xref: /linux/net/sched/em_meta.c (revision d67b569f5f620c0fb95d5212642746b7ba9d29e4)
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: INDEV |                           | id: VALUE |
31  * 	      | data:     |                           | data: 3   |
32  * 	      +-----------+                           +-----------+
33  * 	            |                                       |
34  * 	            ---> meta_ops[INT][INDEV](...)          |
35  *	                      |                             |
36  * 	            -----------                             |
37  * 	            V                                       V
38  * 	      +-----------+                           +-----------+
39  * 	      | type: INT |                           | type: INT |
40  * 	 obj  | id: INDEV |                           | 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/config.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/tc_ematch/tc_em_meta.h>
70 #include <net/dst.h>
71 #include <net/route.h>
72 #include <net/pkt_cls.h>
73 #include <net/sock.h>
74 
75 struct meta_obj
76 {
77 	unsigned long		value;
78 	unsigned int		len;
79 };
80 
81 struct meta_value
82 {
83 	struct tcf_meta_val	hdr;
84 	unsigned long		val;
85 	unsigned int		len;
86 };
87 
88 struct meta_match
89 {
90 	struct meta_value	lvalue;
91 	struct meta_value	rvalue;
92 };
93 
94 static inline int meta_id(struct meta_value *v)
95 {
96 	return TCF_META_ID(v->hdr.kind);
97 }
98 
99 static inline int meta_type(struct meta_value *v)
100 {
101 	return TCF_META_TYPE(v->hdr.kind);
102 }
103 
104 #define META_COLLECTOR(FUNC) static void meta_##FUNC(struct sk_buff *skb, \
105 	struct tcf_pkt_info *info, struct meta_value *v, \
106 	struct meta_obj *dst, int *err)
107 
108 /**************************************************************************
109  * System status & misc
110  **************************************************************************/
111 
112 META_COLLECTOR(int_random)
113 {
114 	get_random_bytes(&dst->value, sizeof(dst->value));
115 }
116 
117 static inline unsigned long fixed_loadavg(int load)
118 {
119 	int rnd_load = load + (FIXED_1/200);
120 	int rnd_frac = ((rnd_load & (FIXED_1-1)) * 100) >> FSHIFT;
121 
122 	return ((rnd_load >> FSHIFT) * 100) + rnd_frac;
123 }
124 
125 META_COLLECTOR(int_loadavg_0)
126 {
127 	dst->value = fixed_loadavg(avenrun[0]);
128 }
129 
130 META_COLLECTOR(int_loadavg_1)
131 {
132 	dst->value = fixed_loadavg(avenrun[1]);
133 }
134 
135 META_COLLECTOR(int_loadavg_2)
136 {
137 	dst->value = fixed_loadavg(avenrun[2]);
138 }
139 
140 /**************************************************************************
141  * Device names & indices
142  **************************************************************************/
143 
144 static inline int int_dev(struct net_device *dev, struct meta_obj *dst)
145 {
146 	if (unlikely(dev == NULL))
147 		return -1;
148 
149 	dst->value = dev->ifindex;
150 	return 0;
151 }
152 
153 static inline int var_dev(struct net_device *dev, struct meta_obj *dst)
154 {
155 	if (unlikely(dev == NULL))
156 		return -1;
157 
158 	dst->value = (unsigned long) dev->name;
159 	dst->len = strlen(dev->name);
160 	return 0;
161 }
162 
163 META_COLLECTOR(int_dev)
164 {
165 	*err = int_dev(skb->dev, dst);
166 }
167 
168 META_COLLECTOR(var_dev)
169 {
170 	*err = var_dev(skb->dev, dst);
171 }
172 
173 META_COLLECTOR(int_indev)
174 {
175 	*err = int_dev(skb->input_dev, dst);
176 }
177 
178 META_COLLECTOR(var_indev)
179 {
180 	*err = var_dev(skb->input_dev, dst);
181 }
182 
183 META_COLLECTOR(int_realdev)
184 {
185 	*err = int_dev(skb->real_dev, dst);
186 }
187 
188 META_COLLECTOR(var_realdev)
189 {
190 	*err = var_dev(skb->real_dev, dst);
191 }
192 
193 /**************************************************************************
194  * skb attributes
195  **************************************************************************/
196 
197 META_COLLECTOR(int_priority)
198 {
199 	dst->value = skb->priority;
200 }
201 
202 META_COLLECTOR(int_protocol)
203 {
204 	/* Let userspace take care of the byte ordering */
205 	dst->value = skb->protocol;
206 }
207 
208 META_COLLECTOR(int_pkttype)
209 {
210 	dst->value = skb->pkt_type;
211 }
212 
213 META_COLLECTOR(int_pktlen)
214 {
215 	dst->value = skb->len;
216 }
217 
218 META_COLLECTOR(int_datalen)
219 {
220 	dst->value = skb->data_len;
221 }
222 
223 META_COLLECTOR(int_maclen)
224 {
225 	dst->value = skb->mac_len;
226 }
227 
228 /**************************************************************************
229  * Netfilter
230  **************************************************************************/
231 
232 #ifdef CONFIG_NETFILTER
233 META_COLLECTOR(int_nfmark)
234 {
235 	dst->value = skb->nfmark;
236 }
237 #endif
238 
239 /**************************************************************************
240  * Traffic Control
241  **************************************************************************/
242 
243 META_COLLECTOR(int_tcindex)
244 {
245 	dst->value = skb->tc_index;
246 }
247 
248 #ifdef CONFIG_NET_CLS_ACT
249 META_COLLECTOR(int_tcverd)
250 {
251 	dst->value = skb->tc_verd;
252 }
253 
254 META_COLLECTOR(int_tcclassid)
255 {
256 	dst->value = skb->tc_classid;
257 }
258 #endif
259 
260 /**************************************************************************
261  * Routing
262  **************************************************************************/
263 
264 #ifdef CONFIG_NET_CLS_ROUTE
265 META_COLLECTOR(int_rtclassid)
266 {
267 	if (unlikely(skb->dst == NULL))
268 		*err = -1;
269 	else
270 		dst->value = skb->dst->tclassid;
271 }
272 #endif
273 
274 META_COLLECTOR(int_rtiif)
275 {
276 	if (unlikely(skb->dst == NULL))
277 		*err = -1;
278 	else
279 		dst->value = ((struct rtable*) skb->dst)->fl.iif;
280 }
281 
282 /**************************************************************************
283  * Socket Attributes
284  **************************************************************************/
285 
286 #define SKIP_NONLOCAL(skb)			\
287 	if (unlikely(skb->sk == NULL)) {	\
288 		*err = -1;			\
289 		return;				\
290 	}
291 
292 META_COLLECTOR(int_sk_family)
293 {
294 	SKIP_NONLOCAL(skb);
295 	dst->value = skb->sk->sk_family;
296 }
297 
298 META_COLLECTOR(int_sk_state)
299 {
300 	SKIP_NONLOCAL(skb);
301 	dst->value = skb->sk->sk_state;
302 }
303 
304 META_COLLECTOR(int_sk_reuse)
305 {
306 	SKIP_NONLOCAL(skb);
307 	dst->value = skb->sk->sk_reuse;
308 }
309 
310 META_COLLECTOR(int_sk_bound_if)
311 {
312 	SKIP_NONLOCAL(skb);
313 	/* No error if bound_dev_if is 0, legal userspace check */
314 	dst->value = skb->sk->sk_bound_dev_if;
315 }
316 
317 META_COLLECTOR(var_sk_bound_if)
318 {
319 	SKIP_NONLOCAL(skb);
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 		dev = dev_get_by_index(skb->sk->sk_bound_dev_if);
328 		*err = var_dev(dev, dst);
329 		if (dev)
330 			dev_put(dev);
331 	 }
332 }
333 
334 META_COLLECTOR(int_sk_refcnt)
335 {
336 	SKIP_NONLOCAL(skb);
337 	dst->value = atomic_read(&skb->sk->sk_refcnt);
338 }
339 
340 META_COLLECTOR(int_sk_rcvbuf)
341 {
342 	SKIP_NONLOCAL(skb);
343 	dst->value = skb->sk->sk_rcvbuf;
344 }
345 
346 META_COLLECTOR(int_sk_shutdown)
347 {
348 	SKIP_NONLOCAL(skb);
349 	dst->value = skb->sk->sk_shutdown;
350 }
351 
352 META_COLLECTOR(int_sk_proto)
353 {
354 	SKIP_NONLOCAL(skb);
355 	dst->value = skb->sk->sk_protocol;
356 }
357 
358 META_COLLECTOR(int_sk_type)
359 {
360 	SKIP_NONLOCAL(skb);
361 	dst->value = skb->sk->sk_type;
362 }
363 
364 META_COLLECTOR(int_sk_rmem_alloc)
365 {
366 	SKIP_NONLOCAL(skb);
367 	dst->value = atomic_read(&skb->sk->sk_rmem_alloc);
368 }
369 
370 META_COLLECTOR(int_sk_wmem_alloc)
371 {
372 	SKIP_NONLOCAL(skb);
373 	dst->value = atomic_read(&skb->sk->sk_wmem_alloc);
374 }
375 
376 META_COLLECTOR(int_sk_omem_alloc)
377 {
378 	SKIP_NONLOCAL(skb);
379 	dst->value = atomic_read(&skb->sk->sk_omem_alloc);
380 }
381 
382 META_COLLECTOR(int_sk_rcv_qlen)
383 {
384 	SKIP_NONLOCAL(skb);
385 	dst->value = skb->sk->sk_receive_queue.qlen;
386 }
387 
388 META_COLLECTOR(int_sk_snd_qlen)
389 {
390 	SKIP_NONLOCAL(skb);
391 	dst->value = skb->sk->sk_write_queue.qlen;
392 }
393 
394 META_COLLECTOR(int_sk_wmem_queued)
395 {
396 	SKIP_NONLOCAL(skb);
397 	dst->value = skb->sk->sk_wmem_queued;
398 }
399 
400 META_COLLECTOR(int_sk_fwd_alloc)
401 {
402 	SKIP_NONLOCAL(skb);
403 	dst->value = skb->sk->sk_forward_alloc;
404 }
405 
406 META_COLLECTOR(int_sk_sndbuf)
407 {
408 	SKIP_NONLOCAL(skb);
409 	dst->value = skb->sk->sk_sndbuf;
410 }
411 
412 META_COLLECTOR(int_sk_alloc)
413 {
414 	SKIP_NONLOCAL(skb);
415 	dst->value = skb->sk->sk_allocation;
416 }
417 
418 META_COLLECTOR(int_sk_route_caps)
419 {
420 	SKIP_NONLOCAL(skb);
421 	dst->value = skb->sk->sk_route_caps;
422 }
423 
424 META_COLLECTOR(int_sk_hashent)
425 {
426 	SKIP_NONLOCAL(skb);
427 	dst->value = skb->sk->sk_hashent;
428 }
429 
430 META_COLLECTOR(int_sk_lingertime)
431 {
432 	SKIP_NONLOCAL(skb);
433 	dst->value = skb->sk->sk_lingertime / HZ;
434 }
435 
436 META_COLLECTOR(int_sk_err_qlen)
437 {
438 	SKIP_NONLOCAL(skb);
439 	dst->value = skb->sk->sk_error_queue.qlen;
440 }
441 
442 META_COLLECTOR(int_sk_ack_bl)
443 {
444 	SKIP_NONLOCAL(skb);
445 	dst->value = skb->sk->sk_ack_backlog;
446 }
447 
448 META_COLLECTOR(int_sk_max_ack_bl)
449 {
450 	SKIP_NONLOCAL(skb);
451 	dst->value = skb->sk->sk_max_ack_backlog;
452 }
453 
454 META_COLLECTOR(int_sk_prio)
455 {
456 	SKIP_NONLOCAL(skb);
457 	dst->value = skb->sk->sk_priority;
458 }
459 
460 META_COLLECTOR(int_sk_rcvlowat)
461 {
462 	SKIP_NONLOCAL(skb);
463 	dst->value = skb->sk->sk_rcvlowat;
464 }
465 
466 META_COLLECTOR(int_sk_rcvtimeo)
467 {
468 	SKIP_NONLOCAL(skb);
469 	dst->value = skb->sk->sk_rcvtimeo / HZ;
470 }
471 
472 META_COLLECTOR(int_sk_sndtimeo)
473 {
474 	SKIP_NONLOCAL(skb);
475 	dst->value = skb->sk->sk_sndtimeo / HZ;
476 }
477 
478 META_COLLECTOR(int_sk_sendmsg_off)
479 {
480 	SKIP_NONLOCAL(skb);
481 	dst->value = skb->sk->sk_sndmsg_off;
482 }
483 
484 META_COLLECTOR(int_sk_write_pend)
485 {
486 	SKIP_NONLOCAL(skb);
487 	dst->value = skb->sk->sk_write_pending;
488 }
489 
490 /**************************************************************************
491  * Meta value collectors assignment table
492  **************************************************************************/
493 
494 struct meta_ops
495 {
496 	void		(*get)(struct sk_buff *, struct tcf_pkt_info *,
497 			       struct meta_value *, struct meta_obj *, int *);
498 };
499 
500 #define META_ID(name) TCF_META_ID_##name
501 #define META_FUNC(name) { .get = meta_##name }
502 
503 /* Meta value operations table listing all meta value collectors and
504  * assigns them to a type and meta id. */
505 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX+1][TCF_META_ID_MAX+1] = {
506 	[TCF_META_TYPE_VAR] = {
507 		[META_ID(DEV)]			= META_FUNC(var_dev),
508 		[META_ID(INDEV)]		= META_FUNC(var_indev),
509 		[META_ID(REALDEV)]		= META_FUNC(var_realdev),
510 		[META_ID(SK_BOUND_IF)] 		= META_FUNC(var_sk_bound_if),
511 	},
512 	[TCF_META_TYPE_INT] = {
513 		[META_ID(RANDOM)]		= META_FUNC(int_random),
514 		[META_ID(LOADAVG_0)]		= META_FUNC(int_loadavg_0),
515 		[META_ID(LOADAVG_1)]		= META_FUNC(int_loadavg_1),
516 		[META_ID(LOADAVG_2)]		= META_FUNC(int_loadavg_2),
517 		[META_ID(DEV)]			= META_FUNC(int_dev),
518 		[META_ID(INDEV)]		= META_FUNC(int_indev),
519 		[META_ID(REALDEV)]		= META_FUNC(int_realdev),
520 		[META_ID(PRIORITY)]		= META_FUNC(int_priority),
521 		[META_ID(PROTOCOL)]		= META_FUNC(int_protocol),
522 		[META_ID(PKTTYPE)]		= META_FUNC(int_pkttype),
523 		[META_ID(PKTLEN)]		= META_FUNC(int_pktlen),
524 		[META_ID(DATALEN)]		= META_FUNC(int_datalen),
525 		[META_ID(MACLEN)]		= META_FUNC(int_maclen),
526 #ifdef CONFIG_NETFILTER
527 		[META_ID(NFMARK)]		= META_FUNC(int_nfmark),
528 #endif
529 		[META_ID(TCINDEX)]		= META_FUNC(int_tcindex),
530 #ifdef CONFIG_NET_CLS_ACT
531 		[META_ID(TCVERDICT)]		= META_FUNC(int_tcverd),
532 		[META_ID(TCCLASSID)]		= META_FUNC(int_tcclassid),
533 #endif
534 #ifdef CONFIG_NET_CLS_ROUTE
535 		[META_ID(RTCLASSID)]		= META_FUNC(int_rtclassid),
536 #endif
537 		[META_ID(RTIIF)]		= META_FUNC(int_rtiif),
538 		[META_ID(SK_FAMILY)]		= META_FUNC(int_sk_family),
539 		[META_ID(SK_STATE)]		= META_FUNC(int_sk_state),
540 		[META_ID(SK_REUSE)]		= META_FUNC(int_sk_reuse),
541 		[META_ID(SK_BOUND_IF)]		= META_FUNC(int_sk_bound_if),
542 		[META_ID(SK_REFCNT)]		= META_FUNC(int_sk_refcnt),
543 		[META_ID(SK_RCVBUF)]		= META_FUNC(int_sk_rcvbuf),
544 		[META_ID(SK_SNDBUF)]		= META_FUNC(int_sk_sndbuf),
545 		[META_ID(SK_SHUTDOWN)]		= META_FUNC(int_sk_shutdown),
546 		[META_ID(SK_PROTO)]		= META_FUNC(int_sk_proto),
547 		[META_ID(SK_TYPE)]		= META_FUNC(int_sk_type),
548 		[META_ID(SK_RMEM_ALLOC)]	= META_FUNC(int_sk_rmem_alloc),
549 		[META_ID(SK_WMEM_ALLOC)]	= META_FUNC(int_sk_wmem_alloc),
550 		[META_ID(SK_OMEM_ALLOC)]	= META_FUNC(int_sk_omem_alloc),
551 		[META_ID(SK_WMEM_QUEUED)]	= META_FUNC(int_sk_wmem_queued),
552 		[META_ID(SK_RCV_QLEN)]		= META_FUNC(int_sk_rcv_qlen),
553 		[META_ID(SK_SND_QLEN)]		= META_FUNC(int_sk_snd_qlen),
554 		[META_ID(SK_ERR_QLEN)]		= META_FUNC(int_sk_err_qlen),
555 		[META_ID(SK_FORWARD_ALLOCS)]	= META_FUNC(int_sk_fwd_alloc),
556 		[META_ID(SK_ALLOCS)]		= META_FUNC(int_sk_alloc),
557 		[META_ID(SK_ROUTE_CAPS)]	= META_FUNC(int_sk_route_caps),
558 		[META_ID(SK_HASHENT)]		= META_FUNC(int_sk_hashent),
559 		[META_ID(SK_LINGERTIME)]	= META_FUNC(int_sk_lingertime),
560 		[META_ID(SK_ACK_BACKLOG)]	= META_FUNC(int_sk_ack_bl),
561 		[META_ID(SK_MAX_ACK_BACKLOG)]	= META_FUNC(int_sk_max_ack_bl),
562 		[META_ID(SK_PRIO)]		= META_FUNC(int_sk_prio),
563 		[META_ID(SK_RCVLOWAT)]		= META_FUNC(int_sk_rcvlowat),
564 		[META_ID(SK_RCVTIMEO)]		= META_FUNC(int_sk_rcvtimeo),
565 		[META_ID(SK_SNDTIMEO)]		= META_FUNC(int_sk_sndtimeo),
566 		[META_ID(SK_SENDMSG_OFF)]	= META_FUNC(int_sk_sendmsg_off),
567 		[META_ID(SK_WRITE_PENDING)]	= META_FUNC(int_sk_write_pend),
568 	}
569 };
570 
571 static inline struct meta_ops * meta_ops(struct meta_value *val)
572 {
573 	return &__meta_ops[meta_type(val)][meta_id(val)];
574 }
575 
576 /**************************************************************************
577  * Type specific operations for TCF_META_TYPE_VAR
578  **************************************************************************/
579 
580 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
581 {
582 	int r = a->len - b->len;
583 
584 	if (r == 0)
585 		r = memcmp((void *) a->value, (void *) b->value, a->len);
586 
587 	return r;
588 }
589 
590 static int meta_var_change(struct meta_value *dst, struct rtattr *rta)
591 {
592 	int len = RTA_PAYLOAD(rta);
593 
594 	dst->val = (unsigned long) kmalloc(len, GFP_KERNEL);
595 	if (dst->val == 0UL)
596 		return -ENOMEM;
597 	memcpy((void *) dst->val, RTA_DATA(rta), len);
598 	dst->len = len;
599 	return 0;
600 }
601 
602 static void meta_var_destroy(struct meta_value *v)
603 {
604 	if (v->val)
605 		kfree((void *) v->val);
606 }
607 
608 static void meta_var_apply_extras(struct meta_value *v,
609 				  struct meta_obj *dst)
610 {
611 	int shift = v->hdr.shift;
612 
613 	if (shift && shift < dst->len)
614 		dst->len -= shift;
615 }
616 
617 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
618 {
619 	if (v->val && v->len)
620 		RTA_PUT(skb, tlv, v->len, (void *) v->val);
621 	return 0;
622 
623 rtattr_failure:
624 	return -1;
625 }
626 
627 /**************************************************************************
628  * Type specific operations for TCF_META_TYPE_INT
629  **************************************************************************/
630 
631 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
632 {
633 	/* Let gcc optimize it, the unlikely is not really based on
634 	 * some numbers but jump free code for mismatches seems
635 	 * more logical. */
636 	if (unlikely(a->value == b->value))
637 		return 0;
638 	else if (a->value < b->value)
639 		return -1;
640 	else
641 		return 1;
642 }
643 
644 static int meta_int_change(struct meta_value *dst, struct rtattr *rta)
645 {
646 	if (RTA_PAYLOAD(rta) >= sizeof(unsigned long)) {
647 		dst->val = *(unsigned long *) RTA_DATA(rta);
648 		dst->len = sizeof(unsigned long);
649 	} else if (RTA_PAYLOAD(rta) == sizeof(u32)) {
650 		dst->val = *(u32 *) RTA_DATA(rta);
651 		dst->len = sizeof(u32);
652 	} else
653 		return -EINVAL;
654 
655 	return 0;
656 }
657 
658 static void meta_int_apply_extras(struct meta_value *v,
659 				  struct meta_obj *dst)
660 {
661 	if (v->hdr.shift)
662 		dst->value >>= v->hdr.shift;
663 
664 	if (v->val)
665 		dst->value &= v->val;
666 }
667 
668 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
669 {
670 	if (v->len == sizeof(unsigned long))
671 		RTA_PUT(skb, tlv, sizeof(unsigned long), &v->val);
672 	else if (v->len == sizeof(u32)) {
673 		u32 d = v->val;
674 		RTA_PUT(skb, tlv, sizeof(d), &d);
675 	}
676 
677 	return 0;
678 
679 rtattr_failure:
680 	return -1;
681 }
682 
683 /**************************************************************************
684  * Type specific operations table
685  **************************************************************************/
686 
687 struct meta_type_ops
688 {
689 	void	(*destroy)(struct meta_value *);
690 	int	(*compare)(struct meta_obj *, struct meta_obj *);
691 	int	(*change)(struct meta_value *, struct rtattr *);
692 	void	(*apply_extras)(struct meta_value *, struct meta_obj *);
693 	int	(*dump)(struct sk_buff *, struct meta_value *, int);
694 };
695 
696 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX+1] = {
697 	[TCF_META_TYPE_VAR] = {
698 		.destroy = meta_var_destroy,
699 		.compare = meta_var_compare,
700 		.change = meta_var_change,
701 		.apply_extras = meta_var_apply_extras,
702 		.dump = meta_var_dump
703 	},
704 	[TCF_META_TYPE_INT] = {
705 		.compare = meta_int_compare,
706 		.change = meta_int_change,
707 		.apply_extras = meta_int_apply_extras,
708 		.dump = meta_int_dump
709 	}
710 };
711 
712 static inline struct meta_type_ops * meta_type_ops(struct meta_value *v)
713 {
714 	return &__meta_type_ops[meta_type(v)];
715 }
716 
717 /**************************************************************************
718  * Core
719  **************************************************************************/
720 
721 static inline int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
722 			   struct meta_value *v, struct meta_obj *dst)
723 {
724 	int err = 0;
725 
726 	if (meta_id(v) == TCF_META_ID_VALUE) {
727 		dst->value = v->val;
728 		dst->len = v->len;
729 		return 0;
730 	}
731 
732 	meta_ops(v)->get(skb, info, v, dst, &err);
733 	if (err < 0)
734 		return err;
735 
736 	if (meta_type_ops(v)->apply_extras)
737 	    meta_type_ops(v)->apply_extras(v, dst);
738 
739 	return 0;
740 }
741 
742 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
743 			 struct tcf_pkt_info *info)
744 {
745 	int r;
746 	struct meta_match *meta = (struct meta_match *) m->data;
747 	struct meta_obj l_value, r_value;
748 
749 	if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
750 	    meta_get(skb, info, &meta->rvalue, &r_value) < 0)
751 		return 0;
752 
753 	r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
754 
755 	switch (meta->lvalue.hdr.op) {
756 		case TCF_EM_OPND_EQ:
757 			return !r;
758 		case TCF_EM_OPND_LT:
759 			return r < 0;
760 		case TCF_EM_OPND_GT:
761 			return r > 0;
762 	}
763 
764 	return 0;
765 }
766 
767 static inline void meta_delete(struct meta_match *meta)
768 {
769 	struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
770 
771 	if (ops && ops->destroy) {
772 		ops->destroy(&meta->lvalue);
773 		ops->destroy(&meta->rvalue);
774 	}
775 
776 	kfree(meta);
777 }
778 
779 static inline int meta_change_data(struct meta_value *dst, struct rtattr *rta)
780 {
781 	if (rta) {
782 		if (RTA_PAYLOAD(rta) == 0)
783 			return -EINVAL;
784 
785 		return meta_type_ops(dst)->change(dst, rta);
786 	}
787 
788 	return 0;
789 }
790 
791 static inline int meta_is_supported(struct meta_value *val)
792 {
793 	return (!meta_id(val) || meta_ops(val)->get);
794 }
795 
796 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
797 			  struct tcf_ematch *m)
798 {
799 	int err = -EINVAL;
800 	struct rtattr *tb[TCA_EM_META_MAX];
801 	struct tcf_meta_hdr *hdr;
802 	struct meta_match *meta = NULL;
803 
804 	if (rtattr_parse(tb, TCA_EM_META_MAX, data, len) < 0)
805 		goto errout;
806 
807 	if (tb[TCA_EM_META_HDR-1] == NULL ||
808 	    RTA_PAYLOAD(tb[TCA_EM_META_HDR-1]) < sizeof(*hdr))
809 		goto errout;
810 	hdr = RTA_DATA(tb[TCA_EM_META_HDR-1]);
811 
812 	if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
813 	    TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
814 	    TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
815 	    TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
816 		goto errout;
817 
818 	meta = kmalloc(sizeof(*meta), GFP_KERNEL);
819 	if (meta == NULL)
820 		goto errout;
821 	memset(meta, 0, sizeof(*meta));
822 
823 	memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
824 	memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
825 
826 	if (!meta_is_supported(&meta->lvalue) ||
827 	    !meta_is_supported(&meta->rvalue)) {
828 		err = -EOPNOTSUPP;
829 		goto errout;
830 	}
831 
832 	if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE-1]) < 0 ||
833 	    meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE-1]) < 0)
834 		goto errout;
835 
836 	m->datalen = sizeof(*meta);
837 	m->data = (unsigned long) meta;
838 
839 	err = 0;
840 errout:
841 	if (err && meta)
842 		meta_delete(meta);
843 	return err;
844 }
845 
846 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
847 {
848 	if (m)
849 		meta_delete((struct meta_match *) m->data);
850 }
851 
852 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
853 {
854 	struct meta_match *meta = (struct meta_match *) em->data;
855 	struct tcf_meta_hdr hdr;
856 	struct meta_type_ops *ops;
857 
858 	memset(&hdr, 0, sizeof(hdr));
859 	memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
860 	memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
861 
862 	RTA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
863 
864 	ops = meta_type_ops(&meta->lvalue);
865 	if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
866 	    ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
867 		goto rtattr_failure;
868 
869 	return 0;
870 
871 rtattr_failure:
872 	return -1;
873 }
874 
875 static struct tcf_ematch_ops em_meta_ops = {
876 	.kind	  = TCF_EM_META,
877 	.change	  = em_meta_change,
878 	.match	  = em_meta_match,
879 	.destroy  = em_meta_destroy,
880 	.dump	  = em_meta_dump,
881 	.owner	  = THIS_MODULE,
882 	.link	  = LIST_HEAD_INIT(em_meta_ops.link)
883 };
884 
885 static int __init init_em_meta(void)
886 {
887 	return tcf_em_register(&em_meta_ops);
888 }
889 
890 static void __exit exit_em_meta(void)
891 {
892 	tcf_em_unregister(&em_meta_ops);
893 }
894 
895 MODULE_LICENSE("GPL");
896 
897 module_init(init_em_meta);
898 module_exit(exit_em_meta);
899