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 <linux/err.h>
  27 #include <net/net_namespace.h>
  28 #include <net/sock.h>
  29 #include <net/netlink.h>
  30 #include <net/pkt_sched.h>
  31 #include <net/pkt_cls.h>
  32
  33 /* The list of all installed classifier types */
  34
  35 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
  36
  37 /* Protects list of registered TC modules. It is pure SMP lock. */
  38 static DEFINE_RWLOCK(cls_mod_lock);
  39
  40 /* Find classifier type by string name */
  41
  42 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
  43 {
  44         struct tcf_proto_ops *t = NULL;
  45
  46         if (kind) {
  47                 read_lock(&cls_mod_lock);
  48                 for (t = tcf_proto_base; t; t = t->next) {
  49                         if (nla_strcmp(kind, t->kind) == 0) {
  50                                 if (!try_module_get(t->owner))
  51                                         t = NULL;
  52                                 break;
  53                         }
  54                 }
  55                 read_unlock(&cls_mod_lock);
  56         }
  57         return t;
  58 }
  59
  60 /* Register(unregister) new classifier type */
  61
  62 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  63 {
  64         struct tcf_proto_ops *t, **tp;
  65         int rc = -EEXIST;
  66
  67         write_lock(&cls_mod_lock);
  68         for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
  69                 if (!strcmp(ops->kind, t->kind))
  70                         goto out;
  71
  72         ops->next = NULL;
  73         *tp = ops;
  74         rc = 0;
  75 out:
  76         write_unlock(&cls_mod_lock);
  77         return rc;
  78 }
  79 EXPORT_SYMBOL(register_tcf_proto_ops);
  80
  81 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
  82 {
  83         struct tcf_proto_ops *t, **tp;
  84         int rc = -ENOENT;
  85
  86         write_lock(&cls_mod_lock);
  87         for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
  88                 if (t == ops)
  89                         break;
  90
  91         if (!t)
  92                 goto out;
  93         *tp = t->next;
  94         rc = 0;
  95 out:
  96         write_unlock(&cls_mod_lock);
  97         return rc;
  98 }
  99 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 100
 101 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 102                           struct tcf_proto *tp, unsigned long fh, int event);
 103
 104
 105 /* Select new prio value from the range, managed by kernel. */
 106
 107 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 108 {
 109         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 110
 111         if (tp)
 112                 first = tp->prio-1;
 113
 114         return first;
 115 }
 116
 117 /* Add/change/delete/get a filter node */
 118
 119 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
 120 {
 121         struct net *net = skb->sk->sk_net;
 122         struct nlattr *tca[TCA_MAX + 1];
 123         struct tcmsg *t;
 124         u32 protocol;
 125         u32 prio;
 126         u32 nprio;
 127         u32 parent;
 128         struct net_device *dev;
 129         struct Qdisc  *q;
 130         struct tcf_proto **back, **chain;
 131         struct tcf_proto *tp;
 132         struct tcf_proto_ops *tp_ops;
 133         const struct Qdisc_class_ops *cops;
 134         unsigned long cl;
 135         unsigned long fh;
 136         int err;
 137
 138         if (net != &init_net)
 139                 return -EINVAL;
 140
 141 replay:
 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         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
 164         if (err < 0)
 165                 return err;
 166
 167         /* Find qdisc */
 168         if (!parent) {
 169                 q = dev->qdisc_sleeping;
 170                 parent = q->handle;
 171         } else {
 172                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
 173                 if (q == NULL)
 174                         return -EINVAL;
 175         }
 176
 177         /* Is it classful? */
 178         if ((cops = q->ops->cl_ops) == NULL)
 179                 return -EINVAL;
 180
 181         /* Do we search for filter, attached to class? */
 182         if (TC_H_MIN(parent)) {
 183                 cl = cops->get(q, parent);
 184                 if (cl == 0)
 185                         return -ENOENT;
 186         }
 187
 188         /* And the last stroke */
 189         chain = cops->tcf_chain(q, cl);
 190         err = -EINVAL;
 191         if (chain == NULL)
 192                 goto errout;
 193
 194         /* Check the chain for existence of proto-tcf with this priority */
 195         for (back = chain; (tp=*back) != NULL; back = &tp->next) {
 196                 if (tp->prio >= prio) {
 197                         if (tp->prio == prio) {
 198                                 if (!nprio || (tp->protocol != protocol && protocol))
 199                                         goto errout;
 200                         } else
 201                                 tp = NULL;
 202                         break;
 203                 }
 204         }
 205
 206         if (tp == NULL) {
 207                 /* Proto-tcf does not exist, create new one */
 208
 209                 if (tca[TCA_KIND] == NULL || !protocol)
 210                         goto errout;
 211
 212                 err = -ENOENT;
 213                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
 214                         goto errout;
 215
 216
 217                 /* Create new proto tcf */
 218
 219                 err = -ENOBUFS;
 220                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 221                 if (tp == NULL)
 222                         goto errout;
 223                 err = -EINVAL;
 224                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
 225                 if (tp_ops == NULL) {
 226 #ifdef CONFIG_KMOD
 227                         struct nlattr *kind = tca[TCA_KIND];
 228                         char name[IFNAMSIZ];
 229
 230                         if (kind != NULL &&
 231                             nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
 232                                 rtnl_unlock();
 233                                 request_module("cls_%s", name);
 234                                 rtnl_lock();
 235                                 tp_ops = tcf_proto_lookup_ops(kind);
 236                                 /* We dropped the RTNL semaphore in order to
 237                                  * perform the module load.  So, even if we
 238                                  * succeeded in loading the module we have to
 239                                  * replay the request.  We indicate this using
 240                                  * -EAGAIN.
 241                                  */
 242                                 if (tp_ops != NULL) {
 243                                         module_put(tp_ops->owner);
 244                                         err = -EAGAIN;
 245                                 }
 246                         }
 247 #endif
 248                         kfree(tp);
 249                         goto errout;
 250                 }
 251                 tp->ops = tp_ops;
 252                 tp->protocol = protocol;
 253                 tp->prio = nprio ? : tcf_auto_prio(*back);
 254                 tp->q = q;
 255                 tp->classify = tp_ops->classify;
 256                 tp->classid = parent;
 257
 258                 err = tp_ops->init(tp);
 259                 if (err != 0) {
 260                         module_put(tp_ops->owner);
 261                         kfree(tp);
 262                         goto errout;
 263                 }
 264
 265                 qdisc_lock_tree(dev);
 266                 tp->next = *back;
 267                 *back = tp;
 268                 qdisc_unlock_tree(dev);
 269
 270         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
 271                 goto errout;
 272
 273         fh = tp->ops->get(tp, t->tcm_handle);
 274
 275         if (fh == 0) {
 276                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
 277                         qdisc_lock_tree(dev);
 278                         *back = tp->next;
 279                         qdisc_unlock_tree(dev);
 280
 281                         tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 282                         tcf_destroy(tp);
 283                         err = 0;
 284                         goto errout;
 285                 }
 286
 287                 err = -ENOENT;
 288                 if (n->nlmsg_type != RTM_NEWTFILTER ||
 289                     !(n->nlmsg_flags & NLM_F_CREATE))
 290                         goto errout;
 291         } else {
 292                 switch (n->nlmsg_type) {
 293                 case RTM_NEWTFILTER:
 294                         err = -EEXIST;
 295                         if (n->nlmsg_flags & NLM_F_EXCL)
 296                                 goto errout;
 297                         break;
 298                 case RTM_DELTFILTER:
 299                         err = tp->ops->delete(tp, fh);
 300                         if (err == 0)
 301                                 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 302                         goto errout;
 303                 case RTM_GETTFILTER:
 304                         err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 305                         goto errout;
 306                 default:
 307                         err = -EINVAL;
 308                         goto errout;
 309                 }
 310         }
 311
 312         err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
 313         if (err == 0)
 314                 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 315
 316 errout:
 317         if (cl)
 318                 cops->put(q, cl);
 319         if (err == -EAGAIN)
 320                 /* Replay the request. */
 321                 goto replay;
 322         return err;
 323 }
 324
 325 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
 326                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
 327 {
 328         struct tcmsg *tcm;
 329         struct nlmsghdr  *nlh;
 330         unsigned char *b = skb_tail_pointer(skb);
 331
 332         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
 333         tcm = NLMSG_DATA(nlh);
 334         tcm->tcm_family = AF_UNSPEC;
 335         tcm->tcm__pad1 = 0;
 336         tcm->tcm__pad1 = 0;
 337         tcm->tcm_ifindex = tp->q->dev->ifindex;
 338         tcm->tcm_parent = tp->classid;
 339         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
 340         NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
 341         tcm->tcm_handle = fh;
 342         if (RTM_DELTFILTER != event) {
 343                 tcm->tcm_handle = 0;
 344                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
 345                         goto nla_put_failure;
 346         }
 347         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 348         return skb->len;
 349
 350 nlmsg_failure:
 351 nla_put_failure:
 352         nlmsg_trim(skb, b);
 353         return -1;
 354 }
 355
 356 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 357                           struct tcf_proto *tp, unsigned long fh, int event)
 358 {
 359         struct sk_buff *skb;
 360         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
 361
 362         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 363         if (!skb)
 364                 return -ENOBUFS;
 365
 366         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
 367                 kfree_skb(skb);
 368                 return -EINVAL;
 369         }
 370
 371         return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
 372                               n->nlmsg_flags & NLM_F_ECHO);
 373 }
 374
 375 struct tcf_dump_args {
 376         struct tcf_walker w;
 377         struct sk_buff *skb;
 378         struct netlink_callback *cb;
 379 };
 380
 381 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
 382                          struct tcf_walker *arg)
 383 {
 384         struct tcf_dump_args *a = (void *)arg;
 385
 386         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
 387                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
 388 }
 389
 390 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
 391 {
 392         struct net *net = skb->sk->sk_net;
 393         int t;
 394         int s_t;
 395         struct net_device *dev;
 396         struct Qdisc *q;
 397         struct tcf_proto *tp, **chain;
 398         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
 399         unsigned long cl = 0;
 400         const struct Qdisc_class_ops *cops;
 401         struct tcf_dump_args arg;
 402
 403         if (net != &init_net)
 404                 return 0;
 405
 406         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
 407                 return skb->len;
 408         if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
 409                 return skb->len;
 410
 411         if (!tcm->tcm_parent)
 412                 q = dev->qdisc_sleeping;
 413         else
 414                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
 415         if (!q)
 416                 goto out;
 417         if ((cops = q->ops->cl_ops) == NULL)
 418                 goto errout;
 419         if (TC_H_MIN(tcm->tcm_parent)) {
 420                 cl = cops->get(q, tcm->tcm_parent);
 421                 if (cl == 0)
 422                         goto errout;
 423         }
 424         chain = cops->tcf_chain(q, cl);
 425         if (chain == NULL)
 426                 goto errout;
 427
 428         s_t = cb->args[0];
 429
 430         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
 431                 if (t < s_t) continue;
 432                 if (TC_H_MAJ(tcm->tcm_info) &&
 433                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
 434                         continue;
 435                 if (TC_H_MIN(tcm->tcm_info) &&
 436                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
 437                         continue;
 438                 if (t > s_t)
 439                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
 440                 if (cb->args[1] == 0) {
 441                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
 442                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
 443                                           RTM_NEWTFILTER) <= 0)
 444                                 break;
 445
 446                         cb->args[1] = 1;
 447                 }
 448                 if (tp->ops->walk == NULL)
 449                         continue;
 450                 arg.w.fn = tcf_node_dump;
 451                 arg.skb = skb;
 452                 arg.cb = cb;
 453                 arg.w.stop = 0;
 454                 arg.w.skip = cb->args[1]-1;
 455                 arg.w.count = 0;
 456                 tp->ops->walk(tp, &arg.w);
 457                 cb->args[1] = arg.w.count+1;
 458                 if (arg.w.stop)
 459                         break;
 460         }
 461
 462         cb->args[0] = t;
 463
 464 errout:
 465         if (cl)
 466                 cops->put(q, cl);
 467 out:
 468         dev_put(dev);
 469         return skb->len;
 470 }
 471
 472 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
 473 {
 474 #ifdef CONFIG_NET_CLS_ACT
 475         if (exts->action) {
 476                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
 477                 exts->action = NULL;
 478         }
 479 #endif
 480 }
 481 EXPORT_SYMBOL(tcf_exts_destroy);
 482
 483 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
 484                   struct nlattr *rate_tlv, struct tcf_exts *exts,
 485                   const struct tcf_ext_map *map)
 486 {
 487         memset(exts, 0, sizeof(*exts));
 488
 489 #ifdef CONFIG_NET_CLS_ACT
 490         {
 491                 struct tc_action *act;
 492
 493                 if (map->police && tb[map->police]) {
 494                         act = tcf_action_init_1(tb[map->police], rate_tlv,
 495                                                 "police", TCA_ACT_NOREPLACE,
 496                                                 TCA_ACT_BIND);
 497                         if (IS_ERR(act))
 498                                 return PTR_ERR(act);
 499
 500                         act->type = TCA_OLD_COMPAT;
 501                         exts->action = act;
 502                 } else if (map->action && tb[map->action]) {
 503                         act = tcf_action_init(tb[map->action], rate_tlv, NULL,
 504                                               TCA_ACT_NOREPLACE, TCA_ACT_BIND);
 505                         if (IS_ERR(act))
 506                                 return PTR_ERR(act);
 507
 508                         exts->action = act;
 509                 }
 510         }
 511 #else
 512         if ((map->action && tb[map->action]) ||
 513             (map->police && tb[map->police]))
 514                 return -EOPNOTSUPP;
 515 #endif
 516
 517         return 0;
 518 }
 519 EXPORT_SYMBOL(tcf_exts_validate);
 520
 521 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
 522                      struct tcf_exts *src)
 523 {
 524 #ifdef CONFIG_NET_CLS_ACT
 525         if (src->action) {
 526                 struct tc_action *act;
 527                 tcf_tree_lock(tp);
 528                 act = xchg(&dst->action, src->action);
 529                 tcf_tree_unlock(tp);
 530                 if (act)
 531                         tcf_action_destroy(act, TCA_ACT_UNBIND);
 532         }
 533 #endif
 534 }
 535 EXPORT_SYMBOL(tcf_exts_change);
 536
 537 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
 538                   const struct tcf_ext_map *map)
 539 {
 540 #ifdef CONFIG_NET_CLS_ACT
 541         if (map->action && exts->action) {
 542                 /*
 543                  * again for backward compatible mode - we want
 544                  * to work with both old and new modes of entering
 545                  * tc data even if iproute2  was newer - jhs
 546                  */
 547                 struct nlattr *nest;
 548
 549                 if (exts->action->type != TCA_OLD_COMPAT) {
 550                         nest = nla_nest_start(skb, map->action);
 551                         if (nest == NULL)
 552                                 goto nla_put_failure;
 553                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
 554                                 goto nla_put_failure;
 555                         nla_nest_end(skb, nest);
 556                 } else if (map->police) {
 557                         nest = nla_nest_start(skb, map->police);
 558                         if (nest == NULL)
 559                                 goto nla_put_failure;
 560                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
 561                                 goto nla_put_failure;
 562                         nla_nest_end(skb, nest);
 563                 }
 564         }
 565 #endif
 566         return 0;
 567 nla_put_failure: __attribute__ ((unused))
 568         return -1;
 569 }
 570 EXPORT_SYMBOL(tcf_exts_dump);
 571
 572
 573 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
 574                         const struct tcf_ext_map *map)
 575 {
 576 #ifdef CONFIG_NET_CLS_ACT
 577         if (exts->action)
 578                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
 579                         goto nla_put_failure;
 580 #endif
 581         return 0;
 582 nla_put_failure: __attribute__ ((unused))
 583         return -1;
 584 }
 585 EXPORT_SYMBOL(tcf_exts_dump_stats);
 586
 587 static int __init tc_filter_init(void)
 588 {
 589         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
 590         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
 591         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
 592                                                  tc_dump_tfilter);
 593
 594         return 0;
 595 }
 596
 597 subsys_initcall(tc_filter_init);