2 * net/sched/act_api.c Packet action API.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Author: Jamal Hadi Salim
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/slab.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/kmod.h>
22 #include <linux/err.h>
23 #include <linux/module.h>
24 #include <linux/rhashtable.h>
25 #include <linux/list.h>
26 #include <net/net_namespace.h>
28 #include <net/sch_generic.h>
29 #include <net/pkt_cls.h>
30 #include <net/act_api.h>
31 #include <net/netlink.h>
33 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
35 u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
39 a->goto_chain = tcf_chain_get_by_act(tp->chain->block, chain_index);
45 static void tcf_action_goto_chain_fini(struct tc_action *a)
47 tcf_chain_put_by_act(a->goto_chain);
50 static void tcf_action_goto_chain_exec(const struct tc_action *a,
51 struct tcf_result *res)
53 const struct tcf_chain *chain = a->goto_chain;
55 res->goto_tp = rcu_dereference_bh(chain->filter_chain);
58 static void tcf_free_cookie_rcu(struct rcu_head *p)
60 struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
66 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
67 struct tc_cookie *new_cookie)
69 struct tc_cookie *old;
71 old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
73 call_rcu(&old->rcu, tcf_free_cookie_rcu);
76 /* XXX: For standalone actions, we don't need a RCU grace period either, because
77 * actions are always connected to filters and filters are already destroyed in
78 * RCU callbacks, so after a RCU grace period actions are already disconnected
79 * from filters. Readers later can not find us.
81 static void free_tcf(struct tc_action *p)
83 free_percpu(p->cpu_bstats);
84 free_percpu(p->cpu_bstats_hw);
85 free_percpu(p->cpu_qstats);
87 tcf_set_action_cookie(&p->act_cookie, NULL);
89 tcf_action_goto_chain_fini(p);
94 static void tcf_action_cleanup(struct tc_action *p)
99 gen_kill_estimator(&p->tcfa_rate_est);
103 static int __tcf_action_put(struct tc_action *p, bool bind)
105 struct tcf_idrinfo *idrinfo = p->idrinfo;
107 if (refcount_dec_and_mutex_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
109 atomic_dec(&p->tcfa_bindcnt);
110 idr_remove(&idrinfo->action_idr, p->tcfa_index);
111 mutex_unlock(&idrinfo->lock);
113 tcf_action_cleanup(p);
118 atomic_dec(&p->tcfa_bindcnt);
123 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
127 /* Release with strict==1 and bind==0 is only called through act API
128 * interface (classifiers always bind). Only case when action with
129 * positive reference count and zero bind count can exist is when it was
130 * also created with act API (unbinding last classifier will destroy the
131 * action if it was created by classifier). So only case when bind count
132 * can be changed after initial check is when unbound action is
133 * destroyed by act API while classifier binds to action with same id
134 * concurrently. This result either creation of new action(same behavior
135 * as before), or reusing existing action if concurrent process
136 * increments reference count before action is deleted. Both scenarios
140 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
143 if (__tcf_action_put(p, bind))
149 EXPORT_SYMBOL(__tcf_idr_release);
151 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
153 struct tc_cookie *act_cookie;
157 act_cookie = rcu_dereference(act->act_cookie);
160 cookie_len = nla_total_size(act_cookie->len);
163 return nla_total_size(0) /* action number nested */
164 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
165 + cookie_len /* TCA_ACT_COOKIE */
166 + nla_total_size(0) /* TCA_ACT_STATS nested */
167 /* TCA_STATS_BASIC */
168 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
169 /* TCA_STATS_QUEUE */
170 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
171 + nla_total_size(0) /* TCA_OPTIONS nested */
172 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
175 static size_t tcf_action_full_attrs_size(size_t sz)
177 return NLMSG_HDRLEN /* struct nlmsghdr */
178 + sizeof(struct tcamsg)
179 + nla_total_size(0) /* TCA_ACT_TAB nested */
183 static size_t tcf_action_fill_size(const struct tc_action *act)
185 size_t sz = tcf_action_shared_attrs_size(act);
187 if (act->ops->get_fill_size)
188 return act->ops->get_fill_size(act) + sz;
192 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
193 struct netlink_callback *cb)
195 int err = 0, index = -1, s_i = 0, n_i = 0;
196 u32 act_flags = cb->args[2];
197 unsigned long jiffy_since = cb->args[3];
199 struct idr *idr = &idrinfo->action_idr;
201 unsigned long id = 1;
203 mutex_lock(&idrinfo->lock);
207 idr_for_each_entry_ul(idr, p, id) {
213 time_after(jiffy_since,
214 (unsigned long)p->tcfa_tm.lastuse))
217 nest = nla_nest_start(skb, n_i);
220 goto nla_put_failure;
222 err = tcf_action_dump_1(skb, p, 0, 0);
225 nlmsg_trim(skb, nest);
228 nla_nest_end(skb, nest);
230 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
231 n_i >= TCA_ACT_MAX_PRIO)
236 cb->args[0] = index + 1;
238 mutex_unlock(&idrinfo->lock);
240 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
246 nla_nest_cancel(skb, nest);
250 static int tcf_idr_release_unsafe(struct tc_action *p)
252 if (atomic_read(&p->tcfa_bindcnt) > 0)
255 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
256 idr_remove(&p->idrinfo->action_idr, p->tcfa_index);
257 tcf_action_cleanup(p);
258 return ACT_P_DELETED;
264 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
265 const struct tc_action_ops *ops)
270 struct idr *idr = &idrinfo->action_idr;
272 unsigned long id = 1;
274 nest = nla_nest_start(skb, 0);
276 goto nla_put_failure;
277 if (nla_put_string(skb, TCA_KIND, ops->kind))
278 goto nla_put_failure;
280 mutex_lock(&idrinfo->lock);
281 idr_for_each_entry_ul(idr, p, id) {
282 ret = tcf_idr_release_unsafe(p);
283 if (ret == ACT_P_DELETED) {
284 module_put(ops->owner);
286 } else if (ret < 0) {
287 mutex_unlock(&idrinfo->lock);
288 goto nla_put_failure;
291 mutex_unlock(&idrinfo->lock);
293 if (nla_put_u32(skb, TCA_FCNT, n_i))
294 goto nla_put_failure;
295 nla_nest_end(skb, nest);
299 nla_nest_cancel(skb, nest);
303 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
304 struct netlink_callback *cb, int type,
305 const struct tc_action_ops *ops,
306 struct netlink_ext_ack *extack)
308 struct tcf_idrinfo *idrinfo = tn->idrinfo;
310 if (type == RTM_DELACTION) {
311 return tcf_del_walker(idrinfo, skb, ops);
312 } else if (type == RTM_GETACTION) {
313 return tcf_dump_walker(idrinfo, skb, cb);
315 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
316 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
320 EXPORT_SYMBOL(tcf_generic_walker);
322 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
324 struct tcf_idrinfo *idrinfo = tn->idrinfo;
327 mutex_lock(&idrinfo->lock);
328 p = idr_find(&idrinfo->action_idr, index);
332 refcount_inc(&p->tcfa_refcnt);
333 mutex_unlock(&idrinfo->lock);
341 EXPORT_SYMBOL(tcf_idr_search);
343 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
348 mutex_lock(&idrinfo->lock);
349 p = idr_find(&idrinfo->action_idr, index);
351 mutex_unlock(&idrinfo->lock);
355 if (!atomic_read(&p->tcfa_bindcnt)) {
356 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
357 struct module *owner = p->ops->owner;
359 WARN_ON(p != idr_remove(&idrinfo->action_idr,
361 mutex_unlock(&idrinfo->lock);
363 tcf_action_cleanup(p);
372 mutex_unlock(&idrinfo->lock);
376 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
377 struct tc_action **a, const struct tc_action_ops *ops,
378 int bind, bool cpustats)
380 struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
381 struct tcf_idrinfo *idrinfo = tn->idrinfo;
386 refcount_set(&p->tcfa_refcnt, 1);
388 atomic_set(&p->tcfa_bindcnt, 1);
391 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
394 p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
395 if (!p->cpu_bstats_hw)
397 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
401 spin_lock_init(&p->tcfa_lock);
402 p->tcfa_index = index;
403 p->tcfa_tm.install = jiffies;
404 p->tcfa_tm.lastuse = jiffies;
405 p->tcfa_tm.firstuse = 0;
407 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
409 &p->tcfa_lock, NULL, est);
414 p->idrinfo = idrinfo;
419 free_percpu(p->cpu_qstats);
421 free_percpu(p->cpu_bstats_hw);
423 free_percpu(p->cpu_bstats);
428 EXPORT_SYMBOL(tcf_idr_create);
430 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
432 struct tcf_idrinfo *idrinfo = tn->idrinfo;
434 mutex_lock(&idrinfo->lock);
435 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
436 WARN_ON(!IS_ERR(idr_replace(&idrinfo->action_idr, a, a->tcfa_index)));
437 mutex_unlock(&idrinfo->lock);
439 EXPORT_SYMBOL(tcf_idr_insert);
441 /* Cleanup idr index that was allocated but not initialized. */
443 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
445 struct tcf_idrinfo *idrinfo = tn->idrinfo;
447 mutex_lock(&idrinfo->lock);
448 /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
449 WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
450 mutex_unlock(&idrinfo->lock);
452 EXPORT_SYMBOL(tcf_idr_cleanup);
454 /* Check if action with specified index exists. If actions is found, increments
455 * its reference and bind counters, and return 1. Otherwise insert temporary
456 * error pointer (to prevent concurrent users from inserting actions with same
457 * index) and return 0.
460 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
461 struct tc_action **a, int bind)
463 struct tcf_idrinfo *idrinfo = tn->idrinfo;
468 mutex_lock(&idrinfo->lock);
470 p = idr_find(&idrinfo->action_idr, *index);
472 /* This means that another process allocated
473 * index but did not assign the pointer yet.
475 mutex_unlock(&idrinfo->lock);
480 refcount_inc(&p->tcfa_refcnt);
482 atomic_inc(&p->tcfa_bindcnt);
487 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
490 idr_replace(&idrinfo->action_idr,
491 ERR_PTR(-EBUSY), *index);
496 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
497 UINT_MAX, GFP_KERNEL);
499 idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
502 mutex_unlock(&idrinfo->lock);
505 EXPORT_SYMBOL(tcf_idr_check_alloc);
507 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
508 struct tcf_idrinfo *idrinfo)
510 struct idr *idr = &idrinfo->action_idr;
513 unsigned long id = 1;
515 idr_for_each_entry_ul(idr, p, id) {
516 ret = __tcf_idr_release(p, false, true);
517 if (ret == ACT_P_DELETED)
518 module_put(ops->owner);
522 idr_destroy(&idrinfo->action_idr);
524 EXPORT_SYMBOL(tcf_idrinfo_destroy);
526 static LIST_HEAD(act_base);
527 static DEFINE_RWLOCK(act_mod_lock);
529 int tcf_register_action(struct tc_action_ops *act,
530 struct pernet_operations *ops)
532 struct tc_action_ops *a;
535 if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
538 /* We have to register pernet ops before making the action ops visible,
539 * otherwise tcf_action_init_1() could get a partially initialized
542 ret = register_pernet_subsys(ops);
546 write_lock(&act_mod_lock);
547 list_for_each_entry(a, &act_base, head) {
548 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
549 write_unlock(&act_mod_lock);
550 unregister_pernet_subsys(ops);
554 list_add_tail(&act->head, &act_base);
555 write_unlock(&act_mod_lock);
559 EXPORT_SYMBOL(tcf_register_action);
561 int tcf_unregister_action(struct tc_action_ops *act,
562 struct pernet_operations *ops)
564 struct tc_action_ops *a;
567 write_lock(&act_mod_lock);
568 list_for_each_entry(a, &act_base, head) {
570 list_del(&act->head);
575 write_unlock(&act_mod_lock);
577 unregister_pernet_subsys(ops);
580 EXPORT_SYMBOL(tcf_unregister_action);
583 static struct tc_action_ops *tc_lookup_action_n(char *kind)
585 struct tc_action_ops *a, *res = NULL;
588 read_lock(&act_mod_lock);
589 list_for_each_entry(a, &act_base, head) {
590 if (strcmp(kind, a->kind) == 0) {
591 if (try_module_get(a->owner))
596 read_unlock(&act_mod_lock);
601 /* lookup by nlattr */
602 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
604 struct tc_action_ops *a, *res = NULL;
607 read_lock(&act_mod_lock);
608 list_for_each_entry(a, &act_base, head) {
609 if (nla_strcmp(kind, a->kind) == 0) {
610 if (try_module_get(a->owner))
615 read_unlock(&act_mod_lock);
620 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
621 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
622 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
623 int nr_actions, struct tcf_result *res)
626 u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
630 if (skb_skip_tc_classify(skb))
634 for (i = 0; i < nr_actions; i++) {
635 const struct tc_action *a = actions[i];
637 if (jmp_prgcnt > 0) {
642 ret = a->ops->act(skb, a, res);
643 if (ret == TC_ACT_REPEAT)
644 goto repeat; /* we need a ttl - JHS */
646 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
647 jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
648 if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
649 /* faulty opcode, stop pipeline */
654 goto restart_act_graph;
655 else /* faulty graph, stop pipeline */
658 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
659 tcf_action_goto_chain_exec(a, res);
662 if (ret != TC_ACT_PIPE)
668 EXPORT_SYMBOL(tcf_action_exec);
670 int tcf_action_destroy(struct tc_action *actions[], int bind)
672 const struct tc_action_ops *ops;
676 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
680 ret = __tcf_idr_release(a, bind, true);
681 if (ret == ACT_P_DELETED)
682 module_put(ops->owner);
689 static int tcf_action_destroy_1(struct tc_action *a, int bind)
691 struct tc_action *actions[] = { a, NULL };
693 return tcf_action_destroy(actions, bind);
696 static int tcf_action_put(struct tc_action *p)
698 return __tcf_action_put(p, false);
701 /* Put all actions in this array, skip those NULL's. */
702 static void tcf_action_put_many(struct tc_action *actions[])
706 for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
707 struct tc_action *a = actions[i];
708 const struct tc_action_ops *ops;
713 if (tcf_action_put(a))
714 module_put(ops->owner);
719 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
721 return a->ops->dump(skb, a, bind, ref);
725 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
728 unsigned char *b = skb_tail_pointer(skb);
730 struct tc_cookie *cookie;
732 if (nla_put_string(skb, TCA_KIND, a->ops->kind))
733 goto nla_put_failure;
734 if (tcf_action_copy_stats(skb, a, 0))
735 goto nla_put_failure;
738 cookie = rcu_dereference(a->act_cookie);
740 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
742 goto nla_put_failure;
747 nest = nla_nest_start(skb, TCA_OPTIONS);
749 goto nla_put_failure;
750 err = tcf_action_dump_old(skb, a, bind, ref);
752 nla_nest_end(skb, nest);
760 EXPORT_SYMBOL(tcf_action_dump_1);
762 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
766 int err = -EINVAL, i;
769 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
771 nest = nla_nest_start(skb, a->order);
773 goto nla_put_failure;
774 err = tcf_action_dump_1(skb, a, bind, ref);
777 nla_nest_end(skb, nest);
785 nla_nest_cancel(skb, nest);
789 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
791 struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
795 c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
800 c->len = nla_len(tb[TCA_ACT_COOKIE]);
805 static bool tcf_action_valid(int action)
807 int opcode = TC_ACT_EXT_OPCODE(action);
810 return action <= TC_ACT_VALUE_MAX;
811 return opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC;
814 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
815 struct nlattr *nla, struct nlattr *est,
816 char *name, int ovr, int bind,
818 struct netlink_ext_ack *extack)
821 struct tc_action_ops *a_o;
822 struct tc_cookie *cookie = NULL;
823 char act_name[IFNAMSIZ];
824 struct nlattr *tb[TCA_ACT_MAX + 1];
829 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
833 kind = tb[TCA_ACT_KIND];
835 NL_SET_ERR_MSG(extack, "TC action kind must be specified");
838 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
839 NL_SET_ERR_MSG(extack, "TC action name too long");
842 if (tb[TCA_ACT_COOKIE]) {
843 int cklen = nla_len(tb[TCA_ACT_COOKIE]);
845 if (cklen > TC_COOKIE_MAX_SIZE) {
846 NL_SET_ERR_MSG(extack, "TC cookie size above the maximum");
850 cookie = nla_memdup_cookie(tb);
852 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
858 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
859 NL_SET_ERR_MSG(extack, "TC action name too long");
865 a_o = tc_lookup_action_n(act_name);
867 #ifdef CONFIG_MODULES
870 request_module("act_%s", act_name);
874 a_o = tc_lookup_action_n(act_name);
876 /* We dropped the RTNL semaphore in order to
877 * perform the module load. So, even if we
878 * succeeded in loading the module we have to
879 * tell the caller to replay the request. We
880 * indicate this using -EAGAIN.
887 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
892 /* backward compatibility for policer */
894 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
897 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
902 if (!name && tb[TCA_ACT_COOKIE])
903 tcf_set_action_cookie(&a->act_cookie, cookie);
905 /* module count goes up only when brand new policy is created
906 * if it exists and is only bound to in a_o->init() then
907 * ACT_P_CREATED is not returned (a zero is).
909 if (err != ACT_P_CREATED)
910 module_put(a_o->owner);
912 if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
913 err = tcf_action_goto_chain_init(a, tp);
915 tcf_action_destroy_1(a, bind);
916 NL_SET_ERR_MSG(extack, "Failed to init TC action chain");
921 if (!tcf_action_valid(a->tcfa_action)) {
922 tcf_action_destroy_1(a, bind);
923 NL_SET_ERR_MSG(extack, "Invalid control action value");
924 return ERR_PTR(-EINVAL);
930 module_put(a_o->owner);
939 /* Returns numbers of initialized actions or negative error. */
941 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
942 struct nlattr *est, char *name, int ovr, int bind,
943 struct tc_action *actions[], size_t *attr_size,
944 bool rtnl_held, struct netlink_ext_ack *extack)
946 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
947 struct tc_action *act;
952 err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
956 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
957 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
964 sz += tcf_action_fill_size(act);
965 /* Start from index 0 */
966 actions[i - 1] = act;
969 *attr_size = tcf_action_full_attrs_size(sz);
973 tcf_action_destroy(actions, bind);
977 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
986 /* compat_mode being true specifies a call that is supposed
987 * to add additional backward compatibility statistic TLVs.
990 if (p->type == TCA_OLD_COMPAT)
991 err = gnet_stats_start_copy_compat(skb, 0,
999 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1000 &p->tcfa_lock, &d, TCA_ACT_PAD);
1005 if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
1006 gnet_stats_copy_basic_hw(NULL, &d, p->cpu_bstats_hw,
1007 &p->tcfa_bstats_hw) < 0 ||
1008 gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1009 gnet_stats_copy_queue(&d, p->cpu_qstats,
1011 p->tcfa_qstats.qlen) < 0)
1014 if (gnet_stats_finish_copy(&d) < 0)
1023 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1024 u32 portid, u32 seq, u16 flags, int event, int bind,
1028 struct nlmsghdr *nlh;
1029 unsigned char *b = skb_tail_pointer(skb);
1030 struct nlattr *nest;
1032 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1034 goto out_nlmsg_trim;
1035 t = nlmsg_data(nlh);
1036 t->tca_family = AF_UNSPEC;
1040 nest = nla_nest_start(skb, TCA_ACT_TAB);
1042 goto out_nlmsg_trim;
1044 if (tcf_action_dump(skb, actions, bind, ref) < 0)
1045 goto out_nlmsg_trim;
1047 nla_nest_end(skb, nest);
1049 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1058 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1059 struct tc_action *actions[], int event,
1060 struct netlink_ext_ack *extack)
1062 struct sk_buff *skb;
1064 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1067 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1069 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1074 return rtnl_unicast(skb, net, portid);
1077 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1078 struct nlmsghdr *n, u32 portid,
1079 struct netlink_ext_ack *extack)
1081 struct nlattr *tb[TCA_ACT_MAX + 1];
1082 const struct tc_action_ops *ops;
1083 struct tc_action *a;
1087 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1092 if (tb[TCA_ACT_INDEX] == NULL ||
1093 nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1094 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1097 index = nla_get_u32(tb[TCA_ACT_INDEX]);
1100 ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1101 if (!ops) { /* could happen in batch of actions */
1102 NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1106 if (ops->lookup(net, &a, index) == 0) {
1107 NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1111 module_put(ops->owner);
1115 module_put(ops->owner);
1117 return ERR_PTR(err);
1120 static int tca_action_flush(struct net *net, struct nlattr *nla,
1121 struct nlmsghdr *n, u32 portid,
1122 struct netlink_ext_ack *extack)
1124 struct sk_buff *skb;
1126 struct nlmsghdr *nlh;
1128 struct netlink_callback dcb;
1129 struct nlattr *nest;
1130 struct nlattr *tb[TCA_ACT_MAX + 1];
1131 const struct tc_action_ops *ops;
1132 struct nlattr *kind;
1135 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1139 b = skb_tail_pointer(skb);
1141 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1146 kind = tb[TCA_ACT_KIND];
1147 ops = tc_lookup_action(kind);
1148 if (!ops) { /*some idjot trying to flush unknown action */
1149 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1153 nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1156 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1157 goto out_module_put;
1159 t = nlmsg_data(nlh);
1160 t->tca_family = AF_UNSPEC;
1164 nest = nla_nest_start(skb, TCA_ACT_TAB);
1166 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1167 goto out_module_put;
1170 err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1172 nla_nest_cancel(skb, nest);
1173 goto out_module_put;
1176 nla_nest_end(skb, nest);
1178 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1179 nlh->nlmsg_flags |= NLM_F_ROOT;
1180 module_put(ops->owner);
1181 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1182 n->nlmsg_flags & NLM_F_ECHO);
1186 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1191 module_put(ops->owner);
1197 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1201 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1202 struct tc_action *a = actions[i];
1203 const struct tc_action_ops *ops = a->ops;
1204 /* Actions can be deleted concurrently so we must save their
1205 * type and id to search again after reference is released.
1207 struct tcf_idrinfo *idrinfo = a->idrinfo;
1208 u32 act_index = a->tcfa_index;
1211 if (tcf_action_put(a)) {
1212 /* last reference, action was deleted concurrently */
1213 module_put(ops->owner);
1217 /* now do the delete */
1218 ret = tcf_idr_delete_index(idrinfo, act_index);
1227 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1228 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1231 struct sk_buff *skb;
1233 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1238 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1240 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1245 /* now do the delete */
1246 ret = tcf_action_delete(net, actions);
1248 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1253 ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1254 n->nlmsg_flags & NLM_F_ECHO);
1261 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1262 u32 portid, int event, struct netlink_ext_ack *extack)
1265 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1266 struct tc_action *act;
1267 size_t attr_size = 0;
1268 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1270 ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
1274 if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1276 return tca_action_flush(net, tb[1], n, portid, extack);
1278 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1282 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1283 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1289 attr_size += tcf_action_fill_size(act);
1290 actions[i - 1] = act;
1293 attr_size = tcf_action_full_attrs_size(attr_size);
1295 if (event == RTM_GETACTION)
1296 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1298 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1304 tcf_action_put_many(actions);
1309 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1310 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1312 struct sk_buff *skb;
1315 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1320 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1321 RTM_NEWACTION, 0, 0) <= 0) {
1322 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1327 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1328 n->nlmsg_flags & NLM_F_ECHO);
1334 static int tcf_action_add(struct net *net, struct nlattr *nla,
1335 struct nlmsghdr *n, u32 portid, int ovr,
1336 struct netlink_ext_ack *extack)
1338 size_t attr_size = 0;
1340 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1342 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, actions,
1343 &attr_size, true, extack);
1346 ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1348 tcf_action_put_many(actions);
1353 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1354 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1355 [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1356 .validation_data = &tcaa_root_flags_allowed },
1357 [TCA_ROOT_TIME_DELTA] = { .type = NLA_U32 },
1360 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1361 struct netlink_ext_ack *extack)
1363 struct net *net = sock_net(skb->sk);
1364 struct nlattr *tca[TCA_ROOT_MAX + 1];
1365 u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1366 int ret = 0, ovr = 0;
1368 if ((n->nlmsg_type != RTM_GETACTION) &&
1369 !netlink_capable(skb, CAP_NET_ADMIN))
1372 ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1377 if (tca[TCA_ACT_TAB] == NULL) {
1378 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1382 /* n->nlmsg_flags & NLM_F_CREATE */
1383 switch (n->nlmsg_type) {
1385 /* we are going to assume all other flags
1386 * imply create only if it doesn't exist
1387 * Note that CREATE | EXCL implies that
1388 * but since we want avoid ambiguity (eg when flags
1389 * is zero) then just set this
1391 if (n->nlmsg_flags & NLM_F_REPLACE)
1394 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1400 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1401 portid, RTM_DELACTION, extack);
1404 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1405 portid, RTM_GETACTION, extack);
1414 static struct nlattr *find_dump_kind(struct nlattr **nla)
1416 struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1417 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1418 struct nlattr *kind;
1420 tb1 = nla[TCA_ACT_TAB];
1424 if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1425 NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1430 if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1432 kind = tb2[TCA_ACT_KIND];
1437 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1439 struct net *net = sock_net(skb->sk);
1440 struct nlmsghdr *nlh;
1441 unsigned char *b = skb_tail_pointer(skb);
1442 struct nlattr *nest;
1443 struct tc_action_ops *a_o;
1445 struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1446 struct nlattr *tb[TCA_ROOT_MAX + 1];
1447 struct nlattr *count_attr = NULL;
1448 unsigned long jiffy_since = 0;
1449 struct nlattr *kind = NULL;
1450 struct nla_bitfield32 bf;
1451 u32 msecs_since = 0;
1454 ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1455 tcaa_policy, cb->extack);
1459 kind = find_dump_kind(tb);
1461 pr_info("tc_dump_action: action bad kind\n");
1465 a_o = tc_lookup_action(kind);
1470 if (tb[TCA_ROOT_FLAGS]) {
1471 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1472 cb->args[2] = bf.value;
1475 if (tb[TCA_ROOT_TIME_DELTA]) {
1476 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1479 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1480 cb->nlh->nlmsg_type, sizeof(*t), 0);
1482 goto out_module_put;
1485 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1487 t = nlmsg_data(nlh);
1488 t->tca_family = AF_UNSPEC;
1491 cb->args[3] = jiffy_since;
1492 count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1494 goto out_module_put;
1496 nest = nla_nest_start(skb, TCA_ACT_TAB);
1498 goto out_module_put;
1500 ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1502 goto out_module_put;
1505 nla_nest_end(skb, nest);
1507 act_count = cb->args[1];
1508 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1513 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1514 if (NETLINK_CB(cb->skb).portid && ret)
1515 nlh->nlmsg_flags |= NLM_F_MULTI;
1516 module_put(a_o->owner);
1520 module_put(a_o->owner);
1525 struct tcf_action_net {
1526 struct rhashtable egdev_ht;
1529 static unsigned int tcf_action_net_id;
1531 struct tcf_action_egdev_cb {
1532 struct list_head list;
1537 struct tcf_action_egdev {
1538 struct rhash_head ht_node;
1539 const struct net_device *dev;
1540 unsigned int refcnt;
1541 struct list_head cb_list;
1544 static const struct rhashtable_params tcf_action_egdev_ht_params = {
1545 .key_offset = offsetof(struct tcf_action_egdev, dev),
1546 .head_offset = offsetof(struct tcf_action_egdev, ht_node),
1547 .key_len = sizeof(const struct net_device *),
1550 static struct tcf_action_egdev *
1551 tcf_action_egdev_lookup(const struct net_device *dev)
1553 struct net *net = dev_net(dev);
1554 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1556 return rhashtable_lookup_fast(&tan->egdev_ht, &dev,
1557 tcf_action_egdev_ht_params);
1560 static struct tcf_action_egdev *
1561 tcf_action_egdev_get(const struct net_device *dev)
1563 struct tcf_action_egdev *egdev;
1564 struct tcf_action_net *tan;
1566 egdev = tcf_action_egdev_lookup(dev);
1570 egdev = kzalloc(sizeof(*egdev), GFP_KERNEL);
1573 INIT_LIST_HEAD(&egdev->cb_list);
1575 tan = net_generic(dev_net(dev), tcf_action_net_id);
1576 rhashtable_insert_fast(&tan->egdev_ht, &egdev->ht_node,
1577 tcf_action_egdev_ht_params);
1584 static void tcf_action_egdev_put(struct tcf_action_egdev *egdev)
1586 struct tcf_action_net *tan;
1588 if (--egdev->refcnt)
1590 tan = net_generic(dev_net(egdev->dev), tcf_action_net_id);
1591 rhashtable_remove_fast(&tan->egdev_ht, &egdev->ht_node,
1592 tcf_action_egdev_ht_params);
1596 static struct tcf_action_egdev_cb *
1597 tcf_action_egdev_cb_lookup(struct tcf_action_egdev *egdev,
1598 tc_setup_cb_t *cb, void *cb_priv)
1600 struct tcf_action_egdev_cb *egdev_cb;
1602 list_for_each_entry(egdev_cb, &egdev->cb_list, list)
1603 if (egdev_cb->cb == cb && egdev_cb->cb_priv == cb_priv)
1608 static int tcf_action_egdev_cb_call(struct tcf_action_egdev *egdev,
1609 enum tc_setup_type type,
1610 void *type_data, bool err_stop)
1612 struct tcf_action_egdev_cb *egdev_cb;
1616 list_for_each_entry(egdev_cb, &egdev->cb_list, list) {
1617 err = egdev_cb->cb(type, type_data, egdev_cb->cb_priv);
1628 static int tcf_action_egdev_cb_add(struct tcf_action_egdev *egdev,
1629 tc_setup_cb_t *cb, void *cb_priv)
1631 struct tcf_action_egdev_cb *egdev_cb;
1633 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1634 if (WARN_ON(egdev_cb))
1636 egdev_cb = kzalloc(sizeof(*egdev_cb), GFP_KERNEL);
1640 egdev_cb->cb_priv = cb_priv;
1641 list_add(&egdev_cb->list, &egdev->cb_list);
1645 static void tcf_action_egdev_cb_del(struct tcf_action_egdev *egdev,
1646 tc_setup_cb_t *cb, void *cb_priv)
1648 struct tcf_action_egdev_cb *egdev_cb;
1650 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1651 if (WARN_ON(!egdev_cb))
1653 list_del(&egdev_cb->list);
1657 static int __tc_setup_cb_egdev_register(const struct net_device *dev,
1658 tc_setup_cb_t *cb, void *cb_priv)
1660 struct tcf_action_egdev *egdev = tcf_action_egdev_get(dev);
1665 err = tcf_action_egdev_cb_add(egdev, cb, cb_priv);
1671 tcf_action_egdev_put(egdev);
1674 int tc_setup_cb_egdev_register(const struct net_device *dev,
1675 tc_setup_cb_t *cb, void *cb_priv)
1680 err = __tc_setup_cb_egdev_register(dev, cb, cb_priv);
1684 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_register);
1686 static void __tc_setup_cb_egdev_unregister(const struct net_device *dev,
1687 tc_setup_cb_t *cb, void *cb_priv)
1689 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1691 if (WARN_ON(!egdev))
1693 tcf_action_egdev_cb_del(egdev, cb, cb_priv);
1694 tcf_action_egdev_put(egdev);
1696 void tc_setup_cb_egdev_unregister(const struct net_device *dev,
1697 tc_setup_cb_t *cb, void *cb_priv)
1700 __tc_setup_cb_egdev_unregister(dev, cb, cb_priv);
1703 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_unregister);
1705 int tc_setup_cb_egdev_call(const struct net_device *dev,
1706 enum tc_setup_type type, void *type_data,
1709 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1713 return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
1715 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
1717 static __net_init int tcf_action_net_init(struct net *net)
1719 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1721 return rhashtable_init(&tan->egdev_ht, &tcf_action_egdev_ht_params);
1724 static void __net_exit tcf_action_net_exit(struct net *net)
1726 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1728 rhashtable_destroy(&tan->egdev_ht);
1731 static struct pernet_operations tcf_action_net_ops = {
1732 .init = tcf_action_net_init,
1733 .exit = tcf_action_net_exit,
1734 .id = &tcf_action_net_id,
1735 .size = sizeof(struct tcf_action_net),
1738 static int __init tc_action_init(void)
1742 err = register_pernet_subsys(&tcf_action_net_ops);
1746 rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1747 rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1748 rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1754 subsys_initcall(tc_action_init);