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)
 120                 goto errout;
 121
 122         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL);
 123         if (err < 0)
 124                 return err;
 125
 126         err = -EINVAL;
 127         if (tb[TCA_DSMARK_MASK]) {
 128                 if (nla_len(tb[TCA_DSMARK_MASK]) < sizeof(u8))
 129                         goto errout;
 130                 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
 131         }
 132         if (tb[TCA_DSMARK_VALUE]) {
 133                 if (nla_len(tb[TCA_DSMARK_VALUE]) < sizeof(u8))
 134                         goto errout;
 135                 p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
 136         }
 137
 138         if (tb[TCA_DSMARK_MASK])
 139                 p->mask[*arg-1] = mask;
 140
 141         err = 0;
 142
 143 errout:
 144         return err;
 145 }
 146
 147 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 148 {
 149         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 150
 151         if (!dsmark_valid_index(p, arg))
 152                 return -EINVAL;
 153
 154         p->mask[arg-1] = 0xff;
 155         p->value[arg-1] = 0;
 156
 157         return 0;
 158 }
 159
 160 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 161 {
 162         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 163         int i;
 164
 165         pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
 166
 167         if (walker->stop)
 168                 return;
 169
 170         for (i = 0; i < p->indices; i++) {
 171                 if (p->mask[i] == 0xff && !p->value[i])
 172                         goto ignore;
 173                 if (walker->count >= walker->skip) {
 174                         if (walker->fn(sch, i+1, walker) < 0) {
 175                                 walker->stop = 1;
 176                                 break;
 177                         }
 178                 }
 179 ignore:
 180                 walker->count++;
 181         }
 182 }
 183
 184 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
 185                                                  unsigned long cl)
 186 {
 187         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 188         return &p->filter_list;
 189 }
 190
 191 /* --------------------------- Qdisc operations ---------------------------- */
 192
 193 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 194 {
 195         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 196         int err;
 197
 198         pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 199
 200         if (p->set_tc_index) {
 201                 switch (skb->protocol) {
 202                 case __constant_htons(ETH_P_IP):
 203                         if (skb_cow_head(skb, sizeof(struct iphdr)))
 204                                 goto drop;
 205
 206                         skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
 207                                 & ~INET_ECN_MASK;
 208                         break;
 209
 210                 case __constant_htons(ETH_P_IPV6):
 211                         if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
 212                                 goto drop;
 213
 214                         skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 215                                 & ~INET_ECN_MASK;
 216                         break;
 217                 default:
 218                         skb->tc_index = 0;
 219                         break;
 220                 }
 221         }
 222
 223         if (TC_H_MAJ(skb->priority) == sch->handle)
 224                 skb->tc_index = TC_H_MIN(skb->priority);
 225         else {
 226                 struct tcf_result res;
 227                 int result = tc_classify(skb, p->filter_list, &res);
 228
 229                 pr_debug("result %d class 0x%04x\n", result, res.classid);
 230
 231                 switch (result) {
 232 #ifdef CONFIG_NET_CLS_ACT
 233                 case TC_ACT_QUEUED:
 234                 case TC_ACT_STOLEN:
 235                         kfree_skb(skb);
 236                         return NET_XMIT_SUCCESS;
 237
 238                 case TC_ACT_SHOT:
 239                         goto drop;
 240 #endif
 241                 case TC_ACT_OK:
 242                         skb->tc_index = TC_H_MIN(res.classid);
 243                         break;
 244
 245                 default:
 246                         if (p->default_index != NO_DEFAULT_INDEX)
 247                                 skb->tc_index = p->default_index;
 248                         break;
 249                 }
 250         }
 251
 252         err = p->q->enqueue(skb, p->q);
 253         if (err != NET_XMIT_SUCCESS) {
 254                 sch->qstats.drops++;
 255                 return err;
 256         }
 257
 258         sch->bstats.bytes += skb->len;
 259         sch->bstats.packets++;
 260         sch->q.qlen++;
 261
 262         return NET_XMIT_SUCCESS;
 263
 264 drop:
 265         kfree_skb(skb);
 266         sch->qstats.drops++;
 267         return NET_XMIT_BYPASS;
 268 }
 269
 270 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 271 {
 272         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 273         struct sk_buff *skb;
 274         u32 index;
 275
 276         pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
 277
 278         skb = p->q->ops->dequeue(p->q);
 279         if (skb == NULL)
 280                 return NULL;
 281
 282         sch->q.qlen--;
 283
 284         index = skb->tc_index & (p->indices - 1);
 285         pr_debug("index %d->%d\n", skb->tc_index, index);
 286
 287         switch (skb->protocol) {
 288         case __constant_htons(ETH_P_IP):
 289                 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
 290                                     p->value[index]);
 291                         break;
 292         case __constant_htons(ETH_P_IPV6):
 293                 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
 294                                     p->value[index]);
 295                         break;
 296         default:
 297                 /*
 298                  * Only complain if a change was actually attempted.
 299                  * This way, we can send non-IP traffic through dsmark
 300                  * and don't need yet another qdisc as a bypass.
 301                  */
 302                 if (p->mask[index] != 0xff || p->value[index])
 303                         printk(KERN_WARNING
 304                                "dsmark_dequeue: unsupported protocol %d\n",
 305                                ntohs(skb->protocol));
 306                 break;
 307         }
 308
 309         return skb;
 310 }
 311
 312 static int dsmark_requeue(struct sk_buff *skb, struct Qdisc *sch)
 313 {
 314         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 315         int err;
 316
 317         pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
 318
 319         err = p->q->ops->requeue(skb, p->q);
 320         if (err != NET_XMIT_SUCCESS) {
 321                 sch->qstats.drops++;
 322                 return err;
 323         }
 324
 325         sch->q.qlen++;
 326         sch->qstats.requeues++;
 327
 328         return NET_XMIT_SUCCESS;
 329 }
 330
 331 static unsigned int dsmark_drop(struct Qdisc *sch)
 332 {
 333         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 334         unsigned int len;
 335
 336         pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 337
 338         if (p->q->ops->drop == NULL)
 339                 return 0;
 340
 341         len = p->q->ops->drop(p->q);
 342         if (len)
 343                 sch->q.qlen--;
 344
 345         return len;
 346 }
 347
 348 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
 349 {
 350         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 351         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 352         int err = -EINVAL;
 353         u32 default_index = NO_DEFAULT_INDEX;
 354         u16 indices;
 355         u8 *mask;
 356
 357         pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
 358
 359         if (!opt)
 360                 goto errout;
 361
 362         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL);
 363         if (err < 0)
 364                 goto errout;
 365
 366         err = -EINVAL;
 367         if (nla_len(tb[TCA_DSMARK_INDICES]) < sizeof(u16))
 368                 goto errout;
 369         indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
 370
 371         if (hweight32(indices) != 1)
 372                 goto errout;
 373
 374         if (tb[TCA_DSMARK_DEFAULT_INDEX]) {
 375                 if (nla_len(tb[TCA_DSMARK_DEFAULT_INDEX]) < sizeof(u16))
 376                         goto errout;
 377                 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
 378         }
 379
 380         mask = kmalloc(indices * 2, GFP_KERNEL);
 381         if (mask == NULL) {
 382                 err = -ENOMEM;
 383                 goto errout;
 384         }
 385
 386         p->mask = mask;
 387         memset(p->mask, 0xff, indices);
 388
 389         p->value = p->mask + indices;
 390         memset(p->value, 0, indices);
 391
 392         p->indices = indices;
 393         p->default_index = default_index;
 394         p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
 395
 396         p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
 397         if (p->q == NULL)
 398                 p->q = &noop_qdisc;
 399
 400         pr_debug("dsmark_init: qdisc %p\n", p->q);
 401
 402         err = 0;
 403 errout:
 404         return err;
 405 }
 406
 407 static void dsmark_reset(struct Qdisc *sch)
 408 {
 409         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 410
 411         pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
 412         qdisc_reset(p->q);
 413         sch->q.qlen = 0;
 414 }
 415
 416 static void dsmark_destroy(struct Qdisc *sch)
 417 {
 418         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 419
 420         pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
 421
 422         tcf_destroy_chain(p->filter_list);
 423         qdisc_destroy(p->q);
 424         kfree(p->mask);
 425 }
 426
 427 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 428                              struct sk_buff *skb, struct tcmsg *tcm)
 429 {
 430         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 431         struct nlattr *opts = NULL;
 432
 433         pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
 434
 435         if (!dsmark_valid_index(p, cl))
 436                 return -EINVAL;
 437
 438         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
 439         tcm->tcm_info = p->q->handle;
 440
 441         opts = nla_nest_start(skb, TCA_OPTIONS);
 442         if (opts == NULL)
 443                 goto nla_put_failure;
 444         NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
 445         NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
 446
 447         return nla_nest_end(skb, opts);
 448
 449 nla_put_failure:
 450         return nla_nest_cancel(skb, opts);
 451 }
 452
 453 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 454 {
 455         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 456         struct nlattr *opts = NULL;
 457
 458         opts = nla_nest_start(skb, TCA_OPTIONS);
 459         if (opts == NULL)
 460                 goto nla_put_failure;
 461         NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
 462
 463         if (p->default_index != NO_DEFAULT_INDEX)
 464                 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
 465
 466         if (p->set_tc_index)
 467                 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
 468
 469         return nla_nest_end(skb, opts);
 470
 471 nla_put_failure:
 472         return nla_nest_cancel(skb, opts);
 473 }
 474
 475 static const struct Qdisc_class_ops dsmark_class_ops = {
 476         .graft          =       dsmark_graft,
 477         .leaf           =       dsmark_leaf,
 478         .get            =       dsmark_get,
 479         .put            =       dsmark_put,
 480         .change         =       dsmark_change,
 481         .delete         =       dsmark_delete,
 482         .walk           =       dsmark_walk,
 483         .tcf_chain      =       dsmark_find_tcf,
 484         .bind_tcf       =       dsmark_bind_filter,
 485         .unbind_tcf     =       dsmark_put,
 486         .dump           =       dsmark_dump_class,
 487 };
 488
 489 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 490         .next           =       NULL,
 491         .cl_ops         =       &dsmark_class_ops,
 492         .id             =       "dsmark",
 493         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 494         .enqueue        =       dsmark_enqueue,
 495         .dequeue        =       dsmark_dequeue,
 496         .requeue        =       dsmark_requeue,
 497         .drop           =       dsmark_drop,
 498         .init           =       dsmark_init,
 499         .reset          =       dsmark_reset,
 500         .destroy        =       dsmark_destroy,
 501         .change         =       NULL,
 502         .dump           =       dsmark_dump,
 503         .owner          =       THIS_MODULE,
 504 };
 505
 506 static int __init dsmark_module_init(void)
 507 {
 508         return register_qdisc(&dsmark_qdisc_ops);
 509 }
 510
 511 static void __exit dsmark_module_exit(void)
 512 {
 513         unregister_qdisc(&dsmark_qdisc_ops);
 514 }
 515
 516 module_init(dsmark_module_init)
 517 module_exit(dsmark_module_exit)
 518
 519 MODULE_LICENSE("GPL");