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 tcf_common *p = container_of(head, struct tcf_common, tcfc_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 *a)
  40 {
  41         struct tcf_common *p = (struct tcf_common *)a;
  42
  43         spin_lock_bh(&hinfo->lock);
  44         hlist_del(&p->tcfc_head);
  45         spin_unlock_bh(&hinfo->lock);
  46         gen_kill_estimator(&p->tcfc_bstats,
  47                            &p->tcfc_rate_est);
  48         /*
  49          * gen_estimator est_timer() might access p->tcfc_lock
  50          * or bstats, wait a RCU grace period before freeing p
  51          */
  52         call_rcu(&p->tcfc_rcu, free_tcf);
  53 }
  54
  55 int __tcf_hash_release(struct tc_action *a, bool bind, bool strict)
  56 {
  57         struct tcf_common *p = (struct tcf_common *)a;
  58         int ret = 0;
  59
  60         if (p) {
  61                 if (bind)
  62                         p->tcfc_bindcnt--;
  63                 else if (strict && p->tcfc_bindcnt > 0)
  64                         return -EPERM;
  65
  66                 p->tcfc_refcnt--;
  67                 if (p->tcfc_bindcnt <= 0 && p->tcfc_refcnt <= 0) {
  68                         if (a->ops->cleanup)
  69                                 a->ops->cleanup(a, bind);
  70                         list_del(&a->list);
  71                         tcf_hash_destroy(a->hinfo, a);
  72                         ret = ACT_P_DELETED;
  73                 }
  74         }
  75
  76         return ret;
  77 }
  78 EXPORT_SYMBOL(__tcf_hash_release);
  79
  80 static int tcf_dump_walker(struct tcf_hashinfo *hinfo, struct sk_buff *skb,
  81                            struct netlink_callback *cb)
  82 {
  83         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
  84         struct nlattr *nest;
  85
  86         spin_lock_bh(&hinfo->lock);
  87
  88         s_i = cb->args[0];
  89
  90         for (i = 0; i < (hinfo->hmask + 1); i++) {
  91                 struct hlist_head *head;
  92                 struct tcf_common *p;
  93
  94                 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
  95
  96                 hlist_for_each_entry_rcu(p, head, tcfc_head) {
  97                         index++;
  98                         if (index < s_i)
  99                                 continue;
 100
 101                         nest = nla_nest_start(skb, n_i);
 102                         if (nest == NULL)
 103                                 goto nla_put_failure;
 104                         err = tcf_action_dump_1(skb, (struct tc_action *)p, 0, 0);
 105                         if (err < 0) {
 106                                 index--;
 107                                 nlmsg_trim(skb, nest);
 108                                 goto done;
 109                         }
 110                         nla_nest_end(skb, nest);
 111                         n_i++;
 112                         if (n_i >= TCA_ACT_MAX_PRIO)
 113                                 goto done;
 114                 }
 115         }
 116 done:
 117         spin_unlock_bh(&hinfo->lock);
 118         if (n_i)
 119                 cb->args[0] += n_i;
 120         return n_i;
 121
 122 nla_put_failure:
 123         nla_nest_cancel(skb, nest);
 124         goto done;
 125 }
 126
 127 static int tcf_del_walker(struct tcf_hashinfo *hinfo, struct sk_buff *skb,
 128                           const struct tc_action_ops *ops)
 129 {
 130         struct nlattr *nest;
 131         int i = 0, n_i = 0;
 132         int ret = -EINVAL;
 133
 134         nest = nla_nest_start(skb, 0);
 135         if (nest == NULL)
 136                 goto nla_put_failure;
 137         if (nla_put_string(skb, TCA_KIND, ops->kind))
 138                 goto nla_put_failure;
 139         for (i = 0; i < (hinfo->hmask + 1); i++) {
 140                 struct hlist_head *head;
 141                 struct hlist_node *n;
 142                 struct tcf_common *p;
 143
 144                 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
 145                 hlist_for_each_entry_safe(p, n, head, tcfc_head) {
 146                         ret = __tcf_hash_release((struct tc_action *)p, false, true);
 147                         if (ret == ACT_P_DELETED) {
 148                                 module_put(p->tcfc_act.ops->owner);
 149                                 n_i++;
 150                         } else if (ret < 0)
 151                                 goto nla_put_failure;
 152                 }
 153         }
 154         if (nla_put_u32(skb, TCA_FCNT, n_i))
 155                 goto nla_put_failure;
 156         nla_nest_end(skb, nest);
 157
 158         return n_i;
 159 nla_put_failure:
 160         nla_nest_cancel(skb, nest);
 161         return ret;
 162 }
 163
 164 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 165                        struct netlink_callback *cb, int type,
 166                        const struct tc_action_ops *ops)
 167 {
 168         struct tcf_hashinfo *hinfo = tn->hinfo;
 169
 170         if (type == RTM_DELACTION) {
 171                 return tcf_del_walker(hinfo, skb, ops);
 172         } else if (type == RTM_GETACTION) {
 173                 return tcf_dump_walker(hinfo, skb, cb);
 174         } else {
 175                 WARN(1, "tcf_generic_walker: unknown action %d\n", type);
 176                 return -EINVAL;
 177         }
 178 }
 179 EXPORT_SYMBOL(tcf_generic_walker);
 180
 181 static struct tcf_common *tcf_hash_lookup(u32 index, struct tcf_hashinfo *hinfo)
 182 {
 183         struct tcf_common *p = NULL;
 184         struct hlist_head *head;
 185
 186         spin_lock_bh(&hinfo->lock);
 187         head = &hinfo->htab[tcf_hash(index, hinfo->hmask)];
 188         hlist_for_each_entry_rcu(p, head, tcfc_head)
 189                 if (p->tcfc_index == index)
 190                         break;
 191         spin_unlock_bh(&hinfo->lock);
 192
 193         return p;
 194 }
 195
 196 u32 tcf_hash_new_index(struct tc_action_net *tn)
 197 {
 198         struct tcf_hashinfo *hinfo = tn->hinfo;
 199         u32 val = hinfo->index;
 200
 201         do {
 202                 if (++val == 0)
 203                         val = 1;
 204         } while (tcf_hash_lookup(val, hinfo));
 205
 206         hinfo->index = val;
 207         return val;
 208 }
 209 EXPORT_SYMBOL(tcf_hash_new_index);
 210
 211 int tcf_hash_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 212 {
 213         struct tcf_hashinfo *hinfo = tn->hinfo;
 214         struct tcf_common *p = tcf_hash_lookup(index, hinfo);
 215
 216         if (p) {
 217                 *a = &p->tcfc_act;
 218                 return 1;
 219         }
 220         return 0;
 221 }
 222 EXPORT_SYMBOL(tcf_hash_search);
 223
 224 bool tcf_hash_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
 225                     int bind)
 226 {
 227         struct tcf_hashinfo *hinfo = tn->hinfo;
 228         struct tcf_common *p = NULL;
 229         if (index && (p = tcf_hash_lookup(index, hinfo)) != NULL) {
 230                 if (bind)
 231                         p->tcfc_bindcnt++;
 232                 p->tcfc_refcnt++;
 233                 *a = &p->tcfc_act;
 234                 return true;
 235         }
 236         return false;
 237 }
 238 EXPORT_SYMBOL(tcf_hash_check);
 239
 240 void tcf_hash_cleanup(struct tc_action *a, struct nlattr *est)
 241 {
 242         struct tcf_common *pc = (struct tcf_common *)a;
 243         if (est)
 244                 gen_kill_estimator(&pc->tcfc_bstats,
 245                                    &pc->tcfc_rate_est);
 246         call_rcu(&pc->tcfc_rcu, free_tcf);
 247 }
 248 EXPORT_SYMBOL(tcf_hash_cleanup);
 249
 250 int tcf_hash_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 251                     struct tc_action **a, const struct tc_action_ops *ops,
 252                     int bind, bool cpustats)
 253 {
 254         struct tcf_common *p = kzalloc(ops->size, GFP_KERNEL);
 255         struct tcf_hashinfo *hinfo = tn->hinfo;
 256         int err = -ENOMEM;
 257
 258         if (unlikely(!p))
 259                 return -ENOMEM;
 260         p->tcfc_refcnt = 1;
 261         if (bind)
 262                 p->tcfc_bindcnt = 1;
 263
 264         if (cpustats) {
 265                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 266                 if (!p->cpu_bstats) {
 267 err1:
 268                         kfree(p);
 269                         return err;
 270                 }
 271                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 272                 if (!p->cpu_qstats) {
 273 err2:
 274                         free_percpu(p->cpu_bstats);
 275                         goto err1;
 276                 }
 277         }
 278         spin_lock_init(&p->tcfc_lock);
 279         INIT_HLIST_NODE(&p->tcfc_head);
 280         p->tcfc_index = index ? index : tcf_hash_new_index(tn);
 281         p->tcfc_tm.install = jiffies;
 282         p->tcfc_tm.lastuse = jiffies;
 283         p->tcfc_tm.firstuse = 0;
 284         if (est) {
 285                 err = gen_new_estimator(&p->tcfc_bstats, p->cpu_bstats,
 286                                         &p->tcfc_rate_est,
 287                                         &p->tcfc_lock, NULL, est);
 288                 if (err) {
 289                         free_percpu(p->cpu_qstats);
 290                         goto err2;
 291                 }
 292         }
 293
 294         p->tcfc_act.hinfo = hinfo;
 295         p->tcfc_act.ops = ops;
 296         INIT_LIST_HEAD(&p->tcfc_act.list);
 297         *a = &p->tcfc_act;
 298         return 0;
 299 }
 300 EXPORT_SYMBOL(tcf_hash_create);
 301
 302 void tcf_hash_insert(struct tc_action_net *tn, struct tc_action *a)
 303 {
 304         struct tcf_common *p = (struct tcf_common *)a;
 305         struct tcf_hashinfo *hinfo = tn->hinfo;
 306         unsigned int h = tcf_hash(p->tcfc_index, hinfo->hmask);
 307
 308         spin_lock_bh(&hinfo->lock);
 309         hlist_add_head(&p->tcfc_head, &hinfo->htab[h]);
 310         spin_unlock_bh(&hinfo->lock);
 311 }
 312 EXPORT_SYMBOL(tcf_hash_insert);
 313
 314 void tcf_hashinfo_destroy(const struct tc_action_ops *ops,
 315                           struct tcf_hashinfo *hinfo)
 316 {
 317         int i;
 318
 319         for (i = 0; i < hinfo->hmask + 1; i++) {
 320                 struct tcf_common *p;
 321                 struct hlist_node *n;
 322
 323                 hlist_for_each_entry_safe(p, n, &hinfo->htab[i], tcfc_head) {
 324                         int ret;
 325
 326                         ret = __tcf_hash_release((struct tc_action *)p, false, true);
 327                         if (ret == ACT_P_DELETED)
 328                                 module_put(ops->owner);
 329                         else if (ret < 0)
 330                                 return;
 331                 }
 332         }
 333         kfree(hinfo->htab);
 334 }
 335 EXPORT_SYMBOL(tcf_hashinfo_destroy);
 336
 337 static LIST_HEAD(act_base);
 338 static DEFINE_RWLOCK(act_mod_lock);
 339
 340 int tcf_register_action(struct tc_action_ops *act,
 341                         struct pernet_operations *ops)
 342 {
 343         struct tc_action_ops *a;
 344         int ret;
 345
 346         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 347                 return -EINVAL;
 348
 349         write_lock(&act_mod_lock);
 350         list_for_each_entry(a, &act_base, head) {
 351                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
 352                         write_unlock(&act_mod_lock);
 353                         return -EEXIST;
 354                 }
 355         }
 356         list_add_tail(&act->head, &act_base);
 357         write_unlock(&act_mod_lock);
 358
 359         ret = register_pernet_subsys(ops);
 360         if (ret) {
 361                 tcf_unregister_action(act, ops);
 362                 return ret;
 363         }
 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         unregister_pernet_subsys(ops);
 376
 377         write_lock(&act_mod_lock);
 378         list_for_each_entry(a, &act_base, head) {
 379                 if (a == act) {
 380                         list_del(&act->head);
 381                         err = 0;
 382                         break;
 383                 }
 384         }
 385         write_unlock(&act_mod_lock);
 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, const struct list_head *actions,
 429                     struct tcf_result *res)
 430 {
 431         const struct tc_action *a;
 432         int ret = -1;
 433
 434         if (skb->tc_verd & TC_NCLS) {
 435                 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
 436                 ret = TC_ACT_OK;
 437                 goto exec_done;
 438         }
 439         list_for_each_entry(a, actions, list) {
 440 repeat:
 441                 ret = a->ops->act(skb, a, res);
 442                 if (ret == TC_ACT_REPEAT)
 443                         goto repeat;    /* we need a ttl - JHS */
 444                 if (ret != TC_ACT_PIPE)
 445                         goto exec_done;
 446         }
 447 exec_done:
 448         return ret;
 449 }
 450 EXPORT_SYMBOL(tcf_action_exec);
 451
 452 int tcf_action_destroy(struct list_head *actions, int bind)
 453 {
 454         struct tc_action *a, *tmp;
 455         int ret = 0;
 456
 457         list_for_each_entry_safe(a, tmp, actions, list) {
 458                 ret = __tcf_hash_release(a, bind, true);
 459                 if (ret == ACT_P_DELETED)
 460                         module_put(a->ops->owner);
 461                 else if (ret < 0)
 462                         return ret;
 463         }
 464         return ret;
 465 }
 466
 467 int
 468 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 469 {
 470         return a->ops->dump(skb, a, bind, ref);
 471 }
 472
 473 int
 474 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 475 {
 476         int err = -EINVAL;
 477         unsigned char *b = skb_tail_pointer(skb);
 478         struct nlattr *nest;
 479
 480         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 481                 goto nla_put_failure;
 482         if (tcf_action_copy_stats(skb, a, 0))
 483                 goto nla_put_failure;
 484         nest = nla_nest_start(skb, TCA_OPTIONS);
 485         if (nest == NULL)
 486                 goto nla_put_failure;
 487         err = tcf_action_dump_old(skb, a, bind, ref);
 488         if (err > 0) {
 489                 nla_nest_end(skb, nest);
 490                 return err;
 491         }
 492
 493 nla_put_failure:
 494         nlmsg_trim(skb, b);
 495         return -1;
 496 }
 497 EXPORT_SYMBOL(tcf_action_dump_1);
 498
 499 int tcf_action_dump(struct sk_buff *skb, struct list_head *actions,
 500                     int bind, int ref)
 501 {
 502         struct tc_action *a;
 503         int err = -EINVAL;
 504         struct nlattr *nest;
 505
 506         list_for_each_entry(a, actions, list) {
 507                 nest = nla_nest_start(skb, a->order);
 508                 if (nest == NULL)
 509                         goto nla_put_failure;
 510                 err = tcf_action_dump_1(skb, a, bind, ref);
 511                 if (err < 0)
 512                         goto errout;
 513                 nla_nest_end(skb, nest);
 514         }
 515
 516         return 0;
 517
 518 nla_put_failure:
 519         err = -EINVAL;
 520 errout:
 521         nla_nest_cancel(skb, nest);
 522         return err;
 523 }
 524
 525 struct tc_action *tcf_action_init_1(struct net *net, struct nlattr *nla,
 526                                     struct nlattr *est, char *name, int ovr,
 527                                     int bind)
 528 {
 529         struct tc_action *a;
 530         struct tc_action_ops *a_o;
 531         char act_name[IFNAMSIZ];
 532         struct nlattr *tb[TCA_ACT_MAX + 1];
 533         struct nlattr *kind;
 534         int err;
 535
 536         if (name == NULL) {
 537                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 538                 if (err < 0)
 539                         goto err_out;
 540                 err = -EINVAL;
 541                 kind = tb[TCA_ACT_KIND];
 542                 if (kind == NULL)
 543                         goto err_out;
 544                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
 545                         goto err_out;
 546         } else {
 547                 err = -EINVAL;
 548                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
 549                         goto err_out;
 550         }
 551
 552         a_o = tc_lookup_action_n(act_name);
 553         if (a_o == NULL) {
 554 #ifdef CONFIG_MODULES
 555                 rtnl_unlock();
 556                 request_module("act_%s", act_name);
 557                 rtnl_lock();
 558
 559                 a_o = tc_lookup_action_n(act_name);
 560
 561                 /* We dropped the RTNL semaphore in order to
 562                  * perform the module load.  So, even if we
 563                  * succeeded in loading the module we have to
 564                  * tell the caller to replay the request.  We
 565                  * indicate this using -EAGAIN.
 566                  */
 567                 if (a_o != NULL) {
 568                         err = -EAGAIN;
 569                         goto err_mod;
 570                 }
 571 #endif
 572                 err = -ENOENT;
 573                 goto err_out;
 574         }
 575
 576         /* backward compatibility for policer */
 577         if (name == NULL)
 578                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind);
 579         else
 580                 err = a_o->init(net, nla, est, &a, ovr, bind);
 581         if (err < 0)
 582                 goto err_mod;
 583
 584         /* module count goes up only when brand new policy is created
 585          * if it exists and is only bound to in a_o->init() then
 586          * ACT_P_CREATED is not returned (a zero is).
 587          */
 588         if (err != ACT_P_CREATED)
 589                 module_put(a_o->owner);
 590
 591         return a;
 592
 593 err_mod:
 594         module_put(a_o->owner);
 595 err_out:
 596         return ERR_PTR(err);
 597 }
 598
 599 int tcf_action_init(struct net *net, struct nlattr *nla,
 600                                   struct nlattr *est, char *name, int ovr,
 601                                   int bind, struct list_head *actions)
 602 {
 603         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 604         struct tc_action *act;
 605         int err;
 606         int i;
 607
 608         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
 609         if (err < 0)
 610                 return err;
 611
 612         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 613                 act = tcf_action_init_1(net, tb[i], est, name, ovr, bind);
 614                 if (IS_ERR(act)) {
 615                         err = PTR_ERR(act);
 616                         goto err;
 617                 }
 618                 act->order = i;
 619                 list_add_tail(&act->list, actions);
 620         }
 621         return 0;
 622
 623 err:
 624         tcf_action_destroy(actions, bind);
 625         return err;
 626 }
 627
 628 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *a,
 629                           int compat_mode)
 630 {
 631         int err = 0;
 632         struct gnet_dump d;
 633         struct tcf_common *p = (struct tcf_common *)a;
 634
 635         if (p == NULL)
 636                 goto errout;
 637
 638         /* compat_mode being true specifies a call that is supposed
 639          * to add additional backward compatibility statistic TLVs.
 640          */
 641         if (compat_mode) {
 642                 if (a->type == TCA_OLD_COMPAT)
 643                         err = gnet_stats_start_copy_compat(skb, 0,
 644                                                            TCA_STATS,
 645                                                            TCA_XSTATS,
 646                                                            &p->tcfc_lock, &d,
 647                                                            TCA_PAD);
 648                 else
 649                         return 0;
 650         } else
 651                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
 652                                             &p->tcfc_lock, &d, TCA_ACT_PAD);
 653
 654         if (err < 0)
 655                 goto errout;
 656
 657         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfc_bstats) < 0 ||
 658             gnet_stats_copy_rate_est(&d, &p->tcfc_bstats,
 659                                      &p->tcfc_rate_est) < 0 ||
 660             gnet_stats_copy_queue(&d, p->cpu_qstats,
 661                                   &p->tcfc_qstats,
 662                                   p->tcfc_qstats.qlen) < 0)
 663                 goto errout;
 664
 665         if (gnet_stats_finish_copy(&d) < 0)
 666                 goto errout;
 667
 668         return 0;
 669
 670 errout:
 671         return -1;
 672 }
 673
 674 static int tca_get_fill(struct sk_buff *skb, struct list_head *actions,
 675                         u32 portid, u32 seq, u16 flags, int event, int bind,
 676                         int ref)
 677 {
 678         struct tcamsg *t;
 679         struct nlmsghdr *nlh;
 680         unsigned char *b = skb_tail_pointer(skb);
 681         struct nlattr *nest;
 682
 683         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
 684         if (!nlh)
 685                 goto out_nlmsg_trim;
 686         t = nlmsg_data(nlh);
 687         t->tca_family = AF_UNSPEC;
 688         t->tca__pad1 = 0;
 689         t->tca__pad2 = 0;
 690
 691         nest = nla_nest_start(skb, TCA_ACT_TAB);
 692         if (nest == NULL)
 693                 goto out_nlmsg_trim;
 694
 695         if (tcf_action_dump(skb, actions, bind, ref) < 0)
 696                 goto out_nlmsg_trim;
 697
 698         nla_nest_end(skb, nest);
 699
 700         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 701         return skb->len;
 702
 703 out_nlmsg_trim:
 704         nlmsg_trim(skb, b);
 705         return -1;
 706 }
 707
 708 static int
 709 act_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
 710                struct list_head *actions, int event)
 711 {
 712         struct sk_buff *skb;
 713
 714         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 715         if (!skb)
 716                 return -ENOBUFS;
 717         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
 718                          0, 0) <= 0) {
 719                 kfree_skb(skb);
 720                 return -EINVAL;
 721         }
 722
 723         return rtnl_unicast(skb, net, portid);
 724 }
 725
 726 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
 727                                           struct nlmsghdr *n, u32 portid)
 728 {
 729         struct nlattr *tb[TCA_ACT_MAX + 1];
 730         const struct tc_action_ops *ops;
 731         struct tc_action *a;
 732         int index;
 733         int err;
 734
 735         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 736         if (err < 0)
 737                 goto err_out;
 738
 739         err = -EINVAL;
 740         if (tb[TCA_ACT_INDEX] == NULL ||
 741             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
 742                 goto err_out;
 743         index = nla_get_u32(tb[TCA_ACT_INDEX]);
 744
 745         err = -EINVAL;
 746         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
 747         if (!ops) /* could happen in batch of actions */
 748                 goto err_out;
 749         err = -ENOENT;
 750         if (ops->lookup(net, &a, index) == 0)
 751                 goto err_mod;
 752
 753         module_put(ops->owner);
 754         return a;
 755
 756 err_mod:
 757         module_put(ops->owner);
 758 err_out:
 759         return ERR_PTR(err);
 760 }
 761
 762 static void cleanup_a(struct list_head *actions)
 763 {
 764         struct tc_action *a, *tmp;
 765
 766         list_for_each_entry_safe(a, tmp, actions, list) {
 767                 list_del(&a->list);
 768                 kfree(a);
 769         }
 770 }
 771
 772 static int tca_action_flush(struct net *net, struct nlattr *nla,
 773                             struct nlmsghdr *n, u32 portid)
 774 {
 775         struct sk_buff *skb;
 776         unsigned char *b;
 777         struct nlmsghdr *nlh;
 778         struct tcamsg *t;
 779         struct netlink_callback dcb;
 780         struct nlattr *nest;
 781         struct nlattr *tb[TCA_ACT_MAX + 1];
 782         const struct tc_action_ops *ops;
 783         struct nlattr *kind;
 784         int err = -ENOMEM;
 785
 786         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 787         if (!skb) {
 788                 pr_debug("tca_action_flush: failed skb alloc\n");
 789                 return err;
 790         }
 791
 792         b = skb_tail_pointer(skb);
 793
 794         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 795         if (err < 0)
 796                 goto err_out;
 797
 798         err = -EINVAL;
 799         kind = tb[TCA_ACT_KIND];
 800         ops = tc_lookup_action(kind);
 801         if (!ops) /*some idjot trying to flush unknown action */
 802                 goto err_out;
 803
 804         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
 805                         sizeof(*t), 0);
 806         if (!nlh)
 807                 goto out_module_put;
 808         t = nlmsg_data(nlh);
 809         t->tca_family = AF_UNSPEC;
 810         t->tca__pad1 = 0;
 811         t->tca__pad2 = 0;
 812
 813         nest = nla_nest_start(skb, TCA_ACT_TAB);
 814         if (nest == NULL)
 815                 goto out_module_put;
 816
 817         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops);
 818         if (err < 0)
 819                 goto out_module_put;
 820         if (err == 0)
 821                 goto noflush_out;
 822
 823         nla_nest_end(skb, nest);
 824
 825         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 826         nlh->nlmsg_flags |= NLM_F_ROOT;
 827         module_put(ops->owner);
 828         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 829                              n->nlmsg_flags & NLM_F_ECHO);
 830         if (err > 0)
 831                 return 0;
 832
 833         return err;
 834
 835 out_module_put:
 836         module_put(ops->owner);
 837 err_out:
 838 noflush_out:
 839         kfree_skb(skb);
 840         return err;
 841 }
 842
 843 static int
 844 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
 845                u32 portid)
 846 {
 847         int ret;
 848         struct sk_buff *skb;
 849
 850         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 851         if (!skb)
 852                 return -ENOBUFS;
 853
 854         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
 855                          0, 1) <= 0) {
 856                 kfree_skb(skb);
 857                 return -EINVAL;
 858         }
 859
 860         /* now do the delete */
 861         ret = tcf_action_destroy(actions, 0);
 862         if (ret < 0) {
 863                 kfree_skb(skb);
 864                 return ret;
 865         }
 866
 867         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 868                              n->nlmsg_flags & NLM_F_ECHO);
 869         if (ret > 0)
 870                 return 0;
 871         return ret;
 872 }
 873
 874 static int
 875 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
 876               u32 portid, int event)
 877 {
 878         int i, ret;
 879         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 880         struct tc_action *act;
 881         LIST_HEAD(actions);
 882
 883         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
 884         if (ret < 0)
 885                 return ret;
 886
 887         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
 888                 if (tb[1] != NULL)
 889                         return tca_action_flush(net, tb[1], n, portid);
 890                 else
 891                         return -EINVAL;
 892         }
 893
 894         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 895                 act = tcf_action_get_1(net, tb[i], n, portid);
 896                 if (IS_ERR(act)) {
 897                         ret = PTR_ERR(act);
 898                         goto err;
 899                 }
 900                 act->order = i;
 901                 list_add_tail(&act->list, &actions);
 902         }
 903
 904         if (event == RTM_GETACTION)
 905                 ret = act_get_notify(net, portid, n, &actions, event);
 906         else { /* delete */
 907                 ret = tcf_del_notify(net, n, &actions, portid);
 908                 if (ret)
 909                         goto err;
 910                 return ret;
 911         }
 912 err:
 913         cleanup_a(&actions);
 914         return ret;
 915 }
 916
 917 static int
 918 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
 919                u32 portid)
 920 {
 921         struct sk_buff *skb;
 922         int err = 0;
 923
 924         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 925         if (!skb)
 926                 return -ENOBUFS;
 927
 928         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
 929                          RTM_NEWACTION, 0, 0) <= 0) {
 930                 kfree_skb(skb);
 931                 return -EINVAL;
 932         }
 933
 934         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 935                              n->nlmsg_flags & NLM_F_ECHO);
 936         if (err > 0)
 937                 err = 0;
 938         return err;
 939 }
 940
 941 static int
 942 tcf_action_add(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
 943                u32 portid, int ovr)
 944 {
 945         int ret = 0;
 946         LIST_HEAD(actions);
 947
 948         ret = tcf_action_init(net, nla, NULL, NULL, ovr, 0, &actions);
 949         if (ret)
 950                 goto done;
 951
 952         /* dump then free all the actions after update; inserted policy
 953          * stays intact
 954          */
 955         ret = tcf_add_notify(net, n, &actions, portid);
 956         cleanup_a(&actions);
 957 done:
 958         return ret;
 959 }
 960
 961 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n)
 962 {
 963         struct net *net = sock_net(skb->sk);
 964         struct nlattr *tca[TCA_ACT_MAX + 1];
 965         u32 portid = skb ? NETLINK_CB(skb).portid : 0;
 966         int ret = 0, ovr = 0;
 967
 968         if ((n->nlmsg_type != RTM_GETACTION) &&
 969             !netlink_capable(skb, CAP_NET_ADMIN))
 970                 return -EPERM;
 971
 972         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ACT_MAX, NULL);
 973         if (ret < 0)
 974                 return ret;
 975
 976         if (tca[TCA_ACT_TAB] == NULL) {
 977                 pr_notice("tc_ctl_action: received NO action attribs\n");
 978                 return -EINVAL;
 979         }
 980
 981         /* n->nlmsg_flags & NLM_F_CREATE */
 982         switch (n->nlmsg_type) {
 983         case RTM_NEWACTION:
 984                 /* we are going to assume all other flags
 985                  * imply create only if it doesn't exist
 986                  * Note that CREATE | EXCL implies that
 987                  * but since we want avoid ambiguity (eg when flags
 988                  * is zero) then just set this
 989                  */
 990                 if (n->nlmsg_flags & NLM_F_REPLACE)
 991                         ovr = 1;
 992 replay:
 993                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr);
 994                 if (ret == -EAGAIN)
 995                         goto replay;
 996                 break;
 997         case RTM_DELACTION:
 998                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
 999                                     portid, RTM_DELACTION);
1000                 break;
1001         case RTM_GETACTION:
1002                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1003                                     portid, RTM_GETACTION);
1004                 break;
1005         default:
1006                 BUG();
1007         }
1008
1009         return ret;
1010 }
1011
1012 static struct nlattr *
1013 find_dump_kind(const struct nlmsghdr *n)
1014 {
1015         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1016         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1017         struct nlattr *nla[TCAA_MAX + 1];
1018         struct nlattr *kind;
1019
1020         if (nlmsg_parse(n, sizeof(struct tcamsg), nla, TCAA_MAX, NULL) < 0)
1021                 return NULL;
1022         tb1 = nla[TCA_ACT_TAB];
1023         if (tb1 == NULL)
1024                 return NULL;
1025
1026         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1027                       NLMSG_ALIGN(nla_len(tb1)), NULL) < 0)
1028                 return NULL;
1029
1030         if (tb[1] == NULL)
1031                 return NULL;
1032         if (nla_parse(tb2, TCA_ACT_MAX, nla_data(tb[1]),
1033                       nla_len(tb[1]), NULL) < 0)
1034                 return NULL;
1035         kind = tb2[TCA_ACT_KIND];
1036
1037         return kind;
1038 }
1039
1040 static int
1041 tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1042 {
1043         struct net *net = sock_net(skb->sk);
1044         struct nlmsghdr *nlh;
1045         unsigned char *b = skb_tail_pointer(skb);
1046         struct nlattr *nest;
1047         struct tc_action_ops *a_o;
1048         int ret = 0;
1049         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1050         struct nlattr *kind = find_dump_kind(cb->nlh);
1051
1052         if (kind == NULL) {
1053                 pr_info("tc_dump_action: action bad kind\n");
1054                 return 0;
1055         }
1056
1057         a_o = tc_lookup_action(kind);
1058         if (a_o == NULL)
1059                 return 0;
1060
1061         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1062                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1063         if (!nlh)
1064                 goto out_module_put;
1065         t = nlmsg_data(nlh);
1066         t->tca_family = AF_UNSPEC;
1067         t->tca__pad1 = 0;
1068         t->tca__pad2 = 0;
1069
1070         nest = nla_nest_start(skb, TCA_ACT_TAB);
1071         if (nest == NULL)
1072                 goto out_module_put;
1073
1074         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o);
1075         if (ret < 0)
1076                 goto out_module_put;
1077
1078         if (ret > 0) {
1079                 nla_nest_end(skb, nest);
1080                 ret = skb->len;
1081         } else
1082                 nlmsg_trim(skb, b);
1083
1084         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1085         if (NETLINK_CB(cb->skb).portid && ret)
1086                 nlh->nlmsg_flags |= NLM_F_MULTI;
1087         module_put(a_o->owner);
1088         return skb->len;
1089
1090 out_module_put:
1091         module_put(a_o->owner);
1092         nlmsg_trim(skb, b);
1093         return skb->len;
1094 }
1095
1096 static int __init tc_action_init(void)
1097 {
1098         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, NULL);
1099         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, NULL);
1100         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1101                       NULL);
1102
1103         return 0;
1104 }
1105
1106 subsys_initcall(tc_action_init);