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         /* We have to register pernet ops before making the action ops visible,
 345          * otherwise tcf_action_init_1() could get a partially initialized
 346          * netns.
 347          */
 348         ret = register_pernet_subsys(ops);
 349         if (ret)
 350                 return ret;
 351
 352         write_lock(&act_mod_lock);
 353         list_for_each_entry(a, &act_base, head) {
 354                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
 355                         write_unlock(&act_mod_lock);
 356                         unregister_pernet_subsys(ops);
 357                         return -EEXIST;
 358                 }
 359         }
 360         list_add_tail(&act->head, &act_base);
 361         write_unlock(&act_mod_lock);
 362
 363         return 0;
 364 }
 365 EXPORT_SYMBOL(tcf_register_action);
 366
 367 int tcf_unregister_action(struct tc_action_ops *act,
 368                           struct pernet_operations *ops)
 369 {
 370         struct tc_action_ops *a;
 371         int err = -ENOENT;
 372
 373         write_lock(&act_mod_lock);
 374         list_for_each_entry(a, &act_base, head) {
 375                 if (a == act) {
 376                         list_del(&act->head);
 377                         err = 0;
 378                         break;
 379                 }
 380         }
 381         write_unlock(&act_mod_lock);
 382         if (!err)
 383                 unregister_pernet_subsys(ops);
 384         return err;
 385 }
 386 EXPORT_SYMBOL(tcf_unregister_action);
 387
 388 /* lookup by name */
 389 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 390 {
 391         struct tc_action_ops *a, *res = NULL;
 392
 393         if (kind) {
 394                 read_lock(&act_mod_lock);
 395                 list_for_each_entry(a, &act_base, head) {
 396                         if (strcmp(kind, a->kind) == 0) {
 397                                 if (try_module_get(a->owner))
 398                                         res = a;
 399                                 break;
 400                         }
 401                 }
 402                 read_unlock(&act_mod_lock);
 403         }
 404         return res;
 405 }
 406
 407 /* lookup by nlattr */
 408 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 409 {
 410         struct tc_action_ops *a, *res = NULL;
 411
 412         if (kind) {
 413                 read_lock(&act_mod_lock);
 414                 list_for_each_entry(a, &act_base, head) {
 415                         if (nla_strcmp(kind, a->kind) == 0) {
 416                                 if (try_module_get(a->owner))
 417                                         res = a;
 418                                 break;
 419                         }
 420                 }
 421                 read_unlock(&act_mod_lock);
 422         }
 423         return res;
 424 }
 425
 426 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 427                     int nr_actions, struct tcf_result *res)
 428 {
 429         int ret = -1, i;
 430
 431         if (skb->tc_verd & TC_NCLS) {
 432                 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
 433                 ret = TC_ACT_OK;
 434                 goto exec_done;
 435         }
 436         for (i = 0; i < nr_actions; i++) {
 437                 const struct tc_action *a = actions[i];
 438
 439 repeat:
 440                 ret = a->ops->act(skb, a, res);
 441                 if (ret == TC_ACT_REPEAT)
 442                         goto repeat;    /* we need a ttl - JHS */
 443                 if (ret != TC_ACT_PIPE)
 444                         goto exec_done;
 445         }
 446 exec_done:
 447         return ret;
 448 }
 449 EXPORT_SYMBOL(tcf_action_exec);
 450
 451 int tcf_action_destroy(struct list_head *actions, int bind)
 452 {
 453         struct tc_action *a, *tmp;
 454         int ret = 0;
 455
 456         list_for_each_entry_safe(a, tmp, actions, list) {
 457                 ret = __tcf_hash_release(a, bind, true);
 458                 if (ret == ACT_P_DELETED)
 459                         module_put(a->ops->owner);
 460                 else if (ret < 0)
 461                         return ret;
 462         }
 463         return ret;
 464 }
 465
 466 int
 467 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 468 {
 469         return a->ops->dump(skb, a, bind, ref);
 470 }
 471
 472 int
 473 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 474 {
 475         int err = -EINVAL;
 476         unsigned char *b = skb_tail_pointer(skb);
 477         struct nlattr *nest;
 478
 479         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 480                 goto nla_put_failure;
 481         if (tcf_action_copy_stats(skb, a, 0))
 482                 goto nla_put_failure;
 483         nest = nla_nest_start(skb, TCA_OPTIONS);
 484         if (nest == NULL)
 485                 goto nla_put_failure;
 486         err = tcf_action_dump_old(skb, a, bind, ref);
 487         if (err > 0) {
 488                 nla_nest_end(skb, nest);
 489                 return err;
 490         }
 491
 492 nla_put_failure:
 493         nlmsg_trim(skb, b);
 494         return -1;
 495 }
 496 EXPORT_SYMBOL(tcf_action_dump_1);
 497
 498 int tcf_action_dump(struct sk_buff *skb, struct list_head *actions,
 499                     int bind, int ref)
 500 {
 501         struct tc_action *a;
 502         int err = -EINVAL;
 503         struct nlattr *nest;
 504
 505         list_for_each_entry(a, actions, list) {
 506                 nest = nla_nest_start(skb, a->order);
 507                 if (nest == NULL)
 508                         goto nla_put_failure;
 509                 err = tcf_action_dump_1(skb, a, bind, ref);
 510                 if (err < 0)
 511                         goto errout;
 512                 nla_nest_end(skb, nest);
 513         }
 514
 515         return 0;
 516
 517 nla_put_failure:
 518         err = -EINVAL;
 519 errout:
 520         nla_nest_cancel(skb, nest);
 521         return err;
 522 }
 523
 524 struct tc_action *tcf_action_init_1(struct net *net, struct nlattr *nla,
 525                                     struct nlattr *est, char *name, int ovr,
 526                                     int bind)
 527 {
 528         struct tc_action *a;
 529         struct tc_action_ops *a_o;
 530         char act_name[IFNAMSIZ];
 531         struct nlattr *tb[TCA_ACT_MAX + 1];
 532         struct nlattr *kind;
 533         int err;
 534
 535         if (name == NULL) {
 536                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 537                 if (err < 0)
 538                         goto err_out;
 539                 err = -EINVAL;
 540                 kind = tb[TCA_ACT_KIND];
 541                 if (kind == NULL)
 542                         goto err_out;
 543                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
 544                         goto err_out;
 545         } else {
 546                 err = -EINVAL;
 547                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
 548                         goto err_out;
 549         }
 550
 551         a_o = tc_lookup_action_n(act_name);
 552         if (a_o == NULL) {
 553 #ifdef CONFIG_MODULES
 554                 rtnl_unlock();
 555                 request_module("act_%s", act_name);
 556                 rtnl_lock();
 557
 558                 a_o = tc_lookup_action_n(act_name);
 559
 560                 /* We dropped the RTNL semaphore in order to
 561                  * perform the module load.  So, even if we
 562                  * succeeded in loading the module we have to
 563                  * tell the caller to replay the request.  We
 564                  * indicate this using -EAGAIN.
 565                  */
 566                 if (a_o != NULL) {
 567                         err = -EAGAIN;
 568                         goto err_mod;
 569                 }
 570 #endif
 571                 err = -ENOENT;
 572                 goto err_out;
 573         }
 574
 575         /* backward compatibility for policer */
 576         if (name == NULL)
 577                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind);
 578         else
 579                 err = a_o->init(net, nla, est, &a, ovr, bind);
 580         if (err < 0)
 581                 goto err_mod;
 582
 583         /* module count goes up only when brand new policy is created
 584          * if it exists and is only bound to in a_o->init() then
 585          * ACT_P_CREATED is not returned (a zero is).
 586          */
 587         if (err != ACT_P_CREATED)
 588                 module_put(a_o->owner);
 589
 590         return a;
 591
 592 err_mod:
 593         module_put(a_o->owner);
 594 err_out:
 595         return ERR_PTR(err);
 596 }
 597
 598 static void cleanup_a(struct list_head *actions, int ovr)
 599 {
 600         struct tc_action *a;
 601
 602         if (!ovr)
 603                 return;
 604
 605         list_for_each_entry(a, actions, list)
 606                 a->tcfa_refcnt--;
 607 }
 608
 609 int tcf_action_init(struct net *net, struct nlattr *nla, struct nlattr *est,
 610                     char *name, int ovr, int bind, struct list_head *actions)
 611 {
 612         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 613         struct tc_action *act;
 614         int err;
 615         int i;
 616
 617         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
 618         if (err < 0)
 619                 return err;
 620
 621         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 622                 act = tcf_action_init_1(net, tb[i], est, name, ovr, bind);
 623                 if (IS_ERR(act)) {
 624                         err = PTR_ERR(act);
 625                         goto err;
 626                 }
 627                 act->order = i;
 628                 if (ovr)
 629                         act->tcfa_refcnt++;
 630                 list_add_tail(&act->list, actions);
 631         }
 632
 633         /* Remove the temp refcnt which was necessary to protect against
 634          * destroying an existing action which was being replaced
 635          */
 636         cleanup_a(actions, ovr);
 637         return 0;
 638
 639 err:
 640         tcf_action_destroy(actions, bind);
 641         return err;
 642 }
 643
 644 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
 645                           int compat_mode)
 646 {
 647         int err = 0;
 648         struct gnet_dump d;
 649
 650         if (p == NULL)
 651                 goto errout;
 652
 653         /* compat_mode being true specifies a call that is supposed
 654          * to add additional backward compatibility statistic TLVs.
 655          */
 656         if (compat_mode) {
 657                 if (p->type == TCA_OLD_COMPAT)
 658                         err = gnet_stats_start_copy_compat(skb, 0,
 659                                                            TCA_STATS,
 660                                                            TCA_XSTATS,
 661                                                            &p->tcfa_lock, &d,
 662                                                            TCA_PAD);
 663                 else
 664                         return 0;
 665         } else
 666                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
 667                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
 668
 669         if (err < 0)
 670                 goto errout;
 671
 672         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
 673             gnet_stats_copy_rate_est(&d, &p->tcfa_bstats,
 674                                      &p->tcfa_rate_est) < 0 ||
 675             gnet_stats_copy_queue(&d, p->cpu_qstats,
 676                                   &p->tcfa_qstats,
 677                                   p->tcfa_qstats.qlen) < 0)
 678                 goto errout;
 679
 680         if (gnet_stats_finish_copy(&d) < 0)
 681                 goto errout;
 682
 683         return 0;
 684
 685 errout:
 686         return -1;
 687 }
 688
 689 static int tca_get_fill(struct sk_buff *skb, struct list_head *actions,
 690                         u32 portid, u32 seq, u16 flags, int event, int bind,
 691                         int ref)
 692 {
 693         struct tcamsg *t;
 694         struct nlmsghdr *nlh;
 695         unsigned char *b = skb_tail_pointer(skb);
 696         struct nlattr *nest;
 697
 698         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
 699         if (!nlh)
 700                 goto out_nlmsg_trim;
 701         t = nlmsg_data(nlh);
 702         t->tca_family = AF_UNSPEC;
 703         t->tca__pad1 = 0;
 704         t->tca__pad2 = 0;
 705
 706         nest = nla_nest_start(skb, TCA_ACT_TAB);
 707         if (nest == NULL)
 708                 goto out_nlmsg_trim;
 709
 710         if (tcf_action_dump(skb, actions, bind, ref) < 0)
 711                 goto out_nlmsg_trim;
 712
 713         nla_nest_end(skb, nest);
 714
 715         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 716         return skb->len;
 717
 718 out_nlmsg_trim:
 719         nlmsg_trim(skb, b);
 720         return -1;
 721 }
 722
 723 static int
 724 act_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
 725                struct list_head *actions, int event)
 726 {
 727         struct sk_buff *skb;
 728
 729         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 730         if (!skb)
 731                 return -ENOBUFS;
 732         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
 733                          0, 0) <= 0) {
 734                 kfree_skb(skb);
 735                 return -EINVAL;
 736         }
 737
 738         return rtnl_unicast(skb, net, portid);
 739 }
 740
 741 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
 742                                           struct nlmsghdr *n, u32 portid)
 743 {
 744         struct nlattr *tb[TCA_ACT_MAX + 1];
 745         const struct tc_action_ops *ops;
 746         struct tc_action *a;
 747         int index;
 748         int err;
 749
 750         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 751         if (err < 0)
 752                 goto err_out;
 753
 754         err = -EINVAL;
 755         if (tb[TCA_ACT_INDEX] == NULL ||
 756             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
 757                 goto err_out;
 758         index = nla_get_u32(tb[TCA_ACT_INDEX]);
 759
 760         err = -EINVAL;
 761         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
 762         if (!ops) /* could happen in batch of actions */
 763                 goto err_out;
 764         err = -ENOENT;
 765         if (ops->lookup(net, &a, index) == 0)
 766                 goto err_mod;
 767
 768         module_put(ops->owner);
 769         return a;
 770
 771 err_mod:
 772         module_put(ops->owner);
 773 err_out:
 774         return ERR_PTR(err);
 775 }
 776
 777 static int tca_action_flush(struct net *net, struct nlattr *nla,
 778                             struct nlmsghdr *n, u32 portid)
 779 {
 780         struct sk_buff *skb;
 781         unsigned char *b;
 782         struct nlmsghdr *nlh;
 783         struct tcamsg *t;
 784         struct netlink_callback dcb;
 785         struct nlattr *nest;
 786         struct nlattr *tb[TCA_ACT_MAX + 1];
 787         const struct tc_action_ops *ops;
 788         struct nlattr *kind;
 789         int err = -ENOMEM;
 790
 791         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 792         if (!skb) {
 793                 pr_debug("tca_action_flush: failed skb alloc\n");
 794                 return err;
 795         }
 796
 797         b = skb_tail_pointer(skb);
 798
 799         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
 800         if (err < 0)
 801                 goto err_out;
 802
 803         err = -EINVAL;
 804         kind = tb[TCA_ACT_KIND];
 805         ops = tc_lookup_action(kind);
 806         if (!ops) /*some idjot trying to flush unknown action */
 807                 goto err_out;
 808
 809         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
 810                         sizeof(*t), 0);
 811         if (!nlh)
 812                 goto out_module_put;
 813         t = nlmsg_data(nlh);
 814         t->tca_family = AF_UNSPEC;
 815         t->tca__pad1 = 0;
 816         t->tca__pad2 = 0;
 817
 818         nest = nla_nest_start(skb, TCA_ACT_TAB);
 819         if (nest == NULL)
 820                 goto out_module_put;
 821
 822         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops);
 823         if (err < 0)
 824                 goto out_module_put;
 825         if (err == 0)
 826                 goto noflush_out;
 827
 828         nla_nest_end(skb, nest);
 829
 830         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 831         nlh->nlmsg_flags |= NLM_F_ROOT;
 832         module_put(ops->owner);
 833         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 834                              n->nlmsg_flags & NLM_F_ECHO);
 835         if (err > 0)
 836                 return 0;
 837
 838         return err;
 839
 840 out_module_put:
 841         module_put(ops->owner);
 842 err_out:
 843 noflush_out:
 844         kfree_skb(skb);
 845         return err;
 846 }
 847
 848 static int
 849 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
 850                u32 portid)
 851 {
 852         int ret;
 853         struct sk_buff *skb;
 854
 855         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 856         if (!skb)
 857                 return -ENOBUFS;
 858
 859         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
 860                          0, 1) <= 0) {
 861                 kfree_skb(skb);
 862                 return -EINVAL;
 863         }
 864
 865         /* now do the delete */
 866         ret = tcf_action_destroy(actions, 0);
 867         if (ret < 0) {
 868                 kfree_skb(skb);
 869                 return ret;
 870         }
 871
 872         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 873                              n->nlmsg_flags & NLM_F_ECHO);
 874         if (ret > 0)
 875                 return 0;
 876         return ret;
 877 }
 878
 879 static int
 880 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
 881               u32 portid, int event)
 882 {
 883         int i, ret;
 884         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 885         struct tc_action *act;
 886         LIST_HEAD(actions);
 887
 888         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
 889         if (ret < 0)
 890                 return ret;
 891
 892         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
 893                 if (tb[1] != NULL)
 894                         return tca_action_flush(net, tb[1], n, portid);
 895                 else
 896                         return -EINVAL;
 897         }
 898
 899         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 900                 act = tcf_action_get_1(net, tb[i], n, portid);
 901                 if (IS_ERR(act)) {
 902                         ret = PTR_ERR(act);
 903                         goto err;
 904                 }
 905                 act->order = i;
 906                 if (event == RTM_GETACTION)
 907                         act->tcfa_refcnt++;
 908                 list_add_tail(&act->list, &actions);
 909         }
 910
 911         if (event == RTM_GETACTION)
 912                 ret = act_get_notify(net, portid, n, &actions, event);
 913         else { /* delete */
 914                 ret = tcf_del_notify(net, n, &actions, portid);
 915                 if (ret)
 916                         goto err;
 917                 return ret;
 918         }
 919 err:
 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 ret = 0;
 952         LIST_HEAD(actions);
 953
 954         ret = tcf_action_init(net, nla, NULL, NULL, ovr, 0, &actions);
 955         if (ret)
 956                 return ret;
 957
 958         return tcf_add_notify(net, n, &actions, portid);
 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 *find_dump_kind(const struct nlmsghdr *n)
1013 {
1014         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1015         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1016         struct nlattr *nla[TCAA_MAX + 1];
1017         struct nlattr *kind;
1018
1019         if (nlmsg_parse(n, sizeof(struct tcamsg), nla, TCAA_MAX, NULL) < 0)
1020                 return NULL;
1021         tb1 = nla[TCA_ACT_TAB];
1022         if (tb1 == NULL)
1023                 return NULL;
1024
1025         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1026                       NLMSG_ALIGN(nla_len(tb1)), NULL) < 0)
1027                 return NULL;
1028
1029         if (tb[1] == NULL)
1030                 return NULL;
1031         if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL) < 0)
1032                 return NULL;
1033         kind = tb2[TCA_ACT_KIND];
1034
1035         return kind;
1036 }
1037
1038 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1039 {
1040         struct net *net = sock_net(skb->sk);
1041         struct nlmsghdr *nlh;
1042         unsigned char *b = skb_tail_pointer(skb);
1043         struct nlattr *nest;
1044         struct tc_action_ops *a_o;
1045         int ret = 0;
1046         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1047         struct nlattr *kind = find_dump_kind(cb->nlh);
1048
1049         if (kind == NULL) {
1050                 pr_info("tc_dump_action: action bad kind\n");
1051                 return 0;
1052         }
1053
1054         a_o = tc_lookup_action(kind);
1055         if (a_o == NULL)
1056                 return 0;
1057
1058         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1059                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1060         if (!nlh)
1061                 goto out_module_put;
1062         t = nlmsg_data(nlh);
1063         t->tca_family = AF_UNSPEC;
1064         t->tca__pad1 = 0;
1065         t->tca__pad2 = 0;
1066
1067         nest = nla_nest_start(skb, TCA_ACT_TAB);
1068         if (nest == NULL)
1069                 goto out_module_put;
1070
1071         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o);
1072         if (ret < 0)
1073                 goto out_module_put;
1074
1075         if (ret > 0) {
1076                 nla_nest_end(skb, nest);
1077                 ret = skb->len;
1078         } else
1079                 nlmsg_trim(skb, b);
1080
1081         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1082         if (NETLINK_CB(cb->skb).portid && ret)
1083                 nlh->nlmsg_flags |= NLM_F_MULTI;
1084         module_put(a_o->owner);
1085         return skb->len;
1086
1087 out_module_put:
1088         module_put(a_o->owner);
1089         nlmsg_trim(skb, b);
1090         return skb->len;
1091 }
1092
1093 static int __init tc_action_init(void)
1094 {
1095         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, NULL);
1096         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, NULL);
1097         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1098                       NULL);
1099
1100         return 0;
1101 }
1102
1103 subsys_initcall(tc_action_init);