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/dsfield.h>
17 #include <net/inet_ecn.h>
18 #include <asm/byteorder.h>
19
20 /*
21 * classid class marking
22 * ------- ----- -------
23 * n/a 0 n/a
24 * x:0 1 use entry [0]
25 * ... ... ...
26 * x:y y>0 y+1 use entry [y]
27 * ... ... ...
28 * x:indices-1 indices use entry [indices-1]
29 * ... ... ...
30 * x:y y+1 use entry [y & (indices-1)]
31 * ... ... ...
32 * 0xffff 0x10000 use entry [indices-1]
33 */
34
35
36 #define NO_DEFAULT_INDEX (1 << 16)
37
38 struct dsmark_qdisc_data {
39 struct Qdisc *q;
40 struct tcf_proto __rcu *filter_list;
41 u8 *mask; /* "owns" the array */
42 u8 *value;
43 u16 indices;
44 u32 default_index; /* index range is 0...0xffff */
45 int set_tc_index;
46 };
47
dsmark_valid_index(struct dsmark_qdisc_data * p,u16 index)48 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
49 {
50 return index <= p->indices && index > 0;
51 }
52
53 /* ------------------------- Class/flow operations ------------------------- */
54
dsmark_graft(struct Qdisc * sch,unsigned long arg,struct Qdisc * new,struct Qdisc ** old)55 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
56 struct Qdisc *new, struct Qdisc **old)
57 {
58 struct dsmark_qdisc_data *p = qdisc_priv(sch);
59
60 pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
61 __func__, sch, p, new, old);
62
63 if (new == NULL) {
64 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
65 sch->handle);
66 if (new == NULL)
67 new = &noop_qdisc;
68 }
69
70 *old = qdisc_replace(sch, new, &p->q);
71 return 0;
72 }
73
dsmark_leaf(struct Qdisc * sch,unsigned long arg)74 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
75 {
76 struct dsmark_qdisc_data *p = qdisc_priv(sch);
77 return p->q;
78 }
79
dsmark_get(struct Qdisc * sch,u32 classid)80 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
81 {
82 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
83 __func__, sch, qdisc_priv(sch), classid);
84
85 return TC_H_MIN(classid) + 1;
86 }
87
dsmark_bind_filter(struct Qdisc * sch,unsigned long parent,u32 classid)88 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
89 unsigned long parent, u32 classid)
90 {
91 return dsmark_get(sch, classid);
92 }
93
dsmark_put(struct Qdisc * sch,unsigned long cl)94 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
95 {
96 }
97
98 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
99 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
100 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
101 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
102 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
103 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
104 };
105
dsmark_change(struct Qdisc * sch,u32 classid,u32 parent,struct nlattr ** tca,unsigned long * arg)106 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
107 struct nlattr **tca, unsigned long *arg)
108 {
109 struct dsmark_qdisc_data *p = qdisc_priv(sch);
110 struct nlattr *opt = tca[TCA_OPTIONS];
111 struct nlattr *tb[TCA_DSMARK_MAX + 1];
112 int err = -EINVAL;
113 u8 mask = 0;
114
115 pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
116 __func__, sch, p, classid, parent, *arg);
117
118 if (!dsmark_valid_index(p, *arg)) {
119 err = -ENOENT;
120 goto errout;
121 }
122
123 if (!opt)
124 goto errout;
125
126 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
127 if (err < 0)
128 goto errout;
129
130 if (tb[TCA_DSMARK_MASK])
131 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
132
133 if (tb[TCA_DSMARK_VALUE])
134 p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
135
136 if (tb[TCA_DSMARK_MASK])
137 p->mask[*arg - 1] = mask;
138
139 err = 0;
140
141 errout:
142 return err;
143 }
144
dsmark_delete(struct Qdisc * sch,unsigned long arg)145 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
146 {
147 struct dsmark_qdisc_data *p = qdisc_priv(sch);
148
149 if (!dsmark_valid_index(p, arg))
150 return -EINVAL;
151
152 p->mask[arg - 1] = 0xff;
153 p->value[arg - 1] = 0;
154
155 return 0;
156 }
157
dsmark_walk(struct Qdisc * sch,struct qdisc_walker * walker)158 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
159 {
160 struct dsmark_qdisc_data *p = qdisc_priv(sch);
161 int i;
162
163 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
164 __func__, sch, p, walker);
165
166 if (walker->stop)
167 return;
168
169 for (i = 0; i < p->indices; i++) {
170 if (p->mask[i] == 0xff && !p->value[i])
171 goto ignore;
172 if (walker->count >= walker->skip) {
173 if (walker->fn(sch, i + 1, walker) < 0) {
174 walker->stop = 1;
175 break;
176 }
177 }
178 ignore:
179 walker->count++;
180 }
181 }
182
dsmark_find_tcf(struct Qdisc * sch,unsigned long cl)183 static inline struct tcf_proto __rcu **dsmark_find_tcf(struct Qdisc *sch,
184 unsigned long cl)
185 {
186 struct dsmark_qdisc_data *p = qdisc_priv(sch);
187 return &p->filter_list;
188 }
189
190 /* --------------------------- Qdisc operations ---------------------------- */
191
dsmark_enqueue(struct sk_buff * skb,struct Qdisc * sch)192 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
193 {
194 struct dsmark_qdisc_data *p = qdisc_priv(sch);
195 int err;
196
197 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
198
199 if (p->set_tc_index) {
200 switch (skb->protocol) {
201 case htons(ETH_P_IP):
202 if (skb_cow_head(skb, sizeof(struct iphdr)))
203 goto drop;
204
205 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
206 & ~INET_ECN_MASK;
207 break;
208
209 case htons(ETH_P_IPV6):
210 if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
211 goto drop;
212
213 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
214 & ~INET_ECN_MASK;
215 break;
216 default:
217 skb->tc_index = 0;
218 break;
219 }
220 }
221
222 if (TC_H_MAJ(skb->priority) == sch->handle)
223 skb->tc_index = TC_H_MIN(skb->priority);
224 else {
225 struct tcf_result res;
226 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
227 int result = tc_classify(skb, fl, &res);
228
229 pr_debug("result %d class 0x%04x\n", result, res.classid);
230
231 switch (result) {
232 #ifdef CONFIG_NET_CLS_ACT
233 case TC_ACT_QUEUED:
234 case TC_ACT_STOLEN:
235 kfree_skb(skb);
236 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
237
238 case TC_ACT_SHOT:
239 goto drop;
240 #endif
241 case TC_ACT_OK:
242 skb->tc_index = TC_H_MIN(res.classid);
243 break;
244
245 default:
246 if (p->default_index != NO_DEFAULT_INDEX)
247 skb->tc_index = p->default_index;
248 break;
249 }
250 }
251
252 err = qdisc_enqueue(skb, p->q);
253 if (err != NET_XMIT_SUCCESS) {
254 if (net_xmit_drop_count(err))
255 qdisc_qstats_drop(sch);
256 return err;
257 }
258
259 qdisc_qstats_backlog_inc(sch, skb);
260 sch->q.qlen++;
261
262 return NET_XMIT_SUCCESS;
263
264 drop:
265 qdisc_drop(skb, sch);
266 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
267 }
268
dsmark_dequeue(struct Qdisc * sch)269 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
270 {
271 struct dsmark_qdisc_data *p = qdisc_priv(sch);
272 struct sk_buff *skb;
273 u32 index;
274
275 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
276
277 skb = p->q->ops->dequeue(p->q);
278 if (skb == NULL)
279 return NULL;
280
281 qdisc_bstats_update(sch, skb);
282 qdisc_qstats_backlog_dec(sch, skb);
283 sch->q.qlen--;
284
285 index = skb->tc_index & (p->indices - 1);
286 pr_debug("index %d->%d\n", skb->tc_index, index);
287
288 switch (skb->protocol) {
289 case htons(ETH_P_IP):
290 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
291 p->value[index]);
292 break;
293 case htons(ETH_P_IPV6):
294 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
295 p->value[index]);
296 break;
297 default:
298 /*
299 * Only complain if a change was actually attempted.
300 * This way, we can send non-IP traffic through dsmark
301 * and don't need yet another qdisc as a bypass.
302 */
303 if (p->mask[index] != 0xff || p->value[index])
304 pr_warn("%s: unsupported protocol %d\n",
305 __func__, ntohs(skb->protocol));
306 break;
307 }
308
309 return skb;
310 }
311
dsmark_peek(struct Qdisc * sch)312 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
313 {
314 struct dsmark_qdisc_data *p = qdisc_priv(sch);
315
316 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
317
318 return p->q->ops->peek(p->q);
319 }
320
dsmark_drop(struct Qdisc * sch)321 static unsigned int dsmark_drop(struct Qdisc *sch)
322 {
323 struct dsmark_qdisc_data *p = qdisc_priv(sch);
324 unsigned int len;
325
326 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
327
328 if (p->q->ops->drop == NULL)
329 return 0;
330
331 len = p->q->ops->drop(p->q);
332 if (len)
333 sch->q.qlen--;
334
335 return len;
336 }
337
dsmark_init(struct Qdisc * sch,struct nlattr * opt)338 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
339 {
340 struct dsmark_qdisc_data *p = qdisc_priv(sch);
341 struct nlattr *tb[TCA_DSMARK_MAX + 1];
342 int err = -EINVAL;
343 u32 default_index = NO_DEFAULT_INDEX;
344 u16 indices;
345 u8 *mask;
346
347 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
348
349 if (!opt)
350 goto errout;
351
352 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
353 if (err < 0)
354 goto errout;
355
356 err = -EINVAL;
357 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
358
359 if (hweight32(indices) != 1)
360 goto errout;
361
362 if (tb[TCA_DSMARK_DEFAULT_INDEX])
363 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
364
365 mask = kmalloc(indices * 2, GFP_KERNEL);
366 if (mask == NULL) {
367 err = -ENOMEM;
368 goto errout;
369 }
370
371 p->mask = mask;
372 memset(p->mask, 0xff, indices);
373
374 p->value = p->mask + indices;
375 memset(p->value, 0, indices);
376
377 p->indices = indices;
378 p->default_index = default_index;
379 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
380
381 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
382 if (p->q == NULL)
383 p->q = &noop_qdisc;
384
385 pr_debug("%s: qdisc %p\n", __func__, p->q);
386
387 err = 0;
388 errout:
389 return err;
390 }
391
dsmark_reset(struct Qdisc * sch)392 static void dsmark_reset(struct Qdisc *sch)
393 {
394 struct dsmark_qdisc_data *p = qdisc_priv(sch);
395
396 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
397 qdisc_reset(p->q);
398 sch->qstats.backlog = 0;
399 sch->q.qlen = 0;
400 }
401
dsmark_destroy(struct Qdisc * sch)402 static void dsmark_destroy(struct Qdisc *sch)
403 {
404 struct dsmark_qdisc_data *p = qdisc_priv(sch);
405
406 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
407
408 tcf_destroy_chain(&p->filter_list);
409 qdisc_destroy(p->q);
410 kfree(p->mask);
411 }
412
dsmark_dump_class(struct Qdisc * sch,unsigned long cl,struct sk_buff * skb,struct tcmsg * tcm)413 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
414 struct sk_buff *skb, struct tcmsg *tcm)
415 {
416 struct dsmark_qdisc_data *p = qdisc_priv(sch);
417 struct nlattr *opts = NULL;
418
419 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
420
421 if (!dsmark_valid_index(p, cl))
422 return -EINVAL;
423
424 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
425 tcm->tcm_info = p->q->handle;
426
427 opts = nla_nest_start(skb, TCA_OPTIONS);
428 if (opts == NULL)
429 goto nla_put_failure;
430 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]) ||
431 nla_put_u8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]))
432 goto nla_put_failure;
433
434 return nla_nest_end(skb, opts);
435
436 nla_put_failure:
437 nla_nest_cancel(skb, opts);
438 return -EMSGSIZE;
439 }
440
dsmark_dump(struct Qdisc * sch,struct sk_buff * skb)441 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
442 {
443 struct dsmark_qdisc_data *p = qdisc_priv(sch);
444 struct nlattr *opts = NULL;
445
446 opts = nla_nest_start(skb, TCA_OPTIONS);
447 if (opts == NULL)
448 goto nla_put_failure;
449 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
450 goto nla_put_failure;
451
452 if (p->default_index != NO_DEFAULT_INDEX &&
453 nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
454 goto nla_put_failure;
455
456 if (p->set_tc_index &&
457 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
458 goto nla_put_failure;
459
460 return nla_nest_end(skb, opts);
461
462 nla_put_failure:
463 nla_nest_cancel(skb, opts);
464 return -EMSGSIZE;
465 }
466
467 static const struct Qdisc_class_ops dsmark_class_ops = {
468 .graft = dsmark_graft,
469 .leaf = dsmark_leaf,
470 .get = dsmark_get,
471 .put = dsmark_put,
472 .change = dsmark_change,
473 .delete = dsmark_delete,
474 .walk = dsmark_walk,
475 .tcf_chain = dsmark_find_tcf,
476 .bind_tcf = dsmark_bind_filter,
477 .unbind_tcf = dsmark_put,
478 .dump = dsmark_dump_class,
479 };
480
481 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
482 .next = NULL,
483 .cl_ops = &dsmark_class_ops,
484 .id = "dsmark",
485 .priv_size = sizeof(struct dsmark_qdisc_data),
486 .enqueue = dsmark_enqueue,
487 .dequeue = dsmark_dequeue,
488 .peek = dsmark_peek,
489 .drop = dsmark_drop,
490 .init = dsmark_init,
491 .reset = dsmark_reset,
492 .destroy = dsmark_destroy,
493 .change = NULL,
494 .dump = dsmark_dump,
495 .owner = THIS_MODULE,
496 };
497
dsmark_module_init(void)498 static int __init dsmark_module_init(void)
499 {
500 return register_qdisc(&dsmark_qdisc_ops);
501 }
502
dsmark_module_exit(void)503 static void __exit dsmark_module_exit(void)
504 {
505 unregister_qdisc(&dsmark_qdisc_ops);
506 }
507
508 module_init(dsmark_module_init)
509 module_exit(dsmark_module_exit)
510
511 MODULE_LICENSE("GPL");
512