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 rtattr **tca, unsigned long *arg)
 104 {
 105         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 106         struct rtattr *opt = tca[TCA_OPTIONS-1];
 107         struct rtattr *tb[TCA_DSMARK_MAX];
 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 rtattr_failure;
 117         }
 118
 119         if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt))
 120                 goto rtattr_failure;
 121
 122         if (tb[TCA_DSMARK_MASK-1])
 123                 mask = RTA_GET_U8(tb[TCA_DSMARK_MASK-1]);
 124
 125         if (tb[TCA_DSMARK_VALUE-1])
 126                 p->value[*arg-1] = RTA_GET_U8(tb[TCA_DSMARK_VALUE-1]);
 127
 128         if (tb[TCA_DSMARK_MASK-1])
 129                 p->mask[*arg-1] = mask;
 130
 131         err = 0;
 132
 133 rtattr_failure:
 134         return err;
 135 }
 136
 137 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 138 {
 139         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 140
 141         if (!dsmark_valid_index(p, arg))
 142                 return -EINVAL;
 143
 144         p->mask[arg-1] = 0xff;
 145         p->value[arg-1] = 0;
 146
 147         return 0;
 148 }
 149
 150 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 151 {
 152         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 153         int i;
 154
 155         pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
 156
 157         if (walker->stop)
 158                 return;
 159
 160         for (i = 0; i < p->indices; i++) {
 161                 if (p->mask[i] == 0xff && !p->value[i])
 162                         goto ignore;
 163                 if (walker->count >= walker->skip) {
 164                         if (walker->fn(sch, i+1, walker) < 0) {
 165                                 walker->stop = 1;
 166                                 break;
 167                         }
 168                 }
 169 ignore:
 170                 walker->count++;
 171         }
 172 }
 173
 174 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
 175                                                  unsigned long cl)
 176 {
 177         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 178         return &p->filter_list;
 179 }
 180
 181 /* --------------------------- Qdisc operations ---------------------------- */
 182
 183 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 184 {
 185         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 186         int err;
 187
 188         pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 189
 190         if (p->set_tc_index) {
 191                 switch (skb->protocol) {
 192                 case __constant_htons(ETH_P_IP):
 193                         if (skb_cow_head(skb, sizeof(struct iphdr)))
 194                                 goto drop;
 195
 196                         skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
 197                                 & ~INET_ECN_MASK;
 198                         break;
 199
 200                 case __constant_htons(ETH_P_IPV6):
 201                         if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
 202                                 goto drop;
 203
 204                         skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 205                                 & ~INET_ECN_MASK;
 206                         break;
 207                 default:
 208                         skb->tc_index = 0;
 209                         break;
 210                 }
 211         }
 212
 213         if (TC_H_MAJ(skb->priority) == sch->handle)
 214                 skb->tc_index = TC_H_MIN(skb->priority);
 215         else {
 216                 struct tcf_result res;
 217                 int result = tc_classify(skb, p->filter_list, &res);
 218
 219                 pr_debug("result %d class 0x%04x\n", result, res.classid);
 220
 221                 switch (result) {
 222 #ifdef CONFIG_NET_CLS_ACT
 223                 case TC_ACT_QUEUED:
 224                 case TC_ACT_STOLEN:
 225                         kfree_skb(skb);
 226                         return NET_XMIT_SUCCESS;
 227
 228                 case TC_ACT_SHOT:
 229                         goto drop;
 230 #endif
 231                 case TC_ACT_OK:
 232                         skb->tc_index = TC_H_MIN(res.classid);
 233                         break;
 234
 235                 default:
 236                         if (p->default_index != NO_DEFAULT_INDEX)
 237                                 skb->tc_index = p->default_index;
 238                         break;
 239                 }
 240         }
 241
 242         err = p->q->enqueue(skb, p->q);
 243         if (err != NET_XMIT_SUCCESS) {
 244                 sch->qstats.drops++;
 245                 return err;
 246         }
 247
 248         sch->bstats.bytes += skb->len;
 249         sch->bstats.packets++;
 250         sch->q.qlen++;
 251
 252         return NET_XMIT_SUCCESS;
 253
 254 drop:
 255         kfree_skb(skb);
 256         sch->qstats.drops++;
 257         return NET_XMIT_BYPASS;
 258 }
 259
 260 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 261 {
 262         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 263         struct sk_buff *skb;
 264         u32 index;
 265
 266         pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
 267
 268         skb = p->q->ops->dequeue(p->q);
 269         if (skb == NULL)
 270                 return NULL;
 271
 272         sch->q.qlen--;
 273
 274         index = skb->tc_index & (p->indices - 1);
 275         pr_debug("index %d->%d\n", skb->tc_index, index);
 276
 277         switch (skb->protocol) {
 278         case __constant_htons(ETH_P_IP):
 279                 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
 280                                     p->value[index]);
 281                         break;
 282         case __constant_htons(ETH_P_IPV6):
 283                 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
 284                                     p->value[index]);
 285                         break;
 286         default:
 287                 /*
 288                  * Only complain if a change was actually attempted.
 289                  * This way, we can send non-IP traffic through dsmark
 290                  * and don't need yet another qdisc as a bypass.
 291                  */
 292                 if (p->mask[index] != 0xff || p->value[index])
 293                         printk(KERN_WARNING
 294                                "dsmark_dequeue: unsupported protocol %d\n",
 295                                ntohs(skb->protocol));
 296                 break;
 297         }
 298
 299         return skb;
 300 }
 301
 302 static int dsmark_requeue(struct sk_buff *skb, struct Qdisc *sch)
 303 {
 304         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 305         int err;
 306
 307         pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 308
 309         err = p->q->ops->requeue(skb, p->q);
 310         if (err != NET_XMIT_SUCCESS) {
 311                 sch->qstats.drops++;
 312                 return err;
 313         }
 314
 315         sch->q.qlen++;
 316         sch->qstats.requeues++;
 317
 318         return NET_XMIT_SUCCESS;
 319 }
 320
 321 static unsigned int dsmark_drop(struct Qdisc *sch)
 322 {
 323         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 324         unsigned int len;
 325
 326         pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 327
 328         if (p->q->ops->drop == NULL)
 329                 return 0;
 330
 331         len = p->q->ops->drop(p->q);
 332         if (len)
 333                 sch->q.qlen--;
 334
 335         return len;
 336 }
 337
 338 static int dsmark_init(struct Qdisc *sch, struct rtattr *opt)
 339 {
 340         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 341         struct rtattr *tb[TCA_DSMARK_MAX];
 342         int err = -EINVAL;
 343         u32 default_index = NO_DEFAULT_INDEX;
 344         u16 indices;
 345         u8 *mask;
 346
 347         pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
 348
 349         if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt) < 0)
 350                 goto errout;
 351
 352         indices = RTA_GET_U16(tb[TCA_DSMARK_INDICES-1]);
 353
 354         if (hweight32(indices) != 1)
 355                 goto errout;
 356
 357         if (tb[TCA_DSMARK_DEFAULT_INDEX-1])
 358                 default_index = RTA_GET_U16(tb[TCA_DSMARK_DEFAULT_INDEX-1]);
 359
 360         mask = kmalloc(indices * 2, GFP_KERNEL);
 361         if (mask == NULL) {
 362                 err = -ENOMEM;
 363                 goto errout;
 364         }
 365
 366         p->mask = mask;
 367         memset(p->mask, 0xff, indices);
 368
 369         p->value = p->mask + indices;
 370         memset(p->value, 0, indices);
 371
 372         p->indices = indices;
 373         p->default_index = default_index;
 374         p->set_tc_index = RTA_GET_FLAG(tb[TCA_DSMARK_SET_TC_INDEX-1]);
 375
 376         p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
 377         if (p->q == NULL)
 378                 p->q = &noop_qdisc;
 379
 380         pr_debug("dsmark_init: qdisc %p\n", p->q);
 381
 382         err = 0;
 383 errout:
 384 rtattr_failure:
 385         return err;
 386 }
 387
 388 static void dsmark_reset(struct Qdisc *sch)
 389 {
 390         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 391
 392         pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 393         qdisc_reset(p->q);
 394         sch->q.qlen = 0;
 395 }
 396
 397 static void dsmark_destroy(struct Qdisc *sch)
 398 {
 399         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 400
 401         pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
 402
 403         tcf_destroy_chain(p->filter_list);
 404         qdisc_destroy(p->q);
 405         kfree(p->mask);
 406 }
 407
 408 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 409                              struct sk_buff *skb, struct tcmsg *tcm)
 410 {
 411         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 412         struct rtattr *opts = NULL;
 413
 414         pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
 415
 416         if (!dsmark_valid_index(p, cl))
 417                 return -EINVAL;
 418
 419         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
 420         tcm->tcm_info = p->q->handle;
 421
 422         opts = RTA_NEST(skb, TCA_OPTIONS);
 423         RTA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
 424         RTA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
 425
 426         return RTA_NEST_END(skb, opts);
 427
 428 rtattr_failure:
 429         return RTA_NEST_CANCEL(skb, opts);
 430 }
 431
 432 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 433 {
 434         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 435         struct rtattr *opts = NULL;
 436
 437         opts = RTA_NEST(skb, TCA_OPTIONS);
 438         RTA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
 439
 440         if (p->default_index != NO_DEFAULT_INDEX)
 441                 RTA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
 442
 443         if (p->set_tc_index)
 444                 RTA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
 445
 446         return RTA_NEST_END(skb, opts);
 447
 448 rtattr_failure:
 449         return RTA_NEST_CANCEL(skb, opts);
 450 }
 451
 452 static const struct Qdisc_class_ops dsmark_class_ops = {
 453         .graft          =       dsmark_graft,
 454         .leaf           =       dsmark_leaf,
 455         .get            =       dsmark_get,
 456         .put            =       dsmark_put,
 457         .change         =       dsmark_change,
 458         .delete         =       dsmark_delete,
 459         .walk           =       dsmark_walk,
 460         .tcf_chain      =       dsmark_find_tcf,
 461         .bind_tcf       =       dsmark_bind_filter,
 462         .unbind_tcf     =       dsmark_put,
 463         .dump           =       dsmark_dump_class,
 464 };
 465
 466 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 467         .next           =       NULL,
 468         .cl_ops         =       &dsmark_class_ops,
 469         .id             =       "dsmark",
 470         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 471         .enqueue        =       dsmark_enqueue,
 472         .dequeue        =       dsmark_dequeue,
 473         .requeue        =       dsmark_requeue,
 474         .drop           =       dsmark_drop,
 475         .init           =       dsmark_init,
 476         .reset          =       dsmark_reset,
 477         .destroy        =       dsmark_destroy,
 478         .change         =       NULL,
 479         .dump           =       dsmark_dump,
 480         .owner          =       THIS_MODULE,
 481 };
 482
 483 static int __init dsmark_module_init(void)
 484 {
 485         return register_qdisc(&dsmark_qdisc_ops);
 486 }
 487
 488 static void __exit dsmark_module_exit(void)
 489 {
 490         unregister_qdisc(&dsmark_qdisc_ops);
 491 }
 492
 493 module_init(dsmark_module_init)
 494 module_exit(dsmark_module_exit)
 495
 496 MODULE_LICENSE("GPL");