xref: /openbmc/linux/net/sched/em_meta.c (revision 239480ab)
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: DEV   |                           | id: VALUE |
31  * 	      | data:     |                           | data: 3   |
32  * 	      +-----------+                           +-----------+
33  * 	            |                                       |
34  * 	            ---> meta_ops[INT][DEV](...)            |
35  *	                      |                             |
36  * 	            -----------                             |
37  * 	            V                                       V
38  * 	      +-----------+                           +-----------+
39  * 	      | type: INT |                           | type: INT |
40  * 	 obj  | id: DEV |                             | 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  * 	Additionally, type dependent 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 subsystem is enabled in the kernel.
59  */
60 
61 #include <linux/slab.h>
62 #include <linux/module.h>
63 #include <linux/types.h>
64 #include <linux/kernel.h>
65 #include <linux/sched.h>
66 #include <linux/sched/loadavg.h>
67 #include <linux/string.h>
68 #include <linux/skbuff.h>
69 #include <linux/random.h>
70 #include <linux/if_vlan.h>
71 #include <linux/tc_ematch/tc_em_meta.h>
72 #include <net/dst.h>
73 #include <net/route.h>
74 #include <net/pkt_cls.h>
75 #include <net/sock.h>
76 
77 struct meta_obj {
78 	unsigned long		value;
79 	unsigned int		len;
80 };
81 
82 struct meta_value {
83 	struct tcf_meta_val	hdr;
84 	unsigned long		val;
85 	unsigned int		len;
86 };
87 
88 struct meta_match {
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 /**************************************************************************
173  * vlan tag
174  **************************************************************************/
175 
176 META_COLLECTOR(int_vlan_tag)
177 {
178 	unsigned short tag;
179 
180 	if (skb_vlan_tag_present(skb))
181 		dst->value = skb_vlan_tag_get(skb);
182 	else if (!__vlan_get_tag(skb, &tag))
183 		dst->value = tag;
184 	else
185 		*err = -1;
186 }
187 
188 
189 
190 /**************************************************************************
191  * skb attributes
192  **************************************************************************/
193 
194 META_COLLECTOR(int_priority)
195 {
196 	dst->value = skb->priority;
197 }
198 
199 META_COLLECTOR(int_protocol)
200 {
201 	/* Let userspace take care of the byte ordering */
202 	dst->value = tc_skb_protocol(skb);
203 }
204 
205 META_COLLECTOR(int_pkttype)
206 {
207 	dst->value = skb->pkt_type;
208 }
209 
210 META_COLLECTOR(int_pktlen)
211 {
212 	dst->value = skb->len;
213 }
214 
215 META_COLLECTOR(int_datalen)
216 {
217 	dst->value = skb->data_len;
218 }
219 
220 META_COLLECTOR(int_maclen)
221 {
222 	dst->value = skb->mac_len;
223 }
224 
225 META_COLLECTOR(int_rxhash)
226 {
227 	dst->value = skb_get_hash(skb);
228 }
229 
230 /**************************************************************************
231  * Netfilter
232  **************************************************************************/
233 
234 META_COLLECTOR(int_mark)
235 {
236 	dst->value = skb->mark;
237 }
238 
239 /**************************************************************************
240  * Traffic Control
241  **************************************************************************/
242 
243 META_COLLECTOR(int_tcindex)
244 {
245 	dst->value = skb->tc_index;
246 }
247 
248 /**************************************************************************
249  * Routing
250  **************************************************************************/
251 
252 META_COLLECTOR(int_rtclassid)
253 {
254 	if (unlikely(skb_dst(skb) == NULL))
255 		*err = -1;
256 	else
257 #ifdef CONFIG_IP_ROUTE_CLASSID
258 		dst->value = skb_dst(skb)->tclassid;
259 #else
260 		dst->value = 0;
261 #endif
262 }
263 
264 META_COLLECTOR(int_rtiif)
265 {
266 	if (unlikely(skb_rtable(skb) == NULL))
267 		*err = -1;
268 	else
269 		dst->value = inet_iif(skb);
270 }
271 
272 /**************************************************************************
273  * Socket Attributes
274  **************************************************************************/
275 
276 #define skip_nonlocal(skb) \
277 	(unlikely(skb->sk == NULL))
278 
279 META_COLLECTOR(int_sk_family)
280 {
281 	if (skip_nonlocal(skb)) {
282 		*err = -1;
283 		return;
284 	}
285 	dst->value = skb->sk->sk_family;
286 }
287 
288 META_COLLECTOR(int_sk_state)
289 {
290 	if (skip_nonlocal(skb)) {
291 		*err = -1;
292 		return;
293 	}
294 	dst->value = skb->sk->sk_state;
295 }
296 
297 META_COLLECTOR(int_sk_reuse)
298 {
299 	if (skip_nonlocal(skb)) {
300 		*err = -1;
301 		return;
302 	}
303 	dst->value = skb->sk->sk_reuse;
304 }
305 
306 META_COLLECTOR(int_sk_bound_if)
307 {
308 	if (skip_nonlocal(skb)) {
309 		*err = -1;
310 		return;
311 	}
312 	/* No error if bound_dev_if is 0, legal userspace check */
313 	dst->value = skb->sk->sk_bound_dev_if;
314 }
315 
316 META_COLLECTOR(var_sk_bound_if)
317 {
318 	if (skip_nonlocal(skb)) {
319 		*err = -1;
320 		return;
321 	}
322 
323 	if (skb->sk->sk_bound_dev_if == 0) {
324 		dst->value = (unsigned long) "any";
325 		dst->len = 3;
326 	} else {
327 		struct net_device *dev;
328 
329 		rcu_read_lock();
330 		dev = dev_get_by_index_rcu(sock_net(skb->sk),
331 					   skb->sk->sk_bound_dev_if);
332 		*err = var_dev(dev, dst);
333 		rcu_read_unlock();
334 	}
335 }
336 
337 META_COLLECTOR(int_sk_refcnt)
338 {
339 	if (skip_nonlocal(skb)) {
340 		*err = -1;
341 		return;
342 	}
343 	dst->value = atomic_read(&skb->sk->sk_refcnt);
344 }
345 
346 META_COLLECTOR(int_sk_rcvbuf)
347 {
348 	const struct sock *sk = skb_to_full_sk(skb);
349 
350 	if (!sk) {
351 		*err = -1;
352 		return;
353 	}
354 	dst->value = sk->sk_rcvbuf;
355 }
356 
357 META_COLLECTOR(int_sk_shutdown)
358 {
359 	const struct sock *sk = skb_to_full_sk(skb);
360 
361 	if (!sk) {
362 		*err = -1;
363 		return;
364 	}
365 	dst->value = sk->sk_shutdown;
366 }
367 
368 META_COLLECTOR(int_sk_proto)
369 {
370 	const struct sock *sk = skb_to_full_sk(skb);
371 
372 	if (!sk) {
373 		*err = -1;
374 		return;
375 	}
376 	dst->value = sk->sk_protocol;
377 }
378 
379 META_COLLECTOR(int_sk_type)
380 {
381 	const struct sock *sk = skb_to_full_sk(skb);
382 
383 	if (!sk) {
384 		*err = -1;
385 		return;
386 	}
387 	dst->value = sk->sk_type;
388 }
389 
390 META_COLLECTOR(int_sk_rmem_alloc)
391 {
392 	const struct sock *sk = skb_to_full_sk(skb);
393 
394 	if (!sk) {
395 		*err = -1;
396 		return;
397 	}
398 	dst->value = sk_rmem_alloc_get(sk);
399 }
400 
401 META_COLLECTOR(int_sk_wmem_alloc)
402 {
403 	const struct sock *sk = skb_to_full_sk(skb);
404 
405 	if (!sk) {
406 		*err = -1;
407 		return;
408 	}
409 	dst->value = sk_wmem_alloc_get(sk);
410 }
411 
412 META_COLLECTOR(int_sk_omem_alloc)
413 {
414 	const struct sock *sk = skb_to_full_sk(skb);
415 
416 	if (!sk) {
417 		*err = -1;
418 		return;
419 	}
420 	dst->value = atomic_read(&sk->sk_omem_alloc);
421 }
422 
423 META_COLLECTOR(int_sk_rcv_qlen)
424 {
425 	const struct sock *sk = skb_to_full_sk(skb);
426 
427 	if (!sk) {
428 		*err = -1;
429 		return;
430 	}
431 	dst->value = sk->sk_receive_queue.qlen;
432 }
433 
434 META_COLLECTOR(int_sk_snd_qlen)
435 {
436 	const struct sock *sk = skb_to_full_sk(skb);
437 
438 	if (!sk) {
439 		*err = -1;
440 		return;
441 	}
442 	dst->value = sk->sk_write_queue.qlen;
443 }
444 
445 META_COLLECTOR(int_sk_wmem_queued)
446 {
447 	const struct sock *sk = skb_to_full_sk(skb);
448 
449 	if (!sk) {
450 		*err = -1;
451 		return;
452 	}
453 	dst->value = sk->sk_wmem_queued;
454 }
455 
456 META_COLLECTOR(int_sk_fwd_alloc)
457 {
458 	const struct sock *sk = skb_to_full_sk(skb);
459 
460 	if (!sk) {
461 		*err = -1;
462 		return;
463 	}
464 	dst->value = sk->sk_forward_alloc;
465 }
466 
467 META_COLLECTOR(int_sk_sndbuf)
468 {
469 	const struct sock *sk = skb_to_full_sk(skb);
470 
471 	if (!sk) {
472 		*err = -1;
473 		return;
474 	}
475 	dst->value = sk->sk_sndbuf;
476 }
477 
478 META_COLLECTOR(int_sk_alloc)
479 {
480 	const struct sock *sk = skb_to_full_sk(skb);
481 
482 	if (!sk) {
483 		*err = -1;
484 		return;
485 	}
486 	dst->value = (__force int) sk->sk_allocation;
487 }
488 
489 META_COLLECTOR(int_sk_hash)
490 {
491 	if (skip_nonlocal(skb)) {
492 		*err = -1;
493 		return;
494 	}
495 	dst->value = skb->sk->sk_hash;
496 }
497 
498 META_COLLECTOR(int_sk_lingertime)
499 {
500 	const struct sock *sk = skb_to_full_sk(skb);
501 
502 	if (!sk) {
503 		*err = -1;
504 		return;
505 	}
506 	dst->value = sk->sk_lingertime / HZ;
507 }
508 
509 META_COLLECTOR(int_sk_err_qlen)
510 {
511 	const struct sock *sk = skb_to_full_sk(skb);
512 
513 	if (!sk) {
514 		*err = -1;
515 		return;
516 	}
517 	dst->value = sk->sk_error_queue.qlen;
518 }
519 
520 META_COLLECTOR(int_sk_ack_bl)
521 {
522 	const struct sock *sk = skb_to_full_sk(skb);
523 
524 	if (!sk) {
525 		*err = -1;
526 		return;
527 	}
528 	dst->value = sk->sk_ack_backlog;
529 }
530 
531 META_COLLECTOR(int_sk_max_ack_bl)
532 {
533 	const struct sock *sk = skb_to_full_sk(skb);
534 
535 	if (!sk) {
536 		*err = -1;
537 		return;
538 	}
539 	dst->value = sk->sk_max_ack_backlog;
540 }
541 
542 META_COLLECTOR(int_sk_prio)
543 {
544 	const struct sock *sk = skb_to_full_sk(skb);
545 
546 	if (!sk) {
547 		*err = -1;
548 		return;
549 	}
550 	dst->value = sk->sk_priority;
551 }
552 
553 META_COLLECTOR(int_sk_rcvlowat)
554 {
555 	const struct sock *sk = skb_to_full_sk(skb);
556 
557 	if (!sk) {
558 		*err = -1;
559 		return;
560 	}
561 	dst->value = sk->sk_rcvlowat;
562 }
563 
564 META_COLLECTOR(int_sk_rcvtimeo)
565 {
566 	const struct sock *sk = skb_to_full_sk(skb);
567 
568 	if (!sk) {
569 		*err = -1;
570 		return;
571 	}
572 	dst->value = sk->sk_rcvtimeo / HZ;
573 }
574 
575 META_COLLECTOR(int_sk_sndtimeo)
576 {
577 	const struct sock *sk = skb_to_full_sk(skb);
578 
579 	if (!sk) {
580 		*err = -1;
581 		return;
582 	}
583 	dst->value = sk->sk_sndtimeo / HZ;
584 }
585 
586 META_COLLECTOR(int_sk_sendmsg_off)
587 {
588 	const struct sock *sk = skb_to_full_sk(skb);
589 
590 	if (!sk) {
591 		*err = -1;
592 		return;
593 	}
594 	dst->value = sk->sk_frag.offset;
595 }
596 
597 META_COLLECTOR(int_sk_write_pend)
598 {
599 	const struct sock *sk = skb_to_full_sk(skb);
600 
601 	if (!sk) {
602 		*err = -1;
603 		return;
604 	}
605 	dst->value = sk->sk_write_pending;
606 }
607 
608 /**************************************************************************
609  * Meta value collectors assignment table
610  **************************************************************************/
611 
612 struct meta_ops {
613 	void		(*get)(struct sk_buff *, struct tcf_pkt_info *,
614 			       struct meta_value *, struct meta_obj *, int *);
615 };
616 
617 #define META_ID(name) TCF_META_ID_##name
618 #define META_FUNC(name) { .get = meta_##name }
619 
620 /* Meta value operations table listing all meta value collectors and
621  * assigns them to a type and meta id. */
622 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX + 1][TCF_META_ID_MAX + 1] = {
623 	[TCF_META_TYPE_VAR] = {
624 		[META_ID(DEV)]			= META_FUNC(var_dev),
625 		[META_ID(SK_BOUND_IF)] 		= META_FUNC(var_sk_bound_if),
626 	},
627 	[TCF_META_TYPE_INT] = {
628 		[META_ID(RANDOM)]		= META_FUNC(int_random),
629 		[META_ID(LOADAVG_0)]		= META_FUNC(int_loadavg_0),
630 		[META_ID(LOADAVG_1)]		= META_FUNC(int_loadavg_1),
631 		[META_ID(LOADAVG_2)]		= META_FUNC(int_loadavg_2),
632 		[META_ID(DEV)]			= META_FUNC(int_dev),
633 		[META_ID(PRIORITY)]		= META_FUNC(int_priority),
634 		[META_ID(PROTOCOL)]		= META_FUNC(int_protocol),
635 		[META_ID(PKTTYPE)]		= META_FUNC(int_pkttype),
636 		[META_ID(PKTLEN)]		= META_FUNC(int_pktlen),
637 		[META_ID(DATALEN)]		= META_FUNC(int_datalen),
638 		[META_ID(MACLEN)]		= META_FUNC(int_maclen),
639 		[META_ID(NFMARK)]		= META_FUNC(int_mark),
640 		[META_ID(TCINDEX)]		= META_FUNC(int_tcindex),
641 		[META_ID(RTCLASSID)]		= META_FUNC(int_rtclassid),
642 		[META_ID(RTIIF)]		= META_FUNC(int_rtiif),
643 		[META_ID(SK_FAMILY)]		= META_FUNC(int_sk_family),
644 		[META_ID(SK_STATE)]		= META_FUNC(int_sk_state),
645 		[META_ID(SK_REUSE)]		= META_FUNC(int_sk_reuse),
646 		[META_ID(SK_BOUND_IF)]		= META_FUNC(int_sk_bound_if),
647 		[META_ID(SK_REFCNT)]		= META_FUNC(int_sk_refcnt),
648 		[META_ID(SK_RCVBUF)]		= META_FUNC(int_sk_rcvbuf),
649 		[META_ID(SK_SNDBUF)]		= META_FUNC(int_sk_sndbuf),
650 		[META_ID(SK_SHUTDOWN)]		= META_FUNC(int_sk_shutdown),
651 		[META_ID(SK_PROTO)]		= META_FUNC(int_sk_proto),
652 		[META_ID(SK_TYPE)]		= META_FUNC(int_sk_type),
653 		[META_ID(SK_RMEM_ALLOC)]	= META_FUNC(int_sk_rmem_alloc),
654 		[META_ID(SK_WMEM_ALLOC)]	= META_FUNC(int_sk_wmem_alloc),
655 		[META_ID(SK_OMEM_ALLOC)]	= META_FUNC(int_sk_omem_alloc),
656 		[META_ID(SK_WMEM_QUEUED)]	= META_FUNC(int_sk_wmem_queued),
657 		[META_ID(SK_RCV_QLEN)]		= META_FUNC(int_sk_rcv_qlen),
658 		[META_ID(SK_SND_QLEN)]		= META_FUNC(int_sk_snd_qlen),
659 		[META_ID(SK_ERR_QLEN)]		= META_FUNC(int_sk_err_qlen),
660 		[META_ID(SK_FORWARD_ALLOCS)]	= META_FUNC(int_sk_fwd_alloc),
661 		[META_ID(SK_ALLOCS)]		= META_FUNC(int_sk_alloc),
662 		[META_ID(SK_HASH)]		= META_FUNC(int_sk_hash),
663 		[META_ID(SK_LINGERTIME)]	= META_FUNC(int_sk_lingertime),
664 		[META_ID(SK_ACK_BACKLOG)]	= META_FUNC(int_sk_ack_bl),
665 		[META_ID(SK_MAX_ACK_BACKLOG)]	= META_FUNC(int_sk_max_ack_bl),
666 		[META_ID(SK_PRIO)]		= META_FUNC(int_sk_prio),
667 		[META_ID(SK_RCVLOWAT)]		= META_FUNC(int_sk_rcvlowat),
668 		[META_ID(SK_RCVTIMEO)]		= META_FUNC(int_sk_rcvtimeo),
669 		[META_ID(SK_SNDTIMEO)]		= META_FUNC(int_sk_sndtimeo),
670 		[META_ID(SK_SENDMSG_OFF)]	= META_FUNC(int_sk_sendmsg_off),
671 		[META_ID(SK_WRITE_PENDING)]	= META_FUNC(int_sk_write_pend),
672 		[META_ID(VLAN_TAG)]		= META_FUNC(int_vlan_tag),
673 		[META_ID(RXHASH)]		= META_FUNC(int_rxhash),
674 	}
675 };
676 
677 static inline struct meta_ops *meta_ops(struct meta_value *val)
678 {
679 	return &__meta_ops[meta_type(val)][meta_id(val)];
680 }
681 
682 /**************************************************************************
683  * Type specific operations for TCF_META_TYPE_VAR
684  **************************************************************************/
685 
686 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
687 {
688 	int r = a->len - b->len;
689 
690 	if (r == 0)
691 		r = memcmp((void *) a->value, (void *) b->value, a->len);
692 
693 	return r;
694 }
695 
696 static int meta_var_change(struct meta_value *dst, struct nlattr *nla)
697 {
698 	int len = nla_len(nla);
699 
700 	dst->val = (unsigned long)kmemdup(nla_data(nla), len, GFP_KERNEL);
701 	if (dst->val == 0UL)
702 		return -ENOMEM;
703 	dst->len = len;
704 	return 0;
705 }
706 
707 static void meta_var_destroy(struct meta_value *v)
708 {
709 	kfree((void *) v->val);
710 }
711 
712 static void meta_var_apply_extras(struct meta_value *v,
713 				  struct meta_obj *dst)
714 {
715 	int shift = v->hdr.shift;
716 
717 	if (shift && shift < dst->len)
718 		dst->len -= shift;
719 }
720 
721 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
722 {
723 	if (v->val && v->len &&
724 	    nla_put(skb, tlv, v->len, (void *) v->val))
725 		goto nla_put_failure;
726 	return 0;
727 
728 nla_put_failure:
729 	return -1;
730 }
731 
732 /**************************************************************************
733  * Type specific operations for TCF_META_TYPE_INT
734  **************************************************************************/
735 
736 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
737 {
738 	/* Let gcc optimize it, the unlikely is not really based on
739 	 * some numbers but jump free code for mismatches seems
740 	 * more logical. */
741 	if (unlikely(a->value == b->value))
742 		return 0;
743 	else if (a->value < b->value)
744 		return -1;
745 	else
746 		return 1;
747 }
748 
749 static int meta_int_change(struct meta_value *dst, struct nlattr *nla)
750 {
751 	if (nla_len(nla) >= sizeof(unsigned long)) {
752 		dst->val = *(unsigned long *) nla_data(nla);
753 		dst->len = sizeof(unsigned long);
754 	} else if (nla_len(nla) == sizeof(u32)) {
755 		dst->val = nla_get_u32(nla);
756 		dst->len = sizeof(u32);
757 	} else
758 		return -EINVAL;
759 
760 	return 0;
761 }
762 
763 static void meta_int_apply_extras(struct meta_value *v,
764 				  struct meta_obj *dst)
765 {
766 	if (v->hdr.shift)
767 		dst->value >>= v->hdr.shift;
768 
769 	if (v->val)
770 		dst->value &= v->val;
771 }
772 
773 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
774 {
775 	if (v->len == sizeof(unsigned long)) {
776 		if (nla_put(skb, tlv, sizeof(unsigned long), &v->val))
777 			goto nla_put_failure;
778 	} else if (v->len == sizeof(u32)) {
779 		if (nla_put_u32(skb, tlv, v->val))
780 			goto nla_put_failure;
781 	}
782 
783 	return 0;
784 
785 nla_put_failure:
786 	return -1;
787 }
788 
789 /**************************************************************************
790  * Type specific operations table
791  **************************************************************************/
792 
793 struct meta_type_ops {
794 	void	(*destroy)(struct meta_value *);
795 	int	(*compare)(struct meta_obj *, struct meta_obj *);
796 	int	(*change)(struct meta_value *, struct nlattr *);
797 	void	(*apply_extras)(struct meta_value *, struct meta_obj *);
798 	int	(*dump)(struct sk_buff *, struct meta_value *, int);
799 };
800 
801 static const struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX + 1] = {
802 	[TCF_META_TYPE_VAR] = {
803 		.destroy = meta_var_destroy,
804 		.compare = meta_var_compare,
805 		.change = meta_var_change,
806 		.apply_extras = meta_var_apply_extras,
807 		.dump = meta_var_dump
808 	},
809 	[TCF_META_TYPE_INT] = {
810 		.compare = meta_int_compare,
811 		.change = meta_int_change,
812 		.apply_extras = meta_int_apply_extras,
813 		.dump = meta_int_dump
814 	}
815 };
816 
817 static inline const struct meta_type_ops *meta_type_ops(struct meta_value *v)
818 {
819 	return &__meta_type_ops[meta_type(v)];
820 }
821 
822 /**************************************************************************
823  * Core
824  **************************************************************************/
825 
826 static int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
827 		    struct meta_value *v, struct meta_obj *dst)
828 {
829 	int err = 0;
830 
831 	if (meta_id(v) == TCF_META_ID_VALUE) {
832 		dst->value = v->val;
833 		dst->len = v->len;
834 		return 0;
835 	}
836 
837 	meta_ops(v)->get(skb, info, v, dst, &err);
838 	if (err < 0)
839 		return err;
840 
841 	if (meta_type_ops(v)->apply_extras)
842 		meta_type_ops(v)->apply_extras(v, dst);
843 
844 	return 0;
845 }
846 
847 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
848 			 struct tcf_pkt_info *info)
849 {
850 	int r;
851 	struct meta_match *meta = (struct meta_match *) m->data;
852 	struct meta_obj l_value, r_value;
853 
854 	if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
855 	    meta_get(skb, info, &meta->rvalue, &r_value) < 0)
856 		return 0;
857 
858 	r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
859 
860 	switch (meta->lvalue.hdr.op) {
861 	case TCF_EM_OPND_EQ:
862 		return !r;
863 	case TCF_EM_OPND_LT:
864 		return r < 0;
865 	case TCF_EM_OPND_GT:
866 		return r > 0;
867 	}
868 
869 	return 0;
870 }
871 
872 static void meta_delete(struct meta_match *meta)
873 {
874 	if (meta) {
875 		const struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
876 
877 		if (ops && ops->destroy) {
878 			ops->destroy(&meta->lvalue);
879 			ops->destroy(&meta->rvalue);
880 		}
881 	}
882 
883 	kfree(meta);
884 }
885 
886 static inline int meta_change_data(struct meta_value *dst, struct nlattr *nla)
887 {
888 	if (nla) {
889 		if (nla_len(nla) == 0)
890 			return -EINVAL;
891 
892 		return meta_type_ops(dst)->change(dst, nla);
893 	}
894 
895 	return 0;
896 }
897 
898 static inline int meta_is_supported(struct meta_value *val)
899 {
900 	return !meta_id(val) || meta_ops(val)->get;
901 }
902 
903 static const struct nla_policy meta_policy[TCA_EM_META_MAX + 1] = {
904 	[TCA_EM_META_HDR]	= { .len = sizeof(struct tcf_meta_hdr) },
905 };
906 
907 static int em_meta_change(struct net *net, void *data, int len,
908 			  struct tcf_ematch *m)
909 {
910 	int err;
911 	struct nlattr *tb[TCA_EM_META_MAX + 1];
912 	struct tcf_meta_hdr *hdr;
913 	struct meta_match *meta = NULL;
914 
915 	err = nla_parse(tb, TCA_EM_META_MAX, data, len, meta_policy, NULL);
916 	if (err < 0)
917 		goto errout;
918 
919 	err = -EINVAL;
920 	if (tb[TCA_EM_META_HDR] == NULL)
921 		goto errout;
922 	hdr = nla_data(tb[TCA_EM_META_HDR]);
923 
924 	if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
925 	    TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
926 	    TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
927 	    TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
928 		goto errout;
929 
930 	meta = kzalloc(sizeof(*meta), GFP_KERNEL);
931 	if (meta == NULL) {
932 		err = -ENOMEM;
933 		goto errout;
934 	}
935 
936 	memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
937 	memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
938 
939 	if (!meta_is_supported(&meta->lvalue) ||
940 	    !meta_is_supported(&meta->rvalue)) {
941 		err = -EOPNOTSUPP;
942 		goto errout;
943 	}
944 
945 	if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE]) < 0 ||
946 	    meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE]) < 0)
947 		goto errout;
948 
949 	m->datalen = sizeof(*meta);
950 	m->data = (unsigned long) meta;
951 
952 	err = 0;
953 errout:
954 	if (err && meta)
955 		meta_delete(meta);
956 	return err;
957 }
958 
959 static void em_meta_destroy(struct tcf_ematch *m)
960 {
961 	if (m)
962 		meta_delete((struct meta_match *) m->data);
963 }
964 
965 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
966 {
967 	struct meta_match *meta = (struct meta_match *) em->data;
968 	struct tcf_meta_hdr hdr;
969 	const struct meta_type_ops *ops;
970 
971 	memset(&hdr, 0, sizeof(hdr));
972 	memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
973 	memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
974 
975 	if (nla_put(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr))
976 		goto nla_put_failure;
977 
978 	ops = meta_type_ops(&meta->lvalue);
979 	if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
980 	    ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
981 		goto nla_put_failure;
982 
983 	return 0;
984 
985 nla_put_failure:
986 	return -1;
987 }
988 
989 static struct tcf_ematch_ops em_meta_ops = {
990 	.kind	  = TCF_EM_META,
991 	.change	  = em_meta_change,
992 	.match	  = em_meta_match,
993 	.destroy  = em_meta_destroy,
994 	.dump	  = em_meta_dump,
995 	.owner	  = THIS_MODULE,
996 	.link	  = LIST_HEAD_INIT(em_meta_ops.link)
997 };
998 
999 static int __init init_em_meta(void)
1000 {
1001 	return tcf_em_register(&em_meta_ops);
1002 }
1003 
1004 static void __exit exit_em_meta(void)
1005 {
1006 	tcf_em_unregister(&em_meta_ops);
1007 }
1008 
1009 MODULE_LICENSE("GPL");
1010 
1011 module_init(init_em_meta);
1012 module_exit(exit_em_meta);
1013 
1014 MODULE_ALIAS_TCF_EMATCH(TCF_EM_META);
1015