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 mask_value {
  39         u8                      mask;
  40         u8                      value;
  41 };
  42
  43 struct dsmark_qdisc_data {
  44         struct Qdisc            *q;
  45         struct tcf_proto __rcu  *filter_list;
  46         struct mask_value       *mv;
  47         u16                     indices;
  48         u8                      set_tc_index;
  49         u32                     default_index;  /* index range is 0...0xffff */
  50 #define DSMARK_EMBEDDED_SZ      16
  51         struct mask_value       embedded[DSMARK_EMBEDDED_SZ];
  52 };
  53
  54 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  55 {
  56         return index <= p->indices && index > 0;
  57 }
  58
  59 /* ------------------------- Class/flow operations ------------------------- */
  60
  61 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  62                         struct Qdisc *new, struct Qdisc **old)
  63 {
  64         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  65
  66         pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
  67                  __func__, sch, p, new, old);
  68
  69         if (new == NULL) {
  70                 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
  71                                         sch->handle);
  72                 if (new == NULL)
  73                         new = &noop_qdisc;
  74         }
  75
  76         sch_tree_lock(sch);
  77         *old = p->q;
  78         p->q = new;
  79         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
  80         qdisc_reset(*old);
  81         sch_tree_unlock(sch);
  82
  83         return 0;
  84 }
  85
  86 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
  87 {
  88         struct dsmark_qdisc_data *p = qdisc_priv(sch);
  89         return p->q;
  90 }
  91
  92 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
  93 {
  94         pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
  95                  __func__, sch, qdisc_priv(sch), classid);
  96
  97         return TC_H_MIN(classid) + 1;
  98 }
  99
 100 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
 101                                         unsigned long parent, u32 classid)
 102 {
 103         return dsmark_get(sch, classid);
 104 }
 105
 106 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
 107 {
 108 }
 109
 110 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
 111         [TCA_DSMARK_INDICES]            = { .type = NLA_U16 },
 112         [TCA_DSMARK_DEFAULT_INDEX]      = { .type = NLA_U16 },
 113         [TCA_DSMARK_SET_TC_INDEX]       = { .type = NLA_FLAG },
 114         [TCA_DSMARK_MASK]               = { .type = NLA_U8 },
 115         [TCA_DSMARK_VALUE]              = { .type = NLA_U8 },
 116 };
 117
 118 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
 119                          struct nlattr **tca, unsigned long *arg)
 120 {
 121         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 122         struct nlattr *opt = tca[TCA_OPTIONS];
 123         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 124         int err = -EINVAL;
 125
 126         pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
 127                  __func__, sch, p, classid, parent, *arg);
 128
 129         if (!dsmark_valid_index(p, *arg)) {
 130                 err = -ENOENT;
 131                 goto errout;
 132         }
 133
 134         if (!opt)
 135                 goto errout;
 136
 137         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
 138         if (err < 0)
 139                 goto errout;
 140
 141         if (tb[TCA_DSMARK_VALUE])
 142                 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
 143
 144         if (tb[TCA_DSMARK_MASK])
 145                 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
 146
 147         err = 0;
 148
 149 errout:
 150         return err;
 151 }
 152
 153 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
 154 {
 155         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 156
 157         if (!dsmark_valid_index(p, arg))
 158                 return -EINVAL;
 159
 160         p->mv[arg - 1].mask = 0xff;
 161         p->mv[arg - 1].value = 0;
 162
 163         return 0;
 164 }
 165
 166 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 167 {
 168         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 169         int i;
 170
 171         pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
 172                  __func__, sch, p, walker);
 173
 174         if (walker->stop)
 175                 return;
 176
 177         for (i = 0; i < p->indices; i++) {
 178                 if (p->mv[i].mask == 0xff && !p->mv[i].value)
 179                         goto ignore;
 180                 if (walker->count >= walker->skip) {
 181                         if (walker->fn(sch, i + 1, walker) < 0) {
 182                                 walker->stop = 1;
 183                                 break;
 184                         }
 185                 }
 186 ignore:
 187                 walker->count++;
 188         }
 189 }
 190
 191 static inline struct tcf_proto __rcu **dsmark_find_tcf(struct Qdisc *sch,
 192                                                        unsigned long cl)
 193 {
 194         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 195         return &p->filter_list;
 196 }
 197
 198 /* --------------------------- Qdisc operations ---------------------------- */
 199
 200 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 201 {
 202         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 203         int err;
 204
 205         pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
 206
 207         if (p->set_tc_index) {
 208                 switch (tc_skb_protocol(skb)) {
 209                 case htons(ETH_P_IP):
 210                         if (skb_cow_head(skb, sizeof(struct iphdr)))
 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                         if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
 219                                 goto drop;
 220
 221                         skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
 222                                 & ~INET_ECN_MASK;
 223                         break;
 224                 default:
 225                         skb->tc_index = 0;
 226                         break;
 227                 }
 228         }
 229
 230         if (TC_H_MAJ(skb->priority) == sch->handle)
 231                 skb->tc_index = TC_H_MIN(skb->priority);
 232         else {
 233                 struct tcf_result res;
 234                 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
 235                 int result = tc_classify(skb, fl, &res, false);
 236
 237                 pr_debug("result %d class 0x%04x\n", result, res.classid);
 238
 239                 switch (result) {
 240 #ifdef CONFIG_NET_CLS_ACT
 241                 case TC_ACT_QUEUED:
 242                 case TC_ACT_STOLEN:
 243                         kfree_skb(skb);
 244                         return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 245
 246                 case TC_ACT_SHOT:
 247                         goto drop;
 248 #endif
 249                 case TC_ACT_OK:
 250                         skb->tc_index = TC_H_MIN(res.classid);
 251                         break;
 252
 253                 default:
 254                         if (p->default_index != NO_DEFAULT_INDEX)
 255                                 skb->tc_index = p->default_index;
 256                         break;
 257                 }
 258         }
 259
 260         err = qdisc_enqueue(skb, p->q);
 261         if (err != NET_XMIT_SUCCESS) {
 262                 if (net_xmit_drop_count(err))
 263                         qdisc_qstats_drop(sch);
 264                 return err;
 265         }
 266
 267         sch->q.qlen++;
 268
 269         return NET_XMIT_SUCCESS;
 270
 271 drop:
 272         qdisc_drop(skb, sch);
 273         return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 274 }
 275
 276 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
 277 {
 278         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 279         struct sk_buff *skb;
 280         u32 index;
 281
 282         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 283
 284         skb = p->q->ops->dequeue(p->q);
 285         if (skb == NULL)
 286                 return NULL;
 287
 288         qdisc_bstats_update(sch, skb);
 289         sch->q.qlen--;
 290
 291         index = skb->tc_index & (p->indices - 1);
 292         pr_debug("index %d->%d\n", skb->tc_index, index);
 293
 294         switch (tc_skb_protocol(skb)) {
 295         case htons(ETH_P_IP):
 296                 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
 297                                     p->mv[index].value);
 298                         break;
 299         case htons(ETH_P_IPV6):
 300                 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
 301                                     p->mv[index].value);
 302                         break;
 303         default:
 304                 /*
 305                  * Only complain if a change was actually attempted.
 306                  * This way, we can send non-IP traffic through dsmark
 307                  * and don't need yet another qdisc as a bypass.
 308                  */
 309                 if (p->mv[index].mask != 0xff || p->mv[index].value)
 310                         pr_warn("%s: unsupported protocol %d\n",
 311                                 __func__, ntohs(tc_skb_protocol(skb)));
 312                 break;
 313         }
 314
 315         return skb;
 316 }
 317
 318 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
 319 {
 320         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 321
 322         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 323
 324         return p->q->ops->peek(p->q);
 325 }
 326
 327 static unsigned int dsmark_drop(struct Qdisc *sch)
 328 {
 329         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 330         unsigned int len;
 331
 332         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 333
 334         if (p->q->ops->drop == NULL)
 335                 return 0;
 336
 337         len = p->q->ops->drop(p->q);
 338         if (len)
 339                 sch->q.qlen--;
 340
 341         return len;
 342 }
 343
 344 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
 345 {
 346         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 347         struct nlattr *tb[TCA_DSMARK_MAX + 1];
 348         int err = -EINVAL;
 349         u32 default_index = NO_DEFAULT_INDEX;
 350         u16 indices;
 351         int i;
 352
 353         pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
 354
 355         if (!opt)
 356                 goto errout;
 357
 358         err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
 359         if (err < 0)
 360                 goto errout;
 361
 362         err = -EINVAL;
 363         indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
 364
 365         if (hweight32(indices) != 1)
 366                 goto errout;
 367
 368         if (tb[TCA_DSMARK_DEFAULT_INDEX])
 369                 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
 370
 371         if (indices <= DSMARK_EMBEDDED_SZ)
 372                 p->mv = p->embedded;
 373         else
 374                 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
 375         if (!p->mv) {
 376                 err = -ENOMEM;
 377                 goto errout;
 378         }
 379         for (i = 0; i < indices; i++) {
 380                 p->mv[i].mask = 0xff;
 381                 p->mv[i].value = 0;
 382         }
 383         p->indices = indices;
 384         p->default_index = default_index;
 385         p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
 386
 387         p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
 388         if (p->q == NULL)
 389                 p->q = &noop_qdisc;
 390
 391         pr_debug("%s: qdisc %p\n", __func__, p->q);
 392
 393         err = 0;
 394 errout:
 395         return err;
 396 }
 397
 398 static void dsmark_reset(struct Qdisc *sch)
 399 {
 400         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 401
 402         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 403         qdisc_reset(p->q);
 404         sch->q.qlen = 0;
 405 }
 406
 407 static void dsmark_destroy(struct Qdisc *sch)
 408 {
 409         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 410
 411         pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
 412
 413         tcf_destroy_chain(&p->filter_list);
 414         qdisc_destroy(p->q);
 415         if (p->mv != p->embedded)
 416                 kfree(p->mv);
 417 }
 418
 419 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
 420                              struct sk_buff *skb, struct tcmsg *tcm)
 421 {
 422         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 423         struct nlattr *opts = NULL;
 424
 425         pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
 426
 427         if (!dsmark_valid_index(p, cl))
 428                 return -EINVAL;
 429
 430         tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
 431         tcm->tcm_info = p->q->handle;
 432
 433         opts = nla_nest_start(skb, TCA_OPTIONS);
 434         if (opts == NULL)
 435                 goto nla_put_failure;
 436         if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
 437             nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
 438                 goto nla_put_failure;
 439
 440         return nla_nest_end(skb, opts);
 441
 442 nla_put_failure:
 443         nla_nest_cancel(skb, opts);
 444         return -EMSGSIZE;
 445 }
 446
 447 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
 448 {
 449         struct dsmark_qdisc_data *p = qdisc_priv(sch);
 450         struct nlattr *opts = NULL;
 451
 452         opts = nla_nest_start(skb, TCA_OPTIONS);
 453         if (opts == NULL)
 454                 goto nla_put_failure;
 455         if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
 456                 goto nla_put_failure;
 457
 458         if (p->default_index != NO_DEFAULT_INDEX &&
 459             nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
 460                 goto nla_put_failure;
 461
 462         if (p->set_tc_index &&
 463             nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
 464                 goto nla_put_failure;
 465
 466         return nla_nest_end(skb, opts);
 467
 468 nla_put_failure:
 469         nla_nest_cancel(skb, opts);
 470         return -EMSGSIZE;
 471 }
 472
 473 static const struct Qdisc_class_ops dsmark_class_ops = {
 474         .graft          =       dsmark_graft,
 475         .leaf           =       dsmark_leaf,
 476         .get            =       dsmark_get,
 477         .put            =       dsmark_put,
 478         .change         =       dsmark_change,
 479         .delete         =       dsmark_delete,
 480         .walk           =       dsmark_walk,
 481         .tcf_chain      =       dsmark_find_tcf,
 482         .bind_tcf       =       dsmark_bind_filter,
 483         .unbind_tcf     =       dsmark_put,
 484         .dump           =       dsmark_dump_class,
 485 };
 486
 487 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
 488         .next           =       NULL,
 489         .cl_ops         =       &dsmark_class_ops,
 490         .id             =       "dsmark",
 491         .priv_size      =       sizeof(struct dsmark_qdisc_data),
 492         .enqueue        =       dsmark_enqueue,
 493         .dequeue        =       dsmark_dequeue,
 494         .peek           =       dsmark_peek,
 495         .drop           =       dsmark_drop,
 496         .init           =       dsmark_init,
 497         .reset          =       dsmark_reset,
 498         .destroy        =       dsmark_destroy,
 499         .change         =       NULL,
 500         .dump           =       dsmark_dump,
 501         .owner          =       THIS_MODULE,
 502 };
 503
 504 static int __init dsmark_module_init(void)
 505 {
 506         return register_qdisc(&dsmark_qdisc_ops);
 507 }
 508
 509 static void __exit dsmark_module_exit(void)
 510 {
 511         unregister_qdisc(&dsmark_qdisc_ops);
 512 }
 513
 514 module_init(dsmark_module_init)
 515 module_exit(dsmark_module_exit)
 516
 517 MODULE_LICENSE("GPL");