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,
65 struct netlink_ext_ack *extack)
67 struct dsmark_qdisc_data *p = qdisc_priv(sch);
69 pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
70 __func__, sch, p, new, old);
73 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
79 *old = qdisc_replace(sch, new, &p->q);
83 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
85 struct dsmark_qdisc_data *p = qdisc_priv(sch);
89 static unsigned long dsmark_find(struct Qdisc *sch, u32 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 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
98 __func__, sch, qdisc_priv(sch), classid);
100 return dsmark_find(sch, classid);
103 static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
107 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
108 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
109 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
110 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
111 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
112 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
115 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
116 struct nlattr **tca, unsigned long *arg,
117 struct netlink_ext_ack *extack)
119 struct dsmark_qdisc_data *p = qdisc_priv(sch);
120 struct nlattr *opt = tca[TCA_OPTIONS];
121 struct nlattr *tb[TCA_DSMARK_MAX + 1];
124 pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
125 __func__, sch, p, classid, parent, *arg);
127 if (!dsmark_valid_index(p, *arg)) {
135 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
139 if (tb[TCA_DSMARK_VALUE])
140 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
142 if (tb[TCA_DSMARK_MASK])
143 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_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->mv[arg - 1].mask = 0xff;
159 p->mv[arg - 1].value = 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("%s(sch %p,[qdisc %p],walker %p)\n",
170 __func__, sch, p, walker);
175 for (i = 0; i < p->indices; i++) {
176 if (p->mv[i].mask == 0xff && !p->mv[i].value)
178 if (walker->count >= walker->skip) {
179 if (walker->fn(sch, i + 1, walker) < 0) {
189 static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl,
190 struct netlink_ext_ack *extack)
192 struct dsmark_qdisc_data *p = qdisc_priv(sch);
197 /* --------------------------- Qdisc operations ---------------------------- */
199 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
200 struct sk_buff **to_free)
202 unsigned int len = qdisc_pkt_len(skb);
203 struct dsmark_qdisc_data *p = qdisc_priv(sch);
206 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
208 if (p->set_tc_index) {
209 int wlen = skb_network_offset(skb);
211 switch (tc_skb_protocol(skb)) {
212 case htons(ETH_P_IP):
213 wlen += sizeof(struct iphdr);
214 if (!pskb_may_pull(skb, wlen) ||
215 skb_try_make_writable(skb, wlen))
218 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
222 case htons(ETH_P_IPV6):
223 wlen += sizeof(struct ipv6hdr);
224 if (!pskb_may_pull(skb, wlen) ||
225 skb_try_make_writable(skb, wlen))
228 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
237 if (TC_H_MAJ(skb->priority) == sch->handle)
238 skb->tc_index = TC_H_MIN(skb->priority);
240 struct tcf_result res;
241 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
242 int result = tcf_classify(skb, fl, &res, false);
244 pr_debug("result %d class 0x%04x\n", result, res.classid);
247 #ifdef CONFIG_NET_CLS_ACT
251 __qdisc_drop(skb, to_free);
252 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
258 skb->tc_index = TC_H_MIN(res.classid);
262 if (p->default_index != NO_DEFAULT_INDEX)
263 skb->tc_index = p->default_index;
268 err = qdisc_enqueue(skb, p->q, to_free);
269 if (err != NET_XMIT_SUCCESS) {
270 if (net_xmit_drop_count(err))
271 qdisc_qstats_drop(sch);
275 sch->qstats.backlog += len;
278 return NET_XMIT_SUCCESS;
281 qdisc_drop(skb, sch, to_free);
282 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
285 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
287 struct dsmark_qdisc_data *p = qdisc_priv(sch);
291 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
293 skb = qdisc_dequeue_peeked(p->q);
297 qdisc_bstats_update(sch, skb);
298 qdisc_qstats_backlog_dec(sch, skb);
301 index = skb->tc_index & (p->indices - 1);
302 pr_debug("index %d->%d\n", skb->tc_index, index);
304 switch (tc_skb_protocol(skb)) {
305 case htons(ETH_P_IP):
306 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
309 case htons(ETH_P_IPV6):
310 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
315 * Only complain if a change was actually attempted.
316 * This way, we can send non-IP traffic through dsmark
317 * and don't need yet another qdisc as a bypass.
319 if (p->mv[index].mask != 0xff || p->mv[index].value)
320 pr_warn("%s: unsupported protocol %d\n",
321 __func__, ntohs(tc_skb_protocol(skb)));
328 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
330 struct dsmark_qdisc_data *p = qdisc_priv(sch);
332 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
334 return p->q->ops->peek(p->q);
337 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt,
338 struct netlink_ext_ack *extack)
340 struct dsmark_qdisc_data *p = qdisc_priv(sch);
341 struct nlattr *tb[TCA_DSMARK_MAX + 1];
343 u32 default_index = NO_DEFAULT_INDEX;
347 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
352 err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
356 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
361 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
363 if (hweight32(indices) != 1)
366 if (tb[TCA_DSMARK_DEFAULT_INDEX])
367 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
369 if (indices <= DSMARK_EMBEDDED_SZ)
372 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
377 for (i = 0; i < indices; i++) {
378 p->mv[i].mask = 0xff;
381 p->indices = indices;
382 p->default_index = default_index;
383 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
385 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
390 qdisc_hash_add(p->q, true);
392 pr_debug("%s: qdisc %p\n", __func__, p->q);
399 static void dsmark_reset(struct Qdisc *sch)
401 struct dsmark_qdisc_data *p = qdisc_priv(sch);
403 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
405 sch->qstats.backlog = 0;
409 static void dsmark_destroy(struct Qdisc *sch)
411 struct dsmark_qdisc_data *p = qdisc_priv(sch);
413 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
415 tcf_block_put(p->block);
417 if (p->mv != p->embedded)
421 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
422 struct sk_buff *skb, struct tcmsg *tcm)
424 struct dsmark_qdisc_data *p = qdisc_priv(sch);
425 struct nlattr *opts = NULL;
427 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
429 if (!dsmark_valid_index(p, cl))
432 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
433 tcm->tcm_info = p->q->handle;
435 opts = nla_nest_start(skb, TCA_OPTIONS);
437 goto nla_put_failure;
438 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
439 nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
440 goto nla_put_failure;
442 return nla_nest_end(skb, opts);
445 nla_nest_cancel(skb, opts);
449 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
451 struct dsmark_qdisc_data *p = qdisc_priv(sch);
452 struct nlattr *opts = NULL;
454 opts = nla_nest_start(skb, TCA_OPTIONS);
456 goto nla_put_failure;
457 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
458 goto nla_put_failure;
460 if (p->default_index != NO_DEFAULT_INDEX &&
461 nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
462 goto nla_put_failure;
464 if (p->set_tc_index &&
465 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
466 goto nla_put_failure;
468 return nla_nest_end(skb, opts);
471 nla_nest_cancel(skb, opts);
475 static const struct Qdisc_class_ops dsmark_class_ops = {
476 .graft = dsmark_graft,
479 .change = dsmark_change,
480 .delete = dsmark_delete,
482 .tcf_block = dsmark_tcf_block,
483 .bind_tcf = dsmark_bind_filter,
484 .unbind_tcf = dsmark_unbind_filter,
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,
497 .reset = dsmark_reset,
498 .destroy = dsmark_destroy,
501 .owner = THIS_MODULE,
504 static int __init dsmark_module_init(void)
506 return register_qdisc(&dsmark_qdisc_ops);
509 static void __exit dsmark_module_exit(void)
511 unregister_qdisc(&dsmark_qdisc_ops);
514 module_init(dsmark_module_init)
515 module_exit(dsmark_module_exit)
517 MODULE_LICENSE("GPL");