sched/act_ipt.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/act_ipt.c          iptables target interface
   4  *
   5  *TODO: Add other tables. For now we only support the ipv4 table targets
   6  *
   7  * Copyright:   Jamal Hadi Salim (2002-13)
   8  */
   9
  10 #include <linux/types.h>
  11 #include <linux/kernel.h>
  12 #include <linux/string.h>
  13 #include <linux/errno.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/rtnetlink.h>
  16 #include <linux/module.h>
  17 #include <linux/init.h>
  18 #include <linux/slab.h>
  19 #include <net/netlink.h>
  20 #include <net/pkt_sched.h>
  21 #include <linux/tc_act/tc_ipt.h>
  22 #include <net/tc_act/tc_ipt.h>
  23
  24 #include <linux/netfilter_ipv4/ip_tables.h>
  25
  26
  27 static struct tc_action_ops act_ipt_ops;
  28 static struct tc_action_ops act_xt_ops;
  29
  30 static int ipt_init_target(struct net *net, struct xt_entry_target *t,
  31                            char *table, unsigned int hook)
  32 {
  33         struct xt_tgchk_param par;
  34         struct xt_target *target;
  35         struct ipt_entry e = {};
  36         int ret = 0;
  37
  38         target = xt_request_find_target(AF_INET, t->u.user.name,
  39                                         t->u.user.revision);
  40         if (IS_ERR(target))
  41                 return PTR_ERR(target);
  42
  43         t->u.kernel.target = target;
  44         memset(&par, 0, sizeof(par));
  45         par.net       = net;
  46         par.table     = table;
  47         par.entryinfo = &e;
  48         par.target    = target;
  49         par.targinfo  = t->data;
  50         par.hook_mask = 1 << hook;
  51         par.family    = NFPROTO_IPV4;
  52
  53         ret = xt_check_target(&par, t->u.target_size - sizeof(*t), 0, false);
  54         if (ret < 0) {
  55                 module_put(t->u.kernel.target->me);
  56                 return ret;
  57         }
  58         return 0;
  59 }
  60
  61 static void ipt_destroy_target(struct xt_entry_target *t, struct net *net)
  62 {
  63         struct xt_tgdtor_param par = {
  64                 .target   = t->u.kernel.target,
  65                 .targinfo = t->data,
  66                 .family   = NFPROTO_IPV4,
  67                 .net      = net,
  68         };
  69         if (par.target->destroy != NULL)
  70                 par.target->destroy(&par);
  71         module_put(par.target->me);
  72 }
  73
  74 static void tcf_ipt_release(struct tc_action *a)
  75 {
  76         struct tcf_ipt *ipt = to_ipt(a);
  77
  78         if (ipt->tcfi_t) {
  79                 ipt_destroy_target(ipt->tcfi_t, a->idrinfo->net);
  80                 kfree(ipt->tcfi_t);
  81         }
  82         kfree(ipt->tcfi_tname);
  83 }
  84
  85 static const struct nla_policy ipt_policy[TCA_IPT_MAX + 1] = {
  86         [TCA_IPT_TABLE] = { .type = NLA_STRING, .len = IFNAMSIZ },
  87         [TCA_IPT_HOOK]  = NLA_POLICY_RANGE(NLA_U32, NF_INET_PRE_ROUTING,
  88                                            NF_INET_NUMHOOKS),
  89         [TCA_IPT_INDEX] = { .type = NLA_U32 },
  90         [TCA_IPT_TARG]  = { .len = sizeof(struct xt_entry_target) },
  91 };
  92
  93 static int __tcf_ipt_init(struct net *net, unsigned int id, struct nlattr *nla,
  94                           struct nlattr *est, struct tc_action **a,
  95                           const struct tc_action_ops *ops,
  96                           struct tcf_proto *tp, u32 flags)
  97 {
  98         struct tc_action_net *tn = net_generic(net, id);
  99         bool bind = flags & TCA_ACT_FLAGS_BIND;
 100         struct nlattr *tb[TCA_IPT_MAX + 1];
 101         struct tcf_ipt *ipt;
 102         struct xt_entry_target *td, *t;
 103         char *tname;
 104         bool exists = false;
 105         int ret = 0, err;
 106         u32 hook = 0;
 107         u32 index = 0;
 108
 109         if (nla == NULL)
 110                 return -EINVAL;
 111
 112         err = nla_parse_nested_deprecated(tb, TCA_IPT_MAX, nla, ipt_policy,
 113                                           NULL);
 114         if (err < 0)
 115                 return err;
 116
 117         if (tb[TCA_IPT_INDEX] != NULL)
 118                 index = nla_get_u32(tb[TCA_IPT_INDEX]);
 119
 120         err = tcf_idr_check_alloc(tn, &index, a, bind);
 121         if (err < 0)
 122                 return err;
 123         exists = err;
 124         if (exists && bind)
 125                 return 0;
 126
 127         if (tb[TCA_IPT_HOOK] == NULL || tb[TCA_IPT_TARG] == NULL) {
 128                 if (exists)
 129                         tcf_idr_release(*a, bind);
 130                 else
 131                         tcf_idr_cleanup(tn, index);
 132                 return -EINVAL;
 133         }
 134
 135         td = (struct xt_entry_target *)nla_data(tb[TCA_IPT_TARG]);
 136         if (nla_len(tb[TCA_IPT_TARG]) != td->u.target_size) {
 137                 if (exists)
 138                         tcf_idr_release(*a, bind);
 139                 else
 140                         tcf_idr_cleanup(tn, index);
 141                 return -EINVAL;
 142         }
 143
 144         if (!exists) {
 145                 ret = tcf_idr_create(tn, index, est, a, ops, bind,
 146                                      false, flags);
 147                 if (ret) {
 148                         tcf_idr_cleanup(tn, index);
 149                         return ret;
 150                 }
 151                 ret = ACT_P_CREATED;
 152         } else {
 153                 if (bind)/* dont override defaults */
 154                         return 0;
 155
 156                 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
 157                         tcf_idr_release(*a, bind);
 158                         return -EEXIST;
 159                 }
 160         }
 161
 162         err = -EINVAL;
 163         hook = nla_get_u32(tb[TCA_IPT_HOOK]);
 164         switch (hook) {
 165         case NF_INET_PRE_ROUTING:
 166                 break;
 167         case NF_INET_POST_ROUTING:
 168                 break;
 169         default:
 170                 goto err1;
 171         }
 172
 173         if (tb[TCA_IPT_TABLE]) {
 174                 /* mangle only for now */
 175                 if (nla_strcmp(tb[TCA_IPT_TABLE], "mangle"))
 176                         goto err1;
 177         }
 178
 179         tname = kstrdup("mangle", GFP_KERNEL);
 180         if (unlikely(!tname))
 181                 goto err1;
 182
 183         t = kmemdup(td, td->u.target_size, GFP_KERNEL);
 184         if (unlikely(!t))
 185                 goto err2;
 186
 187         err = ipt_init_target(net, t, tname, hook);
 188         if (err < 0)
 189                 goto err3;
 190
 191         ipt = to_ipt(*a);
 192
 193         spin_lock_bh(&ipt->tcf_lock);
 194         if (ret != ACT_P_CREATED) {
 195                 ipt_destroy_target(ipt->tcfi_t, net);
 196                 kfree(ipt->tcfi_tname);
 197                 kfree(ipt->tcfi_t);
 198         }
 199         ipt->tcfi_tname = tname;
 200         ipt->tcfi_t     = t;
 201         ipt->tcfi_hook  = hook;
 202         spin_unlock_bh(&ipt->tcf_lock);
 203         return ret;
 204
 205 err3:
 206         kfree(t);
 207 err2:
 208         kfree(tname);
 209 err1:
 210         tcf_idr_release(*a, bind);
 211         return err;
 212 }
 213
 214 static int tcf_ipt_init(struct net *net, struct nlattr *nla,
 215                         struct nlattr *est, struct tc_action **a,
 216                         struct tcf_proto *tp,
 217                         u32 flags, struct netlink_ext_ack *extack)
 218 {
 219         return __tcf_ipt_init(net, act_ipt_ops.net_id, nla, est,
 220                               a, &act_ipt_ops, tp, flags);
 221 }
 222
 223 static int tcf_xt_init(struct net *net, struct nlattr *nla,
 224                        struct nlattr *est, struct tc_action **a,
 225                        struct tcf_proto *tp,
 226                        u32 flags, struct netlink_ext_ack *extack)
 227 {
 228         return __tcf_ipt_init(net, act_xt_ops.net_id, nla, est,
 229                               a, &act_xt_ops, tp, flags);
 230 }
 231
 232 static bool tcf_ipt_act_check(struct sk_buff *skb)
 233 {
 234         const struct iphdr *iph;
 235         unsigned int nhoff, len;
 236
 237         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 238                 return false;
 239
 240         nhoff = skb_network_offset(skb);
 241         iph = ip_hdr(skb);
 242         if (iph->ihl < 5 || iph->version != 4)
 243                 return false;
 244
 245         len = skb_ip_totlen(skb);
 246         if (skb->len < nhoff + len || len < (iph->ihl * 4u))
 247                 return false;
 248
 249         return pskb_may_pull(skb, iph->ihl * 4u);
 250 }
 251
 252 static int tcf_ipt_act(struct sk_buff *skb, const struct tc_action *a,
 253                        struct tcf_result *res)
 254 {
 255         int ret = 0, result = 0;
 256         struct tcf_ipt *ipt = to_ipt(a);
 257         struct xt_action_param par;
 258         struct nf_hook_state state = {
 259                 .net    = dev_net(skb->dev),
 260                 .in     = skb->dev,
 261                 .hook   = ipt->tcfi_hook,
 262                 .pf     = NFPROTO_IPV4,
 263         };
 264
 265         if (skb_protocol(skb, false) != htons(ETH_P_IP))
 266                 return TC_ACT_UNSPEC;
 267
 268         if (skb_unclone(skb, GFP_ATOMIC))
 269                 return TC_ACT_UNSPEC;
 270
 271         if (!tcf_ipt_act_check(skb))
 272                 return TC_ACT_UNSPEC;
 273
 274         if (state.hook == NF_INET_POST_ROUTING) {
 275                 if (!skb_dst(skb))
 276                         return TC_ACT_UNSPEC;
 277
 278                 state.out = skb->dev;
 279         }
 280
 281         spin_lock(&ipt->tcf_lock);
 282
 283         tcf_lastuse_update(&ipt->tcf_tm);
 284         bstats_update(&ipt->tcf_bstats, skb);
 285
 286         /* yes, we have to worry about both in and out dev
 287          * worry later - danger - this API seems to have changed
 288          * from earlier kernels
 289          */
 290         par.state    = &state;
 291         par.target   = ipt->tcfi_t->u.kernel.target;
 292         par.targinfo = ipt->tcfi_t->data;
 293         ret = par.target->target(skb, &par);
 294
 295         switch (ret) {
 296         case NF_ACCEPT:
 297                 result = TC_ACT_OK;
 298                 break;
 299         case NF_DROP:
 300                 result = TC_ACT_SHOT;
 301                 ipt->tcf_qstats.drops++;
 302                 break;
 303         case XT_CONTINUE:
 304                 result = TC_ACT_PIPE;
 305                 break;
 306         default:
 307                 net_notice_ratelimited("tc filter: Bogus netfilter code %d assume ACCEPT\n",
 308                                        ret);
 309                 result = TC_ACT_OK;
 310                 break;
 311         }
 312         spin_unlock(&ipt->tcf_lock);
 313         return result;
 314
 315 }
 316
 317 static int tcf_ipt_dump(struct sk_buff *skb, struct tc_action *a, int bind,
 318                         int ref)
 319 {
 320         unsigned char *b = skb_tail_pointer(skb);
 321         struct tcf_ipt *ipt = to_ipt(a);
 322         struct xt_entry_target *t;
 323         struct tcf_t tm;
 324         struct tc_cnt c;
 325
 326         /* for simple targets kernel size == user size
 327          * user name = target name
 328          * for foolproof you need to not assume this
 329          */
 330
 331         spin_lock_bh(&ipt->tcf_lock);
 332         t = kmemdup(ipt->tcfi_t, ipt->tcfi_t->u.user.target_size, GFP_ATOMIC);
 333         if (unlikely(!t))
 334                 goto nla_put_failure;
 335
 336         c.bindcnt = atomic_read(&ipt->tcf_bindcnt) - bind;
 337         c.refcnt = refcount_read(&ipt->tcf_refcnt) - ref;
 338         strcpy(t->u.user.name, ipt->tcfi_t->u.kernel.target->name);
 339
 340         if (nla_put(skb, TCA_IPT_TARG, ipt->tcfi_t->u.user.target_size, t) ||
 341             nla_put_u32(skb, TCA_IPT_INDEX, ipt->tcf_index) ||
 342             nla_put_u32(skb, TCA_IPT_HOOK, ipt->tcfi_hook) ||
 343             nla_put(skb, TCA_IPT_CNT, sizeof(struct tc_cnt), &c) ||
 344             nla_put_string(skb, TCA_IPT_TABLE, ipt->tcfi_tname))
 345                 goto nla_put_failure;
 346
 347         tcf_tm_dump(&tm, &ipt->tcf_tm);
 348         if (nla_put_64bit(skb, TCA_IPT_TM, sizeof(tm), &tm, TCA_IPT_PAD))
 349                 goto nla_put_failure;
 350
 351         spin_unlock_bh(&ipt->tcf_lock);
 352         kfree(t);
 353         return skb->len;
 354
 355 nla_put_failure:
 356         spin_unlock_bh(&ipt->tcf_lock);
 357         nlmsg_trim(skb, b);
 358         kfree(t);
 359         return -1;
 360 }
 361
 362 static struct tc_action_ops act_ipt_ops = {
 363         .kind           =       "ipt",
 364         .id             =       TCA_ID_IPT,
 365         .owner          =       THIS_MODULE,
 366         .act            =       tcf_ipt_act,
 367         .dump           =       tcf_ipt_dump,
 368         .cleanup        =       tcf_ipt_release,
 369         .init           =       tcf_ipt_init,
 370         .size           =       sizeof(struct tcf_ipt),
 371 };
 372
 373 static __net_init int ipt_init_net(struct net *net)
 374 {
 375         struct tc_action_net *tn = net_generic(net, act_ipt_ops.net_id);
 376
 377         return tc_action_net_init(net, tn, &act_ipt_ops);
 378 }
 379
 380 static void __net_exit ipt_exit_net(struct list_head *net_list)
 381 {
 382         tc_action_net_exit(net_list, act_ipt_ops.net_id);
 383 }
 384
 385 static struct pernet_operations ipt_net_ops = {
 386         .init = ipt_init_net,
 387         .exit_batch = ipt_exit_net,
 388         .id   = &act_ipt_ops.net_id,
 389         .size = sizeof(struct tc_action_net),
 390 };
 391
 392 static struct tc_action_ops act_xt_ops = {
 393         .kind           =       "xt",
 394         .id             =       TCA_ID_XT,
 395         .owner          =       THIS_MODULE,
 396         .act            =       tcf_ipt_act,
 397         .dump           =       tcf_ipt_dump,
 398         .cleanup        =       tcf_ipt_release,
 399         .init           =       tcf_xt_init,
 400         .size           =       sizeof(struct tcf_ipt),
 401 };
 402
 403 static __net_init int xt_init_net(struct net *net)
 404 {
 405         struct tc_action_net *tn = net_generic(net, act_xt_ops.net_id);
 406
 407         return tc_action_net_init(net, tn, &act_xt_ops);
 408 }
 409
 410 static void __net_exit xt_exit_net(struct list_head *net_list)
 411 {
 412         tc_action_net_exit(net_list, act_xt_ops.net_id);
 413 }
 414
 415 static struct pernet_operations xt_net_ops = {
 416         .init = xt_init_net,
 417         .exit_batch = xt_exit_net,
 418         .id   = &act_xt_ops.net_id,
 419         .size = sizeof(struct tc_action_net),
 420 };
 421
 422 MODULE_AUTHOR("Jamal Hadi Salim(2002-13)");
 423 MODULE_DESCRIPTION("Iptables target actions");
 424 MODULE_LICENSE("GPL");
 425 MODULE_ALIAS("act_xt");
 426
 427 static int __init ipt_init_module(void)
 428 {
 429         int ret1, ret2;
 430
 431         ret1 = tcf_register_action(&act_xt_ops, &xt_net_ops);
 432         if (ret1 < 0)
 433                 pr_err("Failed to load xt action\n");
 434
 435         ret2 = tcf_register_action(&act_ipt_ops, &ipt_net_ops);
 436         if (ret2 < 0)
 437                 pr_err("Failed to load ipt action\n");
 438
 439         if (ret1 < 0 && ret2 < 0) {
 440                 return ret1;
 441         } else
 442                 return 0;
 443 }
 444
 445 static void __exit ipt_cleanup_module(void)
 446 {
 447         tcf_unregister_action(&act_ipt_ops, &ipt_net_ops);
 448         tcf_unregister_action(&act_xt_ops, &xt_net_ops);
 449 }
 450
 451 module_init(ipt_init_module);
 452 module_exit(ipt_cleanup_module);