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_deprecated(tb, TCA_DSMARK_MAX, opt,
136 dsmark_policy, NULL);
140 if (tb[TCA_DSMARK_VALUE])
141 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
143 if (tb[TCA_DSMARK_MASK])
144 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
152 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
154 struct dsmark_qdisc_data *p = qdisc_priv(sch);
156 if (!dsmark_valid_index(p, arg))
159 p->mv[arg - 1].mask = 0xff;
160 p->mv[arg - 1].value = 0;
165 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
167 struct dsmark_qdisc_data *p = qdisc_priv(sch);
170 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
171 __func__, sch, p, walker);
176 for (i = 0; i < p->indices; i++) {
177 if (p->mv[i].mask == 0xff && !p->mv[i].value)
179 if (walker->count >= walker->skip) {
180 if (walker->fn(sch, i + 1, walker) < 0) {
190 static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl,
191 struct netlink_ext_ack *extack)
193 struct dsmark_qdisc_data *p = qdisc_priv(sch);
198 /* --------------------------- Qdisc operations ---------------------------- */
200 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
201 struct sk_buff **to_free)
203 unsigned int len = qdisc_pkt_len(skb);
204 struct dsmark_qdisc_data *p = qdisc_priv(sch);
207 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
209 if (p->set_tc_index) {
210 int wlen = skb_network_offset(skb);
212 switch (tc_skb_protocol(skb)) {
213 case htons(ETH_P_IP):
214 wlen += sizeof(struct iphdr);
215 if (!pskb_may_pull(skb, wlen) ||
216 skb_try_make_writable(skb, wlen))
219 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
223 case htons(ETH_P_IPV6):
224 wlen += sizeof(struct ipv6hdr);
225 if (!pskb_may_pull(skb, wlen) ||
226 skb_try_make_writable(skb, wlen))
229 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
238 if (TC_H_MAJ(skb->priority) == sch->handle)
239 skb->tc_index = TC_H_MIN(skb->priority);
241 struct tcf_result res;
242 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
243 int result = tcf_classify(skb, fl, &res, false);
245 pr_debug("result %d class 0x%04x\n", result, res.classid);
248 #ifdef CONFIG_NET_CLS_ACT
252 __qdisc_drop(skb, to_free);
253 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
259 skb->tc_index = TC_H_MIN(res.classid);
263 if (p->default_index != NO_DEFAULT_INDEX)
264 skb->tc_index = p->default_index;
269 err = qdisc_enqueue(skb, p->q, to_free);
270 if (err != NET_XMIT_SUCCESS) {
271 if (net_xmit_drop_count(err))
272 qdisc_qstats_drop(sch);
276 sch->qstats.backlog += len;
279 return NET_XMIT_SUCCESS;
282 qdisc_drop(skb, sch, to_free);
283 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
286 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
288 struct dsmark_qdisc_data *p = qdisc_priv(sch);
292 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
294 skb = qdisc_dequeue_peeked(p->q);
298 qdisc_bstats_update(sch, skb);
299 qdisc_qstats_backlog_dec(sch, skb);
302 index = skb->tc_index & (p->indices - 1);
303 pr_debug("index %d->%d\n", skb->tc_index, index);
305 switch (tc_skb_protocol(skb)) {
306 case htons(ETH_P_IP):
307 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
310 case htons(ETH_P_IPV6):
311 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
316 * Only complain if a change was actually attempted.
317 * This way, we can send non-IP traffic through dsmark
318 * and don't need yet another qdisc as a bypass.
320 if (p->mv[index].mask != 0xff || p->mv[index].value)
321 pr_warn("%s: unsupported protocol %d\n",
322 __func__, ntohs(tc_skb_protocol(skb)));
329 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
331 struct dsmark_qdisc_data *p = qdisc_priv(sch);
333 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
335 return p->q->ops->peek(p->q);
338 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt,
339 struct netlink_ext_ack *extack)
341 struct dsmark_qdisc_data *p = qdisc_priv(sch);
342 struct nlattr *tb[TCA_DSMARK_MAX + 1];
344 u32 default_index = NO_DEFAULT_INDEX;
348 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
353 err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
357 err = nla_parse_nested_deprecated(tb, TCA_DSMARK_MAX, opt,
358 dsmark_policy, NULL);
363 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
365 if (hweight32(indices) != 1)
368 if (tb[TCA_DSMARK_DEFAULT_INDEX])
369 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
371 if (indices <= DSMARK_EMBEDDED_SZ)
374 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
379 for (i = 0; i < indices; i++) {
380 p->mv[i].mask = 0xff;
383 p->indices = indices;
384 p->default_index = default_index;
385 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
387 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
392 qdisc_hash_add(p->q, true);
394 pr_debug("%s: qdisc %p\n", __func__, p->q);
401 static void dsmark_reset(struct Qdisc *sch)
403 struct dsmark_qdisc_data *p = qdisc_priv(sch);
405 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
407 sch->qstats.backlog = 0;
411 static void dsmark_destroy(struct Qdisc *sch)
413 struct dsmark_qdisc_data *p = qdisc_priv(sch);
415 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
417 tcf_block_put(p->block);
419 if (p->mv != p->embedded)
423 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
424 struct sk_buff *skb, struct tcmsg *tcm)
426 struct dsmark_qdisc_data *p = qdisc_priv(sch);
427 struct nlattr *opts = NULL;
429 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
431 if (!dsmark_valid_index(p, cl))
434 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
435 tcm->tcm_info = p->q->handle;
437 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
439 goto nla_put_failure;
440 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
441 nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
442 goto nla_put_failure;
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_noflag(skb, TCA_OPTIONS);
458 goto nla_put_failure;
459 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
460 goto nla_put_failure;
462 if (p->default_index != NO_DEFAULT_INDEX &&
463 nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
464 goto nla_put_failure;
466 if (p->set_tc_index &&
467 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
468 goto nla_put_failure;
470 return nla_nest_end(skb, opts);
473 nla_nest_cancel(skb, opts);
477 static const struct Qdisc_class_ops dsmark_class_ops = {
478 .graft = dsmark_graft,
481 .change = dsmark_change,
482 .delete = dsmark_delete,
484 .tcf_block = dsmark_tcf_block,
485 .bind_tcf = dsmark_bind_filter,
486 .unbind_tcf = dsmark_unbind_filter,
487 .dump = dsmark_dump_class,
490 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
492 .cl_ops = &dsmark_class_ops,
494 .priv_size = sizeof(struct dsmark_qdisc_data),
495 .enqueue = dsmark_enqueue,
496 .dequeue = dsmark_dequeue,
499 .reset = dsmark_reset,
500 .destroy = dsmark_destroy,
503 .owner = THIS_MODULE,
506 static int __init dsmark_module_init(void)
508 return register_qdisc(&dsmark_qdisc_ops);
511 static void __exit dsmark_module_exit(void)
513 unregister_qdisc(&dsmark_qdisc_ops);
516 module_init(dsmark_module_init)
517 module_exit(dsmark_module_exit)
519 MODULE_LICENSE("GPL");