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_qstats);
86 tcf_set_action_cookie(&p->act_cookie, NULL);
88 tcf_action_goto_chain_fini(p);
93 static void tcf_action_cleanup(struct tc_action *p)
98 gen_kill_estimator(&p->tcfa_rate_est);
102 static int __tcf_action_put(struct tc_action *p, bool bind)
104 struct tcf_idrinfo *idrinfo = p->idrinfo;
106 if (refcount_dec_and_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
108 atomic_dec(&p->tcfa_bindcnt);
109 idr_remove(&idrinfo->action_idr, p->tcfa_index);
110 spin_unlock(&idrinfo->lock);
112 tcf_action_cleanup(p);
117 atomic_dec(&p->tcfa_bindcnt);
122 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
126 /* Release with strict==1 and bind==0 is only called through act API
127 * interface (classifiers always bind). Only case when action with
128 * positive reference count and zero bind count can exist is when it was
129 * also created with act API (unbinding last classifier will destroy the
130 * action if it was created by classifier). So only case when bind count
131 * can be changed after initial check is when unbound action is
132 * destroyed by act API while classifier binds to action with same id
133 * concurrently. This result either creation of new action(same behavior
134 * as before), or reusing existing action if concurrent process
135 * increments reference count before action is deleted. Both scenarios
139 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
142 if (__tcf_action_put(p, bind))
148 EXPORT_SYMBOL(__tcf_idr_release);
150 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
152 struct tc_cookie *act_cookie;
156 act_cookie = rcu_dereference(act->act_cookie);
159 cookie_len = nla_total_size(act_cookie->len);
162 return nla_total_size(0) /* action number nested */
163 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
164 + cookie_len /* TCA_ACT_COOKIE */
165 + nla_total_size(0) /* TCA_ACT_STATS nested */
166 /* TCA_STATS_BASIC */
167 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
168 /* TCA_STATS_QUEUE */
169 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
170 + nla_total_size(0) /* TCA_OPTIONS nested */
171 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
174 static size_t tcf_action_full_attrs_size(size_t sz)
176 return NLMSG_HDRLEN /* struct nlmsghdr */
177 + sizeof(struct tcamsg)
178 + nla_total_size(0) /* TCA_ACT_TAB nested */
182 static size_t tcf_action_fill_size(const struct tc_action *act)
184 size_t sz = tcf_action_shared_attrs_size(act);
186 if (act->ops->get_fill_size)
187 return act->ops->get_fill_size(act) + sz;
191 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
192 struct netlink_callback *cb)
194 int err = 0, index = -1, s_i = 0, n_i = 0;
195 u32 act_flags = cb->args[2];
196 unsigned long jiffy_since = cb->args[3];
198 struct idr *idr = &idrinfo->action_idr;
200 unsigned long id = 1;
202 spin_lock(&idrinfo->lock);
206 idr_for_each_entry_ul(idr, p, id) {
212 time_after(jiffy_since,
213 (unsigned long)p->tcfa_tm.lastuse))
216 nest = nla_nest_start(skb, n_i);
219 goto nla_put_failure;
221 err = tcf_action_dump_1(skb, p, 0, 0);
224 nlmsg_trim(skb, nest);
227 nla_nest_end(skb, nest);
229 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
230 n_i >= TCA_ACT_MAX_PRIO)
235 cb->args[0] = index + 1;
237 spin_unlock(&idrinfo->lock);
239 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
245 nla_nest_cancel(skb, nest);
249 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
250 const struct tc_action_ops *ops)
255 struct idr *idr = &idrinfo->action_idr;
257 unsigned long id = 1;
259 nest = nla_nest_start(skb, 0);
261 goto nla_put_failure;
262 if (nla_put_string(skb, TCA_KIND, ops->kind))
263 goto nla_put_failure;
265 idr_for_each_entry_ul(idr, p, id) {
266 ret = __tcf_idr_release(p, false, true);
267 if (ret == ACT_P_DELETED) {
268 module_put(ops->owner);
270 } else if (ret < 0) {
271 goto nla_put_failure;
274 if (nla_put_u32(skb, TCA_FCNT, n_i))
275 goto nla_put_failure;
276 nla_nest_end(skb, nest);
280 nla_nest_cancel(skb, nest);
284 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
285 struct netlink_callback *cb, int type,
286 const struct tc_action_ops *ops,
287 struct netlink_ext_ack *extack)
289 struct tcf_idrinfo *idrinfo = tn->idrinfo;
291 if (type == RTM_DELACTION) {
292 return tcf_del_walker(idrinfo, skb, ops);
293 } else if (type == RTM_GETACTION) {
294 return tcf_dump_walker(idrinfo, skb, cb);
296 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
297 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
301 EXPORT_SYMBOL(tcf_generic_walker);
303 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
305 struct tcf_idrinfo *idrinfo = tn->idrinfo;
308 spin_lock(&idrinfo->lock);
309 p = idr_find(&idrinfo->action_idr, index);
313 refcount_inc(&p->tcfa_refcnt);
314 spin_unlock(&idrinfo->lock);
322 EXPORT_SYMBOL(tcf_idr_search);
324 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
329 spin_lock(&idrinfo->lock);
330 p = idr_find(&idrinfo->action_idr, index);
332 spin_unlock(&idrinfo->lock);
336 if (!atomic_read(&p->tcfa_bindcnt)) {
337 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
338 struct module *owner = p->ops->owner;
340 WARN_ON(p != idr_remove(&idrinfo->action_idr,
342 spin_unlock(&idrinfo->lock);
344 tcf_action_cleanup(p);
353 spin_unlock(&idrinfo->lock);
357 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
358 struct tc_action **a, const struct tc_action_ops *ops,
359 int bind, bool cpustats)
361 struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
362 struct tcf_idrinfo *idrinfo = tn->idrinfo;
367 refcount_set(&p->tcfa_refcnt, 1);
369 atomic_set(&p->tcfa_bindcnt, 1);
372 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
375 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
379 spin_lock_init(&p->tcfa_lock);
380 p->tcfa_index = index;
381 p->tcfa_tm.install = jiffies;
382 p->tcfa_tm.lastuse = jiffies;
383 p->tcfa_tm.firstuse = 0;
385 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
387 &p->tcfa_lock, NULL, est);
392 p->idrinfo = idrinfo;
397 free_percpu(p->cpu_qstats);
399 free_percpu(p->cpu_bstats);
404 EXPORT_SYMBOL(tcf_idr_create);
406 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
408 struct tcf_idrinfo *idrinfo = tn->idrinfo;
410 spin_lock(&idrinfo->lock);
411 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
412 WARN_ON(!IS_ERR(idr_replace(&idrinfo->action_idr, a, a->tcfa_index)));
413 spin_unlock(&idrinfo->lock);
415 EXPORT_SYMBOL(tcf_idr_insert);
417 /* Cleanup idr index that was allocated but not initialized. */
419 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
421 struct tcf_idrinfo *idrinfo = tn->idrinfo;
423 spin_lock(&idrinfo->lock);
424 /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
425 WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
426 spin_unlock(&idrinfo->lock);
428 EXPORT_SYMBOL(tcf_idr_cleanup);
430 /* Check if action with specified index exists. If actions is found, increments
431 * its reference and bind counters, and return 1. Otherwise insert temporary
432 * error pointer (to prevent concurrent users from inserting actions with same
433 * index) and return 0.
436 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
437 struct tc_action **a, int bind)
439 struct tcf_idrinfo *idrinfo = tn->idrinfo;
444 spin_lock(&idrinfo->lock);
446 p = idr_find(&idrinfo->action_idr, *index);
448 /* This means that another process allocated
449 * index but did not assign the pointer yet.
451 spin_unlock(&idrinfo->lock);
456 refcount_inc(&p->tcfa_refcnt);
458 atomic_inc(&p->tcfa_bindcnt);
463 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
466 idr_replace(&idrinfo->action_idr,
467 ERR_PTR(-EBUSY), *index);
472 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
473 UINT_MAX, GFP_ATOMIC);
475 idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
478 spin_unlock(&idrinfo->lock);
481 EXPORT_SYMBOL(tcf_idr_check_alloc);
483 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
484 struct tcf_idrinfo *idrinfo)
486 struct idr *idr = &idrinfo->action_idr;
489 unsigned long id = 1;
491 idr_for_each_entry_ul(idr, p, id) {
492 ret = __tcf_idr_release(p, false, true);
493 if (ret == ACT_P_DELETED)
494 module_put(ops->owner);
498 idr_destroy(&idrinfo->action_idr);
500 EXPORT_SYMBOL(tcf_idrinfo_destroy);
502 static LIST_HEAD(act_base);
503 static DEFINE_RWLOCK(act_mod_lock);
505 int tcf_register_action(struct tc_action_ops *act,
506 struct pernet_operations *ops)
508 struct tc_action_ops *a;
511 if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
514 /* We have to register pernet ops before making the action ops visible,
515 * otherwise tcf_action_init_1() could get a partially initialized
518 ret = register_pernet_subsys(ops);
522 write_lock(&act_mod_lock);
523 list_for_each_entry(a, &act_base, head) {
524 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
525 write_unlock(&act_mod_lock);
526 unregister_pernet_subsys(ops);
530 list_add_tail(&act->head, &act_base);
531 write_unlock(&act_mod_lock);
535 EXPORT_SYMBOL(tcf_register_action);
537 int tcf_unregister_action(struct tc_action_ops *act,
538 struct pernet_operations *ops)
540 struct tc_action_ops *a;
543 write_lock(&act_mod_lock);
544 list_for_each_entry(a, &act_base, head) {
546 list_del(&act->head);
551 write_unlock(&act_mod_lock);
553 unregister_pernet_subsys(ops);
556 EXPORT_SYMBOL(tcf_unregister_action);
559 static struct tc_action_ops *tc_lookup_action_n(char *kind)
561 struct tc_action_ops *a, *res = NULL;
564 read_lock(&act_mod_lock);
565 list_for_each_entry(a, &act_base, head) {
566 if (strcmp(kind, a->kind) == 0) {
567 if (try_module_get(a->owner))
572 read_unlock(&act_mod_lock);
577 /* lookup by nlattr */
578 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
580 struct tc_action_ops *a, *res = NULL;
583 read_lock(&act_mod_lock);
584 list_for_each_entry(a, &act_base, head) {
585 if (nla_strcmp(kind, a->kind) == 0) {
586 if (try_module_get(a->owner))
591 read_unlock(&act_mod_lock);
596 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
597 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
598 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
599 int nr_actions, struct tcf_result *res)
602 u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
606 if (skb_skip_tc_classify(skb))
610 for (i = 0; i < nr_actions; i++) {
611 const struct tc_action *a = actions[i];
614 if (jmp_prgcnt > 0) {
621 ret = a->ops->act(skb, a, res);
623 if (unlikely(ret == TC_ACT_REPEAT)) {
624 if (--repeat_ttl != 0)
626 /* suspicious opcode, stop pipeline */
627 net_warn_ratelimited("TC_ACT_REPEAT abuse ?\n");
631 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
632 jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
633 if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
634 /* faulty opcode, stop pipeline */
639 goto restart_act_graph;
640 else /* faulty graph, stop pipeline */
643 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
644 tcf_action_goto_chain_exec(a, res);
647 if (ret != TC_ACT_PIPE)
653 EXPORT_SYMBOL(tcf_action_exec);
655 int tcf_action_destroy(struct tc_action *actions[], int bind)
657 const struct tc_action_ops *ops;
661 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
665 ret = __tcf_idr_release(a, bind, true);
666 if (ret == ACT_P_DELETED)
667 module_put(ops->owner);
674 static int tcf_action_destroy_1(struct tc_action *a, int bind)
676 struct tc_action *actions[] = { a, NULL };
678 return tcf_action_destroy(actions, bind);
681 static int tcf_action_put(struct tc_action *p)
683 return __tcf_action_put(p, false);
686 /* Put all actions in this array, skip those NULL's. */
687 static void tcf_action_put_many(struct tc_action *actions[])
691 for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
692 struct tc_action *a = actions[i];
693 const struct tc_action_ops *ops;
698 if (tcf_action_put(a))
699 module_put(ops->owner);
704 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
706 return a->ops->dump(skb, a, bind, ref);
710 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
713 unsigned char *b = skb_tail_pointer(skb);
715 struct tc_cookie *cookie;
717 if (nla_put_string(skb, TCA_KIND, a->ops->kind))
718 goto nla_put_failure;
719 if (tcf_action_copy_stats(skb, a, 0))
720 goto nla_put_failure;
723 cookie = rcu_dereference(a->act_cookie);
725 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
727 goto nla_put_failure;
732 nest = nla_nest_start(skb, TCA_OPTIONS);
734 goto nla_put_failure;
735 err = tcf_action_dump_old(skb, a, bind, ref);
737 nla_nest_end(skb, nest);
745 EXPORT_SYMBOL(tcf_action_dump_1);
747 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
751 int err = -EINVAL, i;
754 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
756 nest = nla_nest_start(skb, i + 1);
758 goto nla_put_failure;
759 err = tcf_action_dump_1(skb, a, bind, ref);
762 nla_nest_end(skb, nest);
770 nla_nest_cancel(skb, nest);
774 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
776 struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
780 c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
785 c->len = nla_len(tb[TCA_ACT_COOKIE]);
790 static bool tcf_action_valid(int action)
792 int opcode = TC_ACT_EXT_OPCODE(action);
795 return action <= TC_ACT_VALUE_MAX;
796 return opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC;
799 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
800 struct nlattr *nla, struct nlattr *est,
801 char *name, int ovr, int bind,
803 struct netlink_ext_ack *extack)
806 struct tc_action_ops *a_o;
807 struct tc_cookie *cookie = NULL;
808 char act_name[IFNAMSIZ];
809 struct nlattr *tb[TCA_ACT_MAX + 1];
814 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
818 kind = tb[TCA_ACT_KIND];
820 NL_SET_ERR_MSG(extack, "TC action kind must be specified");
823 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
824 NL_SET_ERR_MSG(extack, "TC action name too long");
827 if (tb[TCA_ACT_COOKIE]) {
828 int cklen = nla_len(tb[TCA_ACT_COOKIE]);
830 if (cklen > TC_COOKIE_MAX_SIZE) {
831 NL_SET_ERR_MSG(extack, "TC cookie size above the maximum");
835 cookie = nla_memdup_cookie(tb);
837 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
843 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
844 NL_SET_ERR_MSG(extack, "TC action name too long");
850 a_o = tc_lookup_action_n(act_name);
852 #ifdef CONFIG_MODULES
855 request_module("act_%s", act_name);
859 a_o = tc_lookup_action_n(act_name);
861 /* We dropped the RTNL semaphore in order to
862 * perform the module load. So, even if we
863 * succeeded in loading the module we have to
864 * tell the caller to replay the request. We
865 * indicate this using -EAGAIN.
872 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
877 /* backward compatibility for policer */
879 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
882 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
887 if (!name && tb[TCA_ACT_COOKIE])
888 tcf_set_action_cookie(&a->act_cookie, cookie);
890 /* module count goes up only when brand new policy is created
891 * if it exists and is only bound to in a_o->init() then
892 * ACT_P_CREATED is not returned (a zero is).
894 if (err != ACT_P_CREATED)
895 module_put(a_o->owner);
897 if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
898 err = tcf_action_goto_chain_init(a, tp);
900 tcf_action_destroy_1(a, bind);
901 NL_SET_ERR_MSG(extack, "Failed to init TC action chain");
906 if (!tcf_action_valid(a->tcfa_action)) {
907 tcf_action_destroy_1(a, bind);
908 NL_SET_ERR_MSG(extack, "Invalid control action value");
909 return ERR_PTR(-EINVAL);
912 if (!bind && ovr && err == ACT_P_CREATED)
913 refcount_set(&a->tcfa_refcnt, 2);
918 module_put(a_o->owner);
927 /* Returns numbers of initialized actions or negative error. */
929 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
930 struct nlattr *est, char *name, int ovr, int bind,
931 struct tc_action *actions[], size_t *attr_size,
932 bool rtnl_held, struct netlink_ext_ack *extack)
934 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
935 struct tc_action *act;
940 err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
944 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
945 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
952 sz += tcf_action_fill_size(act);
953 /* Start from index 0 */
954 actions[i - 1] = act;
957 *attr_size = tcf_action_full_attrs_size(sz);
961 tcf_action_destroy(actions, bind);
965 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
974 /* compat_mode being true specifies a call that is supposed
975 * to add additional backward compatibility statistic TLVs.
978 if (p->type == TCA_OLD_COMPAT)
979 err = gnet_stats_start_copy_compat(skb, 0,
987 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
988 &p->tcfa_lock, &d, TCA_ACT_PAD);
993 if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
994 gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
995 gnet_stats_copy_queue(&d, p->cpu_qstats,
997 p->tcfa_qstats.qlen) < 0)
1000 if (gnet_stats_finish_copy(&d) < 0)
1009 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1010 u32 portid, u32 seq, u16 flags, int event, int bind,
1014 struct nlmsghdr *nlh;
1015 unsigned char *b = skb_tail_pointer(skb);
1016 struct nlattr *nest;
1018 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1020 goto out_nlmsg_trim;
1021 t = nlmsg_data(nlh);
1022 t->tca_family = AF_UNSPEC;
1026 nest = nla_nest_start(skb, TCA_ACT_TAB);
1028 goto out_nlmsg_trim;
1030 if (tcf_action_dump(skb, actions, bind, ref) < 0)
1031 goto out_nlmsg_trim;
1033 nla_nest_end(skb, nest);
1035 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1044 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1045 struct tc_action *actions[], int event,
1046 struct netlink_ext_ack *extack)
1048 struct sk_buff *skb;
1050 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1053 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1055 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1060 return rtnl_unicast(skb, net, portid);
1063 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1064 struct nlmsghdr *n, u32 portid,
1065 struct netlink_ext_ack *extack)
1067 struct nlattr *tb[TCA_ACT_MAX + 1];
1068 const struct tc_action_ops *ops;
1069 struct tc_action *a;
1073 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1078 if (tb[TCA_ACT_INDEX] == NULL ||
1079 nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1080 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1083 index = nla_get_u32(tb[TCA_ACT_INDEX]);
1086 ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1087 if (!ops) { /* could happen in batch of actions */
1088 NL_SET_ERR_MSG(extack, "Specified TC action not found");
1092 if (ops->lookup(net, &a, index, extack) == 0)
1095 module_put(ops->owner);
1099 module_put(ops->owner);
1101 return ERR_PTR(err);
1104 static int tca_action_flush(struct net *net, struct nlattr *nla,
1105 struct nlmsghdr *n, u32 portid,
1106 struct netlink_ext_ack *extack)
1108 struct sk_buff *skb;
1110 struct nlmsghdr *nlh;
1112 struct netlink_callback dcb;
1113 struct nlattr *nest;
1114 struct nlattr *tb[TCA_ACT_MAX + 1];
1115 const struct tc_action_ops *ops;
1116 struct nlattr *kind;
1119 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1123 b = skb_tail_pointer(skb);
1125 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1130 kind = tb[TCA_ACT_KIND];
1131 ops = tc_lookup_action(kind);
1132 if (!ops) { /*some idjot trying to flush unknown action */
1133 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1137 nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1140 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1141 goto out_module_put;
1143 t = nlmsg_data(nlh);
1144 t->tca_family = AF_UNSPEC;
1148 nest = nla_nest_start(skb, TCA_ACT_TAB);
1150 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1151 goto out_module_put;
1154 err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1156 nla_nest_cancel(skb, nest);
1157 goto out_module_put;
1160 nla_nest_end(skb, nest);
1162 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1163 nlh->nlmsg_flags |= NLM_F_ROOT;
1164 module_put(ops->owner);
1165 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1166 n->nlmsg_flags & NLM_F_ECHO);
1170 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1175 module_put(ops->owner);
1181 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1185 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1186 struct tc_action *a = actions[i];
1187 const struct tc_action_ops *ops = a->ops;
1188 /* Actions can be deleted concurrently so we must save their
1189 * type and id to search again after reference is released.
1191 struct tcf_idrinfo *idrinfo = a->idrinfo;
1192 u32 act_index = a->tcfa_index;
1195 if (tcf_action_put(a)) {
1196 /* last reference, action was deleted concurrently */
1197 module_put(ops->owner);
1201 /* now do the delete */
1202 ret = tcf_idr_delete_index(idrinfo, act_index);
1211 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1212 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1215 struct sk_buff *skb;
1217 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1222 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1224 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1229 /* now do the delete */
1230 ret = tcf_action_delete(net, actions);
1232 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1237 ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1238 n->nlmsg_flags & NLM_F_ECHO);
1245 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1246 u32 portid, int event, struct netlink_ext_ack *extack)
1249 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1250 struct tc_action *act;
1251 size_t attr_size = 0;
1252 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1254 ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
1258 if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1260 return tca_action_flush(net, tb[1], n, portid, extack);
1262 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1266 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1267 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1272 attr_size += tcf_action_fill_size(act);
1273 actions[i - 1] = act;
1276 attr_size = tcf_action_full_attrs_size(attr_size);
1278 if (event == RTM_GETACTION)
1279 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1281 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1287 tcf_action_put_many(actions);
1292 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1293 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1295 struct sk_buff *skb;
1298 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1303 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1304 RTM_NEWACTION, 0, 0) <= 0) {
1305 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1310 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1311 n->nlmsg_flags & NLM_F_ECHO);
1317 static int tcf_action_add(struct net *net, struct nlattr *nla,
1318 struct nlmsghdr *n, u32 portid, int ovr,
1319 struct netlink_ext_ack *extack)
1321 size_t attr_size = 0;
1323 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1325 for (loop = 0; loop < 10; loop++) {
1326 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0,
1327 actions, &attr_size, true, extack);
1334 ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1336 tcf_action_put_many(actions);
1341 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1342 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1343 [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1344 .validation_data = &tcaa_root_flags_allowed },
1345 [TCA_ROOT_TIME_DELTA] = { .type = NLA_U32 },
1348 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1349 struct netlink_ext_ack *extack)
1351 struct net *net = sock_net(skb->sk);
1352 struct nlattr *tca[TCA_ROOT_MAX + 1];
1353 u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1354 int ret = 0, ovr = 0;
1356 if ((n->nlmsg_type != RTM_GETACTION) &&
1357 !netlink_capable(skb, CAP_NET_ADMIN))
1360 ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1365 if (tca[TCA_ACT_TAB] == NULL) {
1366 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1370 /* n->nlmsg_flags & NLM_F_CREATE */
1371 switch (n->nlmsg_type) {
1373 /* we are going to assume all other flags
1374 * imply create only if it doesn't exist
1375 * Note that CREATE | EXCL implies that
1376 * but since we want avoid ambiguity (eg when flags
1377 * is zero) then just set this
1379 if (n->nlmsg_flags & NLM_F_REPLACE)
1381 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1385 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1386 portid, RTM_DELACTION, extack);
1389 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1390 portid, RTM_GETACTION, extack);
1399 static struct nlattr *find_dump_kind(struct nlattr **nla)
1401 struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1402 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1403 struct nlattr *kind;
1405 tb1 = nla[TCA_ACT_TAB];
1409 if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1410 NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1415 if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1417 kind = tb2[TCA_ACT_KIND];
1422 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1424 struct net *net = sock_net(skb->sk);
1425 struct nlmsghdr *nlh;
1426 unsigned char *b = skb_tail_pointer(skb);
1427 struct nlattr *nest;
1428 struct tc_action_ops *a_o;
1430 struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1431 struct nlattr *tb[TCA_ROOT_MAX + 1];
1432 struct nlattr *count_attr = NULL;
1433 unsigned long jiffy_since = 0;
1434 struct nlattr *kind = NULL;
1435 struct nla_bitfield32 bf;
1436 u32 msecs_since = 0;
1439 ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1444 kind = find_dump_kind(tb);
1446 pr_info("tc_dump_action: action bad kind\n");
1450 a_o = tc_lookup_action(kind);
1455 if (tb[TCA_ROOT_FLAGS]) {
1456 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1457 cb->args[2] = bf.value;
1460 if (tb[TCA_ROOT_TIME_DELTA]) {
1461 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1464 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1465 cb->nlh->nlmsg_type, sizeof(*t), 0);
1467 goto out_module_put;
1470 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1472 t = nlmsg_data(nlh);
1473 t->tca_family = AF_UNSPEC;
1476 cb->args[3] = jiffy_since;
1477 count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1479 goto out_module_put;
1481 nest = nla_nest_start(skb, TCA_ACT_TAB);
1483 goto out_module_put;
1485 ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1487 goto out_module_put;
1490 nla_nest_end(skb, nest);
1492 act_count = cb->args[1];
1493 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1498 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1499 if (NETLINK_CB(cb->skb).portid && ret)
1500 nlh->nlmsg_flags |= NLM_F_MULTI;
1501 module_put(a_o->owner);
1505 module_put(a_o->owner);
1510 struct tcf_action_net {
1511 struct rhashtable egdev_ht;
1514 static unsigned int tcf_action_net_id;
1516 struct tcf_action_egdev_cb {
1517 struct list_head list;
1522 struct tcf_action_egdev {
1523 struct rhash_head ht_node;
1524 const struct net_device *dev;
1525 unsigned int refcnt;
1526 struct list_head cb_list;
1529 static const struct rhashtable_params tcf_action_egdev_ht_params = {
1530 .key_offset = offsetof(struct tcf_action_egdev, dev),
1531 .head_offset = offsetof(struct tcf_action_egdev, ht_node),
1532 .key_len = sizeof(const struct net_device *),
1535 static struct tcf_action_egdev *
1536 tcf_action_egdev_lookup(const struct net_device *dev)
1538 struct net *net = dev_net(dev);
1539 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1541 return rhashtable_lookup_fast(&tan->egdev_ht, &dev,
1542 tcf_action_egdev_ht_params);
1545 static struct tcf_action_egdev *
1546 tcf_action_egdev_get(const struct net_device *dev)
1548 struct tcf_action_egdev *egdev;
1549 struct tcf_action_net *tan;
1551 egdev = tcf_action_egdev_lookup(dev);
1555 egdev = kzalloc(sizeof(*egdev), GFP_KERNEL);
1558 INIT_LIST_HEAD(&egdev->cb_list);
1560 tan = net_generic(dev_net(dev), tcf_action_net_id);
1561 rhashtable_insert_fast(&tan->egdev_ht, &egdev->ht_node,
1562 tcf_action_egdev_ht_params);
1569 static void tcf_action_egdev_put(struct tcf_action_egdev *egdev)
1571 struct tcf_action_net *tan;
1573 if (--egdev->refcnt)
1575 tan = net_generic(dev_net(egdev->dev), tcf_action_net_id);
1576 rhashtable_remove_fast(&tan->egdev_ht, &egdev->ht_node,
1577 tcf_action_egdev_ht_params);
1581 static struct tcf_action_egdev_cb *
1582 tcf_action_egdev_cb_lookup(struct tcf_action_egdev *egdev,
1583 tc_setup_cb_t *cb, void *cb_priv)
1585 struct tcf_action_egdev_cb *egdev_cb;
1587 list_for_each_entry(egdev_cb, &egdev->cb_list, list)
1588 if (egdev_cb->cb == cb && egdev_cb->cb_priv == cb_priv)
1593 static int tcf_action_egdev_cb_call(struct tcf_action_egdev *egdev,
1594 enum tc_setup_type type,
1595 void *type_data, bool err_stop)
1597 struct tcf_action_egdev_cb *egdev_cb;
1601 list_for_each_entry(egdev_cb, &egdev->cb_list, list) {
1602 err = egdev_cb->cb(type, type_data, egdev_cb->cb_priv);
1613 static int tcf_action_egdev_cb_add(struct tcf_action_egdev *egdev,
1614 tc_setup_cb_t *cb, void *cb_priv)
1616 struct tcf_action_egdev_cb *egdev_cb;
1618 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1619 if (WARN_ON(egdev_cb))
1621 egdev_cb = kzalloc(sizeof(*egdev_cb), GFP_KERNEL);
1625 egdev_cb->cb_priv = cb_priv;
1626 list_add(&egdev_cb->list, &egdev->cb_list);
1630 static void tcf_action_egdev_cb_del(struct tcf_action_egdev *egdev,
1631 tc_setup_cb_t *cb, void *cb_priv)
1633 struct tcf_action_egdev_cb *egdev_cb;
1635 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1636 if (WARN_ON(!egdev_cb))
1638 list_del(&egdev_cb->list);
1642 static int __tc_setup_cb_egdev_register(const struct net_device *dev,
1643 tc_setup_cb_t *cb, void *cb_priv)
1645 struct tcf_action_egdev *egdev = tcf_action_egdev_get(dev);
1650 err = tcf_action_egdev_cb_add(egdev, cb, cb_priv);
1656 tcf_action_egdev_put(egdev);
1659 int tc_setup_cb_egdev_register(const struct net_device *dev,
1660 tc_setup_cb_t *cb, void *cb_priv)
1665 err = __tc_setup_cb_egdev_register(dev, cb, cb_priv);
1669 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_register);
1671 static void __tc_setup_cb_egdev_unregister(const struct net_device *dev,
1672 tc_setup_cb_t *cb, void *cb_priv)
1674 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1676 if (WARN_ON(!egdev))
1678 tcf_action_egdev_cb_del(egdev, cb, cb_priv);
1679 tcf_action_egdev_put(egdev);
1681 void tc_setup_cb_egdev_unregister(const struct net_device *dev,
1682 tc_setup_cb_t *cb, void *cb_priv)
1685 __tc_setup_cb_egdev_unregister(dev, cb, cb_priv);
1688 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_unregister);
1690 int tc_setup_cb_egdev_call(const struct net_device *dev,
1691 enum tc_setup_type type, void *type_data,
1694 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1698 return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
1700 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
1702 static __net_init int tcf_action_net_init(struct net *net)
1704 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1706 return rhashtable_init(&tan->egdev_ht, &tcf_action_egdev_ht_params);
1709 static void __net_exit tcf_action_net_exit(struct net *net)
1711 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1713 rhashtable_destroy(&tan->egdev_ht);
1716 static struct pernet_operations tcf_action_net_ops = {
1717 .init = tcf_action_net_init,
1718 .exit = tcf_action_net_exit,
1719 .id = &tcf_action_net_id,
1720 .size = sizeof(struct tcf_action_net),
1723 static int __init tc_action_init(void)
1727 err = register_pernet_subsys(&tcf_action_net_ops);
1731 rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1732 rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1733 rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1739 subsys_initcall(tc_action_init);