Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / net / sched / act_api.c
diff --git a/kernel/net/sched/act_api.c b/kernel/net/sched/act_api.c
new file mode 100644 (file)
index 0000000..3d43e49
--- /dev/null
@@ -0,0 +1,1094 @@
+/*
+ * net/sched/act_api.c Packet action API.
+ *
+ *             This program is free software; you can redistribute it and/or
+ *             modify it under the terms of the GNU General Public License
+ *             as published by the Free Software Foundation; either version
+ *             2 of the License, or (at your option) any later version.
+ *
+ * Author:     Jamal Hadi Salim
+ *
+ *
+ */
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/slab.h>
+#include <linux/skbuff.h>
+#include <linux/init.h>
+#include <linux/kmod.h>
+#include <linux/err.h>
+#include <linux/module.h>
+#include <net/net_namespace.h>
+#include <net/sock.h>
+#include <net/sch_generic.h>
+#include <net/act_api.h>
+#include <net/netlink.h>
+
+void tcf_hash_destroy(struct tc_action *a)
+{
+       struct tcf_common *p = a->priv;
+       struct tcf_hashinfo *hinfo = a->ops->hinfo;
+
+       spin_lock_bh(&hinfo->lock);
+       hlist_del(&p->tcfc_head);
+       spin_unlock_bh(&hinfo->lock);
+       gen_kill_estimator(&p->tcfc_bstats,
+                          &p->tcfc_rate_est);
+       /*
+        * gen_estimator est_timer() might access p->tcfc_lock
+        * or bstats, wait a RCU grace period before freeing p
+        */
+       kfree_rcu(p, tcfc_rcu);
+}
+EXPORT_SYMBOL(tcf_hash_destroy);
+
+int tcf_hash_release(struct tc_action *a, int bind)
+{
+       struct tcf_common *p = a->priv;
+       int ret = 0;
+
+       if (p) {
+               if (bind)
+                       p->tcfc_bindcnt--;
+               else if (p->tcfc_bindcnt > 0)
+                       return -EPERM;
+
+               p->tcfc_refcnt--;
+               if (p->tcfc_bindcnt <= 0 && p->tcfc_refcnt <= 0) {
+                       if (a->ops->cleanup)
+                               a->ops->cleanup(a, bind);
+                       tcf_hash_destroy(a);
+                       ret = 1;
+               }
+       }
+       return ret;
+}
+EXPORT_SYMBOL(tcf_hash_release);
+
+static int tcf_dump_walker(struct sk_buff *skb, struct netlink_callback *cb,
+                          struct tc_action *a)
+{
+       struct tcf_hashinfo *hinfo = a->ops->hinfo;
+       struct hlist_head *head;
+       struct tcf_common *p;
+       int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
+       struct nlattr *nest;
+
+       spin_lock_bh(&hinfo->lock);
+
+       s_i = cb->args[0];
+
+       for (i = 0; i < (hinfo->hmask + 1); i++) {
+               head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
+
+               hlist_for_each_entry_rcu(p, head, tcfc_head) {
+                       index++;
+                       if (index < s_i)
+                               continue;
+                       a->priv = p;
+                       a->order = n_i;
+
+                       nest = nla_nest_start(skb, a->order);
+                       if (nest == NULL)
+                               goto nla_put_failure;
+                       err = tcf_action_dump_1(skb, a, 0, 0);
+                       if (err < 0) {
+                               index--;
+                               nlmsg_trim(skb, nest);
+                               goto done;
+                       }
+                       nla_nest_end(skb, nest);
+                       n_i++;
+                       if (n_i >= TCA_ACT_MAX_PRIO)
+                               goto done;
+               }
+       }
+done:
+       spin_unlock_bh(&hinfo->lock);
+       if (n_i)
+               cb->args[0] += n_i;
+       return n_i;
+
+nla_put_failure:
+       nla_nest_cancel(skb, nest);
+       goto done;
+}
+
+static int tcf_del_walker(struct sk_buff *skb, struct tc_action *a)
+{
+       struct tcf_hashinfo *hinfo = a->ops->hinfo;
+       struct hlist_head *head;
+       struct hlist_node *n;
+       struct tcf_common *p;
+       struct nlattr *nest;
+       int i = 0, n_i = 0;
+       int ret = -EINVAL;
+
+       nest = nla_nest_start(skb, a->order);
+       if (nest == NULL)
+               goto nla_put_failure;
+       if (nla_put_string(skb, TCA_KIND, a->ops->kind))
+               goto nla_put_failure;
+       for (i = 0; i < (hinfo->hmask + 1); i++) {
+               head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
+               hlist_for_each_entry_safe(p, n, head, tcfc_head) {
+                       a->priv = p;
+                       ret = tcf_hash_release(a, 0);
+                       if (ret == ACT_P_DELETED) {
+                               module_put(a->ops->owner);
+                               n_i++;
+                       } else if (ret < 0)
+                               goto nla_put_failure;
+               }
+       }
+       if (nla_put_u32(skb, TCA_FCNT, n_i))
+               goto nla_put_failure;
+       nla_nest_end(skb, nest);
+
+       return n_i;
+nla_put_failure:
+       nla_nest_cancel(skb, nest);
+       return ret;
+}
+
+static int tcf_generic_walker(struct sk_buff *skb, struct netlink_callback *cb,
+                             int type, struct tc_action *a)
+{
+       if (type == RTM_DELACTION) {
+               return tcf_del_walker(skb, a);
+       } else if (type == RTM_GETACTION) {
+               return tcf_dump_walker(skb, cb, a);
+       } else {
+               WARN(1, "tcf_generic_walker: unknown action %d\n", type);
+               return -EINVAL;
+       }
+}
+
+static struct tcf_common *tcf_hash_lookup(u32 index, struct tcf_hashinfo *hinfo)
+{
+       struct tcf_common *p = NULL;
+       struct hlist_head *head;
+
+       spin_lock_bh(&hinfo->lock);
+       head = &hinfo->htab[tcf_hash(index, hinfo->hmask)];
+       hlist_for_each_entry_rcu(p, head, tcfc_head)
+               if (p->tcfc_index == index)
+                       break;
+       spin_unlock_bh(&hinfo->lock);
+
+       return p;
+}
+
+u32 tcf_hash_new_index(struct tcf_hashinfo *hinfo)
+{
+       u32 val = hinfo->index;
+
+       do {
+               if (++val == 0)
+                       val = 1;
+       } while (tcf_hash_lookup(val, hinfo));
+
+       hinfo->index = val;
+       return val;
+}
+EXPORT_SYMBOL(tcf_hash_new_index);
+
+int tcf_hash_search(struct tc_action *a, u32 index)
+{
+       struct tcf_hashinfo *hinfo = a->ops->hinfo;
+       struct tcf_common *p = tcf_hash_lookup(index, hinfo);
+
+       if (p) {
+               a->priv = p;
+               return 1;
+       }
+       return 0;
+}
+EXPORT_SYMBOL(tcf_hash_search);
+
+int tcf_hash_check(u32 index, struct tc_action *a, int bind)
+{
+       struct tcf_hashinfo *hinfo = a->ops->hinfo;
+       struct tcf_common *p = NULL;
+       if (index && (p = tcf_hash_lookup(index, hinfo)) != NULL) {
+               if (bind)
+                       p->tcfc_bindcnt++;
+               p->tcfc_refcnt++;
+               a->priv = p;
+               return 1;
+       }
+       return 0;
+}
+EXPORT_SYMBOL(tcf_hash_check);
+
+void tcf_hash_cleanup(struct tc_action *a, struct nlattr *est)
+{
+       struct tcf_common *pc = a->priv;
+       if (est)
+               gen_kill_estimator(&pc->tcfc_bstats,
+                                  &pc->tcfc_rate_est);
+       kfree_rcu(pc, tcfc_rcu);
+}
+EXPORT_SYMBOL(tcf_hash_cleanup);
+
+int tcf_hash_create(u32 index, struct nlattr *est, struct tc_action *a,
+                   int size, int bind)
+{
+       struct tcf_hashinfo *hinfo = a->ops->hinfo;
+       struct tcf_common *p = kzalloc(size, GFP_KERNEL);
+
+       if (unlikely(!p))
+               return -ENOMEM;
+       p->tcfc_refcnt = 1;
+       if (bind)
+               p->tcfc_bindcnt = 1;
+
+       spin_lock_init(&p->tcfc_lock);
+       INIT_HLIST_NODE(&p->tcfc_head);
+       p->tcfc_index = index ? index : tcf_hash_new_index(hinfo);
+       p->tcfc_tm.install = jiffies;
+       p->tcfc_tm.lastuse = jiffies;
+       if (est) {
+               int err = gen_new_estimator(&p->tcfc_bstats, NULL,
+                                           &p->tcfc_rate_est,
+                                           &p->tcfc_lock, est);
+               if (err) {
+                       kfree(p);
+                       return err;
+               }
+       }
+
+       a->priv = (void *) p;
+       return 0;
+}
+EXPORT_SYMBOL(tcf_hash_create);
+
+void tcf_hash_insert(struct tc_action *a)
+{
+       struct tcf_common *p = a->priv;
+       struct tcf_hashinfo *hinfo = a->ops->hinfo;
+       unsigned int h = tcf_hash(p->tcfc_index, hinfo->hmask);
+
+       spin_lock_bh(&hinfo->lock);
+       hlist_add_head(&p->tcfc_head, &hinfo->htab[h]);
+       spin_unlock_bh(&hinfo->lock);
+}
+EXPORT_SYMBOL(tcf_hash_insert);
+
+static LIST_HEAD(act_base);
+static DEFINE_RWLOCK(act_mod_lock);
+
+int tcf_register_action(struct tc_action_ops *act, unsigned int mask)
+{
+       struct tc_action_ops *a;
+       int err;
+
+       /* Must supply act, dump and init */
+       if (!act->act || !act->dump || !act->init)
+               return -EINVAL;
+
+       /* Supply defaults */
+       if (!act->lookup)
+               act->lookup = tcf_hash_search;
+       if (!act->walk)
+               act->walk = tcf_generic_walker;
+
+       act->hinfo = kmalloc(sizeof(struct tcf_hashinfo), GFP_KERNEL);
+       if (!act->hinfo)
+               return -ENOMEM;
+       err = tcf_hashinfo_init(act->hinfo, mask);
+       if (err) {
+               kfree(act->hinfo);
+               return err;
+       }
+
+       write_lock(&act_mod_lock);
+       list_for_each_entry(a, &act_base, head) {
+               if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
+                       write_unlock(&act_mod_lock);
+                       tcf_hashinfo_destroy(act->hinfo);
+                       kfree(act->hinfo);
+                       return -EEXIST;
+               }
+       }
+       list_add_tail(&act->head, &act_base);
+       write_unlock(&act_mod_lock);
+       return 0;
+}
+EXPORT_SYMBOL(tcf_register_action);
+
+int tcf_unregister_action(struct tc_action_ops *act)
+{
+       struct tc_action_ops *a;
+       int err = -ENOENT;
+
+       write_lock(&act_mod_lock);
+       list_for_each_entry(a, &act_base, head) {
+               if (a == act) {
+                       list_del(&act->head);
+                       tcf_hashinfo_destroy(act->hinfo);
+                       kfree(act->hinfo);
+                       err = 0;
+                       break;
+               }
+       }
+       write_unlock(&act_mod_lock);
+       return err;
+}
+EXPORT_SYMBOL(tcf_unregister_action);
+
+/* lookup by name */
+static struct tc_action_ops *tc_lookup_action_n(char *kind)
+{
+       struct tc_action_ops *a, *res = NULL;
+
+       if (kind) {
+               read_lock(&act_mod_lock);
+               list_for_each_entry(a, &act_base, head) {
+                       if (strcmp(kind, a->kind) == 0) {
+                               if (try_module_get(a->owner))
+                                       res = a;
+                               break;
+                       }
+               }
+               read_unlock(&act_mod_lock);
+       }
+       return res;
+}
+
+/* lookup by nlattr */
+static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
+{
+       struct tc_action_ops *a, *res = NULL;
+
+       if (kind) {
+               read_lock(&act_mod_lock);
+               list_for_each_entry(a, &act_base, head) {
+                       if (nla_strcmp(kind, a->kind) == 0) {
+                               if (try_module_get(a->owner))
+                                       res = a;
+                               break;
+                       }
+               }
+               read_unlock(&act_mod_lock);
+       }
+       return res;
+}
+
+int tcf_action_exec(struct sk_buff *skb, const struct list_head *actions,
+                   struct tcf_result *res)
+{
+       const struct tc_action *a;
+       int ret = -1;
+
+       if (skb->tc_verd & TC_NCLS) {
+               skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
+               ret = TC_ACT_OK;
+               goto exec_done;
+       }
+       list_for_each_entry(a, actions, list) {
+repeat:
+               ret = a->ops->act(skb, a, res);
+               if (TC_MUNGED & skb->tc_verd) {
+                       /* copied already, allow trampling */
+                       skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
+                       skb->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
+               }
+               if (ret == TC_ACT_REPEAT)
+                       goto repeat;    /* we need a ttl - JHS */
+               if (ret != TC_ACT_PIPE)
+                       goto exec_done;
+       }
+exec_done:
+       return ret;
+}
+EXPORT_SYMBOL(tcf_action_exec);
+
+int tcf_action_destroy(struct list_head *actions, int bind)
+{
+       struct tc_action *a, *tmp;
+       int ret = 0;
+
+       list_for_each_entry_safe(a, tmp, actions, list) {
+               ret = tcf_hash_release(a, bind);
+               if (ret == ACT_P_DELETED)
+                       module_put(a->ops->owner);
+               else if (ret < 0)
+                       return ret;
+               list_del(&a->list);
+               kfree(a);
+       }
+       return ret;
+}
+
+int
+tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
+{
+       return a->ops->dump(skb, a, bind, ref);
+}
+
+int
+tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
+{
+       int err = -EINVAL;
+       unsigned char *b = skb_tail_pointer(skb);
+       struct nlattr *nest;
+
+       if (nla_put_string(skb, TCA_KIND, a->ops->kind))
+               goto nla_put_failure;
+       if (tcf_action_copy_stats(skb, a, 0))
+               goto nla_put_failure;
+       nest = nla_nest_start(skb, TCA_OPTIONS);
+       if (nest == NULL)
+               goto nla_put_failure;
+       err = tcf_action_dump_old(skb, a, bind, ref);
+       if (err > 0) {
+               nla_nest_end(skb, nest);
+               return err;
+       }
+
+nla_put_failure:
+       nlmsg_trim(skb, b);
+       return -1;
+}
+EXPORT_SYMBOL(tcf_action_dump_1);
+
+int
+tcf_action_dump(struct sk_buff *skb, struct list_head *actions, int bind, int ref)
+{
+       struct tc_action *a;
+       int err = -EINVAL;
+       struct nlattr *nest;
+
+       list_for_each_entry(a, actions, list) {
+               nest = nla_nest_start(skb, a->order);
+               if (nest == NULL)
+                       goto nla_put_failure;
+               err = tcf_action_dump_1(skb, a, bind, ref);
+               if (err < 0)
+                       goto errout;
+               nla_nest_end(skb, nest);
+       }
+
+       return 0;
+
+nla_put_failure:
+       err = -EINVAL;
+errout:
+       nla_nest_cancel(skb, nest);
+       return err;
+}
+
+struct tc_action *tcf_action_init_1(struct net *net, struct nlattr *nla,
+                                   struct nlattr *est, char *name, int ovr,
+                                   int bind)
+{
+       struct tc_action *a;
+       struct tc_action_ops *a_o;
+       char act_name[IFNAMSIZ];
+       struct nlattr *tb[TCA_ACT_MAX + 1];
+       struct nlattr *kind;
+       int err;
+
+       if (name == NULL) {
+               err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
+               if (err < 0)
+                       goto err_out;
+               err = -EINVAL;
+               kind = tb[TCA_ACT_KIND];
+               if (kind == NULL)
+                       goto err_out;
+               if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
+                       goto err_out;
+       } else {
+               err = -EINVAL;
+               if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
+                       goto err_out;
+       }
+
+       a_o = tc_lookup_action_n(act_name);
+       if (a_o == NULL) {
+#ifdef CONFIG_MODULES
+               rtnl_unlock();
+               request_module("act_%s", act_name);
+               rtnl_lock();
+
+               a_o = tc_lookup_action_n(act_name);
+
+               /* We dropped the RTNL semaphore in order to
+                * perform the module load.  So, even if we
+                * succeeded in loading the module we have to
+                * tell the caller to replay the request.  We
+                * indicate this using -EAGAIN.
+                */
+               if (a_o != NULL) {
+                       err = -EAGAIN;
+                       goto err_mod;
+               }
+#endif
+               err = -ENOENT;
+               goto err_out;
+       }
+
+       err = -ENOMEM;
+       a = kzalloc(sizeof(*a), GFP_KERNEL);
+       if (a == NULL)
+               goto err_mod;
+
+       a->ops = a_o;
+       INIT_LIST_HEAD(&a->list);
+       /* backward compatibility for policer */
+       if (name == NULL)
+               err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, a, ovr, bind);
+       else
+               err = a_o->init(net, nla, est, a, ovr, bind);
+       if (err < 0)
+               goto err_free;
+
+       /* module count goes up only when brand new policy is created
+        * if it exists and is only bound to in a_o->init() then
+        * ACT_P_CREATED is not returned (a zero is).
+        */
+       if (err != ACT_P_CREATED)
+               module_put(a_o->owner);
+
+       return a;
+
+err_free:
+       kfree(a);
+err_mod:
+       module_put(a_o->owner);
+err_out:
+       return ERR_PTR(err);
+}
+
+int tcf_action_init(struct net *net, struct nlattr *nla,
+                                 struct nlattr *est, char *name, int ovr,
+                                 int bind, struct list_head *actions)
+{
+       struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
+       struct tc_action *act;
+       int err;
+       int i;
+
+       err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
+       if (err < 0)
+               return err;
+
+       for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
+               act = tcf_action_init_1(net, tb[i], est, name, ovr, bind);
+               if (IS_ERR(act)) {
+                       err = PTR_ERR(act);
+                       goto err;
+               }
+               act->order = i;
+               list_add_tail(&act->list, actions);
+       }
+       return 0;
+
+err:
+       tcf_action_destroy(actions, bind);
+       return err;
+}
+
+int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *a,
+                         int compat_mode)
+{
+       int err = 0;
+       struct gnet_dump d;
+       struct tcf_common *p = a->priv;
+
+       if (p == NULL)
+               goto errout;
+
+       /* compat_mode being true specifies a call that is supposed
+        * to add additional backward compatibility statistic TLVs.
+        */
+       if (compat_mode) {
+               if (a->type == TCA_OLD_COMPAT)
+                       err = gnet_stats_start_copy_compat(skb, 0,
+                               TCA_STATS, TCA_XSTATS, &p->tcfc_lock, &d);
+               else
+                       return 0;
+       } else
+               err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
+                                           &p->tcfc_lock, &d);
+
+       if (err < 0)
+               goto errout;
+
+       if (gnet_stats_copy_basic(&d, NULL, &p->tcfc_bstats) < 0 ||
+           gnet_stats_copy_rate_est(&d, &p->tcfc_bstats,
+                                    &p->tcfc_rate_est) < 0 ||
+           gnet_stats_copy_queue(&d, NULL,
+                                 &p->tcfc_qstats,
+                                 p->tcfc_qstats.qlen) < 0)
+               goto errout;
+
+       if (gnet_stats_finish_copy(&d) < 0)
+               goto errout;
+
+       return 0;
+
+errout:
+       return -1;
+}
+
+static int
+tca_get_fill(struct sk_buff *skb, struct list_head *actions, u32 portid, u32 seq,
+            u16 flags, int event, int bind, int ref)
+{
+       struct tcamsg *t;
+       struct nlmsghdr *nlh;
+       unsigned char *b = skb_tail_pointer(skb);
+       struct nlattr *nest;
+
+       nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
+       if (!nlh)
+               goto out_nlmsg_trim;
+       t = nlmsg_data(nlh);
+       t->tca_family = AF_UNSPEC;
+       t->tca__pad1 = 0;
+       t->tca__pad2 = 0;
+
+       nest = nla_nest_start(skb, TCA_ACT_TAB);
+       if (nest == NULL)
+               goto out_nlmsg_trim;
+
+       if (tcf_action_dump(skb, actions, bind, ref) < 0)
+               goto out_nlmsg_trim;
+
+       nla_nest_end(skb, nest);
+
+       nlh->nlmsg_len = skb_tail_pointer(skb) - b;
+       return skb->len;
+
+out_nlmsg_trim:
+       nlmsg_trim(skb, b);
+       return -1;
+}
+
+static int
+act_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
+              struct list_head *actions, int event)
+{
+       struct sk_buff *skb;
+
+       skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
+       if (!skb)
+               return -ENOBUFS;
+       if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event, 0, 0) <= 0) {
+               kfree_skb(skb);
+               return -EINVAL;
+       }
+
+       return rtnl_unicast(skb, net, portid);
+}
+
+static struct tc_action *create_a(int i)
+{
+       struct tc_action *act;
+
+       act = kzalloc(sizeof(*act), GFP_KERNEL);
+       if (act == NULL) {
+               pr_debug("create_a: failed to alloc!\n");
+               return NULL;
+       }
+       act->order = i;
+       INIT_LIST_HEAD(&act->list);
+       return act;
+}
+
+static struct tc_action *
+tcf_action_get_1(struct nlattr *nla, struct nlmsghdr *n, u32 portid)
+{
+       struct nlattr *tb[TCA_ACT_MAX + 1];
+       struct tc_action *a;
+       int index;
+       int err;
+
+       err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
+       if (err < 0)
+               goto err_out;
+
+       err = -EINVAL;
+       if (tb[TCA_ACT_INDEX] == NULL ||
+           nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
+               goto err_out;
+       index = nla_get_u32(tb[TCA_ACT_INDEX]);
+
+       err = -ENOMEM;
+       a = create_a(0);
+       if (a == NULL)
+               goto err_out;
+
+       err = -EINVAL;
+       a->ops = tc_lookup_action(tb[TCA_ACT_KIND]);
+       if (a->ops == NULL) /* could happen in batch of actions */
+               goto err_free;
+       err = -ENOENT;
+       if (a->ops->lookup(a, index) == 0)
+               goto err_mod;
+
+       module_put(a->ops->owner);
+       return a;
+
+err_mod:
+       module_put(a->ops->owner);
+err_free:
+       kfree(a);
+err_out:
+       return ERR_PTR(err);
+}
+
+static void cleanup_a(struct list_head *actions)
+{
+       struct tc_action *a, *tmp;
+
+       list_for_each_entry_safe(a, tmp, actions, list) {
+               list_del(&a->list);
+               kfree(a);
+       }
+}
+
+static int tca_action_flush(struct net *net, struct nlattr *nla,
+                           struct nlmsghdr *n, u32 portid)
+{
+       struct sk_buff *skb;
+       unsigned char *b;
+       struct nlmsghdr *nlh;
+       struct tcamsg *t;
+       struct netlink_callback dcb;
+       struct nlattr *nest;
+       struct nlattr *tb[TCA_ACT_MAX + 1];
+       struct nlattr *kind;
+       struct tc_action a;
+       int err = -ENOMEM;
+
+       skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
+       if (!skb) {
+               pr_debug("tca_action_flush: failed skb alloc\n");
+               return err;
+       }
+
+       b = skb_tail_pointer(skb);
+
+       err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
+       if (err < 0)
+               goto err_out;
+
+       err = -EINVAL;
+       kind = tb[TCA_ACT_KIND];
+       memset(&a, 0, sizeof(struct tc_action));
+       INIT_LIST_HEAD(&a.list);
+       a.ops = tc_lookup_action(kind);
+       if (a.ops == NULL) /*some idjot trying to flush unknown action */
+               goto err_out;
+
+       nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION, sizeof(*t), 0);
+       if (!nlh)
+               goto out_module_put;
+       t = nlmsg_data(nlh);
+       t->tca_family = AF_UNSPEC;
+       t->tca__pad1 = 0;
+       t->tca__pad2 = 0;
+
+       nest = nla_nest_start(skb, TCA_ACT_TAB);
+       if (nest == NULL)
+               goto out_module_put;
+
+       err = a.ops->walk(skb, &dcb, RTM_DELACTION, &a);
+       if (err < 0)
+               goto out_module_put;
+       if (err == 0)
+               goto noflush_out;
+
+       nla_nest_end(skb, nest);
+
+       nlh->nlmsg_len = skb_tail_pointer(skb) - b;
+       nlh->nlmsg_flags |= NLM_F_ROOT;
+       module_put(a.ops->owner);
+       err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
+                            n->nlmsg_flags & NLM_F_ECHO);
+       if (err > 0)
+               return 0;
+
+       return err;
+
+out_module_put:
+       module_put(a.ops->owner);
+err_out:
+noflush_out:
+       kfree_skb(skb);
+       return err;
+}
+
+static int
+tcf_del_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
+              u32 portid)
+{
+       int ret;
+       struct sk_buff *skb;
+
+       skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
+       if (!skb)
+               return -ENOBUFS;
+
+       if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
+                        0, 1) <= 0) {
+               kfree_skb(skb);
+               return -EINVAL;
+       }
+
+       /* now do the delete */
+       ret = tcf_action_destroy(actions, 0);
+       if (ret < 0) {
+               kfree_skb(skb);
+               return ret;
+       }
+
+       ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
+                            n->nlmsg_flags & NLM_F_ECHO);
+       if (ret > 0)
+               return 0;
+       return ret;
+}
+
+static int
+tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
+             u32 portid, int event)
+{
+       int i, ret;
+       struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
+       struct tc_action *act;
+       LIST_HEAD(actions);
+
+       ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
+       if (ret < 0)
+               return ret;
+
+       if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
+               if (tb[1] != NULL)
+                       return tca_action_flush(net, tb[1], n, portid);
+               else
+                       return -EINVAL;
+       }
+
+       for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
+               act = tcf_action_get_1(tb[i], n, portid);
+               if (IS_ERR(act)) {
+                       ret = PTR_ERR(act);
+                       goto err;
+               }
+               act->order = i;
+               list_add_tail(&act->list, &actions);
+       }
+
+       if (event == RTM_GETACTION)
+               ret = act_get_notify(net, portid, n, &actions, event);
+       else { /* delete */
+               ret = tcf_del_notify(net, n, &actions, portid);
+               if (ret)
+                       goto err;
+               return ret;
+       }
+err:
+       cleanup_a(&actions);
+       return ret;
+}
+
+static int
+tcf_add_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
+              u32 portid)
+{
+       struct sk_buff *skb;
+       int err = 0;
+
+       skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
+       if (!skb)
+               return -ENOBUFS;
+
+       if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
+                        RTM_NEWACTION, 0, 0) <= 0) {
+               kfree_skb(skb);
+               return -EINVAL;
+       }
+
+       err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
+                            n->nlmsg_flags & NLM_F_ECHO);
+       if (err > 0)
+               err = 0;
+       return err;
+}
+
+static int
+tcf_action_add(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
+              u32 portid, int ovr)
+{
+       int ret = 0;
+       LIST_HEAD(actions);
+
+       ret = tcf_action_init(net, nla, NULL, NULL, ovr, 0, &actions);
+       if (ret)
+               goto done;
+
+       /* dump then free all the actions after update; inserted policy
+        * stays intact
+        */
+       ret = tcf_add_notify(net, n, &actions, portid);
+       cleanup_a(&actions);
+done:
+       return ret;
+}
+
+static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n)
+{
+       struct net *net = sock_net(skb->sk);
+       struct nlattr *tca[TCA_ACT_MAX + 1];
+       u32 portid = skb ? NETLINK_CB(skb).portid : 0;
+       int ret = 0, ovr = 0;
+
+       if ((n->nlmsg_type != RTM_GETACTION) && !netlink_capable(skb, CAP_NET_ADMIN))
+               return -EPERM;
+
+       ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ACT_MAX, NULL);
+       if (ret < 0)
+               return ret;
+
+       if (tca[TCA_ACT_TAB] == NULL) {
+               pr_notice("tc_ctl_action: received NO action attribs\n");
+               return -EINVAL;
+       }
+
+       /* n->nlmsg_flags & NLM_F_CREATE */
+       switch (n->nlmsg_type) {
+       case RTM_NEWACTION:
+               /* we are going to assume all other flags
+                * imply create only if it doesn't exist
+                * Note that CREATE | EXCL implies that
+                * but since we want avoid ambiguity (eg when flags
+                * is zero) then just set this
+                */
+               if (n->nlmsg_flags & NLM_F_REPLACE)
+                       ovr = 1;
+replay:
+               ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr);
+               if (ret == -EAGAIN)
+                       goto replay;
+               break;
+       case RTM_DELACTION:
+               ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
+                                   portid, RTM_DELACTION);
+               break;
+       case RTM_GETACTION:
+               ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
+                                   portid, RTM_GETACTION);
+               break;
+       default:
+               BUG();
+       }
+
+       return ret;
+}
+
+static struct nlattr *
+find_dump_kind(const struct nlmsghdr *n)
+{
+       struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
+       struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
+       struct nlattr *nla[TCAA_MAX + 1];
+       struct nlattr *kind;
+
+       if (nlmsg_parse(n, sizeof(struct tcamsg), nla, TCAA_MAX, NULL) < 0)
+               return NULL;
+       tb1 = nla[TCA_ACT_TAB];
+       if (tb1 == NULL)
+               return NULL;
+
+       if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
+                     NLMSG_ALIGN(nla_len(tb1)), NULL) < 0)
+               return NULL;
+
+       if (tb[1] == NULL)
+               return NULL;
+       if (nla_parse(tb2, TCA_ACT_MAX, nla_data(tb[1]),
+                     nla_len(tb[1]), NULL) < 0)
+               return NULL;
+       kind = tb2[TCA_ACT_KIND];
+
+       return kind;
+}
+
+static int
+tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
+{
+       struct nlmsghdr *nlh;
+       unsigned char *b = skb_tail_pointer(skb);
+       struct nlattr *nest;
+       struct tc_action_ops *a_o;
+       struct tc_action a;
+       int ret = 0;
+       struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
+       struct nlattr *kind = find_dump_kind(cb->nlh);
+
+       if (kind == NULL) {
+               pr_info("tc_dump_action: action bad kind\n");
+               return 0;
+       }
+
+       a_o = tc_lookup_action(kind);
+       if (a_o == NULL)
+               return 0;
+
+       memset(&a, 0, sizeof(struct tc_action));
+       a.ops = a_o;
+
+       nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
+                       cb->nlh->nlmsg_type, sizeof(*t), 0);
+       if (!nlh)
+               goto out_module_put;
+       t = nlmsg_data(nlh);
+       t->tca_family = AF_UNSPEC;
+       t->tca__pad1 = 0;
+       t->tca__pad2 = 0;
+
+       nest = nla_nest_start(skb, TCA_ACT_TAB);
+       if (nest == NULL)
+               goto out_module_put;
+
+       ret = a_o->walk(skb, cb, RTM_GETACTION, &a);
+       if (ret < 0)
+               goto out_module_put;
+
+       if (ret > 0) {
+               nla_nest_end(skb, nest);
+               ret = skb->len;
+       } else
+               nla_nest_cancel(skb, nest);
+
+       nlh->nlmsg_len = skb_tail_pointer(skb) - b;
+       if (NETLINK_CB(cb->skb).portid && ret)
+               nlh->nlmsg_flags |= NLM_F_MULTI;
+       module_put(a_o->owner);
+       return skb->len;
+
+out_module_put:
+       module_put(a_o->owner);
+       nlmsg_trim(skb, b);
+       return skb->len;
+}
+
+static int __init tc_action_init(void)
+{
+       rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, NULL);
+       rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, NULL);
+       rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
+                     NULL);
+
+       return 0;
+}
+
+subsys_initcall(tc_action_init);