net/sched/cls_api.c

   1 /*
   2  * net/sched/cls_api.c  Packet classifier 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  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  * Changes:
  12  *
  13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  14  *
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/init.h>
  24 #include <linux/kmod.h>
  25 #include <linux/netlink.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28 #include <net/netlink.h>
  29 #include <net/pkt_sched.h>
  30 #include <net/pkt_cls.h>
  31
  32 /* The list of all installed classifier types */
  33
  34 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
  35
  36 /* Protects list of registered TC modules. It is pure SMP lock. */
  37 static DEFINE_RWLOCK(cls_mod_lock);
  38
  39 /* Find classifier type by string name */
  40
  41 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct rtattr *kind)
  42 {
  43         struct tcf_proto_ops *t = NULL;
  44
  45         if (kind) {
  46                 read_lock(&cls_mod_lock);
  47                 for (t = tcf_proto_base; t; t = t->next) {
  48                         if (rtattr_strcmp(kind, t->kind) == 0) {
  49                                 if (!try_module_get(t->owner))
  50                                         t = NULL;
  51                                 break;
  52                         }
  53                 }
  54                 read_unlock(&cls_mod_lock);
  55         }
  56         return t;
  57 }
  58
  59 /* Register(unregister) new classifier type */
  60
  61 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  62 {
  63         struct tcf_proto_ops *t, **tp;
  64         int rc = -EEXIST;
  65
  66         write_lock(&cls_mod_lock);
  67         for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
  68                 if (!strcmp(ops->kind, t->kind))
  69                         goto out;
  70
  71         ops->next = NULL;
  72         *tp = ops;
  73         rc = 0;
  74 out:
  75         write_unlock(&cls_mod_lock);
  76         return rc;
  77 }
  78 EXPORT_SYMBOL(register_tcf_proto_ops);
  79
  80 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
  81 {
  82         struct tcf_proto_ops *t, **tp;
  83         int rc = -ENOENT;
  84
  85         write_lock(&cls_mod_lock);
  86         for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
  87                 if (t == ops)
  88                         break;
  89
  90         if (!t)
  91                 goto out;
  92         *tp = t->next;
  93         rc = 0;
  94 out:
  95         write_unlock(&cls_mod_lock);
  96         return rc;
  97 }
  98 EXPORT_SYMBOL(unregister_tcf_proto_ops);
  99
 100 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 101                           struct tcf_proto *tp, unsigned long fh, int event);
 102
 103
 104 /* Select new prio value from the range, managed by kernel. */
 105
 106 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 107 {
 108         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 109
 110         if (tp)
 111                 first = tp->prio-1;
 112
 113         return first;
 114 }
 115
 116 /* Add/change/delete/get a filter node */
 117
 118 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
 119 {
 120         struct net *net = skb->sk->sk_net;
 121         struct rtattr **tca;
 122         struct tcmsg *t;
 123         u32 protocol;
 124         u32 prio;
 125         u32 nprio;
 126         u32 parent;
 127         struct net_device *dev;
 128         struct Qdisc  *q;
 129         struct tcf_proto **back, **chain;
 130         struct tcf_proto *tp;
 131         struct tcf_proto_ops *tp_ops;
 132         const struct Qdisc_class_ops *cops;
 133         unsigned long cl;
 134         unsigned long fh;
 135         int err;
 136
 137         if (net != &init_net)
 138                 return -EINVAL;
 139
 140 replay:
 141         tca = arg;
 142         t = NLMSG_DATA(n);
 143         protocol = TC_H_MIN(t->tcm_info);
 144         prio = TC_H_MAJ(t->tcm_info);
 145         nprio = prio;
 146         parent = t->tcm_parent;
 147         cl = 0;
 148
 149         if (prio == 0) {
 150                 /* If no priority is given, user wants we allocated it. */
 151                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
 152                         return -ENOENT;
 153                 prio = TC_H_MAKE(0x80000000U, 0U);
 154         }
 155
 156         /* Find head of filter chain. */
 157
 158         /* Find link */
 159         dev = __dev_get_by_index(&init_net, t->tcm_ifindex);
 160         if (dev == NULL)
 161                 return -ENODEV;
 162
 163         /* Find qdisc */
 164         if (!parent) {
 165                 q = dev->qdisc_sleeping;
 166                 parent = q->handle;
 167         } else {
 168                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
 169                 if (q == NULL)
 170                         return -EINVAL;
 171         }
 172
 173         /* Is it classful? */
 174         if ((cops = q->ops->cl_ops) == NULL)
 175                 return -EINVAL;
 176
 177         /* Do we search for filter, attached to class? */
 178         if (TC_H_MIN(parent)) {
 179                 cl = cops->get(q, parent);
 180                 if (cl == 0)
 181                         return -ENOENT;
 182         }
 183
 184         /* And the last stroke */
 185         chain = cops->tcf_chain(q, cl);
 186         err = -EINVAL;
 187         if (chain == NULL)
 188                 goto errout;
 189
 190         /* Check the chain for existence of proto-tcf with this priority */
 191         for (back = chain; (tp=*back) != NULL; back = &tp->next) {
 192                 if (tp->prio >= prio) {
 193                         if (tp->prio == prio) {
 194                                 if (!nprio || (tp->protocol != protocol && protocol))
 195                                         goto errout;
 196                         } else
 197                                 tp = NULL;
 198                         break;
 199                 }
 200         }
 201
 202         if (tp == NULL) {
 203                 /* Proto-tcf does not exist, create new one */
 204
 205                 if (tca[TCA_KIND-1] == NULL || !protocol)
 206                         goto errout;
 207
 208                 err = -ENOENT;
 209                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
 210                         goto errout;
 211
 212
 213                 /* Create new proto tcf */
 214
 215                 err = -ENOBUFS;
 216                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 217                 if (tp == NULL)
 218                         goto errout;
 219                 err = -EINVAL;
 220                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]);
 221                 if (tp_ops == NULL) {
 222 #ifdef CONFIG_KMOD
 223                         struct rtattr *kind = tca[TCA_KIND-1];
 224                         char name[IFNAMSIZ];
 225
 226                         if (kind != NULL &&
 227                             rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
 228                                 rtnl_unlock();
 229                                 request_module("cls_%s", name);
 230                                 rtnl_lock();
 231                                 tp_ops = tcf_proto_lookup_ops(kind);
 232                                 /* We dropped the RTNL semaphore in order to
 233                                  * perform the module load.  So, even if we
 234                                  * succeeded in loading the module we have to
 235                                  * replay the request.  We indicate this using
 236                                  * -EAGAIN.
 237                                  */
 238                                 if (tp_ops != NULL) {
 239                                         module_put(tp_ops->owner);
 240                                         err = -EAGAIN;
 241                                 }
 242                         }
 243 #endif
 244                         kfree(tp);
 245                         goto errout;
 246                 }
 247                 tp->ops = tp_ops;
 248                 tp->protocol = protocol;
 249                 tp->prio = nprio ? : tcf_auto_prio(*back);
 250                 tp->q = q;
 251                 tp->classify = tp_ops->classify;
 252                 tp->classid = parent;
 253
 254                 err = tp_ops->init(tp);
 255                 if (err != 0) {
 256                         module_put(tp_ops->owner);
 257                         kfree(tp);
 258                         goto errout;
 259                 }
 260
 261                 qdisc_lock_tree(dev);
 262                 tp->next = *back;
 263                 *back = tp;
 264                 qdisc_unlock_tree(dev);
 265
 266         } else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind))
 267                 goto errout;
 268
 269         fh = tp->ops->get(tp, t->tcm_handle);
 270
 271         if (fh == 0) {
 272                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
 273                         qdisc_lock_tree(dev);
 274                         *back = tp->next;
 275                         qdisc_unlock_tree(dev);
 276
 277                         tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 278                         tcf_destroy(tp);
 279                         err = 0;
 280                         goto errout;
 281                 }
 282
 283                 err = -ENOENT;
 284                 if (n->nlmsg_type != RTM_NEWTFILTER ||
 285                     !(n->nlmsg_flags & NLM_F_CREATE))
 286                         goto errout;
 287         } else {
 288                 switch (n->nlmsg_type) {
 289                 case RTM_NEWTFILTER:
 290                         err = -EEXIST;
 291                         if (n->nlmsg_flags & NLM_F_EXCL)
 292                                 goto errout;
 293                         break;
 294                 case RTM_DELTFILTER:
 295                         err = tp->ops->delete(tp, fh);
 296                         if (err == 0)
 297                                 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 298                         goto errout;
 299                 case RTM_GETTFILTER:
 300                         err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 301                         goto errout;
 302                 default:
 303                         err = -EINVAL;
 304                         goto errout;
 305                 }
 306         }
 307
 308         err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
 309         if (err == 0)
 310                 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 311
 312 errout:
 313         if (cl)
 314                 cops->put(q, cl);
 315         if (err == -EAGAIN)
 316                 /* Replay the request. */
 317                 goto replay;
 318         return err;
 319 }
 320
 321 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
 322                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
 323 {
 324         struct tcmsg *tcm;
 325         struct nlmsghdr  *nlh;
 326         unsigned char *b = skb_tail_pointer(skb);
 327
 328         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
 329         tcm = NLMSG_DATA(nlh);
 330         tcm->tcm_family = AF_UNSPEC;
 331         tcm->tcm__pad1 = 0;
 332         tcm->tcm__pad1 = 0;
 333         tcm->tcm_ifindex = tp->q->dev->ifindex;
 334         tcm->tcm_parent = tp->classid;
 335         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
 336         RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind);
 337         tcm->tcm_handle = fh;
 338         if (RTM_DELTFILTER != event) {
 339                 tcm->tcm_handle = 0;
 340                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
 341                         goto rtattr_failure;
 342         }
 343         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 344         return skb->len;
 345
 346 nlmsg_failure:
 347 rtattr_failure:
 348         nlmsg_trim(skb, b);
 349         return -1;
 350 }
 351
 352 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 353                           struct tcf_proto *tp, unsigned long fh, int event)
 354 {
 355         struct sk_buff *skb;
 356         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
 357
 358         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 359         if (!skb)
 360                 return -ENOBUFS;
 361
 362         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
 363                 kfree_skb(skb);
 364                 return -EINVAL;
 365         }
 366
 367         return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
 368                               n->nlmsg_flags & NLM_F_ECHO);
 369 }
 370
 371 struct tcf_dump_args {
 372         struct tcf_walker w;
 373         struct sk_buff *skb;
 374         struct netlink_callback *cb;
 375 };
 376
 377 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
 378                          struct tcf_walker *arg)
 379 {
 380         struct tcf_dump_args *a = (void *)arg;
 381
 382         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
 383                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
 384 }
 385
 386 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
 387 {
 388         struct net *net = skb->sk->sk_net;
 389         int t;
 390         int s_t;
 391         struct net_device *dev;
 392         struct Qdisc *q;
 393         struct tcf_proto *tp, **chain;
 394         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
 395         unsigned long cl = 0;
 396         const struct Qdisc_class_ops *cops;
 397         struct tcf_dump_args arg;
 398
 399         if (net != &init_net)
 400                 return 0;
 401
 402         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
 403                 return skb->len;
 404         if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
 405                 return skb->len;
 406
 407         if (!tcm->tcm_parent)
 408                 q = dev->qdisc_sleeping;
 409         else
 410                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
 411         if (!q)
 412                 goto out;
 413         if ((cops = q->ops->cl_ops) == NULL)
 414                 goto errout;
 415         if (TC_H_MIN(tcm->tcm_parent)) {
 416                 cl = cops->get(q, tcm->tcm_parent);
 417                 if (cl == 0)
 418                         goto errout;
 419         }
 420         chain = cops->tcf_chain(q, cl);
 421         if (chain == NULL)
 422                 goto errout;
 423
 424         s_t = cb->args[0];
 425
 426         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
 427                 if (t < s_t) continue;
 428                 if (TC_H_MAJ(tcm->tcm_info) &&
 429                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
 430                         continue;
 431                 if (TC_H_MIN(tcm->tcm_info) &&
 432                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
 433                         continue;
 434                 if (t > s_t)
 435                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
 436                 if (cb->args[1] == 0) {
 437                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
 438                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
 439                                           RTM_NEWTFILTER) <= 0)
 440                                 break;
 441
 442                         cb->args[1] = 1;
 443                 }
 444                 if (tp->ops->walk == NULL)
 445                         continue;
 446                 arg.w.fn = tcf_node_dump;
 447                 arg.skb = skb;
 448                 arg.cb = cb;
 449                 arg.w.stop = 0;
 450                 arg.w.skip = cb->args[1]-1;
 451                 arg.w.count = 0;
 452                 tp->ops->walk(tp, &arg.w);
 453                 cb->args[1] = arg.w.count+1;
 454                 if (arg.w.stop)
 455                         break;
 456         }
 457
 458         cb->args[0] = t;
 459
 460 errout:
 461         if (cl)
 462                 cops->put(q, cl);
 463 out:
 464         dev_put(dev);
 465         return skb->len;
 466 }
 467
 468 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
 469 {
 470 #ifdef CONFIG_NET_CLS_ACT
 471         if (exts->action) {
 472                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
 473                 exts->action = NULL;
 474         }
 475 #endif
 476 }
 477 EXPORT_SYMBOL(tcf_exts_destroy);
 478
 479 int tcf_exts_validate(struct tcf_proto *tp, struct rtattr **tb,
 480                   struct rtattr *rate_tlv, struct tcf_exts *exts,
 481                   struct tcf_ext_map *map)
 482 {
 483         memset(exts, 0, sizeof(*exts));
 484
 485 #ifdef CONFIG_NET_CLS_ACT
 486         {
 487                 int err;
 488                 struct tc_action *act;
 489
 490                 if (map->police && tb[map->police-1]) {
 491                         act = tcf_action_init_1(tb[map->police-1], rate_tlv,
 492                                                 "police", TCA_ACT_NOREPLACE,
 493                                                 TCA_ACT_BIND, &err);
 494                         if (act == NULL)
 495                                 return err;
 496
 497                         act->type = TCA_OLD_COMPAT;
 498                         exts->action = act;
 499                 } else if (map->action && tb[map->action-1]) {
 500                         act = tcf_action_init(tb[map->action-1], rate_tlv, NULL,
 501                                 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
 502                         if (act == NULL)
 503                                 return err;
 504
 505                         exts->action = act;
 506                 }
 507         }
 508 #else
 509         if ((map->action && tb[map->action-1]) ||
 510             (map->police && tb[map->police-1]))
 511                 return -EOPNOTSUPP;
 512 #endif
 513
 514         return 0;
 515 }
 516 EXPORT_SYMBOL(tcf_exts_validate);
 517
 518 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
 519                      struct tcf_exts *src)
 520 {
 521 #ifdef CONFIG_NET_CLS_ACT
 522         if (src->action) {
 523                 struct tc_action *act;
 524                 tcf_tree_lock(tp);
 525                 act = xchg(&dst->action, src->action);
 526                 tcf_tree_unlock(tp);
 527                 if (act)
 528                         tcf_action_destroy(act, TCA_ACT_UNBIND);
 529         }
 530 #endif
 531 }
 532 EXPORT_SYMBOL(tcf_exts_change);
 533
 534 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
 535               struct tcf_ext_map *map)
 536 {
 537 #ifdef CONFIG_NET_CLS_ACT
 538         if (map->action && exts->action) {
 539                 /*
 540                  * again for backward compatible mode - we want
 541                  * to work with both old and new modes of entering
 542                  * tc data even if iproute2  was newer - jhs
 543                  */
 544                 struct rtattr *p_rta = (struct rtattr *)skb_tail_pointer(skb);
 545
 546                 if (exts->action->type != TCA_OLD_COMPAT) {
 547                         RTA_PUT(skb, map->action, 0, NULL);
 548                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
 549                                 goto rtattr_failure;
 550                         p_rta->rta_len = skb_tail_pointer(skb) - (u8 *)p_rta;
 551                 } else if (map->police) {
 552                         RTA_PUT(skb, map->police, 0, NULL);
 553                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
 554                                 goto rtattr_failure;
 555                         p_rta->rta_len = skb_tail_pointer(skb) - (u8 *)p_rta;
 556                 }
 557         }
 558 #endif
 559         return 0;
 560 rtattr_failure: __attribute__ ((unused))
 561         return -1;
 562 }
 563 EXPORT_SYMBOL(tcf_exts_dump);
 564
 565
 566 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
 567                         struct tcf_ext_map *map)
 568 {
 569 #ifdef CONFIG_NET_CLS_ACT
 570         if (exts->action)
 571                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
 572                         goto rtattr_failure;
 573 #endif
 574         return 0;
 575 rtattr_failure: __attribute__ ((unused))
 576         return -1;
 577 }
 578 EXPORT_SYMBOL(tcf_exts_dump_stats);
 579
 580 static int __init tc_filter_init(void)
 581 {
 582         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
 583         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
 584         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
 585                                                  tc_dump_tfilter);
 586
 587         return 0;
 588 }
 589
 590 subsys_initcall(tc_filter_init);