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 <net/pkt_sched.h>
  14 #include <net/dsfield.h>
  15 #include <net/inet_ecn.h>
  16 #include <asm/byteorder.h>
  17
  18
  19 #if 0 /* control */
  20 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
  21 #else
  22 #define DPRINTK(format,args...)
  23 #endif
  24
  25 #if 0 /* data */
  26 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
  27 #else
  28 #define D2PRINTK(format,args...)
  29 #endif
  30
  31
  32 #define PRIV(sch) ((struct dsmark_qdisc_data *) qdisc_priv(sch))
  33
  34
  35 /*
  36  * classid      class           marking
  37  * -------      -----           -------
  38  *   n/a          0             n/a
  39  *   x:0          1             use entry [0]
  40  *   ...         ...            ...
  41  *   x:y y>0     y+1            use entry [y]
  42  *   ...         ...            ...
  43  * x:indices-1  indices         use entry [indices-1]
  44  *   ...         ...            ...
  45  *   x:y         y+1            use entry [y & (indices-1)]
  46  *   ...         ...            ...
  47  * 0xffff       0x10000         use entry [indices-1]
  48  */
  49
  50
  51 #define NO_DEFAULT_INDEX        (1 << 16)
  52
  53 struct dsmark_qdisc_data {
  54         struct Qdisc            *q;
  55         struct tcf_proto        *filter_list;
  56         u8                      *mask;  /* "owns" the array */
  57         u8                      *value;
  58         u16                     indices;
  59         u32                     default_index;  /* index range is 0...0xffff */
  60         int                     set_tc_index;
  61 };
  62
  63 static inline int dsmark_valid_indices(u16 indices)
  64 {
  65         while (indices != 1) {
  66                 if (indices & 1)
  67                         return 0;
  68                 indices >>= 1;
  69         }
  70
  71         return 1;
  72 }
  73
  74 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  75 {
  76         return (index <= p->indices && index > 0);
  77 }
  78
  79 /* ------------------------- Class/flow operations ------------------------- */
  80
  81 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  82                         struct Qdisc *new, struct Qdisc **old)
  83 {
  84         struct dsmark_qdisc_data *p = PRIV(sch);
  85
  86         DPRINTK("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
  87                 sch, p, new, old);
  88
  89         if (new == NULL) {
  90                 new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
  91                                         sch->handle);
  92                 if (new == NULL)
  93                         new = &noop_qdisc;
  94         }
  95
  96         sch_tree_lock(sch);
  97         *old = xchg(&p->q, new);
  98         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
  99         qdisc_reset(*old);
 100         sch_tree_unlock(sch);
 101
 102         return 0;
 103 }
 104
 105 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
 106 {
 107         return PRIV(sch)->q;
 108 }
 109
 110 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
 111 {
 112         DPRINTK("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
 113                 sch, PRIV(sch), classid);
 114
 115         return TC_H_MIN(classid) + 1;
 116 }
 117
 118 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
 119                                         unsigned long parent, u32 classid)
 120 {
 121         return dsmark_get(sch, classid);
 122 }
 123
 124 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
 125 {
 126 }
 127
 128 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
 129                          struct rtattr **tca, unsigned long *arg)
 130 {
 131         struct dsmark_qdisc_data *p = PRIV(sch);
 132         struct rtattr *opt = tca[TCA_OPTIONS-1];
 133         struct rtattr *tb[TCA_DSMARK_MAX];
 134         int err = -EINVAL;
 135         u8 mask = 0;
 136
 137         DPRINTK("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
 138                 "arg 0x%lx\n", sch, p, classid, parent, *arg);
 139
 140         if (!dsmark_valid_index(p, *arg)) {
 141                 err = -ENOENT;
 142                 goto rtattr_failure;
 143         }
 144
 145         if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt))
 146                 goto rtattr_failure;
 147
 148         if (tb[TCA_DSMARK_MASK-1])
 149                 mask = RTA_GET_U8(tb[TCA_DSMARK_MASK-1]);
 150
 151         if (tb[TCA_DSMARK_VALUE-1])
 152                 p->value[*arg-1] = RTA_GET_U8(tb[TCA_DSMARK_VALUE-1]);
 153
 154         if (tb[TCA_DSMARK_MASK-1])
 155                 p->mask[*arg-1] = mask;
 156
 157         err = 0;
 158
 159 rtattr_failure:
 160         return err;
 161 }
 162
 163 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 164 {
 165         struct dsmark_qdisc_data *p = PRIV(sch);
 166
 167         if (!dsmark_valid_index(p, arg))
 168                 return -EINVAL;
 169
 170         p->mask[arg-1] = 0xff;
 171         p->value[arg-1] = 0;
 172
 173         return 0;
 174 }
 175
 176 static void dsmark_walk(struct Qdisc *sch,struct qdisc_walker *walker)
 177 {
 178         struct dsmark_qdisc_data *p = PRIV(sch);
 179         int i;
 180
 181         DPRINTK("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
 182
 183         if (walker->stop)
 184                 return;
 185
 186         for (i = 0; i < p->indices; i++) {
 187                 if (p->mask[i] == 0xff && !p->value[i])
 188                         goto ignore;
 189                 if (walker->count >= walker->skip) {
 190                         if (walker->fn(sch, i+1, walker) < 0) {
 191                                 walker->stop = 1;
 192                                 break;
 193                         }
 194                 }
 195 ignore:
 196                 walker->count++;
 197         }
 198 }
 199
 200 static struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,unsigned long cl)
 201 {
 202         return &PRIV(sch)->filter_list;
 203 }
 204
 205 /* --------------------------- Qdisc operations ---------------------------- */
 206
 207 static int dsmark_enqueue(struct sk_buff *skb,struct Qdisc *sch)
 208 {
 209         struct dsmark_qdisc_data *p = PRIV(sch);
 210         int err;
 211
 212         D2PRINTK("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 213
 214         if (p->set_tc_index) {
 215                 /* FIXME: Safe with non-linear skbs? --RR */
 216                 switch (skb->protocol) {
 217                         case __constant_htons(ETH_P_IP):
 218                                 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
 219                                         & ~INET_ECN_MASK;
 220                                 break;
 221                         case __constant_htons(ETH_P_IPV6):
 222                                 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 223                                         & ~INET_ECN_MASK;
 224                                 break;
 225                         default:
 226                                 skb->tc_index = 0;
 227                                 break;
 228                 }
 229         }
 230
 231         if (TC_H_MAJ(skb->priority) == sch->handle)
 232                 skb->tc_index = TC_H_MIN(skb->priority);
 233         else {
 234                 struct tcf_result res;
 235                 int result = tc_classify(skb, p->filter_list, &res);
 236
 237                 D2PRINTK("result %d class 0x%04x\n", result, res.classid);
 238
 239                 switch (result) {
 240 #ifdef CONFIG_NET_CLS_ACT
 241                 case TC_ACT_QUEUED:
 242                 case TC_ACT_STOLEN:
 243                         kfree_skb(skb);
 244                         return NET_XMIT_SUCCESS;
 245                 case TC_ACT_SHOT:
 246                         kfree_skb(skb);
 247                         sch->qstats.drops++;
 248                         return NET_XMIT_BYPASS;
 249 #endif
 250                 case TC_ACT_OK:
 251                         skb->tc_index = TC_H_MIN(res.classid);
 252                         break;
 253                 default:
 254                         if (p->default_index != NO_DEFAULT_INDEX)
 255                                 skb->tc_index = p->default_index;
 256                         break;
 257                 }
 258         }
 259
 260         err = p->q->enqueue(skb,p->q);
 261         if (err != NET_XMIT_SUCCESS) {
 262                 sch->qstats.drops++;
 263                 return err;
 264         }
 265
 266         sch->bstats.bytes += skb->len;
 267         sch->bstats.packets++;
 268         sch->q.qlen++;
 269
 270         return NET_XMIT_SUCCESS;
 271 }
 272
 273 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 274 {
 275         struct dsmark_qdisc_data *p = PRIV(sch);
 276         struct sk_buff *skb;
 277         u32 index;
 278
 279         D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
 280
 281         skb = p->q->ops->dequeue(p->q);
 282         if (skb == NULL)
 283                 return NULL;
 284
 285         sch->q.qlen--;
 286
 287         index = skb->tc_index & (p->indices - 1);
 288         D2PRINTK("index %d->%d\n", skb->tc_index, index);
 289
 290         switch (skb->protocol) {
 291                 case __constant_htons(ETH_P_IP):
 292                         ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
 293                                             p->value[index]);
 294                         break;
 295                 case __constant_htons(ETH_P_IPV6):
 296                         ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
 297                                             p->value[index]);
 298                         break;
 299                 default:
 300                         /*
 301                          * Only complain if a change was actually attempted.
 302                          * This way, we can send non-IP traffic through dsmark
 303                          * and don't need yet another qdisc as a bypass.
 304                          */
 305                         if (p->mask[index] != 0xff || p->value[index])
 306                                 printk(KERN_WARNING "dsmark_dequeue: "
 307                                        "unsupported protocol %d\n",
 308                                        ntohs(skb->protocol));
 309                         break;
 310         }
 311
 312         return skb;
 313 }
 314
 315 static int dsmark_requeue(struct sk_buff *skb,struct Qdisc *sch)
 316 {
 317         struct dsmark_qdisc_data *p = PRIV(sch);
 318         int err;
 319
 320         D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 321
 322         err = p->q->ops->requeue(skb, p->q);
 323         if (err != NET_XMIT_SUCCESS) {
 324                 sch->qstats.drops++;
 325                 return err;
 326         }
 327
 328         sch->q.qlen++;
 329         sch->qstats.requeues++;
 330
 331         return NET_XMIT_SUCCESS;
 332 }
 333
 334 static unsigned int dsmark_drop(struct Qdisc *sch)
 335 {
 336         struct dsmark_qdisc_data *p = PRIV(sch);
 337         unsigned int len;
 338
 339         DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 340
 341         if (p->q->ops->drop == NULL)
 342                 return 0;
 343
 344         len = p->q->ops->drop(p->q);
 345         if (len)
 346                 sch->q.qlen--;
 347
 348         return len;
 349 }
 350
 351 static int dsmark_init(struct Qdisc *sch, struct rtattr *opt)
 352 {
 353         struct dsmark_qdisc_data *p = PRIV(sch);
 354         struct rtattr *tb[TCA_DSMARK_MAX];
 355         int err = -EINVAL;
 356         u32 default_index = NO_DEFAULT_INDEX;
 357         u16 indices;
 358         u8 *mask;
 359
 360         DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
 361
 362         if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt) < 0)
 363                 goto errout;
 364
 365         indices = RTA_GET_U16(tb[TCA_DSMARK_INDICES-1]);
 366         if (!indices || !dsmark_valid_indices(indices))
 367                 goto errout;
 368
 369         if (tb[TCA_DSMARK_DEFAULT_INDEX-1])
 370                 default_index = RTA_GET_U16(tb[TCA_DSMARK_DEFAULT_INDEX-1]);
 371
 372         mask = kmalloc(indices * 2, GFP_KERNEL);
 373         if (mask == NULL) {
 374                 err = -ENOMEM;
 375                 goto errout;
 376         }
 377
 378         p->mask = mask;
 379         memset(p->mask, 0xff, indices);
 380
 381         p->value = p->mask + indices;
 382         memset(p->value, 0, indices);
 383
 384         p->indices = indices;
 385         p->default_index = default_index;
 386         p->set_tc_index = RTA_GET_FLAG(tb[TCA_DSMARK_SET_TC_INDEX-1]);
 387
 388         p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
 389         if (p->q == NULL)
 390                 p->q = &noop_qdisc;
 391
 392         DPRINTK("dsmark_init: qdisc %p\n", p->q);
 393
 394         err = 0;
 395 errout:
 396 rtattr_failure:
 397         return err;
 398 }
 399
 400 static void dsmark_reset(struct Qdisc *sch)
 401 {
 402         struct dsmark_qdisc_data *p = PRIV(sch);
 403
 404         DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 405         qdisc_reset(p->q);
 406         sch->q.qlen = 0;
 407 }
 408
 409 static void dsmark_destroy(struct Qdisc *sch)
 410 {
 411         struct dsmark_qdisc_data *p = PRIV(sch);
 412
 413         DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
 414
 415         tcf_destroy_chain(p->filter_list);
 416         qdisc_destroy(p->q);
 417         kfree(p->mask);
 418 }
 419
 420 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 421                              struct sk_buff *skb, struct tcmsg *tcm)
 422 {
 423         struct dsmark_qdisc_data *p = PRIV(sch);
 424         struct rtattr *opts = NULL;
 425
 426         DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
 427
 428         if (!dsmark_valid_index(p, cl))
 429                 return -EINVAL;
 430
 431         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
 432         tcm->tcm_info = p->q->handle;
 433
 434         opts = RTA_NEST(skb, TCA_OPTIONS);
 435         RTA_PUT_U8(skb,TCA_DSMARK_MASK, p->mask[cl-1]);
 436         RTA_PUT_U8(skb,TCA_DSMARK_VALUE, p->value[cl-1]);
 437
 438         return RTA_NEST_END(skb, opts);
 439
 440 rtattr_failure:
 441         return RTA_NEST_CANCEL(skb, opts);
 442 }
 443
 444 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 445 {
 446         struct dsmark_qdisc_data *p = PRIV(sch);
 447         struct rtattr *opts = NULL;
 448
 449         opts = RTA_NEST(skb, TCA_OPTIONS);
 450         RTA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
 451
 452         if (p->default_index != NO_DEFAULT_INDEX)
 453                 RTA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
 454
 455         if (p->set_tc_index)
 456                 RTA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
 457
 458         return RTA_NEST_END(skb, opts);
 459
 460 rtattr_failure:
 461         return RTA_NEST_CANCEL(skb, opts);
 462 }
 463
 464 static const struct Qdisc_class_ops dsmark_class_ops = {
 465         .graft          =       dsmark_graft,
 466         .leaf           =       dsmark_leaf,
 467         .get            =       dsmark_get,
 468         .put            =       dsmark_put,
 469         .change         =       dsmark_change,
 470         .delete         =       dsmark_delete,
 471         .walk           =       dsmark_walk,
 472         .tcf_chain      =       dsmark_find_tcf,
 473         .bind_tcf       =       dsmark_bind_filter,
 474         .unbind_tcf     =       dsmark_put,
 475         .dump           =       dsmark_dump_class,
 476 };
 477
 478 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 479         .next           =       NULL,
 480         .cl_ops         =       &dsmark_class_ops,
 481         .id             =       "dsmark",
 482         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 483         .enqueue        =       dsmark_enqueue,
 484         .dequeue        =       dsmark_dequeue,
 485         .requeue        =       dsmark_requeue,
 486         .drop           =       dsmark_drop,
 487         .init           =       dsmark_init,
 488         .reset          =       dsmark_reset,
 489         .destroy        =       dsmark_destroy,
 490         .change         =       NULL,
 491         .dump           =       dsmark_dump,
 492         .owner          =       THIS_MODULE,
 493 };
 494
 495 static int __init dsmark_module_init(void)
 496 {
 497         return register_qdisc(&dsmark_qdisc_ops);
 498 }
 499
 500 static void __exit dsmark_module_exit(void)
 501 {
 502         unregister_qdisc(&dsmark_qdisc_ops);
 503 }
 504
 505 module_init(dsmark_module_init)
 506 module_exit(dsmark_module_exit)
 507
 508 MODULE_LICENSE("GPL");