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/pkt_cls.h>
  17 #include <net/dsfield.h>
  18 #include <net/inet_ecn.h>
  19 #include <asm/byteorder.h>
  20
  21 /*
  22  * classid      class           marking
  23  * -------      -----           -------
  24  *   n/a          0             n/a
  25  *   x:0          1             use entry [0]
  26  *   ...         ...            ...
  27  *   x:y y>0     y+1            use entry [y]
  28  *   ...         ...            ...
  29  * x:indices-1  indices         use entry [indices-1]
  30  *   ...         ...            ...
  31  *   x:y         y+1            use entry [y & (indices-1)]
  32  *   ...         ...            ...
  33  * 0xffff       0x10000         use entry [indices-1]
  34  */
  35
  36
  37 #define NO_DEFAULT_INDEX        (1 << 16)
  38
  39 struct mask_value {
  40         u8                      mask;
  41         u8                      value;
  42 };
  43
  44 struct dsmark_qdisc_data {
  45         struct Qdisc            *q;
  46         struct tcf_proto __rcu  *filter_list;
  47         struct mask_value       *mv;
  48         u16                     indices;
  49         u8                      set_tc_index;
  50         u32                     default_index;  /* index range is 0...0xffff */
  51 #define DSMARK_EMBEDDED_SZ      16
  52         struct mask_value       embedded[DSMARK_EMBEDDED_SZ];
  53 };
  54
  55 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  56 {
  57         return index <= p->indices && index > 0;
  58 }
  59
  60 /* ------------------------- Class/flow operations ------------------------- */
  61
  62 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  63                         struct Qdisc *new, struct Qdisc **old)
  64 {
  65         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  66
  67         pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
  68                  __func__, sch, p, new, old);
  69
  70         if (new == NULL) {
  71                 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
  72                                         sch->handle);
  73                 if (new == NULL)
  74                         new = &noop_qdisc;
  75         }
  76
  77         *old = qdisc_replace(sch, new, &p->q);
  78         return 0;
  79 }
  80
  81 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
  82 {
  83         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  84         return p->q;
  85 }
  86
  87 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
  88 {
  89         pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
  90                  __func__, sch, qdisc_priv(sch), classid);
  91
  92         return TC_H_MIN(classid) + 1;
  93 }
  94
  95 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
  96                                         unsigned long parent, u32 classid)
  97 {
  98         return dsmark_get(sch, classid);
  99 }
 100
 101 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
 102 {
 103 }
 104
 105 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
 106         [TCA_DSMARK_INDICES]            = { .type = NLA_U16 },
 107         [TCA_DSMARK_DEFAULT_INDEX]      = { .type = NLA_U16 },
 108         [TCA_DSMARK_SET_TC_INDEX]       = { .type = NLA_FLAG },
 109         [TCA_DSMARK_MASK]               = { .type = NLA_U8 },
 110         [TCA_DSMARK_VALUE]              = { .type = NLA_U8 },
 111 };
 112
 113 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
 114                          struct nlattr **tca, unsigned long *arg)
 115 {
 116         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 117         struct nlattr *opt = tca[TCA_OPTIONS];
 118         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 119         int err = -EINVAL;
 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, NULL);
 133         if (err < 0)
 134                 goto errout;
 135
 136         if (tb[TCA_DSMARK_VALUE])
 137                 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
 138
 139         if (tb[TCA_DSMARK_MASK])
 140                 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
 141
 142         err = 0;
 143
 144 errout:
 145         return err;
 146 }
 147
 148 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 149 {
 150         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 151
 152         if (!dsmark_valid_index(p, arg))
 153                 return -EINVAL;
 154
 155         p->mv[arg - 1].mask = 0xff;
 156         p->mv[arg - 1].value = 0;
 157
 158         return 0;
 159 }
 160
 161 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 162 {
 163         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 164         int i;
 165
 166         pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
 167                  __func__, sch, p, walker);
 168
 169         if (walker->stop)
 170                 return;
 171
 172         for (i = 0; i < p->indices; i++) {
 173                 if (p->mv[i].mask == 0xff && !p->mv[i].value)
 174                         goto ignore;
 175                 if (walker->count >= walker->skip) {
 176                         if (walker->fn(sch, i + 1, walker) < 0) {
 177                                 walker->stop = 1;
 178                                 break;
 179                         }
 180                 }
 181 ignore:
 182                 walker->count++;
 183         }
 184 }
 185
 186 static inline struct tcf_proto __rcu **dsmark_find_tcf(struct Qdisc *sch,
 187                                                        unsigned long cl)
 188 {
 189         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 190         return &p->filter_list;
 191 }
 192
 193 /* --------------------------- Qdisc operations ---------------------------- */
 194
 195 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 196                           struct sk_buff **to_free)
 197 {
 198         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 199         int err;
 200
 201         pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
 202
 203         if (p->set_tc_index) {
 204                 int wlen = skb_network_offset(skb);
 205
 206                 switch (tc_skb_protocol(skb)) {
 207                 case htons(ETH_P_IP):
 208                         wlen += sizeof(struct iphdr);
 209                         if (!pskb_may_pull(skb, wlen) ||
 210                             skb_try_make_writable(skb, wlen))
 211                                 goto drop;
 212
 213                         skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
 214                                 & ~INET_ECN_MASK;
 215                         break;
 216
 217                 case htons(ETH_P_IPV6):
 218                         wlen += sizeof(struct ipv6hdr);
 219                         if (!pskb_may_pull(skb, wlen) ||
 220                             skb_try_make_writable(skb, wlen))
 221                                 goto drop;
 222
 223                         skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 224                                 & ~INET_ECN_MASK;
 225                         break;
 226                 default:
 227                         skb->tc_index = 0;
 228                         break;
 229                 }
 230         }
 231
 232         if (TC_H_MAJ(skb->priority) == sch->handle)
 233                 skb->tc_index = TC_H_MIN(skb->priority);
 234         else {
 235                 struct tcf_result res;
 236                 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
 237                 int result = tc_classify(skb, fl, &res, false);
 238
 239                 pr_debug("result %d class 0x%04x\n", result, res.classid);
 240
 241                 switch (result) {
 242 #ifdef CONFIG_NET_CLS_ACT
 243                 case TC_ACT_QUEUED:
 244                 case TC_ACT_STOLEN:
 245                         __qdisc_drop(skb, to_free);
 246                         return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 247
 248                 case TC_ACT_SHOT:
 249                         goto drop;
 250 #endif
 251                 case TC_ACT_OK:
 252                         skb->tc_index = TC_H_MIN(res.classid);
 253                         break;
 254
 255                 default:
 256                         if (p->default_index != NO_DEFAULT_INDEX)
 257                                 skb->tc_index = p->default_index;
 258                         break;
 259                 }
 260         }
 261
 262         err = qdisc_enqueue(skb, p->q, to_free);
 263         if (err != NET_XMIT_SUCCESS) {
 264                 if (net_xmit_drop_count(err))
 265                         qdisc_qstats_drop(sch);
 266                 return err;
 267         }
 268
 269         qdisc_qstats_backlog_inc(sch, skb);
 270         sch->q.qlen++;
 271
 272         return NET_XMIT_SUCCESS;
 273
 274 drop:
 275         qdisc_drop(skb, sch, to_free);
 276         return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 277 }
 278
 279 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 280 {
 281         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 282         struct sk_buff *skb;
 283         u32 index;
 284
 285         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 286
 287         skb = qdisc_dequeue_peeked(p->q);
 288         if (skb == NULL)
 289                 return NULL;
 290
 291         qdisc_bstats_update(sch, skb);
 292         qdisc_qstats_backlog_dec(sch, skb);
 293         sch->q.qlen--;
 294
 295         index = skb->tc_index & (p->indices - 1);
 296         pr_debug("index %d->%d\n", skb->tc_index, index);
 297
 298         switch (tc_skb_protocol(skb)) {
 299         case htons(ETH_P_IP):
 300                 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
 301                                     p->mv[index].value);
 302                         break;
 303         case htons(ETH_P_IPV6):
 304                 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
 305                                     p->mv[index].value);
 306                         break;
 307         default:
 308                 /*
 309                  * Only complain if a change was actually attempted.
 310                  * This way, we can send non-IP traffic through dsmark
 311                  * and don't need yet another qdisc as a bypass.
 312                  */
 313                 if (p->mv[index].mask != 0xff || p->mv[index].value)
 314                         pr_warn("%s: unsupported protocol %d\n",
 315                                 __func__, ntohs(tc_skb_protocol(skb)));
 316                 break;
 317         }
 318
 319         return skb;
 320 }
 321
 322 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
 323 {
 324         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 325
 326         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 327
 328         return p->q->ops->peek(p->q);
 329 }
 330
 331 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
 332 {
 333         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 334         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 335         int err = -EINVAL;
 336         u32 default_index = NO_DEFAULT_INDEX;
 337         u16 indices;
 338         int i;
 339
 340         pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
 341
 342         if (!opt)
 343                 goto errout;
 344
 345         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
 346         if (err < 0)
 347                 goto errout;
 348
 349         err = -EINVAL;
 350         indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
 351
 352         if (hweight32(indices) != 1)
 353                 goto errout;
 354
 355         if (tb[TCA_DSMARK_DEFAULT_INDEX])
 356                 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
 357
 358         if (indices <= DSMARK_EMBEDDED_SZ)
 359                 p->mv = p->embedded;
 360         else
 361                 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
 362         if (!p->mv) {
 363                 err = -ENOMEM;
 364                 goto errout;
 365         }
 366         for (i = 0; i < indices; i++) {
 367                 p->mv[i].mask = 0xff;
 368                 p->mv[i].value = 0;
 369         }
 370         p->indices = indices;
 371         p->default_index = default_index;
 372         p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
 373
 374         p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
 375         if (p->q == NULL)
 376                 p->q = &noop_qdisc;
 377         else
 378                 qdisc_hash_add(p->q, true);
 379
 380         pr_debug("%s: qdisc %p\n", __func__, p->q);
 381
 382         err = 0;
 383 errout:
 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("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 392         qdisc_reset(p->q);
 393         sch->qstats.backlog = 0;
 394         sch->q.qlen = 0;
 395 }
 396
 397 static void dsmark_destroy(struct Qdisc *sch)
 398 {
 399         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 400
 401         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 402
 403         tcf_destroy_chain(&p->filter_list);
 404         qdisc_destroy(p->q);
 405         if (p->mv != p->embedded)
 406                 kfree(p->mv);
 407 }
 408
 409 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 410                              struct sk_buff *skb, struct tcmsg *tcm)
 411 {
 412         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 413         struct nlattr *opts = NULL;
 414
 415         pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
 416
 417         if (!dsmark_valid_index(p, cl))
 418                 return -EINVAL;
 419
 420         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
 421         tcm->tcm_info = p->q->handle;
 422
 423         opts = nla_nest_start(skb, TCA_OPTIONS);
 424         if (opts == NULL)
 425                 goto nla_put_failure;
 426         if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
 427             nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
 428                 goto nla_put_failure;
 429
 430         return nla_nest_end(skb, opts);
 431
 432 nla_put_failure:
 433         nla_nest_cancel(skb, opts);
 434         return -EMSGSIZE;
 435 }
 436
 437 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 438 {
 439         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 440         struct nlattr *opts = NULL;
 441
 442         opts = nla_nest_start(skb, TCA_OPTIONS);
 443         if (opts == NULL)
 444                 goto nla_put_failure;
 445         if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
 446                 goto nla_put_failure;
 447
 448         if (p->default_index != NO_DEFAULT_INDEX &&
 449             nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
 450                 goto nla_put_failure;
 451
 452         if (p->set_tc_index &&
 453             nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
 454                 goto nla_put_failure;
 455
 456         return nla_nest_end(skb, opts);
 457
 458 nla_put_failure:
 459         nla_nest_cancel(skb, opts);
 460         return -EMSGSIZE;
 461 }
 462
 463 static const struct Qdisc_class_ops dsmark_class_ops = {
 464         .graft          =       dsmark_graft,
 465         .leaf           =       dsmark_leaf,
 466         .get            =       dsmark_get,
 467         .put            =       dsmark_put,
 468         .change         =       dsmark_change,
 469         .delete         =       dsmark_delete,
 470         .walk           =       dsmark_walk,
 471         .tcf_chain      =       dsmark_find_tcf,
 472         .bind_tcf       =       dsmark_bind_filter,
 473         .unbind_tcf     =       dsmark_put,
 474         .dump           =       dsmark_dump_class,
 475 };
 476
 477 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 478         .next           =       NULL,
 479         .cl_ops         =       &dsmark_class_ops,
 480         .id             =       "dsmark",
 481         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 482         .enqueue        =       dsmark_enqueue,
 483         .dequeue        =       dsmark_dequeue,
 484         .peek           =       dsmark_peek,
 485         .init           =       dsmark_init,
 486         .reset          =       dsmark_reset,
 487         .destroy        =       dsmark_destroy,
 488         .change         =       NULL,
 489         .dump           =       dsmark_dump,
 490         .owner          =       THIS_MODULE,
 491 };
 492
 493 static int __init dsmark_module_init(void)
 494 {
 495         return register_qdisc(&dsmark_qdisc_ops);
 496 }
 497
 498 static void __exit dsmark_module_exit(void)
 499 {
 500         unregister_qdisc(&dsmark_qdisc_ops);
 501 }
 502
 503 module_init(dsmark_module_init)
 504 module_exit(dsmark_module_exit)
 505
 506 MODULE_LICENSE("GPL");