net/sched/cls_tcindex.c

   1 /*
   2  * net/sched/cls_tcindex.c      Packet classifier for skb->tc_index
   3  *
   4  * Written 1998,1999 by Werner Almesberger, EPFL ICA
   5  */
   6
   7 #include <linux/module.h>
   8 #include <linux/types.h>
   9 #include <linux/kernel.h>
  10 #include <linux/skbuff.h>
  11 #include <linux/errno.h>
  12 #include <linux/slab.h>
  13 #include <net/act_api.h>
  14 #include <net/netlink.h>
  15 #include <net/pkt_cls.h>
  16
  17 /*
  18  * Passing parameters to the root seems to be done more awkwardly than really
  19  * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
  20  * verified. FIXME.
  21  */
  22
  23 #define PERFECT_HASH_THRESHOLD  64      /* use perfect hash if not bigger */
  24 #define DEFAULT_HASH_SIZE       64      /* optimized for diffserv */
  25
  26
  27 struct tcindex_filter_result {
  28         struct tcf_exts         exts;
  29         struct tcf_result       res;
  30 };
  31
  32 struct tcindex_filter {
  33         u16 key;
  34         struct tcindex_filter_result result;
  35         struct tcindex_filter *next;
  36 };
  37
  38
  39 struct tcindex_data {
  40         struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
  41         struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
  42                                       NULL if unused */
  43         u16 mask;               /* AND key with mask */
  44         int shift;              /* shift ANDed key to the right */
  45         int hash;               /* hash table size; 0 if undefined */
  46         int alloc_hash;         /* allocated size */
  47         int fall_through;       /* 0: only classify if explicit match */
  48 };
  49
  50 static inline int
  51 tcindex_filter_is_set(struct tcindex_filter_result *r)
  52 {
  53         return tcf_exts_is_predicative(&r->exts) || r->res.classid;
  54 }
  55
  56 static struct tcindex_filter_result *
  57 tcindex_lookup(struct tcindex_data *p, u16 key)
  58 {
  59         struct tcindex_filter *f;
  60
  61         if (p->perfect)
  62                 return tcindex_filter_is_set(p->perfect + key) ?
  63                         p->perfect + key : NULL;
  64         else if (p->h) {
  65                 for (f = p->h[key % p->hash]; f; f = f->next)
  66                         if (f->key == key)
  67                                 return &f->result;
  68         }
  69
  70         return NULL;
  71 }
  72
  73
  74 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
  75                             struct tcf_result *res)
  76 {
  77         struct tcindex_data *p = tp->root;
  78         struct tcindex_filter_result *f;
  79         int key = (skb->tc_index & p->mask) >> p->shift;
  80
  81         pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
  82                  skb, tp, res, p);
  83
  84         f = tcindex_lookup(p, key);
  85         if (!f) {
  86                 if (!p->fall_through)
  87                         return -1;
  88                 res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
  89                 res->class = 0;
  90                 pr_debug("alg 0x%x\n", res->classid);
  91                 return 0;
  92         }
  93         *res = f->res;
  94         pr_debug("map 0x%x\n", res->classid);
  95
  96         return tcf_exts_exec(skb, &f->exts, res);
  97 }
  98
  99
 100 static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
 101 {
 102         struct tcindex_data *p = tp->root;
 103         struct tcindex_filter_result *r;
 104
 105         pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
 106         if (p->perfect && handle >= p->alloc_hash)
 107                 return 0;
 108         r = tcindex_lookup(p, handle);
 109         return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
 110 }
 111
 112
 113 static void tcindex_put(struct tcf_proto *tp, unsigned long f)
 114 {
 115         pr_debug("tcindex_put(tp %p,f 0x%lx)\n", tp, f);
 116 }
 117
 118
 119 static int tcindex_init(struct tcf_proto *tp)
 120 {
 121         struct tcindex_data *p;
 122
 123         pr_debug("tcindex_init(tp %p)\n", tp);
 124         p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
 125         if (!p)
 126                 return -ENOMEM;
 127
 128         p->mask = 0xffff;
 129         p->hash = DEFAULT_HASH_SIZE;
 130         p->fall_through = 1;
 131
 132         tp->root = p;
 133         return 0;
 134 }
 135
 136
 137 static int
 138 __tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
 139 {
 140         struct tcindex_data *p = tp->root;
 141         struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
 142         struct tcindex_filter *f = NULL;
 143
 144         pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n", tp, arg, p, f);
 145         if (p->perfect) {
 146                 if (!r->res.class)
 147                         return -ENOENT;
 148         } else {
 149                 int i;
 150                 struct tcindex_filter **walk = NULL;
 151
 152                 for (i = 0; i < p->hash; i++)
 153                         for (walk = p->h+i; *walk; walk = &(*walk)->next)
 154                                 if (&(*walk)->result == r)
 155                                         goto found;
 156                 return -ENOENT;
 157
 158 found:
 159                 f = *walk;
 160                 if (lock)
 161                         tcf_tree_lock(tp);
 162                 *walk = f->next;
 163                 if (lock)
 164                         tcf_tree_unlock(tp);
 165         }
 166         tcf_unbind_filter(tp, &r->res);
 167         tcf_exts_destroy(tp, &r->exts);
 168         kfree(f);
 169         return 0;
 170 }
 171
 172 static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
 173 {
 174         return __tcindex_delete(tp, arg, 1);
 175 }
 176
 177 static inline int
 178 valid_perfect_hash(struct tcindex_data *p)
 179 {
 180         return  p->hash > (p->mask >> p->shift);
 181 }
 182
 183 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
 184         [TCA_TCINDEX_HASH]              = { .type = NLA_U32 },
 185         [TCA_TCINDEX_MASK]              = { .type = NLA_U16 },
 186         [TCA_TCINDEX_SHIFT]             = { .type = NLA_U32 },
 187         [TCA_TCINDEX_FALL_THROUGH]      = { .type = NLA_U32 },
 188         [TCA_TCINDEX_CLASSID]           = { .type = NLA_U32 },
 189 };
 190
 191 static void tcindex_filter_result_init(struct tcindex_filter_result *r)
 192 {
 193         memset(r, 0, sizeof(*r));
 194         tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
 195 }
 196
 197 static int
 198 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
 199                   u32 handle, struct tcindex_data *p,
 200                   struct tcindex_filter_result *r, struct nlattr **tb,
 201                  struct nlattr *est)
 202 {
 203         int err, balloc = 0;
 204         struct tcindex_filter_result new_filter_result, *old_r = r;
 205         struct tcindex_filter_result cr;
 206         struct tcindex_data cp;
 207         struct tcindex_filter *f = NULL; /* make gcc behave */
 208         struct tcf_exts e;
 209
 210         tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
 211         err = tcf_exts_validate(net, tp, tb, est, &e);
 212         if (err < 0)
 213                 return err;
 214
 215         memcpy(&cp, p, sizeof(cp));
 216         tcindex_filter_result_init(&new_filter_result);
 217
 218         tcindex_filter_result_init(&cr);
 219         if (old_r)
 220                 cr.res = r->res;
 221
 222         if (tb[TCA_TCINDEX_HASH])
 223                 cp.hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
 224
 225         if (tb[TCA_TCINDEX_MASK])
 226                 cp.mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
 227
 228         if (tb[TCA_TCINDEX_SHIFT])
 229                 cp.shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
 230
 231         err = -EBUSY;
 232         /* Hash already allocated, make sure that we still meet the
 233          * requirements for the allocated hash.
 234          */
 235         if (cp.perfect) {
 236                 if (!valid_perfect_hash(&cp) ||
 237                     cp.hash > cp.alloc_hash)
 238                         goto errout;
 239         } else if (cp.h && cp.hash != cp.alloc_hash)
 240                 goto errout;
 241
 242         err = -EINVAL;
 243         if (tb[TCA_TCINDEX_FALL_THROUGH])
 244                 cp.fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
 245
 246         if (!cp.hash) {
 247                 /* Hash not specified, use perfect hash if the upper limit
 248                  * of the hashing index is below the threshold.
 249                  */
 250                 if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
 251                         cp.hash = (cp.mask >> cp.shift) + 1;
 252                 else
 253                         cp.hash = DEFAULT_HASH_SIZE;
 254         }
 255
 256         if (!cp.perfect && !cp.h)
 257                 cp.alloc_hash = cp.hash;
 258
 259         /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
 260          * but then, we'd fail handles that may become valid after some future
 261          * mask change. While this is extremely unlikely to ever matter,
 262          * the check below is safer (and also more backwards-compatible).
 263          */
 264         if (cp.perfect || valid_perfect_hash(&cp))
 265                 if (handle >= cp.alloc_hash)
 266                         goto errout;
 267
 268
 269         err = -ENOMEM;
 270         if (!cp.perfect && !cp.h) {
 271                 if (valid_perfect_hash(&cp)) {
 272                         int i;
 273
 274                         cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
 275                         if (!cp.perfect)
 276                                 goto errout;
 277                         for (i = 0; i < cp.hash; i++)
 278                                 tcf_exts_init(&cp.perfect[i].exts, TCA_TCINDEX_ACT,
 279                                               TCA_TCINDEX_POLICE);
 280                         balloc = 1;
 281                 } else {
 282                         cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
 283                         if (!cp.h)
 284                                 goto errout;
 285                         balloc = 2;
 286                 }
 287         }
 288
 289         if (cp.perfect)
 290                 r = cp.perfect + handle;
 291         else
 292                 r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
 293
 294         if (r == &new_filter_result) {
 295                 f = kzalloc(sizeof(*f), GFP_KERNEL);
 296                 if (!f)
 297                         goto errout_alloc;
 298         }
 299
 300         if (tb[TCA_TCINDEX_CLASSID]) {
 301                 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
 302                 tcf_bind_filter(tp, &cr.res, base);
 303         }
 304
 305         if (old_r)
 306                 tcf_exts_change(tp, &r->exts, &e);
 307         else
 308                 tcf_exts_change(tp, &cr.exts, &e);
 309
 310         tcf_tree_lock(tp);
 311         if (old_r && old_r != r)
 312                 tcindex_filter_result_init(old_r);
 313
 314         memcpy(p, &cp, sizeof(cp));
 315         r->res = cr.res;
 316
 317         if (r == &new_filter_result) {
 318                 struct tcindex_filter **fp;
 319
 320                 f->key = handle;
 321                 f->result = new_filter_result;
 322                 f->next = NULL;
 323                 for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
 324                         /* nothing */;
 325                 *fp = f;
 326         }
 327         tcf_tree_unlock(tp);
 328
 329         return 0;
 330
 331 errout_alloc:
 332         if (balloc == 1)
 333                 kfree(cp.perfect);
 334         else if (balloc == 2)
 335                 kfree(cp.h);
 336 errout:
 337         tcf_exts_destroy(tp, &e);
 338         return err;
 339 }
 340
 341 static int
 342 tcindex_change(struct net *net, struct sk_buff *in_skb,
 343                struct tcf_proto *tp, unsigned long base, u32 handle,
 344                struct nlattr **tca, unsigned long *arg)
 345 {
 346         struct nlattr *opt = tca[TCA_OPTIONS];
 347         struct nlattr *tb[TCA_TCINDEX_MAX + 1];
 348         struct tcindex_data *p = tp->root;
 349         struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
 350         int err;
 351
 352         pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
 353             "p %p,r %p,*arg 0x%lx\n",
 354             tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
 355
 356         if (!opt)
 357                 return 0;
 358
 359         err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
 360         if (err < 0)
 361                 return err;
 362
 363         return tcindex_set_parms(net, tp, base, handle, p, r, tb,
 364                                  tca[TCA_RATE]);
 365 }
 366
 367
 368 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
 369 {
 370         struct tcindex_data *p = tp->root;
 371         struct tcindex_filter *f, *next;
 372         int i;
 373
 374         pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
 375         if (p->perfect) {
 376                 for (i = 0; i < p->hash; i++) {
 377                         if (!p->perfect[i].res.class)
 378                                 continue;
 379                         if (walker->count >= walker->skip) {
 380                                 if (walker->fn(tp,
 381                                     (unsigned long) (p->perfect+i), walker)
 382                                      < 0) {
 383                                         walker->stop = 1;
 384                                         return;
 385                                 }
 386                         }
 387                         walker->count++;
 388                 }
 389         }
 390         if (!p->h)
 391                 return;
 392         for (i = 0; i < p->hash; i++) {
 393                 for (f = p->h[i]; f; f = next) {
 394                         next = f->next;
 395                         if (walker->count >= walker->skip) {
 396                                 if (walker->fn(tp, (unsigned long) &f->result,
 397                                     walker) < 0) {
 398                                         walker->stop = 1;
 399                                         return;
 400                                 }
 401                         }
 402                         walker->count++;
 403                 }
 404         }
 405 }
 406
 407
 408 static int tcindex_destroy_element(struct tcf_proto *tp,
 409     unsigned long arg, struct tcf_walker *walker)
 410 {
 411         return __tcindex_delete(tp, arg, 0);
 412 }
 413
 414
 415 static void tcindex_destroy(struct tcf_proto *tp)
 416 {
 417         struct tcindex_data *p = tp->root;
 418         struct tcf_walker walker;
 419
 420         pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
 421         walker.count = 0;
 422         walker.skip = 0;
 423         walker.fn = &tcindex_destroy_element;
 424         tcindex_walk(tp, &walker);
 425         kfree(p->perfect);
 426         kfree(p->h);
 427         kfree(p);
 428         tp->root = NULL;
 429 }
 430
 431
 432 static int tcindex_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
 433     struct sk_buff *skb, struct tcmsg *t)
 434 {
 435         struct tcindex_data *p = tp->root;
 436         struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
 437         unsigned char *b = skb_tail_pointer(skb);
 438         struct nlattr *nest;
 439
 440         pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
 441                  tp, fh, skb, t, p, r, b);
 442         pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
 443
 444         nest = nla_nest_start(skb, TCA_OPTIONS);
 445         if (nest == NULL)
 446                 goto nla_put_failure;
 447
 448         if (!fh) {
 449                 t->tcm_handle = ~0; /* whatever ... */
 450                 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
 451                     nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
 452                     nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
 453                     nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
 454                         goto nla_put_failure;
 455                 nla_nest_end(skb, nest);
 456         } else {
 457                 if (p->perfect) {
 458                         t->tcm_handle = r-p->perfect;
 459                 } else {
 460                         struct tcindex_filter *f;
 461                         int i;
 462
 463                         t->tcm_handle = 0;
 464                         for (i = 0; !t->tcm_handle && i < p->hash; i++) {
 465                                 for (f = p->h[i]; !t->tcm_handle && f;
 466                                      f = f->next) {
 467                                         if (&f->result == r)
 468                                                 t->tcm_handle = f->key;
 469                                 }
 470                         }
 471                 }
 472                 pr_debug("handle = %d\n", t->tcm_handle);
 473                 if (r->res.class &&
 474                     nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
 475                         goto nla_put_failure;
 476
 477                 if (tcf_exts_dump(skb, &r->exts) < 0)
 478                         goto nla_put_failure;
 479                 nla_nest_end(skb, nest);
 480
 481                 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
 482                         goto nla_put_failure;
 483         }
 484
 485         return skb->len;
 486
 487 nla_put_failure:
 488         nlmsg_trim(skb, b);
 489         return -1;
 490 }
 491
 492 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
 493         .kind           =       "tcindex",
 494         .classify       =       tcindex_classify,
 495         .init           =       tcindex_init,
 496         .destroy        =       tcindex_destroy,
 497         .get            =       tcindex_get,
 498         .put            =       tcindex_put,
 499         .change         =       tcindex_change,
 500         .delete         =       tcindex_delete,
 501         .walk           =       tcindex_walk,
 502         .dump           =       tcindex_dump,
 503         .owner          =       THIS_MODULE,
 504 };
 505
 506 static int __init init_tcindex(void)
 507 {
 508         return register_tcf_proto_ops(&cls_tcindex_ops);
 509 }
 510
 511 static void __exit exit_tcindex(void)
 512 {
 513         unregister_tcf_proto_ops(&cls_tcindex_ops);
 514 }
 515
 516 module_init(init_tcindex)
 517 module_exit(exit_tcindex)
 518 MODULE_LICENSE("GPL");