void tcf_block_put(struct tcf_block *block)
1568 struct tcf_block_ext_info ei = {0, };
1570 if (!block)
1571 return;
1572 tcf_block_put_ext(block, block->q, &ei);
1575 EXPORT_SYMBOL(tcf_block_put);
1577 static int
1578 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1579 void *cb_priv, bool add, bool offload_in_use,
1580 struct netlink_ext_ack *extack)
1582 struct tcf_chain *chain, *chain_prev;
1583 struct tcf_proto *tp, *tp_prev;
1584 int err;
1586 lockdep_assert_held(&block->cb_lock);
1588 for (chain = __tcf_get_next_chain(block, NULL);
1589 chain;
1590 chain_prev = chain,
1591 chain = __tcf_get_next_chain(block, chain),
1592 tcf_chain_put(chain_prev)) {
1593 if (chain->tmplt_ops && add)
1594 chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1595 cb_priv);
1596 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1597 tp_prev = tp,
1598 tp = __tcf_get_next_proto(chain, tp),
1599 tcf_proto_put(tp_prev, true, NULL)) {
1600 if (tp->ops->reoffload) {
1601 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1602 extack);
1603 if (err && add)
1604 goto err_playback_remove;
1605 } else if (add && offload_in_use) {
1606 err = -EOPNOTSUPP;
1607 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1608 goto err_playback_remove;
1611 if (chain->tmplt_ops && !add)
1612 chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1613 cb_priv);
1616 return 0;
1618 err_playback_remove:
1619 tcf_proto_put(tp, true, NULL);
1620 tcf_chain_put(chain);
1621 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1622 extack);
1623 return err;
1626 static int tcf_block_bind(struct tcf_block *block,
1627 struct flow_block_offload *bo)
1629 struct flow_block_cb *block_cb, *next;
1630 int err, i = 0;
1632 lockdep_assert_held(&block->cb_lock);
1634 list_for_each_entry(block_cb, &bo->cb_list, list) {
1635 err = tcf_block_playback_offloads(block, block_cb->cb,
1636 block_cb->cb_priv, true,
1637 tcf_block_offload_in_use(block),
1638 bo->extack);
1639 if (err)
1640 goto err_unroll;
1641 if (!bo->unlocked_driver_cb)
1642 block->lockeddevcnt++;
1644 i++;
1646 list_splice(&bo->cb_list, &block->flow_block.cb_list);
1648 return 0;
1650 err_unroll:
1651 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1652 list_del(&block_cb->driver_list);
1653 if (i-- > 0) {
1654 list_del(&block_cb->list);
1655 tcf_block_playback_offloads(block, block_cb->cb,
1656 block_cb->cb_priv, false,
1657 tcf_block_offload_in_use(block),
1658 NULL);
1659 if (!bo->unlocked_driver_cb)
1660 block->lockeddevcnt--;
1662 flow_block_cb_free(block_cb);
1665 return err;
1668 static void tcf_block_unbind(struct tcf_block *block,
1669 struct flow_block_offload *bo)
1671 struct flow_block_cb *block_cb, *next;
1673 lockdep_assert_held(&block->cb_lock);
1675 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1676 tcf_block_playback_offloads(block, block_cb->cb,
1677 block_cb->cb_priv, false,
1678 tcf_block_offload_in_use(block),
1679 NULL);
1680 list_del(&block_cb->list);
1681 flow_block_cb_free(block_cb);
1682 if (!bo->unlocked_driver_cb)
1683 block->lockeddevcnt--;
1687 static int tcf_block_setup(struct tcf_block *block,
1688 struct flow_block_offload *bo)
1690 int err;
1692 switch (bo->command) {
1693 case FLOW_BLOCK_BIND:
1694 err = tcf_block_bind(block, bo);
1695 break;
1696 case FLOW_BLOCK_UNBIND:
1697 err = 0;
1698 tcf_block_unbind(block, bo);
1699 break;
1700 default:
1701 WARN_ON_ONCE(1);
1702 err = -EOPNOTSUPP;
1705 return err;
1708 /* Main classifier routine: scans classifier chain attached
1709 * to this qdisc, (optionally) tests for protocol and asks
1710 * specific classifiers.
1712 static inline int __tcf_classify(struct sk_buff *skb,
1713 const struct tcf_proto *tp,
1714 const struct tcf_proto *orig_tp,
1715 struct tcf_result *res,
1716 bool compat_mode,
1717 struct tcf_exts_miss_cookie_node *n,
1718 int act_index,
1719 u32 *last_executed_chain)
1721 #ifdef CONFIG_NET_CLS_ACT
1722 const int max_reclassify_loop = 16;
1723 const struct tcf_proto *first_tp;
1724 int limit = 0;
1726 reclassify:
1727 #endif
1728 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1729 __be16 protocol = skb_protocol(skb, false);
1730 int err = 0;
1732 if (n) {
1733 struct tcf_exts *exts;
1735 if (n->tp_prio != tp->prio)
1736 continue;
1738 /* We re-lookup the tp and chain based on index instead
1739 * of having hard refs and locks to them, so do a sanity
1740 * check if any of tp,chain,exts was replaced by the
1741 * time we got here with a cookie from hardware.
1743 if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1744 !tp->ops->get_exts)) {
1745 tcf_set_drop_reason(skb,
1746 SKB_DROP_REASON_TC_COOKIE_ERROR);
1747 return TC_ACT_SHOT;
1750 exts = tp->ops->get_exts(tp, n->handle);
1751 if (unlikely(!exts || n->exts != exts)) {
1752 tcf_set_drop_reason(skb,
1753 SKB_DROP_REASON_TC_COOKIE_ERROR);
1754 return TC_ACT_SHOT;
1757 n = NULL;
1758 err = tcf_exts_exec_ex(skb, exts, act_index, res);
1759 } else {
1760 if (tp->protocol != protocol &&
1761 tp->protocol != htons(ETH_P_ALL))
1762 continue;
1764 err = tc_classify(skb, tp, res);
1766 #ifdef CONFIG_NET_CLS_ACT
1767 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1768 first_tp = orig_tp;
1769 *last_executed_chain = first_tp->chain->index;
1770 goto reset;
1771 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1772 first_tp = res->goto_tp;
1773 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1774 goto reset;
1776 #endif
1777 if (err >= 0)
1778 return err;
1781 if (unlikely(n)) {
1782 tcf_set_drop_reason(skb,
1783 SKB_DROP_REASON_TC_COOKIE_ERROR);
1784 return TC_ACT_SHOT;
1787 return TC_ACT_UNSPEC; /* signal: continue lookup */
1788 #ifdef CONFIG_NET_CLS_ACT
1789 reset:
1790 if (unlikely(limit++ >= max_reclassify_loop)) {
1791 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1792 tp->chain->block->index,
1793 tp->prio & 0xffff,
1794 ntohs(tp->protocol));
1795 tcf_set_drop_reason(skb,
1796 SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1797 return TC_ACT_SHOT;
1800 tp = first_tp;
1801 goto reclassify;
1802 #endif
1805 int tcf_classify(struct sk_buff *skb,
1806 const struct tcf_block *block,
1807 const struct tcf_proto *tp,
1808 struct tcf_result *res, bool compat_mode)
1810 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1811 u32 last_executed_chain = 0;
1813 return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1814 &last_executed_chain);
1815 #else
1816 u32 last_executed_chain = tp ? tp->chain->index : 0;
1817 struct tcf_exts_miss_cookie_node *n = NULL;
1818 const struct tcf_proto *orig_tp = tp;
1819 struct tc_skb_ext *ext;
1820 int act_index = 0;
1821 int ret;
1823 if (block) {
1824 ext = skb_ext_find(skb, TC_SKB_EXT);
1826 if (ext && (ext->chain || ext->act_miss)) {
1827 struct tcf_chain *fchain;
1828 u32 chain;
1830 if (ext->act_miss) {
1831 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1832 &act_index);
1833 if (!n) {
1834 tcf_set_drop_reason(skb,
1835 SKB_DROP_REASON_TC_COOKIE_ERROR);
1836 return TC_ACT_SHOT;
1839 chain = n->chain_index;
1840 } else {
1841 chain = ext->chain;
1844 fchain = tcf_chain_lookup_rcu(block, chain);
1845 if (!fchain) {
1846 tcf_set_drop_reason(skb,
1847 SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1849 return TC_ACT_SHOT;
1852 /* Consume, so cloned/redirect skbs won't inherit ext */
1853 skb_ext_del(skb, TC_SKB_EXT);
1855 tp = rcu_dereference_bh(fchain->filter_chain);
1856 last_executed_chain = fchain->index;
1860 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1861 &last_executed_chain);
1863 if (tc_skb_ext_tc_enabled()) {
1864 /* If we missed on some chain */
1865 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1866 struct tc_skb_cb *cb = tc_skb_cb(skb);
1868 ext = tc_skb_ext_alloc(skb);
1869 if (WARN_ON_ONCE(!ext)) {
1870 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1871 return TC_ACT_SHOT;
1873 ext->chain = last_executed_chain;
1874 ext->mru = cb->mru;
1875 ext->post_ct = cb->post_ct;
1876 ext->post_ct_snat = cb->post_ct_snat;
1877 ext->post_ct_dnat = cb->post_ct_dnat;
1878 ext->zone = cb->zone;
1882 return ret;
1883 #endif
1885 EXPORT_SYMBOL(tcf_classify);
1887 struct tcf_chain_info {
1888 struct tcf_proto __rcu **pprev;
1889 struct tcf_proto __rcu *next;
1892 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1893 struct tcf_chain_info *chain_info)
1895 return tcf_chain_dereference(*chain_info->pprev, chain);
1898 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1899 struct tcf_chain_info *chain_info,
1900 struct tcf_proto *tp)
1902 if (chain->flushing)
1903 return -EAGAIN;
1905 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1906 if (*chain_info->pprev == chain->filter_chain)
1907 tcf_chain0_head_change(chain, tp);
1908 tcf_proto_get(tp);
1909 rcu_assign_pointer(*chain_info->pprev, tp);
1911 return 0;
1914 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1915 struct tcf_chain_info *chain_info,
1916 struct tcf_proto *tp)
1918 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1920 tcf_proto_mark_delete(tp);
1921 if (tp == chain->filter_chain)
1922 tcf_chain0_head_change(chain, next);
1923 RCU_INIT_POINTER(*chain_info->pprev, next);
1926 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1927 struct tcf_chain_info *chain_info,
1928 u32 protocol, u32 prio,
1929 bool prio_allocate,
1930 struct netlink_ext_ack *extack);
1932 /* Try to insert new proto.
1933 * If proto with specified priority already exists, free new proto
1934 * and return existing one.
1937 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1938 struct tcf_proto *tp_new,
1939 u32 protocol, u32 prio,
1940 bool rtnl_held)
1942 struct tcf_chain_info chain_info;
1943 struct tcf_proto *tp;
1944 int err = 0;
1946 mutex_lock(&chain->filter_chain_lock);
1948 if (tcf_proto_exists_destroying(chain, tp_new)) {
1949 mutex_unlock(&chain->filter_chain_lock);
1950 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1951 return ERR_PTR(-EAGAIN);
1954 tp = tcf_chain_tp_find(chain, &chain_info, protocol, prio, false, NULL);
1955 if (!tp)
1956 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1957 mutex_unlock(&chain->filter_chain_lock);
1959 if (tp) {
1960 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1961 tp_new = tp;
1962 } else if (err) {
1963 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1964 tp_new = ERR_PTR(err);
1967 return tp_new;
1970 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1971 struct tcf_proto *tp, bool rtnl_held,
1972 struct netlink_ext_ack *extack)
1974 struct tcf_chain_info chain_info;
1975 struct tcf_proto *tp_iter;
1976 struct tcf_proto **pprev;
1977 struct tcf_proto *next;
1979 mutex_lock(&chain->filter_chain_lock);
1981 /* Atomically find and remove tp from chain. */
1982 for (pprev = &chain->filter_chain;
1983 (tp_iter = tcf_chain_dereference(*pprev, chain));
1984 pprev = &tp_iter->next) {
1985 if (tp_iter == tp) {
1986 chain_info.pprev = pprev;
1987 chain_info.next = tp_iter->next;
1988 WARN_ON(tp_iter->deleting);
1989 break;
1992 /* Verify that tp still exists and no new filters were inserted
1993 * concurrently.
1994 * Mark tp for deletion if it is empty.
1996 if (!tp_iter || !tcf_proto_check_delete(tp)) {
1997 mutex_unlock(&chain->filter_chain_lock);
1998 return;
2001 tcf_proto_signal_destroying(chain, tp);
2002 next = tcf_chain_dereference(chain_info.next, chain);
2003 if (tp == chain->filter_chain)
2004 tcf_chain0_head_change(chain, next);
2005 RCU_INIT_POINTER(*chain_info.pprev, next);
2006 mutex_unlock(&chain->filter_chain_lock);
2008 tcf_proto_put(tp, rtnl_held, extack);
2011 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
2012 struct tcf_chain_info *chain_info,
2013 u32 protocol, u32 prio,
2014 bool prio_allocate,
2015 struct netlink_ext_ack *extack)
2017 struct tcf_proto **pprev;
2018 struct tcf_proto *tp;
2020 /* Check the chain for existence of proto-tcf with this priority */
2021 for (pprev = &chain->filter_chain;
2022 (tp = tcf_chain_dereference(*pprev, chain));
2023 pprev = &tp->next) {
2024 if (tp->prio >= prio) {
2025 if (tp->prio == prio) {
2026 if (prio_allocate) {
2027 NL_SET_ERR_MSG(extack, "Lowest ID from auto-alloc range already in use");
2028 return ERR_PTR(-ENOSPC);
2030 if (tp->protocol != protocol && protocol) {
2031 NL_SET_ERR_MSG(extack, "Protocol mismatch for filter with specified priority");
2032 return ERR_PTR(-EINVAL);
2034 } else {
2035 tp = NULL;
2037 break;
2040 chain_info->pprev = pprev;
2041 if (tp) {
2042 chain_info->next = tp->next;
2043 tcf_proto_get(tp);
2044 } else {
2045 chain_info->next = NULL;
2047 return tp;
2050 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2051 struct tcf_proto *tp, struct tcf_block *block,
2052 struct Qdisc *q, u32 parent, void *fh,
2053 u32 portid, u32 seq, u16 flags, int event,
2054 bool terse_dump, bool rtnl_held,
2055 struct netlink_ext_ack *extack)
2057 struct tcmsg *tcm;
2058 struct nlmsghdr *nlh;
2059 unsigned char *b = skb_tail_pointer(skb);
2061 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2062 if (!nlh)
2063 goto out_nlmsg_trim;
2064 tcm = nlmsg_data(nlh);
2065 tcm->tcm_family = AF_UNSPEC;
2066 tcm->tcm__pad1 = 0;
2067 tcm->tcm__pad2 = 0;
2068 if (q) {
2069 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2070 tcm->tcm_parent = parent;
2071 } else {
2072 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2073 tcm->tcm_block_index = block->index;
2075 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2076 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2077 goto nla_put_failure;
2078 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2079 goto nla_put_failure;
2080 if (!fh) {
2081 tcm->tcm_handle = 0;
2082 } else if (terse_dump) {
2083 if (tp->ops->terse_dump) {
2084 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2085 rtnl_held) < 0)
2086 goto nla_put_failure;
2087 } else {
2088 goto cls_op_not_supp;
2090 } else {
2091 if (tp->ops->dump &&
2092 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2093 goto nla_put_failure;
2096 if (extack && extack->_msg &&
2097 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2098 goto nla_put_failure;
2100 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2102 return skb->len;
2104 out_nlmsg_trim:
2105 nla_put_failure:
2106 cls_op_not_supp:
2107 nlmsg_trim(skb, b);
2108 return -1;
2111 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2112 struct nlmsghdr *n, struct tcf_proto *tp,
2113 struct tcf_block *block, struct Qdisc *q,
2114 u32 parent, void *fh, int event, bool unicast,
2115 bool rtnl_held, struct netlink_ext_ack *extack)
2117 struct sk_buff *skb;
2118 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2119 int err = 0;
2121 if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2122 return 0;
2124 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2125 if (!skb)
2126 return -ENOBUFS;
2128 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2129 n->nlmsg_seq, n->nlmsg_flags, event,
2130 false, rtnl_held, extack) <= 0) {
2131 kfree_skb(skb);
2132 return -EINVAL;
2135 if (unicast)
2136 err = rtnl_unicast(skb, net, portid);
2137 else
2138 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2139 n->nlmsg_flags & NLM_F_ECHO);
2140 return err;
2143 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2144 struct nlmsghdr *n, struct tcf_proto *tp,
2145 struct tcf_block *block, struct Qdisc *q,
2146 u32 parent, void *fh, bool *last, bool rtnl_held,
2147 struct netlink_ext_ack *extack)
2149 struct sk_buff *skb;
2150 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2151 int err;
2153 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2154 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2156 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2157 if (!skb)
2158 return -ENOBUFS;
2160 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2161 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2162 false, rtnl_held, extack) <= 0) {
2163 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2164 kfree_skb(skb);
2165 return -EINVAL;
2168 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2169 if (err) {
2170 kfree_skb(skb);
2171 return err;
2174 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2175 n->nlmsg_flags & NLM_F_ECHO);
2176 if (err < 0)
2177 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2179 return err;
2182 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2183 struct tcf_block *block, struct Qdisc *q,
2184 u32 parent, struct nlmsghdr *n,
2185 struct tcf_chain *chain, int event,
2186 struct netlink_ext_ack *extack)
2188 struct tcf_proto *tp;
2190 for (tp = tcf_get_next_proto(chain, NULL);
2191 tp; tp = tcf_get_next_proto(chain, tp))
2192 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2193 event, false, true, extack);
2196 static void tfilter_put(struct tcf_proto *tp, void *fh)
2198 if (tp->ops->put && fh)
2199 tp->ops->put(tp, fh);
2202 static bool is_qdisc_ingress(__u32 classid)
2204 return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2207 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2208 struct netlink_ext_ack *extack)
2210 struct net *net = sock_net(skb->sk);
2211 struct nlattr *tca[TCA_MAX + 1];
2212 char name[IFNAMSIZ];
2213 struct tcmsg *t;
2214 u32 protocol;
2215 u32 prio;
2216 bool prio_allocate;
2217 u32 parent;
2218 u32 chain_index;
2219 struct Qdisc *q;
2220 struct tcf_chain_info chain_info;
2221 struct tcf_chain *chain;
2222 struct tcf_block *block;
2223 struct tcf_proto *tp;
2224 unsigned long cl;
2225 void *fh;
2226 int err;
2227 int tp_created;
2228 bool rtnl_held = false;
2229 u32 flags;
2231 replay:
2232 tp_created = 0;
2234 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2235 rtm_tca_policy, extack);
2236 if (err < 0)
2237 return err;
2239 t = nlmsg_data(n);
2240 protocol = TC_H_MIN(t->tcm_info);
2241 prio = TC_H_MAJ(t->tcm_info);
2242 prio_allocate = false;
2243 parent = t->tcm_parent;
2244 tp = NULL;
2245 cl = 0;
2246 block = NULL;
2247 q = NULL;
2248 chain = NULL;
2249 flags = 0;
2251 if (prio == 0) {
2252 /* If no priority is provided by the user,
2253 * we allocate one.
2255 if (n->nlmsg_flags & NLM_F_CREATE) {
2256 prio = TC_H_MAKE(0x80000000U, 0U);
2257 prio_allocate = true;
2258 } else {
2259 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2260 return -ENOENT;
2264 /* Find head of filter chain. */
2266 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2267 if (err)
2268 return err;
2270 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2271 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2272 err = -EINVAL;
2273 goto errout;
2276 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2277 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2278 * type is not specified, classifier is not unlocked.
2280 if (rtnl_held ||
2281 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2282 !tcf_proto_is_unlocked(name)) {
2283 rtnl_held = true;
2284 rtnl_lock();
2287 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2288 if (err)
2289 goto errout;
2291 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2292 extack);
2293 if (IS_ERR(block)) {
2294 err = PTR_ERR(block);
2295 goto errout;
2297 block->classid = parent;
2299 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2300 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2301 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2302 err = -EINVAL;
2303 goto errout;
2305 chain = tcf_chain_get(block, chain_index, true);
2306 if (!chain) {
2307 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2308 err = -ENOMEM;
2309 goto errout;
2312 mutex_lock(&chain->filter_chain_lock);
2313 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2314 prio, prio_allocate, extack);
2315 if (IS_ERR(tp)) {
2316 err = PTR_ERR(tp);
2317 goto errout_locked;
2320 if (tp == NULL) {
2321 struct tcf_proto *tp_new = NULL;
2323 if (chain->flushing) {
2324 err = -EAGAIN;
2325 goto errout_locked;
2328 /* Proto-tcf does not exist, create new one */
2330 if (tca[TCA_KIND] == NULL || !protocol) {
2331 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2332 err = -EINVAL;
2333 goto errout_locked;
2336 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2337 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2338 err = -ENOENT;
2339 goto errout_locked;
2342 if (prio_allocate)
2343 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2344 &chain_info));
2346 mutex_unlock(&chain->filter_chain_lock);
2347 tp_new = tcf_proto_create(name, protocol, prio, chain,
2348 rtnl_held, extack);
2349 if (IS_ERR(tp_new)) {
2350 err = PTR_ERR(tp_new);
2351 goto errout_tp;
2354 tp_created = 1;
2355 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2356 rtnl_held);
2357 if (IS_ERR(tp)) {
2358 err = PTR_ERR(tp);
2359 goto errout_tp;
2361 } else {
2362 mutex_unlock(&chain->filter_chain_lock);
2365 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2366 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2367 err = -EINVAL;
2368 goto errout;
2371 fh = tp->ops->get(tp, t->tcm_handle);
2373 if (!fh) {
2374 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2375 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2376 err = -ENOENT;
2377 goto errout;
2379 } else if (n->nlmsg_flags & NLM_F_EXCL) {
2380 tfilter_put(tp, fh);
2381 NL_SET_ERR_MSG(extack, "Filter already exists");
2382 err = -EEXIST;
2383 goto errout;
2386 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2387 tfilter_put(tp, fh);
2388 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2389 err = -EINVAL;
2390 goto errout;
2393 if (!(n->nlmsg_flags & NLM_F_CREATE))
2394 flags |= TCA_ACT_FLAGS_REPLACE;
2395 if (!rtnl_held)
2396 flags |= TCA_ACT_FLAGS_NO_RTNL;
2397 if (is_qdisc_ingress(parent))
2398 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2399 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2400 flags, extack);
2401 if (err == 0) {
2402 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2403 RTM_NEWTFILTER, false, rtnl_held, extack);
2404 tfilter_put(tp, fh);
2405 tcf_proto_count_usesw(tp, true);
2406 /* q pointer is NULL for shared blocks */
2407 if (q)
2408 q->flags &= ~TCQ_F_CAN_BYPASS;
2411 errout:
2412 if (err && tp_created)
2413 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2414 errout_tp:
2415 if (chain) {
2416 if (tp && !IS_ERR(tp))
2417 tcf_proto_put(tp, rtnl_held, NULL);
2418 if (!tp_created)
2419 tcf_chain_put(chain);
2421 tcf_block_release(q, block, rtnl_held);
2423 if (rtnl_held)
2424 rtnl_unlock();
2426 if (err == -EAGAIN) {
2427 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2428 * of target chain.
2430 rtnl_held = true;
2431 /* Replay the request. */
2432 goto replay;
2434 return err;
2436 errout_locked:
2437 mutex_unlock(&chain->filter_chain_lock);
2438 goto errout;
2441 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2442 struct netlink_ext_ack *extack)
2444 struct net *net = sock_net(skb->sk);
2445 struct nlattr *tca[TCA_MAX + 1];
2446 char name[IFNAMSIZ];
2447 struct tcmsg *t;
2448 u32 protocol;
2449 u32 prio;
2450 u32 parent;
2451 u32 chain_index;
2452 struct Qdisc *q = NULL;
2453 struct tcf_chain_info chain_info;
2454 struct tcf_chain *chain = NULL;
2455 struct tcf_block *block = NULL;
2456 struct tcf_proto *tp = NULL;
2457 unsigned long cl = 0;
2458 void *fh = NULL;
2459 int err;
2460 bool rtnl_held = false;
2462 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2463 rtm_tca_policy, extack);
2464 if (err < 0)
2465 return err;
2467 t = nlmsg_data(n);
2468 protocol = TC_H_MIN(t->tcm_info);
2469 prio = TC_H_MAJ(t->tcm_info);
2470 parent = t->tcm_parent;
2472 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2473 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2474 return -ENOENT;
2477 /* Find head of filter chain. */
2479 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2480 if (err)
2481 return err;
2483 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2484 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2485 err = -EINVAL;
2486 goto errout;
2488 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2489 * found), qdisc is not unlocked, classifier type is not specified,
2490 * classifier is not unlocked.
2492 if (!prio ||
2493 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2494 !tcf_proto_is_unlocked(name)) {
2495 rtnl_held = true;
2496 rtnl_lock();
2499 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2500 if (err)
2501 goto errout;
2503 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2504 extack);
2505 if (IS_ERR(block)) {
2506 err = PTR_ERR(block);
2507 goto errout;
2510 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2511 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2512 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2513 err = -EINVAL;
2514 goto errout;
2516 chain = tcf_chain_get(block, chain_index, false);
2517 if (!chain) {
2518 /* User requested flush on non-existent chain. Nothing to do,
2519 * so just return success.
2521 if (prio == 0) {
2522 err = 0;
2523 goto errout;
2525 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2526 err = -ENOENT;
2527 goto errout;
2530 if (prio == 0) {
2531 tfilter_notify_chain(net, skb, block, q, parent, n,
2532 chain, RTM_DELTFILTER, extack);
2533 tcf_chain_flush(chain, rtnl_held);
2534 err = 0;
2535 goto errout;
2538 mutex_lock(&chain->filter_chain_lock);
2539 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2540 prio, false, extack);
2541 if (!tp) {
2542 err = -ENOENT;
2543 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2544 goto errout_locked;
2545 } else if (IS_ERR(tp)) {
2546 err = PTR_ERR(tp);
2547 goto errout_locked;
2548 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2549 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2550 err = -EINVAL;
2551 goto errout_locked;
2552 } else if (t->tcm_handle == 0) {
2553 tcf_proto_signal_destroying(chain, tp);
2554 tcf_chain_tp_remove(chain, &chain_info, tp);
2555 mutex_unlock(&chain->filter_chain_lock);
2557 tcf_proto_put(tp, rtnl_held, NULL);
2558 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2559 RTM_DELTFILTER, false, rtnl_held, extack);
2560 err = 0;
2561 goto errout;
2563 mutex_unlock(&chain->filter_chain_lock);
2565 fh = tp->ops->get(tp, t->tcm_handle);
2567 if (!fh) {
2568 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2569 err = -ENOENT;
2570 } else {
2571 bool last;
2573 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2574 &last, rtnl_held, extack);
2576 if (err)
2577 goto errout;
2578 if (last)
2579 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2582 errout:
2583 if (chain) {
2584 if (tp && !IS_ERR(tp))
2585 tcf_proto_put(tp, rtnl_held, NULL);
2586 tcf_chain_put(chain);
2588 tcf_block_release(q, block, rtnl_held);
2590 if (rtnl_held)
2591 rtnl_unlock();
2593 return err;
2595 errout_locked:
2596 mutex_unlock(&chain->filter_chain_lock);
2597 goto errout;
2600 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2601 struct netlink_ext_ack *extack)
2603 struct net *net = sock_net(skb->sk);
2604 struct nlattr *tca[TCA_MAX + 1];
2605 char name[IFNAMSIZ];
2606 struct tcmsg *t;
2607 u32 protocol;
2608 u32 prio;
2609 u32 parent;
2610 u32 chain_index;
2611 struct Qdisc *q = NULL;
2612 struct tcf_chain_info chain_info;
2613 struct tcf_chain *chain = NULL;
2614 struct tcf_block *block = NULL;
2615 struct tcf_proto *tp = NULL;
2616 unsigned long cl = 0;
2617 void *fh = NULL;
2618 int err;
2619 bool rtnl_held = false;
2621 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2622 rtm_tca_policy, extack);
2623 if (err < 0)
2624 return err;
2626 t = nlmsg_data(n);
2627 protocol = TC_H_MIN(t->tcm_info);
2628 prio = TC_H_MAJ(t->tcm_info);
2629 parent = t->tcm_parent;
2631 if (prio == 0) {
2632 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2633 return -ENOENT;
2636 /* Find head of filter chain. */
2638 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2639 if (err)
2640 return err;
2642 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2643 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2644 err = -EINVAL;
2645 goto errout;
2647 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2648 * unlocked, classifier type is not specified, classifier is not
2649 * unlocked.
2651 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2652 !tcf_proto_is_unlocked(name)) {
2653 rtnl_held = true;
2654 rtnl_lock();
2657 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2658 if (err)
2659 goto errout;
2661 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2662 extack);
2663 if (IS_ERR(block)) {
2664 err = PTR_ERR(block);
2665 goto errout;
2668 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2669 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2670 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2671 err = -EINVAL;
2672 goto errout;
2674 chain = tcf_chain_get(block, chain_index, false);
2675 if (!chain) {
2676 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2677 err = -EINVAL;
2678 goto errout;
2681 mutex_lock(&chain->filter_chain_lock);
2682 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2683 prio, false, extack);
2684 mutex_unlock(&chain->filter_chain_lock);
2685 if (!tp) {
2686 err = -ENOENT;
2687 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2688 goto errout;
2689 } else if (IS_ERR(tp)) {
2690 err = PTR_ERR(tp);
2691 goto errout;
2692 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2693 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2694 err = -EINVAL;
2695 goto errout;
2698 fh = tp->ops->get(tp, t->tcm_handle);
2700 if (!fh) {
2701 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2702 err = -ENOENT;
2703 } else {
2704 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2705 fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2706 if (err < 0)
2707 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2710 tfilter_put(tp, fh);
2711 errout:
2712 if (chain) {
2713 if (tp && !IS_ERR(tp))
2714 tcf_proto_put(tp, rtnl_held, NULL);
2715 tcf_chain_put(chain);
2717 tcf_block_release(q, block, rtnl_held);
2719 if (rtnl_held)
2720 rtnl_unlock();
2722 return err;
2725 struct tcf_dump_args {
2726 struct tcf_walker w;
2727 struct sk_buff *skb;
2728 struct netlink_callback *cb;
2729 struct tcf_block *block;
2730 struct Qdisc *q;
2731 u32 parent;
2732 bool terse_dump;
2735 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2737 struct tcf_dump_args *a = (void *)arg;
2738 struct net *net = sock_net(a->skb->sk);
2740 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2741 n, NETLINK_CB(a->cb->skb).portid,
2742 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2743 RTM_NEWTFILTER, a->terse_dump, true, NULL);
2746 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2747 struct sk_buff *skb, struct netlink_callback *cb,
2748 long index_start, long *p_index, bool terse)
2750 struct net *net = sock_net(skb->sk);
2751 struct tcf_block *block = chain->block;
2752 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2753 struct tcf_proto *tp, *tp_prev;
2754 struct tcf_dump_args arg;
2756 for (tp = __tcf_get_next_proto(chain, NULL);
2758 tp_prev = tp,
2759 tp = __tcf_get_next_proto(chain, tp),
2760 tcf_proto_put(tp_prev, true, NULL),
2761 (*p_index)++) {
2762 if (*p_index < index_start)
2763 continue;
2764 if (TC_H_MAJ(tcm->tcm_info) &&
2765 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2766 continue;
2767 if (TC_H_MIN(tcm->tcm_info) &&
2768 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2769 continue;
2770 if (*p_index > index_start)
2771 memset(&cb->args[1], 0,
2772 sizeof(cb->args) - sizeof(cb->args[0]));
2773 if (cb->args[1] == 0) {
2774 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2775 NETLINK_CB(cb->skb).portid,
2776 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2777 RTM_NEWTFILTER, false, true, NULL) <= 0)
2778 goto errout;
2779 cb->args[1] = 1;
2781 if (!tp->ops->walk)
2782 continue;
2783 arg.w.fn = tcf_node_dump;
2784 arg.skb = skb;
2785 arg.cb = cb;
2786 arg.block = block;
2787 arg.q = q;
2788 arg.parent = parent;
2789 arg.w.stop = 0;
2790 arg.w.skip = cb->args[1] - 1;
2791 arg.w.count = 0;
2792 arg.w.cookie = cb->args[2];
2793 arg.terse_dump = terse;
2794 tp->ops->walk(tp, &arg.w, true);
2795 cb->args[2] = arg.w.cookie;
2796 cb->args[1] = arg.w.count + 1;
2797 if (arg.w.stop)
2798 goto errout;
2800 return true;
2802 errout:
2803 tcf_proto_put(tp, true, NULL);
2804 return false;
2807 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2808 [TCA_CHAIN] = { .type = NLA_U32 },
2809 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2812 /* called with RTNL */
2813 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2815 struct tcf_chain *chain, *chain_prev;
2816 struct net *net = sock_net(skb->sk);
2817 struct nlattr *tca[TCA_MAX + 1];
2818 struct Qdisc *q = NULL;
2819 struct tcf_block *block;
2820 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2821 bool terse_dump = false;
2822 long index_start;
2823 long index;
2824 u32 parent;
2825 int err;
2827 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2828 return skb->len;
2830 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2831 tcf_tfilter_dump_policy, cb->extack);
2832 if (err)
2833 return err;
2835 if (tca[TCA_DUMP_FLAGS]) {
2836 struct nla_bitfield32 flags =
2837 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2839 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2842 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2843 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2844 if (!block)
2845 goto out;
2846 /* If we work with block index, q is NULL and parent value
2847 * will never be used in the following code. The check
2848 * in tcf_fill_node prevents it. However, compiler does not
2849 * see that far, so set parent to zero to silence the warning
2850 * about parent being uninitialized.
2852 parent = 0;
2853 } else {
2854 const struct Qdisc_class_ops *cops;
2855 struct net_device *dev;
2856 unsigned long cl = 0;
2858 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2859 if (!dev)
2860 return skb->len;
2862 parent = tcm->tcm_parent;
2863 if (!parent)
2864 q = rtnl_dereference(dev->qdisc);
2865 else
2866 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2867 if (!q)
2868 goto out;
2869 cops = q->ops->cl_ops;
2870 if (!cops)
2871 goto out;
2872 if (!cops->tcf_block)
2873 goto out;
2874 if (TC_H_MIN(tcm->tcm_parent)) {
2875 cl = cops->find(q, tcm->tcm_parent);
2876 if (cl == 0)
2877 goto out;
2879 block = cops->tcf_block(q, cl, NULL);
2880 if (!block)
2881 goto out;
2882 parent = block->classid;
2883 if (tcf_block_shared(block))
2884 q = NULL;
2887 index_start = cb->args[0];
2888 index = 0;
2890 for (chain = __tcf_get_next_chain(block, NULL);
2891 chain;
2892 chain_prev = chain,
2893 chain = __tcf_get_next_chain(block, chain),
2894 tcf_chain_put(chain_prev)) {
2895 if (tca[TCA_CHAIN] &&
2896 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2897 continue;
2898 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2899 index_start, &index, terse_dump)) {
2900 tcf_chain_put(chain);
2901 err = -EMSGSIZE;
2902 break;
2906 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2907 tcf_block_refcnt_put(block, true);
2908 cb->args[0] = index;
2910 out:
2911 /* If we did no progress, the error (EMSGSIZE) is real */
2912 if (skb->len == 0 && err)
2913 return err;
2914 return skb->len;
2917 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2918 void *tmplt_priv, u32 chain_index,
2919 struct net *net, struct sk_buff *skb,
2920 struct tcf_block *block,
2921 u32 portid, u32 seq, u16 flags, int event,
2922 struct netlink_ext_ack *extack)
2924 unsigned char *b = skb_tail_pointer(skb);
2925 const struct tcf_proto_ops *ops;
2926 struct nlmsghdr *nlh;
2927 struct tcmsg *tcm;
2928 void *priv;
2930 ops = tmplt_ops;
2931 priv = tmplt_priv;
2933 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2934 if (!nlh)
2935 goto out_nlmsg_trim;
2936 tcm = nlmsg_data(nlh);
2937 tcm->tcm_family = AF_UNSPEC;
2938 tcm->tcm__pad1 = 0;
2939 tcm->tcm__pad2 = 0;
2940 tcm->tcm_handle = 0;
2941 if (block->q) {
2942 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2943 tcm->tcm_parent = block->q->handle;
2944 } else {
2945 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2946 tcm->tcm_block_index = block->index;
2949 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2950 goto nla_put_failure;
2952 if (ops) {
2953 if (nla_put_string(skb, TCA_KIND, ops->kind))
2954 goto nla_put_failure;
2955 if (ops->tmplt_dump(skb, net, priv) < 0)
2956 goto nla_put_failure;
2959 if (extack && extack->_msg &&
2960 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2961 goto out_nlmsg_trim;
2963 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2965 return skb->len;
2967 out_nlmsg_trim:
2968 nla_put_failure:
2969 nlmsg_trim(skb, b);
2970 return -EMSGSIZE;
2973 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2974 u32 seq, u16 flags, int event, bool unicast,
2975 struct netlink_ext_ack *extack)
2977 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2978 struct tcf_block *block = chain->block;
2979 struct net *net = block->net;
2980 struct sk_buff *skb;
2981 int err = 0;
2983 if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2984 return 0;
2986 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2987 if (!skb)
2988 return -ENOBUFS;
2990 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2991 chain->index, net, skb, block, portid,
2992 seq, flags, event, extack) <= 0) {
2993 kfree_skb(skb);
2994 return -EINVAL;
2997 if (unicast)
2998 err = rtnl_unicast(skb, net, portid);
2999 else
3000 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
3001 flags & NLM_F_ECHO);
3003 return err;
3006 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
3007 void *tmplt_priv, u32 chain_index,
3008 struct tcf_block *block, struct sk_buff *oskb,
3009 u32 seq, u16 flags)
3011 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3012 struct net *net = block->net;
3013 struct sk_buff *skb;
3015 if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
3016 return 0;
3018 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3019 if (!skb)
3020 return -ENOBUFS;
3022 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
3023 block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
3024 kfree_skb(skb);
3025 return -EINVAL;
3028 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
3031 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
3032 struct nlattr **tca,
3033 struct netlink_ext_ack *extack)
3035 const struct tcf_proto_ops *ops;
3036 char name[IFNAMSIZ];
3037 void *tmplt_priv;
3039 /* If kind is not set, user did not specify template. */
3040 if (!tca[TCA_KIND])
3041 return 0;
3043 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3044 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3045 return -EINVAL;
3048 ops = tcf_proto_lookup_ops(name, true, extack);
3049 if (IS_ERR(ops))
3050 return PTR_ERR(ops);
3051 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3052 !ops->tmplt_reoffload) {
3053 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3054 module_put(ops->owner);
3055 return -EOPNOTSUPP;
3058 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3059 if (IS_ERR(tmplt_priv)) {
3060 module_put(ops->owner);
3061 return PTR_ERR(tmplt_priv);
3063 chain->tmplt_ops = ops;
3064 chain->tmplt_priv = tmplt_priv;
3065 return 0;
3068 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3069 void *tmplt_priv)
3071 /* If template ops are set, no work to do for us. */
3072 if (!tmplt_ops)
3073 return;
3075 tmplt_ops->tmplt_destroy(tmplt_priv);
3076 module_put(tmplt_ops->owner);
3079 /* Add/delete/get a chain */
3081 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3082 struct netlink_ext_ack *extack)
3084 struct net *net = sock_net(skb->sk);
3085 struct nlattr *tca[TCA_MAX + 1];
3086 struct tcmsg *t;
3087 u32 parent;
3088 u32 chain_index;
3089 struct Qdisc *q;
3090 struct tcf_chain *chain;
3091 struct tcf_block *block;
3092 unsigned long cl;
3093 int err;
3095 replay:
3096 q = NULL;
3097 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3098 rtm_tca_policy, extack);
3099 if (err < 0)
3100 return err;
3102 t = nlmsg_data(n);
3103 parent = t->tcm_parent;
3104 cl = 0;
3106 block = tcf_block_find(net, &q, &parent, &cl,
3107 t->tcm_ifindex, t->tcm_block_index, extack);
3108 if (IS_ERR(block))
3109 return PTR_ERR(block);
3111 chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
3112 if (chain_index > TC_ACT_EXT_VAL_MASK) {
3113 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3114 err = -EINVAL;
3115 goto errout_block;
3118 mutex_lock(&block->lock);
3119 chain = tcf_chain_lookup(block, chain_index);
3120 if (n->nlmsg_type == RTM_NEWCHAIN) {
3121 if (chain) {
3122 if (tcf_chain_held_by_acts_only(chain)) {
3123 /* The chain exists only because there is
3124 * some action referencing it.
3126 tcf_chain_hold(chain);
3127 } else {
3128 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3129 err = -EEXIST;
3130 goto errout_block_locked;
3132 } else {
3133 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3134 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3135 err = -ENOENT;
3136 goto errout_block_locked;
3138 chain = tcf_chain_create(block, chain_index);
3139 if (!chain) {
3140 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3141 err = -ENOMEM;
3142 goto errout_block_locked;
3145 } else {
3146 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3147 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3148 err = -EINVAL;
3149 goto errout_block_locked;
3151 tcf_chain_hold(chain);
3154 if (n->nlmsg_type == RTM_NEWCHAIN) {
3155 /* Modifying chain requires holding parent block lock. In case
3156 * the chain was successfully added, take a reference to the
3157 * chain. This ensures that an empty chain does not disappear at
3158 * the end of this function.
3160 tcf_chain_hold(chain);
3161 chain->explicitly_created = true;
3163 mutex_unlock(&block->lock);
3165 switch (n->nlmsg_type) {
3166 case RTM_NEWCHAIN:
3167 err = tc_chain_tmplt_add(chain, net, tca, extack);
3168 if (err) {
3169 tcf_chain_put_explicitly_created(chain);
3170 goto errout;
3173 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3174 RTM_NEWCHAIN, false, extack);
3175 break;
3176 case RTM_DELCHAIN:
3177 tfilter_notify_chain(net, skb, block, q, parent, n,
3178 chain, RTM_DELTFILTER, extack);
3179 /* Flush the chain first as the user requested chain removal. */
3180 tcf_chain_flush(chain, true);
3181 /* In case the chain was successfully deleted, put a reference
3182 * to the chain previously taken during addition.
3184 tcf_chain_put_explicitly_created(chain);
3185 break;
3186 case RTM_GETCHAIN:
3187 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3188 n->nlmsg_flags, n->nlmsg_type, true, extack);
3189 if (err < 0)
3190 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3191 break;
3192 default:
3193 err = -EOPNOTSUPP;
3194 NL_SET_ERR_MSG(extack, "Unsupported message type");
3195 goto errout;
3198 errout:
3199 tcf_chain_put(chain);
3200 errout_block:
3201 tcf_block_release(q, block, true);
3202 if (err == -EAGAIN)
3203 /* Replay the request. */
3204 goto replay;
3205 return err;
3207 errout_block_locked:
3208 mutex_unlock(&block->lock);
3209 goto errout_block;
3212 /* called with RTNL */
3213 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3215 struct net *net = sock_net(skb->sk);
3216 struct nlattr *tca[TCA_MAX + 1];
3217 struct Qdisc *q = NULL;
3218 struct tcf_block *block;
3219 struct tcmsg *tcm = nlmsg_data(cb->nlh);
3220 struct tcf_chain *chain;
3221 long index_start;
3222 long index;
3223 int err;
3225 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3226 return skb->len;
3228 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3229 rtm_tca_policy, cb->extack);
3230 if (err)
3231 return err;
3233 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3234 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3235 if (!block)
3236 goto out;
3237 } else {
3238 const struct Qdisc_class_ops *cops;
3239 struct net_device *dev;
3240 unsigned long cl = 0;
3242 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3243 if (!dev)
3244 return skb->len;
3246 if (!tcm->tcm_parent)
3247 q = rtnl_dereference(dev->qdisc);
3248 else
3249 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3251 if (!q)
3252 goto out;
3253 cops = q->ops->cl_ops;
3254 if (!cops)
3255 goto out;
3256 if (!cops->tcf_block)
3257 goto out;
3258 if (TC_H_MIN(tcm->tcm_parent)) {
3259 cl = cops->find(q, tcm->tcm_parent);
3260 if (cl == 0)
3261 goto out;
3263 block = cops->tcf_block(q, cl, NULL);
3264 if (!block)
3265 goto out;
3266 if (tcf_block_shared(block))
3267 q = NULL;
3270 index_start = cb->args[0];
3271 index = 0;
3273 mutex_lock(&block->lock);
3274 list_for_each_entry(chain, &block->chain_list, list) {
3275 if ((tca[TCA_CHAIN] &&
3276 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3277 continue;
3278 if (index < index_start) {
3279 index++;
3280 continue;
3282 if (tcf_chain_held_by_acts_only(chain))
3283 continue;
3284 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3285 chain->index, net, skb, block,
3286 NETLINK_CB(cb->skb).portid,
3287 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3288 RTM_NEWCHAIN, NULL);
3289 if (err <= 0)
3290 break;
3291 index++;
3293 mutex_unlock(&block->lock);
3295 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3296 tcf_block_refcnt_put(block, true);
3297 cb->args[0] = index;
3299 out:
3300 /* If we did no progress, the error (EMSGSIZE) is real */
3301 if (skb->len == 0 && err)
3302 return err;
3303 return skb->len;
3306 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3307 int police, struct tcf_proto *tp, u32 handle,
3308 bool use_action_miss)
3310 int err = 0;
3312 #ifdef CONFIG_NET_CLS_ACT
3313 exts->type = 0;
3314 exts->nr_actions = 0;
3315 exts->miss_cookie_node = NULL;
3316 /* Note: we do not own yet a reference on net.
3317 * This reference might be taken later from tcf_exts_get_net().
3319 exts->net = net;
3320 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3321 GFP_KERNEL);
3322 if (!exts->actions)
3323 return -ENOMEM;
3324 #endif
3326 exts->action = action;
3327 exts->police = police;
3329 if (!use_action_miss)
3330 return 0;
3332 err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3333 if (err)
3334 goto err_miss_alloc;
3336 return 0;
3338 err_miss_alloc:
3339 tcf_exts_destroy(exts);
3340 #ifdef CONFIG_NET_CLS_ACT
3341 exts->actions = NULL;
3342 #endif
3343 return err;
3345 EXPORT_SYMBOL(tcf_exts_init_ex);
3347 void tcf_exts_destroy(struct tcf_exts *exts)
3349 tcf_exts_miss_cookie_base_destroy(exts);
3351 #ifdef CONFIG_NET_CLS_ACT
3352 if (exts->actions) {
3353 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3354 kfree(exts->actions);
3356 exts->nr_actions = 0;
3357 #endif
3359 EXPORT_SYMBOL(tcf_exts_destroy);
3361 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3362 struct nlattr *rate_tlv, struct tcf_exts *exts,
3363 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3365 #ifdef CONFIG_NET_CLS_ACT
3367 int init_res[TCA_ACT_MAX_PRIO] = {};
3368 struct tc_action *act;
3369 size_t attr_size = 0;
3371 if (exts->police && tb[exts->police]) {
3372 struct tc_action_ops *a_o;
3374 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3375 a_o = tc_action_load_ops(tb[exts->police], flags,
3376 extack);
3377 if (IS_ERR(a_o))
3378 return PTR_ERR(a_o);
3379 act = tcf_action_init_1(net, tp, tb[exts->police],
3380 rate_tlv, a_o, init_res, flags,
3381 extack);
3382 module_put(a_o->owner);
3383 if (IS_ERR(act))
3384 return PTR_ERR(act);
3386 act->type = exts->type = TCA_OLD_COMPAT;
3387 exts->actions[0] = act;
3388 exts->nr_actions = 1;
3389 tcf_idr_insert_many(exts->actions, init_res);
3390 } else if (exts->action && tb[exts->action]) {
3391 int err;
3393 flags |= TCA_ACT_FLAGS_BIND;
3394 err = tcf_action_init(net, tp, tb[exts->action],
3395 rate_tlv, exts->actions, init_res,
3396 &attr_size, flags, fl_flags,
3397 extack);
3398 if (err < 0)
3399 return err;
3400 exts->nr_actions = err;
3403 #else
3404 if ((exts->action && tb[exts->action]) ||
3405 (exts->police && tb[exts->police])) {
3406 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3407 return -EOPNOTSUPP;
3409 #endif
3411 return 0;
3413 EXPORT_SYMBOL(tcf_exts_validate_ex);
3415 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3416 struct nlattr *rate_tlv, struct tcf_exts *exts,
3417 u32 flags, struct netlink_ext_ack *extack)
3419 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3420 flags, 0, extack);
3422 EXPORT_SYMBOL(tcf_exts_validate);
3424 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3426 #ifdef CONFIG_NET_CLS_ACT
3427 struct tcf_exts old = *dst;
3429 *dst = *src;
3430 tcf_exts_destroy(&old);
3431 #endif
3433 EXPORT_SYMBOL(tcf_exts_change);
3435 #ifdef CONFIG_NET_CLS_ACT
3436 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3438 if (exts->nr_actions == 0)
3439 return NULL;
3440 else
3441 return exts->actions[0];
3443 #endif
3445 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3447 #ifdef CONFIG_NET_CLS_ACT
3448 struct nlattr *nest;
3450 if (exts->action && tcf_exts_has_actions(exts)) {
3452 * again for backward compatible mode - we want
3453 * to work with both old and new modes of entering
3454 * tc data even if iproute2 was newer - jhs
3456 if (exts->type != TCA_OLD_COMPAT) {
3457 nest = nla_nest_start_noflag(skb, exts->action);
3458 if (nest == NULL)
3459 goto nla_put_failure;
3461 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3462 < 0)
3463 goto nla_put_failure;
3464 nla_nest_end(skb, nest);
3465 } else if (exts->police) {
3466 struct tc_action *act = tcf_exts_first_act(exts);
3467 nest = nla_nest_start_noflag(skb, exts->police);
3468 if (nest == NULL || !act)
3469 goto nla_put_failure;
3470 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3471 goto nla_put_failure;
3472 nla_nest_end(skb, nest);
3475 return 0;
3477 nla_put_failure:
3478 nla_nest_cancel(skb, nest);
3479 return -1;
3480 #else
3481 return 0;
3482 #endif
3484 EXPORT_SYMBOL(tcf_exts_dump);
3486 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3488 #ifdef CONFIG_NET_CLS_ACT
3489 struct nlattr *nest;
3491 if (!exts->action || !tcf_exts_has_actions(exts))
3492 return 0;
3494 nest = nla_nest_start_noflag(skb, exts->action);
3495 if (!nest)
3496 goto nla_put_failure;
3498 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3499 goto nla_put_failure;
3500 nla_nest_end(skb, nest);
3501 return 0;
3503 nla_put_failure:
3504 nla_nest_cancel(skb, nest);
3505 return -1;
3506 #else
3507 return 0;
3508 #endif
3510 EXPORT_SYMBOL(tcf_exts_terse_dump);
3512 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3514 #ifdef CONFIG_NET_CLS_ACT
3515 struct tc_action *a = tcf_exts_first_act(exts);
3516 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3517 return -1;
3518 #endif
3519 return 0;
3521 EXPORT_SYMBOL(tcf_exts_dump_stats);
3523 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3525 if (*flags & TCA_CLS_FLAGS_IN_HW)
3526 return;
3527 *flags |= TCA_CLS_FLAGS_IN_HW;
3528 atomic_inc(&block->offloadcnt);
3531 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3533 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3534 return;
3535 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3536 atomic_dec(&block->offloadcnt);
3539 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3540 struct tcf_proto *tp, u32 *cnt,
3541 u32 *flags, u32 diff, bool add)
3543 lockdep_assert_held(&block->cb_lock);
3545 spin_lock(&tp->lock);
3546 if (add) {
3547 if (!*cnt)
3548 tcf_block_offload_inc(block, flags);
3549 *cnt += diff;
3550 } else {
3551 *cnt -= diff;
3552 if (!*cnt)
3553 tcf_block_offload_dec(block, flags);
3555 spin_unlock(&tp->lock);
3558 static void
3559 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3560 u32 *cnt, u32 *flags)
3562 lockdep_assert_held(&block->cb_lock);
3564 spin_lock(&tp->lock);
3565 tcf_block_offload_dec(block, flags);
3566 *cnt = 0;
3567 spin_unlock(&tp->lock);
3570 static int
3571 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3572 void *type_data, bool err_stop)
3574 struct flow_block_cb *block_cb;
3575 int ok_count = 0;
3576 int err;
3578 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3579 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3580 if (err) {
3581 if (err_stop)
3582 return err;
3583 } else {
3584 ok_count++;
3587 return ok_count;
3590 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3591 void *type_data, bool err_stop, bool rtnl_held)
3593 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3594 int ok_count;
3596 retry:
3597 if (take_rtnl)
3598 rtnl_lock();
3599 down_read(&block->cb_lock);
3600 /* Need to obtain rtnl lock if block is bound to devs that require it.
3601 * In block bind code cb_lock is obtained while holding rtnl, so we must
3602 * obtain the locks in same order here.
3604 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3605 up_read(&block->cb_lock);
3606 take_rtnl = true;
3607 goto retry;
3610 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3612 up_read(&block->cb_lock);
3613 if (take_rtnl)
3614 rtnl_unlock();
3615 return ok_count;
3617 EXPORT_SYMBOL(tc_setup_cb_call);
3619 /* Non-destructive filter add. If filter that wasn't already in hardware is
3620 * successfully offloaded, increment block offloads counter. On failure,
3621 * previously offloaded filter is considered to be intact and offloads counter
3622 * is not decremented.
3625 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3626 enum tc_setup_type type, void *type_data, bool err_stop,
3627 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3629 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3630 int ok_count;
3632 retry:
3633 if (take_rtnl)
3634 rtnl_lock();
3635 down_read(&block->cb_lock);
3636 /* Need to obtain rtnl lock if block is bound to devs that require it.
3637 * In block bind code cb_lock is obtained while holding rtnl, so we must
3638 * obtain the locks in same order here.
3640 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3641 up_read(&block->cb_lock);
3642 take_rtnl = true;
3643 goto retry;
3646 /* Make sure all netdevs sharing this block are offload-capable. */
3647 if (block->nooffloaddevcnt && err_stop) {
3648 ok_count = -EOPNOTSUPP;
3649 goto err_unlock;
3652 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3653 if (ok_count < 0)
3654 goto err_unlock;
3656 if (tp->ops->hw_add)
3657 tp->ops->hw_add(tp, type_data);
3658 if (ok_count > 0)
3659 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3660 ok_count, true);
3661 err_unlock:
3662 up_read(&block->cb_lock);
3663 if (take_rtnl)
3664 rtnl_unlock();
3665 return min(ok_count, 0);
3667 EXPORT_SYMBOL(tc_setup_cb_add);
3669 /* Destructive filter replace. If filter that wasn't already in hardware is
3670 * successfully offloaded, increment block offload counter. On failure,
3671 * previously offloaded filter is considered to be destroyed and offload counter
3672 * is decremented.
3675 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3676 enum tc_setup_type type, void *type_data, bool err_stop,
3677 u32 *old_flags, unsigned int *old_in_hw_count,
3678 u32 *new_flags, unsigned int *new_in_hw_count,
3679 bool rtnl_held)
3681 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3682 int ok_count;
3684 retry:
3685 if (take_rtnl)
3686 rtnl_lock();
3687 down_read(&block->cb_lock);
3688 /* Need to obtain rtnl lock if block is bound to devs that require it.
3689 * In block bind code cb_lock is obtained while holding rtnl, so we must
3690 * obtain the locks in same order here.
3692 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3693 up_read(&block->cb_lock);
3694 take_rtnl = true;
3695 goto retry;
3698 /* Make sure all netdevs sharing this block are offload-capable. */
3699 if (block->nooffloaddevcnt && err_stop) {
3700 ok_count = -EOPNOTSUPP;
3701 goto err_unlock;
3704 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3705 if (tp->ops->hw_del)
3706 tp->ops->hw_del(tp, type_data);
3708 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3709 if (ok_count < 0)
3710 goto err_unlock;
3712 if (tp->ops->hw_add)
3713 tp->ops->hw_add(tp, type_data);
3714 if (ok_count > 0)
3715 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3716 new_flags, ok_count, true);
3717 err_unlock:
3718 up_read(&block->cb_lock);
3719 if (take_rtnl)
3720 rtnl_unlock();
3721 return min(ok_count, 0);
3723 EXPORT_SYMBOL(tc_setup_cb_replace);
3725 /* Destroy filter and decrement block offload counter, if filter was previously
3726 * offloaded.
3729 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3730 enum tc_setup_type type, void *type_data, bool err_stop,
3731 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3733 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3734 int ok_count;
3736 retry:
3737 if (take_rtnl)
3738 rtnl_lock();
3739 down_read(&block->cb_lock);
3740 /* Need to obtain rtnl lock if block is bound to devs that require it.
3741 * In block bind code cb_lock is obtained while holding rtnl, so we must
3742 * obtain the locks in same order here.
3744 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3745 up_read(&block->cb_lock);
3746 take_rtnl = true;
3747 goto retry;
3750 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3752 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3753 if (tp->ops->hw_del)
3754 tp->ops->hw_del(tp, type_data);
3756 up_read(&block->cb_lock);
3757 if (take_rtnl)
3758 rtnl_unlock();
3759 return min(ok_count, 0);
3761 EXPORT_SYMBOL(tc_setup_cb_destroy);
3763 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3764 bool add, flow_setup_cb_t *cb,
3765 enum tc_setup_type type, void *type_data,
3766 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3768 int err = cb(type, type_data, cb_priv);
3770 if (err) {
3771 if (add && tc_skip_sw(*flags))
3772 return err;
3773 } else {
3774 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3775 add);
3778 return 0;
3780 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3782 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3783 const struct tc_action *act)
3785 struct tc_cookie *user_cookie;
3786 int err = 0;
3788 rcu_read_lock();
3789 user_cookie = rcu_dereference(act->user_cookie);
3790 if (user_cookie) {
3791 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3792 user_cookie->len,
3793 GFP_ATOMIC);
3794 if (!entry->user_cookie)
3795 err = -ENOMEM;
3797 rcu_read_unlock();
3798 return err;
3801 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3803 flow_action_cookie_destroy(entry->user_cookie);
3806 void tc_cleanup_offload_action(struct flow_action *flow_action)
3808 struct flow_action_entry *entry;
3809 int i;
3811 flow_action_for_each(i, entry, flow_action) {
3812 tcf_act_put_user_cookie(entry);
3813 if (entry->destructor)
3814 entry->destructor(entry->destructor_priv);
3817 EXPORT_SYMBOL(tc_cleanup_offload_action);
3819 static int tc_setup_offload_act(struct tc_action *act,
3820 struct flow_action_entry *entry,
3821 u32 *index_inc,
3822 struct netlink_ext_ack *extack)
3824 #ifdef CONFIG_NET_CLS_ACT
3825 if (act->ops->offload_act_setup) {
3826 return act->ops->offload_act_setup(act, entry, index_inc, true,
3827 extack);
3828 } else {
3829 NL_SET_ERR_MSG(extack, "Action does not support offload");
3830 return -EOPNOTSUPP;
3832 #else
3833 return 0;
3834 #endif
3837 int tc_setup_action(struct flow_action *flow_action,
3838 struct tc_action *actions[],
3839 u32 miss_cookie_base,
3840 struct netlink_ext_ack *extack)
3842 int i, j, k, index, err = 0;
3843 struct tc_action *act;
3845 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3846 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3847 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3849 if (!actions)
3850 return 0;
3852 j = 0;
3853 tcf_act_for_each_action(i, act, actions) {
3854 struct flow_action_entry *entry;
3856 entry = &flow_action->entries[j];
3857 spin_lock_bh(&act->tcfa_lock);
3858 err = tcf_act_get_user_cookie(entry, act);
3859 if (err)
3860 goto err_out_locked;
3862 index = 0;
3863 err = tc_setup_offload_act(act, entry, &index, extack);
3864 if (err)
3865 goto err_out_locked;
3867 for (k = 0; k < index ; k++) {
3868 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3869 entry[k].hw_index = act->tcfa_index;
3870 entry[k].cookie = (unsigned long)act;
3871 entry[k].miss_cookie =
3872 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3875 j += index;
3877 spin_unlock_bh(&act->tcfa_lock);
3880 err_out:
3881 if (err)
3882 tc_cleanup_offload_action(flow_action);
3884 return err;
3885 err_out_locked:
3886 spin_unlock_bh(&act->tcfa_lock);
3887 goto err_out;
3890 int tc_setup_offload_action(struct flow_action *flow_action,
3891 const struct tcf_exts *exts,
3892 struct netlink_ext_ack *extack)
3894 #ifdef CONFIG_NET_CLS_ACT
3895 u32 miss_cookie_base;
3897 if (!exts)
3898 return 0;
3900 miss_cookie_base = exts->miss_cookie_node ?
3901 exts->miss_cookie_node->miss_cookie_base : 0;
3902 return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3903 extack);
3904 #else
3905 return 0;
3906 #endif
3908 EXPORT_SYMBOL(tc_setup_offload_action);
3910 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3912 unsigned int num_acts = 0;
3913 struct tc_action *act;
3914 int i;
3916 tcf_exts_for_each_action(i, act, exts) {
3917 if (is_tcf_pedit(act))
3918 num_acts += tcf_pedit_nkeys(act);
3919 else
3920 num_acts++;
3922 return num_acts;
3924 EXPORT_SYMBOL(tcf_exts_num_actions);
3926 #ifdef CONFIG_NET_CLS_ACT
3927 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3928 u32 *p_block_index,
3929 struct netlink_ext_ack *extack)
3931 *p_block_index = nla_get_u32(block_index_attr);
3932 if (!*p_block_index) {
3933 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3934 return -EINVAL;
3937 return 0;
3940 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3941 enum flow_block_binder_type binder_type,
3942 struct nlattr *block_index_attr,
3943 struct netlink_ext_ack *extack)
3945 u32 block_index;
3946 int err;
3948 if (!block_index_attr)
3949 return 0;
3951 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3952 if (err)
3953 return err;
3955 qe->info.binder_type = binder_type;
3956 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3957 qe->info.chain_head_change_priv = &qe->filter_chain;
3958 qe->info.block_index = block_index;
3960 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3962 EXPORT_SYMBOL(tcf_qevent_init);
3964 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3966 if (qe->info.block_index)
3967 tcf_block_put_ext(qe->block, sch, &qe->info);
3969 EXPORT_SYMBOL(tcf_qevent_destroy);
3971 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3972 struct netlink_ext_ack *extack)
3974 u32 block_index;
3975 int err;
3977 if (!block_index_attr)
3978 return 0;
3980 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3981 if (err)
3982 return err;
3984 /* Bounce newly-configured block or change in block. */
3985 if (block_index != qe->info.block_index) {
3986 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3987 return -EINVAL;
3990 return 0;
3992 EXPORT_SYMBOL(tcf_qevent_validate_change);
3994 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3995 struct sk_buff **to_free, int *ret)
3997 struct tcf_result cl_res;
3998 struct tcf_proto *fl;
4000 if (!qe->info.block_index)
4001 return skb;
4003 fl = rcu_dereference_bh(qe->filter_chain);
4005 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
4006 case TC_ACT_SHOT:
4007 qdisc_qstats_drop(sch);
4008 __qdisc_drop(skb, to_free);
4009 *ret = __NET_XMIT_BYPASS;
4010 return NULL;
4011 case TC_ACT_STOLEN:
4012 case TC_ACT_QUEUED:
4013 case TC_ACT_TRAP:
4014 __qdisc_drop(skb, to_free);
4015 *ret = __NET_XMIT_STOLEN;
4016 return NULL;
4017 case TC_ACT_REDIRECT:
4018 skb_do_redirect(skb);
4019 *ret = __NET_XMIT_STOLEN;
4020 return NULL;
4023 return skb;
4025 EXPORT_SYMBOL(tcf_qevent_handle);
4027 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
4029 if (!qe->info.block_index)
4030 return 0;
4031 return nla_put_u32(skb, attr_name, qe->info.block_index);
4033 EXPORT_SYMBOL(tcf_qevent_dump);
4034 #endif
4036 static __net_init int tcf_net_init(struct net *net)
4038 struct tcf_net *tn = net_generic(net, tcf_net_id);
4040 spin_lock_init(&tn->idr_lock);
4041 idr_init(&tn->idr);
4042 return 0;
4045 static void __net_exit tcf_net_exit(struct net *net)
4047 struct tcf_net *tn = net_generic(net, tcf_net_id);
4049 idr_destroy(&tn->idr);
4052 static struct pernet_operations tcf_net_ops = {
4053 .init = tcf_net_init,
4054 .exit = tcf_net_exit,
4055 .id = &tcf_net_id,
4056 .size = sizeof(struct tcf_net),
4059 static const struct rtnl_msg_handler tc_filter_rtnl_msg_handlers[] __initconst = {
4060 {.msgtype = RTM_NEWTFILTER, .doit = tc_new_tfilter,
4061 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4062 {.msgtype = RTM_DELTFILTER, .doit = tc_del_tfilter,
4063 .flags = RTNL_FLAG_DOIT_UNLOCKED},
4064 {.msgtype = RTM_GETTFILTER, .doit = tc_get_tfilter,
4065 .dumpit = tc_dump_tfilter, .flags = RTNL_FLAG_DOIT_UNLOCKED},
4066 {.msgtype = RTM_NEWCHAIN, .doit = tc_ctl_chain},
4067 {.msgtype = RTM_DELCHAIN, .doit = tc_ctl_chain},
4068 {.msgtype = RTM_GETCHAIN, .doit = tc_ctl_chain,
4069 .dumpit = tc_dump_chain},
4072 static int __init tc_filter_init(void)
4074 int err;
4076 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4077 if (!tc_filter_wq)
4078 return -ENOMEM;
4080 err = register_pernet_subsys(&tcf_net_ops);
4081 if (err)
4082 goto err_register_pernet_subsys;
4084 xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4085 rtnl_register_many(tc_filter_rtnl_msg_handlers);
4087 return 0;
4089 err_register_pernet_subsys:
4090 destroy_workqueue(tc_filter_wq);
4091 return err;
4094 subsys_initcall(tc_filter_init);