Loading...
Note: File does not exist in v6.13.7.
1/* net/sched/sch_dsmark.c - Differentiated Services field marker */
2
3/* Written 1998-2000 by Werner Almesberger, EPFL ICA */
4
5
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>
20
21/*
22 * classid class marking
23 * ------- ----- -------
24 * n/a 0 n/a
25 * x:0 1 use entry [0]
26 * ... ... ...
27 * x:y y>0 y+1 use entry [y]
28 * ... ... ...
29 * x:indices-1 indices use entry [indices-1]
30 * ... ... ...
31 * x:y y+1 use entry [y & (indices-1)]
32 * ... ... ...
33 * 0xffff 0x10000 use entry [indices-1]
34 */
35
36
37#define NO_DEFAULT_INDEX (1 << 16)
38
39struct mask_value {
40 u8 mask;
41 u8 value;
42};
43
44struct dsmark_qdisc_data {
45 struct Qdisc *q;
46 struct tcf_proto __rcu *filter_list;
47 struct tcf_block *block;
48 struct mask_value *mv;
49 u16 indices;
50 u8 set_tc_index;
51 u32 default_index; /* index range is 0...0xffff */
52#define DSMARK_EMBEDDED_SZ 16
53 struct mask_value embedded[DSMARK_EMBEDDED_SZ];
54};
55
56static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
57{
58 return index <= p->indices && index > 0;
59}
60
61/* ------------------------- Class/flow operations ------------------------- */
62
63static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
64 struct Qdisc *new, struct Qdisc **old,
65 struct netlink_ext_ack *extack)
66{
67 struct dsmark_qdisc_data *p = qdisc_priv(sch);
68
69 pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
70 __func__, sch, p, new, old);
71
72 if (new == NULL) {
73 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
74 sch->handle, NULL);
75 if (new == NULL)
76 new = &noop_qdisc;
77 }
78
79 *old = qdisc_replace(sch, new, &p->q);
80 return 0;
81}
82
83static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
84{
85 struct dsmark_qdisc_data *p = qdisc_priv(sch);
86 return p->q;
87}
88
89static unsigned long dsmark_find(struct Qdisc *sch, u32 classid)
90{
91 return TC_H_MIN(classid) + 1;
92}
93
94static unsigned long dsmark_bind_filter(struct Qdisc *sch,
95 unsigned long parent, u32 classid)
96{
97 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
98 __func__, sch, qdisc_priv(sch), classid);
99
100 return dsmark_find(sch, classid);
101}
102
103static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
104{
105}
106
107static 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 },
113};
114
115static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
116 struct nlattr **tca, unsigned long *arg,
117 struct netlink_ext_ack *extack)
118{
119 struct dsmark_qdisc_data *p = qdisc_priv(sch);
120 struct nlattr *opt = tca[TCA_OPTIONS];
121 struct nlattr *tb[TCA_DSMARK_MAX + 1];
122 int err = -EINVAL;
123
124 pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
125 __func__, sch, p, classid, parent, *arg);
126
127 if (!dsmark_valid_index(p, *arg)) {
128 err = -ENOENT;
129 goto errout;
130 }
131
132 if (!opt)
133 goto errout;
134
135 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
136 if (err < 0)
137 goto errout;
138
139 if (tb[TCA_DSMARK_VALUE])
140 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
141
142 if (tb[TCA_DSMARK_MASK])
143 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
144
145 err = 0;
146
147errout:
148 return err;
149}
150
151static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
152{
153 struct dsmark_qdisc_data *p = qdisc_priv(sch);
154
155 if (!dsmark_valid_index(p, arg))
156 return -EINVAL;
157
158 p->mv[arg - 1].mask = 0xff;
159 p->mv[arg - 1].value = 0;
160
161 return 0;
162}
163
164static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
165{
166 struct dsmark_qdisc_data *p = qdisc_priv(sch);
167 int i;
168
169 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
170 __func__, sch, p, walker);
171
172 if (walker->stop)
173 return;
174
175 for (i = 0; i < p->indices; i++) {
176 if (p->mv[i].mask == 0xff && !p->mv[i].value)
177 goto ignore;
178 if (walker->count >= walker->skip) {
179 if (walker->fn(sch, i + 1, walker) < 0) {
180 walker->stop = 1;
181 break;
182 }
183 }
184ignore:
185 walker->count++;
186 }
187}
188
189static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl,
190 struct netlink_ext_ack *extack)
191{
192 struct dsmark_qdisc_data *p = qdisc_priv(sch);
193
194 return p->block;
195}
196
197/* --------------------------- Qdisc operations ---------------------------- */
198
199static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
200 struct sk_buff **to_free)
201{
202 struct dsmark_qdisc_data *p = qdisc_priv(sch);
203 int err;
204
205 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
206
207 if (p->set_tc_index) {
208 int wlen = skb_network_offset(skb);
209
210 switch (tc_skb_protocol(skb)) {
211 case htons(ETH_P_IP):
212 wlen += sizeof(struct iphdr);
213 if (!pskb_may_pull(skb, wlen) ||
214 skb_try_make_writable(skb, wlen))
215 goto drop;
216
217 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
218 & ~INET_ECN_MASK;
219 break;
220
221 case htons(ETH_P_IPV6):
222 wlen += sizeof(struct ipv6hdr);
223 if (!pskb_may_pull(skb, wlen) ||
224 skb_try_make_writable(skb, wlen))
225 goto drop;
226
227 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
228 & ~INET_ECN_MASK;
229 break;
230 default:
231 skb->tc_index = 0;
232 break;
233 }
234 }
235
236 if (TC_H_MAJ(skb->priority) == sch->handle)
237 skb->tc_index = TC_H_MIN(skb->priority);
238 else {
239 struct tcf_result res;
240 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
241 int result = tcf_classify(skb, fl, &res, false);
242
243 pr_debug("result %d class 0x%04x\n", result, res.classid);
244
245 switch (result) {
246#ifdef CONFIG_NET_CLS_ACT
247 case TC_ACT_QUEUED:
248 case TC_ACT_STOLEN:
249 case TC_ACT_TRAP:
250 __qdisc_drop(skb, to_free);
251 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
252
253 case TC_ACT_SHOT:
254 goto drop;
255#endif
256 case TC_ACT_OK:
257 skb->tc_index = TC_H_MIN(res.classid);
258 break;
259
260 default:
261 if (p->default_index != NO_DEFAULT_INDEX)
262 skb->tc_index = p->default_index;
263 break;
264 }
265 }
266
267 err = qdisc_enqueue(skb, p->q, to_free);
268 if (err != NET_XMIT_SUCCESS) {
269 if (net_xmit_drop_count(err))
270 qdisc_qstats_drop(sch);
271 return err;
272 }
273
274 qdisc_qstats_backlog_inc(sch, skb);
275 sch->q.qlen++;
276
277 return NET_XMIT_SUCCESS;
278
279drop:
280 qdisc_drop(skb, sch, to_free);
281 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
282}
283
284static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
285{
286 struct dsmark_qdisc_data *p = qdisc_priv(sch);
287 struct sk_buff *skb;
288 u32 index;
289
290 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
291
292 skb = qdisc_dequeue_peeked(p->q);
293 if (skb == NULL)
294 return NULL;
295
296 qdisc_bstats_update(sch, skb);
297 qdisc_qstats_backlog_dec(sch, skb);
298 sch->q.qlen--;
299
300 index = skb->tc_index & (p->indices - 1);
301 pr_debug("index %d->%d\n", skb->tc_index, index);
302
303 switch (tc_skb_protocol(skb)) {
304 case htons(ETH_P_IP):
305 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
306 p->mv[index].value);
307 break;
308 case htons(ETH_P_IPV6):
309 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
310 p->mv[index].value);
311 break;
312 default:
313 /*
314 * Only complain if a change was actually attempted.
315 * This way, we can send non-IP traffic through dsmark
316 * and don't need yet another qdisc as a bypass.
317 */
318 if (p->mv[index].mask != 0xff || p->mv[index].value)
319 pr_warn("%s: unsupported protocol %d\n",
320 __func__, ntohs(tc_skb_protocol(skb)));
321 break;
322 }
323
324 return skb;
325}
326
327static struct sk_buff *dsmark_peek(struct Qdisc *sch)
328{
329 struct dsmark_qdisc_data *p = qdisc_priv(sch);
330
331 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
332
333 return p->q->ops->peek(p->q);
334}
335
336static int dsmark_init(struct Qdisc *sch, struct nlattr *opt,
337 struct netlink_ext_ack *extack)
338{
339 struct dsmark_qdisc_data *p = qdisc_priv(sch);
340 struct nlattr *tb[TCA_DSMARK_MAX + 1];
341 int err = -EINVAL;
342 u32 default_index = NO_DEFAULT_INDEX;
343 u16 indices;
344 int i;
345
346 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
347
348 if (!opt)
349 goto errout;
350
351 err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
352 if (err)
353 return err;
354
355 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
356 if (err < 0)
357 goto errout;
358
359 err = -EINVAL;
360 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
361
362 if (hweight32(indices) != 1)
363 goto errout;
364
365 if (tb[TCA_DSMARK_DEFAULT_INDEX])
366 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
367
368 if (indices <= DSMARK_EMBEDDED_SZ)
369 p->mv = p->embedded;
370 else
371 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
372 if (!p->mv) {
373 err = -ENOMEM;
374 goto errout;
375 }
376 for (i = 0; i < indices; i++) {
377 p->mv[i].mask = 0xff;
378 p->mv[i].value = 0;
379 }
380 p->indices = indices;
381 p->default_index = default_index;
382 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
383
384 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
385 NULL);
386 if (p->q == NULL)
387 p->q = &noop_qdisc;
388 else
389 qdisc_hash_add(p->q, true);
390
391 pr_debug("%s: qdisc %p\n", __func__, p->q);
392
393 err = 0;
394errout:
395 return err;
396}
397
398static void dsmark_reset(struct Qdisc *sch)
399{
400 struct dsmark_qdisc_data *p = qdisc_priv(sch);
401
402 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
403 qdisc_reset(p->q);
404 sch->qstats.backlog = 0;
405 sch->q.qlen = 0;
406}
407
408static void dsmark_destroy(struct Qdisc *sch)
409{
410 struct dsmark_qdisc_data *p = qdisc_priv(sch);
411
412 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
413
414 tcf_block_put(p->block);
415 qdisc_destroy(p->q);
416 if (p->mv != p->embedded)
417 kfree(p->mv);
418}
419
420static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
421 struct sk_buff *skb, struct tcmsg *tcm)
422{
423 struct dsmark_qdisc_data *p = qdisc_priv(sch);
424 struct nlattr *opts = NULL;
425
426 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
427
428 if (!dsmark_valid_index(p, cl))
429 return -EINVAL;
430
431 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
432 tcm->tcm_info = p->q->handle;
433
434 opts = nla_nest_start(skb, TCA_OPTIONS);
435 if (opts == NULL)
436 goto nla_put_failure;
437 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
438 nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
439 goto nla_put_failure;
440
441 return nla_nest_end(skb, opts);
442
443nla_put_failure:
444 nla_nest_cancel(skb, opts);
445 return -EMSGSIZE;
446}
447
448static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
449{
450 struct dsmark_qdisc_data *p = qdisc_priv(sch);
451 struct nlattr *opts = NULL;
452
453 opts = nla_nest_start(skb, TCA_OPTIONS);
454 if (opts == NULL)
455 goto nla_put_failure;
456 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
457 goto nla_put_failure;
458
459 if (p->default_index != NO_DEFAULT_INDEX &&
460 nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
461 goto nla_put_failure;
462
463 if (p->set_tc_index &&
464 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
465 goto nla_put_failure;
466
467 return nla_nest_end(skb, opts);
468
469nla_put_failure:
470 nla_nest_cancel(skb, opts);
471 return -EMSGSIZE;
472}
473
474static const struct Qdisc_class_ops dsmark_class_ops = {
475 .graft = dsmark_graft,
476 .leaf = dsmark_leaf,
477 .find = dsmark_find,
478 .change = dsmark_change,
479 .delete = dsmark_delete,
480 .walk = dsmark_walk,
481 .tcf_block = dsmark_tcf_block,
482 .bind_tcf = dsmark_bind_filter,
483 .unbind_tcf = dsmark_unbind_filter,
484 .dump = dsmark_dump_class,
485};
486
487static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
488 .next = NULL,
489 .cl_ops = &dsmark_class_ops,
490 .id = "dsmark",
491 .priv_size = sizeof(struct dsmark_qdisc_data),
492 .enqueue = dsmark_enqueue,
493 .dequeue = dsmark_dequeue,
494 .peek = dsmark_peek,
495 .init = dsmark_init,
496 .reset = dsmark_reset,
497 .destroy = dsmark_destroy,
498 .change = NULL,
499 .dump = dsmark_dump,
500 .owner = THIS_MODULE,
501};
502
503static int __init dsmark_module_init(void)
504{
505 return register_qdisc(&dsmark_qdisc_ops);
506}
507
508static void __exit dsmark_module_exit(void)
509{
510 unregister_qdisc(&dsmark_qdisc_ops);
511}
512
513module_init(dsmark_module_init)
514module_exit(dsmark_module_exit)
515
516MODULE_LICENSE("GPL");