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