xref: /openbmc/linux/net/sched/em_meta.c (revision 0e17c50f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/em_meta.c	Metadata ematch
4  *
5  * Authors:	Thomas Graf <tgraf@suug.ch>
6  *
7  * ==========================================================================
8  *
9  * 	The metadata ematch compares two meta objects where each object
10  * 	represents either a meta value stored in the kernel or a static
11  * 	value provided by userspace. The objects are not provided by
12  * 	userspace itself but rather a definition providing the information
13  * 	to build them. Every object is of a certain type which must be
14  * 	equal to the object it is being compared to.
15  *
16  * 	The definition of a objects conists of the type (meta type), a
17  * 	identifier (meta id) and additional type specific information.
18  * 	The meta id is either TCF_META_TYPE_VALUE for values provided by
19  * 	userspace or a index to the meta operations table consisting of
20  * 	function pointers to type specific meta data collectors returning
21  * 	the value of the requested meta value.
22  *
23  * 	         lvalue                                   rvalue
24  * 	      +-----------+                           +-----------+
25  * 	      | type: INT |                           | type: INT |
26  * 	 def  | id: DEV   |                           | id: VALUE |
27  * 	      | data:     |                           | data: 3   |
28  * 	      +-----------+                           +-----------+
29  * 	            |                                       |
30  * 	            ---> meta_ops[INT][DEV](...)            |
31  *	                      |                             |
32  * 	            -----------                             |
33  * 	            V                                       V
34  * 	      +-----------+                           +-----------+
35  * 	      | type: INT |                           | type: INT |
36  * 	 obj  | id: DEV |                             | id: VALUE |
37  * 	      | data: 2   |<--data got filled out     | data: 3   |
38  * 	      +-----------+                           +-----------+
39  * 	            |                                         |
40  * 	            --------------> 2  equals 3 <--------------
41  *
42  * 	This is a simplified schema, the complexity varies depending
43  * 	on the meta type. Obviously, the length of the data must also
44  * 	be provided for non-numeric types.
45  *
46  * 	Additionally, type dependent modifiers such as shift operators
47  * 	or mask may be applied to extend the functionaliy. As of now,
48  * 	the variable length type supports shifting the byte string to
49  * 	the right, eating up any number of octets and thus supporting
50  * 	wildcard interface name comparisons such as "ppp%" matching
51  * 	ppp0..9.
52  *
53  * 	NOTE: Certain meta values depend on other subsystems and are
54  * 	      only available if that subsystem is enabled in the kernel.
55  */
56 
57 #include <linux/slab.h>
58 #include <linux/module.h>
59 #include <linux/types.h>
60 #include <linux/kernel.h>
61 #include <linux/sched.h>
62 #include <linux/sched/loadavg.h>
63 #include <linux/string.h>
64 #include <linux/skbuff.h>
65 #include <linux/random.h>
66 #include <linux/if_vlan.h>
67 #include <linux/tc_ematch/tc_em_meta.h>
68 #include <net/dst.h>
69 #include <net/route.h>
70 #include <net/pkt_cls.h>
71 #include <net/sock.h>
72 
73 struct meta_obj {
74 	unsigned long		value;
75 	unsigned int		len;
76 };
77 
78 struct meta_value {
79 	struct tcf_meta_val	hdr;
80 	unsigned long		val;
81 	unsigned int		len;
82 };
83 
84 struct meta_match {
85 	struct meta_value	lvalue;
86 	struct meta_value	rvalue;
87 };
88 
89 static inline int meta_id(struct meta_value *v)
90 {
91 	return TCF_META_ID(v->hdr.kind);
92 }
93 
94 static inline int meta_type(struct meta_value *v)
95 {
96 	return TCF_META_TYPE(v->hdr.kind);
97 }
98 
99 #define META_COLLECTOR(FUNC) static void meta_##FUNC(struct sk_buff *skb, \
100 	struct tcf_pkt_info *info, struct meta_value *v, \
101 	struct meta_obj *dst, int *err)
102 
103 /**************************************************************************
104  * System status & misc
105  **************************************************************************/
106 
107 META_COLLECTOR(int_random)
108 {
109 	get_random_bytes(&dst->value, sizeof(dst->value));
110 }
111 
112 static inline unsigned long fixed_loadavg(int load)
113 {
114 	int rnd_load = load + (FIXED_1/200);
115 	int rnd_frac = ((rnd_load & (FIXED_1-1)) * 100) >> FSHIFT;
116 
117 	return ((rnd_load >> FSHIFT) * 100) + rnd_frac;
118 }
119 
120 META_COLLECTOR(int_loadavg_0)
121 {
122 	dst->value = fixed_loadavg(avenrun[0]);
123 }
124 
125 META_COLLECTOR(int_loadavg_1)
126 {
127 	dst->value = fixed_loadavg(avenrun[1]);
128 }
129 
130 META_COLLECTOR(int_loadavg_2)
131 {
132 	dst->value = fixed_loadavg(avenrun[2]);
133 }
134 
135 /**************************************************************************
136  * Device names & indices
137  **************************************************************************/
138 
139 static inline int int_dev(struct net_device *dev, struct meta_obj *dst)
140 {
141 	if (unlikely(dev == NULL))
142 		return -1;
143 
144 	dst->value = dev->ifindex;
145 	return 0;
146 }
147 
148 static inline int var_dev(struct net_device *dev, struct meta_obj *dst)
149 {
150 	if (unlikely(dev == NULL))
151 		return -1;
152 
153 	dst->value = (unsigned long) dev->name;
154 	dst->len = strlen(dev->name);
155 	return 0;
156 }
157 
158 META_COLLECTOR(int_dev)
159 {
160 	*err = int_dev(skb->dev, dst);
161 }
162 
163 META_COLLECTOR(var_dev)
164 {
165 	*err = var_dev(skb->dev, dst);
166 }
167 
168 /**************************************************************************
169  * vlan tag
170  **************************************************************************/
171 
172 META_COLLECTOR(int_vlan_tag)
173 {
174 	unsigned short tag;
175 
176 	if (skb_vlan_tag_present(skb))
177 		dst->value = skb_vlan_tag_get(skb);
178 	else if (!__vlan_get_tag(skb, &tag))
179 		dst->value = tag;
180 	else
181 		*err = -1;
182 }
183 
184 
185 
186 /**************************************************************************
187  * skb attributes
188  **************************************************************************/
189 
190 META_COLLECTOR(int_priority)
191 {
192 	dst->value = skb->priority;
193 }
194 
195 META_COLLECTOR(int_protocol)
196 {
197 	/* Let userspace take care of the byte ordering */
198 	dst->value = tc_skb_protocol(skb);
199 }
200 
201 META_COLLECTOR(int_pkttype)
202 {
203 	dst->value = skb->pkt_type;
204 }
205 
206 META_COLLECTOR(int_pktlen)
207 {
208 	dst->value = skb->len;
209 }
210 
211 META_COLLECTOR(int_datalen)
212 {
213 	dst->value = skb->data_len;
214 }
215 
216 META_COLLECTOR(int_maclen)
217 {
218 	dst->value = skb->mac_len;
219 }
220 
221 META_COLLECTOR(int_rxhash)
222 {
223 	dst->value = skb_get_hash(skb);
224 }
225 
226 /**************************************************************************
227  * Netfilter
228  **************************************************************************/
229 
230 META_COLLECTOR(int_mark)
231 {
232 	dst->value = skb->mark;
233 }
234 
235 /**************************************************************************
236  * Traffic Control
237  **************************************************************************/
238 
239 META_COLLECTOR(int_tcindex)
240 {
241 	dst->value = skb->tc_index;
242 }
243 
244 /**************************************************************************
245  * Routing
246  **************************************************************************/
247 
248 META_COLLECTOR(int_rtclassid)
249 {
250 	if (unlikely(skb_dst(skb) == NULL))
251 		*err = -1;
252 	else
253 #ifdef CONFIG_IP_ROUTE_CLASSID
254 		dst->value = skb_dst(skb)->tclassid;
255 #else
256 		dst->value = 0;
257 #endif
258 }
259 
260 META_COLLECTOR(int_rtiif)
261 {
262 	if (unlikely(skb_rtable(skb) == NULL))
263 		*err = -1;
264 	else
265 		dst->value = inet_iif(skb);
266 }
267 
268 /**************************************************************************
269  * Socket Attributes
270  **************************************************************************/
271 
272 #define skip_nonlocal(skb) \
273 	(unlikely(skb->sk == NULL))
274 
275 META_COLLECTOR(int_sk_family)
276 {
277 	if (skip_nonlocal(skb)) {
278 		*err = -1;
279 		return;
280 	}
281 	dst->value = skb->sk->sk_family;
282 }
283 
284 META_COLLECTOR(int_sk_state)
285 {
286 	if (skip_nonlocal(skb)) {
287 		*err = -1;
288 		return;
289 	}
290 	dst->value = skb->sk->sk_state;
291 }
292 
293 META_COLLECTOR(int_sk_reuse)
294 {
295 	if (skip_nonlocal(skb)) {
296 		*err = -1;
297 		return;
298 	}
299 	dst->value = skb->sk->sk_reuse;
300 }
301 
302 META_COLLECTOR(int_sk_bound_if)
303 {
304 	if (skip_nonlocal(skb)) {
305 		*err = -1;
306 		return;
307 	}
308 	/* No error if bound_dev_if is 0, legal userspace check */
309 	dst->value = skb->sk->sk_bound_dev_if;
310 }
311 
312 META_COLLECTOR(var_sk_bound_if)
313 {
314 	if (skip_nonlocal(skb)) {
315 		*err = -1;
316 		return;
317 	}
318 
319 	if (skb->sk->sk_bound_dev_if == 0) {
320 		dst->value = (unsigned long) "any";
321 		dst->len = 3;
322 	} else {
323 		struct net_device *dev;
324 
325 		rcu_read_lock();
326 		dev = dev_get_by_index_rcu(sock_net(skb->sk),
327 					   skb->sk->sk_bound_dev_if);
328 		*err = var_dev(dev, dst);
329 		rcu_read_unlock();
330 	}
331 }
332 
333 META_COLLECTOR(int_sk_refcnt)
334 {
335 	if (skip_nonlocal(skb)) {
336 		*err = -1;
337 		return;
338 	}
339 	dst->value = refcount_read(&skb->sk->sk_refcnt);
340 }
341 
342 META_COLLECTOR(int_sk_rcvbuf)
343 {
344 	const struct sock *sk = skb_to_full_sk(skb);
345 
346 	if (!sk) {
347 		*err = -1;
348 		return;
349 	}
350 	dst->value = sk->sk_rcvbuf;
351 }
352 
353 META_COLLECTOR(int_sk_shutdown)
354 {
355 	const struct sock *sk = skb_to_full_sk(skb);
356 
357 	if (!sk) {
358 		*err = -1;
359 		return;
360 	}
361 	dst->value = sk->sk_shutdown;
362 }
363 
364 META_COLLECTOR(int_sk_proto)
365 {
366 	const struct sock *sk = skb_to_full_sk(skb);
367 
368 	if (!sk) {
369 		*err = -1;
370 		return;
371 	}
372 	dst->value = sk->sk_protocol;
373 }
374 
375 META_COLLECTOR(int_sk_type)
376 {
377 	const struct sock *sk = skb_to_full_sk(skb);
378 
379 	if (!sk) {
380 		*err = -1;
381 		return;
382 	}
383 	dst->value = sk->sk_type;
384 }
385 
386 META_COLLECTOR(int_sk_rmem_alloc)
387 {
388 	const struct sock *sk = skb_to_full_sk(skb);
389 
390 	if (!sk) {
391 		*err = -1;
392 		return;
393 	}
394 	dst->value = sk_rmem_alloc_get(sk);
395 }
396 
397 META_COLLECTOR(int_sk_wmem_alloc)
398 {
399 	const struct sock *sk = skb_to_full_sk(skb);
400 
401 	if (!sk) {
402 		*err = -1;
403 		return;
404 	}
405 	dst->value = sk_wmem_alloc_get(sk);
406 }
407 
408 META_COLLECTOR(int_sk_omem_alloc)
409 {
410 	const struct sock *sk = skb_to_full_sk(skb);
411 
412 	if (!sk) {
413 		*err = -1;
414 		return;
415 	}
416 	dst->value = atomic_read(&sk->sk_omem_alloc);
417 }
418 
419 META_COLLECTOR(int_sk_rcv_qlen)
420 {
421 	const struct sock *sk = skb_to_full_sk(skb);
422 
423 	if (!sk) {
424 		*err = -1;
425 		return;
426 	}
427 	dst->value = sk->sk_receive_queue.qlen;
428 }
429 
430 META_COLLECTOR(int_sk_snd_qlen)
431 {
432 	const struct sock *sk = skb_to_full_sk(skb);
433 
434 	if (!sk) {
435 		*err = -1;
436 		return;
437 	}
438 	dst->value = sk->sk_write_queue.qlen;
439 }
440 
441 META_COLLECTOR(int_sk_wmem_queued)
442 {
443 	const struct sock *sk = skb_to_full_sk(skb);
444 
445 	if (!sk) {
446 		*err = -1;
447 		return;
448 	}
449 	dst->value = READ_ONCE(sk->sk_wmem_queued);
450 }
451 
452 META_COLLECTOR(int_sk_fwd_alloc)
453 {
454 	const struct sock *sk = skb_to_full_sk(skb);
455 
456 	if (!sk) {
457 		*err = -1;
458 		return;
459 	}
460 	dst->value = sk->sk_forward_alloc;
461 }
462 
463 META_COLLECTOR(int_sk_sndbuf)
464 {
465 	const struct sock *sk = skb_to_full_sk(skb);
466 
467 	if (!sk) {
468 		*err = -1;
469 		return;
470 	}
471 	dst->value = sk->sk_sndbuf;
472 }
473 
474 META_COLLECTOR(int_sk_alloc)
475 {
476 	const struct sock *sk = skb_to_full_sk(skb);
477 
478 	if (!sk) {
479 		*err = -1;
480 		return;
481 	}
482 	dst->value = (__force int) sk->sk_allocation;
483 }
484 
485 META_COLLECTOR(int_sk_hash)
486 {
487 	if (skip_nonlocal(skb)) {
488 		*err = -1;
489 		return;
490 	}
491 	dst->value = skb->sk->sk_hash;
492 }
493 
494 META_COLLECTOR(int_sk_lingertime)
495 {
496 	const struct sock *sk = skb_to_full_sk(skb);
497 
498 	if (!sk) {
499 		*err = -1;
500 		return;
501 	}
502 	dst->value = sk->sk_lingertime / HZ;
503 }
504 
505 META_COLLECTOR(int_sk_err_qlen)
506 {
507 	const struct sock *sk = skb_to_full_sk(skb);
508 
509 	if (!sk) {
510 		*err = -1;
511 		return;
512 	}
513 	dst->value = sk->sk_error_queue.qlen;
514 }
515 
516 META_COLLECTOR(int_sk_ack_bl)
517 {
518 	const struct sock *sk = skb_to_full_sk(skb);
519 
520 	if (!sk) {
521 		*err = -1;
522 		return;
523 	}
524 	dst->value = READ_ONCE(sk->sk_ack_backlog);
525 }
526 
527 META_COLLECTOR(int_sk_max_ack_bl)
528 {
529 	const struct sock *sk = skb_to_full_sk(skb);
530 
531 	if (!sk) {
532 		*err = -1;
533 		return;
534 	}
535 	dst->value = READ_ONCE(sk->sk_max_ack_backlog);
536 }
537 
538 META_COLLECTOR(int_sk_prio)
539 {
540 	const struct sock *sk = skb_to_full_sk(skb);
541 
542 	if (!sk) {
543 		*err = -1;
544 		return;
545 	}
546 	dst->value = sk->sk_priority;
547 }
548 
549 META_COLLECTOR(int_sk_rcvlowat)
550 {
551 	const struct sock *sk = skb_to_full_sk(skb);
552 
553 	if (!sk) {
554 		*err = -1;
555 		return;
556 	}
557 	dst->value = READ_ONCE(sk->sk_rcvlowat);
558 }
559 
560 META_COLLECTOR(int_sk_rcvtimeo)
561 {
562 	const struct sock *sk = skb_to_full_sk(skb);
563 
564 	if (!sk) {
565 		*err = -1;
566 		return;
567 	}
568 	dst->value = sk->sk_rcvtimeo / HZ;
569 }
570 
571 META_COLLECTOR(int_sk_sndtimeo)
572 {
573 	const struct sock *sk = skb_to_full_sk(skb);
574 
575 	if (!sk) {
576 		*err = -1;
577 		return;
578 	}
579 	dst->value = sk->sk_sndtimeo / HZ;
580 }
581 
582 META_COLLECTOR(int_sk_sendmsg_off)
583 {
584 	const struct sock *sk = skb_to_full_sk(skb);
585 
586 	if (!sk) {
587 		*err = -1;
588 		return;
589 	}
590 	dst->value = sk->sk_frag.offset;
591 }
592 
593 META_COLLECTOR(int_sk_write_pend)
594 {
595 	const struct sock *sk = skb_to_full_sk(skb);
596 
597 	if (!sk) {
598 		*err = -1;
599 		return;
600 	}
601 	dst->value = sk->sk_write_pending;
602 }
603 
604 /**************************************************************************
605  * Meta value collectors assignment table
606  **************************************************************************/
607 
608 struct meta_ops {
609 	void		(*get)(struct sk_buff *, struct tcf_pkt_info *,
610 			       struct meta_value *, struct meta_obj *, int *);
611 };
612 
613 #define META_ID(name) TCF_META_ID_##name
614 #define META_FUNC(name) { .get = meta_##name }
615 
616 /* Meta value operations table listing all meta value collectors and
617  * assigns them to a type and meta id. */
618 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX + 1][TCF_META_ID_MAX + 1] = {
619 	[TCF_META_TYPE_VAR] = {
620 		[META_ID(DEV)]			= META_FUNC(var_dev),
621 		[META_ID(SK_BOUND_IF)] 		= META_FUNC(var_sk_bound_if),
622 	},
623 	[TCF_META_TYPE_INT] = {
624 		[META_ID(RANDOM)]		= META_FUNC(int_random),
625 		[META_ID(LOADAVG_0)]		= META_FUNC(int_loadavg_0),
626 		[META_ID(LOADAVG_1)]		= META_FUNC(int_loadavg_1),
627 		[META_ID(LOADAVG_2)]		= META_FUNC(int_loadavg_2),
628 		[META_ID(DEV)]			= META_FUNC(int_dev),
629 		[META_ID(PRIORITY)]		= META_FUNC(int_priority),
630 		[META_ID(PROTOCOL)]		= META_FUNC(int_protocol),
631 		[META_ID(PKTTYPE)]		= META_FUNC(int_pkttype),
632 		[META_ID(PKTLEN)]		= META_FUNC(int_pktlen),
633 		[META_ID(DATALEN)]		= META_FUNC(int_datalen),
634 		[META_ID(MACLEN)]		= META_FUNC(int_maclen),
635 		[META_ID(NFMARK)]		= META_FUNC(int_mark),
636 		[META_ID(TCINDEX)]		= META_FUNC(int_tcindex),
637 		[META_ID(RTCLASSID)]		= META_FUNC(int_rtclassid),
638 		[META_ID(RTIIF)]		= META_FUNC(int_rtiif),
639 		[META_ID(SK_FAMILY)]		= META_FUNC(int_sk_family),
640 		[META_ID(SK_STATE)]		= META_FUNC(int_sk_state),
641 		[META_ID(SK_REUSE)]		= META_FUNC(int_sk_reuse),
642 		[META_ID(SK_BOUND_IF)]		= META_FUNC(int_sk_bound_if),
643 		[META_ID(SK_REFCNT)]		= META_FUNC(int_sk_refcnt),
644 		[META_ID(SK_RCVBUF)]		= META_FUNC(int_sk_rcvbuf),
645 		[META_ID(SK_SNDBUF)]		= META_FUNC(int_sk_sndbuf),
646 		[META_ID(SK_SHUTDOWN)]		= META_FUNC(int_sk_shutdown),
647 		[META_ID(SK_PROTO)]		= META_FUNC(int_sk_proto),
648 		[META_ID(SK_TYPE)]		= META_FUNC(int_sk_type),
649 		[META_ID(SK_RMEM_ALLOC)]	= META_FUNC(int_sk_rmem_alloc),
650 		[META_ID(SK_WMEM_ALLOC)]	= META_FUNC(int_sk_wmem_alloc),
651 		[META_ID(SK_OMEM_ALLOC)]	= META_FUNC(int_sk_omem_alloc),
652 		[META_ID(SK_WMEM_QUEUED)]	= META_FUNC(int_sk_wmem_queued),
653 		[META_ID(SK_RCV_QLEN)]		= META_FUNC(int_sk_rcv_qlen),
654 		[META_ID(SK_SND_QLEN)]		= META_FUNC(int_sk_snd_qlen),
655 		[META_ID(SK_ERR_QLEN)]		= META_FUNC(int_sk_err_qlen),
656 		[META_ID(SK_FORWARD_ALLOCS)]	= META_FUNC(int_sk_fwd_alloc),
657 		[META_ID(SK_ALLOCS)]		= META_FUNC(int_sk_alloc),
658 		[META_ID(SK_HASH)]		= META_FUNC(int_sk_hash),
659 		[META_ID(SK_LINGERTIME)]	= META_FUNC(int_sk_lingertime),
660 		[META_ID(SK_ACK_BACKLOG)]	= META_FUNC(int_sk_ack_bl),
661 		[META_ID(SK_MAX_ACK_BACKLOG)]	= META_FUNC(int_sk_max_ack_bl),
662 		[META_ID(SK_PRIO)]		= META_FUNC(int_sk_prio),
663 		[META_ID(SK_RCVLOWAT)]		= META_FUNC(int_sk_rcvlowat),
664 		[META_ID(SK_RCVTIMEO)]		= META_FUNC(int_sk_rcvtimeo),
665 		[META_ID(SK_SNDTIMEO)]		= META_FUNC(int_sk_sndtimeo),
666 		[META_ID(SK_SENDMSG_OFF)]	= META_FUNC(int_sk_sendmsg_off),
667 		[META_ID(SK_WRITE_PENDING)]	= META_FUNC(int_sk_write_pend),
668 		[META_ID(VLAN_TAG)]		= META_FUNC(int_vlan_tag),
669 		[META_ID(RXHASH)]		= META_FUNC(int_rxhash),
670 	}
671 };
672 
673 static inline struct meta_ops *meta_ops(struct meta_value *val)
674 {
675 	return &__meta_ops[meta_type(val)][meta_id(val)];
676 }
677 
678 /**************************************************************************
679  * Type specific operations for TCF_META_TYPE_VAR
680  **************************************************************************/
681 
682 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
683 {
684 	int r = a->len - b->len;
685 
686 	if (r == 0)
687 		r = memcmp((void *) a->value, (void *) b->value, a->len);
688 
689 	return r;
690 }
691 
692 static int meta_var_change(struct meta_value *dst, struct nlattr *nla)
693 {
694 	int len = nla_len(nla);
695 
696 	dst->val = (unsigned long)kmemdup(nla_data(nla), len, GFP_KERNEL);
697 	if (dst->val == 0UL)
698 		return -ENOMEM;
699 	dst->len = len;
700 	return 0;
701 }
702 
703 static void meta_var_destroy(struct meta_value *v)
704 {
705 	kfree((void *) v->val);
706 }
707 
708 static void meta_var_apply_extras(struct meta_value *v,
709 				  struct meta_obj *dst)
710 {
711 	int shift = v->hdr.shift;
712 
713 	if (shift && shift < dst->len)
714 		dst->len -= shift;
715 }
716 
717 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
718 {
719 	if (v->val && v->len &&
720 	    nla_put(skb, tlv, v->len, (void *) v->val))
721 		goto nla_put_failure;
722 	return 0;
723 
724 nla_put_failure:
725 	return -1;
726 }
727 
728 /**************************************************************************
729  * Type specific operations for TCF_META_TYPE_INT
730  **************************************************************************/
731 
732 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
733 {
734 	/* Let gcc optimize it, the unlikely is not really based on
735 	 * some numbers but jump free code for mismatches seems
736 	 * more logical. */
737 	if (unlikely(a->value == b->value))
738 		return 0;
739 	else if (a->value < b->value)
740 		return -1;
741 	else
742 		return 1;
743 }
744 
745 static int meta_int_change(struct meta_value *dst, struct nlattr *nla)
746 {
747 	if (nla_len(nla) >= sizeof(unsigned long)) {
748 		dst->val = *(unsigned long *) nla_data(nla);
749 		dst->len = sizeof(unsigned long);
750 	} else if (nla_len(nla) == sizeof(u32)) {
751 		dst->val = nla_get_u32(nla);
752 		dst->len = sizeof(u32);
753 	} else
754 		return -EINVAL;
755 
756 	return 0;
757 }
758 
759 static void meta_int_apply_extras(struct meta_value *v,
760 				  struct meta_obj *dst)
761 {
762 	if (v->hdr.shift)
763 		dst->value >>= v->hdr.shift;
764 
765 	if (v->val)
766 		dst->value &= v->val;
767 }
768 
769 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
770 {
771 	if (v->len == sizeof(unsigned long)) {
772 		if (nla_put(skb, tlv, sizeof(unsigned long), &v->val))
773 			goto nla_put_failure;
774 	} else if (v->len == sizeof(u32)) {
775 		if (nla_put_u32(skb, tlv, v->val))
776 			goto nla_put_failure;
777 	}
778 
779 	return 0;
780 
781 nla_put_failure:
782 	return -1;
783 }
784 
785 /**************************************************************************
786  * Type specific operations table
787  **************************************************************************/
788 
789 struct meta_type_ops {
790 	void	(*destroy)(struct meta_value *);
791 	int	(*compare)(struct meta_obj *, struct meta_obj *);
792 	int	(*change)(struct meta_value *, struct nlattr *);
793 	void	(*apply_extras)(struct meta_value *, struct meta_obj *);
794 	int	(*dump)(struct sk_buff *, struct meta_value *, int);
795 };
796 
797 static const struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX + 1] = {
798 	[TCF_META_TYPE_VAR] = {
799 		.destroy = meta_var_destroy,
800 		.compare = meta_var_compare,
801 		.change = meta_var_change,
802 		.apply_extras = meta_var_apply_extras,
803 		.dump = meta_var_dump
804 	},
805 	[TCF_META_TYPE_INT] = {
806 		.compare = meta_int_compare,
807 		.change = meta_int_change,
808 		.apply_extras = meta_int_apply_extras,
809 		.dump = meta_int_dump
810 	}
811 };
812 
813 static inline const struct meta_type_ops *meta_type_ops(struct meta_value *v)
814 {
815 	return &__meta_type_ops[meta_type(v)];
816 }
817 
818 /**************************************************************************
819  * Core
820  **************************************************************************/
821 
822 static int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
823 		    struct meta_value *v, struct meta_obj *dst)
824 {
825 	int err = 0;
826 
827 	if (meta_id(v) == TCF_META_ID_VALUE) {
828 		dst->value = v->val;
829 		dst->len = v->len;
830 		return 0;
831 	}
832 
833 	meta_ops(v)->get(skb, info, v, dst, &err);
834 	if (err < 0)
835 		return err;
836 
837 	if (meta_type_ops(v)->apply_extras)
838 		meta_type_ops(v)->apply_extras(v, dst);
839 
840 	return 0;
841 }
842 
843 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
844 			 struct tcf_pkt_info *info)
845 {
846 	int r;
847 	struct meta_match *meta = (struct meta_match *) m->data;
848 	struct meta_obj l_value, r_value;
849 
850 	if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
851 	    meta_get(skb, info, &meta->rvalue, &r_value) < 0)
852 		return 0;
853 
854 	r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
855 
856 	switch (meta->lvalue.hdr.op) {
857 	case TCF_EM_OPND_EQ:
858 		return !r;
859 	case TCF_EM_OPND_LT:
860 		return r < 0;
861 	case TCF_EM_OPND_GT:
862 		return r > 0;
863 	}
864 
865 	return 0;
866 }
867 
868 static void meta_delete(struct meta_match *meta)
869 {
870 	if (meta) {
871 		const struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
872 
873 		if (ops && ops->destroy) {
874 			ops->destroy(&meta->lvalue);
875 			ops->destroy(&meta->rvalue);
876 		}
877 	}
878 
879 	kfree(meta);
880 }
881 
882 static inline int meta_change_data(struct meta_value *dst, struct nlattr *nla)
883 {
884 	if (nla) {
885 		if (nla_len(nla) == 0)
886 			return -EINVAL;
887 
888 		return meta_type_ops(dst)->change(dst, nla);
889 	}
890 
891 	return 0;
892 }
893 
894 static inline int meta_is_supported(struct meta_value *val)
895 {
896 	return !meta_id(val) || meta_ops(val)->get;
897 }
898 
899 static const struct nla_policy meta_policy[TCA_EM_META_MAX + 1] = {
900 	[TCA_EM_META_HDR]	= { .len = sizeof(struct tcf_meta_hdr) },
901 };
902 
903 static int em_meta_change(struct net *net, void *data, int len,
904 			  struct tcf_ematch *m)
905 {
906 	int err;
907 	struct nlattr *tb[TCA_EM_META_MAX + 1];
908 	struct tcf_meta_hdr *hdr;
909 	struct meta_match *meta = NULL;
910 
911 	err = nla_parse_deprecated(tb, TCA_EM_META_MAX, data, len,
912 				   meta_policy, NULL);
913 	if (err < 0)
914 		goto errout;
915 
916 	err = -EINVAL;
917 	if (tb[TCA_EM_META_HDR] == NULL)
918 		goto errout;
919 	hdr = nla_data(tb[TCA_EM_META_HDR]);
920 
921 	if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
922 	    TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
923 	    TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
924 	    TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
925 		goto errout;
926 
927 	meta = kzalloc(sizeof(*meta), GFP_KERNEL);
928 	if (meta == NULL) {
929 		err = -ENOMEM;
930 		goto errout;
931 	}
932 
933 	memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
934 	memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
935 
936 	if (!meta_is_supported(&meta->lvalue) ||
937 	    !meta_is_supported(&meta->rvalue)) {
938 		err = -EOPNOTSUPP;
939 		goto errout;
940 	}
941 
942 	if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE]) < 0 ||
943 	    meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE]) < 0)
944 		goto errout;
945 
946 	m->datalen = sizeof(*meta);
947 	m->data = (unsigned long) meta;
948 
949 	err = 0;
950 errout:
951 	if (err && meta)
952 		meta_delete(meta);
953 	return err;
954 }
955 
956 static void em_meta_destroy(struct tcf_ematch *m)
957 {
958 	if (m)
959 		meta_delete((struct meta_match *) m->data);
960 }
961 
962 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
963 {
964 	struct meta_match *meta = (struct meta_match *) em->data;
965 	struct tcf_meta_hdr hdr;
966 	const struct meta_type_ops *ops;
967 
968 	memset(&hdr, 0, sizeof(hdr));
969 	memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
970 	memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
971 
972 	if (nla_put(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr))
973 		goto nla_put_failure;
974 
975 	ops = meta_type_ops(&meta->lvalue);
976 	if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
977 	    ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
978 		goto nla_put_failure;
979 
980 	return 0;
981 
982 nla_put_failure:
983 	return -1;
984 }
985 
986 static struct tcf_ematch_ops em_meta_ops = {
987 	.kind	  = TCF_EM_META,
988 	.change	  = em_meta_change,
989 	.match	  = em_meta_match,
990 	.destroy  = em_meta_destroy,
991 	.dump	  = em_meta_dump,
992 	.owner	  = THIS_MODULE,
993 	.link	  = LIST_HEAD_INIT(em_meta_ops.link)
994 };
995 
996 static int __init init_em_meta(void)
997 {
998 	return tcf_em_register(&em_meta_ops);
999 }
1000 
1001 static void __exit exit_em_meta(void)
1002 {
1003 	tcf_em_unregister(&em_meta_ops);
1004 }
1005 
1006 MODULE_LICENSE("GPL");
1007 
1008 module_init(init_em_meta);
1009 module_exit(exit_em_meta);
1010 
1011 MODULE_ALIAS_TCF_EMATCH(TCF_EM_META);
1012