1 /* net/sched/sch_dsmark.c - Differentiated Services field marker */
3 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
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>
22 * classid class marking
23 * ------- ----- -------
27 * x:y y>0 y+1 use entry [y]
29 * x:indices-1 indices use entry [indices-1]
31 * x:y y+1 use entry [y & (indices-1)]
33 * 0xffff 0x10000 use entry [indices-1]
37 #define NO_DEFAULT_INDEX (1 << 16)
44 struct dsmark_qdisc_data {
46 struct tcf_proto __rcu *filter_list;
47 struct tcf_block *block;
48 struct mask_value *mv;
51 u32 default_index; /* index range is 0...0xffff */
52 #define DSMARK_EMBEDDED_SZ 16
53 struct mask_value embedded[DSMARK_EMBEDDED_SZ];
56 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
58 return index <= p->indices && index > 0;
61 /* ------------------------- Class/flow operations ------------------------- */
63 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
64 struct Qdisc *new, struct Qdisc **old)
66 struct dsmark_qdisc_data *p = qdisc_priv(sch);
68 pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
69 __func__, sch, p, new, old);
72 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
78 *old = qdisc_replace(sch, new, &p->q);
82 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
84 struct dsmark_qdisc_data *p = qdisc_priv(sch);
88 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
90 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
91 __func__, sch, qdisc_priv(sch), classid);
93 return TC_H_MIN(classid) + 1;
96 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
97 unsigned long parent, u32 classid)
99 return dsmark_get(sch, classid);
102 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
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 },
114 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
115 struct nlattr **tca, unsigned long *arg)
117 struct dsmark_qdisc_data *p = qdisc_priv(sch);
118 struct nlattr *opt = tca[TCA_OPTIONS];
119 struct nlattr *tb[TCA_DSMARK_MAX + 1];
122 pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
123 __func__, sch, p, classid, parent, *arg);
125 if (!dsmark_valid_index(p, *arg)) {
133 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
137 if (tb[TCA_DSMARK_VALUE])
138 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
140 if (tb[TCA_DSMARK_MASK])
141 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
149 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
151 struct dsmark_qdisc_data *p = qdisc_priv(sch);
153 if (!dsmark_valid_index(p, arg))
156 p->mv[arg - 1].mask = 0xff;
157 p->mv[arg - 1].value = 0;
162 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
164 struct dsmark_qdisc_data *p = qdisc_priv(sch);
167 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
168 __func__, sch, p, walker);
173 for (i = 0; i < p->indices; i++) {
174 if (p->mv[i].mask == 0xff && !p->mv[i].value)
176 if (walker->count >= walker->skip) {
177 if (walker->fn(sch, i + 1, walker) < 0) {
187 static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl)
189 struct dsmark_qdisc_data *p = qdisc_priv(sch);
194 /* --------------------------- Qdisc operations ---------------------------- */
196 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
197 struct sk_buff **to_free)
199 struct dsmark_qdisc_data *p = qdisc_priv(sch);
202 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
204 if (p->set_tc_index) {
205 int wlen = skb_network_offset(skb);
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))
214 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
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))
224 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
233 if (TC_H_MAJ(skb->priority) == sch->handle)
234 skb->tc_index = TC_H_MIN(skb->priority);
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);
240 pr_debug("result %d class 0x%04x\n", result, res.classid);
243 #ifdef CONFIG_NET_CLS_ACT
247 __qdisc_drop(skb, to_free);
248 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
254 skb->tc_index = TC_H_MIN(res.classid);
258 if (p->default_index != NO_DEFAULT_INDEX)
259 skb->tc_index = p->default_index;
264 err = qdisc_enqueue(skb, p->q, to_free);
265 if (err != NET_XMIT_SUCCESS) {
266 if (net_xmit_drop_count(err))
267 qdisc_qstats_drop(sch);
271 qdisc_qstats_backlog_inc(sch, skb);
274 return NET_XMIT_SUCCESS;
277 qdisc_drop(skb, sch, to_free);
278 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
281 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
283 struct dsmark_qdisc_data *p = qdisc_priv(sch);
287 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
289 skb = qdisc_dequeue_peeked(p->q);
293 qdisc_bstats_update(sch, skb);
294 qdisc_qstats_backlog_dec(sch, skb);
297 index = skb->tc_index & (p->indices - 1);
298 pr_debug("index %d->%d\n", skb->tc_index, index);
300 switch (tc_skb_protocol(skb)) {
301 case htons(ETH_P_IP):
302 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
305 case htons(ETH_P_IPV6):
306 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
311 * Only complain if a change was actually attempted.
312 * This way, we can send non-IP traffic through dsmark
313 * and don't need yet another qdisc as a bypass.
315 if (p->mv[index].mask != 0xff || p->mv[index].value)
316 pr_warn("%s: unsupported protocol %d\n",
317 __func__, ntohs(tc_skb_protocol(skb)));
324 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
326 struct dsmark_qdisc_data *p = qdisc_priv(sch);
328 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
330 return p->q->ops->peek(p->q);
333 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
335 struct dsmark_qdisc_data *p = qdisc_priv(sch);
336 struct nlattr *tb[TCA_DSMARK_MAX + 1];
338 u32 default_index = NO_DEFAULT_INDEX;
342 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
347 err = tcf_block_get(&p->block, &p->filter_list);
351 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
356 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
358 if (hweight32(indices) != 1)
361 if (tb[TCA_DSMARK_DEFAULT_INDEX])
362 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
364 if (indices <= DSMARK_EMBEDDED_SZ)
367 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
372 for (i = 0; i < indices; i++) {
373 p->mv[i].mask = 0xff;
376 p->indices = indices;
377 p->default_index = default_index;
378 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
380 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
384 qdisc_hash_add(p->q, true);
386 pr_debug("%s: qdisc %p\n", __func__, p->q);
393 static void dsmark_reset(struct Qdisc *sch)
395 struct dsmark_qdisc_data *p = qdisc_priv(sch);
397 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
399 sch->qstats.backlog = 0;
403 static void dsmark_destroy(struct Qdisc *sch)
405 struct dsmark_qdisc_data *p = qdisc_priv(sch);
407 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
409 tcf_block_put(p->block);
411 if (p->mv != p->embedded)
415 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
416 struct sk_buff *skb, struct tcmsg *tcm)
418 struct dsmark_qdisc_data *p = qdisc_priv(sch);
419 struct nlattr *opts = NULL;
421 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
423 if (!dsmark_valid_index(p, cl))
426 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
427 tcm->tcm_info = p->q->handle;
429 opts = nla_nest_start(skb, TCA_OPTIONS);
431 goto nla_put_failure;
432 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
433 nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
434 goto nla_put_failure;
436 return nla_nest_end(skb, opts);
439 nla_nest_cancel(skb, opts);
443 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
445 struct dsmark_qdisc_data *p = qdisc_priv(sch);
446 struct nlattr *opts = NULL;
448 opts = nla_nest_start(skb, TCA_OPTIONS);
450 goto nla_put_failure;
451 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
452 goto nla_put_failure;
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;
458 if (p->set_tc_index &&
459 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
460 goto nla_put_failure;
462 return nla_nest_end(skb, opts);
465 nla_nest_cancel(skb, opts);
469 static const struct Qdisc_class_ops dsmark_class_ops = {
470 .graft = dsmark_graft,
474 .change = dsmark_change,
475 .delete = dsmark_delete,
477 .tcf_block = dsmark_tcf_block,
478 .bind_tcf = dsmark_bind_filter,
479 .unbind_tcf = dsmark_put,
480 .dump = dsmark_dump_class,
483 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
485 .cl_ops = &dsmark_class_ops,
487 .priv_size = sizeof(struct dsmark_qdisc_data),
488 .enqueue = dsmark_enqueue,
489 .dequeue = dsmark_dequeue,
492 .reset = dsmark_reset,
493 .destroy = dsmark_destroy,
496 .owner = THIS_MODULE,
499 static int __init dsmark_module_init(void)
501 return register_qdisc(&dsmark_qdisc_ops);
504 static void __exit dsmark_module_exit(void)
506 unregister_qdisc(&dsmark_qdisc_ops);
509 module_init(dsmark_module_init)
510 module_exit(dsmark_module_exit)
512 MODULE_LICENSE("GPL");