sched/cls_route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/cls_route.c        ROUTE4 classifier.
   4  *
   5  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/slab.h>
  10 #include <linux/types.h>
  11 #include <linux/kernel.h>
  12 #include <linux/string.h>
  13 #include <linux/errno.h>
  14 #include <linux/skbuff.h>
  15 #include <net/dst.h>
  16 #include <net/route.h>
  17 #include <net/netlink.h>
  18 #include <net/act_api.h>
  19 #include <net/pkt_cls.h>
  20
  21 /*
  22  * 1. For now we assume that route tags < 256.
  23  *    It allows to use direct table lookups, instead of hash tables.
  24  * 2. For now we assume that "from TAG" and "fromdev DEV" statements
  25  *    are mutually  exclusive.
  26  * 3. "to TAG from ANY" has higher priority, than "to ANY from XXX"
  27  */
  28 struct route4_fastmap {
  29         struct route4_filter            *filter;
  30         u32                             id;
  31         int                             iif;
  32 };
  33
  34 struct route4_head {
  35         struct route4_fastmap           fastmap[16];
  36         struct route4_bucket __rcu      *table[256 + 1];
  37         struct rcu_head                 rcu;
  38 };
  39
  40 struct route4_bucket {
  41         /* 16 FROM buckets + 16 IIF buckets + 1 wildcard bucket */
  42         struct route4_filter __rcu      *ht[16 + 16 + 1];
  43         struct rcu_head                 rcu;
  44 };
  45
  46 struct route4_filter {
  47         struct route4_filter __rcu      *next;
  48         u32                     id;
  49         int                     iif;
  50
  51         struct tcf_result       res;
  52         struct tcf_exts         exts;
  53         u32                     handle;
  54         struct route4_bucket    *bkt;
  55         struct tcf_proto        *tp;
  56         struct rcu_work         rwork;
  57 };
  58
  59 #define ROUTE4_FAILURE ((struct route4_filter *)(-1L))
  60
  61 static inline int route4_fastmap_hash(u32 id, int iif)
  62 {
  63         return id & 0xF;
  64 }
  65
  66 static DEFINE_SPINLOCK(fastmap_lock);
  67 static void
  68 route4_reset_fastmap(struct route4_head *head)
  69 {
  70         spin_lock_bh(&fastmap_lock);
  71         memset(head->fastmap, 0, sizeof(head->fastmap));
  72         spin_unlock_bh(&fastmap_lock);
  73 }
  74
  75 static void
  76 route4_set_fastmap(struct route4_head *head, u32 id, int iif,
  77                    struct route4_filter *f)
  78 {
  79         int h = route4_fastmap_hash(id, iif);
  80
  81         /* fastmap updates must look atomic to aling id, iff, filter */
  82         spin_lock_bh(&fastmap_lock);
  83         head->fastmap[h].id = id;
  84         head->fastmap[h].iif = iif;
  85         head->fastmap[h].filter = f;
  86         spin_unlock_bh(&fastmap_lock);
  87 }
  88
  89 static inline int route4_hash_to(u32 id)
  90 {
  91         return id & 0xFF;
  92 }
  93
  94 static inline int route4_hash_from(u32 id)
  95 {
  96         return (id >> 16) & 0xF;
  97 }
  98
  99 static inline int route4_hash_iif(int iif)
 100 {
 101         return 16 + ((iif >> 16) & 0xF);
 102 }
 103
 104 static inline int route4_hash_wild(void)
 105 {
 106         return 32;
 107 }
 108
 109 #define ROUTE4_APPLY_RESULT()                                   \
 110 {                                                               \
 111         *res = f->res;                                          \
 112         if (tcf_exts_has_actions(&f->exts)) {                   \
 113                 int r = tcf_exts_exec(skb, &f->exts, res);      \
 114                 if (r < 0) {                                    \
 115                         dont_cache = 1;                         \
 116                         continue;                               \
 117                 }                                               \
 118                 return r;                                       \
 119         } else if (!dont_cache)                                 \
 120                 route4_set_fastmap(head, id, iif, f);           \
 121         return 0;                                               \
 122 }
 123
 124 static int route4_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 125                            struct tcf_result *res)
 126 {
 127         struct route4_head *head = rcu_dereference_bh(tp->root);
 128         struct dst_entry *dst;
 129         struct route4_bucket *b;
 130         struct route4_filter *f;
 131         u32 id, h;
 132         int iif, dont_cache = 0;
 133
 134         dst = skb_dst(skb);
 135         if (!dst)
 136                 goto failure;
 137
 138         id = dst->tclassid;
 139
 140         iif = inet_iif(skb);
 141
 142         h = route4_fastmap_hash(id, iif);
 143
 144         spin_lock(&fastmap_lock);
 145         if (id == head->fastmap[h].id &&
 146             iif == head->fastmap[h].iif &&
 147             (f = head->fastmap[h].filter) != NULL) {
 148                 if (f == ROUTE4_FAILURE) {
 149                         spin_unlock(&fastmap_lock);
 150                         goto failure;
 151                 }
 152
 153                 *res = f->res;
 154                 spin_unlock(&fastmap_lock);
 155                 return 0;
 156         }
 157         spin_unlock(&fastmap_lock);
 158
 159         h = route4_hash_to(id);
 160
 161 restart:
 162         b = rcu_dereference_bh(head->table[h]);
 163         if (b) {
 164                 for (f = rcu_dereference_bh(b->ht[route4_hash_from(id)]);
 165                      f;
 166                      f = rcu_dereference_bh(f->next))
 167                         if (f->id == id)
 168                                 ROUTE4_APPLY_RESULT();
 169
 170                 for (f = rcu_dereference_bh(b->ht[route4_hash_iif(iif)]);
 171                      f;
 172                      f = rcu_dereference_bh(f->next))
 173                         if (f->iif == iif)
 174                                 ROUTE4_APPLY_RESULT();
 175
 176                 for (f = rcu_dereference_bh(b->ht[route4_hash_wild()]);
 177                      f;
 178                      f = rcu_dereference_bh(f->next))
 179                         ROUTE4_APPLY_RESULT();
 180         }
 181         if (h < 256) {
 182                 h = 256;
 183                 id &= ~0xFFFF;
 184                 goto restart;
 185         }
 186
 187         if (!dont_cache)
 188                 route4_set_fastmap(head, id, iif, ROUTE4_FAILURE);
 189 failure:
 190         return -1;
 191 }
 192
 193 static inline u32 to_hash(u32 id)
 194 {
 195         u32 h = id & 0xFF;
 196
 197         if (id & 0x8000)
 198                 h += 256;
 199         return h;
 200 }
 201
 202 static inline u32 from_hash(u32 id)
 203 {
 204         id &= 0xFFFF;
 205         if (id == 0xFFFF)
 206                 return 32;
 207         if (!(id & 0x8000)) {
 208                 if (id > 255)
 209                         return 256;
 210                 return id & 0xF;
 211         }
 212         return 16 + (id & 0xF);
 213 }
 214
 215 static void *route4_get(struct tcf_proto *tp, u32 handle)
 216 {
 217         struct route4_head *head = rtnl_dereference(tp->root);
 218         struct route4_bucket *b;
 219         struct route4_filter *f;
 220         unsigned int h1, h2;
 221
 222         h1 = to_hash(handle);
 223         if (h1 > 256)
 224                 return NULL;
 225
 226         h2 = from_hash(handle >> 16);
 227         if (h2 > 32)
 228                 return NULL;
 229
 230         b = rtnl_dereference(head->table[h1]);
 231         if (b) {
 232                 for (f = rtnl_dereference(b->ht[h2]);
 233                      f;
 234                      f = rtnl_dereference(f->next))
 235                         if (f->handle == handle)
 236                                 return f;
 237         }
 238         return NULL;
 239 }
 240
 241 static int route4_init(struct tcf_proto *tp)
 242 {
 243         struct route4_head *head;
 244
 245         head = kzalloc(sizeof(struct route4_head), GFP_KERNEL);
 246         if (head == NULL)
 247                 return -ENOBUFS;
 248
 249         rcu_assign_pointer(tp->root, head);
 250         return 0;
 251 }
 252
 253 static void __route4_delete_filter(struct route4_filter *f)
 254 {
 255         tcf_exts_destroy(&f->exts);
 256         tcf_exts_put_net(&f->exts);
 257         kfree(f);
 258 }
 259
 260 static void route4_delete_filter_work(struct work_struct *work)
 261 {
 262         struct route4_filter *f = container_of(to_rcu_work(work),
 263                                                struct route4_filter,
 264                                                rwork);
 265         rtnl_lock();
 266         __route4_delete_filter(f);
 267         rtnl_unlock();
 268 }
 269
 270 static void route4_queue_work(struct route4_filter *f)
 271 {
 272         tcf_queue_work(&f->rwork, route4_delete_filter_work);
 273 }
 274
 275 static void route4_destroy(struct tcf_proto *tp, bool rtnl_held,
 276                            struct netlink_ext_ack *extack)
 277 {
 278         struct route4_head *head = rtnl_dereference(tp->root);
 279         int h1, h2;
 280
 281         if (head == NULL)
 282                 return;
 283
 284         for (h1 = 0; h1 <= 256; h1++) {
 285                 struct route4_bucket *b;
 286
 287                 b = rtnl_dereference(head->table[h1]);
 288                 if (b) {
 289                         for (h2 = 0; h2 <= 32; h2++) {
 290                                 struct route4_filter *f;
 291
 292                                 while ((f = rtnl_dereference(b->ht[h2])) != NULL) {
 293                                         struct route4_filter *next;
 294
 295                                         next = rtnl_dereference(f->next);
 296                                         RCU_INIT_POINTER(b->ht[h2], next);
 297                                         tcf_unbind_filter(tp, &f->res);
 298                                         if (tcf_exts_get_net(&f->exts))
 299                                                 route4_queue_work(f);
 300                                         else
 301                                                 __route4_delete_filter(f);
 302                                 }
 303                         }
 304                         RCU_INIT_POINTER(head->table[h1], NULL);
 305                         kfree_rcu(b, rcu);
 306                 }
 307         }
 308         kfree_rcu(head, rcu);
 309 }
 310
 311 static int route4_delete(struct tcf_proto *tp, void *arg, bool *last,
 312                          bool rtnl_held, struct netlink_ext_ack *extack)
 313 {
 314         struct route4_head *head = rtnl_dereference(tp->root);
 315         struct route4_filter *f = arg;
 316         struct route4_filter __rcu **fp;
 317         struct route4_filter *nf;
 318         struct route4_bucket *b;
 319         unsigned int h = 0;
 320         int i, h1;
 321
 322         if (!head || !f)
 323                 return -EINVAL;
 324
 325         h = f->handle;
 326         b = f->bkt;
 327
 328         fp = &b->ht[from_hash(h >> 16)];
 329         for (nf = rtnl_dereference(*fp); nf;
 330              fp = &nf->next, nf = rtnl_dereference(*fp)) {
 331                 if (nf == f) {
 332                         /* unlink it */
 333                         RCU_INIT_POINTER(*fp, rtnl_dereference(f->next));
 334
 335                         /* Remove any fastmap lookups that might ref filter
 336                          * notice we unlink'd the filter so we can't get it
 337                          * back in the fastmap.
 338                          */
 339                         route4_reset_fastmap(head);
 340
 341                         /* Delete it */
 342                         tcf_unbind_filter(tp, &f->res);
 343                         tcf_exts_get_net(&f->exts);
 344                         tcf_queue_work(&f->rwork, route4_delete_filter_work);
 345
 346                         /* Strip RTNL protected tree */
 347                         for (i = 0; i <= 32; i++) {
 348                                 struct route4_filter *rt;
 349
 350                                 rt = rtnl_dereference(b->ht[i]);
 351                                 if (rt)
 352                                         goto out;
 353                         }
 354
 355                         /* OK, session has no flows */
 356                         RCU_INIT_POINTER(head->table[to_hash(h)], NULL);
 357                         kfree_rcu(b, rcu);
 358                         break;
 359                 }
 360         }
 361
 362 out:
 363         *last = true;
 364         for (h1 = 0; h1 <= 256; h1++) {
 365                 if (rcu_access_pointer(head->table[h1])) {
 366                         *last = false;
 367                         break;
 368                 }
 369         }
 370
 371         return 0;
 372 }
 373
 374 static const struct nla_policy route4_policy[TCA_ROUTE4_MAX + 1] = {
 375         [TCA_ROUTE4_CLASSID]    = { .type = NLA_U32 },
 376         [TCA_ROUTE4_TO]         = { .type = NLA_U32 },
 377         [TCA_ROUTE4_FROM]       = { .type = NLA_U32 },
 378         [TCA_ROUTE4_IIF]        = { .type = NLA_U32 },
 379 };
 380
 381 static int route4_set_parms(struct net *net, struct tcf_proto *tp,
 382                             unsigned long base, struct route4_filter *f,
 383                             u32 handle, struct route4_head *head,
 384                             struct nlattr **tb, struct nlattr *est, int new,
 385                             u32 flags, struct netlink_ext_ack *extack)
 386 {
 387         u32 id = 0, to = 0, nhandle = 0x8000;
 388         struct route4_filter *fp;
 389         unsigned int h1;
 390         struct route4_bucket *b;
 391         int err;
 392
 393         err = tcf_exts_validate(net, tp, tb, est, &f->exts, flags, extack);
 394         if (err < 0)
 395                 return err;
 396
 397         if (tb[TCA_ROUTE4_TO]) {
 398                 if (new && handle & 0x8000)
 399                         return -EINVAL;
 400                 to = nla_get_u32(tb[TCA_ROUTE4_TO]);
 401                 if (to > 0xFF)
 402                         return -EINVAL;
 403                 nhandle = to;
 404         }
 405
 406         if (tb[TCA_ROUTE4_FROM]) {
 407                 if (tb[TCA_ROUTE4_IIF])
 408                         return -EINVAL;
 409                 id = nla_get_u32(tb[TCA_ROUTE4_FROM]);
 410                 if (id > 0xFF)
 411                         return -EINVAL;
 412                 nhandle |= id << 16;
 413         } else if (tb[TCA_ROUTE4_IIF]) {
 414                 id = nla_get_u32(tb[TCA_ROUTE4_IIF]);
 415                 if (id > 0x7FFF)
 416                         return -EINVAL;
 417                 nhandle |= (id | 0x8000) << 16;
 418         } else
 419                 nhandle |= 0xFFFF << 16;
 420
 421         if (handle && new) {
 422                 nhandle |= handle & 0x7F00;
 423                 if (nhandle != handle)
 424                         return -EINVAL;
 425         }
 426
 427         if (!nhandle) {
 428                 NL_SET_ERR_MSG(extack, "Replacing with handle of 0 is invalid");
 429                 return -EINVAL;
 430         }
 431
 432         h1 = to_hash(nhandle);
 433         b = rtnl_dereference(head->table[h1]);
 434         if (!b) {
 435                 b = kzalloc(sizeof(struct route4_bucket), GFP_KERNEL);
 436                 if (b == NULL)
 437                         return -ENOBUFS;
 438
 439                 rcu_assign_pointer(head->table[h1], b);
 440         } else {
 441                 unsigned int h2 = from_hash(nhandle >> 16);
 442
 443                 for (fp = rtnl_dereference(b->ht[h2]);
 444                      fp;
 445                      fp = rtnl_dereference(fp->next))
 446                         if (fp->handle == f->handle)
 447                                 return -EEXIST;
 448         }
 449
 450         if (tb[TCA_ROUTE4_TO])
 451                 f->id = to;
 452
 453         if (tb[TCA_ROUTE4_FROM])
 454                 f->id = to | id<<16;
 455         else if (tb[TCA_ROUTE4_IIF])
 456                 f->iif = id;
 457
 458         f->handle = nhandle;
 459         f->bkt = b;
 460         f->tp = tp;
 461
 462         if (tb[TCA_ROUTE4_CLASSID]) {
 463                 f->res.classid = nla_get_u32(tb[TCA_ROUTE4_CLASSID]);
 464                 tcf_bind_filter(tp, &f->res, base);
 465         }
 466
 467         return 0;
 468 }
 469
 470 static int route4_change(struct net *net, struct sk_buff *in_skb,
 471                          struct tcf_proto *tp, unsigned long base, u32 handle,
 472                          struct nlattr **tca, void **arg, u32 flags,
 473                          struct netlink_ext_ack *extack)
 474 {
 475         struct route4_head *head = rtnl_dereference(tp->root);
 476         struct route4_filter __rcu **fp;
 477         struct route4_filter *fold, *f1, *pfp, *f = NULL;
 478         struct route4_bucket *b;
 479         struct nlattr *opt = tca[TCA_OPTIONS];
 480         struct nlattr *tb[TCA_ROUTE4_MAX + 1];
 481         unsigned int h, th;
 482         int err;
 483         bool new = true;
 484
 485         if (!handle) {
 486                 NL_SET_ERR_MSG(extack, "Creating with handle of 0 is invalid");
 487                 return -EINVAL;
 488         }
 489
 490         if (opt == NULL)
 491                 return -EINVAL;
 492
 493         err = nla_parse_nested_deprecated(tb, TCA_ROUTE4_MAX, opt,
 494                                           route4_policy, NULL);
 495         if (err < 0)
 496                 return err;
 497
 498         fold = *arg;
 499         if (fold && fold->handle != handle)
 500                         return -EINVAL;
 501
 502         err = -ENOBUFS;
 503         f = kzalloc(sizeof(struct route4_filter), GFP_KERNEL);
 504         if (!f)
 505                 goto errout;
 506
 507         err = tcf_exts_init(&f->exts, net, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE);
 508         if (err < 0)
 509                 goto errout;
 510
 511         if (fold) {
 512                 f->id = fold->id;
 513                 f->iif = fold->iif;
 514                 f->handle = fold->handle;
 515
 516                 f->tp = fold->tp;
 517                 f->bkt = fold->bkt;
 518                 new = false;
 519         }
 520
 521         err = route4_set_parms(net, tp, base, f, handle, head, tb,
 522                                tca[TCA_RATE], new, flags, extack);
 523         if (err < 0)
 524                 goto errout;
 525
 526         h = from_hash(f->handle >> 16);
 527         fp = &f->bkt->ht[h];
 528         for (pfp = rtnl_dereference(*fp);
 529              (f1 = rtnl_dereference(*fp)) != NULL;
 530              fp = &f1->next)
 531                 if (f->handle < f1->handle)
 532                         break;
 533
 534         tcf_block_netif_keep_dst(tp->chain->block);
 535         rcu_assign_pointer(f->next, f1);
 536         rcu_assign_pointer(*fp, f);
 537
 538         if (fold) {
 539                 th = to_hash(fold->handle);
 540                 h = from_hash(fold->handle >> 16);
 541                 b = rtnl_dereference(head->table[th]);
 542                 if (b) {
 543                         fp = &b->ht[h];
 544                         for (pfp = rtnl_dereference(*fp); pfp;
 545                              fp = &pfp->next, pfp = rtnl_dereference(*fp)) {
 546                                 if (pfp == fold) {
 547                                         rcu_assign_pointer(*fp, fold->next);
 548                                         break;
 549                                 }
 550                         }
 551                 }
 552         }
 553
 554         route4_reset_fastmap(head);
 555         *arg = f;
 556         if (fold) {
 557                 tcf_unbind_filter(tp, &fold->res);
 558                 tcf_exts_get_net(&fold->exts);
 559                 tcf_queue_work(&fold->rwork, route4_delete_filter_work);
 560         }
 561         return 0;
 562
 563 errout:
 564         if (f)
 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                         bool rtnl_held)
 572 {
 573         struct route4_head *head = rtnl_dereference(tp->root);
 574         unsigned int h, h1;
 575
 576         if (head == NULL || arg->stop)
 577                 return;
 578
 579         for (h = 0; h <= 256; h++) {
 580                 struct route4_bucket *b = rtnl_dereference(head->table[h]);
 581
 582                 if (b) {
 583                         for (h1 = 0; h1 <= 32; h1++) {
 584                                 struct route4_filter *f;
 585
 586                                 for (f = rtnl_dereference(b->ht[h1]);
 587                                      f;
 588                                      f = rtnl_dereference(f->next)) {
 589                                         if (!tc_cls_stats_dump(tp, arg, f))
 590                                                 return;
 591                                 }
 592                         }
 593                 }
 594         }
 595 }
 596
 597 static int route4_dump(struct net *net, struct tcf_proto *tp, void *fh,
 598                        struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
 599 {
 600         struct route4_filter *f = fh;
 601         struct nlattr *nest;
 602         u32 id;
 603
 604         if (f == NULL)
 605                 return skb->len;
 606
 607         t->tcm_handle = f->handle;
 608
 609         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
 610         if (nest == NULL)
 611                 goto nla_put_failure;
 612
 613         if (!(f->handle & 0x8000)) {
 614                 id = f->id & 0xFF;
 615                 if (nla_put_u32(skb, TCA_ROUTE4_TO, id))
 616                         goto nla_put_failure;
 617         }
 618         if (f->handle & 0x80000000) {
 619                 if ((f->handle >> 16) != 0xFFFF &&
 620                     nla_put_u32(skb, TCA_ROUTE4_IIF, f->iif))
 621                         goto nla_put_failure;
 622         } else {
 623                 id = f->id >> 16;
 624                 if (nla_put_u32(skb, TCA_ROUTE4_FROM, id))
 625                         goto nla_put_failure;
 626         }
 627         if (f->res.classid &&
 628             nla_put_u32(skb, TCA_ROUTE4_CLASSID, f->res.classid))
 629                 goto nla_put_failure;
 630
 631         if (tcf_exts_dump(skb, &f->exts) < 0)
 632                 goto nla_put_failure;
 633
 634         nla_nest_end(skb, nest);
 635
 636         if (tcf_exts_dump_stats(skb, &f->exts) < 0)
 637                 goto nla_put_failure;
 638
 639         return skb->len;
 640
 641 nla_put_failure:
 642         nla_nest_cancel(skb, nest);
 643         return -1;
 644 }
 645
 646 static void route4_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
 647                               unsigned long base)
 648 {
 649         struct route4_filter *f = fh;
 650
 651         tc_cls_bind_class(classid, cl, q, &f->res, base);
 652 }
 653
 654 static struct tcf_proto_ops cls_route4_ops __read_mostly = {
 655         .kind           =       "route",
 656         .classify       =       route4_classify,
 657         .init           =       route4_init,
 658         .destroy        =       route4_destroy,
 659         .get            =       route4_get,
 660         .change         =       route4_change,
 661         .delete         =       route4_delete,
 662         .walk           =       route4_walk,
 663         .dump           =       route4_dump,
 664         .bind_class     =       route4_bind_class,
 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");