net/sched/cls_route.c

   1 /*
   2  * net/sched/cls_route.c        ROUTE4 classifier.
   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
  12 #include <linux/module.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/kernel.h>
  16 #include <linux/string.h>
  17 #include <linux/errno.h>
  18 #include <linux/skbuff.h>
  19 #include <net/dst.h>
  20 #include <net/route.h>
  21 #include <net/netlink.h>
  22 #include <net/act_api.h>
  23 #include <net/pkt_cls.h>
  24
  25 /*
  26  * 1. For now we assume that route tags < 256.
  27  *    It allows to use direct table lookups, instead of hash tables.
  28  * 2. For now we assume that "from TAG" and "fromdev DEV" statements
  29  *    are mutually  exclusive.
  30  * 3. "to TAG from ANY" has higher priority, than "to ANY from XXX"
  31  */
  32 struct route4_fastmap {
  33         struct route4_filter            *filter;
  34         u32                             id;
  35         int                             iif;
  36 };
  37
  38 struct route4_head {
  39         struct route4_fastmap           fastmap[16];
  40         struct route4_bucket __rcu      *table[256 + 1];
  41         struct rcu_head                 rcu;
  42 };
  43
  44 struct route4_bucket {
  45         /* 16 FROM buckets + 16 IIF buckets + 1 wildcard bucket */
  46         struct route4_filter __rcu      *ht[16 + 16 + 1];
  47         struct rcu_head                 rcu;
  48 };
  49
  50 struct route4_filter {
  51         struct route4_filter __rcu      *next;
  52         u32                     id;
  53         int                     iif;
  54
  55         struct tcf_result       res;
  56         struct tcf_exts         exts;
  57         u32                     handle;
  58         struct route4_bucket    *bkt;
  59         struct tcf_proto        *tp;
  60         struct rcu_head         rcu;
  61 };
  62
  63 #define ROUTE4_FAILURE ((struct route4_filter *)(-1L))
  64
  65 static inline int route4_fastmap_hash(u32 id, int iif)
  66 {
  67         return id & 0xF;
  68 }
  69
  70 static DEFINE_SPINLOCK(fastmap_lock);
  71 static void
  72 route4_reset_fastmap(struct route4_head *head)
  73 {
  74         spin_lock_bh(&fastmap_lock);
  75         memset(head->fastmap, 0, sizeof(head->fastmap));
  76         spin_unlock_bh(&fastmap_lock);
  77 }
  78
  79 static void
  80 route4_set_fastmap(struct route4_head *head, u32 id, int iif,
  81                    struct route4_filter *f)
  82 {
  83         int h = route4_fastmap_hash(id, iif);
  84
  85         /* fastmap updates must look atomic to aling id, iff, filter */
  86         spin_lock_bh(&fastmap_lock);
  87         head->fastmap[h].id = id;
  88         head->fastmap[h].iif = iif;
  89         head->fastmap[h].filter = f;
  90         spin_unlock_bh(&fastmap_lock);
  91 }
  92
  93 static inline int route4_hash_to(u32 id)
  94 {
  95         return id & 0xFF;
  96 }
  97
  98 static inline int route4_hash_from(u32 id)
  99 {
 100         return (id >> 16) & 0xF;
 101 }
 102
 103 static inline int route4_hash_iif(int iif)
 104 {
 105         return 16 + ((iif >> 16) & 0xF);
 106 }
 107
 108 static inline int route4_hash_wild(void)
 109 {
 110         return 32;
 111 }
 112
 113 #define ROUTE4_APPLY_RESULT()                                   \
 114 {                                                               \
 115         *res = f->res;                                          \
 116         if (tcf_exts_is_available(&f->exts)) {                  \
 117                 int r = tcf_exts_exec(skb, &f->exts, res);      \
 118                 if (r < 0) {                                    \
 119                         dont_cache = 1;                         \
 120                         continue;                               \
 121                 }                                               \
 122                 return r;                                       \
 123         } else if (!dont_cache)                                 \
 124                 route4_set_fastmap(head, id, iif, f);           \
 125         return 0;                                               \
 126 }
 127
 128 static int route4_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 129                            struct tcf_result *res)
 130 {
 131         struct route4_head *head = rcu_dereference_bh(tp->root);
 132         struct dst_entry *dst;
 133         struct route4_bucket *b;
 134         struct route4_filter *f;
 135         u32 id, h;
 136         int iif, dont_cache = 0;
 137
 138         dst = skb_dst(skb);
 139         if (!dst)
 140                 goto failure;
 141
 142         id = dst->tclassid;
 143         if (head == NULL)
 144                 goto old_method;
 145
 146         iif = inet_iif(skb);
 147
 148         h = route4_fastmap_hash(id, iif);
 149
 150         spin_lock(&fastmap_lock);
 151         if (id == head->fastmap[h].id &&
 152             iif == head->fastmap[h].iif &&
 153             (f = head->fastmap[h].filter) != NULL) {
 154                 if (f == ROUTE4_FAILURE) {
 155                         spin_unlock(&fastmap_lock);
 156                         goto failure;
 157                 }
 158
 159                 *res = f->res;
 160                 spin_unlock(&fastmap_lock);
 161                 return 0;
 162         }
 163         spin_unlock(&fastmap_lock);
 164
 165         h = route4_hash_to(id);
 166
 167 restart:
 168         b = rcu_dereference_bh(head->table[h]);
 169         if (b) {
 170                 for (f = rcu_dereference_bh(b->ht[route4_hash_from(id)]);
 171                      f;
 172                      f = rcu_dereference_bh(f->next))
 173                         if (f->id == id)
 174                                 ROUTE4_APPLY_RESULT();
 175
 176                 for (f = rcu_dereference_bh(b->ht[route4_hash_iif(iif)]);
 177                      f;
 178                      f = rcu_dereference_bh(f->next))
 179                         if (f->iif == iif)
 180                                 ROUTE4_APPLY_RESULT();
 181
 182                 for (f = rcu_dereference_bh(b->ht[route4_hash_wild()]);
 183                      f;
 184                      f = rcu_dereference_bh(f->next))
 185                         ROUTE4_APPLY_RESULT();
 186         }
 187         if (h < 256) {
 188                 h = 256;
 189                 id &= ~0xFFFF;
 190                 goto restart;
 191         }
 192
 193         if (!dont_cache)
 194                 route4_set_fastmap(head, id, iif, ROUTE4_FAILURE);
 195 failure:
 196         return -1;
 197
 198 old_method:
 199         if (id && (TC_H_MAJ(id) == 0 ||
 200                    !(TC_H_MAJ(id^tp->q->handle)))) {
 201                 res->classid = id;
 202                 res->class = 0;
 203                 return 0;
 204         }
 205         return -1;
 206 }
 207
 208 static inline u32 to_hash(u32 id)
 209 {
 210         u32 h = id & 0xFF;
 211
 212         if (id & 0x8000)
 213                 h += 256;
 214         return h;
 215 }
 216
 217 static inline u32 from_hash(u32 id)
 218 {
 219         id &= 0xFFFF;
 220         if (id == 0xFFFF)
 221                 return 32;
 222         if (!(id & 0x8000)) {
 223                 if (id > 255)
 224                         return 256;
 225                 return id & 0xF;
 226         }
 227         return 16 + (id & 0xF);
 228 }
 229
 230 static unsigned long route4_get(struct tcf_proto *tp, u32 handle)
 231 {
 232         struct route4_head *head = rtnl_dereference(tp->root);
 233         struct route4_bucket *b;
 234         struct route4_filter *f;
 235         unsigned int h1, h2;
 236
 237         if (!head)
 238                 return 0;
 239
 240         h1 = to_hash(handle);
 241         if (h1 > 256)
 242                 return 0;
 243
 244         h2 = from_hash(handle >> 16);
 245         if (h2 > 32)
 246                 return 0;
 247
 248         b = rtnl_dereference(head->table[h1]);
 249         if (b) {
 250                 for (f = rtnl_dereference(b->ht[h2]);
 251                      f;
 252                      f = rtnl_dereference(f->next))
 253                         if (f->handle == handle)
 254                                 return (unsigned long)f;
 255         }
 256         return 0;
 257 }
 258
 259 static int route4_init(struct tcf_proto *tp)
 260 {
 261         struct route4_head *head;
 262
 263         head = kzalloc(sizeof(struct route4_head), GFP_KERNEL);
 264         if (head == NULL)
 265                 return -ENOBUFS;
 266
 267         rcu_assign_pointer(tp->root, head);
 268         return 0;
 269 }
 270
 271 static void
 272 route4_delete_filter(struct rcu_head *head)
 273 {
 274         struct route4_filter *f = container_of(head, struct route4_filter, rcu);
 275
 276         tcf_exts_destroy(&f->exts);
 277         kfree(f);
 278 }
 279
 280 static bool route4_destroy(struct tcf_proto *tp, bool force)
 281 {
 282         struct route4_head *head = rtnl_dereference(tp->root);
 283         int h1, h2;
 284
 285         if (head == NULL)
 286                 return true;
 287
 288         if (!force) {
 289                 for (h1 = 0; h1 <= 256; h1++) {
 290                         if (rcu_access_pointer(head->table[h1]))
 291                                 return false;
 292                 }
 293         }
 294
 295         for (h1 = 0; h1 <= 256; h1++) {
 296                 struct route4_bucket *b;
 297
 298                 b = rtnl_dereference(head->table[h1]);
 299                 if (b) {
 300                         for (h2 = 0; h2 <= 32; h2++) {
 301                                 struct route4_filter *f;
 302
 303                                 while ((f = rtnl_dereference(b->ht[h2])) != NULL) {
 304                                         struct route4_filter *next;
 305
 306                                         next = rtnl_dereference(f->next);
 307                                         RCU_INIT_POINTER(b->ht[h2], next);
 308                                         tcf_unbind_filter(tp, &f->res);
 309                                         call_rcu(&f->rcu, route4_delete_filter);
 310                                 }
 311                         }
 312                         RCU_INIT_POINTER(head->table[h1], NULL);
 313                         kfree_rcu(b, rcu);
 314                 }
 315         }
 316         RCU_INIT_POINTER(tp->root, NULL);
 317         kfree_rcu(head, rcu);
 318         return true;
 319 }
 320
 321 static int route4_delete(struct tcf_proto *tp, unsigned long arg)
 322 {
 323         struct route4_head *head = rtnl_dereference(tp->root);
 324         struct route4_filter *f = (struct route4_filter *)arg;
 325         struct route4_filter __rcu **fp;
 326         struct route4_filter *nf;
 327         struct route4_bucket *b;
 328         unsigned int h = 0;
 329         int i;
 330
 331         if (!head || !f)
 332                 return -EINVAL;
 333
 334         h = f->handle;
 335         b = f->bkt;
 336
 337         fp = &b->ht[from_hash(h >> 16)];
 338         for (nf = rtnl_dereference(*fp); nf;
 339              fp = &nf->next, nf = rtnl_dereference(*fp)) {
 340                 if (nf == f) {
 341                         /* unlink it */
 342                         RCU_INIT_POINTER(*fp, rtnl_dereference(f->next));
 343
 344                         /* Remove any fastmap lookups that might ref filter
 345                          * notice we unlink'd the filter so we can't get it
 346                          * back in the fastmap.
 347                          */
 348                         route4_reset_fastmap(head);
 349
 350                         /* Delete it */
 351                         tcf_unbind_filter(tp, &f->res);
 352                         call_rcu(&f->rcu, route4_delete_filter);
 353
 354                         /* Strip RTNL protected tree */
 355                         for (i = 0; i <= 32; i++) {
 356                                 struct route4_filter *rt;
 357
 358                                 rt = rtnl_dereference(b->ht[i]);
 359                                 if (rt)
 360                                         return 0;
 361                         }
 362
 363                         /* OK, session has no flows */
 364                         RCU_INIT_POINTER(head->table[to_hash(h)], NULL);
 365                         kfree_rcu(b, rcu);
 366
 367                         return 0;
 368                 }
 369         }
 370         return 0;
 371 }
 372
 373 static const struct nla_policy route4_policy[TCA_ROUTE4_MAX + 1] = {
 374         [TCA_ROUTE4_CLASSID]    = { .type = NLA_U32 },
 375         [TCA_ROUTE4_TO]         = { .type = NLA_U32 },
 376         [TCA_ROUTE4_FROM]       = { .type = NLA_U32 },
 377         [TCA_ROUTE4_IIF]        = { .type = NLA_U32 },
 378 };
 379
 380 static int route4_set_parms(struct net *net, struct tcf_proto *tp,
 381                             unsigned long base, struct route4_filter *f,
 382                             u32 handle, struct route4_head *head,
 383                             struct nlattr **tb, struct nlattr *est, int new,
 384                             bool ovr)
 385 {
 386         u32 id = 0, to = 0, nhandle = 0x8000;
 387         struct route4_filter *fp;
 388         unsigned int h1;
 389         struct route4_bucket *b;
 390         struct tcf_exts e;
 391         int err;
 392
 393         err = tcf_exts_init(&e, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE);
 394         if (err < 0)
 395                 return err;
 396         err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
 397         if (err < 0)
 398                 goto errout;
 399
 400         err = -EINVAL;
 401         if (tb[TCA_ROUTE4_TO]) {
 402                 if (new && handle & 0x8000)
 403                         goto errout;
 404                 to = nla_get_u32(tb[TCA_ROUTE4_TO]);
 405                 if (to > 0xFF)
 406                         goto errout;
 407                 nhandle = to;
 408         }
 409
 410         if (tb[TCA_ROUTE4_FROM]) {
 411                 if (tb[TCA_ROUTE4_IIF])
 412                         goto errout;
 413                 id = nla_get_u32(tb[TCA_ROUTE4_FROM]);
 414                 if (id > 0xFF)
 415                         goto errout;
 416                 nhandle |= id << 16;
 417         } else if (tb[TCA_ROUTE4_IIF]) {
 418                 id = nla_get_u32(tb[TCA_ROUTE4_IIF]);
 419                 if (id > 0x7FFF)
 420                         goto errout;
 421                 nhandle |= (id | 0x8000) << 16;
 422         } else
 423                 nhandle |= 0xFFFF << 16;
 424
 425         if (handle && new) {
 426                 nhandle |= handle & 0x7F00;
 427                 if (nhandle != handle)
 428                         goto errout;
 429         }
 430
 431         h1 = to_hash(nhandle);
 432         b = rtnl_dereference(head->table[h1]);
 433         if (!b) {
 434                 err = -ENOBUFS;
 435                 b = kzalloc(sizeof(struct route4_bucket), GFP_KERNEL);
 436                 if (b == NULL)
 437                         goto errout;
 438
 439                 rcu_assign_pointer(head->table[h1], b);
 440         } else {
 441                 unsigned int h2 = from_hash(nhandle >> 16);
 442
 443                 err = -EEXIST;
 444                 for (fp = rtnl_dereference(b->ht[h2]);
 445                      fp;
 446                      fp = rtnl_dereference(fp->next))
 447                         if (fp->handle == f->handle)
 448                                 goto errout;
 449         }
 450
 451         if (tb[TCA_ROUTE4_TO])
 452                 f->id = to;
 453
 454         if (tb[TCA_ROUTE4_FROM])
 455                 f->id = to | id<<16;
 456         else if (tb[TCA_ROUTE4_IIF])
 457                 f->iif = id;
 458
 459         f->handle = nhandle;
 460         f->bkt = b;
 461         f->tp = tp;
 462
 463         if (tb[TCA_ROUTE4_CLASSID]) {
 464                 f->res.classid = nla_get_u32(tb[TCA_ROUTE4_CLASSID]);
 465                 tcf_bind_filter(tp, &f->res, base);
 466         }
 467
 468         tcf_exts_change(tp, &f->exts, &e);
 469
 470         return 0;
 471 errout:
 472         tcf_exts_destroy(&e);
 473         return err;
 474 }
 475
 476 static int route4_change(struct net *net, struct sk_buff *in_skb,
 477                        struct tcf_proto *tp, unsigned long base,
 478                        u32 handle,
 479                        struct nlattr **tca,
 480                        unsigned long *arg, bool ovr)
 481 {
 482         struct route4_head *head = rtnl_dereference(tp->root);
 483         struct route4_filter __rcu **fp;
 484         struct route4_filter *fold, *f1, *pfp, *f = NULL;
 485         struct route4_bucket *b;
 486         struct nlattr *opt = tca[TCA_OPTIONS];
 487         struct nlattr *tb[TCA_ROUTE4_MAX + 1];
 488         unsigned int h, th;
 489         int err;
 490         bool new = true;
 491
 492         if (opt == NULL)
 493                 return handle ? -EINVAL : 0;
 494
 495         err = nla_parse_nested(tb, TCA_ROUTE4_MAX, opt, route4_policy);
 496         if (err < 0)
 497                 return err;
 498
 499         fold = (struct route4_filter *)*arg;
 500         if (fold && handle && fold->handle != handle)
 501                         return -EINVAL;
 502
 503         err = -ENOBUFS;
 504         f = kzalloc(sizeof(struct route4_filter), GFP_KERNEL);
 505         if (!f)
 506                 goto errout;
 507
 508         err = tcf_exts_init(&f->exts, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE);
 509         if (err < 0)
 510                 goto errout;
 511
 512         if (fold) {
 513                 f->id = fold->id;
 514                 f->iif = fold->iif;
 515                 f->res = fold->res;
 516                 f->handle = fold->handle;
 517
 518                 f->tp = fold->tp;
 519                 f->bkt = fold->bkt;
 520                 new = false;
 521         }
 522
 523         err = route4_set_parms(net, tp, base, f, handle, head, tb,
 524                                tca[TCA_RATE], new, ovr);
 525         if (err < 0)
 526                 goto errout;
 527
 528         h = from_hash(f->handle >> 16);
 529         fp = &f->bkt->ht[h];
 530         for (pfp = rtnl_dereference(*fp);
 531              (f1 = rtnl_dereference(*fp)) != NULL;
 532              fp = &f1->next)
 533                 if (f->handle < f1->handle)
 534                         break;
 535
 536         netif_keep_dst(qdisc_dev(tp->q));
 537         rcu_assign_pointer(f->next, f1);
 538         rcu_assign_pointer(*fp, f);
 539
 540         if (fold && fold->handle && f->handle != fold->handle) {
 541                 th = to_hash(fold->handle);
 542                 h = from_hash(fold->handle >> 16);
 543                 b = rtnl_dereference(head->table[th]);
 544                 if (b) {
 545                         fp = &b->ht[h];
 546                         for (pfp = rtnl_dereference(*fp); pfp;
 547                              fp = &pfp->next, pfp = rtnl_dereference(*fp)) {
 548                                 if (pfp == f) {
 549                                         *fp = f->next;
 550                                         break;
 551                                 }
 552                         }
 553                 }
 554         }
 555
 556         route4_reset_fastmap(head);
 557         *arg = (unsigned long)f;
 558         if (fold) {
 559                 tcf_unbind_filter(tp, &fold->res);
 560                 call_rcu(&fold->rcu, route4_delete_filter);
 561         }
 562         return 0;
 563
 564 errout:
 565         tcf_exts_destroy(&f->exts);
 566         kfree(f);
 567         return err;
 568 }
 569
 570 static void route4_walk(struct tcf_proto *tp, struct tcf_walker *arg)
 571 {
 572         struct route4_head *head = rtnl_dereference(tp->root);
 573         unsigned int h, h1;
 574
 575         if (head == NULL)
 576                 arg->stop = 1;
 577
 578         if (arg->stop)
 579                 return;
 580
 581         for (h = 0; h <= 256; h++) {
 582                 struct route4_bucket *b = rtnl_dereference(head->table[h]);
 583
 584                 if (b) {
 585                         for (h1 = 0; h1 <= 32; h1++) {
 586                                 struct route4_filter *f;
 587
 588                                 for (f = rtnl_dereference(b->ht[h1]);
 589                                      f;
 590                                      f = rtnl_dereference(f->next)) {
 591                                         if (arg->count < arg->skip) {
 592                                                 arg->count++;
 593                                                 continue;
 594                                         }
 595                                         if (arg->fn(tp, (unsigned long)f, arg) < 0) {
 596                                                 arg->stop = 1;
 597                                                 return;
 598                                         }
 599                                         arg->count++;
 600                                 }
 601                         }
 602                 }
 603         }
 604 }
 605
 606 static int route4_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
 607                        struct sk_buff *skb, struct tcmsg *t)
 608 {
 609         struct route4_filter *f = (struct route4_filter *)fh;
 610         struct nlattr *nest;
 611         u32 id;
 612
 613         if (f == NULL)
 614                 return skb->len;
 615
 616         t->tcm_handle = f->handle;
 617
 618         nest = nla_nest_start(skb, TCA_OPTIONS);
 619         if (nest == NULL)
 620                 goto nla_put_failure;
 621
 622         if (!(f->handle & 0x8000)) {
 623                 id = f->id & 0xFF;
 624                 if (nla_put_u32(skb, TCA_ROUTE4_TO, id))
 625                         goto nla_put_failure;
 626         }
 627         if (f->handle & 0x80000000) {
 628                 if ((f->handle >> 16) != 0xFFFF &&
 629                     nla_put_u32(skb, TCA_ROUTE4_IIF, f->iif))
 630                         goto nla_put_failure;
 631         } else {
 632                 id = f->id >> 16;
 633                 if (nla_put_u32(skb, TCA_ROUTE4_FROM, id))
 634                         goto nla_put_failure;
 635         }
 636         if (f->res.classid &&
 637             nla_put_u32(skb, TCA_ROUTE4_CLASSID, f->res.classid))
 638                 goto nla_put_failure;
 639
 640         if (tcf_exts_dump(skb, &f->exts) < 0)
 641                 goto nla_put_failure;
 642
 643         nla_nest_end(skb, nest);
 644
 645         if (tcf_exts_dump_stats(skb, &f->exts) < 0)
 646                 goto nla_put_failure;
 647
 648         return skb->len;
 649
 650 nla_put_failure:
 651         nla_nest_cancel(skb, nest);
 652         return -1;
 653 }
 654
 655 static struct tcf_proto_ops cls_route4_ops __read_mostly = {
 656         .kind           =       "route",
 657         .classify       =       route4_classify,
 658         .init           =       route4_init,
 659         .destroy        =       route4_destroy,
 660         .get            =       route4_get,
 661         .change         =       route4_change,
 662         .delete         =       route4_delete,
 663         .walk           =       route4_walk,
 664         .dump           =       route4_dump,
 665         .owner          =       THIS_MODULE,
 666 };
 667
 668 static int __init init_route4(void)
 669 {
 670         return register_tcf_proto_ops(&cls_route4_ops);
 671 }
 672
 673 static void __exit exit_route4(void)
 674 {
 675         unregister_tcf_proto_ops(&cls_route4_ops);
 676 }
 677
 678 module_init(init_route4)
 679 module_exit(exit_route4)
 680 MODULE_LICENSE("GPL");