]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - net/sched/sch_dsmark.c
multiq: requeue should rewind the current_band
[mirror_ubuntu-artful-kernel.git] / net / sched / sch_dsmark.c
CommitLineData
1da177e4
LT
1/* net/sched/sch_dsmark.c - Differentiated Services field marker */
2
3/* Written 1998-2000 by Werner Almesberger, EPFL ICA */
4
5
1da177e4
LT
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/types.h>
9#include <linux/string.h>
10#include <linux/errno.h>
11#include <linux/skbuff.h>
1da177e4 12#include <linux/rtnetlink.h>
5b0ac72b 13#include <linux/bitops.h>
1da177e4
LT
14#include <net/pkt_sched.h>
15#include <net/dsfield.h>
16#include <net/inet_ecn.h>
17#include <asm/byteorder.h>
18
1da177e4
LT
19/*
20 * classid class marking
21 * ------- ----- -------
22 * n/a 0 n/a
23 * x:0 1 use entry [0]
24 * ... ... ...
25 * x:y y>0 y+1 use entry [y]
26 * ... ... ...
27 * x:indices-1 indices use entry [indices-1]
28 * ... ... ...
29 * x:y y+1 use entry [y & (indices-1)]
30 * ... ... ...
31 * 0xffff 0x10000 use entry [indices-1]
32 */
33
34
35#define NO_DEFAULT_INDEX (1 << 16)
36
37struct dsmark_qdisc_data {
38 struct Qdisc *q;
39 struct tcf_proto *filter_list;
af0d1141
TG
40 u8 *mask; /* "owns" the array */
41 u8 *value;
42 u16 indices;
43 u32 default_index; /* index range is 0...0xffff */
1da177e4
LT
44 int set_tc_index;
45};
46
758cc43c
TG
47static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
48{
49 return (index <= p->indices && index > 0);
50}
1da177e4
LT
51
52/* ------------------------- Class/flow operations ------------------------- */
53
af0d1141
TG
54static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
55 struct Qdisc *new, struct Qdisc **old)
1da177e4 56{
81da99ed 57 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 58
81da99ed 59 pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
af0d1141 60 sch, p, new, old);
486b53e5
TG
61
62 if (new == NULL) {
5ce2d488 63 new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd 64 &pfifo_qdisc_ops,
9f9afec4 65 sch->handle);
486b53e5
TG
66 if (new == NULL)
67 new = &noop_qdisc;
68 }
69
1da177e4 70 sch_tree_lock(sch);
af0d1141 71 *old = xchg(&p->q, new);
5e50da01 72 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
af0d1141 73 qdisc_reset(*old);
af0d1141
TG
74 sch_tree_unlock(sch);
75
10297b99 76 return 0;
1da177e4
LT
77}
78
1da177e4
LT
79static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
80{
81da99ed
SH
81 struct dsmark_qdisc_data *p = qdisc_priv(sch);
82 return p->q;
1da177e4
LT
83}
84
af0d1141 85static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
1da177e4 86{
81da99ed
SH
87 pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
88 sch, qdisc_priv(sch), classid);
1da177e4 89
af0d1141 90 return TC_H_MIN(classid) + 1;
1da177e4
LT
91}
92
1da177e4 93static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141 94 unsigned long parent, u32 classid)
1da177e4 95{
af0d1141 96 return dsmark_get(sch, classid);
1da177e4
LT
97}
98
1da177e4
LT
99static void dsmark_put(struct Qdisc *sch, unsigned long cl)
100{
101}
102
27a3421e
PM
103static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
104 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
105 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
106 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
107 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
108 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
109};
110
1da177e4 111static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
1e90474c 112 struct nlattr **tca, unsigned long *arg)
1da177e4 113{
81da99ed 114 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c
PM
115 struct nlattr *opt = tca[TCA_OPTIONS];
116 struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c
TG
117 int err = -EINVAL;
118 u8 mask = 0;
1da177e4 119
81da99ed 120 pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
758cc43c
TG
121 "arg 0x%lx\n", sch, p, classid, parent, *arg);
122
123 if (!dsmark_valid_index(p, *arg)) {
124 err = -ENOENT;
1e90474c 125 goto errout;
1da177e4 126 }
1da177e4 127
cee63723 128 if (!opt)
1e90474c 129 goto errout;
758cc43c 130
27a3421e 131 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723 132 if (err < 0)
27a3421e 133 goto errout;
cee63723 134
27a3421e 135 if (tb[TCA_DSMARK_MASK])
1e90474c 136 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
27a3421e
PM
137
138 if (tb[TCA_DSMARK_VALUE])
1e90474c 139 p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99 140
1e90474c 141 if (tb[TCA_DSMARK_MASK])
758cc43c
TG
142 p->mask[*arg-1] = mask;
143
144 err = 0;
145
1e90474c 146errout:
758cc43c
TG
147 return err;
148}
1da177e4 149
af0d1141 150static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4 151{
81da99ed 152 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 153
af0d1141 154 if (!dsmark_valid_index(p, arg))
1da177e4 155 return -EINVAL;
10297b99 156
1da177e4
LT
157 p->mask[arg-1] = 0xff;
158 p->value[arg-1] = 0;
af0d1141 159
1da177e4
LT
160 return 0;
161}
162
9d127fbd 163static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
1da177e4 164{
81da99ed 165 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4
LT
166 int i;
167
81da99ed 168 pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
af0d1141 169
1da177e4
LT
170 if (walker->stop)
171 return;
af0d1141 172
1da177e4
LT
173 for (i = 0; i < p->indices; i++) {
174 if (p->mask[i] == 0xff && !p->value[i])
0451eb07 175 goto ignore;
1da177e4
LT
176 if (walker->count >= walker->skip) {
177 if (walker->fn(sch, i+1, walker) < 0) {
178 walker->stop = 1;
179 break;
180 }
181 }
10297b99 182ignore:
0451eb07 183 walker->count++;
10297b99 184 }
1da177e4
LT
185}
186
9d127fbd
SH
187static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
188 unsigned long cl)
1da177e4 189{
81da99ed
SH
190 struct dsmark_qdisc_data *p = qdisc_priv(sch);
191 return &p->filter_list;
1da177e4
LT
192}
193
1da177e4
LT
194/* --------------------------- Qdisc operations ---------------------------- */
195
9d127fbd 196static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1da177e4 197{
81da99ed 198 struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141
TG
199 int err;
200
81da99ed 201 pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
1da177e4 202
1da177e4 203 if (p->set_tc_index) {
1da177e4 204 switch (skb->protocol) {
9d127fbd
SH
205 case __constant_htons(ETH_P_IP):
206 if (skb_cow_head(skb, sizeof(struct iphdr)))
207 goto drop;
4c30719f 208
9d127fbd
SH
209 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
210 & ~INET_ECN_MASK;
211 break;
4c30719f 212
9d127fbd
SH
213 case __constant_htons(ETH_P_IPV6):
214 if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
215 goto drop;
4c30719f 216
9d127fbd
SH
217 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
218 & ~INET_ECN_MASK;
219 break;
220 default:
221 skb->tc_index = 0;
222 break;
3ff50b79 223 }
1da177e4 224 }
af0d1141
TG
225
226 if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4 227 skb->tc_index = TC_H_MIN(skb->priority);
af0d1141
TG
228 else {
229 struct tcf_result res;
230 int result = tc_classify(skb, p->filter_list, &res);
231
81da99ed 232 pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141 233
1da177e4 234 switch (result) {
f6853e2d
PM
235#ifdef CONFIG_NET_CLS_ACT
236 case TC_ACT_QUEUED:
237 case TC_ACT_STOLEN:
238 kfree_skb(skb);
378a2f09 239 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
4c30719f 240
f6853e2d 241 case TC_ACT_SHOT:
4c30719f 242 goto drop;
1da177e4 243#endif
c3bc7cff 244 case TC_ACT_OK:
f6853e2d
PM
245 skb->tc_index = TC_H_MIN(res.classid);
246 break;
4c30719f 247
f6853e2d
PM
248 default:
249 if (p->default_index != NO_DEFAULT_INDEX)
250 skb->tc_index = p->default_index;
251 break;
3ff50b79 252 }
1da177e4 253 }
1da177e4 254
5f86173b 255 err = qdisc_enqueue(skb, p->q);
af0d1141 256 if (err != NET_XMIT_SUCCESS) {
378a2f09
JP
257 if (net_xmit_drop_count(err))
258 sch->qstats.drops++;
af0d1141 259 return err;
1da177e4 260 }
af0d1141 261
0abf77e5 262 sch->bstats.bytes += qdisc_pkt_len(skb);
1da177e4
LT
263 sch->bstats.packets++;
264 sch->q.qlen++;
1da177e4 265
af0d1141 266 return NET_XMIT_SUCCESS;
4c30719f
SH
267
268drop:
269 kfree_skb(skb);
270 sch->qstats.drops++;
c27f339a 271 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
af0d1141 272}
1da177e4
LT
273
274static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
275{
81da99ed 276 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 277 struct sk_buff *skb;
af0d1141
TG
278 u32 index;
279
81da99ed 280 pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
1da177e4 281
1da177e4 282 skb = p->q->ops->dequeue(p->q);
af0d1141 283 if (skb == NULL)
1da177e4 284 return NULL;
af0d1141 285
1da177e4 286 sch->q.qlen--;
af0d1141
TG
287
288 index = skb->tc_index & (p->indices - 1);
81da99ed 289 pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141 290
1da177e4 291 switch (skb->protocol) {
9d127fbd
SH
292 case __constant_htons(ETH_P_IP):
293 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
294 p->value[index]);
1da177e4 295 break;
9d127fbd
SH
296 case __constant_htons(ETH_P_IPV6):
297 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
298 p->value[index]);
1da177e4 299 break;
9d127fbd
SH
300 default:
301 /*
302 * Only complain if a change was actually attempted.
303 * This way, we can send non-IP traffic through dsmark
304 * and don't need yet another qdisc as a bypass.
305 */
306 if (p->mask[index] != 0xff || p->value[index])
307 printk(KERN_WARNING
308 "dsmark_dequeue: unsupported protocol %d\n",
309 ntohs(skb->protocol));
310 break;
3ff50b79 311 }
af0d1141 312
1da177e4
LT
313 return skb;
314}
315
9d127fbd 316static int dsmark_requeue(struct sk_buff *skb, struct Qdisc *sch)
1da177e4 317{
81da99ed 318 struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 319 int err;
1da177e4 320
81da99ed 321 pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
af0d1141
TG
322
323 err = p->q->ops->requeue(skb, p->q);
324 if (err != NET_XMIT_SUCCESS) {
378a2f09
JP
325 if (net_xmit_drop_count(err))
326 sch->qstats.drops++;
af0d1141 327 return err;
1da177e4 328 }
1da177e4 329
af0d1141
TG
330 sch->q.qlen++;
331 sch->qstats.requeues++;
332
333 return NET_XMIT_SUCCESS;
334}
1da177e4
LT
335
336static unsigned int dsmark_drop(struct Qdisc *sch)
337{
81da99ed 338 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 339 unsigned int len;
10297b99 340
81da99ed 341 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
af0d1141
TG
342
343 if (p->q->ops->drop == NULL)
1da177e4 344 return 0;
af0d1141
TG
345
346 len = p->q->ops->drop(p->q);
347 if (len)
348 sch->q.qlen--;
349
1da177e4
LT
350 return len;
351}
352
1e90474c 353static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
1da177e4 354{
81da99ed 355 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c 356 struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c
TG
357 int err = -EINVAL;
358 u32 default_index = NO_DEFAULT_INDEX;
359 u16 indices;
360 u8 *mask;
361
81da99ed 362 pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
758cc43c 363
cee63723
PM
364 if (!opt)
365 goto errout;
366
27a3421e 367 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723 368 if (err < 0)
758cc43c
TG
369 goto errout;
370
cee63723 371 err = -EINVAL;
1e90474c 372 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b
DM
373
374 if (hweight32(indices) != 1)
758cc43c
TG
375 goto errout;
376
27a3421e 377 if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c 378 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c
TG
379
380 mask = kmalloc(indices * 2, GFP_KERNEL);
381 if (mask == NULL) {
382 err = -ENOMEM;
383 goto errout;
1da177e4 384 }
758cc43c
TG
385
386 p->mask = mask;
387 memset(p->mask, 0xff, indices);
388
389 p->value = p->mask + indices;
390 memset(p->value, 0, indices);
391
392 p->indices = indices;
393 p->default_index = default_index;
1e90474c 394 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c 395
5ce2d488 396 p->q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd 397 &pfifo_qdisc_ops, sch->handle);
758cc43c 398 if (p->q == NULL)
1da177e4 399 p->q = &noop_qdisc;
758cc43c 400
81da99ed 401 pr_debug("dsmark_init: qdisc %p\n", p->q);
758cc43c
TG
402
403 err = 0;
404errout:
758cc43c 405 return err;
1da177e4
LT
406}
407
1da177e4
LT
408static void dsmark_reset(struct Qdisc *sch)
409{
81da99ed 410 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 411
81da99ed 412 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
1da177e4
LT
413 qdisc_reset(p->q);
414 sch->q.qlen = 0;
415}
416
1da177e4
LT
417static void dsmark_destroy(struct Qdisc *sch)
418{
81da99ed 419 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 420
81da99ed 421 pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
af0d1141 422
ff31ab56 423 tcf_destroy_chain(&p->filter_list);
1da177e4
LT
424 qdisc_destroy(p->q);
425 kfree(p->mask);
426}
427
1da177e4 428static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09 429 struct sk_buff *skb, struct tcmsg *tcm)
1da177e4 430{
81da99ed 431 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c 432 struct nlattr *opts = NULL;
1da177e4 433
81da99ed 434 pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
02f23f09
TG
435
436 if (!dsmark_valid_index(p, cl))
1da177e4 437 return -EINVAL;
02f23f09
TG
438
439 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
cdc7f8e3 440 tcm->tcm_info = p->q->handle;
02f23f09 441
1e90474c
PM
442 opts = nla_nest_start(skb, TCA_OPTIONS);
443 if (opts == NULL)
444 goto nla_put_failure;
445 NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
446 NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
02f23f09 447
1e90474c 448 return nla_nest_end(skb, opts);
1da177e4 449
1e90474c 450nla_put_failure:
bc3ed28c
TG
451 nla_nest_cancel(skb, opts);
452 return -EMSGSIZE;
1da177e4
LT
453}
454
455static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
456{
81da99ed 457 struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c 458 struct nlattr *opts = NULL;
1da177e4 459
1e90474c
PM
460 opts = nla_nest_start(skb, TCA_OPTIONS);
461 if (opts == NULL)
462 goto nla_put_failure;
463 NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
02f23f09
TG
464
465 if (p->default_index != NO_DEFAULT_INDEX)
1e90474c 466 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
1da177e4 467
1da177e4 468 if (p->set_tc_index)
1e90474c 469 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
02f23f09 470
1e90474c 471 return nla_nest_end(skb, opts);
1da177e4 472
1e90474c 473nla_put_failure:
bc3ed28c
TG
474 nla_nest_cancel(skb, opts);
475 return -EMSGSIZE;
1da177e4
LT
476}
477
20fea08b 478static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4
LT
479 .graft = dsmark_graft,
480 .leaf = dsmark_leaf,
481 .get = dsmark_get,
482 .put = dsmark_put,
483 .change = dsmark_change,
484 .delete = dsmark_delete,
485 .walk = dsmark_walk,
486 .tcf_chain = dsmark_find_tcf,
487 .bind_tcf = dsmark_bind_filter,
488 .unbind_tcf = dsmark_put,
489 .dump = dsmark_dump_class,
490};
491
20fea08b 492static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4
LT
493 .next = NULL,
494 .cl_ops = &dsmark_class_ops,
495 .id = "dsmark",
496 .priv_size = sizeof(struct dsmark_qdisc_data),
497 .enqueue = dsmark_enqueue,
498 .dequeue = dsmark_dequeue,
499 .requeue = dsmark_requeue,
500 .drop = dsmark_drop,
501 .init = dsmark_init,
502 .reset = dsmark_reset,
503 .destroy = dsmark_destroy,
504 .change = NULL,
505 .dump = dsmark_dump,
506 .owner = THIS_MODULE,
507};
508
509static int __init dsmark_module_init(void)
510{
511 return register_qdisc(&dsmark_qdisc_ops);
512}
af0d1141 513
10297b99 514static void __exit dsmark_module_exit(void)
1da177e4
LT
515{
516 unregister_qdisc(&dsmark_qdisc_ops);
517}
af0d1141 518
1da177e4
LT
519module_init(dsmark_module_init)
520module_exit(dsmark_module_exit)
af0d1141 521
1da177e4 522MODULE_LICENSE("GPL");