net/sched/sch_dsmark.c

   1 /* net/sched/sch_dsmark.c - Differentiated Services field marker */
   2
   3 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
   4
   5
   6 #include <linux/module.h>
   7 #include <linux/init.h>
   8 #include <linux/types.h>
   9 #include <linux/string.h>
  10 #include <linux/errno.h>
  11 #include <linux/skbuff.h>
  12 #include <linux/rtnetlink.h>
  13 #include <linux/bitops.h>
  14 #include <net/pkt_sched.h>
  15 #include <net/dsfield.h>
  16 #include <net/inet_ecn.h>
  17 #include <asm/byteorder.h>
  18
  19 /*
  20  * classid      class           marking
  21  * -------      -----           -------
  22  *   n/a          0             n/a
  23  *   x:0          1             use entry [0]
  24  *   ...         ...            ...
  25  *   x:y y>0     y+1            use entry [y]
  26  *   ...         ...            ...
  27  * x:indices-1  indices         use entry [indices-1]
  28  *   ...         ...            ...
  29  *   x:y         y+1            use entry [y & (indices-1)]
  30  *   ...         ...            ...
  31  * 0xffff       0x10000         use entry [indices-1]
  32  */
  33
  34
  35 #define NO_DEFAULT_INDEX        (1 << 16)
  36
  37 struct dsmark_qdisc_data {
  38         struct Qdisc            *q;
  39         struct tcf_proto        *filter_list;
  40         u8                      *mask;  /* "owns" the array */
  41         u8                      *value;
  42         u16                     indices;
  43         u32                     default_index;  /* index range is 0...0xffff */
  44         int                     set_tc_index;
  45 };
  46
  47 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  48 {
  49         return (index <= p->indices && index > 0);
  50 }
  51
  52 /* ------------------------- Class/flow operations ------------------------- */
  53
  54 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  55                         struct Qdisc *new, struct Qdisc **old)
  56 {
  57         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  58
  59         pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
  60                 sch, p, new, old);
  61
  62         if (new == NULL) {
  63                 new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
  64                                         sch->handle);
  65                 if (new == NULL)
  66                         new = &noop_qdisc;
  67         }
  68
  69         sch_tree_lock(sch);
  70         *old = xchg(&p->q, new);
  71         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
  72         qdisc_reset(*old);
  73         sch_tree_unlock(sch);
  74
  75         return 0;
  76 }
  77
  78 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
  79 {
  80         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  81         return p->q;
  82 }
  83
  84 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
  85 {
  86         pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
  87                 sch, qdisc_priv(sch), classid);
  88
  89         return TC_H_MIN(classid) + 1;
  90 }
  91
  92 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
  93                                         unsigned long parent, u32 classid)
  94 {
  95         return dsmark_get(sch, classid);
  96 }
  97
  98 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
  99 {
 100 }
 101
 102 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
 103                          struct nlattr **tca, unsigned long *arg)
 104 {
 105         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 106         struct nlattr *opt = tca[TCA_OPTIONS];
 107         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 108         int err = -EINVAL;
 109         u8 mask = 0;
 110
 111         pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
 112                 "arg 0x%lx\n", sch, p, classid, parent, *arg);
 113
 114         if (!dsmark_valid_index(p, *arg)) {
 115                 err = -ENOENT;
 116                 goto errout;
 117         }
 118
 119         if (!opt || nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL))
 120                 goto errout;
 121
 122         if (tb[TCA_DSMARK_MASK]) {
 123                 if (nla_len(tb[TCA_DSMARK_MASK]) < sizeof(u8))
 124                         goto errout;
 125                 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
 126         }
 127         if (tb[TCA_DSMARK_VALUE]) {
 128                 if (nla_len(tb[TCA_DSMARK_VALUE]) < sizeof(u8))
 129                         goto errout;
 130                 p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
 131         }
 132
 133         if (tb[TCA_DSMARK_MASK])
 134                 p->mask[*arg-1] = mask;
 135
 136         err = 0;
 137
 138 errout:
 139         return err;
 140 }
 141
 142 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 143 {
 144         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 145
 146         if (!dsmark_valid_index(p, arg))
 147                 return -EINVAL;
 148
 149         p->mask[arg-1] = 0xff;
 150         p->value[arg-1] = 0;
 151
 152         return 0;
 153 }
 154
 155 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 156 {
 157         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 158         int i;
 159
 160         pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
 161
 162         if (walker->stop)
 163                 return;
 164
 165         for (i = 0; i < p->indices; i++) {
 166                 if (p->mask[i] == 0xff && !p->value[i])
 167                         goto ignore;
 168                 if (walker->count >= walker->skip) {
 169                         if (walker->fn(sch, i+1, walker) < 0) {
 170                                 walker->stop = 1;
 171                                 break;
 172                         }
 173                 }
 174 ignore:
 175                 walker->count++;
 176         }
 177 }
 178
 179 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
 180                                                  unsigned long cl)
 181 {
 182         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 183         return &p->filter_list;
 184 }
 185
 186 /* --------------------------- Qdisc operations ---------------------------- */
 187
 188 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 189 {
 190         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 191         int err;
 192
 193         pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 194
 195         if (p->set_tc_index) {
 196                 switch (skb->protocol) {
 197                 case __constant_htons(ETH_P_IP):
 198                         if (skb_cow_head(skb, sizeof(struct iphdr)))
 199                                 goto drop;
 200
 201                         skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
 202                                 & ~INET_ECN_MASK;
 203                         break;
 204
 205                 case __constant_htons(ETH_P_IPV6):
 206                         if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
 207                                 goto drop;
 208
 209                         skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 210                                 & ~INET_ECN_MASK;
 211                         break;
 212                 default:
 213                         skb->tc_index = 0;
 214                         break;
 215                 }
 216         }
 217
 218         if (TC_H_MAJ(skb->priority) == sch->handle)
 219                 skb->tc_index = TC_H_MIN(skb->priority);
 220         else {
 221                 struct tcf_result res;
 222                 int result = tc_classify(skb, p->filter_list, &res);
 223
 224                 pr_debug("result %d class 0x%04x\n", result, res.classid);
 225
 226                 switch (result) {
 227 #ifdef CONFIG_NET_CLS_ACT
 228                 case TC_ACT_QUEUED:
 229                 case TC_ACT_STOLEN:
 230                         kfree_skb(skb);
 231                         return NET_XMIT_SUCCESS;
 232
 233                 case TC_ACT_SHOT:
 234                         goto drop;
 235 #endif
 236                 case TC_ACT_OK:
 237                         skb->tc_index = TC_H_MIN(res.classid);
 238                         break;
 239
 240                 default:
 241                         if (p->default_index != NO_DEFAULT_INDEX)
 242                                 skb->tc_index = p->default_index;
 243                         break;
 244                 }
 245         }
 246
 247         err = p->q->enqueue(skb, p->q);
 248         if (err != NET_XMIT_SUCCESS) {
 249                 sch->qstats.drops++;
 250                 return err;
 251         }
 252
 253         sch->bstats.bytes += skb->len;
 254         sch->bstats.packets++;
 255         sch->q.qlen++;
 256
 257         return NET_XMIT_SUCCESS;
 258
 259 drop:
 260         kfree_skb(skb);
 261         sch->qstats.drops++;
 262         return NET_XMIT_BYPASS;
 263 }
 264
 265 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 266 {
 267         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 268         struct sk_buff *skb;
 269         u32 index;
 270
 271         pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
 272
 273         skb = p->q->ops->dequeue(p->q);
 274         if (skb == NULL)
 275                 return NULL;
 276
 277         sch->q.qlen--;
 278
 279         index = skb->tc_index & (p->indices - 1);
 280         pr_debug("index %d->%d\n", skb->tc_index, index);
 281
 282         switch (skb->protocol) {
 283         case __constant_htons(ETH_P_IP):
 284                 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
 285                                     p->value[index]);
 286                         break;
 287         case __constant_htons(ETH_P_IPV6):
 288                 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
 289                                     p->value[index]);
 290                         break;
 291         default:
 292                 /*
 293                  * Only complain if a change was actually attempted.
 294                  * This way, we can send non-IP traffic through dsmark
 295                  * and don't need yet another qdisc as a bypass.
 296                  */
 297                 if (p->mask[index] != 0xff || p->value[index])
 298                         printk(KERN_WARNING
 299                                "dsmark_dequeue: unsupported protocol %d\n",
 300                                ntohs(skb->protocol));
 301                 break;
 302         }
 303
 304         return skb;
 305 }
 306
 307 static int dsmark_requeue(struct sk_buff *skb, struct Qdisc *sch)
 308 {
 309         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 310         int err;
 311
 312         pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 313
 314         err = p->q->ops->requeue(skb, p->q);
 315         if (err != NET_XMIT_SUCCESS) {
 316                 sch->qstats.drops++;
 317                 return err;
 318         }
 319
 320         sch->q.qlen++;
 321         sch->qstats.requeues++;
 322
 323         return NET_XMIT_SUCCESS;
 324 }
 325
 326 static unsigned int dsmark_drop(struct Qdisc *sch)
 327 {
 328         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 329         unsigned int len;
 330
 331         pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 332
 333         if (p->q->ops->drop == NULL)
 334                 return 0;
 335
 336         len = p->q->ops->drop(p->q);
 337         if (len)
 338                 sch->q.qlen--;
 339
 340         return len;
 341 }
 342
 343 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
 344 {
 345         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 346         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 347         int err = -EINVAL;
 348         u32 default_index = NO_DEFAULT_INDEX;
 349         u16 indices;
 350         u8 *mask;
 351
 352         pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
 353
 354         if (!opt || nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL) < 0)
 355                 goto errout;
 356
 357         if (nla_len(tb[TCA_DSMARK_INDICES]) < sizeof(u16))
 358                 goto errout;
 359         indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
 360
 361         if (hweight32(indices) != 1)
 362                 goto errout;
 363
 364         if (tb[TCA_DSMARK_DEFAULT_INDEX]) {
 365                 if (nla_len(tb[TCA_DSMARK_DEFAULT_INDEX]) < sizeof(u16))
 366                         goto errout;
 367                 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
 368         }
 369
 370         mask = kmalloc(indices * 2, GFP_KERNEL);
 371         if (mask == NULL) {
 372                 err = -ENOMEM;
 373                 goto errout;
 374         }
 375
 376         p->mask = mask;
 377         memset(p->mask, 0xff, indices);
 378
 379         p->value = p->mask + indices;
 380         memset(p->value, 0, indices);
 381
 382         p->indices = indices;
 383         p->default_index = default_index;
 384         p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
 385
 386         p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
 387         if (p->q == NULL)
 388                 p->q = &noop_qdisc;
 389
 390         pr_debug("dsmark_init: qdisc %p\n", p->q);
 391
 392         err = 0;
 393 errout:
 394         return err;
 395 }
 396
 397 static void dsmark_reset(struct Qdisc *sch)
 398 {
 399         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 400
 401         pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 402         qdisc_reset(p->q);
 403         sch->q.qlen = 0;
 404 }
 405
 406 static void dsmark_destroy(struct Qdisc *sch)
 407 {
 408         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 409
 410         pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
 411
 412         tcf_destroy_chain(p->filter_list);
 413         qdisc_destroy(p->q);
 414         kfree(p->mask);
 415 }
 416
 417 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 418                              struct sk_buff *skb, struct tcmsg *tcm)
 419 {
 420         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 421         struct nlattr *opts = NULL;
 422
 423         pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
 424
 425         if (!dsmark_valid_index(p, cl))
 426                 return -EINVAL;
 427
 428         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
 429         tcm->tcm_info = p->q->handle;
 430
 431         opts = nla_nest_start(skb, TCA_OPTIONS);
 432         if (opts == NULL)
 433                 goto nla_put_failure;
 434         NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
 435         NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
 436
 437         return nla_nest_end(skb, opts);
 438
 439 nla_put_failure:
 440         return nla_nest_cancel(skb, opts);
 441 }
 442
 443 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 444 {
 445         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 446         struct nlattr *opts = NULL;
 447
 448         opts = nla_nest_start(skb, TCA_OPTIONS);
 449         if (opts == NULL)
 450                 goto nla_put_failure;
 451         NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
 452
 453         if (p->default_index != NO_DEFAULT_INDEX)
 454                 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
 455
 456         if (p->set_tc_index)
 457                 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
 458
 459         return nla_nest_end(skb, opts);
 460
 461 nla_put_failure:
 462         return nla_nest_cancel(skb, opts);
 463 }
 464
 465 static const struct Qdisc_class_ops dsmark_class_ops = {
 466         .graft          =       dsmark_graft,
 467         .leaf           =       dsmark_leaf,
 468         .get            =       dsmark_get,
 469         .put            =       dsmark_put,
 470         .change         =       dsmark_change,
 471         .delete         =       dsmark_delete,
 472         .walk           =       dsmark_walk,
 473         .tcf_chain      =       dsmark_find_tcf,
 474         .bind_tcf       =       dsmark_bind_filter,
 475         .unbind_tcf     =       dsmark_put,
 476         .dump           =       dsmark_dump_class,
 477 };
 478
 479 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 480         .next           =       NULL,
 481         .cl_ops         =       &dsmark_class_ops,
 482         .id             =       "dsmark",
 483         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 484         .enqueue        =       dsmark_enqueue,
 485         .dequeue        =       dsmark_dequeue,
 486         .requeue        =       dsmark_requeue,
 487         .drop           =       dsmark_drop,
 488         .init           =       dsmark_init,
 489         .reset          =       dsmark_reset,
 490         .destroy        =       dsmark_destroy,
 491         .change         =       NULL,
 492         .dump           =       dsmark_dump,
 493         .owner          =       THIS_MODULE,
 494 };
 495
 496 static int __init dsmark_module_init(void)
 497 {
 498         return register_qdisc(&dsmark_qdisc_ops);
 499 }
 500
 501 static void __exit dsmark_module_exit(void)
 502 {
 503         unregister_qdisc(&dsmark_qdisc_ops);
 504 }
 505
 506 module_init(dsmark_module_init)
 507 module_exit(dsmark_module_exit)
 508
 509 MODULE_LICENSE("GPL");