net/sched/act_api.c

   1 /*
   2  * net/sched/act_api.c  Packet action API.
   3  *
   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.
   8  *
   9  * Author:      Jamal Hadi Salim
  10  *
  11  *
  12  */
  13
  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 <net/net_namespace.h>
  25 #include <net/sock.h>
  26 #include <net/sch_generic.h>
  27 #include <net/act_api.h>
  28 #include <net/netlink.h>
  29
  30 static void free_tcf(struct rcu_head *head)
  31 {
  32         struct tc_action *p = container_of(head, struct tc_action, tcfa_rcu);
  33
  34         free_percpu(p->cpu_bstats);
  35         free_percpu(p->cpu_qstats);
  36         kfree(p);
  37 }
  38
  39 static void tcf_hash_destroy(struct tcf_hashinfo *hinfo, struct tc_action *p)
  40 {
  41         spin_lock_bh(&hinfo->lock);
  42         hlist_del(&p->tcfa_head);
  43         spin_unlock_bh(&hinfo->lock);
  44         gen_kill_estimator(&p->tcfa_bstats,
  45                            &p->tcfa_rate_est);
  46         /*
  47          * gen_estimator est_timer() might access p->tcfa_lock
  48          * or bstats, wait a RCU grace period before freeing p
  49          */
  50         call_rcu(&p->tcfa_rcu, free_tcf);
  51 }
  52
  53 int __tcf_hash_release(struct tc_action *p, bool bind, bool strict)
  54 {
  55         int ret = 0;
  56
  57         if (p) {
  58                 if (bind)
  59                         p->tcfa_bindcnt--;
  60                 else if (strict && p->tcfa_bindcnt > 0)
  61                         return -EPERM;
  62
  63                 p->tcfa_refcnt--;
  64                 if (p->tcfa_bindcnt <= 0 && p->tcfa_refcnt <= 0) {
  65                         if (p->ops->cleanup)
  66                                 p->ops->cleanup(p, bind);
  67                         tcf_hash_destroy(p->hinfo, p);
  68                         ret = ACT_P_DELETED;
  69                 }
  70         }
  71
  72         return ret;
  73 }
  74 EXPORT_SYMBOL(__tcf_hash_release);
  75
  76 static int tcf_dump_walker(struct tcf_hashinfo *hinfo, struct sk_buff *skb,
  77                            struct netlink_callback *cb)
  78 {
  79         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
  80         struct nlattr *nest;
  81
  82         spin_lock_bh(&hinfo->lock);
  83
  84         s_i = cb->args[0];
  85
  86         for (i = 0; i < (hinfo->hmask + 1); i++) {
  87                 struct hlist_head *head;
  88                 struct tc_action *p;
  89
  90                 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
  91
  92                 hlist_for_each_entry_rcu(p, head, tcfa_head) {
  93                         index++;
  94                         if (index < s_i)
  95                                 continue;
  96
  97                         nest = nla_nest_start(skb, n_i);
  98                         if (nest == NULL) {
  99                                 index--;
 100                                 goto nla_put_failure;
 101                         }
 102                         err = tcf_action_dump_1(skb, p, 0, 0);
 103                         if (err < 0) {
 104                                 index--;
 105                                 nlmsg_trim(skb, nest);
 106                                 goto done;
 107                         }
 108                         nla_nest_end(skb, nest);
 109                         n_i++;
 110                         if (n_i >= TCA_ACT_MAX_PRIO)
 111                                 goto done;
 112                 }
 113         }
 114 done:
 115         spin_unlock_bh(&hinfo->lock);
 116         if (n_i)
 117                 cb->args[0] += n_i;
 118         return n_i;
 119
 120 nla_put_failure:
 121         nla_nest_cancel(skb, nest);
 122         goto done;
 123 }
 124
 125 static int tcf_del_walker(struct tcf_hashinfo *hinfo, struct sk_buff *skb,
 126                           const struct tc_action_ops *ops)
 127 {
 128         struct nlattr *nest;
 129         int i = 0, n_i = 0;
 130         int ret = -EINVAL;
 131
 132         nest = nla_nest_start(skb, 0);
 133         if (nest == NULL)
 134                 goto nla_put_failure;
 135         if (nla_put_string(skb, TCA_KIND, ops->kind))
 136                 goto nla_put_failure;
 137         for (i = 0; i < (hinfo->hmask + 1); i++) {
 138                 struct hlist_head *head;
 139                 struct hlist_node *n;
 140                 struct tc_action *p;
 141
 142                 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
 143                 hlist_for_each_entry_safe(p, n, head, tcfa_head) {
 144                         ret = __tcf_hash_release(p, false, true);
 145                         if (ret == ACT_P_DELETED) {
 146                                 module_put(ops->owner);
 147                                 n_i++;
 148                         } else if (ret < 0)
 149                                 goto nla_put_failure;
 150                 }
 151         }
 152         if (nla_put_u32(skb, TCA_FCNT, n_i))
 153                 goto nla_put_failure;
 154         nla_nest_end(skb, nest);
 155
 156         return n_i;
 157 nla_put_failure:
 158         nla_nest_cancel(skb, nest);
 159         return ret;
 160 }
 161
 162 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 163                        struct netlink_callback *cb, int type,
 164                        const struct tc_action_ops *ops)
 165 {
 166         struct tcf_hashinfo *hinfo = tn->hinfo;
 167
 168         if (type == RTM_DELACTION) {
 169                 return tcf_del_walker(hinfo, skb, ops);
 170         } else if (type == RTM_GETACTION) {
 171                 return tcf_dump_walker(hinfo, skb, cb);
 172         } else {
 173                 WARN(1, "tcf_generic_walker: unknown action %d\n", type);
 174                 return -EINVAL;
 175         }
 176 }
 177 EXPORT_SYMBOL(tcf_generic_walker);
 178
 179 static struct tc_action *tcf_hash_lookup(u32 index, struct tcf_hashinfo *hinfo)
 180 {
 181         struct tc_action *p = NULL;
 182         struct hlist_head *head;
 183
 184         spin_lock_bh(&hinfo->lock);
 185         head = &hinfo->htab[tcf_hash(index, hinfo->hmask)];
 186         hlist_for_each_entry_rcu(p, head, tcfa_head)
 187                 if (p->tcfa_index == index)
 188                         break;
 189         spin_unlock_bh(&hinfo->lock);
 190
 191         return p;
 192 }
 193
 194 u32 tcf_hash_new_index(struct tc_action_net *tn)
 195 {
 196         struct tcf_hashinfo *hinfo = tn->hinfo;
 197         u32 val = hinfo->index;
 198
 199         do {
 200                 if (++val == 0)
 201                         val = 1;
 202         } while (tcf_hash_lookup(val, hinfo));
 203
 204         hinfo->index = val;
 205         return val;
 206 }
 207 EXPORT_SYMBOL(tcf_hash_new_index);
 208
 209 int tcf_hash_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 210 {
 211         struct tcf_hashinfo *hinfo = tn->hinfo;
 212         struct tc_action *p = tcf_hash_lookup(index, hinfo);
 213
 214         if (p) {
 215                 *a = p;
 216                 return 1;
 217         }
 218         return 0;
 219 }
 220 EXPORT_SYMBOL(tcf_hash_search);
 221
 222 bool tcf_hash_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
 223                     int bind)
 224 {
 225         struct tcf_hashinfo *hinfo = tn->hinfo;
 226         struct tc_action *p = NULL;
 227
 228         if (index && (p = tcf_hash_lookup(index, hinfo)) != NULL) {
 229                 if (bind)
 230                         p->tcfa_bindcnt++;
 231                 p->tcfa_refcnt++;
 232                 *a = p;
 233                 return true;
 234         }
 235         return false;
 236 }
 237 EXPORT_SYMBOL(tcf_hash_check);
 238
 239 void tcf_hash_cleanup(struct tc_action *a, struct nlattr *est)
 240 {
 241         if (est)
 242                 gen_kill_estimator(&a->tcfa_bstats,
 243                                    &a->tcfa_rate_est);
 244         call_rcu(&a->tcfa_rcu, free_tcf);
 245 }
 246 EXPORT_SYMBOL(tcf_hash_cleanup);
 247
 248 int tcf_hash_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 249                     struct tc_action **a, const struct tc_action_ops *ops,
 250                     int bind, bool cpustats)
 251 {
 252         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 253         struct tcf_hashinfo *hinfo = tn->hinfo;
 254         int err = -ENOMEM;
 255
 256         if (unlikely(!p))
 257                 return -ENOMEM;
 258         p->tcfa_refcnt = 1;
 259         if (bind)
 260                 p->tcfa_bindcnt = 1;
 261
 262         if (cpustats) {
 263                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 264                 if (!p->cpu_bstats) {
 265 err1:
 266                         kfree(p);
 267                         return err;
 268                 }
 269                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 270                 if (!p->cpu_qstats) {
 271 err2:
 272                         free_percpu(p->cpu_bstats);
 273                         goto err1;
 274                 }
 275         }
 276         spin_lock_init(&p->tcfa_lock);
 277         INIT_HLIST_NODE(&p->tcfa_head);
 278         p->tcfa_index = index ? index : tcf_hash_new_index(tn);
 279         p->tcfa_tm.install = jiffies;
 280         p->tcfa_tm.lastuse = jiffies;
 281         p->tcfa_tm.firstuse = 0;
 282         if (est) {
 283                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 284                                         &p->tcfa_rate_est,
 285                                         &p->tcfa_lock, NULL, est);
 286                 if (err) {
 287                         free_percpu(p->cpu_qstats);
 288                         goto err2;
 289                 }
 290         }
 291
 292         p->hinfo = hinfo;
 293         p->ops = ops;
 294         INIT_LIST_HEAD(&p->list);
 295         *a = p;
 296         return 0;
 297 }
 298 EXPORT_SYMBOL(tcf_hash_create);
 299
 300 void tcf_hash_insert(struct tc_action_net *tn, struct tc_action *a)
 301 {
 302         struct tcf_hashinfo *hinfo = tn->hinfo;
 303         unsigned int h = tcf_hash(a->tcfa_index, hinfo->hmask);
 304
 305         spin_lock_bh(&hinfo->lock);
 306         hlist_add_head(&a->tcfa_head, &hinfo->htab[h]);
 307         spin_unlock_bh(&hinfo->lock);
 308 }
 309 EXPORT_SYMBOL(tcf_hash_insert);
 310
 311 void tcf_hashinfo_destroy(const struct tc_action_ops *ops,
 312                           struct tcf_hashinfo *hinfo)
 313 {
 314         int i;
 315
 316         for (i = 0; i < hinfo->hmask + 1; i++) {
 317                 struct tc_action *p;
 318                 struct hlist_node *n;
 319
 320                 hlist_for_each_entry_safe(p, n, &hinfo->htab[i], tcfa_head) {
 321                         int ret;
 322
 323                         ret = __tcf_hash_release(p, false, true);
 324                         if (ret == ACT_P_DELETED)
 325                                 module_put(ops->owner);
 326                         else if (ret < 0)
 327                                 return;
 328                 }
 329         }
 330         kfree(hinfo->htab);
 331 }
 332 EXPORT_SYMBOL(tcf_hashinfo_destroy);
 333
 334 static LIST_HEAD(act_base);
 335 static DEFINE_RWLOCK(act_mod_lock);
 336
 337 int tcf_register_action(struct tc_action_ops *act,
 338                         struct pernet_operations *ops)
 339 {
 340         struct tc_action_ops *a;
 341         int ret;
 342
 343         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 344                 return -EINVAL;
 345
 346         /* We have to register pernet ops before making the action ops visible,
 347          * otherwise tcf_action_init_1() could get a partially initialized
 348          * netns.
 349          */
 350         ret = register_pernet_subsys(ops);
 351         if (ret)
 352                 return ret;
 353
 354         write_lock(&act_mod_lock);
 355         list_for_each_entry(a, &act_base, head) {
 356                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
 357                         write_unlock(&act_mod_lock);
 358                         unregister_pernet_subsys(ops);
 359                         return -EEXIST;
 360                 }
 361         }
 362         list_add_tail(&act->head, &act_base);
 363         write_unlock(&act_mod_lock);
 364
 365         return 0;
 366 }
 367 EXPORT_SYMBOL(tcf_register_action);
 368
 369 int tcf_unregister_action(struct tc_action_ops *act,
 370                           struct pernet_operations *ops)
 371 {
 372         struct tc_action_ops *a;
 373         int err = -ENOENT;
 374
 375         write_lock(&act_mod_lock);
 376         list_for_each_entry(a, &act_base, head) {
 377                 if (a == act) {
 378                         list_del(&act->head);
 379                         err = 0;
 380                         break;
 381                 }
 382         }
 383         write_unlock(&act_mod_lock);
 384         if (!err)
 385                 unregister_pernet_subsys(ops);
 386         return err;
 387 }
 388 EXPORT_SYMBOL(tcf_unregister_action);
 389
 390 /* lookup by name */
 391 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 392 {
 393         struct tc_action_ops *a, *res = NULL;
 394
 395         if (kind) {
 396                 read_lock(&act_mod_lock);
 397                 list_for_each_entry(a, &act_base, head) {
 398                         if (strcmp(kind, a->kind) == 0) {
 399                                 if (try_module_get(a->owner))
 400                                         res = a;
 401                                 break;
 402                         }
 403                 }
 404                 read_unlock(&act_mod_lock);
 405         }
 406         return res;
 407 }
 408
 409 /* lookup by nlattr */
 410 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 411 {
 412         struct tc_action_ops *a, *res = NULL;
 413
 414         if (kind) {
 415                 read_lock(&act_mod_lock);
 416                 list_for_each_entry(a, &act_base, head) {
 417                         if (nla_strcmp(kind, a->kind) == 0) {
 418                                 if (try_module_get(a->owner))
 419                                         res = a;
 420                                 break;
 421                         }
 422                 }
 423                 read_unlock(&act_mod_lock);
 424         }
 425         return res;
 426 }
 427
 428 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 429                     int nr_actions, struct tcf_result *res)
 430 {
 431         int ret = -1, i;
 432
 433         if (skb->tc_verd & TC_NCLS) {
 434                 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
 435                 ret = TC_ACT_OK;
 436                 goto exec_done;
 437         }
 438         for (i = 0; i < nr_actions; i++) {
 439                 const struct tc_action *a = actions[i];
 440
 441 repeat:
 442                 ret = a->ops->act(skb, a, res);
 443                 if (ret == TC_ACT_REPEAT)
 444                         goto repeat;    /* we need a ttl - JHS */
 445                 if (ret != TC_ACT_PIPE)
 446                         goto exec_done;
 447         }
 448 exec_done:
 449         return ret;
 450 }
 451 EXPORT_SYMBOL(tcf_action_exec);
 452
 453 int tcf_action_destroy(struct list_head *actions, int bind)
 454 {
 455         const struct tc_action_ops *ops;
 456         struct tc_action *a, *tmp;
 457         int ret = 0;
 458
 459         list_for_each_entry_safe(a, tmp, actions, list) {
 460                 ops = a->ops;
 461                 ret = __tcf_hash_release(a, bind, true);
 462                 if (ret == ACT_P_DELETED)
 463                         module_put(ops->owner);
 464                 else if (ret < 0)
 465                         return ret;
 466         }
 467         return ret;
 468 }
 469
 470 int
 471 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 472 {
 473         return a->ops->dump(skb, a, bind, ref);
 474 }
 475
 476 int
 477 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 478 {
 479         int err = -EINVAL;
 480         unsigned char *b = skb_tail_pointer(skb);
 481         struct nlattr *nest;
 482
 483         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 484                 goto nla_put_failure;
 485         if (tcf_action_copy_stats(skb, a, 0))
 486                 goto nla_put_failure;
 487         nest = nla_nest_start(skb, TCA_OPTIONS);
 488         if (nest == NULL)
 489                 goto nla_put_failure;
 490         err = tcf_action_dump_old(skb, a, bind, ref);
 491         if (err > 0) {
 492                 nla_nest_end(skb, nest);
 493                 return err;
 494         }
 495
 496 nla_put_failure:
 497         nlmsg_trim(skb, b);
 498         return -1;
 499 }
 500 EXPORT_SYMBOL(tcf_action_dump_1);
 501
 502 int tcf_action_dump(struct sk_buff *skb, struct list_head *actions,
 503                     int bind, int ref)
 504 {
 505         struct tc_action *a;
 506         int err = -EINVAL;
 507         struct nlattr *nest;
 508
 509         list_for_each_entry(a, actions, list) {
 510                 nest = nla_nest_start(skb, a->order);
 511                 if (nest == NULL)
 512                         goto nla_put_failure;
 513                 err = tcf_action_dump_1(skb, a, bind, ref);
 514                 if (err < 0)
 515                         goto errout;
 516                 nla_nest_end(skb, nest);
 517         }
 518
 519         return 0;
 520
 521 nla_put_failure:
 522         err = -EINVAL;
 523 errout:
 524         nla_nest_cancel(skb, nest);
 525         return err;
 526 }
 527
 528 struct tc_action *tcf_action_init_1(struct net *net, struct nlattr *nla,
 529                                     struct nlattr *est, char *name, int ovr,
 530                                     int bind)
 531 {
 532         struct tc_action *a;
 533         struct tc_action_ops *a_o;
 534         char act_name[IFNAMSIZ];
 535         struct nlattr *tb[TCA_ACT_MAX + 1];
 536         struct nlattr *kind;
 537         int err;
 538
 539         if (name == NULL) {
 540                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 541                 if (err < 0)
 542                         goto err_out;
 543                 err = -EINVAL;
 544                 kind = tb[TCA_ACT_KIND];
 545                 if (kind == NULL)
 546                         goto err_out;
 547                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
 548                         goto err_out;
 549         } else {
 550                 err = -EINVAL;
 551                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
 552                         goto err_out;
 553         }
 554
 555         a_o = tc_lookup_action_n(act_name);
 556         if (a_o == NULL) {
 557 #ifdef CONFIG_MODULES
 558                 rtnl_unlock();
 559                 request_module("act_%s", act_name);
 560                 rtnl_lock();
 561
 562                 a_o = tc_lookup_action_n(act_name);
 563
 564                 /* We dropped the RTNL semaphore in order to
 565                  * perform the module load.  So, even if we
 566                  * succeeded in loading the module we have to
 567                  * tell the caller to replay the request.  We
 568                  * indicate this using -EAGAIN.
 569                  */
 570                 if (a_o != NULL) {
 571                         err = -EAGAIN;
 572                         goto err_mod;
 573                 }
 574 #endif
 575                 err = -ENOENT;
 576                 goto err_out;
 577         }
 578
 579         /* backward compatibility for policer */
 580         if (name == NULL)
 581                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind);
 582         else
 583                 err = a_o->init(net, nla, est, &a, ovr, bind);
 584         if (err < 0)
 585                 goto err_mod;
 586
 587         /* module count goes up only when brand new policy is created
 588          * if it exists and is only bound to in a_o->init() then
 589          * ACT_P_CREATED is not returned (a zero is).
 590          */
 591         if (err != ACT_P_CREATED)
 592                 module_put(a_o->owner);
 593
 594         return a;
 595
 596 err_mod:
 597         module_put(a_o->owner);
 598 err_out:
 599         return ERR_PTR(err);
 600 }
 601
 602 static void cleanup_a(struct list_head *actions, int ovr)
 603 {
 604         struct tc_action *a;
 605
 606         if (!ovr)
 607                 return;
 608
 609         list_for_each_entry(a, actions, list)
 610                 a->tcfa_refcnt--;
 611 }
 612
 613 int tcf_action_init(struct net *net, struct nlattr *nla, struct nlattr *est,
 614                     char *name, int ovr, int bind, struct list_head *actions)
 615 {
 616         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 617         struct tc_action *act;
 618         int err;
 619         int i;
 620
 621         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
 622         if (err < 0)
 623                 return err;
 624
 625         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 626                 act = tcf_action_init_1(net, tb[i], est, name, ovr, bind);
 627                 if (IS_ERR(act)) {
 628                         err = PTR_ERR(act);
 629                         goto err;
 630                 }
 631                 act->order = i;
 632                 if (ovr)
 633                         act->tcfa_refcnt++;
 634                 list_add_tail(&act->list, actions);
 635         }
 636
 637         /* Remove the temp refcnt which was necessary to protect against
 638          * destroying an existing action which was being replaced
 639          */
 640         cleanup_a(actions, ovr);
 641         return 0;
 642
 643 err:
 644         tcf_action_destroy(actions, bind);
 645         return err;
 646 }
 647
 648 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
 649                           int compat_mode)
 650 {
 651         int err = 0;
 652         struct gnet_dump d;
 653
 654         if (p == NULL)
 655                 goto errout;
 656
 657         /* compat_mode being true specifies a call that is supposed
 658          * to add additional backward compatibility statistic TLVs.
 659          */
 660         if (compat_mode) {
 661                 if (p->type == TCA_OLD_COMPAT)
 662                         err = gnet_stats_start_copy_compat(skb, 0,
 663                                                            TCA_STATS,
 664                                                            TCA_XSTATS,
 665                                                            &p->tcfa_lock, &d,
 666                                                            TCA_PAD);
 667                 else
 668                         return 0;
 669         } else
 670                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
 671                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
 672
 673         if (err < 0)
 674                 goto errout;
 675
 676         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
 677             gnet_stats_copy_rate_est(&d, &p->tcfa_bstats,
 678                                      &p->tcfa_rate_est) < 0 ||
 679             gnet_stats_copy_queue(&d, p->cpu_qstats,
 680                                   &p->tcfa_qstats,
 681                                   p->tcfa_qstats.qlen) < 0)
 682                 goto errout;
 683
 684         if (gnet_stats_finish_copy(&d) < 0)
 685                 goto errout;
 686
 687         return 0;
 688
 689 errout:
 690         return -1;
 691 }
 692
 693 static int tca_get_fill(struct sk_buff *skb, struct list_head *actions,
 694                         u32 portid, u32 seq, u16 flags, int event, int bind,
 695                         int ref)
 696 {
 697         struct tcamsg *t;
 698         struct nlmsghdr *nlh;
 699         unsigned char *b = skb_tail_pointer(skb);
 700         struct nlattr *nest;
 701
 702         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
 703         if (!nlh)
 704                 goto out_nlmsg_trim;
 705         t = nlmsg_data(nlh);
 706         t->tca_family = AF_UNSPEC;
 707         t->tca__pad1 = 0;
 708         t->tca__pad2 = 0;
 709
 710         nest = nla_nest_start(skb, TCA_ACT_TAB);
 711         if (nest == NULL)
 712                 goto out_nlmsg_trim;
 713
 714         if (tcf_action_dump(skb, actions, bind, ref) < 0)
 715                 goto out_nlmsg_trim;
 716
 717         nla_nest_end(skb, nest);
 718
 719         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 720         return skb->len;
 721
 722 out_nlmsg_trim:
 723         nlmsg_trim(skb, b);
 724         return -1;
 725 }
 726
 727 static int
 728 act_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
 729                struct list_head *actions, int event)
 730 {
 731         struct sk_buff *skb;
 732
 733         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 734         if (!skb)
 735                 return -ENOBUFS;
 736         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
 737                          0, 0) <= 0) {
 738                 kfree_skb(skb);
 739                 return -EINVAL;
 740         }
 741
 742         return rtnl_unicast(skb, net, portid);
 743 }
 744
 745 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
 746                                           struct nlmsghdr *n, u32 portid)
 747 {
 748         struct nlattr *tb[TCA_ACT_MAX + 1];
 749         const struct tc_action_ops *ops;
 750         struct tc_action *a;
 751         int index;
 752         int err;
 753
 754         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 755         if (err < 0)
 756                 goto err_out;
 757
 758         err = -EINVAL;
 759         if (tb[TCA_ACT_INDEX] == NULL ||
 760             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
 761                 goto err_out;
 762         index = nla_get_u32(tb[TCA_ACT_INDEX]);
 763
 764         err = -EINVAL;
 765         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
 766         if (!ops) /* could happen in batch of actions */
 767                 goto err_out;
 768         err = -ENOENT;
 769         if (ops->lookup(net, &a, index) == 0)
 770                 goto err_mod;
 771
 772         module_put(ops->owner);
 773         return a;
 774
 775 err_mod:
 776         module_put(ops->owner);
 777 err_out:
 778         return ERR_PTR(err);
 779 }
 780
 781 static int tca_action_flush(struct net *net, struct nlattr *nla,
 782                             struct nlmsghdr *n, u32 portid)
 783 {
 784         struct sk_buff *skb;
 785         unsigned char *b;
 786         struct nlmsghdr *nlh;
 787         struct tcamsg *t;
 788         struct netlink_callback dcb;
 789         struct nlattr *nest;
 790         struct nlattr *tb[TCA_ACT_MAX + 1];
 791         const struct tc_action_ops *ops;
 792         struct nlattr *kind;
 793         int err = -ENOMEM;
 794
 795         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 796         if (!skb) {
 797                 pr_debug("tca_action_flush: failed skb alloc\n");
 798                 return err;
 799         }
 800
 801         b = skb_tail_pointer(skb);
 802
 803         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 804         if (err < 0)
 805                 goto err_out;
 806
 807         err = -EINVAL;
 808         kind = tb[TCA_ACT_KIND];
 809         ops = tc_lookup_action(kind);
 810         if (!ops) /*some idjot trying to flush unknown action */
 811                 goto err_out;
 812
 813         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
 814                         sizeof(*t), 0);
 815         if (!nlh)
 816                 goto out_module_put;
 817         t = nlmsg_data(nlh);
 818         t->tca_family = AF_UNSPEC;
 819         t->tca__pad1 = 0;
 820         t->tca__pad2 = 0;
 821
 822         nest = nla_nest_start(skb, TCA_ACT_TAB);
 823         if (nest == NULL)
 824                 goto out_module_put;
 825
 826         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops);
 827         if (err <= 0)
 828                 goto out_module_put;
 829
 830         nla_nest_end(skb, nest);
 831
 832         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 833         nlh->nlmsg_flags |= NLM_F_ROOT;
 834         module_put(ops->owner);
 835         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 836                              n->nlmsg_flags & NLM_F_ECHO);
 837         if (err > 0)
 838                 return 0;
 839
 840         return err;
 841
 842 out_module_put:
 843         module_put(ops->owner);
 844 err_out:
 845         kfree_skb(skb);
 846         return err;
 847 }
 848
 849 static int
 850 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
 851                u32 portid)
 852 {
 853         int ret;
 854         struct sk_buff *skb;
 855
 856         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 857         if (!skb)
 858                 return -ENOBUFS;
 859
 860         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
 861                          0, 1) <= 0) {
 862                 kfree_skb(skb);
 863                 return -EINVAL;
 864         }
 865
 866         /* now do the delete */
 867         ret = tcf_action_destroy(actions, 0);
 868         if (ret < 0) {
 869                 kfree_skb(skb);
 870                 return ret;
 871         }
 872
 873         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 874                              n->nlmsg_flags & NLM_F_ECHO);
 875         if (ret > 0)
 876                 return 0;
 877         return ret;
 878 }
 879
 880 static int
 881 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
 882               u32 portid, int event)
 883 {
 884         int i, ret;
 885         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 886         struct tc_action *act;
 887         LIST_HEAD(actions);
 888
 889         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
 890         if (ret < 0)
 891                 return ret;
 892
 893         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
 894                 if (tb[1] != NULL)
 895                         return tca_action_flush(net, tb[1], n, portid);
 896                 else
 897                         return -EINVAL;
 898         }
 899
 900         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 901                 act = tcf_action_get_1(net, tb[i], n, portid);
 902                 if (IS_ERR(act)) {
 903                         ret = PTR_ERR(act);
 904                         goto err;
 905                 }
 906                 act->order = i;
 907                 list_add_tail(&act->list, &actions);
 908         }
 909
 910         if (event == RTM_GETACTION)
 911                 ret = act_get_notify(net, portid, n, &actions, event);
 912         else { /* delete */
 913                 ret = tcf_del_notify(net, n, &actions, portid);
 914                 if (ret)
 915                         goto err;
 916                 return ret;
 917         }
 918 err:
 919         if (event != RTM_GETACTION)
 920                 tcf_action_destroy(&actions, 0);
 921         return ret;
 922 }
 923
 924 static int
 925 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
 926                u32 portid)
 927 {
 928         struct sk_buff *skb;
 929         int err = 0;
 930
 931         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 932         if (!skb)
 933                 return -ENOBUFS;
 934
 935         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
 936                          RTM_NEWACTION, 0, 0) <= 0) {
 937                 kfree_skb(skb);
 938                 return -EINVAL;
 939         }
 940
 941         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 942                              n->nlmsg_flags & NLM_F_ECHO);
 943         if (err > 0)
 944                 err = 0;
 945         return err;
 946 }
 947
 948 static int tcf_action_add(struct net *net, struct nlattr *nla,
 949                           struct nlmsghdr *n, u32 portid, int ovr)
 950 {
 951         int loop, ret;
 952         LIST_HEAD(actions);
 953
 954         for (loop = 0; loop < 10; loop++) {
 955                 ret = tcf_action_init(net, nla, NULL, NULL, ovr, 0, &actions);
 956                 if (ret != -EAGAIN)
 957                         break;
 958         }
 959
 960         if (ret)
 961                 return ret;
 962
 963         return tcf_add_notify(net, n, &actions, portid);
 964 }
 965
 966 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n)
 967 {
 968         struct net *net = sock_net(skb->sk);
 969         struct nlattr *tca[TCA_ACT_MAX + 1];
 970         u32 portid = skb ? NETLINK_CB(skb).portid : 0;
 971         int ret = 0, ovr = 0;
 972
 973         if ((n->nlmsg_type != RTM_GETACTION) &&
 974             !netlink_capable(skb, CAP_NET_ADMIN))
 975                 return -EPERM;
 976
 977         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ACT_MAX, NULL);
 978         if (ret < 0)
 979                 return ret;
 980
 981         if (tca[TCA_ACT_TAB] == NULL) {
 982                 pr_notice("tc_ctl_action: received NO action attribs\n");
 983                 return -EINVAL;
 984         }
 985
 986         /* n->nlmsg_flags & NLM_F_CREATE */
 987         switch (n->nlmsg_type) {
 988         case RTM_NEWACTION:
 989                 /* we are going to assume all other flags
 990                  * imply create only if it doesn't exist
 991                  * Note that CREATE | EXCL implies that
 992                  * but since we want avoid ambiguity (eg when flags
 993                  * is zero) then just set this
 994                  */
 995                 if (n->nlmsg_flags & NLM_F_REPLACE)
 996                         ovr = 1;
 997                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr);
 998                 break;
 999         case RTM_DELACTION:
1000                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1001                                     portid, RTM_DELACTION);
1002                 break;
1003         case RTM_GETACTION:
1004                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1005                                     portid, RTM_GETACTION);
1006                 break;
1007         default:
1008                 BUG();
1009         }
1010
1011         return ret;
1012 }
1013
1014 static struct nlattr *find_dump_kind(const struct nlmsghdr *n)
1015 {
1016         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1017         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1018         struct nlattr *nla[TCAA_MAX + 1];
1019         struct nlattr *kind;
1020
1021         if (nlmsg_parse(n, sizeof(struct tcamsg), nla, TCAA_MAX, NULL) < 0)
1022                 return NULL;
1023         tb1 = nla[TCA_ACT_TAB];
1024         if (tb1 == NULL)
1025                 return NULL;
1026
1027         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1028                       NLMSG_ALIGN(nla_len(tb1)), NULL) < 0)
1029                 return NULL;
1030
1031         if (tb[1] == NULL)
1032                 return NULL;
1033         if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL) < 0)
1034                 return NULL;
1035         kind = tb2[TCA_ACT_KIND];
1036
1037         return kind;
1038 }
1039
1040 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1041 {
1042         struct net *net = sock_net(skb->sk);
1043         struct nlmsghdr *nlh;
1044         unsigned char *b = skb_tail_pointer(skb);
1045         struct nlattr *nest;
1046         struct tc_action_ops *a_o;
1047         int ret = 0;
1048         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1049         struct nlattr *kind = find_dump_kind(cb->nlh);
1050
1051         if (kind == NULL) {
1052                 pr_info("tc_dump_action: action bad kind\n");
1053                 return 0;
1054         }
1055
1056         a_o = tc_lookup_action(kind);
1057         if (a_o == NULL)
1058                 return 0;
1059
1060         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1061                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1062         if (!nlh)
1063                 goto out_module_put;
1064         t = nlmsg_data(nlh);
1065         t->tca_family = AF_UNSPEC;
1066         t->tca__pad1 = 0;
1067         t->tca__pad2 = 0;
1068
1069         nest = nla_nest_start(skb, TCA_ACT_TAB);
1070         if (nest == NULL)
1071                 goto out_module_put;
1072
1073         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o);
1074         if (ret < 0)
1075                 goto out_module_put;
1076
1077         if (ret > 0) {
1078                 nla_nest_end(skb, nest);
1079                 ret = skb->len;
1080         } else
1081                 nlmsg_trim(skb, b);
1082
1083         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1084         if (NETLINK_CB(cb->skb).portid && ret)
1085                 nlh->nlmsg_flags |= NLM_F_MULTI;
1086         module_put(a_o->owner);
1087         return skb->len;
1088
1089 out_module_put:
1090         module_put(a_o->owner);
1091         nlmsg_trim(skb, b);
1092         return skb->len;
1093 }
1094
1095 static int __init tc_action_init(void)
1096 {
1097         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, NULL);
1098         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, NULL);
1099         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1100                       NULL);
1101
1102         return 0;
1103 }
1104
1105 subsys_initcall(tc_action_init);