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/pkt_cls.h>
  28 #include <net/act_api.h>
  29 #include <net/netlink.h>
  30
  31 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
  32 {
  33         u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
  34
  35         if (!tp)
  36                 return -EINVAL;
  37         a->goto_chain = tcf_chain_get(tp->chain->block, chain_index, true);
  38         if (!a->goto_chain)
  39                 return -ENOMEM;
  40         return 0;
  41 }
  42
  43 static void tcf_action_goto_chain_fini(struct tc_action *a)
  44 {
  45         tcf_chain_put(a->goto_chain);
  46 }
  47
  48 static void tcf_action_goto_chain_exec(const struct tc_action *a,
  49                                        struct tcf_result *res)
  50 {
  51         const struct tcf_chain *chain = a->goto_chain;
  52
  53         res->goto_tp = rcu_dereference_bh(chain->filter_chain);
  54 }
  55
  56 static void free_tcf(struct rcu_head *head)
  57 {
  58         struct tc_action *p = container_of(head, struct tc_action, tcfa_rcu);
  59
  60         free_percpu(p->cpu_bstats);
  61         free_percpu(p->cpu_qstats);
  62
  63         if (p->act_cookie) {
  64                 kfree(p->act_cookie->data);
  65                 kfree(p->act_cookie);
  66         }
  67         if (p->goto_chain)
  68                 tcf_action_goto_chain_fini(p);
  69
  70         kfree(p);
  71 }
  72
  73 static void tcf_hash_destroy(struct tcf_hashinfo *hinfo, struct tc_action *p)
  74 {
  75         spin_lock_bh(&hinfo->lock);
  76         hlist_del(&p->tcfa_head);
  77         spin_unlock_bh(&hinfo->lock);
  78         gen_kill_estimator(&p->tcfa_rate_est);
  79         /*
  80          * gen_estimator est_timer() might access p->tcfa_lock
  81          * or bstats, wait a RCU grace period before freeing p
  82          */
  83         call_rcu(&p->tcfa_rcu, free_tcf);
  84 }
  85
  86 int __tcf_hash_release(struct tc_action *p, bool bind, bool strict)
  87 {
  88         int ret = 0;
  89
  90         if (p) {
  91                 if (bind)
  92                         p->tcfa_bindcnt--;
  93                 else if (strict && p->tcfa_bindcnt > 0)
  94                         return -EPERM;
  95
  96                 p->tcfa_refcnt--;
  97                 if (p->tcfa_bindcnt <= 0 && p->tcfa_refcnt <= 0) {
  98                         if (p->ops->cleanup)
  99                                 p->ops->cleanup(p, bind);
 100                         tcf_hash_destroy(p->hinfo, p);
 101                         ret = ACT_P_DELETED;
 102                 }
 103         }
 104
 105         return ret;
 106 }
 107 EXPORT_SYMBOL(__tcf_hash_release);
 108
 109 static int tcf_dump_walker(struct tcf_hashinfo *hinfo, struct sk_buff *skb,
 110                            struct netlink_callback *cb)
 111 {
 112         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
 113         struct nlattr *nest;
 114
 115         spin_lock_bh(&hinfo->lock);
 116
 117         s_i = cb->args[0];
 118
 119         for (i = 0; i < (hinfo->hmask + 1); i++) {
 120                 struct hlist_head *head;
 121                 struct tc_action *p;
 122
 123                 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
 124
 125                 hlist_for_each_entry_rcu(p, head, tcfa_head) {
 126                         index++;
 127                         if (index < s_i)
 128                                 continue;
 129
 130                         nest = nla_nest_start(skb, n_i);
 131                         if (nest == NULL)
 132                                 goto nla_put_failure;
 133                         err = tcf_action_dump_1(skb, p, 0, 0);
 134                         if (err < 0) {
 135                                 index--;
 136                                 nlmsg_trim(skb, nest);
 137                                 goto done;
 138                         }
 139                         nla_nest_end(skb, nest);
 140                         n_i++;
 141                         if (n_i >= TCA_ACT_MAX_PRIO)
 142                                 goto done;
 143                 }
 144         }
 145 done:
 146         spin_unlock_bh(&hinfo->lock);
 147         if (n_i)
 148                 cb->args[0] += n_i;
 149         return n_i;
 150
 151 nla_put_failure:
 152         nla_nest_cancel(skb, nest);
 153         goto done;
 154 }
 155
 156 static int tcf_del_walker(struct tcf_hashinfo *hinfo, struct sk_buff *skb,
 157                           const struct tc_action_ops *ops)
 158 {
 159         struct nlattr *nest;
 160         int i = 0, n_i = 0;
 161         int ret = -EINVAL;
 162
 163         nest = nla_nest_start(skb, 0);
 164         if (nest == NULL)
 165                 goto nla_put_failure;
 166         if (nla_put_string(skb, TCA_KIND, ops->kind))
 167                 goto nla_put_failure;
 168         for (i = 0; i < (hinfo->hmask + 1); i++) {
 169                 struct hlist_head *head;
 170                 struct hlist_node *n;
 171                 struct tc_action *p;
 172
 173                 head = &hinfo->htab[tcf_hash(i, hinfo->hmask)];
 174                 hlist_for_each_entry_safe(p, n, head, tcfa_head) {
 175                         ret = __tcf_hash_release(p, false, true);
 176                         if (ret == ACT_P_DELETED) {
 177                                 module_put(p->ops->owner);
 178                                 n_i++;
 179                         } else if (ret < 0)
 180                                 goto nla_put_failure;
 181                 }
 182         }
 183         if (nla_put_u32(skb, TCA_FCNT, n_i))
 184                 goto nla_put_failure;
 185         nla_nest_end(skb, nest);
 186
 187         return n_i;
 188 nla_put_failure:
 189         nla_nest_cancel(skb, nest);
 190         return ret;
 191 }
 192
 193 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 194                        struct netlink_callback *cb, int type,
 195                        const struct tc_action_ops *ops)
 196 {
 197         struct tcf_hashinfo *hinfo = tn->hinfo;
 198
 199         if (type == RTM_DELACTION) {
 200                 return tcf_del_walker(hinfo, skb, ops);
 201         } else if (type == RTM_GETACTION) {
 202                 return tcf_dump_walker(hinfo, skb, cb);
 203         } else {
 204                 WARN(1, "tcf_generic_walker: unknown action %d\n", type);
 205                 return -EINVAL;
 206         }
 207 }
 208 EXPORT_SYMBOL(tcf_generic_walker);
 209
 210 static struct tc_action *tcf_hash_lookup(u32 index, struct tcf_hashinfo *hinfo)
 211 {
 212         struct tc_action *p = NULL;
 213         struct hlist_head *head;
 214
 215         spin_lock_bh(&hinfo->lock);
 216         head = &hinfo->htab[tcf_hash(index, hinfo->hmask)];
 217         hlist_for_each_entry_rcu(p, head, tcfa_head)
 218                 if (p->tcfa_index == index)
 219                         break;
 220         spin_unlock_bh(&hinfo->lock);
 221
 222         return p;
 223 }
 224
 225 u32 tcf_hash_new_index(struct tc_action_net *tn)
 226 {
 227         struct tcf_hashinfo *hinfo = tn->hinfo;
 228         u32 val = hinfo->index;
 229
 230         do {
 231                 if (++val == 0)
 232                         val = 1;
 233         } while (tcf_hash_lookup(val, hinfo));
 234
 235         hinfo->index = val;
 236         return val;
 237 }
 238 EXPORT_SYMBOL(tcf_hash_new_index);
 239
 240 int tcf_hash_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 241 {
 242         struct tcf_hashinfo *hinfo = tn->hinfo;
 243         struct tc_action *p = tcf_hash_lookup(index, hinfo);
 244
 245         if (p) {
 246                 *a = p;
 247                 return 1;
 248         }
 249         return 0;
 250 }
 251 EXPORT_SYMBOL(tcf_hash_search);
 252
 253 bool tcf_hash_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
 254                     int bind)
 255 {
 256         struct tcf_hashinfo *hinfo = tn->hinfo;
 257         struct tc_action *p = NULL;
 258
 259         if (index && (p = tcf_hash_lookup(index, hinfo)) != NULL) {
 260                 if (bind)
 261                         p->tcfa_bindcnt++;
 262                 p->tcfa_refcnt++;
 263                 *a = p;
 264                 return true;
 265         }
 266         return false;
 267 }
 268 EXPORT_SYMBOL(tcf_hash_check);
 269
 270 void tcf_hash_cleanup(struct tc_action *a, struct nlattr *est)
 271 {
 272         if (est)
 273                 gen_kill_estimator(&a->tcfa_rate_est);
 274         call_rcu(&a->tcfa_rcu, free_tcf);
 275 }
 276 EXPORT_SYMBOL(tcf_hash_cleanup);
 277
 278 int tcf_hash_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 279                     struct tc_action **a, const struct tc_action_ops *ops,
 280                     int bind, bool cpustats)
 281 {
 282         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 283         struct tcf_hashinfo *hinfo = tn->hinfo;
 284         int err = -ENOMEM;
 285
 286         if (unlikely(!p))
 287                 return -ENOMEM;
 288         p->tcfa_refcnt = 1;
 289         if (bind)
 290                 p->tcfa_bindcnt = 1;
 291
 292         if (cpustats) {
 293                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 294                 if (!p->cpu_bstats) {
 295 err1:
 296                         kfree(p);
 297                         return err;
 298                 }
 299                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 300                 if (!p->cpu_qstats) {
 301 err2:
 302                         free_percpu(p->cpu_bstats);
 303                         goto err1;
 304                 }
 305         }
 306         spin_lock_init(&p->tcfa_lock);
 307         INIT_HLIST_NODE(&p->tcfa_head);
 308         p->tcfa_index = index ? index : tcf_hash_new_index(tn);
 309         p->tcfa_tm.install = jiffies;
 310         p->tcfa_tm.lastuse = jiffies;
 311         p->tcfa_tm.firstuse = 0;
 312         if (est) {
 313                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 314                                         &p->tcfa_rate_est,
 315                                         &p->tcfa_lock, NULL, est);
 316                 if (err) {
 317                         free_percpu(p->cpu_qstats);
 318                         goto err2;
 319                 }
 320         }
 321
 322         p->hinfo = hinfo;
 323         p->ops = ops;
 324         INIT_LIST_HEAD(&p->list);
 325         *a = p;
 326         return 0;
 327 }
 328 EXPORT_SYMBOL(tcf_hash_create);
 329
 330 void tcf_hash_insert(struct tc_action_net *tn, struct tc_action *a)
 331 {
 332         struct tcf_hashinfo *hinfo = tn->hinfo;
 333         unsigned int h = tcf_hash(a->tcfa_index, hinfo->hmask);
 334
 335         spin_lock_bh(&hinfo->lock);
 336         hlist_add_head(&a->tcfa_head, &hinfo->htab[h]);
 337         spin_unlock_bh(&hinfo->lock);
 338 }
 339 EXPORT_SYMBOL(tcf_hash_insert);
 340
 341 void tcf_hashinfo_destroy(const struct tc_action_ops *ops,
 342                           struct tcf_hashinfo *hinfo)
 343 {
 344         int i;
 345
 346         for (i = 0; i < hinfo->hmask + 1; i++) {
 347                 struct tc_action *p;
 348                 struct hlist_node *n;
 349
 350                 hlist_for_each_entry_safe(p, n, &hinfo->htab[i], tcfa_head) {
 351                         int ret;
 352
 353                         ret = __tcf_hash_release(p, false, true);
 354                         if (ret == ACT_P_DELETED)
 355                                 module_put(ops->owner);
 356                         else if (ret < 0)
 357                                 return;
 358                 }
 359         }
 360         kfree(hinfo->htab);
 361 }
 362 EXPORT_SYMBOL(tcf_hashinfo_destroy);
 363
 364 static LIST_HEAD(act_base);
 365 static DEFINE_RWLOCK(act_mod_lock);
 366
 367 int tcf_register_action(struct tc_action_ops *act,
 368                         struct pernet_operations *ops)
 369 {
 370         struct tc_action_ops *a;
 371         int ret;
 372
 373         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 374                 return -EINVAL;
 375
 376         /* We have to register pernet ops before making the action ops visible,
 377          * otherwise tcf_action_init_1() could get a partially initialized
 378          * netns.
 379          */
 380         ret = register_pernet_subsys(ops);
 381         if (ret)
 382                 return ret;
 383
 384         write_lock(&act_mod_lock);
 385         list_for_each_entry(a, &act_base, head) {
 386                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
 387                         write_unlock(&act_mod_lock);
 388                         unregister_pernet_subsys(ops);
 389                         return -EEXIST;
 390                 }
 391         }
 392         list_add_tail(&act->head, &act_base);
 393         write_unlock(&act_mod_lock);
 394
 395         return 0;
 396 }
 397 EXPORT_SYMBOL(tcf_register_action);
 398
 399 int tcf_unregister_action(struct tc_action_ops *act,
 400                           struct pernet_operations *ops)
 401 {
 402         struct tc_action_ops *a;
 403         int err = -ENOENT;
 404
 405         write_lock(&act_mod_lock);
 406         list_for_each_entry(a, &act_base, head) {
 407                 if (a == act) {
 408                         list_del(&act->head);
 409                         err = 0;
 410                         break;
 411                 }
 412         }
 413         write_unlock(&act_mod_lock);
 414         if (!err)
 415                 unregister_pernet_subsys(ops);
 416         return err;
 417 }
 418 EXPORT_SYMBOL(tcf_unregister_action);
 419
 420 /* lookup by name */
 421 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 422 {
 423         struct tc_action_ops *a, *res = NULL;
 424
 425         if (kind) {
 426                 read_lock(&act_mod_lock);
 427                 list_for_each_entry(a, &act_base, head) {
 428                         if (strcmp(kind, a->kind) == 0) {
 429                                 if (try_module_get(a->owner))
 430                                         res = a;
 431                                 break;
 432                         }
 433                 }
 434                 read_unlock(&act_mod_lock);
 435         }
 436         return res;
 437 }
 438
 439 /* lookup by nlattr */
 440 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 441 {
 442         struct tc_action_ops *a, *res = NULL;
 443
 444         if (kind) {
 445                 read_lock(&act_mod_lock);
 446                 list_for_each_entry(a, &act_base, head) {
 447                         if (nla_strcmp(kind, a->kind) == 0) {
 448                                 if (try_module_get(a->owner))
 449                                         res = a;
 450                                 break;
 451                         }
 452                 }
 453                 read_unlock(&act_mod_lock);
 454         }
 455         return res;
 456 }
 457
 458 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
 459 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 460 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 461                     int nr_actions, struct tcf_result *res)
 462 {
 463         int ret = -1, i;
 464         u32 jmp_prgcnt = 0;
 465         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 466
 467         if (skb_skip_tc_classify(skb))
 468                 return TC_ACT_OK;
 469
 470 restart_act_graph:
 471         for (i = 0; i < nr_actions; i++) {
 472                 const struct tc_action *a = actions[i];
 473
 474                 if (jmp_prgcnt > 0) {
 475                         jmp_prgcnt -= 1;
 476                         continue;
 477                 }
 478 repeat:
 479                 ret = a->ops->act(skb, a, res);
 480                 if (ret == TC_ACT_REPEAT)
 481                         goto repeat;    /* we need a ttl - JHS */
 482
 483                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 484                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 485                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 486                                 /* faulty opcode, stop pipeline */
 487                                 return TC_ACT_OK;
 488                         } else {
 489                                 jmp_ttl -= 1;
 490                                 if (jmp_ttl > 0)
 491                                         goto restart_act_graph;
 492                                 else /* faulty graph, stop pipeline */
 493                                         return TC_ACT_OK;
 494                         }
 495                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 496                         tcf_action_goto_chain_exec(a, res);
 497                 }
 498
 499                 if (ret != TC_ACT_PIPE)
 500                         break;
 501         }
 502
 503         return ret;
 504 }
 505 EXPORT_SYMBOL(tcf_action_exec);
 506
 507 int tcf_action_destroy(struct list_head *actions, int bind)
 508 {
 509         struct tc_action *a, *tmp;
 510         int ret = 0;
 511
 512         list_for_each_entry_safe(a, tmp, actions, list) {
 513                 ret = __tcf_hash_release(a, bind, true);
 514                 if (ret == ACT_P_DELETED)
 515                         module_put(a->ops->owner);
 516                 else if (ret < 0)
 517                         return ret;
 518         }
 519         return ret;
 520 }
 521
 522 int
 523 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 524 {
 525         return a->ops->dump(skb, a, bind, ref);
 526 }
 527
 528 int
 529 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 530 {
 531         int err = -EINVAL;
 532         unsigned char *b = skb_tail_pointer(skb);
 533         struct nlattr *nest;
 534
 535         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 536                 goto nla_put_failure;
 537         if (tcf_action_copy_stats(skb, a, 0))
 538                 goto nla_put_failure;
 539         if (a->act_cookie) {
 540                 if (nla_put(skb, TCA_ACT_COOKIE, a->act_cookie->len,
 541                             a->act_cookie->data))
 542                         goto nla_put_failure;
 543         }
 544
 545         nest = nla_nest_start(skb, TCA_OPTIONS);
 546         if (nest == NULL)
 547                 goto nla_put_failure;
 548         err = tcf_action_dump_old(skb, a, bind, ref);
 549         if (err > 0) {
 550                 nla_nest_end(skb, nest);
 551                 return err;
 552         }
 553
 554 nla_put_failure:
 555         nlmsg_trim(skb, b);
 556         return -1;
 557 }
 558 EXPORT_SYMBOL(tcf_action_dump_1);
 559
 560 int tcf_action_dump(struct sk_buff *skb, struct list_head *actions,
 561                     int bind, int ref)
 562 {
 563         struct tc_action *a;
 564         int err = -EINVAL;
 565         struct nlattr *nest;
 566
 567         list_for_each_entry(a, actions, list) {
 568                 nest = nla_nest_start(skb, a->order);
 569                 if (nest == NULL)
 570                         goto nla_put_failure;
 571                 err = tcf_action_dump_1(skb, a, bind, ref);
 572                 if (err < 0)
 573                         goto errout;
 574                 nla_nest_end(skb, nest);
 575         }
 576
 577         return 0;
 578
 579 nla_put_failure:
 580         err = -EINVAL;
 581 errout:
 582         nla_nest_cancel(skb, nest);
 583         return err;
 584 }
 585
 586 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 587 {
 588         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 589         if (!c)
 590                 return NULL;
 591
 592         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 593         if (!c->data) {
 594                 kfree(c);
 595                 return NULL;
 596         }
 597         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 598
 599         return c;
 600 }
 601
 602 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
 603                                     struct nlattr *nla, struct nlattr *est,
 604                                     char *name, int ovr, int bind)
 605 {
 606         struct tc_action *a;
 607         struct tc_action_ops *a_o;
 608         struct tc_cookie *cookie = NULL;
 609         char act_name[IFNAMSIZ];
 610         struct nlattr *tb[TCA_ACT_MAX + 1];
 611         struct nlattr *kind;
 612         int err;
 613
 614         if (name == NULL) {
 615                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
 616                 if (err < 0)
 617                         goto err_out;
 618                 err = -EINVAL;
 619                 kind = tb[TCA_ACT_KIND];
 620                 if (kind == NULL)
 621                         goto err_out;
 622                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
 623                         goto err_out;
 624                 if (tb[TCA_ACT_COOKIE]) {
 625                         int cklen = nla_len(tb[TCA_ACT_COOKIE]);
 626
 627                         if (cklen > TC_COOKIE_MAX_SIZE)
 628                                 goto err_out;
 629
 630                         cookie = nla_memdup_cookie(tb);
 631                         if (!cookie) {
 632                                 err = -ENOMEM;
 633                                 goto err_out;
 634                         }
 635                 }
 636         } else {
 637                 err = -EINVAL;
 638                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
 639                         goto err_out;
 640         }
 641
 642         a_o = tc_lookup_action_n(act_name);
 643         if (a_o == NULL) {
 644 #ifdef CONFIG_MODULES
 645                 rtnl_unlock();
 646                 request_module("act_%s", act_name);
 647                 rtnl_lock();
 648
 649                 a_o = tc_lookup_action_n(act_name);
 650
 651                 /* We dropped the RTNL semaphore in order to
 652                  * perform the module load.  So, even if we
 653                  * succeeded in loading the module we have to
 654                  * tell the caller to replay the request.  We
 655                  * indicate this using -EAGAIN.
 656                  */
 657                 if (a_o != NULL) {
 658                         err = -EAGAIN;
 659                         goto err_mod;
 660                 }
 661 #endif
 662                 err = -ENOENT;
 663                 goto err_out;
 664         }
 665
 666         /* backward compatibility for policer */
 667         if (name == NULL)
 668                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind);
 669         else
 670                 err = a_o->init(net, nla, est, &a, ovr, bind);
 671         if (err < 0)
 672                 goto err_mod;
 673
 674         if (name == NULL && tb[TCA_ACT_COOKIE]) {
 675                 if (a->act_cookie) {
 676                         kfree(a->act_cookie->data);
 677                         kfree(a->act_cookie);
 678                 }
 679                 a->act_cookie = cookie;
 680         }
 681
 682         /* module count goes up only when brand new policy is created
 683          * if it exists and is only bound to in a_o->init() then
 684          * ACT_P_CREATED is not returned (a zero is).
 685          */
 686         if (err != ACT_P_CREATED)
 687                 module_put(a_o->owner);
 688
 689         if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
 690                 err = tcf_action_goto_chain_init(a, tp);
 691                 if (err) {
 692                         LIST_HEAD(actions);
 693
 694                         list_add_tail(&a->list, &actions);
 695                         tcf_action_destroy(&actions, bind);
 696                         return ERR_PTR(err);
 697                 }
 698         }
 699
 700         return a;
 701
 702 err_mod:
 703         module_put(a_o->owner);
 704 err_out:
 705         if (cookie) {
 706                 kfree(cookie->data);
 707                 kfree(cookie);
 708         }
 709         return ERR_PTR(err);
 710 }
 711
 712 static void cleanup_a(struct list_head *actions, int ovr)
 713 {
 714         struct tc_action *a;
 715
 716         if (!ovr)
 717                 return;
 718
 719         list_for_each_entry(a, actions, list)
 720                 a->tcfa_refcnt--;
 721 }
 722
 723 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
 724                     struct nlattr *est, char *name, int ovr, int bind,
 725                     struct list_head *actions)
 726 {
 727         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 728         struct tc_action *act;
 729         int err;
 730         int i;
 731
 732         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, NULL);
 733         if (err < 0)
 734                 return err;
 735
 736         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 737                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind);
 738                 if (IS_ERR(act)) {
 739                         err = PTR_ERR(act);
 740                         goto err;
 741                 }
 742                 act->order = i;
 743                 if (ovr)
 744                         act->tcfa_refcnt++;
 745                 list_add_tail(&act->list, actions);
 746         }
 747
 748         /* Remove the temp refcnt which was necessary to protect against
 749          * destroying an existing action which was being replaced
 750          */
 751         cleanup_a(actions, ovr);
 752         return 0;
 753
 754 err:
 755         tcf_action_destroy(actions, bind);
 756         return err;
 757 }
 758
 759 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
 760                           int compat_mode)
 761 {
 762         int err = 0;
 763         struct gnet_dump d;
 764
 765         if (p == NULL)
 766                 goto errout;
 767
 768         /* compat_mode being true specifies a call that is supposed
 769          * to add additional backward compatibility statistic TLVs.
 770          */
 771         if (compat_mode) {
 772                 if (p->type == TCA_OLD_COMPAT)
 773                         err = gnet_stats_start_copy_compat(skb, 0,
 774                                                            TCA_STATS,
 775                                                            TCA_XSTATS,
 776                                                            &p->tcfa_lock, &d,
 777                                                            TCA_PAD);
 778                 else
 779                         return 0;
 780         } else
 781                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
 782                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
 783
 784         if (err < 0)
 785                 goto errout;
 786
 787         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
 788             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
 789             gnet_stats_copy_queue(&d, p->cpu_qstats,
 790                                   &p->tcfa_qstats,
 791                                   p->tcfa_qstats.qlen) < 0)
 792                 goto errout;
 793
 794         if (gnet_stats_finish_copy(&d) < 0)
 795                 goto errout;
 796
 797         return 0;
 798
 799 errout:
 800         return -1;
 801 }
 802
 803 static int tca_get_fill(struct sk_buff *skb, struct list_head *actions,
 804                         u32 portid, u32 seq, u16 flags, int event, int bind,
 805                         int ref)
 806 {
 807         struct tcamsg *t;
 808         struct nlmsghdr *nlh;
 809         unsigned char *b = skb_tail_pointer(skb);
 810         struct nlattr *nest;
 811
 812         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
 813         if (!nlh)
 814                 goto out_nlmsg_trim;
 815         t = nlmsg_data(nlh);
 816         t->tca_family = AF_UNSPEC;
 817         t->tca__pad1 = 0;
 818         t->tca__pad2 = 0;
 819
 820         nest = nla_nest_start(skb, TCA_ACT_TAB);
 821         if (nest == NULL)
 822                 goto out_nlmsg_trim;
 823
 824         if (tcf_action_dump(skb, actions, bind, ref) < 0)
 825                 goto out_nlmsg_trim;
 826
 827         nla_nest_end(skb, nest);
 828
 829         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 830         return skb->len;
 831
 832 out_nlmsg_trim:
 833         nlmsg_trim(skb, b);
 834         return -1;
 835 }
 836
 837 static int
 838 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
 839                struct list_head *actions, int event)
 840 {
 841         struct sk_buff *skb;
 842
 843         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 844         if (!skb)
 845                 return -ENOBUFS;
 846         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
 847                          0, 0) <= 0) {
 848                 kfree_skb(skb);
 849                 return -EINVAL;
 850         }
 851
 852         return rtnl_unicast(skb, net, portid);
 853 }
 854
 855 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
 856                                           struct nlmsghdr *n, u32 portid)
 857 {
 858         struct nlattr *tb[TCA_ACT_MAX + 1];
 859         const struct tc_action_ops *ops;
 860         struct tc_action *a;
 861         int index;
 862         int err;
 863
 864         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
 865         if (err < 0)
 866                 goto err_out;
 867
 868         err = -EINVAL;
 869         if (tb[TCA_ACT_INDEX] == NULL ||
 870             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
 871                 goto err_out;
 872         index = nla_get_u32(tb[TCA_ACT_INDEX]);
 873
 874         err = -EINVAL;
 875         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
 876         if (!ops) /* could happen in batch of actions */
 877                 goto err_out;
 878         err = -ENOENT;
 879         if (ops->lookup(net, &a, index) == 0)
 880                 goto err_mod;
 881
 882         module_put(ops->owner);
 883         return a;
 884
 885 err_mod:
 886         module_put(ops->owner);
 887 err_out:
 888         return ERR_PTR(err);
 889 }
 890
 891 static int tca_action_flush(struct net *net, struct nlattr *nla,
 892                             struct nlmsghdr *n, u32 portid)
 893 {
 894         struct sk_buff *skb;
 895         unsigned char *b;
 896         struct nlmsghdr *nlh;
 897         struct tcamsg *t;
 898         struct netlink_callback dcb;
 899         struct nlattr *nest;
 900         struct nlattr *tb[TCA_ACT_MAX + 1];
 901         const struct tc_action_ops *ops;
 902         struct nlattr *kind;
 903         int err = -ENOMEM;
 904
 905         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 906         if (!skb) {
 907                 pr_debug("tca_action_flush: failed skb alloc\n");
 908                 return err;
 909         }
 910
 911         b = skb_tail_pointer(skb);
 912
 913         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
 914         if (err < 0)
 915                 goto err_out;
 916
 917         err = -EINVAL;
 918         kind = tb[TCA_ACT_KIND];
 919         ops = tc_lookup_action(kind);
 920         if (!ops) /*some idjot trying to flush unknown action */
 921                 goto err_out;
 922
 923         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
 924                         sizeof(*t), 0);
 925         if (!nlh)
 926                 goto out_module_put;
 927         t = nlmsg_data(nlh);
 928         t->tca_family = AF_UNSPEC;
 929         t->tca__pad1 = 0;
 930         t->tca__pad2 = 0;
 931
 932         nest = nla_nest_start(skb, TCA_ACT_TAB);
 933         if (nest == NULL)
 934                 goto out_module_put;
 935
 936         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops);
 937         if (err <= 0)
 938                 goto out_module_put;
 939
 940         nla_nest_end(skb, nest);
 941
 942         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 943         nlh->nlmsg_flags |= NLM_F_ROOT;
 944         module_put(ops->owner);
 945         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 946                              n->nlmsg_flags & NLM_F_ECHO);
 947         if (err > 0)
 948                 return 0;
 949
 950         return err;
 951
 952 out_module_put:
 953         module_put(ops->owner);
 954 err_out:
 955         kfree_skb(skb);
 956         return err;
 957 }
 958
 959 static int
 960 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
 961                u32 portid)
 962 {
 963         int ret;
 964         struct sk_buff *skb;
 965
 966         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 967         if (!skb)
 968                 return -ENOBUFS;
 969
 970         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
 971                          0, 1) <= 0) {
 972                 kfree_skb(skb);
 973                 return -EINVAL;
 974         }
 975
 976         /* now do the delete */
 977         ret = tcf_action_destroy(actions, 0);
 978         if (ret < 0) {
 979                 kfree_skb(skb);
 980                 return ret;
 981         }
 982
 983         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 984                              n->nlmsg_flags & NLM_F_ECHO);
 985         if (ret > 0)
 986                 return 0;
 987         return ret;
 988 }
 989
 990 static int
 991 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
 992               u32 portid, int event)
 993 {
 994         int i, ret;
 995         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 996         struct tc_action *act;
 997         LIST_HEAD(actions);
 998
 999         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, NULL);
1000         if (ret < 0)
1001                 return ret;
1002
1003         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1004                 if (tb[1] != NULL)
1005                         return tca_action_flush(net, tb[1], n, portid);
1006                 else
1007                         return -EINVAL;
1008         }
1009
1010         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1011                 act = tcf_action_get_1(net, tb[i], n, portid);
1012                 if (IS_ERR(act)) {
1013                         ret = PTR_ERR(act);
1014                         goto err;
1015                 }
1016                 act->order = i;
1017                 list_add_tail(&act->list, &actions);
1018         }
1019
1020         if (event == RTM_GETACTION)
1021                 ret = tcf_get_notify(net, portid, n, &actions, event);
1022         else { /* delete */
1023                 ret = tcf_del_notify(net, n, &actions, portid);
1024                 if (ret)
1025                         goto err;
1026                 return ret;
1027         }
1028 err:
1029         if (event != RTM_GETACTION)
1030                 tcf_action_destroy(&actions, 0);
1031         return ret;
1032 }
1033
1034 static int
1035 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
1036                u32 portid)
1037 {
1038         struct sk_buff *skb;
1039         int err = 0;
1040
1041         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1042         if (!skb)
1043                 return -ENOBUFS;
1044
1045         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1046                          RTM_NEWACTION, 0, 0) <= 0) {
1047                 kfree_skb(skb);
1048                 return -EINVAL;
1049         }
1050
1051         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1052                              n->nlmsg_flags & NLM_F_ECHO);
1053         if (err > 0)
1054                 err = 0;
1055         return err;
1056 }
1057
1058 static int tcf_action_add(struct net *net, struct nlattr *nla,
1059                           struct nlmsghdr *n, u32 portid, int ovr)
1060 {
1061         int ret = 0;
1062         LIST_HEAD(actions);
1063
1064         ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, &actions);
1065         if (ret)
1066                 return ret;
1067
1068         return tcf_add_notify(net, n, &actions, portid);
1069 }
1070
1071 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1072                          struct netlink_ext_ack *extack)
1073 {
1074         struct net *net = sock_net(skb->sk);
1075         struct nlattr *tca[TCA_ACT_MAX + 1];
1076         u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1077         int ret = 0, ovr = 0;
1078
1079         if ((n->nlmsg_type != RTM_GETACTION) &&
1080             !netlink_capable(skb, CAP_NET_ADMIN))
1081                 return -EPERM;
1082
1083         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ACT_MAX, NULL,
1084                           extack);
1085         if (ret < 0)
1086                 return ret;
1087
1088         if (tca[TCA_ACT_TAB] == NULL) {
1089                 pr_notice("tc_ctl_action: received NO action attribs\n");
1090                 return -EINVAL;
1091         }
1092
1093         /* n->nlmsg_flags & NLM_F_CREATE */
1094         switch (n->nlmsg_type) {
1095         case RTM_NEWACTION:
1096                 /* we are going to assume all other flags
1097                  * imply create only if it doesn't exist
1098                  * Note that CREATE | EXCL implies that
1099                  * but since we want avoid ambiguity (eg when flags
1100                  * is zero) then just set this
1101                  */
1102                 if (n->nlmsg_flags & NLM_F_REPLACE)
1103                         ovr = 1;
1104 replay:
1105                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr);
1106                 if (ret == -EAGAIN)
1107                         goto replay;
1108                 break;
1109         case RTM_DELACTION:
1110                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1111                                     portid, RTM_DELACTION);
1112                 break;
1113         case RTM_GETACTION:
1114                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1115                                     portid, RTM_GETACTION);
1116                 break;
1117         default:
1118                 BUG();
1119         }
1120
1121         return ret;
1122 }
1123
1124 static struct nlattr *find_dump_kind(const struct nlmsghdr *n)
1125 {
1126         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1127         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1128         struct nlattr *nla[TCAA_MAX + 1];
1129         struct nlattr *kind;
1130
1131         if (nlmsg_parse(n, sizeof(struct tcamsg), nla, TCAA_MAX,
1132                         NULL, NULL) < 0)
1133                 return NULL;
1134         tb1 = nla[TCA_ACT_TAB];
1135         if (tb1 == NULL)
1136                 return NULL;
1137
1138         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1139                       NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1140                 return NULL;
1141
1142         if (tb[1] == NULL)
1143                 return NULL;
1144         if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1145                 return NULL;
1146         kind = tb2[TCA_ACT_KIND];
1147
1148         return kind;
1149 }
1150
1151 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1152 {
1153         struct net *net = sock_net(skb->sk);
1154         struct nlmsghdr *nlh;
1155         unsigned char *b = skb_tail_pointer(skb);
1156         struct nlattr *nest;
1157         struct tc_action_ops *a_o;
1158         int ret = 0;
1159         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1160         struct nlattr *kind = find_dump_kind(cb->nlh);
1161
1162         if (kind == NULL) {
1163                 pr_info("tc_dump_action: action bad kind\n");
1164                 return 0;
1165         }
1166
1167         a_o = tc_lookup_action(kind);
1168         if (a_o == NULL)
1169                 return 0;
1170
1171         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1172                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1173         if (!nlh)
1174                 goto out_module_put;
1175         t = nlmsg_data(nlh);
1176         t->tca_family = AF_UNSPEC;
1177         t->tca__pad1 = 0;
1178         t->tca__pad2 = 0;
1179
1180         nest = nla_nest_start(skb, TCA_ACT_TAB);
1181         if (nest == NULL)
1182                 goto out_module_put;
1183
1184         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o);
1185         if (ret < 0)
1186                 goto out_module_put;
1187
1188         if (ret > 0) {
1189                 nla_nest_end(skb, nest);
1190                 ret = skb->len;
1191         } else
1192                 nlmsg_trim(skb, b);
1193
1194         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1195         if (NETLINK_CB(cb->skb).portid && ret)
1196                 nlh->nlmsg_flags |= NLM_F_MULTI;
1197         module_put(a_o->owner);
1198         return skb->len;
1199
1200 out_module_put:
1201         module_put(a_o->owner);
1202         nlmsg_trim(skb, b);
1203         return skb->len;
1204 }
1205
1206 static int __init tc_action_init(void)
1207 {
1208         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, NULL);
1209         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, NULL);
1210         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1211                       NULL);
1212
1213         return 0;
1214 }
1215
1216 subsys_initcall(tc_action_init);