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 static bool __tcf_idr_check(struct tc_action_net *tn, u32 index,
304 struct tc_action **a, int bind)
306 struct tcf_idrinfo *idrinfo = tn->idrinfo;
309 spin_lock(&idrinfo->lock);
310 p = idr_find(&idrinfo->action_idr, index);
314 refcount_inc(&p->tcfa_refcnt);
316 atomic_inc(&p->tcfa_bindcnt);
318 spin_unlock(&idrinfo->lock);
327 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
329 return __tcf_idr_check(tn, index, a, 0);
331 EXPORT_SYMBOL(tcf_idr_search);
333 bool tcf_idr_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
336 return __tcf_idr_check(tn, index, a, bind);
338 EXPORT_SYMBOL(tcf_idr_check);
340 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
345 spin_lock(&idrinfo->lock);
346 p = idr_find(&idrinfo->action_idr, index);
348 spin_unlock(&idrinfo->lock);
352 if (!atomic_read(&p->tcfa_bindcnt)) {
353 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
354 struct module *owner = p->ops->owner;
356 WARN_ON(p != idr_remove(&idrinfo->action_idr,
358 spin_unlock(&idrinfo->lock);
360 tcf_action_cleanup(p);
369 spin_unlock(&idrinfo->lock);
373 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
374 struct tc_action **a, const struct tc_action_ops *ops,
375 int bind, bool cpustats)
377 struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
378 struct tcf_idrinfo *idrinfo = tn->idrinfo;
383 refcount_set(&p->tcfa_refcnt, 1);
385 atomic_set(&p->tcfa_bindcnt, 1);
388 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
391 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
395 spin_lock_init(&p->tcfa_lock);
396 p->tcfa_index = index;
397 p->tcfa_tm.install = jiffies;
398 p->tcfa_tm.lastuse = jiffies;
399 p->tcfa_tm.firstuse = 0;
401 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
403 &p->tcfa_lock, NULL, est);
408 p->idrinfo = idrinfo;
410 INIT_LIST_HEAD(&p->list);
414 free_percpu(p->cpu_qstats);
416 free_percpu(p->cpu_bstats);
421 EXPORT_SYMBOL(tcf_idr_create);
423 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
425 struct tcf_idrinfo *idrinfo = tn->idrinfo;
427 spin_lock(&idrinfo->lock);
428 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
429 WARN_ON(!IS_ERR(idr_replace(&idrinfo->action_idr, a, a->tcfa_index)));
430 spin_unlock(&idrinfo->lock);
432 EXPORT_SYMBOL(tcf_idr_insert);
434 /* Cleanup idr index that was allocated but not initialized. */
436 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
438 struct tcf_idrinfo *idrinfo = tn->idrinfo;
440 spin_lock(&idrinfo->lock);
441 /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
442 WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
443 spin_unlock(&idrinfo->lock);
445 EXPORT_SYMBOL(tcf_idr_cleanup);
447 /* Check if action with specified index exists. If actions is found, increments
448 * its reference and bind counters, and return 1. Otherwise insert temporary
449 * error pointer (to prevent concurrent users from inserting actions with same
450 * index) and return 0.
453 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
454 struct tc_action **a, int bind)
456 struct tcf_idrinfo *idrinfo = tn->idrinfo;
461 spin_lock(&idrinfo->lock);
463 p = idr_find(&idrinfo->action_idr, *index);
465 /* This means that another process allocated
466 * index but did not assign the pointer yet.
468 spin_unlock(&idrinfo->lock);
473 refcount_inc(&p->tcfa_refcnt);
475 atomic_inc(&p->tcfa_bindcnt);
480 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
483 idr_replace(&idrinfo->action_idr,
484 ERR_PTR(-EBUSY), *index);
489 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
490 UINT_MAX, GFP_ATOMIC);
492 idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
495 spin_unlock(&idrinfo->lock);
498 EXPORT_SYMBOL(tcf_idr_check_alloc);
500 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
501 struct tcf_idrinfo *idrinfo)
503 struct idr *idr = &idrinfo->action_idr;
506 unsigned long id = 1;
508 idr_for_each_entry_ul(idr, p, id) {
509 ret = __tcf_idr_release(p, false, true);
510 if (ret == ACT_P_DELETED)
511 module_put(ops->owner);
515 idr_destroy(&idrinfo->action_idr);
517 EXPORT_SYMBOL(tcf_idrinfo_destroy);
519 static LIST_HEAD(act_base);
520 static DEFINE_RWLOCK(act_mod_lock);
522 int tcf_register_action(struct tc_action_ops *act,
523 struct pernet_operations *ops)
525 struct tc_action_ops *a;
528 if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
531 /* We have to register pernet ops before making the action ops visible,
532 * otherwise tcf_action_init_1() could get a partially initialized
535 ret = register_pernet_subsys(ops);
539 write_lock(&act_mod_lock);
540 list_for_each_entry(a, &act_base, head) {
541 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
542 write_unlock(&act_mod_lock);
543 unregister_pernet_subsys(ops);
547 list_add_tail(&act->head, &act_base);
548 write_unlock(&act_mod_lock);
552 EXPORT_SYMBOL(tcf_register_action);
554 int tcf_unregister_action(struct tc_action_ops *act,
555 struct pernet_operations *ops)
557 struct tc_action_ops *a;
560 write_lock(&act_mod_lock);
561 list_for_each_entry(a, &act_base, head) {
563 list_del(&act->head);
568 write_unlock(&act_mod_lock);
570 unregister_pernet_subsys(ops);
573 EXPORT_SYMBOL(tcf_unregister_action);
576 static struct tc_action_ops *tc_lookup_action_n(char *kind)
578 struct tc_action_ops *a, *res = NULL;
581 read_lock(&act_mod_lock);
582 list_for_each_entry(a, &act_base, head) {
583 if (strcmp(kind, a->kind) == 0) {
584 if (try_module_get(a->owner))
589 read_unlock(&act_mod_lock);
594 /* lookup by nlattr */
595 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
597 struct tc_action_ops *a, *res = NULL;
600 read_lock(&act_mod_lock);
601 list_for_each_entry(a, &act_base, head) {
602 if (nla_strcmp(kind, a->kind) == 0) {
603 if (try_module_get(a->owner))
608 read_unlock(&act_mod_lock);
613 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
614 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
615 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
616 int nr_actions, struct tcf_result *res)
619 u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
623 if (skb_skip_tc_classify(skb))
627 for (i = 0; i < nr_actions; i++) {
628 const struct tc_action *a = actions[i];
630 if (jmp_prgcnt > 0) {
635 ret = a->ops->act(skb, a, res);
636 if (ret == TC_ACT_REPEAT)
637 goto repeat; /* we need a ttl - JHS */
639 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
640 jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
641 if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
642 /* faulty opcode, stop pipeline */
647 goto restart_act_graph;
648 else /* faulty graph, stop pipeline */
651 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
652 tcf_action_goto_chain_exec(a, res);
655 if (ret != TC_ACT_PIPE)
661 EXPORT_SYMBOL(tcf_action_exec);
663 int tcf_action_destroy(struct tc_action *actions[], int bind)
665 const struct tc_action_ops *ops;
669 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
673 ret = __tcf_idr_release(a, bind, true);
674 if (ret == ACT_P_DELETED)
675 module_put(ops->owner);
682 static int tcf_action_put(struct tc_action *p)
684 return __tcf_action_put(p, false);
687 /* Put all actions in this array, skip those NULL's. */
688 static void tcf_action_put_many(struct tc_action *actions[])
692 for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
693 struct tc_action *a = actions[i];
694 const struct tc_action_ops *ops;
699 if (tcf_action_put(a))
700 module_put(ops->owner);
705 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
707 return a->ops->dump(skb, a, bind, ref);
711 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
714 unsigned char *b = skb_tail_pointer(skb);
716 struct tc_cookie *cookie;
718 if (nla_put_string(skb, TCA_KIND, a->ops->kind))
719 goto nla_put_failure;
720 if (tcf_action_copy_stats(skb, a, 0))
721 goto nla_put_failure;
724 cookie = rcu_dereference(a->act_cookie);
726 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
728 goto nla_put_failure;
733 nest = nla_nest_start(skb, TCA_OPTIONS);
735 goto nla_put_failure;
736 err = tcf_action_dump_old(skb, a, bind, ref);
738 nla_nest_end(skb, nest);
746 EXPORT_SYMBOL(tcf_action_dump_1);
748 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
752 int err = -EINVAL, i;
755 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
757 nest = nla_nest_start(skb, a->order);
759 goto nla_put_failure;
760 err = tcf_action_dump_1(skb, a, bind, ref);
763 nla_nest_end(skb, nest);
771 nla_nest_cancel(skb, nest);
775 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
777 struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
781 c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
786 c->len = nla_len(tb[TCA_ACT_COOKIE]);
791 static bool tcf_action_valid(int action)
793 int opcode = TC_ACT_EXT_OPCODE(action);
796 return action <= TC_ACT_VALUE_MAX;
797 return opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC;
800 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
801 struct nlattr *nla, struct nlattr *est,
802 char *name, int ovr, int bind,
804 struct netlink_ext_ack *extack)
807 struct tc_action_ops *a_o;
808 struct tc_cookie *cookie = NULL;
809 char act_name[IFNAMSIZ];
810 struct nlattr *tb[TCA_ACT_MAX + 1];
815 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
819 kind = tb[TCA_ACT_KIND];
821 NL_SET_ERR_MSG(extack, "TC action kind must be specified");
824 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
825 NL_SET_ERR_MSG(extack, "TC action name too long");
828 if (tb[TCA_ACT_COOKIE]) {
829 int cklen = nla_len(tb[TCA_ACT_COOKIE]);
831 if (cklen > TC_COOKIE_MAX_SIZE) {
832 NL_SET_ERR_MSG(extack, "TC cookie size above the maximum");
836 cookie = nla_memdup_cookie(tb);
838 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
844 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
845 NL_SET_ERR_MSG(extack, "TC action name too long");
851 a_o = tc_lookup_action_n(act_name);
853 #ifdef CONFIG_MODULES
856 request_module("act_%s", act_name);
860 a_o = tc_lookup_action_n(act_name);
862 /* We dropped the RTNL semaphore in order to
863 * perform the module load. So, even if we
864 * succeeded in loading the module we have to
865 * tell the caller to replay the request. We
866 * indicate this using -EAGAIN.
873 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
878 /* backward compatibility for policer */
880 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
883 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
888 if (!name && tb[TCA_ACT_COOKIE])
889 tcf_set_action_cookie(&a->act_cookie, cookie);
891 /* module count goes up only when brand new policy is created
892 * if it exists and is only bound to in a_o->init() then
893 * ACT_P_CREATED is not returned (a zero is).
895 if (err != ACT_P_CREATED)
896 module_put(a_o->owner);
898 if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
899 err = tcf_action_goto_chain_init(a, tp);
901 struct tc_action *actions[] = { a, NULL };
903 tcf_action_destroy(actions, bind);
904 NL_SET_ERR_MSG(extack, "Failed to init TC action chain");
909 if (!tcf_action_valid(a->tcfa_action)) {
910 NL_SET_ERR_MSG(extack, "invalid action value, using TC_ACT_UNSPEC instead");
911 a->tcfa_action = TC_ACT_UNSPEC;
917 module_put(a_o->owner);
926 /* Returns numbers of initialized actions or negative error. */
928 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
929 struct nlattr *est, char *name, int ovr, int bind,
930 struct tc_action *actions[], size_t *attr_size,
931 bool rtnl_held, struct netlink_ext_ack *extack)
933 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
934 struct tc_action *act;
939 err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
943 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
944 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
951 sz += tcf_action_fill_size(act);
952 /* Start from index 0 */
953 actions[i - 1] = act;
956 *attr_size = tcf_action_full_attrs_size(sz);
960 tcf_action_destroy(actions, bind);
964 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
973 /* compat_mode being true specifies a call that is supposed
974 * to add additional backward compatibility statistic TLVs.
977 if (p->type == TCA_OLD_COMPAT)
978 err = gnet_stats_start_copy_compat(skb, 0,
986 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
987 &p->tcfa_lock, &d, TCA_ACT_PAD);
992 if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
993 gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
994 gnet_stats_copy_queue(&d, p->cpu_qstats,
996 p->tcfa_qstats.qlen) < 0)
999 if (gnet_stats_finish_copy(&d) < 0)
1008 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1009 u32 portid, u32 seq, u16 flags, int event, int bind,
1013 struct nlmsghdr *nlh;
1014 unsigned char *b = skb_tail_pointer(skb);
1015 struct nlattr *nest;
1017 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1019 goto out_nlmsg_trim;
1020 t = nlmsg_data(nlh);
1021 t->tca_family = AF_UNSPEC;
1025 nest = nla_nest_start(skb, TCA_ACT_TAB);
1027 goto out_nlmsg_trim;
1029 if (tcf_action_dump(skb, actions, bind, ref) < 0)
1030 goto out_nlmsg_trim;
1032 nla_nest_end(skb, nest);
1034 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1043 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1044 struct tc_action *actions[], int event,
1045 struct netlink_ext_ack *extack)
1047 struct sk_buff *skb;
1049 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1052 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1054 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1059 return rtnl_unicast(skb, net, portid);
1062 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1063 struct nlmsghdr *n, u32 portid,
1064 struct netlink_ext_ack *extack)
1066 struct nlattr *tb[TCA_ACT_MAX + 1];
1067 const struct tc_action_ops *ops;
1068 struct tc_action *a;
1072 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1077 if (tb[TCA_ACT_INDEX] == NULL ||
1078 nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1079 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1082 index = nla_get_u32(tb[TCA_ACT_INDEX]);
1085 ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1086 if (!ops) { /* could happen in batch of actions */
1087 NL_SET_ERR_MSG(extack, "Specified TC action not found");
1091 if (ops->lookup(net, &a, index, extack) == 0)
1094 module_put(ops->owner);
1098 module_put(ops->owner);
1100 return ERR_PTR(err);
1103 static int tca_action_flush(struct net *net, struct nlattr *nla,
1104 struct nlmsghdr *n, u32 portid,
1105 struct netlink_ext_ack *extack)
1107 struct sk_buff *skb;
1109 struct nlmsghdr *nlh;
1111 struct netlink_callback dcb;
1112 struct nlattr *nest;
1113 struct nlattr *tb[TCA_ACT_MAX + 1];
1114 const struct tc_action_ops *ops;
1115 struct nlattr *kind;
1118 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1122 b = skb_tail_pointer(skb);
1124 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1129 kind = tb[TCA_ACT_KIND];
1130 ops = tc_lookup_action(kind);
1131 if (!ops) { /*some idjot trying to flush unknown action */
1132 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1136 nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1139 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1140 goto out_module_put;
1142 t = nlmsg_data(nlh);
1143 t->tca_family = AF_UNSPEC;
1147 nest = nla_nest_start(skb, TCA_ACT_TAB);
1149 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1150 goto out_module_put;
1153 err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1155 nla_nest_cancel(skb, nest);
1156 goto out_module_put;
1159 nla_nest_end(skb, nest);
1161 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1162 nlh->nlmsg_flags |= NLM_F_ROOT;
1163 module_put(ops->owner);
1164 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1165 n->nlmsg_flags & NLM_F_ECHO);
1169 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1174 module_put(ops->owner);
1180 static int tcf_action_delete(struct net *net, struct tc_action *actions[],
1181 struct netlink_ext_ack *extack)
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;
1194 if (tcf_action_put(a)) {
1195 /* last reference, action was deleted concurrently */
1196 module_put(ops->owner);
1200 /* now do the delete */
1201 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, extack);
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);
1273 attr_size += tcf_action_fill_size(act);
1274 actions[i - 1] = act;
1277 attr_size = tcf_action_full_attrs_size(attr_size);
1279 if (event == RTM_GETACTION)
1280 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1282 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1288 tcf_action_put_many(actions);
1293 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1294 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1296 struct sk_buff *skb;
1299 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1304 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1305 RTM_NEWACTION, 0, 0) <= 0) {
1306 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1311 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1312 n->nlmsg_flags & NLM_F_ECHO);
1318 static int tcf_action_add(struct net *net, struct nlattr *nla,
1319 struct nlmsghdr *n, u32 portid, int ovr,
1320 struct netlink_ext_ack *extack)
1322 size_t attr_size = 0;
1324 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1326 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, actions,
1327 &attr_size, true, extack);
1330 ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1332 tcf_action_put_many(actions);
1337 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1338 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1339 [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1340 .validation_data = &tcaa_root_flags_allowed },
1341 [TCA_ROOT_TIME_DELTA] = { .type = NLA_U32 },
1344 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1345 struct netlink_ext_ack *extack)
1347 struct net *net = sock_net(skb->sk);
1348 struct nlattr *tca[TCA_ROOT_MAX + 1];
1349 u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1350 int ret = 0, ovr = 0;
1352 if ((n->nlmsg_type != RTM_GETACTION) &&
1353 !netlink_capable(skb, CAP_NET_ADMIN))
1356 ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1361 if (tca[TCA_ACT_TAB] == NULL) {
1362 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1366 /* n->nlmsg_flags & NLM_F_CREATE */
1367 switch (n->nlmsg_type) {
1369 /* we are going to assume all other flags
1370 * imply create only if it doesn't exist
1371 * Note that CREATE | EXCL implies that
1372 * but since we want avoid ambiguity (eg when flags
1373 * is zero) then just set this
1375 if (n->nlmsg_flags & NLM_F_REPLACE)
1378 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1384 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1385 portid, RTM_DELACTION, extack);
1388 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1389 portid, RTM_GETACTION, extack);
1398 static struct nlattr *find_dump_kind(struct nlattr **nla)
1400 struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1401 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1402 struct nlattr *kind;
1404 tb1 = nla[TCA_ACT_TAB];
1408 if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1409 NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1414 if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1416 kind = tb2[TCA_ACT_KIND];
1421 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1423 struct net *net = sock_net(skb->sk);
1424 struct nlmsghdr *nlh;
1425 unsigned char *b = skb_tail_pointer(skb);
1426 struct nlattr *nest;
1427 struct tc_action_ops *a_o;
1429 struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1430 struct nlattr *tb[TCA_ROOT_MAX + 1];
1431 struct nlattr *count_attr = NULL;
1432 unsigned long jiffy_since = 0;
1433 struct nlattr *kind = NULL;
1434 struct nla_bitfield32 bf;
1435 u32 msecs_since = 0;
1438 ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1443 kind = find_dump_kind(tb);
1445 pr_info("tc_dump_action: action bad kind\n");
1449 a_o = tc_lookup_action(kind);
1454 if (tb[TCA_ROOT_FLAGS]) {
1455 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1456 cb->args[2] = bf.value;
1459 if (tb[TCA_ROOT_TIME_DELTA]) {
1460 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1463 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1464 cb->nlh->nlmsg_type, sizeof(*t), 0);
1466 goto out_module_put;
1469 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1471 t = nlmsg_data(nlh);
1472 t->tca_family = AF_UNSPEC;
1475 cb->args[3] = jiffy_since;
1476 count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1478 goto out_module_put;
1480 nest = nla_nest_start(skb, TCA_ACT_TAB);
1482 goto out_module_put;
1484 ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1486 goto out_module_put;
1489 nla_nest_end(skb, nest);
1491 act_count = cb->args[1];
1492 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1497 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1498 if (NETLINK_CB(cb->skb).portid && ret)
1499 nlh->nlmsg_flags |= NLM_F_MULTI;
1500 module_put(a_o->owner);
1504 module_put(a_o->owner);
1509 struct tcf_action_net {
1510 struct rhashtable egdev_ht;
1513 static unsigned int tcf_action_net_id;
1515 struct tcf_action_egdev_cb {
1516 struct list_head list;
1521 struct tcf_action_egdev {
1522 struct rhash_head ht_node;
1523 const struct net_device *dev;
1524 unsigned int refcnt;
1525 struct list_head cb_list;
1528 static const struct rhashtable_params tcf_action_egdev_ht_params = {
1529 .key_offset = offsetof(struct tcf_action_egdev, dev),
1530 .head_offset = offsetof(struct tcf_action_egdev, ht_node),
1531 .key_len = sizeof(const struct net_device *),
1534 static struct tcf_action_egdev *
1535 tcf_action_egdev_lookup(const struct net_device *dev)
1537 struct net *net = dev_net(dev);
1538 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1540 return rhashtable_lookup_fast(&tan->egdev_ht, &dev,
1541 tcf_action_egdev_ht_params);
1544 static struct tcf_action_egdev *
1545 tcf_action_egdev_get(const struct net_device *dev)
1547 struct tcf_action_egdev *egdev;
1548 struct tcf_action_net *tan;
1550 egdev = tcf_action_egdev_lookup(dev);
1554 egdev = kzalloc(sizeof(*egdev), GFP_KERNEL);
1557 INIT_LIST_HEAD(&egdev->cb_list);
1559 tan = net_generic(dev_net(dev), tcf_action_net_id);
1560 rhashtable_insert_fast(&tan->egdev_ht, &egdev->ht_node,
1561 tcf_action_egdev_ht_params);
1568 static void tcf_action_egdev_put(struct tcf_action_egdev *egdev)
1570 struct tcf_action_net *tan;
1572 if (--egdev->refcnt)
1574 tan = net_generic(dev_net(egdev->dev), tcf_action_net_id);
1575 rhashtable_remove_fast(&tan->egdev_ht, &egdev->ht_node,
1576 tcf_action_egdev_ht_params);
1580 static struct tcf_action_egdev_cb *
1581 tcf_action_egdev_cb_lookup(struct tcf_action_egdev *egdev,
1582 tc_setup_cb_t *cb, void *cb_priv)
1584 struct tcf_action_egdev_cb *egdev_cb;
1586 list_for_each_entry(egdev_cb, &egdev->cb_list, list)
1587 if (egdev_cb->cb == cb && egdev_cb->cb_priv == cb_priv)
1592 static int tcf_action_egdev_cb_call(struct tcf_action_egdev *egdev,
1593 enum tc_setup_type type,
1594 void *type_data, bool err_stop)
1596 struct tcf_action_egdev_cb *egdev_cb;
1600 list_for_each_entry(egdev_cb, &egdev->cb_list, list) {
1601 err = egdev_cb->cb(type, type_data, egdev_cb->cb_priv);
1612 static int tcf_action_egdev_cb_add(struct tcf_action_egdev *egdev,
1613 tc_setup_cb_t *cb, void *cb_priv)
1615 struct tcf_action_egdev_cb *egdev_cb;
1617 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1618 if (WARN_ON(egdev_cb))
1620 egdev_cb = kzalloc(sizeof(*egdev_cb), GFP_KERNEL);
1624 egdev_cb->cb_priv = cb_priv;
1625 list_add(&egdev_cb->list, &egdev->cb_list);
1629 static void tcf_action_egdev_cb_del(struct tcf_action_egdev *egdev,
1630 tc_setup_cb_t *cb, void *cb_priv)
1632 struct tcf_action_egdev_cb *egdev_cb;
1634 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1635 if (WARN_ON(!egdev_cb))
1637 list_del(&egdev_cb->list);
1641 static int __tc_setup_cb_egdev_register(const struct net_device *dev,
1642 tc_setup_cb_t *cb, void *cb_priv)
1644 struct tcf_action_egdev *egdev = tcf_action_egdev_get(dev);
1649 err = tcf_action_egdev_cb_add(egdev, cb, cb_priv);
1655 tcf_action_egdev_put(egdev);
1658 int tc_setup_cb_egdev_register(const struct net_device *dev,
1659 tc_setup_cb_t *cb, void *cb_priv)
1664 err = __tc_setup_cb_egdev_register(dev, cb, cb_priv);
1668 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_register);
1670 static void __tc_setup_cb_egdev_unregister(const struct net_device *dev,
1671 tc_setup_cb_t *cb, void *cb_priv)
1673 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1675 if (WARN_ON(!egdev))
1677 tcf_action_egdev_cb_del(egdev, cb, cb_priv);
1678 tcf_action_egdev_put(egdev);
1680 void tc_setup_cb_egdev_unregister(const struct net_device *dev,
1681 tc_setup_cb_t *cb, void *cb_priv)
1684 __tc_setup_cb_egdev_unregister(dev, cb, cb_priv);
1687 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_unregister);
1689 int tc_setup_cb_egdev_call(const struct net_device *dev,
1690 enum tc_setup_type type, void *type_data,
1693 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1697 return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
1699 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
1701 static __net_init int tcf_action_net_init(struct net *net)
1703 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1705 return rhashtable_init(&tan->egdev_ht, &tcf_action_egdev_ht_params);
1708 static void __net_exit tcf_action_net_exit(struct net *net)
1710 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1712 rhashtable_destroy(&tan->egdev_ht);
1715 static struct pernet_operations tcf_action_net_ops = {
1716 .init = tcf_action_net_init,
1717 .exit = tcf_action_net_exit,
1718 .id = &tcf_action_net_id,
1719 .size = sizeof(struct tcf_action_net),
1722 static int __init tc_action_init(void)
1726 err = register_pernet_subsys(&tcf_action_net_ops);
1730 rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1731 rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1732 rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1738 subsys_initcall(tc_action_init);