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 tcf_block        *block;
  48         struct mask_value       *mv;
  49         u16                     indices;
  50         u8                      set_tc_index;
  51         u32                     default_index;  /* index range is 0...0xffff */
  52 #define DSMARK_EMBEDDED_SZ      16
  53         struct mask_value       embedded[DSMARK_EMBEDDED_SZ];
  54 };
  55
  56 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  57 {
  58         return index <= p->indices && index > 0;
  59 }
  60
  61 /* ------------------------- Class/flow operations ------------------------- */
  62
  63 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  64                         struct Qdisc *new, struct Qdisc **old)
  65 {
  66         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  67
  68         pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
  69                  __func__, sch, p, new, old);
  70
  71         if (new == NULL) {
  72                 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
  73                                         sch->handle);
  74                 if (new == NULL)
  75                         new = &noop_qdisc;
  76         }
  77
  78         *old = qdisc_replace(sch, new, &p->q);
  79         return 0;
  80 }
  81
  82 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
  83 {
  84         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  85         return p->q;
  86 }
  87
  88 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
  89 {
  90         pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
  91                  __func__, sch, qdisc_priv(sch), classid);
  92
  93         return TC_H_MIN(classid) + 1;
  94 }
  95
  96 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
  97                                         unsigned long parent, u32 classid)
  98 {
  99         return dsmark_get(sch, classid);
 100 }
 101
 102 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
 103 {
 104 }
 105
 106 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
 107         [TCA_DSMARK_INDICES]            = { .type = NLA_U16 },
 108         [TCA_DSMARK_DEFAULT_INDEX]      = { .type = NLA_U16 },
 109         [TCA_DSMARK_SET_TC_INDEX]       = { .type = NLA_FLAG },
 110         [TCA_DSMARK_MASK]               = { .type = NLA_U8 },
 111         [TCA_DSMARK_VALUE]              = { .type = NLA_U8 },
 112 };
 113
 114 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
 115                          struct nlattr **tca, unsigned long *arg)
 116 {
 117         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 118         struct nlattr *opt = tca[TCA_OPTIONS];
 119         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 120         int err = -EINVAL;
 121
 122         pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
 123                  __func__, sch, p, classid, parent, *arg);
 124
 125         if (!dsmark_valid_index(p, *arg)) {
 126                 err = -ENOENT;
 127                 goto errout;
 128         }
 129
 130         if (!opt)
 131                 goto errout;
 132
 133         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
 134         if (err < 0)
 135                 goto errout;
 136
 137         if (tb[TCA_DSMARK_VALUE])
 138                 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
 139
 140         if (tb[TCA_DSMARK_MASK])
 141                 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
 142
 143         err = 0;
 144
 145 errout:
 146         return err;
 147 }
 148
 149 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 150 {
 151         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 152
 153         if (!dsmark_valid_index(p, arg))
 154                 return -EINVAL;
 155
 156         p->mv[arg - 1].mask = 0xff;
 157         p->mv[arg - 1].value = 0;
 158
 159         return 0;
 160 }
 161
 162 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 163 {
 164         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 165         int i;
 166
 167         pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
 168                  __func__, sch, p, walker);
 169
 170         if (walker->stop)
 171                 return;
 172
 173         for (i = 0; i < p->indices; i++) {
 174                 if (p->mv[i].mask == 0xff && !p->mv[i].value)
 175                         goto ignore;
 176                 if (walker->count >= walker->skip) {
 177                         if (walker->fn(sch, i + 1, walker) < 0) {
 178                                 walker->stop = 1;
 179                                 break;
 180                         }
 181                 }
 182 ignore:
 183                 walker->count++;
 184         }
 185 }
 186
 187 static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl)
 188 {
 189         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 190
 191         return p->block;
 192 }
 193
 194 /* --------------------------- Qdisc operations ---------------------------- */
 195
 196 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 197                           struct sk_buff **to_free)
 198 {
 199         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 200         int err;
 201
 202         pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
 203
 204         if (p->set_tc_index) {
 205                 int wlen = skb_network_offset(skb);
 206
 207                 switch (tc_skb_protocol(skb)) {
 208                 case htons(ETH_P_IP):
 209                         wlen += sizeof(struct iphdr);
 210                         if (!pskb_may_pull(skb, wlen) ||
 211                             skb_try_make_writable(skb, wlen))
 212                                 goto drop;
 213
 214                         skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
 215                                 & ~INET_ECN_MASK;
 216                         break;
 217
 218                 case htons(ETH_P_IPV6):
 219                         wlen += sizeof(struct ipv6hdr);
 220                         if (!pskb_may_pull(skb, wlen) ||
 221                             skb_try_make_writable(skb, wlen))
 222                                 goto drop;
 223
 224                         skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 225                                 & ~INET_ECN_MASK;
 226                         break;
 227                 default:
 228                         skb->tc_index = 0;
 229                         break;
 230                 }
 231         }
 232
 233         if (TC_H_MAJ(skb->priority) == sch->handle)
 234                 skb->tc_index = TC_H_MIN(skb->priority);
 235         else {
 236                 struct tcf_result res;
 237                 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
 238                 int result = tcf_classify(skb, fl, &res, false);
 239
 240                 pr_debug("result %d class 0x%04x\n", result, res.classid);
 241
 242                 switch (result) {
 243 #ifdef CONFIG_NET_CLS_ACT
 244                 case TC_ACT_QUEUED:
 245                 case TC_ACT_STOLEN:
 246                         __qdisc_drop(skb, to_free);
 247                         return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 248
 249                 case TC_ACT_SHOT:
 250                         goto drop;
 251 #endif
 252                 case TC_ACT_OK:
 253                         skb->tc_index = TC_H_MIN(res.classid);
 254                         break;
 255
 256                 default:
 257                         if (p->default_index != NO_DEFAULT_INDEX)
 258                                 skb->tc_index = p->default_index;
 259                         break;
 260                 }
 261         }
 262
 263         err = qdisc_enqueue(skb, p->q, to_free);
 264         if (err != NET_XMIT_SUCCESS) {
 265                 if (net_xmit_drop_count(err))
 266                         qdisc_qstats_drop(sch);
 267                 return err;
 268         }
 269
 270         qdisc_qstats_backlog_inc(sch, skb);
 271         sch->q.qlen++;
 272
 273         return NET_XMIT_SUCCESS;
 274
 275 drop:
 276         qdisc_drop(skb, sch, to_free);
 277         return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 278 }
 279
 280 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 281 {
 282         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 283         struct sk_buff *skb;
 284         u32 index;
 285
 286         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 287
 288         skb = qdisc_dequeue_peeked(p->q);
 289         if (skb == NULL)
 290                 return NULL;
 291
 292         qdisc_bstats_update(sch, skb);
 293         qdisc_qstats_backlog_dec(sch, skb);
 294         sch->q.qlen--;
 295
 296         index = skb->tc_index & (p->indices - 1);
 297         pr_debug("index %d->%d\n", skb->tc_index, index);
 298
 299         switch (tc_skb_protocol(skb)) {
 300         case htons(ETH_P_IP):
 301                 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
 302                                     p->mv[index].value);
 303                         break;
 304         case htons(ETH_P_IPV6):
 305                 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
 306                                     p->mv[index].value);
 307                         break;
 308         default:
 309                 /*
 310                  * Only complain if a change was actually attempted.
 311                  * This way, we can send non-IP traffic through dsmark
 312                  * and don't need yet another qdisc as a bypass.
 313                  */
 314                 if (p->mv[index].mask != 0xff || p->mv[index].value)
 315                         pr_warn("%s: unsupported protocol %d\n",
 316                                 __func__, ntohs(tc_skb_protocol(skb)));
 317                 break;
 318         }
 319
 320         return skb;
 321 }
 322
 323 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
 324 {
 325         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 326
 327         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 328
 329         return p->q->ops->peek(p->q);
 330 }
 331
 332 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
 333 {
 334         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 335         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 336         int err = -EINVAL;
 337         u32 default_index = NO_DEFAULT_INDEX;
 338         u16 indices;
 339         int i;
 340
 341         pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
 342
 343         if (!opt)
 344                 goto errout;
 345
 346         err = tcf_block_get(&p->block, &p->filter_list);
 347         if (err)
 348                 return err;
 349
 350         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
 351         if (err < 0)
 352                 goto errout;
 353
 354         err = -EINVAL;
 355         indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
 356
 357         if (hweight32(indices) != 1)
 358                 goto errout;
 359
 360         if (tb[TCA_DSMARK_DEFAULT_INDEX])
 361                 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
 362
 363         if (indices <= DSMARK_EMBEDDED_SZ)
 364                 p->mv = p->embedded;
 365         else
 366                 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
 367         if (!p->mv) {
 368                 err = -ENOMEM;
 369                 goto errout;
 370         }
 371         for (i = 0; i < indices; i++) {
 372                 p->mv[i].mask = 0xff;
 373                 p->mv[i].value = 0;
 374         }
 375         p->indices = indices;
 376         p->default_index = default_index;
 377         p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
 378
 379         p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
 380         if (p->q == NULL)
 381                 p->q = &noop_qdisc;
 382         else
 383                 qdisc_hash_add(p->q, true);
 384
 385         pr_debug("%s: qdisc %p\n", __func__, p->q);
 386
 387         err = 0;
 388 errout:
 389         return err;
 390 }
 391
 392 static void dsmark_reset(struct Qdisc *sch)
 393 {
 394         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 395
 396         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 397         qdisc_reset(p->q);
 398         sch->qstats.backlog = 0;
 399         sch->q.qlen = 0;
 400 }
 401
 402 static void dsmark_destroy(struct Qdisc *sch)
 403 {
 404         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 405
 406         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 407
 408         tcf_block_put(p->block);
 409         qdisc_destroy(p->q);
 410         if (p->mv != p->embedded)
 411                 kfree(p->mv);
 412 }
 413
 414 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 415                              struct sk_buff *skb, struct tcmsg *tcm)
 416 {
 417         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 418         struct nlattr *opts = NULL;
 419
 420         pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
 421
 422         if (!dsmark_valid_index(p, cl))
 423                 return -EINVAL;
 424
 425         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
 426         tcm->tcm_info = p->q->handle;
 427
 428         opts = nla_nest_start(skb, TCA_OPTIONS);
 429         if (opts == NULL)
 430                 goto nla_put_failure;
 431         if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
 432             nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
 433                 goto nla_put_failure;
 434
 435         return nla_nest_end(skb, opts);
 436
 437 nla_put_failure:
 438         nla_nest_cancel(skb, opts);
 439         return -EMSGSIZE;
 440 }
 441
 442 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 443 {
 444         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 445         struct nlattr *opts = NULL;
 446
 447         opts = nla_nest_start(skb, TCA_OPTIONS);
 448         if (opts == NULL)
 449                 goto nla_put_failure;
 450         if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
 451                 goto nla_put_failure;
 452
 453         if (p->default_index != NO_DEFAULT_INDEX &&
 454             nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
 455                 goto nla_put_failure;
 456
 457         if (p->set_tc_index &&
 458             nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
 459                 goto nla_put_failure;
 460
 461         return nla_nest_end(skb, opts);
 462
 463 nla_put_failure:
 464         nla_nest_cancel(skb, opts);
 465         return -EMSGSIZE;
 466 }
 467
 468 static const struct Qdisc_class_ops dsmark_class_ops = {
 469         .graft          =       dsmark_graft,
 470         .leaf           =       dsmark_leaf,
 471         .get            =       dsmark_get,
 472         .put            =       dsmark_put,
 473         .change         =       dsmark_change,
 474         .delete         =       dsmark_delete,
 475         .walk           =       dsmark_walk,
 476         .tcf_block      =       dsmark_tcf_block,
 477         .bind_tcf       =       dsmark_bind_filter,
 478         .unbind_tcf     =       dsmark_put,
 479         .dump           =       dsmark_dump_class,
 480 };
 481
 482 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 483         .next           =       NULL,
 484         .cl_ops         =       &dsmark_class_ops,
 485         .id             =       "dsmark",
 486         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 487         .enqueue        =       dsmark_enqueue,
 488         .dequeue        =       dsmark_dequeue,
 489         .peek           =       dsmark_peek,
 490         .init           =       dsmark_init,
 491         .reset          =       dsmark_reset,
 492         .destroy        =       dsmark_destroy,
 493         .change         =       NULL,
 494         .dump           =       dsmark_dump,
 495         .owner          =       THIS_MODULE,
 496 };
 497
 498 static int __init dsmark_module_init(void)
 499 {
 500         return register_qdisc(&dsmark_qdisc_ops);
 501 }
 502
 503 static void __exit dsmark_module_exit(void)
 504 {
 505         unregister_qdisc(&dsmark_qdisc_ops);
 506 }
 507
 508 module_init(dsmark_module_init)
 509 module_exit(dsmark_module_exit)
 510
 511 MODULE_LICENSE("GPL");