]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - net/sched/sch_dsmark.c
Merge tag 'upstream-5.2-rc1' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-jammy-kernel.git] / net / sched / sch_dsmark.c
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
39 struct mask_value {
40 u8 mask;
41 u8 value;
42 };
43
44 struct 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
56 static 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
63 static 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
83 static 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
89 static unsigned long dsmark_find(struct Qdisc *sch, u32 classid)
90 {
91 return TC_H_MIN(classid) + 1;
92 }
93
94 static 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
103 static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
104 {
105 }
106
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 },
113 };
114
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)
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_deprecated(tb, TCA_DSMARK_MAX, opt,
136 dsmark_policy, NULL);
137 if (err < 0)
138 goto errout;
139
140 if (tb[TCA_DSMARK_VALUE])
141 p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
142
143 if (tb[TCA_DSMARK_MASK])
144 p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
145
146 err = 0;
147
148 errout:
149 return err;
150 }
151
152 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
153 {
154 struct dsmark_qdisc_data *p = qdisc_priv(sch);
155
156 if (!dsmark_valid_index(p, arg))
157 return -EINVAL;
158
159 p->mv[arg - 1].mask = 0xff;
160 p->mv[arg - 1].value = 0;
161
162 return 0;
163 }
164
165 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
166 {
167 struct dsmark_qdisc_data *p = qdisc_priv(sch);
168 int i;
169
170 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
171 __func__, sch, p, walker);
172
173 if (walker->stop)
174 return;
175
176 for (i = 0; i < p->indices; i++) {
177 if (p->mv[i].mask == 0xff && !p->mv[i].value)
178 goto ignore;
179 if (walker->count >= walker->skip) {
180 if (walker->fn(sch, i + 1, walker) < 0) {
181 walker->stop = 1;
182 break;
183 }
184 }
185 ignore:
186 walker->count++;
187 }
188 }
189
190 static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl,
191 struct netlink_ext_ack *extack)
192 {
193 struct dsmark_qdisc_data *p = qdisc_priv(sch);
194
195 return p->block;
196 }
197
198 /* --------------------------- Qdisc operations ---------------------------- */
199
200 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
201 struct sk_buff **to_free)
202 {
203 unsigned int len = qdisc_pkt_len(skb);
204 struct dsmark_qdisc_data *p = qdisc_priv(sch);
205 int err;
206
207 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
208
209 if (p->set_tc_index) {
210 int wlen = skb_network_offset(skb);
211
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))
217 goto drop;
218
219 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
220 & ~INET_ECN_MASK;
221 break;
222
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))
227 goto drop;
228
229 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
230 & ~INET_ECN_MASK;
231 break;
232 default:
233 skb->tc_index = 0;
234 break;
235 }
236 }
237
238 if (TC_H_MAJ(skb->priority) == sch->handle)
239 skb->tc_index = TC_H_MIN(skb->priority);
240 else {
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);
244
245 pr_debug("result %d class 0x%04x\n", result, res.classid);
246
247 switch (result) {
248 #ifdef CONFIG_NET_CLS_ACT
249 case TC_ACT_QUEUED:
250 case TC_ACT_STOLEN:
251 case TC_ACT_TRAP:
252 __qdisc_drop(skb, to_free);
253 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
254
255 case TC_ACT_SHOT:
256 goto drop;
257 #endif
258 case TC_ACT_OK:
259 skb->tc_index = TC_H_MIN(res.classid);
260 break;
261
262 default:
263 if (p->default_index != NO_DEFAULT_INDEX)
264 skb->tc_index = p->default_index;
265 break;
266 }
267 }
268
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);
273 return err;
274 }
275
276 sch->qstats.backlog += len;
277 sch->q.qlen++;
278
279 return NET_XMIT_SUCCESS;
280
281 drop:
282 qdisc_drop(skb, sch, to_free);
283 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
284 }
285
286 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
287 {
288 struct dsmark_qdisc_data *p = qdisc_priv(sch);
289 struct sk_buff *skb;
290 u32 index;
291
292 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
293
294 skb = qdisc_dequeue_peeked(p->q);
295 if (skb == NULL)
296 return NULL;
297
298 qdisc_bstats_update(sch, skb);
299 qdisc_qstats_backlog_dec(sch, skb);
300 sch->q.qlen--;
301
302 index = skb->tc_index & (p->indices - 1);
303 pr_debug("index %d->%d\n", skb->tc_index, index);
304
305 switch (tc_skb_protocol(skb)) {
306 case htons(ETH_P_IP):
307 ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
308 p->mv[index].value);
309 break;
310 case htons(ETH_P_IPV6):
311 ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
312 p->mv[index].value);
313 break;
314 default:
315 /*
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.
319 */
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)));
323 break;
324 }
325
326 return skb;
327 }
328
329 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
330 {
331 struct dsmark_qdisc_data *p = qdisc_priv(sch);
332
333 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
334
335 return p->q->ops->peek(p->q);
336 }
337
338 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt,
339 struct netlink_ext_ack *extack)
340 {
341 struct dsmark_qdisc_data *p = qdisc_priv(sch);
342 struct nlattr *tb[TCA_DSMARK_MAX + 1];
343 int err = -EINVAL;
344 u32 default_index = NO_DEFAULT_INDEX;
345 u16 indices;
346 int i;
347
348 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
349
350 if (!opt)
351 goto errout;
352
353 err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
354 if (err)
355 return err;
356
357 err = nla_parse_nested_deprecated(tb, TCA_DSMARK_MAX, opt,
358 dsmark_policy, NULL);
359 if (err < 0)
360 goto errout;
361
362 err = -EINVAL;
363 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
364
365 if (hweight32(indices) != 1)
366 goto errout;
367
368 if (tb[TCA_DSMARK_DEFAULT_INDEX])
369 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
370
371 if (indices <= DSMARK_EMBEDDED_SZ)
372 p->mv = p->embedded;
373 else
374 p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
375 if (!p->mv) {
376 err = -ENOMEM;
377 goto errout;
378 }
379 for (i = 0; i < indices; i++) {
380 p->mv[i].mask = 0xff;
381 p->mv[i].value = 0;
382 }
383 p->indices = indices;
384 p->default_index = default_index;
385 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
386
387 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
388 NULL);
389 if (p->q == NULL)
390 p->q = &noop_qdisc;
391 else
392 qdisc_hash_add(p->q, true);
393
394 pr_debug("%s: qdisc %p\n", __func__, p->q);
395
396 err = 0;
397 errout:
398 return err;
399 }
400
401 static void dsmark_reset(struct Qdisc *sch)
402 {
403 struct dsmark_qdisc_data *p = qdisc_priv(sch);
404
405 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
406 qdisc_reset(p->q);
407 sch->qstats.backlog = 0;
408 sch->q.qlen = 0;
409 }
410
411 static void dsmark_destroy(struct Qdisc *sch)
412 {
413 struct dsmark_qdisc_data *p = qdisc_priv(sch);
414
415 pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
416
417 tcf_block_put(p->block);
418 qdisc_put(p->q);
419 if (p->mv != p->embedded)
420 kfree(p->mv);
421 }
422
423 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
424 struct sk_buff *skb, struct tcmsg *tcm)
425 {
426 struct dsmark_qdisc_data *p = qdisc_priv(sch);
427 struct nlattr *opts = NULL;
428
429 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
430
431 if (!dsmark_valid_index(p, cl))
432 return -EINVAL;
433
434 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
435 tcm->tcm_info = p->q->handle;
436
437 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
438 if (opts == NULL)
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;
443
444 return nla_nest_end(skb, opts);
445
446 nla_put_failure:
447 nla_nest_cancel(skb, opts);
448 return -EMSGSIZE;
449 }
450
451 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
452 {
453 struct dsmark_qdisc_data *p = qdisc_priv(sch);
454 struct nlattr *opts = NULL;
455
456 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
457 if (opts == NULL)
458 goto nla_put_failure;
459 if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
460 goto nla_put_failure;
461
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;
465
466 if (p->set_tc_index &&
467 nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
468 goto nla_put_failure;
469
470 return nla_nest_end(skb, opts);
471
472 nla_put_failure:
473 nla_nest_cancel(skb, opts);
474 return -EMSGSIZE;
475 }
476
477 static const struct Qdisc_class_ops dsmark_class_ops = {
478 .graft = dsmark_graft,
479 .leaf = dsmark_leaf,
480 .find = dsmark_find,
481 .change = dsmark_change,
482 .delete = dsmark_delete,
483 .walk = dsmark_walk,
484 .tcf_block = dsmark_tcf_block,
485 .bind_tcf = dsmark_bind_filter,
486 .unbind_tcf = dsmark_unbind_filter,
487 .dump = dsmark_dump_class,
488 };
489
490 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
491 .next = NULL,
492 .cl_ops = &dsmark_class_ops,
493 .id = "dsmark",
494 .priv_size = sizeof(struct dsmark_qdisc_data),
495 .enqueue = dsmark_enqueue,
496 .dequeue = dsmark_dequeue,
497 .peek = dsmark_peek,
498 .init = dsmark_init,
499 .reset = dsmark_reset,
500 .destroy = dsmark_destroy,
501 .change = NULL,
502 .dump = dsmark_dump,
503 .owner = THIS_MODULE,
504 };
505
506 static int __init dsmark_module_init(void)
507 {
508 return register_qdisc(&dsmark_qdisc_ops);
509 }
510
511 static void __exit dsmark_module_exit(void)
512 {
513 unregister_qdisc(&dsmark_qdisc_ops);
514 }
515
516 module_init(dsmark_module_init)
517 module_exit(dsmark_module_exit)
518
519 MODULE_LICENSE("GPL");