xref: /openbmc/linux/net/sched/em_meta.c (revision 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2) 
1 /*
2  * net/sched/em_meta.c	Metadata ematch
3  *
4  *		This program is free software; you can redistribute it and/or
5  *		modify it under the terms of the GNU General Public License
6  *		as published by the Free Software Foundation; either version
7  *		2 of the License, or (at your option) any later version.
8  *
9  * Authors:	Thomas Graf <tgraf@suug.ch>
10  *
11  * ==========================================================================
12  *
13  * 	The metadata ematch compares two meta objects where each object
14  * 	represents either a meta value stored in the kernel or a static
15  * 	value provided by userspace. The objects are not provided by
16  * 	userspace itself but rather a definition providing the information
17  * 	to build them. Every object is of a certain type which must be
18  * 	equal to the object it is being compared to.
19  *
20  * 	The definition of a objects conists of the type (meta type), a
21  * 	identifier (meta id) and additional type specific information.
22  * 	The meta id is either TCF_META_TYPE_VALUE for values provided by
23  * 	userspace or a index to the meta operations table consisting of
24  * 	function pointers to type specific meta data collectors returning
25  * 	the value of the requested meta value.
26  *
27  * 	         lvalue                                   rvalue
28  * 	      +-----------+                           +-----------+
29  * 	      | type: INT |                           | type: INT |
30  * 	 def  | id: INDEV |                           | id: VALUE |
31  * 	      | data:     |                           | data: 3   |
32  * 	      +-----------+                           +-----------+
33  * 	            |                                       |
34  * 	            ---> meta_ops[INT][INDEV](...)          |
35  *                            |                            |
36  * 	            -----------                             |
37  * 	            V                                       V
38  * 	      +-----------+                           +-----------+
39  * 	      | type: INT |                           | type: INT |
40  * 	 obj  | id: INDEV |                           | id: VALUE |
41  * 	      | data: 2   |<--data got filled out     | data: 3   |
42  * 	      +-----------+                           +-----------+
43  * 	            |                                         |
44  * 	            --------------> 2  equals 3 <--------------
45  *
46  * 	This is a simplified schema, the complexity varies depending
47  * 	on the meta type. Obviously, the length of the data must also
48  * 	be provided for non-numeric types.
49  *
50  * 	Additionaly, type dependant modifiers such as shift operators
51  * 	or mask may be applied to extend the functionaliy. As of now,
52  * 	the variable length type supports shifting the byte string to
53  * 	the right, eating up any number of octets and thus supporting
54  * 	wildcard interface name comparisons such as "ppp%" matching
55  * 	ppp0..9.
56  *
57  * 	NOTE: Certain meta values depend on other subsystems and are
58  * 	      only available if that subsytem is enabled in the kernel.
59  */
60 
61 #include <linux/config.h>
62 #include <linux/module.h>
63 #include <linux/types.h>
64 #include <linux/kernel.h>
65 #include <linux/sched.h>
66 #include <linux/string.h>
67 #include <linux/skbuff.h>
68 #include <linux/random.h>
69 #include <linux/tc_ematch/tc_em_meta.h>
70 #include <net/dst.h>
71 #include <net/route.h>
72 #include <net/pkt_cls.h>
73 
74 struct meta_obj
75 {
76 	unsigned long		value;
77 	unsigned int		len;
78 };
79 
80 struct meta_value
81 {
82 	struct tcf_meta_val	hdr;
83 	unsigned long		val;
84 	unsigned int		len;
85 };
86 
87 struct meta_match
88 {
89 	struct meta_value	lvalue;
90 	struct meta_value	rvalue;
91 };
92 
93 static inline int meta_id(struct meta_value *v)
94 {
95 	return TCF_META_ID(v->hdr.kind);
96 }
97 
98 static inline int meta_type(struct meta_value *v)
99 {
100 	return TCF_META_TYPE(v->hdr.kind);
101 }
102 
103 #define META_COLLECTOR(FUNC) static void meta_##FUNC(struct sk_buff *skb, \
104 	struct tcf_pkt_info *info, struct meta_value *v, \
105 	struct meta_obj *dst, int *err)
106 
107 /**************************************************************************
108  * System status & misc
109  **************************************************************************/
110 
111 META_COLLECTOR(int_random)
112 {
113 	get_random_bytes(&dst->value, sizeof(dst->value));
114 }
115 
116 static inline unsigned long fixed_loadavg(int load)
117 {
118 	int rnd_load = load + (FIXED_1/200);
119 	int rnd_frac = ((rnd_load & (FIXED_1-1)) * 100) >> FSHIFT;
120 
121 	return ((rnd_load >> FSHIFT) * 100) + rnd_frac;
122 }
123 
124 META_COLLECTOR(int_loadavg_0)
125 {
126 	dst->value = fixed_loadavg(avenrun[0]);
127 }
128 
129 META_COLLECTOR(int_loadavg_1)
130 {
131 	dst->value = fixed_loadavg(avenrun[1]);
132 }
133 
134 META_COLLECTOR(int_loadavg_2)
135 {
136 	dst->value = fixed_loadavg(avenrun[2]);
137 }
138 
139 /**************************************************************************
140  * Device names & indices
141  **************************************************************************/
142 
143 static inline int int_dev(struct net_device *dev, struct meta_obj *dst)
144 {
145 	if (unlikely(dev == NULL))
146 		return -1;
147 
148 	dst->value = dev->ifindex;
149 	return 0;
150 }
151 
152 static inline int var_dev(struct net_device *dev, struct meta_obj *dst)
153 {
154 	if (unlikely(dev == NULL))
155 		return -1;
156 
157 	dst->value = (unsigned long) dev->name;
158 	dst->len = strlen(dev->name);
159 	return 0;
160 }
161 
162 META_COLLECTOR(int_dev)
163 {
164 	*err = int_dev(skb->dev, dst);
165 }
166 
167 META_COLLECTOR(var_dev)
168 {
169 	*err = var_dev(skb->dev, dst);
170 }
171 
172 META_COLLECTOR(int_indev)
173 {
174 	*err = int_dev(skb->input_dev, dst);
175 }
176 
177 META_COLLECTOR(var_indev)
178 {
179 	*err = var_dev(skb->input_dev, dst);
180 }
181 
182 META_COLLECTOR(int_realdev)
183 {
184 	*err = int_dev(skb->real_dev, dst);
185 }
186 
187 META_COLLECTOR(var_realdev)
188 {
189 	*err = var_dev(skb->real_dev, dst);
190 }
191 
192 /**************************************************************************
193  * skb attributes
194  **************************************************************************/
195 
196 META_COLLECTOR(int_priority)
197 {
198 	dst->value = skb->priority;
199 }
200 
201 META_COLLECTOR(int_protocol)
202 {
203 	/* Let userspace take care of the byte ordering */
204 	dst->value = skb->protocol;
205 }
206 
207 META_COLLECTOR(int_security)
208 {
209 	dst->value = skb->security;
210 }
211 
212 META_COLLECTOR(int_pkttype)
213 {
214 	dst->value = skb->pkt_type;
215 }
216 
217 META_COLLECTOR(int_pktlen)
218 {
219 	dst->value = skb->len;
220 }
221 
222 META_COLLECTOR(int_datalen)
223 {
224 	dst->value = skb->data_len;
225 }
226 
227 META_COLLECTOR(int_maclen)
228 {
229 	dst->value = skb->mac_len;
230 }
231 
232 /**************************************************************************
233  * Netfilter
234  **************************************************************************/
235 
236 #ifdef CONFIG_NETFILTER
237 META_COLLECTOR(int_nfmark)
238 {
239 	dst->value = skb->nfmark;
240 }
241 #endif
242 
243 /**************************************************************************
244  * Traffic Control
245  **************************************************************************/
246 
247 META_COLLECTOR(int_tcindex)
248 {
249 	dst->value = skb->tc_index;
250 }
251 
252 #ifdef CONFIG_NET_CLS_ACT
253 META_COLLECTOR(int_tcverd)
254 {
255 	dst->value = skb->tc_verd;
256 }
257 
258 META_COLLECTOR(int_tcclassid)
259 {
260 	dst->value = skb->tc_classid;
261 }
262 #endif
263 
264 /**************************************************************************
265  * Routing
266  **************************************************************************/
267 
268 #ifdef CONFIG_NET_CLS_ROUTE
269 META_COLLECTOR(int_rtclassid)
270 {
271 	if (unlikely(skb->dst == NULL))
272 		*err = -1;
273 	else
274 		dst->value = skb->dst->tclassid;
275 }
276 #endif
277 
278 META_COLLECTOR(int_rtiif)
279 {
280 	if (unlikely(skb->dst == NULL))
281 		*err = -1;
282 	else
283 		dst->value = ((struct rtable*) skb->dst)->fl.iif;
284 }
285 
286 /**************************************************************************
287  * Meta value collectors assignment table
288  **************************************************************************/
289 
290 struct meta_ops
291 {
292 	void		(*get)(struct sk_buff *, struct tcf_pkt_info *,
293 			       struct meta_value *, struct meta_obj *, int *);
294 };
295 
296 /* Meta value operations table listing all meta value collectors and
297  * assigns them to a type and meta id. */
298 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX+1][TCF_META_ID_MAX+1] = {
299 	[TCF_META_TYPE_VAR] = {
300 		[TCF_META_ID_DEV]	= { .get = meta_var_dev },
301 		[TCF_META_ID_INDEV]	= { .get = meta_var_indev },
302 		[TCF_META_ID_REALDEV]	= { .get = meta_var_realdev }
303 	},
304 	[TCF_META_TYPE_INT] = {
305 		[TCF_META_ID_RANDOM]	= { .get = meta_int_random },
306 		[TCF_META_ID_LOADAVG_0]	= { .get = meta_int_loadavg_0 },
307 		[TCF_META_ID_LOADAVG_1]	= { .get = meta_int_loadavg_1 },
308 		[TCF_META_ID_LOADAVG_2]	= { .get = meta_int_loadavg_2 },
309 		[TCF_META_ID_DEV]	= { .get = meta_int_dev },
310 		[TCF_META_ID_INDEV]	= { .get = meta_int_indev },
311 		[TCF_META_ID_REALDEV]	= { .get = meta_int_realdev },
312 		[TCF_META_ID_PRIORITY]	= { .get = meta_int_priority },
313 		[TCF_META_ID_PROTOCOL]	= { .get = meta_int_protocol },
314 		[TCF_META_ID_SECURITY]	= { .get = meta_int_security },
315 		[TCF_META_ID_PKTTYPE]	= { .get = meta_int_pkttype },
316 		[TCF_META_ID_PKTLEN]	= { .get = meta_int_pktlen },
317 		[TCF_META_ID_DATALEN]	= { .get = meta_int_datalen },
318 		[TCF_META_ID_MACLEN]	= { .get = meta_int_maclen },
319 #ifdef CONFIG_NETFILTER
320 		[TCF_META_ID_NFMARK]	= { .get = meta_int_nfmark },
321 #endif
322 		[TCF_META_ID_TCINDEX]	= { .get = meta_int_tcindex },
323 #ifdef CONFIG_NET_CLS_ACT
324 		[TCF_META_ID_TCVERDICT]	= { .get = meta_int_tcverd },
325 		[TCF_META_ID_TCCLASSID]	= { .get = meta_int_tcclassid },
326 #endif
327 #ifdef CONFIG_NET_CLS_ROUTE
328 		[TCF_META_ID_RTCLASSID]	= { .get = meta_int_rtclassid },
329 #endif
330 		[TCF_META_ID_RTIIF]	= { .get = meta_int_rtiif }
331 	}
332 };
333 
334 static inline struct meta_ops * meta_ops(struct meta_value *val)
335 {
336 	return &__meta_ops[meta_type(val)][meta_id(val)];
337 }
338 
339 /**************************************************************************
340  * Type specific operations for TCF_META_TYPE_VAR
341  **************************************************************************/
342 
343 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
344 {
345 	int r = a->len - b->len;
346 
347 	if (r == 0)
348 		r = memcmp((void *) a->value, (void *) b->value, a->len);
349 
350 	return r;
351 }
352 
353 static int meta_var_change(struct meta_value *dst, struct rtattr *rta)
354 {
355 	int len = RTA_PAYLOAD(rta);
356 
357 	dst->val = (unsigned long) kmalloc(len, GFP_KERNEL);
358 	if (dst->val == 0UL)
359 		return -ENOMEM;
360 	memcpy((void *) dst->val, RTA_DATA(rta), len);
361 	dst->len = len;
362 	return 0;
363 }
364 
365 static void meta_var_destroy(struct meta_value *v)
366 {
367 	if (v->val)
368 		kfree((void *) v->val);
369 }
370 
371 static void meta_var_apply_extras(struct meta_value *v,
372 				  struct meta_obj *dst)
373 {
374 	int shift = v->hdr.shift;
375 
376 	if (shift && shift < dst->len)
377 		dst->len -= shift;
378 }
379 
380 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
381 {
382 	if (v->val && v->len)
383 		RTA_PUT(skb, tlv, v->len, (void *) v->val);
384 	return 0;
385 
386 rtattr_failure:
387 	return -1;
388 }
389 
390 /**************************************************************************
391  * Type specific operations for TCF_META_TYPE_INT
392  **************************************************************************/
393 
394 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
395 {
396 	/* Let gcc optimize it, the unlikely is not really based on
397 	 * some numbers but jump free code for mismatches seems
398 	 * more logical. */
399 	if (unlikely(a == b))
400 		return 0;
401 	else if (a < b)
402 		return -1;
403 	else
404 		return 1;
405 }
406 
407 static int meta_int_change(struct meta_value *dst, struct rtattr *rta)
408 {
409 	if (RTA_PAYLOAD(rta) >= sizeof(unsigned long)) {
410 		dst->val = *(unsigned long *) RTA_DATA(rta);
411 		dst->len = sizeof(unsigned long);
412 	} else if (RTA_PAYLOAD(rta) == sizeof(u32)) {
413 		dst->val = *(u32 *) RTA_DATA(rta);
414 		dst->len = sizeof(u32);
415 	} else
416 		return -EINVAL;
417 
418 	return 0;
419 }
420 
421 static void meta_int_apply_extras(struct meta_value *v,
422 				  struct meta_obj *dst)
423 {
424 	if (v->hdr.shift)
425 		dst->value >>= v->hdr.shift;
426 
427 	if (v->val)
428 		dst->value &= v->val;
429 }
430 
431 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
432 {
433 	if (v->len == sizeof(unsigned long))
434 		RTA_PUT(skb, tlv, sizeof(unsigned long), &v->val);
435 	else if (v->len == sizeof(u32)) {
436 		u32 d = v->val;
437 		RTA_PUT(skb, tlv, sizeof(d), &d);
438 	}
439 
440 	return 0;
441 
442 rtattr_failure:
443 	return -1;
444 }
445 
446 /**************************************************************************
447  * Type specific operations table
448  **************************************************************************/
449 
450 struct meta_type_ops
451 {
452 	void	(*destroy)(struct meta_value *);
453 	int	(*compare)(struct meta_obj *, struct meta_obj *);
454 	int	(*change)(struct meta_value *, struct rtattr *);
455 	void	(*apply_extras)(struct meta_value *, struct meta_obj *);
456 	int	(*dump)(struct sk_buff *, struct meta_value *, int);
457 };
458 
459 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX+1] = {
460 	[TCF_META_TYPE_VAR] = {
461 		.destroy = meta_var_destroy,
462 		.compare = meta_var_compare,
463 		.change = meta_var_change,
464 		.apply_extras = meta_var_apply_extras,
465 		.dump = meta_var_dump
466 	},
467 	[TCF_META_TYPE_INT] = {
468 		.compare = meta_int_compare,
469 		.change = meta_int_change,
470 		.apply_extras = meta_int_apply_extras,
471 		.dump = meta_int_dump
472 	}
473 };
474 
475 static inline struct meta_type_ops * meta_type_ops(struct meta_value *v)
476 {
477 	return &__meta_type_ops[meta_type(v)];
478 }
479 
480 /**************************************************************************
481  * Core
482  **************************************************************************/
483 
484 static inline int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
485 			   struct meta_value *v, struct meta_obj *dst)
486 {
487 	int err = 0;
488 
489 	if (meta_id(v) == TCF_META_ID_VALUE) {
490 		dst->value = v->val;
491 		dst->len = v->len;
492 		return 0;
493 	}
494 
495 	meta_ops(v)->get(skb, info, v, dst, &err);
496 	if (err < 0)
497 		return err;
498 
499 	if (meta_type_ops(v)->apply_extras)
500 	    meta_type_ops(v)->apply_extras(v, dst);
501 
502 	return 0;
503 }
504 
505 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
506 			 struct tcf_pkt_info *info)
507 {
508 	int r;
509 	struct meta_match *meta = (struct meta_match *) m->data;
510 	struct meta_obj l_value, r_value;
511 
512 	if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
513 	    meta_get(skb, info, &meta->rvalue, &r_value) < 0)
514 		return 0;
515 
516 	r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
517 
518 	switch (meta->lvalue.hdr.op) {
519 		case TCF_EM_OPND_EQ:
520 			return !r;
521 		case TCF_EM_OPND_LT:
522 			return r < 0;
523 		case TCF_EM_OPND_GT:
524 			return r > 0;
525 	}
526 
527 	return 0;
528 }
529 
530 static inline void meta_delete(struct meta_match *meta)
531 {
532 	struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
533 
534 	if (ops && ops->destroy) {
535 		ops->destroy(&meta->lvalue);
536 		ops->destroy(&meta->rvalue);
537 	}
538 
539 	kfree(meta);
540 }
541 
542 static inline int meta_change_data(struct meta_value *dst, struct rtattr *rta)
543 {
544 	if (rta) {
545 		if (RTA_PAYLOAD(rta) == 0)
546 			return -EINVAL;
547 
548 		return meta_type_ops(dst)->change(dst, rta);
549 	}
550 
551 	return 0;
552 }
553 
554 static inline int meta_is_supported(struct meta_value *val)
555 {
556 	return (!meta_id(val) || meta_ops(val)->get);
557 }
558 
559 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
560 			  struct tcf_ematch *m)
561 {
562 	int err = -EINVAL;
563 	struct rtattr *tb[TCA_EM_META_MAX];
564 	struct tcf_meta_hdr *hdr;
565 	struct meta_match *meta = NULL;
566 
567 	if (rtattr_parse(tb, TCA_EM_META_MAX, data, len) < 0)
568 		goto errout;
569 
570 	if (tb[TCA_EM_META_HDR-1] == NULL ||
571 	    RTA_PAYLOAD(tb[TCA_EM_META_HDR-1]) < sizeof(*hdr))
572 		goto errout;
573 	hdr = RTA_DATA(tb[TCA_EM_META_HDR-1]);
574 
575 	if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
576 	    TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
577 	    TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
578 	    TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
579 		goto errout;
580 
581 	meta = kmalloc(sizeof(*meta), GFP_KERNEL);
582 	if (meta == NULL)
583 		goto errout;
584 	memset(meta, 0, sizeof(*meta));
585 
586 	memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
587 	memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
588 
589 	if (!meta_is_supported(&meta->lvalue) ||
590 	    !meta_is_supported(&meta->rvalue)) {
591 		err = -EOPNOTSUPP;
592 		goto errout;
593 	}
594 
595 	if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE-1]) < 0 ||
596 	    meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE-1]) < 0)
597 		goto errout;
598 
599 	m->datalen = sizeof(*meta);
600 	m->data = (unsigned long) meta;
601 
602 	err = 0;
603 errout:
604 	if (err && meta)
605 		meta_delete(meta);
606 	return err;
607 }
608 
609 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
610 {
611 	if (m)
612 		meta_delete((struct meta_match *) m->data);
613 }
614 
615 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
616 {
617 	struct meta_match *meta = (struct meta_match *) em->data;
618 	struct tcf_meta_hdr hdr;
619 	struct meta_type_ops *ops;
620 
621 	memset(&hdr, 0, sizeof(hdr));
622 	memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
623 	memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
624 
625 	RTA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
626 
627 	ops = meta_type_ops(&meta->lvalue);
628 	if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
629 	    ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
630 		goto rtattr_failure;
631 
632 	return 0;
633 
634 rtattr_failure:
635 	return -1;
636 }
637 
638 static struct tcf_ematch_ops em_meta_ops = {
639 	.kind	  = TCF_EM_META,
640 	.change	  = em_meta_change,
641 	.match	  = em_meta_match,
642 	.destroy  = em_meta_destroy,
643 	.dump	  = em_meta_dump,
644 	.owner	  = THIS_MODULE,
645 	.link	  = LIST_HEAD_INIT(em_meta_ops.link)
646 };
647 
648 static int __init init_em_meta(void)
649 {
650 	return tcf_em_register(&em_meta_ops);
651 }
652 
653 static void __exit exit_em_meta(void)
654 {
655 	tcf_em_unregister(&em_meta_ops);
656 }
657 
658 MODULE_LICENSE("GPL");
659 
660 module_init(init_em_meta);
661 module_exit(exit_em_meta);
662