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/dsfield.h>
17 #include <net/inet_ecn.h>
18 #include <asm/byteorder.h>
21 * classid class marking
22 * ------- ----- -------
26 * x:y y>0 y+1 use entry [y]
28 * x:indices-1 indices use entry [indices-1]
30 * x:y y+1 use entry [y & (indices-1)]
32 * 0xffff 0x10000 use entry [indices-1]
36 #define NO_DEFAULT_INDEX (1 << 16)
43 struct dsmark_qdisc_data {
45 struct tcf_proto __rcu *filter_list;
46 struct mask_value *mv;
49 u32 default_index; /* index range is 0...0xffff */
50 #define DSMARK_EMBEDDED_SZ 16
51 struct mask_value embedded[DSMARK_EMBEDDED_SZ];
54 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
56 return index <= p->indices && index > 0;
59 /* ------------------------- Class/flow operations ------------------------- */
61 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
62 struct Qdisc *new, struct Qdisc **old)
64 struct dsmark_qdisc_data *p = qdisc_priv(sch);
66 pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
67 __func__, sch, p, new, old);
70 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
76 *old = qdisc_replace(sch, new, &p->q);
80 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
82 struct dsmark_qdisc_data *p = qdisc_priv(sch);
86 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
88 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
89 __func__, sch, qdisc_priv(sch), classid);
91 return TC_H_MIN(classid) + 1;
94 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
95 unsigned long parent, u32 classid)
97 return dsmark_get(sch, classid);
100 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
104 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
105 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
106 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
107 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
108 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
109 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
112 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
113 struct nlattr **tca, unsigned long *arg)
115 struct dsmark_qdisc_data *p = qdisc_priv(sch);
116 struct nlattr *opt = tca[TCA_OPTIONS];
117 struct nlattr *tb[TCA_DSMARK_MAX + 1];
120 pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
121 __func__, sch, p, classid, parent, *arg);
123 if (!dsmark_valid_index(p, *arg)) {
131 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
135 if (tb[TCA_DSMARK_VALUE])
136 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
138 if (tb[TCA_DSMARK_MASK])
139 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
147 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
149 struct dsmark_qdisc_data *p = qdisc_priv(sch);
151 if (!dsmark_valid_index(p, arg))
154 p->mv[arg - 1].mask = 0xff;
155 p->mv[arg - 1].value = 0;
160 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
162 struct dsmark_qdisc_data *p = qdisc_priv(sch);
165 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
166 __func__, sch, p, walker);
171 for (i = 0; i < p->indices; i++) {
172 if (p->mv[i].mask == 0xff && !p->mv[i].value)
174 if (walker->count >= walker->skip) {
175 if (walker->fn(sch, i + 1, walker) < 0) {
185 static inline struct tcf_proto __rcu **dsmark_find_tcf(struct Qdisc *sch,
188 struct dsmark_qdisc_data *p = qdisc_priv(sch);
189 return &p->filter_list;
192 /* --------------------------- Qdisc operations ---------------------------- */
194 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
196 struct dsmark_qdisc_data *p = qdisc_priv(sch);
199 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
201 if (p->set_tc_index) {
202 int wlen = skb_network_offset(skb);
204 switch (tc_skb_protocol(skb)) {
205 case htons(ETH_P_IP):
206 wlen += sizeof(struct iphdr);
207 if (!pskb_may_pull(skb, wlen) ||
208 skb_try_make_writable(skb, wlen))
211 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
215 case htons(ETH_P_IPV6):
216 wlen += sizeof(struct ipv6hdr);
217 if (!pskb_may_pull(skb, wlen) ||
218 skb_try_make_writable(skb, wlen))
221 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
230 if (TC_H_MAJ(skb->priority) == sch->handle)
231 skb->tc_index = TC_H_MIN(skb->priority);
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);
237 pr_debug("result %d class 0x%04x\n", result, res.classid);
240 #ifdef CONFIG_NET_CLS_ACT
244 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
250 skb->tc_index = TC_H_MIN(res.classid);
254 if (p->default_index != NO_DEFAULT_INDEX)
255 skb->tc_index = p->default_index;
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);
267 qdisc_qstats_backlog_inc(sch, skb);
270 return NET_XMIT_SUCCESS;
273 qdisc_drop(skb, sch);
274 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
277 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
279 struct dsmark_qdisc_data *p = qdisc_priv(sch);
283 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
285 skb = p->q->ops->dequeue(p->q);
289 qdisc_bstats_update(sch, skb);
290 qdisc_qstats_backlog_dec(sch, skb);
293 index = skb->tc_index & (p->indices - 1);
294 pr_debug("index %d->%d\n", skb->tc_index, index);
296 switch (tc_skb_protocol(skb)) {
297 case htons(ETH_P_IP):
298 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
301 case htons(ETH_P_IPV6):
302 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
307 * Only complain if a change was actually attempted.
308 * This way, we can send non-IP traffic through dsmark
309 * and don't need yet another qdisc as a bypass.
311 if (p->mv[index].mask != 0xff || p->mv[index].value)
312 pr_warn("%s: unsupported protocol %d\n",
313 __func__, ntohs(tc_skb_protocol(skb)));
320 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
322 struct dsmark_qdisc_data *p = qdisc_priv(sch);
324 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
326 return p->q->ops->peek(p->q);
329 static unsigned int dsmark_drop(struct Qdisc *sch)
331 struct dsmark_qdisc_data *p = qdisc_priv(sch);
334 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
336 if (p->q->ops->drop == NULL)
339 len = p->q->ops->drop(p->q);
346 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
348 struct dsmark_qdisc_data *p = qdisc_priv(sch);
349 struct nlattr *tb[TCA_DSMARK_MAX + 1];
351 u32 default_index = NO_DEFAULT_INDEX;
355 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
360 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
365 if (!tb[TCA_DSMARK_INDICES])
367 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
369 if (hweight32(indices) != 1)
372 if (tb[TCA_DSMARK_DEFAULT_INDEX])
373 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
375 if (indices <= DSMARK_EMBEDDED_SZ)
378 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
383 for (i = 0; i < indices; i++) {
384 p->mv[i].mask = 0xff;
387 p->indices = indices;
388 p->default_index = default_index;
389 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
391 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
395 pr_debug("%s: qdisc %p\n", __func__, p->q);
402 static void dsmark_reset(struct Qdisc *sch)
404 struct dsmark_qdisc_data *p = qdisc_priv(sch);
406 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
408 sch->qstats.backlog = 0;
412 static void dsmark_destroy(struct Qdisc *sch)
414 struct dsmark_qdisc_data *p = qdisc_priv(sch);
416 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
418 tcf_destroy_chain(&p->filter_list);
420 if (p->mv != p->embedded)
424 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
425 struct sk_buff *skb, struct tcmsg *tcm)
427 struct dsmark_qdisc_data *p = qdisc_priv(sch);
428 struct nlattr *opts = NULL;
430 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
432 if (!dsmark_valid_index(p, cl))
435 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
436 tcm->tcm_info = p->q->handle;
438 opts = nla_nest_start(skb, TCA_OPTIONS);
440 goto nla_put_failure;
441 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
442 nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
443 goto nla_put_failure;
445 return nla_nest_end(skb, opts);
448 nla_nest_cancel(skb, opts);
452 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
454 struct dsmark_qdisc_data *p = qdisc_priv(sch);
455 struct nlattr *opts = NULL;
457 opts = nla_nest_start(skb, TCA_OPTIONS);
459 goto nla_put_failure;
460 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
461 goto nla_put_failure;
463 if (p->default_index != NO_DEFAULT_INDEX &&
464 nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
465 goto nla_put_failure;
467 if (p->set_tc_index &&
468 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
469 goto nla_put_failure;
471 return nla_nest_end(skb, opts);
474 nla_nest_cancel(skb, opts);
478 static const struct Qdisc_class_ops dsmark_class_ops = {
479 .graft = dsmark_graft,
483 .change = dsmark_change,
484 .delete = dsmark_delete,
486 .tcf_chain = dsmark_find_tcf,
487 .bind_tcf = dsmark_bind_filter,
488 .unbind_tcf = dsmark_put,
489 .dump = dsmark_dump_class,
492 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
494 .cl_ops = &dsmark_class_ops,
496 .priv_size = sizeof(struct dsmark_qdisc_data),
497 .enqueue = dsmark_enqueue,
498 .dequeue = dsmark_dequeue,
502 .reset = dsmark_reset,
503 .destroy = dsmark_destroy,
506 .owner = THIS_MODULE,
509 static int __init dsmark_module_init(void)
511 return register_qdisc(&dsmark_qdisc_ops);
514 static void __exit dsmark_module_exit(void)
516 unregister_qdisc(&dsmark_qdisc_ops);
519 module_init(dsmark_module_init)
520 module_exit(dsmark_module_exit)
522 MODULE_LICENSE("GPL");