xref: /openbmc/linux/net/sched/cls_flower.c (revision 56a0eccd)
1 /*
2  * net/sched/cls_flower.c		Flower classifier
3  *
4  * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  */
11 
12 #include <linux/kernel.h>
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/rhashtable.h>
16 
17 #include <linux/if_ether.h>
18 #include <linux/in6.h>
19 #include <linux/ip.h>
20 
21 #include <net/sch_generic.h>
22 #include <net/pkt_cls.h>
23 #include <net/ip.h>
24 #include <net/flow_dissector.h>
25 
26 struct fl_flow_key {
27 	int	indev_ifindex;
28 	struct flow_dissector_key_control control;
29 	struct flow_dissector_key_basic basic;
30 	struct flow_dissector_key_eth_addrs eth;
31 	struct flow_dissector_key_addrs ipaddrs;
32 	union {
33 		struct flow_dissector_key_ipv4_addrs ipv4;
34 		struct flow_dissector_key_ipv6_addrs ipv6;
35 	};
36 	struct flow_dissector_key_ports tp;
37 } __aligned(BITS_PER_LONG / 8); /* Ensure that we can do comparisons as longs. */
38 
39 struct fl_flow_mask_range {
40 	unsigned short int start;
41 	unsigned short int end;
42 };
43 
44 struct fl_flow_mask {
45 	struct fl_flow_key key;
46 	struct fl_flow_mask_range range;
47 	struct rcu_head	rcu;
48 };
49 
50 struct cls_fl_head {
51 	struct rhashtable ht;
52 	struct fl_flow_mask mask;
53 	struct flow_dissector dissector;
54 	u32 hgen;
55 	bool mask_assigned;
56 	struct list_head filters;
57 	struct rhashtable_params ht_params;
58 	struct rcu_head rcu;
59 };
60 
61 struct cls_fl_filter {
62 	struct rhash_head ht_node;
63 	struct fl_flow_key mkey;
64 	struct tcf_exts exts;
65 	struct tcf_result res;
66 	struct fl_flow_key key;
67 	struct list_head list;
68 	u32 handle;
69 	struct rcu_head	rcu;
70 };
71 
72 static unsigned short int fl_mask_range(const struct fl_flow_mask *mask)
73 {
74 	return mask->range.end - mask->range.start;
75 }
76 
77 static void fl_mask_update_range(struct fl_flow_mask *mask)
78 {
79 	const u8 *bytes = (const u8 *) &mask->key;
80 	size_t size = sizeof(mask->key);
81 	size_t i, first = 0, last = size - 1;
82 
83 	for (i = 0; i < sizeof(mask->key); i++) {
84 		if (bytes[i]) {
85 			if (!first && i)
86 				first = i;
87 			last = i;
88 		}
89 	}
90 	mask->range.start = rounddown(first, sizeof(long));
91 	mask->range.end = roundup(last + 1, sizeof(long));
92 }
93 
94 static void *fl_key_get_start(struct fl_flow_key *key,
95 			      const struct fl_flow_mask *mask)
96 {
97 	return (u8 *) key + mask->range.start;
98 }
99 
100 static void fl_set_masked_key(struct fl_flow_key *mkey, struct fl_flow_key *key,
101 			      struct fl_flow_mask *mask)
102 {
103 	const long *lkey = fl_key_get_start(key, mask);
104 	const long *lmask = fl_key_get_start(&mask->key, mask);
105 	long *lmkey = fl_key_get_start(mkey, mask);
106 	int i;
107 
108 	for (i = 0; i < fl_mask_range(mask); i += sizeof(long))
109 		*lmkey++ = *lkey++ & *lmask++;
110 }
111 
112 static void fl_clear_masked_range(struct fl_flow_key *key,
113 				  struct fl_flow_mask *mask)
114 {
115 	memset(fl_key_get_start(key, mask), 0, fl_mask_range(mask));
116 }
117 
118 static int fl_classify(struct sk_buff *skb, const struct tcf_proto *tp,
119 		       struct tcf_result *res)
120 {
121 	struct cls_fl_head *head = rcu_dereference_bh(tp->root);
122 	struct cls_fl_filter *f;
123 	struct fl_flow_key skb_key;
124 	struct fl_flow_key skb_mkey;
125 
126 	fl_clear_masked_range(&skb_key, &head->mask);
127 	skb_key.indev_ifindex = skb->skb_iif;
128 	/* skb_flow_dissect() does not set n_proto in case an unknown protocol,
129 	 * so do it rather here.
130 	 */
131 	skb_key.basic.n_proto = skb->protocol;
132 	skb_flow_dissect(skb, &head->dissector, &skb_key, 0);
133 
134 	fl_set_masked_key(&skb_mkey, &skb_key, &head->mask);
135 
136 	f = rhashtable_lookup_fast(&head->ht,
137 				   fl_key_get_start(&skb_mkey, &head->mask),
138 				   head->ht_params);
139 	if (f) {
140 		*res = f->res;
141 		return tcf_exts_exec(skb, &f->exts, res);
142 	}
143 	return -1;
144 }
145 
146 static int fl_init(struct tcf_proto *tp)
147 {
148 	struct cls_fl_head *head;
149 
150 	head = kzalloc(sizeof(*head), GFP_KERNEL);
151 	if (!head)
152 		return -ENOBUFS;
153 
154 	INIT_LIST_HEAD_RCU(&head->filters);
155 	rcu_assign_pointer(tp->root, head);
156 
157 	return 0;
158 }
159 
160 static void fl_destroy_filter(struct rcu_head *head)
161 {
162 	struct cls_fl_filter *f = container_of(head, struct cls_fl_filter, rcu);
163 
164 	tcf_exts_destroy(&f->exts);
165 	kfree(f);
166 }
167 
168 static void fl_hw_destroy_filter(struct tcf_proto *tp, unsigned long cookie)
169 {
170 	struct net_device *dev = tp->q->dev_queue->dev;
171 	struct tc_cls_flower_offload offload = {0};
172 	struct tc_to_netdev tc;
173 
174 	if (!tc_should_offload(dev, 0))
175 		return;
176 
177 	offload.command = TC_CLSFLOWER_DESTROY;
178 	offload.cookie = cookie;
179 
180 	tc.type = TC_SETUP_CLSFLOWER;
181 	tc.cls_flower = &offload;
182 
183 	dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, tp->protocol, &tc);
184 }
185 
186 static void fl_hw_replace_filter(struct tcf_proto *tp,
187 				 struct flow_dissector *dissector,
188 				 struct fl_flow_key *mask,
189 				 struct fl_flow_key *key,
190 				 struct tcf_exts *actions,
191 				 unsigned long cookie, u32 flags)
192 {
193 	struct net_device *dev = tp->q->dev_queue->dev;
194 	struct tc_cls_flower_offload offload = {0};
195 	struct tc_to_netdev tc;
196 
197 	if (!tc_should_offload(dev, flags))
198 		return;
199 
200 	offload.command = TC_CLSFLOWER_REPLACE;
201 	offload.cookie = cookie;
202 	offload.dissector = dissector;
203 	offload.mask = mask;
204 	offload.key = key;
205 	offload.exts = actions;
206 
207 	tc.type = TC_SETUP_CLSFLOWER;
208 	tc.cls_flower = &offload;
209 
210 	dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, tp->protocol, &tc);
211 }
212 
213 static bool fl_destroy(struct tcf_proto *tp, bool force)
214 {
215 	struct cls_fl_head *head = rtnl_dereference(tp->root);
216 	struct cls_fl_filter *f, *next;
217 
218 	if (!force && !list_empty(&head->filters))
219 		return false;
220 
221 	list_for_each_entry_safe(f, next, &head->filters, list) {
222 		fl_hw_destroy_filter(tp, (unsigned long)f);
223 		list_del_rcu(&f->list);
224 		call_rcu(&f->rcu, fl_destroy_filter);
225 	}
226 	RCU_INIT_POINTER(tp->root, NULL);
227 	if (head->mask_assigned)
228 		rhashtable_destroy(&head->ht);
229 	kfree_rcu(head, rcu);
230 	return true;
231 }
232 
233 static unsigned long fl_get(struct tcf_proto *tp, u32 handle)
234 {
235 	struct cls_fl_head *head = rtnl_dereference(tp->root);
236 	struct cls_fl_filter *f;
237 
238 	list_for_each_entry(f, &head->filters, list)
239 		if (f->handle == handle)
240 			return (unsigned long) f;
241 	return 0;
242 }
243 
244 static const struct nla_policy fl_policy[TCA_FLOWER_MAX + 1] = {
245 	[TCA_FLOWER_UNSPEC]		= { .type = NLA_UNSPEC },
246 	[TCA_FLOWER_CLASSID]		= { .type = NLA_U32 },
247 	[TCA_FLOWER_INDEV]		= { .type = NLA_STRING,
248 					    .len = IFNAMSIZ },
249 	[TCA_FLOWER_KEY_ETH_DST]	= { .len = ETH_ALEN },
250 	[TCA_FLOWER_KEY_ETH_DST_MASK]	= { .len = ETH_ALEN },
251 	[TCA_FLOWER_KEY_ETH_SRC]	= { .len = ETH_ALEN },
252 	[TCA_FLOWER_KEY_ETH_SRC_MASK]	= { .len = ETH_ALEN },
253 	[TCA_FLOWER_KEY_ETH_TYPE]	= { .type = NLA_U16 },
254 	[TCA_FLOWER_KEY_IP_PROTO]	= { .type = NLA_U8 },
255 	[TCA_FLOWER_KEY_IPV4_SRC]	= { .type = NLA_U32 },
256 	[TCA_FLOWER_KEY_IPV4_SRC_MASK]	= { .type = NLA_U32 },
257 	[TCA_FLOWER_KEY_IPV4_DST]	= { .type = NLA_U32 },
258 	[TCA_FLOWER_KEY_IPV4_DST_MASK]	= { .type = NLA_U32 },
259 	[TCA_FLOWER_KEY_IPV6_SRC]	= { .len = sizeof(struct in6_addr) },
260 	[TCA_FLOWER_KEY_IPV6_SRC_MASK]	= { .len = sizeof(struct in6_addr) },
261 	[TCA_FLOWER_KEY_IPV6_DST]	= { .len = sizeof(struct in6_addr) },
262 	[TCA_FLOWER_KEY_IPV6_DST_MASK]	= { .len = sizeof(struct in6_addr) },
263 	[TCA_FLOWER_KEY_TCP_SRC]	= { .type = NLA_U16 },
264 	[TCA_FLOWER_KEY_TCP_DST]	= { .type = NLA_U16 },
265 	[TCA_FLOWER_KEY_UDP_SRC]	= { .type = NLA_U16 },
266 	[TCA_FLOWER_KEY_UDP_DST]	= { .type = NLA_U16 },
267 };
268 
269 static void fl_set_key_val(struct nlattr **tb,
270 			   void *val, int val_type,
271 			   void *mask, int mask_type, int len)
272 {
273 	if (!tb[val_type])
274 		return;
275 	memcpy(val, nla_data(tb[val_type]), len);
276 	if (mask_type == TCA_FLOWER_UNSPEC || !tb[mask_type])
277 		memset(mask, 0xff, len);
278 	else
279 		memcpy(mask, nla_data(tb[mask_type]), len);
280 }
281 
282 static int fl_set_key(struct net *net, struct nlattr **tb,
283 		      struct fl_flow_key *key, struct fl_flow_key *mask)
284 {
285 #ifdef CONFIG_NET_CLS_IND
286 	if (tb[TCA_FLOWER_INDEV]) {
287 		int err = tcf_change_indev(net, tb[TCA_FLOWER_INDEV]);
288 		if (err < 0)
289 			return err;
290 		key->indev_ifindex = err;
291 		mask->indev_ifindex = 0xffffffff;
292 	}
293 #endif
294 
295 	fl_set_key_val(tb, key->eth.dst, TCA_FLOWER_KEY_ETH_DST,
296 		       mask->eth.dst, TCA_FLOWER_KEY_ETH_DST_MASK,
297 		       sizeof(key->eth.dst));
298 	fl_set_key_val(tb, key->eth.src, TCA_FLOWER_KEY_ETH_SRC,
299 		       mask->eth.src, TCA_FLOWER_KEY_ETH_SRC_MASK,
300 		       sizeof(key->eth.src));
301 
302 	fl_set_key_val(tb, &key->basic.n_proto, TCA_FLOWER_KEY_ETH_TYPE,
303 		       &mask->basic.n_proto, TCA_FLOWER_UNSPEC,
304 		       sizeof(key->basic.n_proto));
305 
306 	if (key->basic.n_proto == htons(ETH_P_IP) ||
307 	    key->basic.n_proto == htons(ETH_P_IPV6)) {
308 		fl_set_key_val(tb, &key->basic.ip_proto, TCA_FLOWER_KEY_IP_PROTO,
309 			       &mask->basic.ip_proto, TCA_FLOWER_UNSPEC,
310 			       sizeof(key->basic.ip_proto));
311 	}
312 
313 	if (tb[TCA_FLOWER_KEY_IPV4_SRC] || tb[TCA_FLOWER_KEY_IPV4_DST]) {
314 		key->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
315 		fl_set_key_val(tb, &key->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC,
316 			       &mask->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC_MASK,
317 			       sizeof(key->ipv4.src));
318 		fl_set_key_val(tb, &key->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST,
319 			       &mask->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST_MASK,
320 			       sizeof(key->ipv4.dst));
321 	} else if (tb[TCA_FLOWER_KEY_IPV6_SRC] || tb[TCA_FLOWER_KEY_IPV6_DST]) {
322 		key->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
323 		fl_set_key_val(tb, &key->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC,
324 			       &mask->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC_MASK,
325 			       sizeof(key->ipv6.src));
326 		fl_set_key_val(tb, &key->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST,
327 			       &mask->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST_MASK,
328 			       sizeof(key->ipv6.dst));
329 	}
330 
331 	if (key->basic.ip_proto == IPPROTO_TCP) {
332 		fl_set_key_val(tb, &key->tp.src, TCA_FLOWER_KEY_TCP_SRC,
333 			       &mask->tp.src, TCA_FLOWER_UNSPEC,
334 			       sizeof(key->tp.src));
335 		fl_set_key_val(tb, &key->tp.dst, TCA_FLOWER_KEY_TCP_DST,
336 			       &mask->tp.dst, TCA_FLOWER_UNSPEC,
337 			       sizeof(key->tp.dst));
338 	} else if (key->basic.ip_proto == IPPROTO_UDP) {
339 		fl_set_key_val(tb, &key->tp.src, TCA_FLOWER_KEY_UDP_SRC,
340 			       &mask->tp.src, TCA_FLOWER_UNSPEC,
341 			       sizeof(key->tp.src));
342 		fl_set_key_val(tb, &key->tp.dst, TCA_FLOWER_KEY_UDP_DST,
343 			       &mask->tp.dst, TCA_FLOWER_UNSPEC,
344 			       sizeof(key->tp.dst));
345 	}
346 
347 	return 0;
348 }
349 
350 static bool fl_mask_eq(struct fl_flow_mask *mask1,
351 		       struct fl_flow_mask *mask2)
352 {
353 	const long *lmask1 = fl_key_get_start(&mask1->key, mask1);
354 	const long *lmask2 = fl_key_get_start(&mask2->key, mask2);
355 
356 	return !memcmp(&mask1->range, &mask2->range, sizeof(mask1->range)) &&
357 	       !memcmp(lmask1, lmask2, fl_mask_range(mask1));
358 }
359 
360 static const struct rhashtable_params fl_ht_params = {
361 	.key_offset = offsetof(struct cls_fl_filter, mkey), /* base offset */
362 	.head_offset = offsetof(struct cls_fl_filter, ht_node),
363 	.automatic_shrinking = true,
364 };
365 
366 static int fl_init_hashtable(struct cls_fl_head *head,
367 			     struct fl_flow_mask *mask)
368 {
369 	head->ht_params = fl_ht_params;
370 	head->ht_params.key_len = fl_mask_range(mask);
371 	head->ht_params.key_offset += mask->range.start;
372 
373 	return rhashtable_init(&head->ht, &head->ht_params);
374 }
375 
376 #define FL_KEY_MEMBER_OFFSET(member) offsetof(struct fl_flow_key, member)
377 #define FL_KEY_MEMBER_SIZE(member) (sizeof(((struct fl_flow_key *) 0)->member))
378 #define FL_KEY_MEMBER_END_OFFSET(member)					\
379 	(FL_KEY_MEMBER_OFFSET(member) + FL_KEY_MEMBER_SIZE(member))
380 
381 #define FL_KEY_IN_RANGE(mask, member)						\
382         (FL_KEY_MEMBER_OFFSET(member) <= (mask)->range.end &&			\
383          FL_KEY_MEMBER_END_OFFSET(member) >= (mask)->range.start)
384 
385 #define FL_KEY_SET(keys, cnt, id, member)					\
386 	do {									\
387 		keys[cnt].key_id = id;						\
388 		keys[cnt].offset = FL_KEY_MEMBER_OFFSET(member);		\
389 		cnt++;								\
390 	} while(0);
391 
392 #define FL_KEY_SET_IF_IN_RANGE(mask, keys, cnt, id, member)			\
393 	do {									\
394 		if (FL_KEY_IN_RANGE(mask, member))				\
395 			FL_KEY_SET(keys, cnt, id, member);			\
396 	} while(0);
397 
398 static void fl_init_dissector(struct cls_fl_head *head,
399 			      struct fl_flow_mask *mask)
400 {
401 	struct flow_dissector_key keys[FLOW_DISSECTOR_KEY_MAX];
402 	size_t cnt = 0;
403 
404 	FL_KEY_SET(keys, cnt, FLOW_DISSECTOR_KEY_CONTROL, control);
405 	FL_KEY_SET(keys, cnt, FLOW_DISSECTOR_KEY_BASIC, basic);
406 	FL_KEY_SET_IF_IN_RANGE(mask, keys, cnt,
407 			       FLOW_DISSECTOR_KEY_ETH_ADDRS, eth);
408 	FL_KEY_SET_IF_IN_RANGE(mask, keys, cnt,
409 			       FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4);
410 	FL_KEY_SET_IF_IN_RANGE(mask, keys, cnt,
411 			       FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6);
412 	FL_KEY_SET_IF_IN_RANGE(mask, keys, cnt,
413 			       FLOW_DISSECTOR_KEY_PORTS, tp);
414 
415 	skb_flow_dissector_init(&head->dissector, keys, cnt);
416 }
417 
418 static int fl_check_assign_mask(struct cls_fl_head *head,
419 				struct fl_flow_mask *mask)
420 {
421 	int err;
422 
423 	if (head->mask_assigned) {
424 		if (!fl_mask_eq(&head->mask, mask))
425 			return -EINVAL;
426 		else
427 			return 0;
428 	}
429 
430 	/* Mask is not assigned yet. So assign it and init hashtable
431 	 * according to that.
432 	 */
433 	err = fl_init_hashtable(head, mask);
434 	if (err)
435 		return err;
436 	memcpy(&head->mask, mask, sizeof(head->mask));
437 	head->mask_assigned = true;
438 
439 	fl_init_dissector(head, mask);
440 
441 	return 0;
442 }
443 
444 static int fl_set_parms(struct net *net, struct tcf_proto *tp,
445 			struct cls_fl_filter *f, struct fl_flow_mask *mask,
446 			unsigned long base, struct nlattr **tb,
447 			struct nlattr *est, bool ovr)
448 {
449 	struct tcf_exts e;
450 	int err;
451 
452 	tcf_exts_init(&e, TCA_FLOWER_ACT, 0);
453 	err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
454 	if (err < 0)
455 		return err;
456 
457 	if (tb[TCA_FLOWER_CLASSID]) {
458 		f->res.classid = nla_get_u32(tb[TCA_FLOWER_CLASSID]);
459 		tcf_bind_filter(tp, &f->res, base);
460 	}
461 
462 	err = fl_set_key(net, tb, &f->key, &mask->key);
463 	if (err)
464 		goto errout;
465 
466 	fl_mask_update_range(mask);
467 	fl_set_masked_key(&f->mkey, &f->key, mask);
468 
469 	tcf_exts_change(tp, &f->exts, &e);
470 
471 	return 0;
472 errout:
473 	tcf_exts_destroy(&e);
474 	return err;
475 }
476 
477 static u32 fl_grab_new_handle(struct tcf_proto *tp,
478 			      struct cls_fl_head *head)
479 {
480 	unsigned int i = 0x80000000;
481 	u32 handle;
482 
483 	do {
484 		if (++head->hgen == 0x7FFFFFFF)
485 			head->hgen = 1;
486 	} while (--i > 0 && fl_get(tp, head->hgen));
487 
488 	if (unlikely(i == 0)) {
489 		pr_err("Insufficient number of handles\n");
490 		handle = 0;
491 	} else {
492 		handle = head->hgen;
493 	}
494 
495 	return handle;
496 }
497 
498 static int fl_change(struct net *net, struct sk_buff *in_skb,
499 		     struct tcf_proto *tp, unsigned long base,
500 		     u32 handle, struct nlattr **tca,
501 		     unsigned long *arg, bool ovr)
502 {
503 	struct cls_fl_head *head = rtnl_dereference(tp->root);
504 	struct cls_fl_filter *fold = (struct cls_fl_filter *) *arg;
505 	struct cls_fl_filter *fnew;
506 	struct nlattr *tb[TCA_FLOWER_MAX + 1];
507 	struct fl_flow_mask mask = {};
508 	u32 flags = 0;
509 	int err;
510 
511 	if (!tca[TCA_OPTIONS])
512 		return -EINVAL;
513 
514 	err = nla_parse_nested(tb, TCA_FLOWER_MAX, tca[TCA_OPTIONS], fl_policy);
515 	if (err < 0)
516 		return err;
517 
518 	if (fold && handle && fold->handle != handle)
519 		return -EINVAL;
520 
521 	fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
522 	if (!fnew)
523 		return -ENOBUFS;
524 
525 	tcf_exts_init(&fnew->exts, TCA_FLOWER_ACT, 0);
526 
527 	if (!handle) {
528 		handle = fl_grab_new_handle(tp, head);
529 		if (!handle) {
530 			err = -EINVAL;
531 			goto errout;
532 		}
533 	}
534 	fnew->handle = handle;
535 
536 	if (tb[TCA_FLOWER_FLAGS])
537 		flags = nla_get_u32(tb[TCA_FLOWER_FLAGS]);
538 
539 	err = fl_set_parms(net, tp, fnew, &mask, base, tb, tca[TCA_RATE], ovr);
540 	if (err)
541 		goto errout;
542 
543 	err = fl_check_assign_mask(head, &mask);
544 	if (err)
545 		goto errout;
546 
547 	err = rhashtable_insert_fast(&head->ht, &fnew->ht_node,
548 				     head->ht_params);
549 	if (err)
550 		goto errout;
551 
552 	fl_hw_replace_filter(tp,
553 			     &head->dissector,
554 			     &mask.key,
555 			     &fnew->key,
556 			     &fnew->exts,
557 			     (unsigned long)fnew,
558 			     flags);
559 
560 	if (fold) {
561 		rhashtable_remove_fast(&head->ht, &fold->ht_node,
562 				       head->ht_params);
563 		fl_hw_destroy_filter(tp, (unsigned long)fold);
564 	}
565 
566 	*arg = (unsigned long) fnew;
567 
568 	if (fold) {
569 		list_replace_rcu(&fold->list, &fnew->list);
570 		tcf_unbind_filter(tp, &fold->res);
571 		call_rcu(&fold->rcu, fl_destroy_filter);
572 	} else {
573 		list_add_tail_rcu(&fnew->list, &head->filters);
574 	}
575 
576 	return 0;
577 
578 errout:
579 	kfree(fnew);
580 	return err;
581 }
582 
583 static int fl_delete(struct tcf_proto *tp, unsigned long arg)
584 {
585 	struct cls_fl_head *head = rtnl_dereference(tp->root);
586 	struct cls_fl_filter *f = (struct cls_fl_filter *) arg;
587 
588 	rhashtable_remove_fast(&head->ht, &f->ht_node,
589 			       head->ht_params);
590 	list_del_rcu(&f->list);
591 	fl_hw_destroy_filter(tp, (unsigned long)f);
592 	tcf_unbind_filter(tp, &f->res);
593 	call_rcu(&f->rcu, fl_destroy_filter);
594 	return 0;
595 }
596 
597 static void fl_walk(struct tcf_proto *tp, struct tcf_walker *arg)
598 {
599 	struct cls_fl_head *head = rtnl_dereference(tp->root);
600 	struct cls_fl_filter *f;
601 
602 	list_for_each_entry_rcu(f, &head->filters, list) {
603 		if (arg->count < arg->skip)
604 			goto skip;
605 		if (arg->fn(tp, (unsigned long) f, arg) < 0) {
606 			arg->stop = 1;
607 			break;
608 		}
609 skip:
610 		arg->count++;
611 	}
612 }
613 
614 static int fl_dump_key_val(struct sk_buff *skb,
615 			   void *val, int val_type,
616 			   void *mask, int mask_type, int len)
617 {
618 	int err;
619 
620 	if (!memchr_inv(mask, 0, len))
621 		return 0;
622 	err = nla_put(skb, val_type, len, val);
623 	if (err)
624 		return err;
625 	if (mask_type != TCA_FLOWER_UNSPEC) {
626 		err = nla_put(skb, mask_type, len, mask);
627 		if (err)
628 			return err;
629 	}
630 	return 0;
631 }
632 
633 static int fl_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
634 		   struct sk_buff *skb, struct tcmsg *t)
635 {
636 	struct cls_fl_head *head = rtnl_dereference(tp->root);
637 	struct cls_fl_filter *f = (struct cls_fl_filter *) fh;
638 	struct nlattr *nest;
639 	struct fl_flow_key *key, *mask;
640 
641 	if (!f)
642 		return skb->len;
643 
644 	t->tcm_handle = f->handle;
645 
646 	nest = nla_nest_start(skb, TCA_OPTIONS);
647 	if (!nest)
648 		goto nla_put_failure;
649 
650 	if (f->res.classid &&
651 	    nla_put_u32(skb, TCA_FLOWER_CLASSID, f->res.classid))
652 		goto nla_put_failure;
653 
654 	key = &f->key;
655 	mask = &head->mask.key;
656 
657 	if (mask->indev_ifindex) {
658 		struct net_device *dev;
659 
660 		dev = __dev_get_by_index(net, key->indev_ifindex);
661 		if (dev && nla_put_string(skb, TCA_FLOWER_INDEV, dev->name))
662 			goto nla_put_failure;
663 	}
664 
665 	if (fl_dump_key_val(skb, key->eth.dst, TCA_FLOWER_KEY_ETH_DST,
666 			    mask->eth.dst, TCA_FLOWER_KEY_ETH_DST_MASK,
667 			    sizeof(key->eth.dst)) ||
668 	    fl_dump_key_val(skb, key->eth.src, TCA_FLOWER_KEY_ETH_SRC,
669 			    mask->eth.src, TCA_FLOWER_KEY_ETH_SRC_MASK,
670 			    sizeof(key->eth.src)) ||
671 	    fl_dump_key_val(skb, &key->basic.n_proto, TCA_FLOWER_KEY_ETH_TYPE,
672 			    &mask->basic.n_proto, TCA_FLOWER_UNSPEC,
673 			    sizeof(key->basic.n_proto)))
674 		goto nla_put_failure;
675 	if ((key->basic.n_proto == htons(ETH_P_IP) ||
676 	     key->basic.n_proto == htons(ETH_P_IPV6)) &&
677 	    fl_dump_key_val(skb, &key->basic.ip_proto, TCA_FLOWER_KEY_IP_PROTO,
678 			    &mask->basic.ip_proto, TCA_FLOWER_UNSPEC,
679 			    sizeof(key->basic.ip_proto)))
680 		goto nla_put_failure;
681 
682 	if (key->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS &&
683 	    (fl_dump_key_val(skb, &key->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC,
684 			     &mask->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC_MASK,
685 			     sizeof(key->ipv4.src)) ||
686 	     fl_dump_key_val(skb, &key->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST,
687 			     &mask->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST_MASK,
688 			     sizeof(key->ipv4.dst))))
689 		goto nla_put_failure;
690 	else if (key->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS &&
691 		 (fl_dump_key_val(skb, &key->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC,
692 				  &mask->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC_MASK,
693 				  sizeof(key->ipv6.src)) ||
694 		  fl_dump_key_val(skb, &key->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST,
695 				  &mask->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST_MASK,
696 				  sizeof(key->ipv6.dst))))
697 		goto nla_put_failure;
698 
699 	if (key->basic.ip_proto == IPPROTO_TCP &&
700 	    (fl_dump_key_val(skb, &key->tp.src, TCA_FLOWER_KEY_TCP_SRC,
701 			     &mask->tp.src, TCA_FLOWER_UNSPEC,
702 			     sizeof(key->tp.src)) ||
703 	     fl_dump_key_val(skb, &key->tp.dst, TCA_FLOWER_KEY_TCP_DST,
704 			     &mask->tp.dst, TCA_FLOWER_UNSPEC,
705 			     sizeof(key->tp.dst))))
706 		goto nla_put_failure;
707 	else if (key->basic.ip_proto == IPPROTO_UDP &&
708 		 (fl_dump_key_val(skb, &key->tp.src, TCA_FLOWER_KEY_UDP_SRC,
709 				  &mask->tp.src, TCA_FLOWER_UNSPEC,
710 				  sizeof(key->tp.src)) ||
711 		  fl_dump_key_val(skb, &key->tp.dst, TCA_FLOWER_KEY_UDP_DST,
712 				  &mask->tp.dst, TCA_FLOWER_UNSPEC,
713 				  sizeof(key->tp.dst))))
714 		goto nla_put_failure;
715 
716 	if (tcf_exts_dump(skb, &f->exts))
717 		goto nla_put_failure;
718 
719 	nla_nest_end(skb, nest);
720 
721 	if (tcf_exts_dump_stats(skb, &f->exts) < 0)
722 		goto nla_put_failure;
723 
724 	return skb->len;
725 
726 nla_put_failure:
727 	nla_nest_cancel(skb, nest);
728 	return -1;
729 }
730 
731 static struct tcf_proto_ops cls_fl_ops __read_mostly = {
732 	.kind		= "flower",
733 	.classify	= fl_classify,
734 	.init		= fl_init,
735 	.destroy	= fl_destroy,
736 	.get		= fl_get,
737 	.change		= fl_change,
738 	.delete		= fl_delete,
739 	.walk		= fl_walk,
740 	.dump		= fl_dump,
741 	.owner		= THIS_MODULE,
742 };
743 
744 static int __init cls_fl_init(void)
745 {
746 	return register_tcf_proto_ops(&cls_fl_ops);
747 }
748 
749 static void __exit cls_fl_exit(void)
750 {
751 	unregister_tcf_proto_ops(&cls_fl_ops);
752 }
753 
754 module_init(cls_fl_init);
755 module_exit(cls_fl_exit);
756 
757 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
758 MODULE_DESCRIPTION("Flower classifier");
759 MODULE_LICENSE("GPL v2");
760