1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/act_api.c Packet action API.
5 * Author: Jamal Hadi Salim
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/string.h>
11 #include <linux/errno.h>
12 #include <linux/slab.h>
13 #include <linux/skbuff.h>
14 #include <linux/init.h>
15 #include <linux/kmod.h>
16 #include <linux/err.h>
17 #include <linux/module.h>
18 #include <net/net_namespace.h>
20 #include <net/sch_generic.h>
21 #include <net/pkt_cls.h>
22 #include <net/act_api.h>
23 #include <net/netlink.h>
25 static void tcf_action_goto_chain_exec(const struct tc_action *a,
26 struct tcf_result *res)
28 const struct tcf_chain *chain = rcu_dereference_bh(a->goto_chain);
30 res->goto_tp = rcu_dereference_bh(chain->filter_chain);
33 static void tcf_free_cookie_rcu(struct rcu_head *p)
35 struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
41 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
42 struct tc_cookie *new_cookie)
44 struct tc_cookie *old;
46 old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
48 call_rcu(&old->rcu, tcf_free_cookie_rcu);
51 int tcf_action_check_ctrlact(int action, struct tcf_proto *tp,
52 struct tcf_chain **newchain,
53 struct netlink_ext_ack *extack)
55 int opcode = TC_ACT_EXT_OPCODE(action), ret = -EINVAL;
59 ret = action > TC_ACT_VALUE_MAX ? -EINVAL : 0;
60 else if (opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC)
63 NL_SET_ERR_MSG(extack, "invalid control action");
67 if (TC_ACT_EXT_CMP(action, TC_ACT_GOTO_CHAIN)) {
68 chain_index = action & TC_ACT_EXT_VAL_MASK;
69 if (!tp || !newchain) {
71 NL_SET_ERR_MSG(extack,
72 "can't goto NULL proto/chain");
75 *newchain = tcf_chain_get_by_act(tp->chain->block, chain_index);
78 NL_SET_ERR_MSG(extack,
79 "can't allocate goto_chain");
85 EXPORT_SYMBOL(tcf_action_check_ctrlact);
87 struct tcf_chain *tcf_action_set_ctrlact(struct tc_action *a, int action,
88 struct tcf_chain *goto_chain)
90 a->tcfa_action = action;
91 goto_chain = rcu_replace_pointer(a->goto_chain, goto_chain, 1);
94 EXPORT_SYMBOL(tcf_action_set_ctrlact);
96 /* XXX: For standalone actions, we don't need a RCU grace period either, because
97 * actions are always connected to filters and filters are already destroyed in
98 * RCU callbacks, so after a RCU grace period actions are already disconnected
99 * from filters. Readers later can not find us.
101 static void free_tcf(struct tc_action *p)
103 struct tcf_chain *chain = rcu_dereference_protected(p->goto_chain, 1);
105 free_percpu(p->cpu_bstats);
106 free_percpu(p->cpu_bstats_hw);
107 free_percpu(p->cpu_qstats);
109 tcf_set_action_cookie(&p->act_cookie, NULL);
111 tcf_chain_put_by_act(chain);
116 static void tcf_action_cleanup(struct tc_action *p)
121 gen_kill_estimator(&p->tcfa_rate_est);
125 static int __tcf_action_put(struct tc_action *p, bool bind)
127 struct tcf_idrinfo *idrinfo = p->idrinfo;
129 if (refcount_dec_and_mutex_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
131 atomic_dec(&p->tcfa_bindcnt);
132 idr_remove(&idrinfo->action_idr, p->tcfa_index);
133 mutex_unlock(&idrinfo->lock);
135 tcf_action_cleanup(p);
140 atomic_dec(&p->tcfa_bindcnt);
145 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
149 /* Release with strict==1 and bind==0 is only called through act API
150 * interface (classifiers always bind). Only case when action with
151 * positive reference count and zero bind count can exist is when it was
152 * also created with act API (unbinding last classifier will destroy the
153 * action if it was created by classifier). So only case when bind count
154 * can be changed after initial check is when unbound action is
155 * destroyed by act API while classifier binds to action with same id
156 * concurrently. This result either creation of new action(same behavior
157 * as before), or reusing existing action if concurrent process
158 * increments reference count before action is deleted. Both scenarios
162 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
165 if (__tcf_action_put(p, bind))
171 EXPORT_SYMBOL(__tcf_idr_release);
173 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
175 struct tc_cookie *act_cookie;
179 act_cookie = rcu_dereference(act->act_cookie);
182 cookie_len = nla_total_size(act_cookie->len);
185 return nla_total_size(0) /* action number nested */
186 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
187 + cookie_len /* TCA_ACT_COOKIE */
188 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_HW_STATS */
189 + nla_total_size(0) /* TCA_ACT_STATS nested */
190 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_FLAGS */
191 /* TCA_STATS_BASIC */
192 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
193 /* TCA_STATS_PKT64 */
194 + nla_total_size_64bit(sizeof(u64))
195 /* TCA_STATS_QUEUE */
196 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
197 + nla_total_size(0) /* TCA_OPTIONS nested */
198 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
201 static size_t tcf_action_full_attrs_size(size_t sz)
203 return NLMSG_HDRLEN /* struct nlmsghdr */
204 + sizeof(struct tcamsg)
205 + nla_total_size(0) /* TCA_ACT_TAB nested */
209 static size_t tcf_action_fill_size(const struct tc_action *act)
211 size_t sz = tcf_action_shared_attrs_size(act);
213 if (act->ops->get_fill_size)
214 return act->ops->get_fill_size(act) + sz;
218 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
219 struct netlink_callback *cb)
221 int err = 0, index = -1, s_i = 0, n_i = 0;
222 u32 act_flags = cb->args[2];
223 unsigned long jiffy_since = cb->args[3];
225 struct idr *idr = &idrinfo->action_idr;
227 unsigned long id = 1;
230 mutex_lock(&idrinfo->lock);
234 idr_for_each_entry_ul(idr, p, tmp, id) {
242 time_after(jiffy_since,
243 (unsigned long)p->tcfa_tm.lastuse))
246 nest = nla_nest_start_noflag(skb, n_i);
249 goto nla_put_failure;
251 err = tcf_action_dump_1(skb, p, 0, 0);
254 nlmsg_trim(skb, nest);
257 nla_nest_end(skb, nest);
259 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
260 n_i >= TCA_ACT_MAX_PRIO)
265 cb->args[0] = index + 1;
267 mutex_unlock(&idrinfo->lock);
269 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
275 nla_nest_cancel(skb, nest);
279 static int tcf_idr_release_unsafe(struct tc_action *p)
281 if (atomic_read(&p->tcfa_bindcnt) > 0)
284 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
285 idr_remove(&p->idrinfo->action_idr, p->tcfa_index);
286 tcf_action_cleanup(p);
287 return ACT_P_DELETED;
293 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
294 const struct tc_action_ops *ops)
299 struct idr *idr = &idrinfo->action_idr;
301 unsigned long id = 1;
304 nest = nla_nest_start_noflag(skb, 0);
306 goto nla_put_failure;
307 if (nla_put_string(skb, TCA_KIND, ops->kind))
308 goto nla_put_failure;
310 mutex_lock(&idrinfo->lock);
311 idr_for_each_entry_ul(idr, p, tmp, id) {
314 ret = tcf_idr_release_unsafe(p);
315 if (ret == ACT_P_DELETED) {
316 module_put(ops->owner);
318 } else if (ret < 0) {
319 mutex_unlock(&idrinfo->lock);
320 goto nla_put_failure;
323 mutex_unlock(&idrinfo->lock);
325 if (nla_put_u32(skb, TCA_FCNT, n_i))
326 goto nla_put_failure;
327 nla_nest_end(skb, nest);
331 nla_nest_cancel(skb, nest);
335 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
336 struct netlink_callback *cb, int type,
337 const struct tc_action_ops *ops,
338 struct netlink_ext_ack *extack)
340 struct tcf_idrinfo *idrinfo = tn->idrinfo;
342 if (type == RTM_DELACTION) {
343 return tcf_del_walker(idrinfo, skb, ops);
344 } else if (type == RTM_GETACTION) {
345 return tcf_dump_walker(idrinfo, skb, cb);
347 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
348 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
352 EXPORT_SYMBOL(tcf_generic_walker);
354 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
356 struct tcf_idrinfo *idrinfo = tn->idrinfo;
359 mutex_lock(&idrinfo->lock);
360 p = idr_find(&idrinfo->action_idr, index);
364 refcount_inc(&p->tcfa_refcnt);
365 mutex_unlock(&idrinfo->lock);
373 EXPORT_SYMBOL(tcf_idr_search);
375 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
380 mutex_lock(&idrinfo->lock);
381 p = idr_find(&idrinfo->action_idr, index);
383 mutex_unlock(&idrinfo->lock);
387 if (!atomic_read(&p->tcfa_bindcnt)) {
388 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
389 struct module *owner = p->ops->owner;
391 WARN_ON(p != idr_remove(&idrinfo->action_idr,
393 mutex_unlock(&idrinfo->lock);
395 tcf_action_cleanup(p);
404 mutex_unlock(&idrinfo->lock);
408 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
409 struct tc_action **a, const struct tc_action_ops *ops,
410 int bind, bool cpustats, u32 flags)
412 struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
413 struct tcf_idrinfo *idrinfo = tn->idrinfo;
418 refcount_set(&p->tcfa_refcnt, 1);
420 atomic_set(&p->tcfa_bindcnt, 1);
423 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
426 p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
427 if (!p->cpu_bstats_hw)
429 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
433 spin_lock_init(&p->tcfa_lock);
434 p->tcfa_index = index;
435 p->tcfa_tm.install = jiffies;
436 p->tcfa_tm.lastuse = jiffies;
437 p->tcfa_tm.firstuse = 0;
438 p->tcfa_flags = flags;
440 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
442 &p->tcfa_lock, NULL, est);
447 p->idrinfo = idrinfo;
452 free_percpu(p->cpu_qstats);
454 free_percpu(p->cpu_bstats_hw);
456 free_percpu(p->cpu_bstats);
461 EXPORT_SYMBOL(tcf_idr_create);
463 int tcf_idr_create_from_flags(struct tc_action_net *tn, u32 index,
464 struct nlattr *est, struct tc_action **a,
465 const struct tc_action_ops *ops, int bind,
468 /* Set cpustats according to actions flags. */
469 return tcf_idr_create(tn, index, est, a, ops, bind,
470 !(flags & TCA_ACT_FLAGS_NO_PERCPU_STATS), flags);
472 EXPORT_SYMBOL(tcf_idr_create_from_flags);
474 /* Cleanup idr index that was allocated but not initialized. */
476 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
478 struct tcf_idrinfo *idrinfo = tn->idrinfo;
480 mutex_lock(&idrinfo->lock);
481 /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
482 WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
483 mutex_unlock(&idrinfo->lock);
485 EXPORT_SYMBOL(tcf_idr_cleanup);
487 /* Check if action with specified index exists. If actions is found, increments
488 * its reference and bind counters, and return 1. Otherwise insert temporary
489 * error pointer (to prevent concurrent users from inserting actions with same
490 * index) and return 0.
493 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
494 struct tc_action **a, int bind)
496 struct tcf_idrinfo *idrinfo = tn->idrinfo;
501 mutex_lock(&idrinfo->lock);
503 p = idr_find(&idrinfo->action_idr, *index);
505 /* This means that another process allocated
506 * index but did not assign the pointer yet.
508 mutex_unlock(&idrinfo->lock);
513 refcount_inc(&p->tcfa_refcnt);
515 atomic_inc(&p->tcfa_bindcnt);
520 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
523 idr_replace(&idrinfo->action_idr,
524 ERR_PTR(-EBUSY), *index);
529 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
530 UINT_MAX, GFP_KERNEL);
532 idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
535 mutex_unlock(&idrinfo->lock);
538 EXPORT_SYMBOL(tcf_idr_check_alloc);
540 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
541 struct tcf_idrinfo *idrinfo)
543 struct idr *idr = &idrinfo->action_idr;
546 unsigned long id = 1;
549 idr_for_each_entry_ul(idr, p, tmp, id) {
550 ret = __tcf_idr_release(p, false, true);
551 if (ret == ACT_P_DELETED)
552 module_put(ops->owner);
556 idr_destroy(&idrinfo->action_idr);
558 EXPORT_SYMBOL(tcf_idrinfo_destroy);
560 static LIST_HEAD(act_base);
561 static DEFINE_RWLOCK(act_mod_lock);
563 int tcf_register_action(struct tc_action_ops *act,
564 struct pernet_operations *ops)
566 struct tc_action_ops *a;
569 if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
572 /* We have to register pernet ops before making the action ops visible,
573 * otherwise tcf_action_init_1() could get a partially initialized
576 ret = register_pernet_subsys(ops);
580 write_lock(&act_mod_lock);
581 list_for_each_entry(a, &act_base, head) {
582 if (act->id == a->id || (strcmp(act->kind, a->kind) == 0)) {
583 write_unlock(&act_mod_lock);
584 unregister_pernet_subsys(ops);
588 list_add_tail(&act->head, &act_base);
589 write_unlock(&act_mod_lock);
593 EXPORT_SYMBOL(tcf_register_action);
595 int tcf_unregister_action(struct tc_action_ops *act,
596 struct pernet_operations *ops)
598 struct tc_action_ops *a;
601 write_lock(&act_mod_lock);
602 list_for_each_entry(a, &act_base, head) {
604 list_del(&act->head);
609 write_unlock(&act_mod_lock);
611 unregister_pernet_subsys(ops);
614 EXPORT_SYMBOL(tcf_unregister_action);
617 static struct tc_action_ops *tc_lookup_action_n(char *kind)
619 struct tc_action_ops *a, *res = NULL;
622 read_lock(&act_mod_lock);
623 list_for_each_entry(a, &act_base, head) {
624 if (strcmp(kind, a->kind) == 0) {
625 if (try_module_get(a->owner))
630 read_unlock(&act_mod_lock);
635 /* lookup by nlattr */
636 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
638 struct tc_action_ops *a, *res = NULL;
641 read_lock(&act_mod_lock);
642 list_for_each_entry(a, &act_base, head) {
643 if (nla_strcmp(kind, a->kind) == 0) {
644 if (try_module_get(a->owner))
649 read_unlock(&act_mod_lock);
654 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
655 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
656 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
657 int nr_actions, struct tcf_result *res)
660 u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
664 if (skb_skip_tc_classify(skb))
668 for (i = 0; i < nr_actions; i++) {
669 const struct tc_action *a = actions[i];
671 if (jmp_prgcnt > 0) {
676 ret = a->ops->act(skb, a, res);
677 if (ret == TC_ACT_REPEAT)
678 goto repeat; /* we need a ttl - JHS */
680 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
681 jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
682 if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
683 /* faulty opcode, stop pipeline */
688 goto restart_act_graph;
689 else /* faulty graph, stop pipeline */
692 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
693 if (unlikely(!rcu_access_pointer(a->goto_chain))) {
694 net_warn_ratelimited("can't go to NULL chain!\n");
697 tcf_action_goto_chain_exec(a, res);
700 if (ret != TC_ACT_PIPE)
706 EXPORT_SYMBOL(tcf_action_exec);
708 int tcf_action_destroy(struct tc_action *actions[], int bind)
710 const struct tc_action_ops *ops;
714 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
718 ret = __tcf_idr_release(a, bind, true);
719 if (ret == ACT_P_DELETED)
720 module_put(ops->owner);
727 static int tcf_action_put(struct tc_action *p)
729 return __tcf_action_put(p, false);
732 /* Put all actions in this array, skip those NULL's. */
733 static void tcf_action_put_many(struct tc_action *actions[])
737 for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
738 struct tc_action *a = actions[i];
739 const struct tc_action_ops *ops;
744 if (tcf_action_put(a))
745 module_put(ops->owner);
750 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
752 return a->ops->dump(skb, a, bind, ref);
756 tcf_action_dump_terse(struct sk_buff *skb, struct tc_action *a)
758 unsigned char *b = skb_tail_pointer(skb);
759 struct tc_cookie *cookie;
761 if (nla_put_string(skb, TCA_KIND, a->ops->kind))
762 goto nla_put_failure;
763 if (tcf_action_copy_stats(skb, a, 0))
764 goto nla_put_failure;
767 cookie = rcu_dereference(a->act_cookie);
769 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
771 goto nla_put_failure;
784 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
787 unsigned char *b = skb_tail_pointer(skb);
790 if (tcf_action_dump_terse(skb, a))
791 goto nla_put_failure;
793 if (a->hw_stats != TCA_ACT_HW_STATS_ANY &&
794 nla_put_bitfield32(skb, TCA_ACT_HW_STATS,
795 a->hw_stats, TCA_ACT_HW_STATS_ANY))
796 goto nla_put_failure;
798 if (a->used_hw_stats_valid &&
799 nla_put_bitfield32(skb, TCA_ACT_USED_HW_STATS,
800 a->used_hw_stats, TCA_ACT_HW_STATS_ANY))
801 goto nla_put_failure;
804 nla_put_bitfield32(skb, TCA_ACT_FLAGS,
805 a->tcfa_flags, a->tcfa_flags))
806 goto nla_put_failure;
808 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
810 goto nla_put_failure;
811 err = tcf_action_dump_old(skb, a, bind, ref);
813 nla_nest_end(skb, nest);
821 EXPORT_SYMBOL(tcf_action_dump_1);
823 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
824 int bind, int ref, bool terse)
827 int err = -EINVAL, i;
830 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
832 nest = nla_nest_start_noflag(skb, i + 1);
834 goto nla_put_failure;
835 err = terse ? tcf_action_dump_terse(skb, a) :
836 tcf_action_dump_1(skb, a, bind, ref);
839 nla_nest_end(skb, nest);
847 nla_nest_cancel(skb, nest);
851 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
853 struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
857 c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
862 c->len = nla_len(tb[TCA_ACT_COOKIE]);
867 static u8 tcf_action_hw_stats_get(struct nlattr *hw_stats_attr)
869 struct nla_bitfield32 hw_stats_bf;
871 /* If the user did not pass the attr, that means he does
872 * not care about the type. Return "any" in that case
873 * which is setting on all supported types.
876 return TCA_ACT_HW_STATS_ANY;
877 hw_stats_bf = nla_get_bitfield32(hw_stats_attr);
878 return hw_stats_bf.value;
881 static const struct nla_policy tcf_action_policy[TCA_ACT_MAX + 1] = {
882 [TCA_ACT_KIND] = { .type = NLA_STRING },
883 [TCA_ACT_INDEX] = { .type = NLA_U32 },
884 [TCA_ACT_COOKIE] = { .type = NLA_BINARY,
885 .len = TC_COOKIE_MAX_SIZE },
886 [TCA_ACT_OPTIONS] = { .type = NLA_NESTED },
887 [TCA_ACT_FLAGS] = NLA_POLICY_BITFIELD32(TCA_ACT_FLAGS_NO_PERCPU_STATS),
888 [TCA_ACT_HW_STATS] = NLA_POLICY_BITFIELD32(TCA_ACT_HW_STATS_ANY),
891 static void tcf_idr_insert_many(struct tc_action *actions[])
895 for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
896 struct tc_action *a = actions[i];
897 struct tcf_idrinfo *idrinfo;
901 idrinfo = a->idrinfo;
902 mutex_lock(&idrinfo->lock);
903 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc if
904 * it is just created, otherwise this is just a nop.
906 idr_replace(&idrinfo->action_idr, a, a->tcfa_index);
907 mutex_unlock(&idrinfo->lock);
911 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
912 struct nlattr *nla, struct nlattr *est,
913 char *name, int ovr, int bind,
915 struct netlink_ext_ack *extack)
917 struct nla_bitfield32 flags = { 0, 0 };
918 u8 hw_stats = TCA_ACT_HW_STATS_ANY;
920 struct tc_action_ops *a_o;
921 struct tc_cookie *cookie = NULL;
922 char act_name[IFNAMSIZ];
923 struct nlattr *tb[TCA_ACT_MAX + 1];
928 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
929 tcf_action_policy, extack);
933 kind = tb[TCA_ACT_KIND];
935 NL_SET_ERR_MSG(extack, "TC action kind must be specified");
938 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
939 NL_SET_ERR_MSG(extack, "TC action name too long");
942 if (tb[TCA_ACT_COOKIE]) {
943 cookie = nla_memdup_cookie(tb);
945 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
950 hw_stats = tcf_action_hw_stats_get(tb[TCA_ACT_HW_STATS]);
951 if (tb[TCA_ACT_FLAGS])
952 flags = nla_get_bitfield32(tb[TCA_ACT_FLAGS]);
954 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
955 NL_SET_ERR_MSG(extack, "TC action name too long");
961 a_o = tc_lookup_action_n(act_name);
963 #ifdef CONFIG_MODULES
966 request_module("act_%s", act_name);
970 a_o = tc_lookup_action_n(act_name);
972 /* We dropped the RTNL semaphore in order to
973 * perform the module load. So, even if we
974 * succeeded in loading the module we have to
975 * tell the caller to replay the request. We
976 * indicate this using -EAGAIN.
983 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
988 /* backward compatibility for policer */
990 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
991 rtnl_held, tp, flags.value, extack);
993 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
994 tp, flags.value, extack);
998 if (!name && tb[TCA_ACT_COOKIE])
999 tcf_set_action_cookie(&a->act_cookie, cookie);
1002 a->hw_stats = hw_stats;
1004 /* module count goes up only when brand new policy is created
1005 * if it exists and is only bound to in a_o->init() then
1006 * ACT_P_CREATED is not returned (a zero is).
1008 if (err != ACT_P_CREATED)
1009 module_put(a_o->owner);
1014 module_put(a_o->owner);
1017 kfree(cookie->data);
1021 return ERR_PTR(err);
1024 /* Returns numbers of initialized actions or negative error. */
1026 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
1027 struct nlattr *est, char *name, int ovr, int bind,
1028 struct tc_action *actions[], size_t *attr_size,
1029 bool rtnl_held, struct netlink_ext_ack *extack)
1031 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1032 struct tc_action *act;
1037 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1042 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1043 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
1049 sz += tcf_action_fill_size(act);
1050 /* Start from index 0 */
1051 actions[i - 1] = act;
1054 /* We have to commit them all together, because if any error happened in
1055 * between, we could not handle the failure gracefully.
1057 tcf_idr_insert_many(actions);
1059 *attr_size = tcf_action_full_attrs_size(sz);
1063 tcf_action_destroy(actions, bind);
1067 void tcf_action_update_stats(struct tc_action *a, u64 bytes, u64 packets,
1070 if (a->cpu_bstats) {
1071 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
1073 this_cpu_ptr(a->cpu_qstats)->drops += drops;
1076 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats_hw),
1081 _bstats_update(&a->tcfa_bstats, bytes, packets);
1082 a->tcfa_qstats.drops += drops;
1084 _bstats_update(&a->tcfa_bstats_hw, bytes, packets);
1086 EXPORT_SYMBOL(tcf_action_update_stats);
1088 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
1097 /* compat_mode being true specifies a call that is supposed
1098 * to add additional backward compatibility statistic TLVs.
1101 if (p->type == TCA_OLD_COMPAT)
1102 err = gnet_stats_start_copy_compat(skb, 0,
1110 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1111 &p->tcfa_lock, &d, TCA_ACT_PAD);
1116 if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
1117 gnet_stats_copy_basic_hw(NULL, &d, p->cpu_bstats_hw,
1118 &p->tcfa_bstats_hw) < 0 ||
1119 gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1120 gnet_stats_copy_queue(&d, p->cpu_qstats,
1122 p->tcfa_qstats.qlen) < 0)
1125 if (gnet_stats_finish_copy(&d) < 0)
1134 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1135 u32 portid, u32 seq, u16 flags, int event, int bind,
1139 struct nlmsghdr *nlh;
1140 unsigned char *b = skb_tail_pointer(skb);
1141 struct nlattr *nest;
1143 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1145 goto out_nlmsg_trim;
1146 t = nlmsg_data(nlh);
1147 t->tca_family = AF_UNSPEC;
1151 nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1153 goto out_nlmsg_trim;
1155 if (tcf_action_dump(skb, actions, bind, ref, false) < 0)
1156 goto out_nlmsg_trim;
1158 nla_nest_end(skb, nest);
1160 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1169 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1170 struct tc_action *actions[], int event,
1171 struct netlink_ext_ack *extack)
1173 struct sk_buff *skb;
1175 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1178 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1180 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1185 return rtnl_unicast(skb, net, portid);
1188 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1189 struct nlmsghdr *n, u32 portid,
1190 struct netlink_ext_ack *extack)
1192 struct nlattr *tb[TCA_ACT_MAX + 1];
1193 const struct tc_action_ops *ops;
1194 struct tc_action *a;
1198 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1199 tcf_action_policy, extack);
1204 if (tb[TCA_ACT_INDEX] == NULL ||
1205 nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1206 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1209 index = nla_get_u32(tb[TCA_ACT_INDEX]);
1212 ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1213 if (!ops) { /* could happen in batch of actions */
1214 NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1218 if (ops->lookup(net, &a, index) == 0) {
1219 NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1223 module_put(ops->owner);
1227 module_put(ops->owner);
1229 return ERR_PTR(err);
1232 static int tca_action_flush(struct net *net, struct nlattr *nla,
1233 struct nlmsghdr *n, u32 portid,
1234 struct netlink_ext_ack *extack)
1236 struct sk_buff *skb;
1238 struct nlmsghdr *nlh;
1240 struct netlink_callback dcb;
1241 struct nlattr *nest;
1242 struct nlattr *tb[TCA_ACT_MAX + 1];
1243 const struct tc_action_ops *ops;
1244 struct nlattr *kind;
1247 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1251 b = skb_tail_pointer(skb);
1253 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1254 tcf_action_policy, extack);
1259 kind = tb[TCA_ACT_KIND];
1260 ops = tc_lookup_action(kind);
1261 if (!ops) { /*some idjot trying to flush unknown action */
1262 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1266 nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1269 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1270 goto out_module_put;
1272 t = nlmsg_data(nlh);
1273 t->tca_family = AF_UNSPEC;
1277 nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1279 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1280 goto out_module_put;
1283 err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1285 nla_nest_cancel(skb, nest);
1286 goto out_module_put;
1289 nla_nest_end(skb, nest);
1291 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1292 nlh->nlmsg_flags |= NLM_F_ROOT;
1293 module_put(ops->owner);
1294 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1295 n->nlmsg_flags & NLM_F_ECHO);
1299 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1304 module_put(ops->owner);
1310 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1314 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1315 struct tc_action *a = actions[i];
1316 const struct tc_action_ops *ops = a->ops;
1317 /* Actions can be deleted concurrently so we must save their
1318 * type and id to search again after reference is released.
1320 struct tcf_idrinfo *idrinfo = a->idrinfo;
1321 u32 act_index = a->tcfa_index;
1324 if (tcf_action_put(a)) {
1325 /* last reference, action was deleted concurrently */
1326 module_put(ops->owner);
1330 /* now do the delete */
1331 ret = tcf_idr_delete_index(idrinfo, act_index);
1340 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1341 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1344 struct sk_buff *skb;
1346 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1351 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1353 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1358 /* now do the delete */
1359 ret = tcf_action_delete(net, actions);
1361 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1366 ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1367 n->nlmsg_flags & NLM_F_ECHO);
1374 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1375 u32 portid, int event, struct netlink_ext_ack *extack)
1378 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1379 struct tc_action *act;
1380 size_t attr_size = 0;
1381 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1383 ret = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1388 if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1390 return tca_action_flush(net, tb[1], n, portid, extack);
1392 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1396 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1397 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1402 attr_size += tcf_action_fill_size(act);
1403 actions[i - 1] = act;
1406 attr_size = tcf_action_full_attrs_size(attr_size);
1408 if (event == RTM_GETACTION)
1409 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1411 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1417 tcf_action_put_many(actions);
1422 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1423 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1425 struct sk_buff *skb;
1428 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1433 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1434 RTM_NEWACTION, 0, 0) <= 0) {
1435 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1440 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1441 n->nlmsg_flags & NLM_F_ECHO);
1447 static int tcf_action_add(struct net *net, struct nlattr *nla,
1448 struct nlmsghdr *n, u32 portid, int ovr,
1449 struct netlink_ext_ack *extack)
1451 size_t attr_size = 0;
1453 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1455 for (loop = 0; loop < 10; loop++) {
1456 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0,
1457 actions, &attr_size, true, extack);
1464 ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1466 tcf_action_put_many(actions);
1471 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1472 [TCA_ROOT_FLAGS] = NLA_POLICY_BITFIELD32(TCA_FLAG_LARGE_DUMP_ON),
1473 [TCA_ROOT_TIME_DELTA] = { .type = NLA_U32 },
1476 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1477 struct netlink_ext_ack *extack)
1479 struct net *net = sock_net(skb->sk);
1480 struct nlattr *tca[TCA_ROOT_MAX + 1];
1481 u32 portid = NETLINK_CB(skb).portid;
1482 int ret = 0, ovr = 0;
1484 if ((n->nlmsg_type != RTM_GETACTION) &&
1485 !netlink_capable(skb, CAP_NET_ADMIN))
1488 ret = nlmsg_parse_deprecated(n, sizeof(struct tcamsg), tca,
1489 TCA_ROOT_MAX, NULL, extack);
1493 if (tca[TCA_ACT_TAB] == NULL) {
1494 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1498 /* n->nlmsg_flags & NLM_F_CREATE */
1499 switch (n->nlmsg_type) {
1501 /* we are going to assume all other flags
1502 * imply create only if it doesn't exist
1503 * Note that CREATE | EXCL implies that
1504 * but since we want avoid ambiguity (eg when flags
1505 * is zero) then just set this
1507 if (n->nlmsg_flags & NLM_F_REPLACE)
1509 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1513 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1514 portid, RTM_DELACTION, extack);
1517 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1518 portid, RTM_GETACTION, extack);
1527 static struct nlattr *find_dump_kind(struct nlattr **nla)
1529 struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1530 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1531 struct nlattr *kind;
1533 tb1 = nla[TCA_ACT_TAB];
1537 if (nla_parse_deprecated(tb, TCA_ACT_MAX_PRIO, nla_data(tb1), NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1542 if (nla_parse_nested_deprecated(tb2, TCA_ACT_MAX, tb[1], tcf_action_policy, NULL) < 0)
1544 kind = tb2[TCA_ACT_KIND];
1549 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1551 struct net *net = sock_net(skb->sk);
1552 struct nlmsghdr *nlh;
1553 unsigned char *b = skb_tail_pointer(skb);
1554 struct nlattr *nest;
1555 struct tc_action_ops *a_o;
1557 struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1558 struct nlattr *tb[TCA_ROOT_MAX + 1];
1559 struct nlattr *count_attr = NULL;
1560 unsigned long jiffy_since = 0;
1561 struct nlattr *kind = NULL;
1562 struct nla_bitfield32 bf;
1563 u32 msecs_since = 0;
1566 ret = nlmsg_parse_deprecated(cb->nlh, sizeof(struct tcamsg), tb,
1567 TCA_ROOT_MAX, tcaa_policy, cb->extack);
1571 kind = find_dump_kind(tb);
1573 pr_info("tc_dump_action: action bad kind\n");
1577 a_o = tc_lookup_action(kind);
1582 if (tb[TCA_ROOT_FLAGS]) {
1583 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1584 cb->args[2] = bf.value;
1587 if (tb[TCA_ROOT_TIME_DELTA]) {
1588 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1591 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1592 cb->nlh->nlmsg_type, sizeof(*t), 0);
1594 goto out_module_put;
1597 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1599 t = nlmsg_data(nlh);
1600 t->tca_family = AF_UNSPEC;
1603 cb->args[3] = jiffy_since;
1604 count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1606 goto out_module_put;
1608 nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1610 goto out_module_put;
1612 ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1614 goto out_module_put;
1617 nla_nest_end(skb, nest);
1619 act_count = cb->args[1];
1620 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1625 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1626 if (NETLINK_CB(cb->skb).portid && ret)
1627 nlh->nlmsg_flags |= NLM_F_MULTI;
1628 module_put(a_o->owner);
1632 module_put(a_o->owner);
1637 static int __init tc_action_init(void)
1639 rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1640 rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1641 rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1647 subsys_initcall(tc_action_init);