xref: /linux/net/sched/em_meta.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
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 = skb_protocol(skb, false);
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 	int bound_dev_if;
315 
316 	if (skip_nonlocal(skb)) {
317 		*err = -1;
318 		return;
319 	}
320 
321 	bound_dev_if = READ_ONCE(skb->sk->sk_bound_dev_if);
322 	if (bound_dev_if == 0) {
323 		dst->value = (unsigned long) "any";
324 		dst->len = 3;
325 	} else {
326 		struct net_device *dev;
327 
328 		rcu_read_lock();
329 		dev = dev_get_by_index_rcu(sock_net(skb->sk),
330 					   bound_dev_if);
331 		*err = var_dev(dev, dst);
332 		rcu_read_unlock();
333 	}
334 }
335 
336 META_COLLECTOR(int_sk_refcnt)
337 {
338 	if (skip_nonlocal(skb)) {
339 		*err = -1;
340 		return;
341 	}
342 	dst->value = refcount_read(&skb->sk->sk_refcnt);
343 }
344 
345 META_COLLECTOR(int_sk_rcvbuf)
346 {
347 	const struct sock *sk = skb_to_full_sk(skb);
348 
349 	if (!sk) {
350 		*err = -1;
351 		return;
352 	}
353 	dst->value = sk->sk_rcvbuf;
354 }
355 
356 META_COLLECTOR(int_sk_shutdown)
357 {
358 	const struct sock *sk = skb_to_full_sk(skb);
359 
360 	if (!sk) {
361 		*err = -1;
362 		return;
363 	}
364 	dst->value = sk->sk_shutdown;
365 }
366 
367 META_COLLECTOR(int_sk_proto)
368 {
369 	const struct sock *sk = skb_to_full_sk(skb);
370 
371 	if (!sk) {
372 		*err = -1;
373 		return;
374 	}
375 	dst->value = sk->sk_protocol;
376 }
377 
378 META_COLLECTOR(int_sk_type)
379 {
380 	const struct sock *sk = skb_to_full_sk(skb);
381 
382 	if (!sk) {
383 		*err = -1;
384 		return;
385 	}
386 	dst->value = sk->sk_type;
387 }
388 
389 META_COLLECTOR(int_sk_rmem_alloc)
390 {
391 	const struct sock *sk = skb_to_full_sk(skb);
392 
393 	if (!sk) {
394 		*err = -1;
395 		return;
396 	}
397 	dst->value = sk_rmem_alloc_get(sk);
398 }
399 
400 META_COLLECTOR(int_sk_wmem_alloc)
401 {
402 	const struct sock *sk = skb_to_full_sk(skb);
403 
404 	if (!sk) {
405 		*err = -1;
406 		return;
407 	}
408 	dst->value = sk_wmem_alloc_get(sk);
409 }
410 
411 META_COLLECTOR(int_sk_omem_alloc)
412 {
413 	const struct sock *sk = skb_to_full_sk(skb);
414 
415 	if (!sk) {
416 		*err = -1;
417 		return;
418 	}
419 	dst->value = atomic_read(&sk->sk_omem_alloc);
420 }
421 
422 META_COLLECTOR(int_sk_rcv_qlen)
423 {
424 	const struct sock *sk = skb_to_full_sk(skb);
425 
426 	if (!sk) {
427 		*err = -1;
428 		return;
429 	}
430 	dst->value = sk->sk_receive_queue.qlen;
431 }
432 
433 META_COLLECTOR(int_sk_snd_qlen)
434 {
435 	const struct sock *sk = skb_to_full_sk(skb);
436 
437 	if (!sk) {
438 		*err = -1;
439 		return;
440 	}
441 	dst->value = sk->sk_write_queue.qlen;
442 }
443 
444 META_COLLECTOR(int_sk_wmem_queued)
445 {
446 	const struct sock *sk = skb_to_full_sk(skb);
447 
448 	if (!sk) {
449 		*err = -1;
450 		return;
451 	}
452 	dst->value = READ_ONCE(sk->sk_wmem_queued);
453 }
454 
455 META_COLLECTOR(int_sk_fwd_alloc)
456 {
457 	const struct sock *sk = skb_to_full_sk(skb);
458 
459 	if (!sk) {
460 		*err = -1;
461 		return;
462 	}
463 	dst->value = sk_forward_alloc_get(sk);
464 }
465 
466 META_COLLECTOR(int_sk_sndbuf)
467 {
468 	const struct sock *sk = skb_to_full_sk(skb);
469 
470 	if (!sk) {
471 		*err = -1;
472 		return;
473 	}
474 	dst->value = sk->sk_sndbuf;
475 }
476 
477 META_COLLECTOR(int_sk_alloc)
478 {
479 	const struct sock *sk = skb_to_full_sk(skb);
480 
481 	if (!sk) {
482 		*err = -1;
483 		return;
484 	}
485 	dst->value = (__force int) sk->sk_allocation;
486 }
487 
488 META_COLLECTOR(int_sk_hash)
489 {
490 	if (skip_nonlocal(skb)) {
491 		*err = -1;
492 		return;
493 	}
494 	dst->value = skb->sk->sk_hash;
495 }
496 
497 META_COLLECTOR(int_sk_lingertime)
498 {
499 	const struct sock *sk = skb_to_full_sk(skb);
500 
501 	if (!sk) {
502 		*err = -1;
503 		return;
504 	}
505 	dst->value = sk->sk_lingertime / HZ;
506 }
507 
508 META_COLLECTOR(int_sk_err_qlen)
509 {
510 	const struct sock *sk = skb_to_full_sk(skb);
511 
512 	if (!sk) {
513 		*err = -1;
514 		return;
515 	}
516 	dst->value = sk->sk_error_queue.qlen;
517 }
518 
519 META_COLLECTOR(int_sk_ack_bl)
520 {
521 	const struct sock *sk = skb_to_full_sk(skb);
522 
523 	if (!sk) {
524 		*err = -1;
525 		return;
526 	}
527 	dst->value = READ_ONCE(sk->sk_ack_backlog);
528 }
529 
530 META_COLLECTOR(int_sk_max_ack_bl)
531 {
532 	const struct sock *sk = skb_to_full_sk(skb);
533 
534 	if (!sk) {
535 		*err = -1;
536 		return;
537 	}
538 	dst->value = READ_ONCE(sk->sk_max_ack_backlog);
539 }
540 
541 META_COLLECTOR(int_sk_prio)
542 {
543 	const struct sock *sk = skb_to_full_sk(skb);
544 
545 	if (!sk) {
546 		*err = -1;
547 		return;
548 	}
549 	dst->value = sk->sk_priority;
550 }
551 
552 META_COLLECTOR(int_sk_rcvlowat)
553 {
554 	const struct sock *sk = skb_to_full_sk(skb);
555 
556 	if (!sk) {
557 		*err = -1;
558 		return;
559 	}
560 	dst->value = READ_ONCE(sk->sk_rcvlowat);
561 }
562 
563 META_COLLECTOR(int_sk_rcvtimeo)
564 {
565 	const struct sock *sk = skb_to_full_sk(skb);
566 
567 	if (!sk) {
568 		*err = -1;
569 		return;
570 	}
571 	dst->value = sk->sk_rcvtimeo / HZ;
572 }
573 
574 META_COLLECTOR(int_sk_sndtimeo)
575 {
576 	const struct sock *sk = skb_to_full_sk(skb);
577 
578 	if (!sk) {
579 		*err = -1;
580 		return;
581 	}
582 	dst->value = sk->sk_sndtimeo / HZ;
583 }
584 
585 META_COLLECTOR(int_sk_sendmsg_off)
586 {
587 	const struct sock *sk = skb_to_full_sk(skb);
588 
589 	if (!sk) {
590 		*err = -1;
591 		return;
592 	}
593 	dst->value = sk->sk_frag.offset;
594 }
595 
596 META_COLLECTOR(int_sk_write_pend)
597 {
598 	const struct sock *sk = skb_to_full_sk(skb);
599 
600 	if (!sk) {
601 		*err = -1;
602 		return;
603 	}
604 	dst->value = sk->sk_write_pending;
605 }
606 
607 /**************************************************************************
608  * Meta value collectors assignment table
609  **************************************************************************/
610 
611 struct meta_ops {
612 	void		(*get)(struct sk_buff *, struct tcf_pkt_info *,
613 			       struct meta_value *, struct meta_obj *, int *);
614 };
615 
616 #define META_ID(name) TCF_META_ID_##name
617 #define META_FUNC(name) { .get = meta_##name }
618 
619 /* Meta value operations table listing all meta value collectors and
620  * assigns them to a type and meta id. */
621 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX + 1][TCF_META_ID_MAX + 1] = {
622 	[TCF_META_TYPE_VAR] = {
623 		[META_ID(DEV)]			= META_FUNC(var_dev),
624 		[META_ID(SK_BOUND_IF)] 		= META_FUNC(var_sk_bound_if),
625 	},
626 	[TCF_META_TYPE_INT] = {
627 		[META_ID(RANDOM)]		= META_FUNC(int_random),
628 		[META_ID(LOADAVG_0)]		= META_FUNC(int_loadavg_0),
629 		[META_ID(LOADAVG_1)]		= META_FUNC(int_loadavg_1),
630 		[META_ID(LOADAVG_2)]		= META_FUNC(int_loadavg_2),
631 		[META_ID(DEV)]			= META_FUNC(int_dev),
632 		[META_ID(PRIORITY)]		= META_FUNC(int_priority),
633 		[META_ID(PROTOCOL)]		= META_FUNC(int_protocol),
634 		[META_ID(PKTTYPE)]		= META_FUNC(int_pkttype),
635 		[META_ID(PKTLEN)]		= META_FUNC(int_pktlen),
636 		[META_ID(DATALEN)]		= META_FUNC(int_datalen),
637 		[META_ID(MACLEN)]		= META_FUNC(int_maclen),
638 		[META_ID(NFMARK)]		= META_FUNC(int_mark),
639 		[META_ID(TCINDEX)]		= META_FUNC(int_tcindex),
640 		[META_ID(RTCLASSID)]		= META_FUNC(int_rtclassid),
641 		[META_ID(RTIIF)]		= META_FUNC(int_rtiif),
642 		[META_ID(SK_FAMILY)]		= META_FUNC(int_sk_family),
643 		[META_ID(SK_STATE)]		= META_FUNC(int_sk_state),
644 		[META_ID(SK_REUSE)]		= META_FUNC(int_sk_reuse),
645 		[META_ID(SK_BOUND_IF)]		= META_FUNC(int_sk_bound_if),
646 		[META_ID(SK_REFCNT)]		= META_FUNC(int_sk_refcnt),
647 		[META_ID(SK_RCVBUF)]		= META_FUNC(int_sk_rcvbuf),
648 		[META_ID(SK_SNDBUF)]		= META_FUNC(int_sk_sndbuf),
649 		[META_ID(SK_SHUTDOWN)]		= META_FUNC(int_sk_shutdown),
650 		[META_ID(SK_PROTO)]		= META_FUNC(int_sk_proto),
651 		[META_ID(SK_TYPE)]		= META_FUNC(int_sk_type),
652 		[META_ID(SK_RMEM_ALLOC)]	= META_FUNC(int_sk_rmem_alloc),
653 		[META_ID(SK_WMEM_ALLOC)]	= META_FUNC(int_sk_wmem_alloc),
654 		[META_ID(SK_OMEM_ALLOC)]	= META_FUNC(int_sk_omem_alloc),
655 		[META_ID(SK_WMEM_QUEUED)]	= META_FUNC(int_sk_wmem_queued),
656 		[META_ID(SK_RCV_QLEN)]		= META_FUNC(int_sk_rcv_qlen),
657 		[META_ID(SK_SND_QLEN)]		= META_FUNC(int_sk_snd_qlen),
658 		[META_ID(SK_ERR_QLEN)]		= META_FUNC(int_sk_err_qlen),
659 		[META_ID(SK_FORWARD_ALLOCS)]	= META_FUNC(int_sk_fwd_alloc),
660 		[META_ID(SK_ALLOCS)]		= META_FUNC(int_sk_alloc),
661 		[META_ID(SK_HASH)]		= META_FUNC(int_sk_hash),
662 		[META_ID(SK_LINGERTIME)]	= META_FUNC(int_sk_lingertime),
663 		[META_ID(SK_ACK_BACKLOG)]	= META_FUNC(int_sk_ack_bl),
664 		[META_ID(SK_MAX_ACK_BACKLOG)]	= META_FUNC(int_sk_max_ack_bl),
665 		[META_ID(SK_PRIO)]		= META_FUNC(int_sk_prio),
666 		[META_ID(SK_RCVLOWAT)]		= META_FUNC(int_sk_rcvlowat),
667 		[META_ID(SK_RCVTIMEO)]		= META_FUNC(int_sk_rcvtimeo),
668 		[META_ID(SK_SNDTIMEO)]		= META_FUNC(int_sk_sndtimeo),
669 		[META_ID(SK_SENDMSG_OFF)]	= META_FUNC(int_sk_sendmsg_off),
670 		[META_ID(SK_WRITE_PENDING)]	= META_FUNC(int_sk_write_pend),
671 		[META_ID(VLAN_TAG)]		= META_FUNC(int_vlan_tag),
672 		[META_ID(RXHASH)]		= META_FUNC(int_rxhash),
673 	}
674 };
675 
676 static inline struct meta_ops *meta_ops(struct meta_value *val)
677 {
678 	return &__meta_ops[meta_type(val)][meta_id(val)];
679 }
680 
681 /**************************************************************************
682  * Type specific operations for TCF_META_TYPE_VAR
683  **************************************************************************/
684 
685 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
686 {
687 	int r = a->len - b->len;
688 
689 	if (r == 0)
690 		r = memcmp((void *) a->value, (void *) b->value, a->len);
691 
692 	return r;
693 }
694 
695 static int meta_var_change(struct meta_value *dst, struct nlattr *nla)
696 {
697 	int len = nla_len(nla);
698 
699 	dst->val = (unsigned long)kmemdup(nla_data(nla), len, GFP_KERNEL);
700 	if (dst->val == 0UL)
701 		return -ENOMEM;
702 	dst->len = len;
703 	return 0;
704 }
705 
706 static void meta_var_destroy(struct meta_value *v)
707 {
708 	kfree((void *) v->val);
709 }
710 
711 static void meta_var_apply_extras(struct meta_value *v,
712 				  struct meta_obj *dst)
713 {
714 	int shift = v->hdr.shift;
715 
716 	if (shift && shift < dst->len)
717 		dst->len -= shift;
718 }
719 
720 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
721 {
722 	if (v->val && v->len &&
723 	    nla_put(skb, tlv, v->len, (void *) v->val))
724 		goto nla_put_failure;
725 	return 0;
726 
727 nla_put_failure:
728 	return -1;
729 }
730 
731 /**************************************************************************
732  * Type specific operations for TCF_META_TYPE_INT
733  **************************************************************************/
734 
735 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
736 {
737 	/* Let gcc optimize it, the unlikely is not really based on
738 	 * some numbers but jump free code for mismatches seems
739 	 * more logical. */
740 	if (unlikely(a->value == b->value))
741 		return 0;
742 	else if (a->value < b->value)
743 		return -1;
744 	else
745 		return 1;
746 }
747 
748 static int meta_int_change(struct meta_value *dst, struct nlattr *nla)
749 {
750 	if (nla_len(nla) >= sizeof(unsigned long)) {
751 		dst->val = *(unsigned long *) nla_data(nla);
752 		dst->len = sizeof(unsigned long);
753 	} else if (nla_len(nla) == sizeof(u32)) {
754 		dst->val = nla_get_u32(nla);
755 		dst->len = sizeof(u32);
756 	} else
757 		return -EINVAL;
758 
759 	return 0;
760 }
761 
762 static void meta_int_apply_extras(struct meta_value *v,
763 				  struct meta_obj *dst)
764 {
765 	if (v->hdr.shift)
766 		dst->value >>= v->hdr.shift;
767 
768 	if (v->val)
769 		dst->value &= v->val;
770 }
771 
772 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
773 {
774 	if (v->len == sizeof(unsigned long)) {
775 		if (nla_put(skb, tlv, sizeof(unsigned long), &v->val))
776 			goto nla_put_failure;
777 	} else if (v->len == sizeof(u32)) {
778 		if (nla_put_u32(skb, tlv, v->val))
779 			goto nla_put_failure;
780 	}
781 
782 	return 0;
783 
784 nla_put_failure:
785 	return -1;
786 }
787 
788 /**************************************************************************
789  * Type specific operations table
790  **************************************************************************/
791 
792 struct meta_type_ops {
793 	void	(*destroy)(struct meta_value *);
794 	int	(*compare)(struct meta_obj *, struct meta_obj *);
795 	int	(*change)(struct meta_value *, struct nlattr *);
796 	void	(*apply_extras)(struct meta_value *, struct meta_obj *);
797 	int	(*dump)(struct sk_buff *, struct meta_value *, int);
798 };
799 
800 static const struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX + 1] = {
801 	[TCF_META_TYPE_VAR] = {
802 		.destroy = meta_var_destroy,
803 		.compare = meta_var_compare,
804 		.change = meta_var_change,
805 		.apply_extras = meta_var_apply_extras,
806 		.dump = meta_var_dump
807 	},
808 	[TCF_META_TYPE_INT] = {
809 		.compare = meta_int_compare,
810 		.change = meta_int_change,
811 		.apply_extras = meta_int_apply_extras,
812 		.dump = meta_int_dump
813 	}
814 };
815 
816 static inline const struct meta_type_ops *meta_type_ops(struct meta_value *v)
817 {
818 	return &__meta_type_ops[meta_type(v)];
819 }
820 
821 /**************************************************************************
822  * Core
823  **************************************************************************/
824 
825 static int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
826 		    struct meta_value *v, struct meta_obj *dst)
827 {
828 	int err = 0;
829 
830 	if (meta_id(v) == TCF_META_ID_VALUE) {
831 		dst->value = v->val;
832 		dst->len = v->len;
833 		return 0;
834 	}
835 
836 	meta_ops(v)->get(skb, info, v, dst, &err);
837 	if (err < 0)
838 		return err;
839 
840 	if (meta_type_ops(v)->apply_extras)
841 		meta_type_ops(v)->apply_extras(v, dst);
842 
843 	return 0;
844 }
845 
846 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
847 			 struct tcf_pkt_info *info)
848 {
849 	int r;
850 	struct meta_match *meta = (struct meta_match *) m->data;
851 	struct meta_obj l_value, r_value;
852 
853 	if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
854 	    meta_get(skb, info, &meta->rvalue, &r_value) < 0)
855 		return 0;
856 
857 	r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
858 
859 	switch (meta->lvalue.hdr.op) {
860 	case TCF_EM_OPND_EQ:
861 		return !r;
862 	case TCF_EM_OPND_LT:
863 		return r < 0;
864 	case TCF_EM_OPND_GT:
865 		return r > 0;
866 	}
867 
868 	return 0;
869 }
870 
871 static void meta_delete(struct meta_match *meta)
872 {
873 	if (meta) {
874 		const struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
875 
876 		if (ops && ops->destroy) {
877 			ops->destroy(&meta->lvalue);
878 			ops->destroy(&meta->rvalue);
879 		}
880 	}
881 
882 	kfree(meta);
883 }
884 
885 static inline int meta_change_data(struct meta_value *dst, struct nlattr *nla)
886 {
887 	if (nla) {
888 		if (nla_len(nla) == 0)
889 			return -EINVAL;
890 
891 		return meta_type_ops(dst)->change(dst, nla);
892 	}
893 
894 	return 0;
895 }
896 
897 static inline int meta_is_supported(struct meta_value *val)
898 {
899 	return !meta_id(val) || meta_ops(val)->get;
900 }
901 
902 static const struct nla_policy meta_policy[TCA_EM_META_MAX + 1] = {
903 	[TCA_EM_META_HDR]	= { .len = sizeof(struct tcf_meta_hdr) },
904 };
905 
906 static int em_meta_change(struct net *net, void *data, int len,
907 			  struct tcf_ematch *m)
908 {
909 	int err;
910 	struct nlattr *tb[TCA_EM_META_MAX + 1];
911 	struct tcf_meta_hdr *hdr;
912 	struct meta_match *meta = NULL;
913 
914 	err = nla_parse_deprecated(tb, TCA_EM_META_MAX, data, len,
915 				   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