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