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/netdevice.h> /* for pkt_sched */
  13 #include <linux/rtnetlink.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 #if 0 /* control */
  21 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
  22 #else
  23 #define DPRINTK(format,args...)
  24 #endif
  25
  26 #if 0 /* data */
  27 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
  28 #else
  29 #define D2PRINTK(format,args...)
  30 #endif
  31
  32
  33 #define PRIV(sch) ((struct dsmark_qdisc_data *) qdisc_priv(sch))
  34
  35
  36 /*
  37  * classid      class           marking
  38  * -------      -----           -------
  39  *   n/a          0             n/a
  40  *   x:0          1             use entry [0]
  41  *   ...         ...            ...
  42  *   x:y y>0     y+1            use entry [y]
  43  *   ...         ...            ...
  44  * x:indices-1  indices         use entry [indices-1]
  45  *   ...         ...            ...
  46  *   x:y         y+1            use entry [y & (indices-1)]
  47  *   ...         ...            ...
  48  * 0xffff       0x10000         use entry [indices-1]
  49  */
  50
  51
  52 #define NO_DEFAULT_INDEX        (1 << 16)
  53
  54 struct dsmark_qdisc_data {
  55         struct Qdisc            *q;
  56         struct tcf_proto        *filter_list;
  57         u8                      *mask;  /* "owns" the array */
  58         u8                      *value;
  59         u16                     indices;
  60         u32                     default_index;  /* index range is 0...0xffff */
  61         int                     set_tc_index;
  62 };
  63
  64 static inline int dsmark_valid_indices(u16 indices)
  65 {
  66         while (indices != 1) {
  67                 if (indices & 1)
  68                         return 0;
  69                 indices >>= 1;
  70         }
  71
  72         return 1;
  73 }
  74
  75 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  76 {
  77         return (index <= p->indices && index > 0);
  78 }
  79
  80 /* ------------------------- Class/flow operations ------------------------- */
  81
  82 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  83                         struct Qdisc *new, struct Qdisc **old)
  84 {
  85         struct dsmark_qdisc_data *p = PRIV(sch);
  86
  87         DPRINTK("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
  88                 sch, p, new, old);
  89
  90         if (new == NULL) {
  91                 new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
  92                 if (new == NULL)
  93                         new = &noop_qdisc;
  94         }
  95
  96         sch_tree_lock(sch);
  97         *old = xchg(&p->q, new);
  98         qdisc_reset(*old);
  99         sch->q.qlen = 0;
 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(skb->nh.iph)
 219                                         & ~INET_ECN_MASK;
 220                                 break;
 221                         case __constant_htons(ETH_P_IPV6):
 222                                 skb->tc_index = ipv6_get_dsfield(skb->nh.ipv6h)
 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_POLICE
 241                         case TC_POLICE_SHOT:
 242                                 kfree_skb(skb);
 243                                 sch->qstats.drops++;
 244                                 return NET_XMIT_POLICED;
 245 #if 0
 246                         case TC_POLICE_RECLASSIFY:
 247                                 /* FIXME: what to do here ??? */
 248 #endif
 249 #endif
 250                         case TC_POLICE_OK:
 251                                 skb->tc_index = TC_H_MIN(res.classid);
 252                                 break;
 253                         case TC_POLICE_UNSPEC:
 254                                 /* fall through */
 255                         default:
 256                                 if (p->default_index != NO_DEFAULT_INDEX)
 257                                         skb->tc_index = p->default_index;
 258                                 break;
 259                 };
 260         }
 261
 262         err = p->q->enqueue(skb,p->q);
 263         if (err != NET_XMIT_SUCCESS) {
 264                 sch->qstats.drops++;
 265                 return err;
 266         }
 267
 268         sch->bstats.bytes += skb->len;
 269         sch->bstats.packets++;
 270         sch->q.qlen++;
 271
 272         return NET_XMIT_SUCCESS;
 273 }
 274
 275 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 276 {
 277         struct dsmark_qdisc_data *p = PRIV(sch);
 278         struct sk_buff *skb;
 279         u32 index;
 280
 281         D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
 282
 283         skb = p->q->ops->dequeue(p->q);
 284         if (skb == NULL)
 285                 return NULL;
 286
 287         sch->q.qlen--;
 288
 289         index = skb->tc_index & (p->indices - 1);
 290         D2PRINTK("index %d->%d\n", skb->tc_index, index);
 291
 292         switch (skb->protocol) {
 293                 case __constant_htons(ETH_P_IP):
 294                         ipv4_change_dsfield(skb->nh.iph, p->mask[index],
 295                                             p->value[index]);
 296                         break;
 297                 case __constant_htons(ETH_P_IPV6):
 298                         ipv6_change_dsfield(skb->nh.ipv6h, p->mask[index],
 299                                             p->value[index]);
 300                         break;
 301                 default:
 302                         /*
 303                          * Only complain if a change was actually attempted.
 304                          * This way, we can send non-IP traffic through dsmark
 305                          * and don't need yet another qdisc as a bypass.
 306                          */
 307                         if (p->mask[index] != 0xff || p->value[index])
 308                                 printk(KERN_WARNING "dsmark_dequeue: "
 309                                        "unsupported protocol %d\n",
 310                                        htons(skb->protocol));
 311                         break;
 312         };
 313
 314         return skb;
 315 }
 316
 317 static int dsmark_requeue(struct sk_buff *skb,struct Qdisc *sch)
 318 {
 319         struct dsmark_qdisc_data *p = PRIV(sch);
 320         int err;
 321
 322         D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 323
 324         err = p->q->ops->requeue(skb, p->q);
 325         if (err != NET_XMIT_SUCCESS) {
 326                 sch->qstats.drops++;
 327                 return err;
 328         }
 329
 330         sch->q.qlen++;
 331         sch->qstats.requeues++;
 332
 333         return NET_XMIT_SUCCESS;
 334 }
 335
 336 static unsigned int dsmark_drop(struct Qdisc *sch)
 337 {
 338         struct dsmark_qdisc_data *p = PRIV(sch);
 339         unsigned int len;
 340
 341         DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 342
 343         if (p->q->ops->drop == NULL)
 344                 return 0;
 345
 346         len = p->q->ops->drop(p->q);
 347         if (len)
 348                 sch->q.qlen--;
 349
 350         return len;
 351 }
 352
 353 static int dsmark_init(struct Qdisc *sch, struct rtattr *opt)
 354 {
 355         struct dsmark_qdisc_data *p = PRIV(sch);
 356         struct rtattr *tb[TCA_DSMARK_MAX];
 357         int err = -EINVAL;
 358         u32 default_index = NO_DEFAULT_INDEX;
 359         u16 indices;
 360         u8 *mask;
 361
 362         DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
 363
 364         if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt) < 0)
 365                 goto errout;
 366
 367         indices = RTA_GET_U16(tb[TCA_DSMARK_INDICES-1]);
 368         if (!indices || !dsmark_valid_indices(indices))
 369                 goto errout;
 370
 371         if (tb[TCA_DSMARK_DEFAULT_INDEX-1])
 372                 default_index = RTA_GET_U16(tb[TCA_DSMARK_DEFAULT_INDEX-1]);
 373
 374         mask = kmalloc(indices * 2, GFP_KERNEL);
 375         if (mask == NULL) {
 376                 err = -ENOMEM;
 377                 goto errout;
 378         }
 379
 380         p->mask = mask;
 381         memset(p->mask, 0xff, indices);
 382
 383         p->value = p->mask + indices;
 384         memset(p->value, 0, indices);
 385
 386         p->indices = indices;
 387         p->default_index = default_index;
 388         p->set_tc_index = RTA_GET_FLAG(tb[TCA_DSMARK_SET_TC_INDEX-1]);
 389
 390         p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
 391         if (p->q == NULL)
 392                 p->q = &noop_qdisc;
 393
 394         DPRINTK("dsmark_init: qdisc %p\n", p->q);
 395
 396         err = 0;
 397 errout:
 398 rtattr_failure:
 399         return err;
 400 }
 401
 402 static void dsmark_reset(struct Qdisc *sch)
 403 {
 404         struct dsmark_qdisc_data *p = PRIV(sch);
 405
 406         DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 407         qdisc_reset(p->q);
 408         sch->q.qlen = 0;
 409 }
 410
 411 static void dsmark_destroy(struct Qdisc *sch)
 412 {
 413         struct dsmark_qdisc_data *p = PRIV(sch);
 414         struct tcf_proto *tp;
 415
 416         DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
 417
 418         while (p->filter_list) {
 419                 tp = p->filter_list;
 420                 p->filter_list = tp->next;
 421                 tcf_destroy(tp);
 422         }
 423
 424         qdisc_destroy(p->q);
 425         kfree(p->mask);
 426 }
 427
 428 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 429                              struct sk_buff *skb, struct tcmsg *tcm)
 430 {
 431         struct dsmark_qdisc_data *p = PRIV(sch);
 432         struct rtattr *opts = NULL;
 433
 434         DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
 435
 436         if (!dsmark_valid_index(p, cl))
 437                 return -EINVAL;
 438
 439         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
 440         tcm->tcm_info = p->q->handle;
 441
 442         opts = RTA_NEST(skb, TCA_OPTIONS);
 443         RTA_PUT_U8(skb,TCA_DSMARK_MASK, p->mask[cl-1]);
 444         RTA_PUT_U8(skb,TCA_DSMARK_VALUE, p->value[cl-1]);
 445
 446         return RTA_NEST_END(skb, opts);
 447
 448 rtattr_failure:
 449         return RTA_NEST_CANCEL(skb, opts);
 450 }
 451
 452 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 453 {
 454         struct dsmark_qdisc_data *p = PRIV(sch);
 455         struct rtattr *opts = NULL;
 456
 457         opts = RTA_NEST(skb, TCA_OPTIONS);
 458         RTA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
 459
 460         if (p->default_index != NO_DEFAULT_INDEX)
 461                 RTA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
 462
 463         if (p->set_tc_index)
 464                 RTA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
 465
 466         return RTA_NEST_END(skb, opts);
 467
 468 rtattr_failure:
 469         return RTA_NEST_CANCEL(skb, opts);
 470 }
 471
 472 static struct Qdisc_class_ops dsmark_class_ops = {
 473         .graft          =       dsmark_graft,
 474         .leaf           =       dsmark_leaf,
 475         .get            =       dsmark_get,
 476         .put            =       dsmark_put,
 477         .change         =       dsmark_change,
 478         .delete         =       dsmark_delete,
 479         .walk           =       dsmark_walk,
 480         .tcf_chain      =       dsmark_find_tcf,
 481         .bind_tcf       =       dsmark_bind_filter,
 482         .unbind_tcf     =       dsmark_put,
 483         .dump           =       dsmark_dump_class,
 484 };
 485
 486 static struct Qdisc_ops dsmark_qdisc_ops = {
 487         .next           =       NULL,
 488         .cl_ops         =       &dsmark_class_ops,
 489         .id             =       "dsmark",
 490         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 491         .enqueue        =       dsmark_enqueue,
 492         .dequeue        =       dsmark_dequeue,
 493         .requeue        =       dsmark_requeue,
 494         .drop           =       dsmark_drop,
 495         .init           =       dsmark_init,
 496         .reset          =       dsmark_reset,
 497         .destroy        =       dsmark_destroy,
 498         .change         =       NULL,
 499         .dump           =       dsmark_dump,
 500         .owner          =       THIS_MODULE,
 501 };
 502
 503 static int __init dsmark_module_init(void)
 504 {
 505         return register_qdisc(&dsmark_qdisc_ops);
 506 }
 507
 508 static void __exit dsmark_module_exit(void)
 509 {
 510         unregister_qdisc(&dsmark_qdisc_ops);
 511 }
 512
 513 module_init(dsmark_module_init)
 514 module_exit(dsmark_module_exit)
 515
 516 MODULE_LICENSE("GPL");