Loading...
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
38struct dsmark_qdisc_data {
39 struct Qdisc *q;
40 struct tcf_proto *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
48static 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
55static 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("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
61 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 sch_tree_lock(sch);
71 *old = p->q;
72 p->q = new;
73 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
74 qdisc_reset(*old);
75 sch_tree_unlock(sch);
76
77 return 0;
78}
79
80static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
81{
82 struct dsmark_qdisc_data *p = qdisc_priv(sch);
83 return p->q;
84}
85
86static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
87{
88 pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
89 sch, qdisc_priv(sch), 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 return dsmark_get(sch, classid);
98}
99
100static void dsmark_put(struct Qdisc *sch, unsigned long cl)
101{
102}
103
104static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
105 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
106 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
107 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
108 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
109 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
110};
111
112static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
113 struct nlattr **tca, unsigned long *arg)
114{
115 struct dsmark_qdisc_data *p = qdisc_priv(sch);
116 struct nlattr *opt = tca[TCA_OPTIONS];
117 struct nlattr *tb[TCA_DSMARK_MAX + 1];
118 int err = -EINVAL;
119 u8 mask = 0;
120
121 pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
122 "arg 0x%lx\n", sch, p, classid, parent, *arg);
123
124 if (!dsmark_valid_index(p, *arg)) {
125 err = -ENOENT;
126 goto errout;
127 }
128
129 if (!opt)
130 goto errout;
131
132 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
133 if (err < 0)
134 goto errout;
135
136 if (tb[TCA_DSMARK_MASK])
137 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
138
139 if (tb[TCA_DSMARK_VALUE])
140 p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
141
142 if (tb[TCA_DSMARK_MASK])
143 p->mask[*arg - 1] = 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->mask[arg - 1] = 0xff;
159 p->value[arg - 1] = 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("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
170
171 if (walker->stop)
172 return;
173
174 for (i = 0; i < p->indices; i++) {
175 if (p->mask[i] == 0xff && !p->value[i])
176 goto ignore;
177 if (walker->count >= walker->skip) {
178 if (walker->fn(sch, i + 1, walker) < 0) {
179 walker->stop = 1;
180 break;
181 }
182 }
183ignore:
184 walker->count++;
185 }
186}
187
188static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
189 unsigned long cl)
190{
191 struct dsmark_qdisc_data *p = qdisc_priv(sch);
192 return &p->filter_list;
193}
194
195/* --------------------------- Qdisc operations ---------------------------- */
196
197static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
198{
199 struct dsmark_qdisc_data *p = qdisc_priv(sch);
200 int err;
201
202 pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
203
204 if (p->set_tc_index) {
205 switch (skb->protocol) {
206 case htons(ETH_P_IP):
207 if (skb_cow_head(skb, sizeof(struct iphdr)))
208 goto drop;
209
210 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
211 & ~INET_ECN_MASK;
212 break;
213
214 case htons(ETH_P_IPV6):
215 if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
216 goto drop;
217
218 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
219 & ~INET_ECN_MASK;
220 break;
221 default:
222 skb->tc_index = 0;
223 break;
224 }
225 }
226
227 if (TC_H_MAJ(skb->priority) == sch->handle)
228 skb->tc_index = TC_H_MIN(skb->priority);
229 else {
230 struct tcf_result res;
231 int result = tc_classify(skb, p->filter_list, &res);
232
233 pr_debug("result %d class 0x%04x\n", result, res.classid);
234
235 switch (result) {
236#ifdef CONFIG_NET_CLS_ACT
237 case TC_ACT_QUEUED:
238 case TC_ACT_STOLEN:
239 kfree_skb(skb);
240 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
241
242 case TC_ACT_SHOT:
243 goto drop;
244#endif
245 case TC_ACT_OK:
246 skb->tc_index = TC_H_MIN(res.classid);
247 break;
248
249 default:
250 if (p->default_index != NO_DEFAULT_INDEX)
251 skb->tc_index = p->default_index;
252 break;
253 }
254 }
255
256 err = qdisc_enqueue(skb, p->q);
257 if (err != NET_XMIT_SUCCESS) {
258 if (net_xmit_drop_count(err))
259 sch->qstats.drops++;
260 return err;
261 }
262
263 sch->q.qlen++;
264
265 return NET_XMIT_SUCCESS;
266
267drop:
268 qdisc_drop(skb, sch);
269 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
270}
271
272static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
273{
274 struct dsmark_qdisc_data *p = qdisc_priv(sch);
275 struct sk_buff *skb;
276 u32 index;
277
278 pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
279
280 skb = p->q->ops->dequeue(p->q);
281 if (skb == NULL)
282 return NULL;
283
284 qdisc_bstats_update(sch, skb);
285 sch->q.qlen--;
286
287 index = skb->tc_index & (p->indices - 1);
288 pr_debug("index %d->%d\n", skb->tc_index, index);
289
290 switch (skb->protocol) {
291 case htons(ETH_P_IP):
292 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
293 p->value[index]);
294 break;
295 case htons(ETH_P_IPV6):
296 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
297 p->value[index]);
298 break;
299 default:
300 /*
301 * Only complain if a change was actually attempted.
302 * This way, we can send non-IP traffic through dsmark
303 * and don't need yet another qdisc as a bypass.
304 */
305 if (p->mask[index] != 0xff || p->value[index])
306 pr_warning("dsmark_dequeue: unsupported protocol %d\n",
307 ntohs(skb->protocol));
308 break;
309 }
310
311 return skb;
312}
313
314static struct sk_buff *dsmark_peek(struct Qdisc *sch)
315{
316 struct dsmark_qdisc_data *p = qdisc_priv(sch);
317
318 pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);
319
320 return p->q->ops->peek(p->q);
321}
322
323static unsigned int dsmark_drop(struct Qdisc *sch)
324{
325 struct dsmark_qdisc_data *p = qdisc_priv(sch);
326 unsigned int len;
327
328 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
329
330 if (p->q->ops->drop == NULL)
331 return 0;
332
333 len = p->q->ops->drop(p->q);
334 if (len)
335 sch->q.qlen--;
336
337 return len;
338}
339
340static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
341{
342 struct dsmark_qdisc_data *p = qdisc_priv(sch);
343 struct nlattr *tb[TCA_DSMARK_MAX + 1];
344 int err = -EINVAL;
345 u32 default_index = NO_DEFAULT_INDEX;
346 u16 indices;
347 u8 *mask;
348
349 pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
350
351 if (!opt)
352 goto errout;
353
354 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
355 if (err < 0)
356 goto errout;
357
358 err = -EINVAL;
359 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
360
361 if (hweight32(indices) != 1)
362 goto errout;
363
364 if (tb[TCA_DSMARK_DEFAULT_INDEX])
365 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
366
367 mask = kmalloc(indices * 2, GFP_KERNEL);
368 if (mask == NULL) {
369 err = -ENOMEM;
370 goto errout;
371 }
372
373 p->mask = mask;
374 memset(p->mask, 0xff, indices);
375
376 p->value = p->mask + indices;
377 memset(p->value, 0, indices);
378
379 p->indices = indices;
380 p->default_index = default_index;
381 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
382
383 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
384 if (p->q == NULL)
385 p->q = &noop_qdisc;
386
387 pr_debug("dsmark_init: qdisc %p\n", p->q);
388
389 err = 0;
390errout:
391 return err;
392}
393
394static void dsmark_reset(struct Qdisc *sch)
395{
396 struct dsmark_qdisc_data *p = qdisc_priv(sch);
397
398 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
399 qdisc_reset(p->q);
400 sch->q.qlen = 0;
401}
402
403static void dsmark_destroy(struct Qdisc *sch)
404{
405 struct dsmark_qdisc_data *p = qdisc_priv(sch);
406
407 pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
408
409 tcf_destroy_chain(&p->filter_list);
410 qdisc_destroy(p->q);
411 kfree(p->mask);
412}
413
414static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
415 struct sk_buff *skb, struct tcmsg *tcm)
416{
417 struct dsmark_qdisc_data *p = qdisc_priv(sch);
418 struct nlattr *opts = NULL;
419
420 pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
421
422 if (!dsmark_valid_index(p, cl))
423 return -EINVAL;
424
425 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
426 tcm->tcm_info = p->q->handle;
427
428 opts = nla_nest_start(skb, TCA_OPTIONS);
429 if (opts == NULL)
430 goto nla_put_failure;
431 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]) ||
432 nla_put_u8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]))
433 goto nla_put_failure;
434
435 return nla_nest_end(skb, opts);
436
437nla_put_failure:
438 nla_nest_cancel(skb, opts);
439 return -EMSGSIZE;
440}
441
442static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
443{
444 struct dsmark_qdisc_data *p = qdisc_priv(sch);
445 struct nlattr *opts = NULL;
446
447 opts = nla_nest_start(skb, TCA_OPTIONS);
448 if (opts == NULL)
449 goto nla_put_failure;
450 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
451 goto nla_put_failure;
452
453 if (p->default_index != NO_DEFAULT_INDEX &&
454 nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
455 goto nla_put_failure;
456
457 if (p->set_tc_index &&
458 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
459 goto nla_put_failure;
460
461 return nla_nest_end(skb, opts);
462
463nla_put_failure:
464 nla_nest_cancel(skb, opts);
465 return -EMSGSIZE;
466}
467
468static const struct Qdisc_class_ops dsmark_class_ops = {
469 .graft = dsmark_graft,
470 .leaf = dsmark_leaf,
471 .get = dsmark_get,
472 .put = dsmark_put,
473 .change = dsmark_change,
474 .delete = dsmark_delete,
475 .walk = dsmark_walk,
476 .tcf_chain = dsmark_find_tcf,
477 .bind_tcf = dsmark_bind_filter,
478 .unbind_tcf = dsmark_put,
479 .dump = dsmark_dump_class,
480};
481
482static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
483 .next = NULL,
484 .cl_ops = &dsmark_class_ops,
485 .id = "dsmark",
486 .priv_size = sizeof(struct dsmark_qdisc_data),
487 .enqueue = dsmark_enqueue,
488 .dequeue = dsmark_dequeue,
489 .peek = dsmark_peek,
490 .drop = dsmark_drop,
491 .init = dsmark_init,
492 .reset = dsmark_reset,
493 .destroy = dsmark_destroy,
494 .change = NULL,
495 .dump = dsmark_dump,
496 .owner = THIS_MODULE,
497};
498
499static int __init dsmark_module_init(void)
500{
501 return register_qdisc(&dsmark_qdisc_ops);
502}
503
504static void __exit dsmark_module_exit(void)
505{
506 unregister_qdisc(&dsmark_qdisc_ops);
507}
508
509module_init(dsmark_module_init)
510module_exit(dsmark_module_exit)
511
512MODULE_LICENSE("GPL");
1// SPDX-License-Identifier: GPL-2.0-only
2/* net/sched/sch_dsmark.c - Differentiated Services field marker */
3
4/* Written 1998-2000 by Werner Almesberger, EPFL ICA */
5
6
7#include <linux/module.h>
8#include <linux/init.h>
9#include <linux/slab.h>
10#include <linux/types.h>
11#include <linux/string.h>
12#include <linux/errno.h>
13#include <linux/skbuff.h>
14#include <linux/rtnetlink.h>
15#include <linux/bitops.h>
16#include <net/pkt_sched.h>
17#include <net/pkt_cls.h>
18#include <net/dsfield.h>
19#include <net/inet_ecn.h>
20#include <asm/byteorder.h>
21
22/*
23 * classid class marking
24 * ------- ----- -------
25 * n/a 0 n/a
26 * x:0 1 use entry [0]
27 * ... ... ...
28 * x:y y>0 y+1 use entry [y]
29 * ... ... ...
30 * x:indices-1 indices use entry [indices-1]
31 * ... ... ...
32 * x:y y+1 use entry [y & (indices-1)]
33 * ... ... ...
34 * 0xffff 0x10000 use entry [indices-1]
35 */
36
37
38#define NO_DEFAULT_INDEX (1 << 16)
39
40struct mask_value {
41 u8 mask;
42 u8 value;
43};
44
45struct dsmark_qdisc_data {
46 struct Qdisc *q;
47 struct tcf_proto __rcu *filter_list;
48 struct tcf_block *block;
49 struct mask_value *mv;
50 u16 indices;
51 u8 set_tc_index;
52 u32 default_index; /* index range is 0...0xffff */
53#define DSMARK_EMBEDDED_SZ 16
54 struct mask_value embedded[DSMARK_EMBEDDED_SZ];
55};
56
57static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
58{
59 return index <= p->indices && index > 0;
60}
61
62/* ------------------------- Class/flow operations ------------------------- */
63
64static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
65 struct Qdisc *new, struct Qdisc **old,
66 struct netlink_ext_ack *extack)
67{
68 struct dsmark_qdisc_data *p = qdisc_priv(sch);
69
70 pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
71 __func__, sch, p, new, old);
72
73 if (new == NULL) {
74 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
75 sch->handle, NULL);
76 if (new == NULL)
77 new = &noop_qdisc;
78 }
79
80 *old = qdisc_replace(sch, new, &p->q);
81 return 0;
82}
83
84static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
85{
86 struct dsmark_qdisc_data *p = qdisc_priv(sch);
87 return p->q;
88}
89
90static unsigned long dsmark_find(struct Qdisc *sch, u32 classid)
91{
92 return TC_H_MIN(classid) + 1;
93}
94
95static unsigned long dsmark_bind_filter(struct Qdisc *sch,
96 unsigned long parent, u32 classid)
97{
98 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
99 __func__, sch, qdisc_priv(sch), classid);
100
101 return dsmark_find(sch, classid);
102}
103
104static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
105{
106}
107
108static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
109 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
110 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
111 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
112 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
113 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
114};
115
116static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
117 struct nlattr **tca, unsigned long *arg,
118 struct netlink_ext_ack *extack)
119{
120 struct dsmark_qdisc_data *p = qdisc_priv(sch);
121 struct nlattr *opt = tca[TCA_OPTIONS];
122 struct nlattr *tb[TCA_DSMARK_MAX + 1];
123 int err = -EINVAL;
124
125 pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
126 __func__, sch, p, classid, parent, *arg);
127
128 if (!dsmark_valid_index(p, *arg)) {
129 err = -ENOENT;
130 goto errout;
131 }
132
133 if (!opt)
134 goto errout;
135
136 err = nla_parse_nested_deprecated(tb, TCA_DSMARK_MAX, opt,
137 dsmark_policy, NULL);
138 if (err < 0)
139 goto errout;
140
141 if (tb[TCA_DSMARK_VALUE])
142 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
143
144 if (tb[TCA_DSMARK_MASK])
145 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
146
147 err = 0;
148
149errout:
150 return err;
151}
152
153static int dsmark_delete(struct Qdisc *sch, unsigned long arg,
154 struct netlink_ext_ack *extack)
155{
156 struct dsmark_qdisc_data *p = qdisc_priv(sch);
157
158 if (!dsmark_valid_index(p, arg))
159 return -EINVAL;
160
161 p->mv[arg - 1].mask = 0xff;
162 p->mv[arg - 1].value = 0;
163
164 return 0;
165}
166
167static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
168{
169 struct dsmark_qdisc_data *p = qdisc_priv(sch);
170 int i;
171
172 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
173 __func__, sch, p, walker);
174
175 if (walker->stop)
176 return;
177
178 for (i = 0; i < p->indices; i++) {
179 if (p->mv[i].mask == 0xff && !p->mv[i].value)
180 goto ignore;
181 if (walker->count >= walker->skip) {
182 if (walker->fn(sch, i + 1, walker) < 0) {
183 walker->stop = 1;
184 break;
185 }
186 }
187ignore:
188 walker->count++;
189 }
190}
191
192static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl,
193 struct netlink_ext_ack *extack)
194{
195 struct dsmark_qdisc_data *p = qdisc_priv(sch);
196
197 return p->block;
198}
199
200/* --------------------------- Qdisc operations ---------------------------- */
201
202static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
203 struct sk_buff **to_free)
204{
205 unsigned int len = qdisc_pkt_len(skb);
206 struct dsmark_qdisc_data *p = qdisc_priv(sch);
207 int err;
208
209 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
210
211 if (p->set_tc_index) {
212 int wlen = skb_network_offset(skb);
213
214 switch (skb_protocol(skb, true)) {
215 case htons(ETH_P_IP):
216 wlen += sizeof(struct iphdr);
217 if (!pskb_may_pull(skb, wlen) ||
218 skb_try_make_writable(skb, wlen))
219 goto drop;
220
221 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
222 & ~INET_ECN_MASK;
223 break;
224
225 case htons(ETH_P_IPV6):
226 wlen += sizeof(struct ipv6hdr);
227 if (!pskb_may_pull(skb, wlen) ||
228 skb_try_make_writable(skb, wlen))
229 goto drop;
230
231 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
232 & ~INET_ECN_MASK;
233 break;
234 default:
235 skb->tc_index = 0;
236 break;
237 }
238 }
239
240 if (TC_H_MAJ(skb->priority) == sch->handle)
241 skb->tc_index = TC_H_MIN(skb->priority);
242 else {
243 struct tcf_result res;
244 struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
245 int result = tcf_classify(skb, fl, &res, false);
246
247 pr_debug("result %d class 0x%04x\n", result, res.classid);
248
249 switch (result) {
250#ifdef CONFIG_NET_CLS_ACT
251 case TC_ACT_QUEUED:
252 case TC_ACT_STOLEN:
253 case TC_ACT_TRAP:
254 __qdisc_drop(skb, to_free);
255 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
256
257 case TC_ACT_SHOT:
258 goto drop;
259#endif
260 case TC_ACT_OK:
261 skb->tc_index = TC_H_MIN(res.classid);
262 break;
263
264 default:
265 if (p->default_index != NO_DEFAULT_INDEX)
266 skb->tc_index = p->default_index;
267 break;
268 }
269 }
270
271 err = qdisc_enqueue(skb, p->q, to_free);
272 if (err != NET_XMIT_SUCCESS) {
273 if (net_xmit_drop_count(err))
274 qdisc_qstats_drop(sch);
275 return err;
276 }
277
278 sch->qstats.backlog += len;
279 sch->q.qlen++;
280
281 return NET_XMIT_SUCCESS;
282
283drop:
284 qdisc_drop(skb, sch, to_free);
285 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
286}
287
288static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
289{
290 struct dsmark_qdisc_data *p = qdisc_priv(sch);
291 struct sk_buff *skb;
292 u32 index;
293
294 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
295
296 skb = qdisc_dequeue_peeked(p->q);
297 if (skb == NULL)
298 return NULL;
299
300 qdisc_bstats_update(sch, skb);
301 qdisc_qstats_backlog_dec(sch, skb);
302 sch->q.qlen--;
303
304 index = skb->tc_index & (p->indices - 1);
305 pr_debug("index %d->%d\n", skb->tc_index, index);
306
307 switch (skb_protocol(skb, true)) {
308 case htons(ETH_P_IP):
309 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
310 p->mv[index].value);
311 break;
312 case htons(ETH_P_IPV6):
313 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
314 p->mv[index].value);
315 break;
316 default:
317 /*
318 * Only complain if a change was actually attempted.
319 * This way, we can send non-IP traffic through dsmark
320 * and don't need yet another qdisc as a bypass.
321 */
322 if (p->mv[index].mask != 0xff || p->mv[index].value)
323 pr_warn("%s: unsupported protocol %d\n",
324 __func__, ntohs(skb_protocol(skb, true)));
325 break;
326 }
327
328 return skb;
329}
330
331static struct sk_buff *dsmark_peek(struct Qdisc *sch)
332{
333 struct dsmark_qdisc_data *p = qdisc_priv(sch);
334
335 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
336
337 return p->q->ops->peek(p->q);
338}
339
340static int dsmark_init(struct Qdisc *sch, struct nlattr *opt,
341 struct netlink_ext_ack *extack)
342{
343 struct dsmark_qdisc_data *p = qdisc_priv(sch);
344 struct nlattr *tb[TCA_DSMARK_MAX + 1];
345 int err = -EINVAL;
346 u32 default_index = NO_DEFAULT_INDEX;
347 u16 indices;
348 int i;
349
350 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
351
352 if (!opt)
353 goto errout;
354
355 err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
356 if (err)
357 return err;
358
359 err = nla_parse_nested_deprecated(tb, TCA_DSMARK_MAX, opt,
360 dsmark_policy, NULL);
361 if (err < 0)
362 goto errout;
363
364 err = -EINVAL;
365 if (!tb[TCA_DSMARK_INDICES])
366 goto errout;
367 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
368
369 if (hweight32(indices) != 1)
370 goto errout;
371
372 if (tb[TCA_DSMARK_DEFAULT_INDEX])
373 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
374
375 if (indices <= DSMARK_EMBEDDED_SZ)
376 p->mv = p->embedded;
377 else
378 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
379 if (!p->mv) {
380 err = -ENOMEM;
381 goto errout;
382 }
383 for (i = 0; i < indices; i++) {
384 p->mv[i].mask = 0xff;
385 p->mv[i].value = 0;
386 }
387 p->indices = indices;
388 p->default_index = default_index;
389 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
390
391 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
392 NULL);
393 if (p->q == NULL)
394 p->q = &noop_qdisc;
395 else
396 qdisc_hash_add(p->q, true);
397
398 pr_debug("%s: qdisc %p\n", __func__, p->q);
399
400 err = 0;
401errout:
402 return err;
403}
404
405static void dsmark_reset(struct Qdisc *sch)
406{
407 struct dsmark_qdisc_data *p = qdisc_priv(sch);
408
409 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
410 if (p->q)
411 qdisc_reset(p->q);
412 sch->qstats.backlog = 0;
413 sch->q.qlen = 0;
414}
415
416static void dsmark_destroy(struct Qdisc *sch)
417{
418 struct dsmark_qdisc_data *p = qdisc_priv(sch);
419
420 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
421
422 tcf_block_put(p->block);
423 qdisc_put(p->q);
424 if (p->mv != p->embedded)
425 kfree(p->mv);
426}
427
428static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
429 struct sk_buff *skb, struct tcmsg *tcm)
430{
431 struct dsmark_qdisc_data *p = qdisc_priv(sch);
432 struct nlattr *opts = NULL;
433
434 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
435
436 if (!dsmark_valid_index(p, cl))
437 return -EINVAL;
438
439 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
440 tcm->tcm_info = p->q->handle;
441
442 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
443 if (opts == NULL)
444 goto nla_put_failure;
445 if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
446 nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
447 goto nla_put_failure;
448
449 return nla_nest_end(skb, opts);
450
451nla_put_failure:
452 nla_nest_cancel(skb, opts);
453 return -EMSGSIZE;
454}
455
456static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
457{
458 struct dsmark_qdisc_data *p = qdisc_priv(sch);
459 struct nlattr *opts = NULL;
460
461 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
462 if (opts == NULL)
463 goto nla_put_failure;
464 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
465 goto nla_put_failure;
466
467 if (p->default_index != NO_DEFAULT_INDEX &&
468 nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
469 goto nla_put_failure;
470
471 if (p->set_tc_index &&
472 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
473 goto nla_put_failure;
474
475 return nla_nest_end(skb, opts);
476
477nla_put_failure:
478 nla_nest_cancel(skb, opts);
479 return -EMSGSIZE;
480}
481
482static const struct Qdisc_class_ops dsmark_class_ops = {
483 .graft = dsmark_graft,
484 .leaf = dsmark_leaf,
485 .find = dsmark_find,
486 .change = dsmark_change,
487 .delete = dsmark_delete,
488 .walk = dsmark_walk,
489 .tcf_block = dsmark_tcf_block,
490 .bind_tcf = dsmark_bind_filter,
491 .unbind_tcf = dsmark_unbind_filter,
492 .dump = dsmark_dump_class,
493};
494
495static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
496 .next = NULL,
497 .cl_ops = &dsmark_class_ops,
498 .id = "dsmark",
499 .priv_size = sizeof(struct dsmark_qdisc_data),
500 .enqueue = dsmark_enqueue,
501 .dequeue = dsmark_dequeue,
502 .peek = dsmark_peek,
503 .init = dsmark_init,
504 .reset = dsmark_reset,
505 .destroy = dsmark_destroy,
506 .change = NULL,
507 .dump = dsmark_dump,
508 .owner = THIS_MODULE,
509};
510
511static int __init dsmark_module_init(void)
512{
513 return register_qdisc(&dsmark_qdisc_ops);
514}
515
516static void __exit dsmark_module_exit(void)
517{
518 unregister_qdisc(&dsmark_qdisc_ops);
519}
520
521module_init(dsmark_module_init)
522module_exit(dsmark_module_exit)
523
524MODULE_LICENSE("GPL");