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)
38 struct dsmark_qdisc_data {
40 struct tcf_proto *filter_list;
41 u8 *mask; /* "owns" the array */
44 u32 default_index; /* index range is 0...0xffff */
48 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
50 return (index <= p->indices && index > 0);
53 /* ------------------------- Class/flow operations ------------------------- */
55 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
56 struct Qdisc *new, struct Qdisc **old)
58 struct dsmark_qdisc_data *p = qdisc_priv(sch);
60 pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
64 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
73 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
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("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
89 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];
121 pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
122 "arg 0x%lx\n", sch, p, classid, parent, *arg);
124 if (!dsmark_valid_index(p, *arg)) {
132 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
136 if (tb[TCA_DSMARK_MASK])
137 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
139 if (tb[TCA_DSMARK_VALUE])
140 p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
142 if (tb[TCA_DSMARK_MASK])
143 p->mask[*arg - 1] = mask;
151 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
153 struct dsmark_qdisc_data *p = qdisc_priv(sch);
155 if (!dsmark_valid_index(p, arg))
158 p->mask[arg - 1] = 0xff;
159 p->value[arg - 1] = 0;
164 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
166 struct dsmark_qdisc_data *p = qdisc_priv(sch);
169 pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
174 for (i = 0; i < p->indices; i++) {
175 if (p->mask[i] == 0xff && !p->value[i])
177 if (walker->count >= walker->skip) {
178 if (walker->fn(sch, i + 1, walker) < 0) {
188 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
191 struct dsmark_qdisc_data *p = qdisc_priv(sch);
192 return &p->filter_list;
195 /* --------------------------- Qdisc operations ---------------------------- */
197 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
199 struct dsmark_qdisc_data *p = qdisc_priv(sch);
202 pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
204 if (p->set_tc_index) {
205 int wlen = skb_network_offset(skb);
207 switch (skb->protocol) {
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 int result = tc_classify(skb, p->filter_list, &res);
239 pr_debug("result %d class 0x%04x\n", result, res.classid);
242 #ifdef CONFIG_NET_CLS_ACT
246 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
252 skb->tc_index = TC_H_MIN(res.classid);
256 if (p->default_index != NO_DEFAULT_INDEX)
257 skb->tc_index = p->default_index;
262 err = qdisc_enqueue(skb, p->q);
263 if (err != NET_XMIT_SUCCESS) {
264 if (net_xmit_drop_count(err))
269 sch->qstats.backlog += qdisc_pkt_len(skb);
272 return NET_XMIT_SUCCESS;
277 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
280 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
282 struct dsmark_qdisc_data *p = qdisc_priv(sch);
286 pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
288 skb = p->q->ops->dequeue(p->q);
292 qdisc_bstats_update(sch, skb);
293 sch->qstats.backlog -= qdisc_pkt_len(skb);
296 index = skb->tc_index & (p->indices - 1);
297 pr_debug("index %d->%d\n", skb->tc_index, index);
299 switch (skb->protocol) {
300 case htons(ETH_P_IP):
301 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
304 case htons(ETH_P_IPV6):
305 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
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.
314 if (p->mask[index] != 0xff || p->value[index])
315 pr_warning("dsmark_dequeue: unsupported protocol %d\n",
316 ntohs(skb->protocol));
323 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
325 struct dsmark_qdisc_data *p = qdisc_priv(sch);
327 pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);
329 return p->q->ops->peek(p->q);
332 static unsigned int dsmark_drop(struct Qdisc *sch)
334 struct dsmark_qdisc_data *p = qdisc_priv(sch);
337 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
339 if (p->q->ops->drop == NULL)
342 len = p->q->ops->drop(p->q);
349 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
351 struct dsmark_qdisc_data *p = qdisc_priv(sch);
352 struct nlattr *tb[TCA_DSMARK_MAX + 1];
354 u32 default_index = NO_DEFAULT_INDEX;
358 pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
363 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
368 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
370 if (hweight32(indices) != 1)
373 if (tb[TCA_DSMARK_DEFAULT_INDEX])
374 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
376 mask = kmalloc(indices * 2, GFP_KERNEL);
383 memset(p->mask, 0xff, indices);
385 p->value = p->mask + indices;
386 memset(p->value, 0, indices);
388 p->indices = indices;
389 p->default_index = default_index;
390 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
392 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
396 pr_debug("dsmark_init: qdisc %p\n", p->q);
403 static void dsmark_reset(struct Qdisc *sch)
405 struct dsmark_qdisc_data *p = qdisc_priv(sch);
407 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
409 sch->qstats.backlog = 0;
413 static void dsmark_destroy(struct Qdisc *sch)
415 struct dsmark_qdisc_data *p = qdisc_priv(sch);
417 pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
419 tcf_destroy_chain(&p->filter_list);
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("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", 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 NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]);
442 NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]);
444 return nla_nest_end(skb, opts);
447 nla_nest_cancel(skb, opts);
451 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
453 struct dsmark_qdisc_data *p = qdisc_priv(sch);
454 struct nlattr *opts = NULL;
456 opts = nla_nest_start(skb, TCA_OPTIONS);
458 goto nla_put_failure;
459 NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
461 if (p->default_index != NO_DEFAULT_INDEX)
462 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
465 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
467 return nla_nest_end(skb, opts);
470 nla_nest_cancel(skb, opts);
474 static const struct Qdisc_class_ops dsmark_class_ops = {
475 .graft = dsmark_graft,
479 .change = dsmark_change,
480 .delete = dsmark_delete,
482 .tcf_chain = dsmark_find_tcf,
483 .bind_tcf = dsmark_bind_filter,
484 .unbind_tcf = dsmark_put,
485 .dump = dsmark_dump_class,
488 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
490 .cl_ops = &dsmark_class_ops,
492 .priv_size = sizeof(struct dsmark_qdisc_data),
493 .enqueue = dsmark_enqueue,
494 .dequeue = dsmark_dequeue,
498 .reset = dsmark_reset,
499 .destroy = dsmark_destroy,
502 .owner = THIS_MODULE,
505 static int __init dsmark_module_init(void)
507 return register_qdisc(&dsmark_qdisc_ops);
510 static void __exit dsmark_module_exit(void)
512 unregister_qdisc(&dsmark_qdisc_ops);
515 module_init(dsmark_module_init)
516 module_exit(dsmark_module_exit)
518 MODULE_LICENSE("GPL");