net/sched/act_connmark.c

   1 /*
   2  * net/sched/act_connmark.c  netfilter connmark retriever action
   3  * skb mark is over-written
   4  *
   5  * Copyright (c) 2011 Felix Fietkau <nbd@openwrt.org>
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11 */
  12
  13 #include <linux/module.h>
  14 #include <linux/init.h>
  15 #include <linux/kernel.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/rtnetlink.h>
  18 #include <linux/pkt_cls.h>
  19 #include <linux/ip.h>
  20 #include <linux/ipv6.h>
  21 #include <net/netlink.h>
  22 #include <net/pkt_sched.h>
  23 #include <net/act_api.h>
  24 #include <uapi/linux/tc_act/tc_connmark.h>
  25 #include <net/tc_act/tc_connmark.h>
  26
  27 #include <net/netfilter/nf_conntrack.h>
  28 #include <net/netfilter/nf_conntrack_core.h>
  29 #include <net/netfilter/nf_conntrack_zones.h>
  30
  31 static unsigned int connmark_net_id;
  32 static struct tc_action_ops act_connmark_ops;
  33
  34 static int tcf_connmark_act(struct sk_buff *skb, const struct tc_action *a,
  35                             struct tcf_result *res)
  36 {
  37         const struct nf_conntrack_tuple_hash *thash;
  38         struct nf_conntrack_tuple tuple;
  39         enum ip_conntrack_info ctinfo;
  40         struct tcf_connmark_info *ca = to_connmark(a);
  41         struct nf_conntrack_zone zone;
  42         struct nf_conn *c;
  43         int proto;
  44
  45         spin_lock(&ca->tcf_lock);
  46         tcf_lastuse_update(&ca->tcf_tm);
  47         bstats_update(&ca->tcf_bstats, skb);
  48
  49         switch (skb_protocol(skb, true)) {
  50         case htons(ETH_P_IP):
  51                 if (skb->len < sizeof(struct iphdr))
  52                         goto out;
  53
  54                 proto = NFPROTO_IPV4;
  55                 break;
  56         case htons(ETH_P_IPV6):
  57                 if (skb->len < sizeof(struct ipv6hdr))
  58                         goto out;
  59
  60                 proto = NFPROTO_IPV6;
  61                 break;
  62         default:
  63                 goto out;
  64         }
  65
  66         c = nf_ct_get(skb, &ctinfo);
  67         if (c) {
  68                 skb->mark = c->mark;
  69                 /* using overlimits stats to count how many packets marked */
  70                 ca->tcf_qstats.overlimits++;
  71                 goto out;
  72         }
  73
  74         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
  75                                proto, ca->net, &tuple))
  76                 goto out;
  77
  78         zone.id = ca->zone;
  79         zone.dir = NF_CT_DEFAULT_ZONE_DIR;
  80
  81         thash = nf_conntrack_find_get(ca->net, &zone, &tuple);
  82         if (!thash)
  83                 goto out;
  84
  85         c = nf_ct_tuplehash_to_ctrack(thash);
  86         /* using overlimits stats to count how many packets marked */
  87         ca->tcf_qstats.overlimits++;
  88         skb->mark = c->mark;
  89         nf_ct_put(c);
  90
  91 out:
  92         spin_unlock(&ca->tcf_lock);
  93         return ca->tcf_action;
  94 }
  95
  96 static const struct nla_policy connmark_policy[TCA_CONNMARK_MAX + 1] = {
  97         [TCA_CONNMARK_PARMS] = { .len = sizeof(struct tc_connmark) },
  98 };
  99
 100 static int tcf_connmark_init(struct net *net, struct nlattr *nla,
 101                              struct nlattr *est, struct tc_action **a,
 102                              int ovr, int bind, bool rtnl_held,
 103                              struct netlink_ext_ack *extack)
 104 {
 105         struct tc_action_net *tn = net_generic(net, connmark_net_id);
 106         struct nlattr *tb[TCA_CONNMARK_MAX + 1];
 107         struct tcf_connmark_info *ci;
 108         struct tc_connmark *parm;
 109         int ret = 0;
 110         u32 index;
 111
 112         if (!nla)
 113                 return -EINVAL;
 114
 115         ret = nla_parse_nested(tb, TCA_CONNMARK_MAX, nla, connmark_policy,
 116                                NULL);
 117         if (ret < 0)
 118                 return ret;
 119
 120         if (!tb[TCA_CONNMARK_PARMS])
 121                 return -EINVAL;
 122
 123         parm = nla_data(tb[TCA_CONNMARK_PARMS]);
 124         index = parm->index;
 125         ret = tcf_idr_check_alloc(tn, &index, a, bind);
 126         if (!ret) {
 127                 ret = tcf_idr_create(tn, index, est, a,
 128                                      &act_connmark_ops, bind, false);
 129                 if (ret) {
 130                         tcf_idr_cleanup(tn, index);
 131                         return ret;
 132                 }
 133
 134                 ci = to_connmark(*a);
 135                 ci->tcf_action = parm->action;
 136                 ci->net = net;
 137                 ci->zone = parm->zone;
 138
 139                 tcf_idr_insert(tn, *a);
 140                 ret = ACT_P_CREATED;
 141         } else if (ret > 0) {
 142                 ci = to_connmark(*a);
 143                 if (bind)
 144                         return 0;
 145                 if (!ovr) {
 146                         tcf_idr_release(*a, bind);
 147                         return -EEXIST;
 148                 }
 149                 /* replacing action and zone */
 150                 ci->tcf_action = parm->action;
 151                 ci->zone = parm->zone;
 152                 ret = 0;
 153         }
 154
 155         return ret;
 156 }
 157
 158 static inline int tcf_connmark_dump(struct sk_buff *skb, struct tc_action *a,
 159                                     int bind, int ref)
 160 {
 161         unsigned char *b = skb_tail_pointer(skb);
 162         struct tcf_connmark_info *ci = to_connmark(a);
 163
 164         struct tc_connmark opt = {
 165                 .index   = ci->tcf_index,
 166                 .refcnt  = refcount_read(&ci->tcf_refcnt) - ref,
 167                 .bindcnt = atomic_read(&ci->tcf_bindcnt) - bind,
 168                 .action  = ci->tcf_action,
 169                 .zone   = ci->zone,
 170         };
 171         struct tcf_t t;
 172
 173         if (nla_put(skb, TCA_CONNMARK_PARMS, sizeof(opt), &opt))
 174                 goto nla_put_failure;
 175
 176         tcf_tm_dump(&t, &ci->tcf_tm);
 177         if (nla_put_64bit(skb, TCA_CONNMARK_TM, sizeof(t), &t,
 178                           TCA_CONNMARK_PAD))
 179                 goto nla_put_failure;
 180
 181         return skb->len;
 182 nla_put_failure:
 183         nlmsg_trim(skb, b);
 184         return -1;
 185 }
 186
 187 static int tcf_connmark_walker(struct net *net, struct sk_buff *skb,
 188                                struct netlink_callback *cb, int type,
 189                                const struct tc_action_ops *ops,
 190                                struct netlink_ext_ack *extack)
 191 {
 192         struct tc_action_net *tn = net_generic(net, connmark_net_id);
 193
 194         return tcf_generic_walker(tn, skb, cb, type, ops, extack);
 195 }
 196
 197 static int tcf_connmark_search(struct net *net, struct tc_action **a, u32 index,
 198                                struct netlink_ext_ack *extack)
 199 {
 200         struct tc_action_net *tn = net_generic(net, connmark_net_id);
 201
 202         return tcf_idr_search(tn, a, index);
 203 }
 204
 205 static struct tc_action_ops act_connmark_ops = {
 206         .kind           =       "connmark",
 207         .type           =       TCA_ACT_CONNMARK,
 208         .owner          =       THIS_MODULE,
 209         .act            =       tcf_connmark_act,
 210         .dump           =       tcf_connmark_dump,
 211         .init           =       tcf_connmark_init,
 212         .walk           =       tcf_connmark_walker,
 213         .lookup         =       tcf_connmark_search,
 214         .size           =       sizeof(struct tcf_connmark_info),
 215 };
 216
 217 static __net_init int connmark_init_net(struct net *net)
 218 {
 219         struct tc_action_net *tn = net_generic(net, connmark_net_id);
 220
 221         return tc_action_net_init(net, tn, &act_connmark_ops);
 222 }
 223
 224 static void __net_exit connmark_exit_net(struct list_head *net_list)
 225 {
 226         tc_action_net_exit(net_list, connmark_net_id);
 227 }
 228
 229 static struct pernet_operations connmark_net_ops = {
 230         .init = connmark_init_net,
 231         .exit_batch = connmark_exit_net,
 232         .id   = &connmark_net_id,
 233         .size = sizeof(struct tc_action_net),
 234 };
 235
 236 static int __init connmark_init_module(void)
 237 {
 238         return tcf_register_action(&act_connmark_ops, &connmark_net_ops);
 239 }
 240
 241 static void __exit connmark_cleanup_module(void)
 242 {
 243         tcf_unregister_action(&act_connmark_ops, &connmark_net_ops);
 244 }
 245
 246 module_init(connmark_init_module);
 247 module_exit(connmark_cleanup_module);
 248 MODULE_AUTHOR("Felix Fietkau <nbd@openwrt.org>");
 249 MODULE_DESCRIPTION("Connection tracking mark restoring");
 250 MODULE_LICENSE("GPL");