xref: /linux/net/sched/em_meta.c (revision 6e8331ac6973435b1e7604c30f2ad394035b46e1)
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  * 	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/module.h>
62 #include <linux/types.h>
63 #include <linux/kernel.h>
64 #include <linux/sched.h>
65 #include <linux/string.h>
66 #include <linux/skbuff.h>
67 #include <linux/random.h>
68 #include <linux/tc_ematch/tc_em_meta.h>
69 #include <net/dst.h>
70 #include <net/route.h>
71 #include <net/pkt_cls.h>
72 #include <net/sock.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 /**************************************************************************
173  * skb attributes
174  **************************************************************************/
175 
176 META_COLLECTOR(int_priority)
177 {
178 	dst->value = skb->priority;
179 }
180 
181 META_COLLECTOR(int_protocol)
182 {
183 	/* Let userspace take care of the byte ordering */
184 	dst->value = skb->protocol;
185 }
186 
187 META_COLLECTOR(int_pkttype)
188 {
189 	dst->value = skb->pkt_type;
190 }
191 
192 META_COLLECTOR(int_pktlen)
193 {
194 	dst->value = skb->len;
195 }
196 
197 META_COLLECTOR(int_datalen)
198 {
199 	dst->value = skb->data_len;
200 }
201 
202 META_COLLECTOR(int_maclen)
203 {
204 	dst->value = skb->mac_len;
205 }
206 
207 /**************************************************************************
208  * Netfilter
209  **************************************************************************/
210 
211 META_COLLECTOR(int_nfmark)
212 {
213 #ifdef CONFIG_NETFILTER
214 	dst->value = skb->nfmark;
215 #else
216 	dst->value = 0;
217 #endif
218 }
219 
220 /**************************************************************************
221  * Traffic Control
222  **************************************************************************/
223 
224 META_COLLECTOR(int_tcindex)
225 {
226 	dst->value = skb->tc_index;
227 }
228 
229 /**************************************************************************
230  * Routing
231  **************************************************************************/
232 
233 META_COLLECTOR(int_rtclassid)
234 {
235 	if (unlikely(skb->dst == NULL))
236 		*err = -1;
237 	else
238 #ifdef CONFIG_NET_CLS_ROUTE
239 		dst->value = skb->dst->tclassid;
240 #else
241 		dst->value = 0;
242 #endif
243 }
244 
245 META_COLLECTOR(int_rtiif)
246 {
247 	if (unlikely(skb->dst == NULL))
248 		*err = -1;
249 	else
250 		dst->value = ((struct rtable*) skb->dst)->fl.iif;
251 }
252 
253 /**************************************************************************
254  * Socket Attributes
255  **************************************************************************/
256 
257 #define SKIP_NONLOCAL(skb)			\
258 	if (unlikely(skb->sk == NULL)) {	\
259 		*err = -1;			\
260 		return;				\
261 	}
262 
263 META_COLLECTOR(int_sk_family)
264 {
265 	SKIP_NONLOCAL(skb);
266 	dst->value = skb->sk->sk_family;
267 }
268 
269 META_COLLECTOR(int_sk_state)
270 {
271 	SKIP_NONLOCAL(skb);
272 	dst->value = skb->sk->sk_state;
273 }
274 
275 META_COLLECTOR(int_sk_reuse)
276 {
277 	SKIP_NONLOCAL(skb);
278 	dst->value = skb->sk->sk_reuse;
279 }
280 
281 META_COLLECTOR(int_sk_bound_if)
282 {
283 	SKIP_NONLOCAL(skb);
284 	/* No error if bound_dev_if is 0, legal userspace check */
285 	dst->value = skb->sk->sk_bound_dev_if;
286 }
287 
288 META_COLLECTOR(var_sk_bound_if)
289 {
290 	SKIP_NONLOCAL(skb);
291 
292 	 if (skb->sk->sk_bound_dev_if == 0) {
293 		dst->value = (unsigned long) "any";
294 		dst->len = 3;
295 	 } else  {
296 		struct net_device *dev;
297 
298 		dev = dev_get_by_index(skb->sk->sk_bound_dev_if);
299 		*err = var_dev(dev, dst);
300 		if (dev)
301 			dev_put(dev);
302 	 }
303 }
304 
305 META_COLLECTOR(int_sk_refcnt)
306 {
307 	SKIP_NONLOCAL(skb);
308 	dst->value = atomic_read(&skb->sk->sk_refcnt);
309 }
310 
311 META_COLLECTOR(int_sk_rcvbuf)
312 {
313 	SKIP_NONLOCAL(skb);
314 	dst->value = skb->sk->sk_rcvbuf;
315 }
316 
317 META_COLLECTOR(int_sk_shutdown)
318 {
319 	SKIP_NONLOCAL(skb);
320 	dst->value = skb->sk->sk_shutdown;
321 }
322 
323 META_COLLECTOR(int_sk_proto)
324 {
325 	SKIP_NONLOCAL(skb);
326 	dst->value = skb->sk->sk_protocol;
327 }
328 
329 META_COLLECTOR(int_sk_type)
330 {
331 	SKIP_NONLOCAL(skb);
332 	dst->value = skb->sk->sk_type;
333 }
334 
335 META_COLLECTOR(int_sk_rmem_alloc)
336 {
337 	SKIP_NONLOCAL(skb);
338 	dst->value = atomic_read(&skb->sk->sk_rmem_alloc);
339 }
340 
341 META_COLLECTOR(int_sk_wmem_alloc)
342 {
343 	SKIP_NONLOCAL(skb);
344 	dst->value = atomic_read(&skb->sk->sk_wmem_alloc);
345 }
346 
347 META_COLLECTOR(int_sk_omem_alloc)
348 {
349 	SKIP_NONLOCAL(skb);
350 	dst->value = atomic_read(&skb->sk->sk_omem_alloc);
351 }
352 
353 META_COLLECTOR(int_sk_rcv_qlen)
354 {
355 	SKIP_NONLOCAL(skb);
356 	dst->value = skb->sk->sk_receive_queue.qlen;
357 }
358 
359 META_COLLECTOR(int_sk_snd_qlen)
360 {
361 	SKIP_NONLOCAL(skb);
362 	dst->value = skb->sk->sk_write_queue.qlen;
363 }
364 
365 META_COLLECTOR(int_sk_wmem_queued)
366 {
367 	SKIP_NONLOCAL(skb);
368 	dst->value = skb->sk->sk_wmem_queued;
369 }
370 
371 META_COLLECTOR(int_sk_fwd_alloc)
372 {
373 	SKIP_NONLOCAL(skb);
374 	dst->value = skb->sk->sk_forward_alloc;
375 }
376 
377 META_COLLECTOR(int_sk_sndbuf)
378 {
379 	SKIP_NONLOCAL(skb);
380 	dst->value = skb->sk->sk_sndbuf;
381 }
382 
383 META_COLLECTOR(int_sk_alloc)
384 {
385 	SKIP_NONLOCAL(skb);
386 	dst->value = skb->sk->sk_allocation;
387 }
388 
389 META_COLLECTOR(int_sk_route_caps)
390 {
391 	SKIP_NONLOCAL(skb);
392 	dst->value = skb->sk->sk_route_caps;
393 }
394 
395 META_COLLECTOR(int_sk_hash)
396 {
397 	SKIP_NONLOCAL(skb);
398 	dst->value = skb->sk->sk_hash;
399 }
400 
401 META_COLLECTOR(int_sk_lingertime)
402 {
403 	SKIP_NONLOCAL(skb);
404 	dst->value = skb->sk->sk_lingertime / HZ;
405 }
406 
407 META_COLLECTOR(int_sk_err_qlen)
408 {
409 	SKIP_NONLOCAL(skb);
410 	dst->value = skb->sk->sk_error_queue.qlen;
411 }
412 
413 META_COLLECTOR(int_sk_ack_bl)
414 {
415 	SKIP_NONLOCAL(skb);
416 	dst->value = skb->sk->sk_ack_backlog;
417 }
418 
419 META_COLLECTOR(int_sk_max_ack_bl)
420 {
421 	SKIP_NONLOCAL(skb);
422 	dst->value = skb->sk->sk_max_ack_backlog;
423 }
424 
425 META_COLLECTOR(int_sk_prio)
426 {
427 	SKIP_NONLOCAL(skb);
428 	dst->value = skb->sk->sk_priority;
429 }
430 
431 META_COLLECTOR(int_sk_rcvlowat)
432 {
433 	SKIP_NONLOCAL(skb);
434 	dst->value = skb->sk->sk_rcvlowat;
435 }
436 
437 META_COLLECTOR(int_sk_rcvtimeo)
438 {
439 	SKIP_NONLOCAL(skb);
440 	dst->value = skb->sk->sk_rcvtimeo / HZ;
441 }
442 
443 META_COLLECTOR(int_sk_sndtimeo)
444 {
445 	SKIP_NONLOCAL(skb);
446 	dst->value = skb->sk->sk_sndtimeo / HZ;
447 }
448 
449 META_COLLECTOR(int_sk_sendmsg_off)
450 {
451 	SKIP_NONLOCAL(skb);
452 	dst->value = skb->sk->sk_sndmsg_off;
453 }
454 
455 META_COLLECTOR(int_sk_write_pend)
456 {
457 	SKIP_NONLOCAL(skb);
458 	dst->value = skb->sk->sk_write_pending;
459 }
460 
461 /**************************************************************************
462  * Meta value collectors assignment table
463  **************************************************************************/
464 
465 struct meta_ops
466 {
467 	void		(*get)(struct sk_buff *, struct tcf_pkt_info *,
468 			       struct meta_value *, struct meta_obj *, int *);
469 };
470 
471 #define META_ID(name) TCF_META_ID_##name
472 #define META_FUNC(name) { .get = meta_##name }
473 
474 /* Meta value operations table listing all meta value collectors and
475  * assigns them to a type and meta id. */
476 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX+1][TCF_META_ID_MAX+1] = {
477 	[TCF_META_TYPE_VAR] = {
478 		[META_ID(DEV)]			= META_FUNC(var_dev),
479 		[META_ID(SK_BOUND_IF)] 		= META_FUNC(var_sk_bound_if),
480 	},
481 	[TCF_META_TYPE_INT] = {
482 		[META_ID(RANDOM)]		= META_FUNC(int_random),
483 		[META_ID(LOADAVG_0)]		= META_FUNC(int_loadavg_0),
484 		[META_ID(LOADAVG_1)]		= META_FUNC(int_loadavg_1),
485 		[META_ID(LOADAVG_2)]		= META_FUNC(int_loadavg_2),
486 		[META_ID(DEV)]			= META_FUNC(int_dev),
487 		[META_ID(PRIORITY)]		= META_FUNC(int_priority),
488 		[META_ID(PROTOCOL)]		= META_FUNC(int_protocol),
489 		[META_ID(PKTTYPE)]		= META_FUNC(int_pkttype),
490 		[META_ID(PKTLEN)]		= META_FUNC(int_pktlen),
491 		[META_ID(DATALEN)]		= META_FUNC(int_datalen),
492 		[META_ID(MACLEN)]		= META_FUNC(int_maclen),
493 		[META_ID(NFMARK)]		= META_FUNC(int_nfmark),
494 		[META_ID(TCINDEX)]		= META_FUNC(int_tcindex),
495 		[META_ID(RTCLASSID)]		= META_FUNC(int_rtclassid),
496 		[META_ID(RTIIF)]		= META_FUNC(int_rtiif),
497 		[META_ID(SK_FAMILY)]		= META_FUNC(int_sk_family),
498 		[META_ID(SK_STATE)]		= META_FUNC(int_sk_state),
499 		[META_ID(SK_REUSE)]		= META_FUNC(int_sk_reuse),
500 		[META_ID(SK_BOUND_IF)]		= META_FUNC(int_sk_bound_if),
501 		[META_ID(SK_REFCNT)]		= META_FUNC(int_sk_refcnt),
502 		[META_ID(SK_RCVBUF)]		= META_FUNC(int_sk_rcvbuf),
503 		[META_ID(SK_SNDBUF)]		= META_FUNC(int_sk_sndbuf),
504 		[META_ID(SK_SHUTDOWN)]		= META_FUNC(int_sk_shutdown),
505 		[META_ID(SK_PROTO)]		= META_FUNC(int_sk_proto),
506 		[META_ID(SK_TYPE)]		= META_FUNC(int_sk_type),
507 		[META_ID(SK_RMEM_ALLOC)]	= META_FUNC(int_sk_rmem_alloc),
508 		[META_ID(SK_WMEM_ALLOC)]	= META_FUNC(int_sk_wmem_alloc),
509 		[META_ID(SK_OMEM_ALLOC)]	= META_FUNC(int_sk_omem_alloc),
510 		[META_ID(SK_WMEM_QUEUED)]	= META_FUNC(int_sk_wmem_queued),
511 		[META_ID(SK_RCV_QLEN)]		= META_FUNC(int_sk_rcv_qlen),
512 		[META_ID(SK_SND_QLEN)]		= META_FUNC(int_sk_snd_qlen),
513 		[META_ID(SK_ERR_QLEN)]		= META_FUNC(int_sk_err_qlen),
514 		[META_ID(SK_FORWARD_ALLOCS)]	= META_FUNC(int_sk_fwd_alloc),
515 		[META_ID(SK_ALLOCS)]		= META_FUNC(int_sk_alloc),
516 		[META_ID(SK_ROUTE_CAPS)]	= META_FUNC(int_sk_route_caps),
517 		[META_ID(SK_HASH)]		= META_FUNC(int_sk_hash),
518 		[META_ID(SK_LINGERTIME)]	= META_FUNC(int_sk_lingertime),
519 		[META_ID(SK_ACK_BACKLOG)]	= META_FUNC(int_sk_ack_bl),
520 		[META_ID(SK_MAX_ACK_BACKLOG)]	= META_FUNC(int_sk_max_ack_bl),
521 		[META_ID(SK_PRIO)]		= META_FUNC(int_sk_prio),
522 		[META_ID(SK_RCVLOWAT)]		= META_FUNC(int_sk_rcvlowat),
523 		[META_ID(SK_RCVTIMEO)]		= META_FUNC(int_sk_rcvtimeo),
524 		[META_ID(SK_SNDTIMEO)]		= META_FUNC(int_sk_sndtimeo),
525 		[META_ID(SK_SENDMSG_OFF)]	= META_FUNC(int_sk_sendmsg_off),
526 		[META_ID(SK_WRITE_PENDING)]	= META_FUNC(int_sk_write_pend),
527 	}
528 };
529 
530 static inline struct meta_ops * meta_ops(struct meta_value *val)
531 {
532 	return &__meta_ops[meta_type(val)][meta_id(val)];
533 }
534 
535 /**************************************************************************
536  * Type specific operations for TCF_META_TYPE_VAR
537  **************************************************************************/
538 
539 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
540 {
541 	int r = a->len - b->len;
542 
543 	if (r == 0)
544 		r = memcmp((void *) a->value, (void *) b->value, a->len);
545 
546 	return r;
547 }
548 
549 static int meta_var_change(struct meta_value *dst, struct rtattr *rta)
550 {
551 	int len = RTA_PAYLOAD(rta);
552 
553 	dst->val = (unsigned long) kmalloc(len, GFP_KERNEL);
554 	if (dst->val == 0UL)
555 		return -ENOMEM;
556 	memcpy((void *) dst->val, RTA_DATA(rta), len);
557 	dst->len = len;
558 	return 0;
559 }
560 
561 static void meta_var_destroy(struct meta_value *v)
562 {
563 	kfree((void *) v->val);
564 }
565 
566 static void meta_var_apply_extras(struct meta_value *v,
567 				  struct meta_obj *dst)
568 {
569 	int shift = v->hdr.shift;
570 
571 	if (shift && shift < dst->len)
572 		dst->len -= shift;
573 }
574 
575 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
576 {
577 	if (v->val && v->len)
578 		RTA_PUT(skb, tlv, v->len, (void *) v->val);
579 	return 0;
580 
581 rtattr_failure:
582 	return -1;
583 }
584 
585 /**************************************************************************
586  * Type specific operations for TCF_META_TYPE_INT
587  **************************************************************************/
588 
589 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
590 {
591 	/* Let gcc optimize it, the unlikely is not really based on
592 	 * some numbers but jump free code for mismatches seems
593 	 * more logical. */
594 	if (unlikely(a->value == b->value))
595 		return 0;
596 	else if (a->value < b->value)
597 		return -1;
598 	else
599 		return 1;
600 }
601 
602 static int meta_int_change(struct meta_value *dst, struct rtattr *rta)
603 {
604 	if (RTA_PAYLOAD(rta) >= sizeof(unsigned long)) {
605 		dst->val = *(unsigned long *) RTA_DATA(rta);
606 		dst->len = sizeof(unsigned long);
607 	} else if (RTA_PAYLOAD(rta) == sizeof(u32)) {
608 		dst->val = *(u32 *) RTA_DATA(rta);
609 		dst->len = sizeof(u32);
610 	} else
611 		return -EINVAL;
612 
613 	return 0;
614 }
615 
616 static void meta_int_apply_extras(struct meta_value *v,
617 				  struct meta_obj *dst)
618 {
619 	if (v->hdr.shift)
620 		dst->value >>= v->hdr.shift;
621 
622 	if (v->val)
623 		dst->value &= v->val;
624 }
625 
626 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
627 {
628 	if (v->len == sizeof(unsigned long))
629 		RTA_PUT(skb, tlv, sizeof(unsigned long), &v->val);
630 	else if (v->len == sizeof(u32)) {
631 		u32 d = v->val;
632 		RTA_PUT(skb, tlv, sizeof(d), &d);
633 	}
634 
635 	return 0;
636 
637 rtattr_failure:
638 	return -1;
639 }
640 
641 /**************************************************************************
642  * Type specific operations table
643  **************************************************************************/
644 
645 struct meta_type_ops
646 {
647 	void	(*destroy)(struct meta_value *);
648 	int	(*compare)(struct meta_obj *, struct meta_obj *);
649 	int	(*change)(struct meta_value *, struct rtattr *);
650 	void	(*apply_extras)(struct meta_value *, struct meta_obj *);
651 	int	(*dump)(struct sk_buff *, struct meta_value *, int);
652 };
653 
654 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX+1] = {
655 	[TCF_META_TYPE_VAR] = {
656 		.destroy = meta_var_destroy,
657 		.compare = meta_var_compare,
658 		.change = meta_var_change,
659 		.apply_extras = meta_var_apply_extras,
660 		.dump = meta_var_dump
661 	},
662 	[TCF_META_TYPE_INT] = {
663 		.compare = meta_int_compare,
664 		.change = meta_int_change,
665 		.apply_extras = meta_int_apply_extras,
666 		.dump = meta_int_dump
667 	}
668 };
669 
670 static inline struct meta_type_ops * meta_type_ops(struct meta_value *v)
671 {
672 	return &__meta_type_ops[meta_type(v)];
673 }
674 
675 /**************************************************************************
676  * Core
677  **************************************************************************/
678 
679 static inline int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
680 			   struct meta_value *v, struct meta_obj *dst)
681 {
682 	int err = 0;
683 
684 	if (meta_id(v) == TCF_META_ID_VALUE) {
685 		dst->value = v->val;
686 		dst->len = v->len;
687 		return 0;
688 	}
689 
690 	meta_ops(v)->get(skb, info, v, dst, &err);
691 	if (err < 0)
692 		return err;
693 
694 	if (meta_type_ops(v)->apply_extras)
695 	    meta_type_ops(v)->apply_extras(v, dst);
696 
697 	return 0;
698 }
699 
700 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
701 			 struct tcf_pkt_info *info)
702 {
703 	int r;
704 	struct meta_match *meta = (struct meta_match *) m->data;
705 	struct meta_obj l_value, r_value;
706 
707 	if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
708 	    meta_get(skb, info, &meta->rvalue, &r_value) < 0)
709 		return 0;
710 
711 	r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
712 
713 	switch (meta->lvalue.hdr.op) {
714 		case TCF_EM_OPND_EQ:
715 			return !r;
716 		case TCF_EM_OPND_LT:
717 			return r < 0;
718 		case TCF_EM_OPND_GT:
719 			return r > 0;
720 	}
721 
722 	return 0;
723 }
724 
725 static inline void meta_delete(struct meta_match *meta)
726 {
727 	struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
728 
729 	if (ops && ops->destroy) {
730 		ops->destroy(&meta->lvalue);
731 		ops->destroy(&meta->rvalue);
732 	}
733 
734 	kfree(meta);
735 }
736 
737 static inline int meta_change_data(struct meta_value *dst, struct rtattr *rta)
738 {
739 	if (rta) {
740 		if (RTA_PAYLOAD(rta) == 0)
741 			return -EINVAL;
742 
743 		return meta_type_ops(dst)->change(dst, rta);
744 	}
745 
746 	return 0;
747 }
748 
749 static inline int meta_is_supported(struct meta_value *val)
750 {
751 	return (!meta_id(val) || meta_ops(val)->get);
752 }
753 
754 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
755 			  struct tcf_ematch *m)
756 {
757 	int err = -EINVAL;
758 	struct rtattr *tb[TCA_EM_META_MAX];
759 	struct tcf_meta_hdr *hdr;
760 	struct meta_match *meta = NULL;
761 
762 	if (rtattr_parse(tb, TCA_EM_META_MAX, data, len) < 0)
763 		goto errout;
764 
765 	if (tb[TCA_EM_META_HDR-1] == NULL ||
766 	    RTA_PAYLOAD(tb[TCA_EM_META_HDR-1]) < sizeof(*hdr))
767 		goto errout;
768 	hdr = RTA_DATA(tb[TCA_EM_META_HDR-1]);
769 
770 	if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
771 	    TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
772 	    TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
773 	    TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
774 		goto errout;
775 
776 	meta = kzalloc(sizeof(*meta), GFP_KERNEL);
777 	if (meta == NULL)
778 		goto errout;
779 
780 	memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
781 	memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
782 
783 	if (!meta_is_supported(&meta->lvalue) ||
784 	    !meta_is_supported(&meta->rvalue)) {
785 		err = -EOPNOTSUPP;
786 		goto errout;
787 	}
788 
789 	if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE-1]) < 0 ||
790 	    meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE-1]) < 0)
791 		goto errout;
792 
793 	m->datalen = sizeof(*meta);
794 	m->data = (unsigned long) meta;
795 
796 	err = 0;
797 errout:
798 	if (err && meta)
799 		meta_delete(meta);
800 	return err;
801 }
802 
803 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
804 {
805 	if (m)
806 		meta_delete((struct meta_match *) m->data);
807 }
808 
809 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
810 {
811 	struct meta_match *meta = (struct meta_match *) em->data;
812 	struct tcf_meta_hdr hdr;
813 	struct meta_type_ops *ops;
814 
815 	memset(&hdr, 0, sizeof(hdr));
816 	memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
817 	memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
818 
819 	RTA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
820 
821 	ops = meta_type_ops(&meta->lvalue);
822 	if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
823 	    ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
824 		goto rtattr_failure;
825 
826 	return 0;
827 
828 rtattr_failure:
829 	return -1;
830 }
831 
832 static struct tcf_ematch_ops em_meta_ops = {
833 	.kind	  = TCF_EM_META,
834 	.change	  = em_meta_change,
835 	.match	  = em_meta_match,
836 	.destroy  = em_meta_destroy,
837 	.dump	  = em_meta_dump,
838 	.owner	  = THIS_MODULE,
839 	.link	  = LIST_HEAD_INIT(em_meta_ops.link)
840 };
841 
842 static int __init init_em_meta(void)
843 {
844 	return tcf_em_register(&em_meta_ops);
845 }
846 
847 static void __exit exit_em_meta(void)
848 {
849 	tcf_em_unregister(&em_meta_ops);
850 }
851 
852 MODULE_LICENSE("GPL");
853 
854 module_init(init_em_meta);
855 module_exit(exit_em_meta);
856