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/slab.h>
   9 #include <linux/types.h>
  10 #include <linux/string.h>
  11 #include <linux/errno.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/rtnetlink.h>
  14 #include <linux/bitops.h>
  15 #include <net/pkt_sched.h>
  16 #include <net/dsfield.h>
  17 #include <net/inet_ecn.h>
  18 #include <asm/byteorder.h>
  19
  20 /*
  21  * classid      class           marking
  22  * -------      -----           -------
  23  *   n/a          0             n/a
  24  *   x:0          1             use entry [0]
  25  *   ...         ...            ...
  26  *   x:y y>0     y+1            use entry [y]
  27  *   ...         ...            ...
  28  * x:indices-1  indices         use entry [indices-1]
  29  *   ...         ...            ...
  30  *   x:y         y+1            use entry [y & (indices-1)]
  31  *   ...         ...            ...
  32  * 0xffff       0x10000         use entry [indices-1]
  33  */
  34
  35
  36 #define NO_DEFAULT_INDEX        (1 << 16)
  37
  38 struct dsmark_qdisc_data {
  39         struct Qdisc            *q;
  40         struct tcf_proto        *filter_list;
  41         u8                      *mask;  /* "owns" the array */
  42         u8                      *value;
  43         u16                     indices;
  44         u32                     default_index;  /* index range is 0...0xffff */
  45         int                     set_tc_index;
  46 };
  47
  48 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  49 {
  50         return index <= p->indices && index > 0;
  51 }
  52
  53 /* ------------------------- Class/flow operations ------------------------- */
  54
  55 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  56                         struct Qdisc *new, struct Qdisc **old)
  57 {
  58         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  59
  60         pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
  61                  __func__, sch, p, new, old);
  62
  63         if (new == NULL) {
  64                 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
  65                                         sch->handle);
  66                 if (new == NULL)
  67                         new = &noop_qdisc;
  68         }
  69
  70         sch_tree_lock(sch);
  71         *old = p->q;
  72         p->q = new;
  73         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
  74         qdisc_reset(*old);
  75         sch_tree_unlock(sch);
  76
  77         return 0;
  78 }
  79
  80 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
  81 {
  82         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  83         return p->q;
  84 }
  85
  86 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
  87 {
  88         pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
  89                  __func__, sch, qdisc_priv(sch), classid);
  90
  91         return TC_H_MIN(classid) + 1;
  92 }
  93
  94 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
  95                                         unsigned long parent, u32 classid)
  96 {
  97         return dsmark_get(sch, classid);
  98 }
  99
 100 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
 101 {
 102 }
 103
 104 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
 105         [TCA_DSMARK_INDICES]            = { .type = NLA_U16 },
 106         [TCA_DSMARK_DEFAULT_INDEX]      = { .type = NLA_U16 },
 107         [TCA_DSMARK_SET_TC_INDEX]       = { .type = NLA_FLAG },
 108         [TCA_DSMARK_MASK]               = { .type = NLA_U8 },
 109         [TCA_DSMARK_VALUE]              = { .type = NLA_U8 },
 110 };
 111
 112 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
 113                          struct nlattr **tca, unsigned long *arg)
 114 {
 115         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 116         struct nlattr *opt = tca[TCA_OPTIONS];
 117         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 118         int err = -EINVAL;
 119         u8 mask = 0;
 120
 121         pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
 122                  __func__, sch, p, classid, parent, *arg);
 123
 124         if (!dsmark_valid_index(p, *arg)) {
 125                 err = -ENOENT;
 126                 goto errout;
 127         }
 128
 129         if (!opt)
 130                 goto errout;
 131
 132         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
 133         if (err < 0)
 134                 goto errout;
 135
 136         if (tb[TCA_DSMARK_MASK])
 137                 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
 138
 139         if (tb[TCA_DSMARK_VALUE])
 140                 p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
 141
 142         if (tb[TCA_DSMARK_MASK])
 143                 p->mask[*arg - 1] = mask;
 144
 145         err = 0;
 146
 147 errout:
 148         return err;
 149 }
 150
 151 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 152 {
 153         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 154
 155         if (!dsmark_valid_index(p, arg))
 156                 return -EINVAL;
 157
 158         p->mask[arg - 1] = 0xff;
 159         p->value[arg - 1] = 0;
 160
 161         return 0;
 162 }
 163
 164 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 165 {
 166         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 167         int i;
 168
 169         pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
 170                  __func__, sch, p, walker);
 171
 172         if (walker->stop)
 173                 return;
 174
 175         for (i = 0; i < p->indices; i++) {
 176                 if (p->mask[i] == 0xff && !p->value[i])
 177                         goto ignore;
 178                 if (walker->count >= walker->skip) {
 179                         if (walker->fn(sch, i + 1, walker) < 0) {
 180                                 walker->stop = 1;
 181                                 break;
 182                         }
 183                 }
 184 ignore:
 185                 walker->count++;
 186         }
 187 }
 188
 189 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
 190                                                  unsigned long cl)
 191 {
 192         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 193         return &p->filter_list;
 194 }
 195
 196 /* --------------------------- Qdisc operations ---------------------------- */
 197
 198 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 199 {
 200         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 201         int err;
 202
 203         pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
 204
 205         if (p->set_tc_index) {
 206                 switch (skb->protocol) {
 207                 case htons(ETH_P_IP):
 208                         if (skb_cow_head(skb, sizeof(struct iphdr)))
 209                                 goto drop;
 210
 211                         skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
 212                                 & ~INET_ECN_MASK;
 213                         break;
 214
 215                 case htons(ETH_P_IPV6):
 216                         if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
 217                                 goto drop;
 218
 219                         skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 220                                 & ~INET_ECN_MASK;
 221                         break;
 222                 default:
 223                         skb->tc_index = 0;
 224                         break;
 225                 }
 226         }
 227
 228         if (TC_H_MAJ(skb->priority) == sch->handle)
 229                 skb->tc_index = TC_H_MIN(skb->priority);
 230         else {
 231                 struct tcf_result res;
 232                 int result = tc_classify(skb, p->filter_list, &res);
 233
 234                 pr_debug("result %d class 0x%04x\n", result, res.classid);
 235
 236                 switch (result) {
 237 #ifdef CONFIG_NET_CLS_ACT
 238                 case TC_ACT_QUEUED:
 239                 case TC_ACT_STOLEN:
 240                         kfree_skb(skb);
 241                         return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 242
 243                 case TC_ACT_SHOT:
 244                         goto drop;
 245 #endif
 246                 case TC_ACT_OK:
 247                         skb->tc_index = TC_H_MIN(res.classid);
 248                         break;
 249
 250                 default:
 251                         if (p->default_index != NO_DEFAULT_INDEX)
 252                                 skb->tc_index = p->default_index;
 253                         break;
 254                 }
 255         }
 256
 257         err = qdisc_enqueue(skb, p->q);
 258         if (err != NET_XMIT_SUCCESS) {
 259                 if (net_xmit_drop_count(err))
 260                         sch->qstats.drops++;
 261                 return err;
 262         }
 263
 264         sch->q.qlen++;
 265
 266         return NET_XMIT_SUCCESS;
 267
 268 drop:
 269         qdisc_drop(skb, sch);
 270         return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 271 }
 272
 273 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 274 {
 275         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 276         struct sk_buff *skb;
 277         u32 index;
 278
 279         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 280
 281         skb = p->q->ops->dequeue(p->q);
 282         if (skb == NULL)
 283                 return NULL;
 284
 285         qdisc_bstats_update(sch, skb);
 286         sch->q.qlen--;
 287
 288         index = skb->tc_index & (p->indices - 1);
 289         pr_debug("index %d->%d\n", skb->tc_index, index);
 290
 291         switch (skb->protocol) {
 292         case htons(ETH_P_IP):
 293                 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
 294                                     p->value[index]);
 295                         break;
 296         case htons(ETH_P_IPV6):
 297                 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
 298                                     p->value[index]);
 299                         break;
 300         default:
 301                 /*
 302                  * Only complain if a change was actually attempted.
 303                  * This way, we can send non-IP traffic through dsmark
 304                  * and don't need yet another qdisc as a bypass.
 305                  */
 306                 if (p->mask[index] != 0xff || p->value[index])
 307                         pr_warn("%s: unsupported protocol %d\n",
 308                                 __func__, ntohs(skb->protocol));
 309                 break;
 310         }
 311
 312         return skb;
 313 }
 314
 315 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
 316 {
 317         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 318
 319         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 320
 321         return p->q->ops->peek(p->q);
 322 }
 323
 324 static unsigned int dsmark_drop(struct Qdisc *sch)
 325 {
 326         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 327         unsigned int len;
 328
 329         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 330
 331         if (p->q->ops->drop == NULL)
 332                 return 0;
 333
 334         len = p->q->ops->drop(p->q);
 335         if (len)
 336                 sch->q.qlen--;
 337
 338         return len;
 339 }
 340
 341 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
 342 {
 343         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 344         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 345         int err = -EINVAL;
 346         u32 default_index = NO_DEFAULT_INDEX;
 347         u16 indices;
 348         u8 *mask;
 349
 350         pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
 351
 352         if (!opt)
 353                 goto errout;
 354
 355         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
 356         if (err < 0)
 357                 goto errout;
 358
 359         err = -EINVAL;
 360         indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
 361
 362         if (hweight32(indices) != 1)
 363                 goto errout;
 364
 365         if (tb[TCA_DSMARK_DEFAULT_INDEX])
 366                 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
 367
 368         mask = kmalloc(indices * 2, GFP_KERNEL);
 369         if (mask == NULL) {
 370                 err = -ENOMEM;
 371                 goto errout;
 372         }
 373
 374         p->mask = mask;
 375         memset(p->mask, 0xff, indices);
 376
 377         p->value = p->mask + indices;
 378         memset(p->value, 0, indices);
 379
 380         p->indices = indices;
 381         p->default_index = default_index;
 382         p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
 383
 384         p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
 385         if (p->q == NULL)
 386                 p->q = &noop_qdisc;
 387
 388         pr_debug("%s: qdisc %p\n", __func__, p->q);
 389
 390         err = 0;
 391 errout:
 392         return err;
 393 }
 394
 395 static void dsmark_reset(struct Qdisc *sch)
 396 {
 397         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 398
 399         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 400         qdisc_reset(p->q);
 401         sch->q.qlen = 0;
 402 }
 403
 404 static void dsmark_destroy(struct Qdisc *sch)
 405 {
 406         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 407
 408         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 409
 410         tcf_destroy_chain(&p->filter_list);
 411         qdisc_destroy(p->q);
 412         kfree(p->mask);
 413 }
 414
 415 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 416                              struct sk_buff *skb, struct tcmsg *tcm)
 417 {
 418         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 419         struct nlattr *opts = NULL;
 420
 421         pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
 422
 423         if (!dsmark_valid_index(p, cl))
 424                 return -EINVAL;
 425
 426         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
 427         tcm->tcm_info = p->q->handle;
 428
 429         opts = nla_nest_start(skb, TCA_OPTIONS);
 430         if (opts == NULL)
 431                 goto nla_put_failure;
 432         if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]) ||
 433             nla_put_u8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]))
 434                 goto nla_put_failure;
 435
 436         return nla_nest_end(skb, opts);
 437
 438 nla_put_failure:
 439         nla_nest_cancel(skb, opts);
 440         return -EMSGSIZE;
 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         if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
 452                 goto nla_put_failure;
 453
 454         if (p->default_index != NO_DEFAULT_INDEX &&
 455             nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
 456                 goto nla_put_failure;
 457
 458         if (p->set_tc_index &&
 459             nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
 460                 goto nla_put_failure;
 461
 462         return nla_nest_end(skb, opts);
 463
 464 nla_put_failure:
 465         nla_nest_cancel(skb, opts);
 466         return -EMSGSIZE;
 467 }
 468
 469 static const struct Qdisc_class_ops dsmark_class_ops = {
 470         .graft          =       dsmark_graft,
 471         .leaf           =       dsmark_leaf,
 472         .get            =       dsmark_get,
 473         .put            =       dsmark_put,
 474         .change         =       dsmark_change,
 475         .delete         =       dsmark_delete,
 476         .walk           =       dsmark_walk,
 477         .tcf_chain      =       dsmark_find_tcf,
 478         .bind_tcf       =       dsmark_bind_filter,
 479         .unbind_tcf     =       dsmark_put,
 480         .dump           =       dsmark_dump_class,
 481 };
 482
 483 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 484         .next           =       NULL,
 485         .cl_ops         =       &dsmark_class_ops,
 486         .id             =       "dsmark",
 487         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 488         .enqueue        =       dsmark_enqueue,
 489         .dequeue        =       dsmark_dequeue,
 490         .peek           =       dsmark_peek,
 491         .drop           =       dsmark_drop,
 492         .init           =       dsmark_init,
 493         .reset          =       dsmark_reset,
 494         .destroy        =       dsmark_destroy,
 495         .change         =       NULL,
 496         .dump           =       dsmark_dump,
 497         .owner          =       THIS_MODULE,
 498 };
 499
 500 static int __init dsmark_module_init(void)
 501 {
 502         return register_qdisc(&dsmark_qdisc_ops);
 503 }
 504
 505 static void __exit dsmark_module_exit(void)
 506 {
 507         unregister_qdisc(&dsmark_qdisc_ops);
 508 }
 509
 510 module_init(dsmark_module_init)
 511 module_exit(dsmark_module_exit)
 512
 513 MODULE_LICENSE("GPL");