xref: /openbmc/linux/net/sched/em_meta.c (revision 93d90ad7) 
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/string.h>
67 #include <linux/skbuff.h>
68 #include <linux/random.h>
69 #include <linux/if_vlan.h>
70 #include <linux/tc_ematch/tc_em_meta.h>
71 #include <net/dst.h>
72 #include <net/route.h>
73 #include <net/pkt_cls.h>
74 #include <net/sock.h>
75 
76 struct meta_obj {
77 	unsigned long		value;
78 	unsigned int		len;
79 };
80 
81 struct meta_value {
82 	struct tcf_meta_val	hdr;
83 	unsigned long		val;
84 	unsigned int		len;
85 };
86 
87 struct meta_match {
88 	struct meta_value	lvalue;
89 	struct meta_value	rvalue;
90 };
91 
92 static inline int meta_id(struct meta_value *v)
93 {
94 	return TCF_META_ID(v->hdr.kind);
95 }
96 
97 static inline int meta_type(struct meta_value *v)
98 {
99 	return TCF_META_TYPE(v->hdr.kind);
100 }
101 
102 #define META_COLLECTOR(FUNC) static void meta_##FUNC(struct sk_buff *skb, \
103 	struct tcf_pkt_info *info, struct meta_value *v, \
104 	struct meta_obj *dst, int *err)
105 
106 /**************************************************************************
107  * System status & misc
108  **************************************************************************/
109 
110 META_COLLECTOR(int_random)
111 {
112 	get_random_bytes(&dst->value, sizeof(dst->value));
113 }
114 
115 static inline unsigned long fixed_loadavg(int load)
116 {
117 	int rnd_load = load + (FIXED_1/200);
118 	int rnd_frac = ((rnd_load & (FIXED_1-1)) * 100) >> FSHIFT;
119 
120 	return ((rnd_load >> FSHIFT) * 100) + rnd_frac;
121 }
122 
123 META_COLLECTOR(int_loadavg_0)
124 {
125 	dst->value = fixed_loadavg(avenrun[0]);
126 }
127 
128 META_COLLECTOR(int_loadavg_1)
129 {
130 	dst->value = fixed_loadavg(avenrun[1]);
131 }
132 
133 META_COLLECTOR(int_loadavg_2)
134 {
135 	dst->value = fixed_loadavg(avenrun[2]);
136 }
137 
138 /**************************************************************************
139  * Device names & indices
140  **************************************************************************/
141 
142 static inline int int_dev(struct net_device *dev, struct meta_obj *dst)
143 {
144 	if (unlikely(dev == NULL))
145 		return -1;
146 
147 	dst->value = dev->ifindex;
148 	return 0;
149 }
150 
151 static inline int var_dev(struct net_device *dev, struct meta_obj *dst)
152 {
153 	if (unlikely(dev == NULL))
154 		return -1;
155 
156 	dst->value = (unsigned long) dev->name;
157 	dst->len = strlen(dev->name);
158 	return 0;
159 }
160 
161 META_COLLECTOR(int_dev)
162 {
163 	*err = int_dev(skb->dev, dst);
164 }
165 
166 META_COLLECTOR(var_dev)
167 {
168 	*err = var_dev(skb->dev, dst);
169 }
170 
171 /**************************************************************************
172  * vlan tag
173  **************************************************************************/
174 
175 META_COLLECTOR(int_vlan_tag)
176 {
177 	unsigned short tag;
178 
179 	tag = vlan_tx_tag_get(skb);
180 	if (!tag && __vlan_get_tag(skb, &tag))
181 		*err = -1;
182 	else
183 		dst->value = tag;
184 }
185 
186 
187 
188 /**************************************************************************
189  * skb attributes
190  **************************************************************************/
191 
192 META_COLLECTOR(int_priority)
193 {
194 	dst->value = skb->priority;
195 }
196 
197 META_COLLECTOR(int_protocol)
198 {
199 	/* Let userspace take care of the byte ordering */
200 	dst->value = skb->protocol;
201 }
202 
203 META_COLLECTOR(int_pkttype)
204 {
205 	dst->value = skb->pkt_type;
206 }
207 
208 META_COLLECTOR(int_pktlen)
209 {
210 	dst->value = skb->len;
211 }
212 
213 META_COLLECTOR(int_datalen)
214 {
215 	dst->value = skb->data_len;
216 }
217 
218 META_COLLECTOR(int_maclen)
219 {
220 	dst->value = skb->mac_len;
221 }
222 
223 META_COLLECTOR(int_rxhash)
224 {
225 	dst->value = skb_get_hash(skb);
226 }
227 
228 /**************************************************************************
229  * Netfilter
230  **************************************************************************/
231 
232 META_COLLECTOR(int_mark)
233 {
234 	dst->value = skb->mark;
235 }
236 
237 /**************************************************************************
238  * Traffic Control
239  **************************************************************************/
240 
241 META_COLLECTOR(int_tcindex)
242 {
243 	dst->value = skb->tc_index;
244 }
245 
246 /**************************************************************************
247  * Routing
248  **************************************************************************/
249 
250 META_COLLECTOR(int_rtclassid)
251 {
252 	if (unlikely(skb_dst(skb) == NULL))
253 		*err = -1;
254 	else
255 #ifdef CONFIG_IP_ROUTE_CLASSID
256 		dst->value = skb_dst(skb)->tclassid;
257 #else
258 		dst->value = 0;
259 #endif
260 }
261 
262 META_COLLECTOR(int_rtiif)
263 {
264 	if (unlikely(skb_rtable(skb) == NULL))
265 		*err = -1;
266 	else
267 		dst->value = inet_iif(skb);
268 }
269 
270 /**************************************************************************
271  * Socket Attributes
272  **************************************************************************/
273 
274 #define skip_nonlocal(skb) \
275 	(unlikely(skb->sk == NULL))
276 
277 META_COLLECTOR(int_sk_family)
278 {
279 	if (skip_nonlocal(skb)) {
280 		*err = -1;
281 		return;
282 	}
283 	dst->value = skb->sk->sk_family;
284 }
285 
286 META_COLLECTOR(int_sk_state)
287 {
288 	if (skip_nonlocal(skb)) {
289 		*err = -1;
290 		return;
291 	}
292 	dst->value = skb->sk->sk_state;
293 }
294 
295 META_COLLECTOR(int_sk_reuse)
296 {
297 	if (skip_nonlocal(skb)) {
298 		*err = -1;
299 		return;
300 	}
301 	dst->value = skb->sk->sk_reuse;
302 }
303 
304 META_COLLECTOR(int_sk_bound_if)
305 {
306 	if (skip_nonlocal(skb)) {
307 		*err = -1;
308 		return;
309 	}
310 	/* No error if bound_dev_if is 0, legal userspace check */
311 	dst->value = skb->sk->sk_bound_dev_if;
312 }
313 
314 META_COLLECTOR(var_sk_bound_if)
315 {
316 	if (skip_nonlocal(skb)) {
317 		*err = -1;
318 		return;
319 	}
320 
321 	if (skb->sk->sk_bound_dev_if == 0) {
322 		dst->value = (unsigned long) "any";
323 		dst->len = 3;
324 	} else {
325 		struct net_device *dev;
326 
327 		rcu_read_lock();
328 		dev = dev_get_by_index_rcu(sock_net(skb->sk),
329 					   skb->sk->sk_bound_dev_if);
330 		*err = var_dev(dev, dst);
331 		rcu_read_unlock();
332 	}
333 }
334 
335 META_COLLECTOR(int_sk_refcnt)
336 {
337 	if (skip_nonlocal(skb)) {
338 		*err = -1;
339 		return;
340 	}
341 	dst->value = atomic_read(&skb->sk->sk_refcnt);
342 }
343 
344 META_COLLECTOR(int_sk_rcvbuf)
345 {
346 	if (skip_nonlocal(skb)) {
347 		*err = -1;
348 		return;
349 	}
350 	dst->value = skb->sk->sk_rcvbuf;
351 }
352 
353 META_COLLECTOR(int_sk_shutdown)
354 {
355 	if (skip_nonlocal(skb)) {
356 		*err = -1;
357 		return;
358 	}
359 	dst->value = skb->sk->sk_shutdown;
360 }
361 
362 META_COLLECTOR(int_sk_proto)
363 {
364 	if (skip_nonlocal(skb)) {
365 		*err = -1;
366 		return;
367 	}
368 	dst->value = skb->sk->sk_protocol;
369 }
370 
371 META_COLLECTOR(int_sk_type)
372 {
373 	if (skip_nonlocal(skb)) {
374 		*err = -1;
375 		return;
376 	}
377 	dst->value = skb->sk->sk_type;
378 }
379 
380 META_COLLECTOR(int_sk_rmem_alloc)
381 {
382 	if (skip_nonlocal(skb)) {
383 		*err = -1;
384 		return;
385 	}
386 	dst->value = sk_rmem_alloc_get(skb->sk);
387 }
388 
389 META_COLLECTOR(int_sk_wmem_alloc)
390 {
391 	if (skip_nonlocal(skb)) {
392 		*err = -1;
393 		return;
394 	}
395 	dst->value = sk_wmem_alloc_get(skb->sk);
396 }
397 
398 META_COLLECTOR(int_sk_omem_alloc)
399 {
400 	if (skip_nonlocal(skb)) {
401 		*err = -1;
402 		return;
403 	}
404 	dst->value = atomic_read(&skb->sk->sk_omem_alloc);
405 }
406 
407 META_COLLECTOR(int_sk_rcv_qlen)
408 {
409 	if (skip_nonlocal(skb)) {
410 		*err = -1;
411 		return;
412 	}
413 	dst->value = skb->sk->sk_receive_queue.qlen;
414 }
415 
416 META_COLLECTOR(int_sk_snd_qlen)
417 {
418 	if (skip_nonlocal(skb)) {
419 		*err = -1;
420 		return;
421 	}
422 	dst->value = skb->sk->sk_write_queue.qlen;
423 }
424 
425 META_COLLECTOR(int_sk_wmem_queued)
426 {
427 	if (skip_nonlocal(skb)) {
428 		*err = -1;
429 		return;
430 	}
431 	dst->value = skb->sk->sk_wmem_queued;
432 }
433 
434 META_COLLECTOR(int_sk_fwd_alloc)
435 {
436 	if (skip_nonlocal(skb)) {
437 		*err = -1;
438 		return;
439 	}
440 	dst->value = skb->sk->sk_forward_alloc;
441 }
442 
443 META_COLLECTOR(int_sk_sndbuf)
444 {
445 	if (skip_nonlocal(skb)) {
446 		*err = -1;
447 		return;
448 	}
449 	dst->value = skb->sk->sk_sndbuf;
450 }
451 
452 META_COLLECTOR(int_sk_alloc)
453 {
454 	if (skip_nonlocal(skb)) {
455 		*err = -1;
456 		return;
457 	}
458 	dst->value = (__force int) skb->sk->sk_allocation;
459 }
460 
461 META_COLLECTOR(int_sk_hash)
462 {
463 	if (skip_nonlocal(skb)) {
464 		*err = -1;
465 		return;
466 	}
467 	dst->value = skb->sk->sk_hash;
468 }
469 
470 META_COLLECTOR(int_sk_lingertime)
471 {
472 	if (skip_nonlocal(skb)) {
473 		*err = -1;
474 		return;
475 	}
476 	dst->value = skb->sk->sk_lingertime / HZ;
477 }
478 
479 META_COLLECTOR(int_sk_err_qlen)
480 {
481 	if (skip_nonlocal(skb)) {
482 		*err = -1;
483 		return;
484 	}
485 	dst->value = skb->sk->sk_error_queue.qlen;
486 }
487 
488 META_COLLECTOR(int_sk_ack_bl)
489 {
490 	if (skip_nonlocal(skb)) {
491 		*err = -1;
492 		return;
493 	}
494 	dst->value = skb->sk->sk_ack_backlog;
495 }
496 
497 META_COLLECTOR(int_sk_max_ack_bl)
498 {
499 	if (skip_nonlocal(skb)) {
500 		*err = -1;
501 		return;
502 	}
503 	dst->value = skb->sk->sk_max_ack_backlog;
504 }
505 
506 META_COLLECTOR(int_sk_prio)
507 {
508 	if (skip_nonlocal(skb)) {
509 		*err = -1;
510 		return;
511 	}
512 	dst->value = skb->sk->sk_priority;
513 }
514 
515 META_COLLECTOR(int_sk_rcvlowat)
516 {
517 	if (skip_nonlocal(skb)) {
518 		*err = -1;
519 		return;
520 	}
521 	dst->value = skb->sk->sk_rcvlowat;
522 }
523 
524 META_COLLECTOR(int_sk_rcvtimeo)
525 {
526 	if (skip_nonlocal(skb)) {
527 		*err = -1;
528 		return;
529 	}
530 	dst->value = skb->sk->sk_rcvtimeo / HZ;
531 }
532 
533 META_COLLECTOR(int_sk_sndtimeo)
534 {
535 	if (skip_nonlocal(skb)) {
536 		*err = -1;
537 		return;
538 	}
539 	dst->value = skb->sk->sk_sndtimeo / HZ;
540 }
541 
542 META_COLLECTOR(int_sk_sendmsg_off)
543 {
544 	if (skip_nonlocal(skb)) {
545 		*err = -1;
546 		return;
547 	}
548 	dst->value = skb->sk->sk_frag.offset;
549 }
550 
551 META_COLLECTOR(int_sk_write_pend)
552 {
553 	if (skip_nonlocal(skb)) {
554 		*err = -1;
555 		return;
556 	}
557 	dst->value = skb->sk->sk_write_pending;
558 }
559 
560 /**************************************************************************
561  * Meta value collectors assignment table
562  **************************************************************************/
563 
564 struct meta_ops {
565 	void		(*get)(struct sk_buff *, struct tcf_pkt_info *,
566 			       struct meta_value *, struct meta_obj *, int *);
567 };
568 
569 #define META_ID(name) TCF_META_ID_##name
570 #define META_FUNC(name) { .get = meta_##name }
571 
572 /* Meta value operations table listing all meta value collectors and
573  * assigns them to a type and meta id. */
574 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX + 1][TCF_META_ID_MAX + 1] = {
575 	[TCF_META_TYPE_VAR] = {
576 		[META_ID(DEV)]			= META_FUNC(var_dev),
577 		[META_ID(SK_BOUND_IF)] 		= META_FUNC(var_sk_bound_if),
578 	},
579 	[TCF_META_TYPE_INT] = {
580 		[META_ID(RANDOM)]		= META_FUNC(int_random),
581 		[META_ID(LOADAVG_0)]		= META_FUNC(int_loadavg_0),
582 		[META_ID(LOADAVG_1)]		= META_FUNC(int_loadavg_1),
583 		[META_ID(LOADAVG_2)]		= META_FUNC(int_loadavg_2),
584 		[META_ID(DEV)]			= META_FUNC(int_dev),
585 		[META_ID(PRIORITY)]		= META_FUNC(int_priority),
586 		[META_ID(PROTOCOL)]		= META_FUNC(int_protocol),
587 		[META_ID(PKTTYPE)]		= META_FUNC(int_pkttype),
588 		[META_ID(PKTLEN)]		= META_FUNC(int_pktlen),
589 		[META_ID(DATALEN)]		= META_FUNC(int_datalen),
590 		[META_ID(MACLEN)]		= META_FUNC(int_maclen),
591 		[META_ID(NFMARK)]		= META_FUNC(int_mark),
592 		[META_ID(TCINDEX)]		= META_FUNC(int_tcindex),
593 		[META_ID(RTCLASSID)]		= META_FUNC(int_rtclassid),
594 		[META_ID(RTIIF)]		= META_FUNC(int_rtiif),
595 		[META_ID(SK_FAMILY)]		= META_FUNC(int_sk_family),
596 		[META_ID(SK_STATE)]		= META_FUNC(int_sk_state),
597 		[META_ID(SK_REUSE)]		= META_FUNC(int_sk_reuse),
598 		[META_ID(SK_BOUND_IF)]		= META_FUNC(int_sk_bound_if),
599 		[META_ID(SK_REFCNT)]		= META_FUNC(int_sk_refcnt),
600 		[META_ID(SK_RCVBUF)]		= META_FUNC(int_sk_rcvbuf),
601 		[META_ID(SK_SNDBUF)]		= META_FUNC(int_sk_sndbuf),
602 		[META_ID(SK_SHUTDOWN)]		= META_FUNC(int_sk_shutdown),
603 		[META_ID(SK_PROTO)]		= META_FUNC(int_sk_proto),
604 		[META_ID(SK_TYPE)]		= META_FUNC(int_sk_type),
605 		[META_ID(SK_RMEM_ALLOC)]	= META_FUNC(int_sk_rmem_alloc),
606 		[META_ID(SK_WMEM_ALLOC)]	= META_FUNC(int_sk_wmem_alloc),
607 		[META_ID(SK_OMEM_ALLOC)]	= META_FUNC(int_sk_omem_alloc),
608 		[META_ID(SK_WMEM_QUEUED)]	= META_FUNC(int_sk_wmem_queued),
609 		[META_ID(SK_RCV_QLEN)]		= META_FUNC(int_sk_rcv_qlen),
610 		[META_ID(SK_SND_QLEN)]		= META_FUNC(int_sk_snd_qlen),
611 		[META_ID(SK_ERR_QLEN)]		= META_FUNC(int_sk_err_qlen),
612 		[META_ID(SK_FORWARD_ALLOCS)]	= META_FUNC(int_sk_fwd_alloc),
613 		[META_ID(SK_ALLOCS)]		= META_FUNC(int_sk_alloc),
614 		[META_ID(SK_HASH)]		= META_FUNC(int_sk_hash),
615 		[META_ID(SK_LINGERTIME)]	= META_FUNC(int_sk_lingertime),
616 		[META_ID(SK_ACK_BACKLOG)]	= META_FUNC(int_sk_ack_bl),
617 		[META_ID(SK_MAX_ACK_BACKLOG)]	= META_FUNC(int_sk_max_ack_bl),
618 		[META_ID(SK_PRIO)]		= META_FUNC(int_sk_prio),
619 		[META_ID(SK_RCVLOWAT)]		= META_FUNC(int_sk_rcvlowat),
620 		[META_ID(SK_RCVTIMEO)]		= META_FUNC(int_sk_rcvtimeo),
621 		[META_ID(SK_SNDTIMEO)]		= META_FUNC(int_sk_sndtimeo),
622 		[META_ID(SK_SENDMSG_OFF)]	= META_FUNC(int_sk_sendmsg_off),
623 		[META_ID(SK_WRITE_PENDING)]	= META_FUNC(int_sk_write_pend),
624 		[META_ID(VLAN_TAG)]		= META_FUNC(int_vlan_tag),
625 		[META_ID(RXHASH)]		= META_FUNC(int_rxhash),
626 	}
627 };
628 
629 static inline struct meta_ops *meta_ops(struct meta_value *val)
630 {
631 	return &__meta_ops[meta_type(val)][meta_id(val)];
632 }
633 
634 /**************************************************************************
635  * Type specific operations for TCF_META_TYPE_VAR
636  **************************************************************************/
637 
638 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
639 {
640 	int r = a->len - b->len;
641 
642 	if (r == 0)
643 		r = memcmp((void *) a->value, (void *) b->value, a->len);
644 
645 	return r;
646 }
647 
648 static int meta_var_change(struct meta_value *dst, struct nlattr *nla)
649 {
650 	int len = nla_len(nla);
651 
652 	dst->val = (unsigned long)kmemdup(nla_data(nla), len, GFP_KERNEL);
653 	if (dst->val == 0UL)
654 		return -ENOMEM;
655 	dst->len = len;
656 	return 0;
657 }
658 
659 static void meta_var_destroy(struct meta_value *v)
660 {
661 	kfree((void *) v->val);
662 }
663 
664 static void meta_var_apply_extras(struct meta_value *v,
665 				  struct meta_obj *dst)
666 {
667 	int shift = v->hdr.shift;
668 
669 	if (shift && shift < dst->len)
670 		dst->len -= shift;
671 }
672 
673 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
674 {
675 	if (v->val && v->len &&
676 	    nla_put(skb, tlv, v->len, (void *) v->val))
677 		goto nla_put_failure;
678 	return 0;
679 
680 nla_put_failure:
681 	return -1;
682 }
683 
684 /**************************************************************************
685  * Type specific operations for TCF_META_TYPE_INT
686  **************************************************************************/
687 
688 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
689 {
690 	/* Let gcc optimize it, the unlikely is not really based on
691 	 * some numbers but jump free code for mismatches seems
692 	 * more logical. */
693 	if (unlikely(a->value == b->value))
694 		return 0;
695 	else if (a->value < b->value)
696 		return -1;
697 	else
698 		return 1;
699 }
700 
701 static int meta_int_change(struct meta_value *dst, struct nlattr *nla)
702 {
703 	if (nla_len(nla) >= sizeof(unsigned long)) {
704 		dst->val = *(unsigned long *) nla_data(nla);
705 		dst->len = sizeof(unsigned long);
706 	} else if (nla_len(nla) == sizeof(u32)) {
707 		dst->val = nla_get_u32(nla);
708 		dst->len = sizeof(u32);
709 	} else
710 		return -EINVAL;
711 
712 	return 0;
713 }
714 
715 static void meta_int_apply_extras(struct meta_value *v,
716 				  struct meta_obj *dst)
717 {
718 	if (v->hdr.shift)
719 		dst->value >>= v->hdr.shift;
720 
721 	if (v->val)
722 		dst->value &= v->val;
723 }
724 
725 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
726 {
727 	if (v->len == sizeof(unsigned long)) {
728 		if (nla_put(skb, tlv, sizeof(unsigned long), &v->val))
729 			goto nla_put_failure;
730 	} else if (v->len == sizeof(u32)) {
731 		if (nla_put_u32(skb, tlv, v->val))
732 			goto nla_put_failure;
733 	}
734 
735 	return 0;
736 
737 nla_put_failure:
738 	return -1;
739 }
740 
741 /**************************************************************************
742  * Type specific operations table
743  **************************************************************************/
744 
745 struct meta_type_ops {
746 	void	(*destroy)(struct meta_value *);
747 	int	(*compare)(struct meta_obj *, struct meta_obj *);
748 	int	(*change)(struct meta_value *, struct nlattr *);
749 	void	(*apply_extras)(struct meta_value *, struct meta_obj *);
750 	int	(*dump)(struct sk_buff *, struct meta_value *, int);
751 };
752 
753 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX + 1] = {
754 	[TCF_META_TYPE_VAR] = {
755 		.destroy = meta_var_destroy,
756 		.compare = meta_var_compare,
757 		.change = meta_var_change,
758 		.apply_extras = meta_var_apply_extras,
759 		.dump = meta_var_dump
760 	},
761 	[TCF_META_TYPE_INT] = {
762 		.compare = meta_int_compare,
763 		.change = meta_int_change,
764 		.apply_extras = meta_int_apply_extras,
765 		.dump = meta_int_dump
766 	}
767 };
768 
769 static inline struct meta_type_ops *meta_type_ops(struct meta_value *v)
770 {
771 	return &__meta_type_ops[meta_type(v)];
772 }
773 
774 /**************************************************************************
775  * Core
776  **************************************************************************/
777 
778 static int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
779 		    struct meta_value *v, struct meta_obj *dst)
780 {
781 	int err = 0;
782 
783 	if (meta_id(v) == TCF_META_ID_VALUE) {
784 		dst->value = v->val;
785 		dst->len = v->len;
786 		return 0;
787 	}
788 
789 	meta_ops(v)->get(skb, info, v, dst, &err);
790 	if (err < 0)
791 		return err;
792 
793 	if (meta_type_ops(v)->apply_extras)
794 		meta_type_ops(v)->apply_extras(v, dst);
795 
796 	return 0;
797 }
798 
799 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
800 			 struct tcf_pkt_info *info)
801 {
802 	int r;
803 	struct meta_match *meta = (struct meta_match *) m->data;
804 	struct meta_obj l_value, r_value;
805 
806 	if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
807 	    meta_get(skb, info, &meta->rvalue, &r_value) < 0)
808 		return 0;
809 
810 	r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
811 
812 	switch (meta->lvalue.hdr.op) {
813 	case TCF_EM_OPND_EQ:
814 		return !r;
815 	case TCF_EM_OPND_LT:
816 		return r < 0;
817 	case TCF_EM_OPND_GT:
818 		return r > 0;
819 	}
820 
821 	return 0;
822 }
823 
824 static void meta_delete(struct meta_match *meta)
825 {
826 	if (meta) {
827 		struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
828 
829 		if (ops && ops->destroy) {
830 			ops->destroy(&meta->lvalue);
831 			ops->destroy(&meta->rvalue);
832 		}
833 	}
834 
835 	kfree(meta);
836 }
837 
838 static inline int meta_change_data(struct meta_value *dst, struct nlattr *nla)
839 {
840 	if (nla) {
841 		if (nla_len(nla) == 0)
842 			return -EINVAL;
843 
844 		return meta_type_ops(dst)->change(dst, nla);
845 	}
846 
847 	return 0;
848 }
849 
850 static inline int meta_is_supported(struct meta_value *val)
851 {
852 	return !meta_id(val) || meta_ops(val)->get;
853 }
854 
855 static const struct nla_policy meta_policy[TCA_EM_META_MAX + 1] = {
856 	[TCA_EM_META_HDR]	= { .len = sizeof(struct tcf_meta_hdr) },
857 };
858 
859 static int em_meta_change(struct net *net, void *data, int len,
860 			  struct tcf_ematch *m)
861 {
862 	int err;
863 	struct nlattr *tb[TCA_EM_META_MAX + 1];
864 	struct tcf_meta_hdr *hdr;
865 	struct meta_match *meta = NULL;
866 
867 	err = nla_parse(tb, TCA_EM_META_MAX, data, len, meta_policy);
868 	if (err < 0)
869 		goto errout;
870 
871 	err = -EINVAL;
872 	if (tb[TCA_EM_META_HDR] == NULL)
873 		goto errout;
874 	hdr = nla_data(tb[TCA_EM_META_HDR]);
875 
876 	if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
877 	    TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
878 	    TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
879 	    TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
880 		goto errout;
881 
882 	meta = kzalloc(sizeof(*meta), GFP_KERNEL);
883 	if (meta == NULL) {
884 		err = -ENOMEM;
885 		goto errout;
886 	}
887 
888 	memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
889 	memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
890 
891 	if (!meta_is_supported(&meta->lvalue) ||
892 	    !meta_is_supported(&meta->rvalue)) {
893 		err = -EOPNOTSUPP;
894 		goto errout;
895 	}
896 
897 	if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE]) < 0 ||
898 	    meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE]) < 0)
899 		goto errout;
900 
901 	m->datalen = sizeof(*meta);
902 	m->data = (unsigned long) meta;
903 
904 	err = 0;
905 errout:
906 	if (err && meta)
907 		meta_delete(meta);
908 	return err;
909 }
910 
911 static void em_meta_destroy(struct tcf_ematch *m)
912 {
913 	if (m)
914 		meta_delete((struct meta_match *) m->data);
915 }
916 
917 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
918 {
919 	struct meta_match *meta = (struct meta_match *) em->data;
920 	struct tcf_meta_hdr hdr;
921 	struct meta_type_ops *ops;
922 
923 	memset(&hdr, 0, sizeof(hdr));
924 	memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
925 	memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
926 
927 	if (nla_put(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr))
928 		goto nla_put_failure;
929 
930 	ops = meta_type_ops(&meta->lvalue);
931 	if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
932 	    ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
933 		goto nla_put_failure;
934 
935 	return 0;
936 
937 nla_put_failure:
938 	return -1;
939 }
940 
941 static struct tcf_ematch_ops em_meta_ops = {
942 	.kind	  = TCF_EM_META,
943 	.change	  = em_meta_change,
944 	.match	  = em_meta_match,
945 	.destroy  = em_meta_destroy,
946 	.dump	  = em_meta_dump,
947 	.owner	  = THIS_MODULE,
948 	.link	  = LIST_HEAD_INIT(em_meta_ops.link)
949 };
950 
951 static int __init init_em_meta(void)
952 {
953 	return tcf_em_register(&em_meta_ops);
954 }
955 
956 static void __exit exit_em_meta(void)
957 {
958 	tcf_em_unregister(&em_meta_ops);
959 }
960 
961 MODULE_LICENSE("GPL");
962 
963 module_init(init_em_meta);
964 module_exit(exit_em_meta);
965 
966 MODULE_ALIAS_TCF_EMATCH(TCF_EM_META);
967