br_fdb.c revision 60063497a95e716c9a689af3be2687d261f115b4
1/*
2 *	Forwarding database
3 *	Linux ethernet bridge
4 *
5 *	Authors:
6 *	Lennert Buytenhek		<buytenh@gnu.org>
7 *
8 *	This program is free software; you can redistribute it and/or
9 *	modify it under the terms of the GNU General Public License
10 *	as published by the Free Software Foundation; either version
11 *	2 of the License, or (at your option) any later version.
12 */
13
14#include <linux/kernel.h>
15#include <linux/init.h>
16#include <linux/rculist.h>
17#include <linux/spinlock.h>
18#include <linux/times.h>
19#include <linux/netdevice.h>
20#include <linux/etherdevice.h>
21#include <linux/jhash.h>
22#include <linux/random.h>
23#include <linux/slab.h>
24#include <linux/atomic.h>
25#include <asm/unaligned.h>
26#include "br_private.h"
27
28static struct kmem_cache *br_fdb_cache __read_mostly;
29static int fdb_insert(struct net_bridge *br, struct net_bridge_port *source,
30		      const unsigned char *addr);
31static void fdb_notify(const struct net_bridge_fdb_entry *, int);
32
33static u32 fdb_salt __read_mostly;
34
35int __init br_fdb_init(void)
36{
37	br_fdb_cache = kmem_cache_create("bridge_fdb_cache",
38					 sizeof(struct net_bridge_fdb_entry),
39					 0,
40					 SLAB_HWCACHE_ALIGN, NULL);
41	if (!br_fdb_cache)
42		return -ENOMEM;
43
44	get_random_bytes(&fdb_salt, sizeof(fdb_salt));
45	return 0;
46}
47
48void br_fdb_fini(void)
49{
50	kmem_cache_destroy(br_fdb_cache);
51}
52
53
54/* if topology_changing then use forward_delay (default 15 sec)
55 * otherwise keep longer (default 5 minutes)
56 */
57static inline unsigned long hold_time(const struct net_bridge *br)
58{
59	return br->topology_change ? br->forward_delay : br->ageing_time;
60}
61
62static inline int has_expired(const struct net_bridge *br,
63				  const struct net_bridge_fdb_entry *fdb)
64{
65	return !fdb->is_static &&
66		time_before_eq(fdb->updated + hold_time(br), jiffies);
67}
68
69static inline int br_mac_hash(const unsigned char *mac)
70{
71	/* use 1 byte of OUI cnd 3 bytes of NIC */
72	u32 key = get_unaligned((u32 *)(mac + 2));
73	return jhash_1word(key, fdb_salt) & (BR_HASH_SIZE - 1);
74}
75
76static void fdb_rcu_free(struct rcu_head *head)
77{
78	struct net_bridge_fdb_entry *ent
79		= container_of(head, struct net_bridge_fdb_entry, rcu);
80	kmem_cache_free(br_fdb_cache, ent);
81}
82
83static inline void fdb_delete(struct net_bridge_fdb_entry *f)
84{
85	fdb_notify(f, RTM_DELNEIGH);
86	hlist_del_rcu(&f->hlist);
87	call_rcu(&f->rcu, fdb_rcu_free);
88}
89
90void br_fdb_changeaddr(struct net_bridge_port *p, const unsigned char *newaddr)
91{
92	struct net_bridge *br = p->br;
93	int i;
94
95	spin_lock_bh(&br->hash_lock);
96
97	/* Search all chains since old address/hash is unknown */
98	for (i = 0; i < BR_HASH_SIZE; i++) {
99		struct hlist_node *h;
100		hlist_for_each(h, &br->hash[i]) {
101			struct net_bridge_fdb_entry *f;
102
103			f = hlist_entry(h, struct net_bridge_fdb_entry, hlist);
104			if (f->dst == p && f->is_local) {
105				/* maybe another port has same hw addr? */
106				struct net_bridge_port *op;
107				list_for_each_entry(op, &br->port_list, list) {
108					if (op != p &&
109					    !compare_ether_addr(op->dev->dev_addr,
110								f->addr.addr)) {
111						f->dst = op;
112						goto insert;
113					}
114				}
115
116				/* delete old one */
117				fdb_delete(f);
118				goto insert;
119			}
120		}
121	}
122 insert:
123	/* insert new address,  may fail if invalid address or dup. */
124	fdb_insert(br, p, newaddr);
125
126	spin_unlock_bh(&br->hash_lock);
127}
128
129void br_fdb_cleanup(unsigned long _data)
130{
131	struct net_bridge *br = (struct net_bridge *)_data;
132	unsigned long delay = hold_time(br);
133	unsigned long next_timer = jiffies + br->ageing_time;
134	int i;
135
136	spin_lock_bh(&br->hash_lock);
137	for (i = 0; i < BR_HASH_SIZE; i++) {
138		struct net_bridge_fdb_entry *f;
139		struct hlist_node *h, *n;
140
141		hlist_for_each_entry_safe(f, h, n, &br->hash[i], hlist) {
142			unsigned long this_timer;
143			if (f->is_static)
144				continue;
145			this_timer = f->updated + delay;
146			if (time_before_eq(this_timer, jiffies))
147				fdb_delete(f);
148			else if (time_before(this_timer, next_timer))
149				next_timer = this_timer;
150		}
151	}
152	spin_unlock_bh(&br->hash_lock);
153
154	mod_timer(&br->gc_timer, round_jiffies_up(next_timer));
155}
156
157/* Completely flush all dynamic entries in forwarding database.*/
158void br_fdb_flush(struct net_bridge *br)
159{
160	int i;
161
162	spin_lock_bh(&br->hash_lock);
163	for (i = 0; i < BR_HASH_SIZE; i++) {
164		struct net_bridge_fdb_entry *f;
165		struct hlist_node *h, *n;
166		hlist_for_each_entry_safe(f, h, n, &br->hash[i], hlist) {
167			if (!f->is_static)
168				fdb_delete(f);
169		}
170	}
171	spin_unlock_bh(&br->hash_lock);
172}
173
174/* Flush all entries referring to a specific port.
175 * if do_all is set also flush static entries
176 */
177void br_fdb_delete_by_port(struct net_bridge *br,
178			   const struct net_bridge_port *p,
179			   int do_all)
180{
181	int i;
182
183	spin_lock_bh(&br->hash_lock);
184	for (i = 0; i < BR_HASH_SIZE; i++) {
185		struct hlist_node *h, *g;
186
187		hlist_for_each_safe(h, g, &br->hash[i]) {
188			struct net_bridge_fdb_entry *f
189				= hlist_entry(h, struct net_bridge_fdb_entry, hlist);
190			if (f->dst != p)
191				continue;
192
193			if (f->is_static && !do_all)
194				continue;
195			/*
196			 * if multiple ports all have the same device address
197			 * then when one port is deleted, assign
198			 * the local entry to other port
199			 */
200			if (f->is_local) {
201				struct net_bridge_port *op;
202				list_for_each_entry(op, &br->port_list, list) {
203					if (op != p &&
204					    !compare_ether_addr(op->dev->dev_addr,
205								f->addr.addr)) {
206						f->dst = op;
207						goto skip_delete;
208					}
209				}
210			}
211
212			fdb_delete(f);
213		skip_delete: ;
214		}
215	}
216	spin_unlock_bh(&br->hash_lock);
217}
218
219/* No locking or refcounting, assumes caller has rcu_read_lock */
220struct net_bridge_fdb_entry *__br_fdb_get(struct net_bridge *br,
221					  const unsigned char *addr)
222{
223	struct hlist_node *h;
224	struct net_bridge_fdb_entry *fdb;
225
226	hlist_for_each_entry_rcu(fdb, h, &br->hash[br_mac_hash(addr)], hlist) {
227		if (!compare_ether_addr(fdb->addr.addr, addr)) {
228			if (unlikely(has_expired(br, fdb)))
229				break;
230			return fdb;
231		}
232	}
233
234	return NULL;
235}
236
237#if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
238/* Interface used by ATM LANE hook to test
239 * if an addr is on some other bridge port */
240int br_fdb_test_addr(struct net_device *dev, unsigned char *addr)
241{
242	struct net_bridge_fdb_entry *fdb;
243	struct net_bridge_port *port;
244	int ret;
245
246	rcu_read_lock();
247	port = br_port_get_rcu(dev);
248	if (!port)
249		ret = 0;
250	else {
251		fdb = __br_fdb_get(port->br, addr);
252		ret = fdb && fdb->dst->dev != dev &&
253			fdb->dst->state == BR_STATE_FORWARDING;
254	}
255	rcu_read_unlock();
256
257	return ret;
258}
259#endif /* CONFIG_ATM_LANE */
260
261/*
262 * Fill buffer with forwarding table records in
263 * the API format.
264 */
265int br_fdb_fillbuf(struct net_bridge *br, void *buf,
266		   unsigned long maxnum, unsigned long skip)
267{
268	struct __fdb_entry *fe = buf;
269	int i, num = 0;
270	struct hlist_node *h;
271	struct net_bridge_fdb_entry *f;
272
273	memset(buf, 0, maxnum*sizeof(struct __fdb_entry));
274
275	rcu_read_lock();
276	for (i = 0; i < BR_HASH_SIZE; i++) {
277		hlist_for_each_entry_rcu(f, h, &br->hash[i], hlist) {
278			if (num >= maxnum)
279				goto out;
280
281			if (has_expired(br, f))
282				continue;
283
284			if (skip) {
285				--skip;
286				continue;
287			}
288
289			/* convert from internal format to API */
290			memcpy(fe->mac_addr, f->addr.addr, ETH_ALEN);
291
292			/* due to ABI compat need to split into hi/lo */
293			fe->port_no = f->dst->port_no;
294			fe->port_hi = f->dst->port_no >> 8;
295
296			fe->is_local = f->is_local;
297			if (!f->is_static)
298				fe->ageing_timer_value = jiffies_to_clock_t(jiffies - f->updated);
299			++fe;
300			++num;
301		}
302	}
303
304 out:
305	rcu_read_unlock();
306
307	return num;
308}
309
310static struct net_bridge_fdb_entry *fdb_find(struct hlist_head *head,
311					     const unsigned char *addr)
312{
313	struct hlist_node *h;
314	struct net_bridge_fdb_entry *fdb;
315
316	hlist_for_each_entry(fdb, h, head, hlist) {
317		if (!compare_ether_addr(fdb->addr.addr, addr))
318			return fdb;
319	}
320	return NULL;
321}
322
323static struct net_bridge_fdb_entry *fdb_find_rcu(struct hlist_head *head,
324						 const unsigned char *addr)
325{
326	struct hlist_node *h;
327	struct net_bridge_fdb_entry *fdb;
328
329	hlist_for_each_entry_rcu(fdb, h, head, hlist) {
330		if (!compare_ether_addr(fdb->addr.addr, addr))
331			return fdb;
332	}
333	return NULL;
334}
335
336static struct net_bridge_fdb_entry *fdb_create(struct hlist_head *head,
337					       struct net_bridge_port *source,
338					       const unsigned char *addr)
339{
340	struct net_bridge_fdb_entry *fdb;
341
342	fdb = kmem_cache_alloc(br_fdb_cache, GFP_ATOMIC);
343	if (fdb) {
344		memcpy(fdb->addr.addr, addr, ETH_ALEN);
345		fdb->dst = source;
346		fdb->is_local = 0;
347		fdb->is_static = 0;
348		fdb->updated = fdb->used = jiffies;
349		hlist_add_head_rcu(&fdb->hlist, head);
350		fdb_notify(fdb, RTM_NEWNEIGH);
351	}
352	return fdb;
353}
354
355static int fdb_insert(struct net_bridge *br, struct net_bridge_port *source,
356		  const unsigned char *addr)
357{
358	struct hlist_head *head = &br->hash[br_mac_hash(addr)];
359	struct net_bridge_fdb_entry *fdb;
360
361	if (!is_valid_ether_addr(addr))
362		return -EINVAL;
363
364	fdb = fdb_find(head, addr);
365	if (fdb) {
366		/* it is okay to have multiple ports with same
367		 * address, just use the first one.
368		 */
369		if (fdb->is_local)
370			return 0;
371		br_warn(br, "adding interface %s with same address "
372		       "as a received packet\n",
373		       source->dev->name);
374		fdb_delete(fdb);
375	}
376
377	fdb = fdb_create(head, source, addr);
378	if (!fdb)
379		return -ENOMEM;
380
381	fdb->is_local = fdb->is_static = 1;
382	return 0;
383}
384
385/* Add entry for local address of interface */
386int br_fdb_insert(struct net_bridge *br, struct net_bridge_port *source,
387		  const unsigned char *addr)
388{
389	int ret;
390
391	spin_lock_bh(&br->hash_lock);
392	ret = fdb_insert(br, source, addr);
393	spin_unlock_bh(&br->hash_lock);
394	return ret;
395}
396
397void br_fdb_update(struct net_bridge *br, struct net_bridge_port *source,
398		   const unsigned char *addr)
399{
400	struct hlist_head *head = &br->hash[br_mac_hash(addr)];
401	struct net_bridge_fdb_entry *fdb;
402
403	/* some users want to always flood. */
404	if (hold_time(br) == 0)
405		return;
406
407	/* ignore packets unless we are using this port */
408	if (!(source->state == BR_STATE_LEARNING ||
409	      source->state == BR_STATE_FORWARDING))
410		return;
411
412	fdb = fdb_find_rcu(head, addr);
413	if (likely(fdb)) {
414		/* attempt to update an entry for a local interface */
415		if (unlikely(fdb->is_local)) {
416			if (net_ratelimit())
417				br_warn(br, "received packet on %s with "
418					"own address as source address\n",
419					source->dev->name);
420		} else {
421			/* fastpath: update of existing entry */
422			fdb->dst = source;
423			fdb->updated = jiffies;
424		}
425	} else {
426		spin_lock(&br->hash_lock);
427		if (likely(!fdb_find(head, addr)))
428			fdb_create(head, source, addr);
429
430		/* else  we lose race and someone else inserts
431		 * it first, don't bother updating
432		 */
433		spin_unlock(&br->hash_lock);
434	}
435}
436
437static int fdb_to_nud(const struct net_bridge_fdb_entry *fdb)
438{
439	if (fdb->is_local)
440		return NUD_PERMANENT;
441	else if (fdb->is_static)
442		return NUD_NOARP;
443	else if (has_expired(fdb->dst->br, fdb))
444		return NUD_STALE;
445	else
446		return NUD_REACHABLE;
447}
448
449static int fdb_fill_info(struct sk_buff *skb,
450			 const struct net_bridge_fdb_entry *fdb,
451			 u32 pid, u32 seq, int type, unsigned int flags)
452{
453	unsigned long now = jiffies;
454	struct nda_cacheinfo ci;
455	struct nlmsghdr *nlh;
456	struct ndmsg *ndm;
457
458	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
459	if (nlh == NULL)
460		return -EMSGSIZE;
461
462
463	ndm = nlmsg_data(nlh);
464	ndm->ndm_family	 = AF_BRIDGE;
465	ndm->ndm_pad1    = 0;
466	ndm->ndm_pad2    = 0;
467	ndm->ndm_flags	 = 0;
468	ndm->ndm_type	 = 0;
469	ndm->ndm_ifindex = fdb->dst->dev->ifindex;
470	ndm->ndm_state   = fdb_to_nud(fdb);
471
472	NLA_PUT(skb, NDA_LLADDR, ETH_ALEN, &fdb->addr);
473
474	ci.ndm_used	 = jiffies_to_clock_t(now - fdb->used);
475	ci.ndm_confirmed = 0;
476	ci.ndm_updated	 = jiffies_to_clock_t(now - fdb->updated);
477	ci.ndm_refcnt	 = 0;
478	NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
479
480	return nlmsg_end(skb, nlh);
481
482nla_put_failure:
483	nlmsg_cancel(skb, nlh);
484	return -EMSGSIZE;
485}
486
487static inline size_t fdb_nlmsg_size(void)
488{
489	return NLMSG_ALIGN(sizeof(struct ndmsg))
490		+ nla_total_size(ETH_ALEN) /* NDA_LLADDR */
491		+ nla_total_size(sizeof(struct nda_cacheinfo));
492}
493
494static void fdb_notify(const struct net_bridge_fdb_entry *fdb, int type)
495{
496	struct net *net = dev_net(fdb->dst->dev);
497	struct sk_buff *skb;
498	int err = -ENOBUFS;
499
500	skb = nlmsg_new(fdb_nlmsg_size(), GFP_ATOMIC);
501	if (skb == NULL)
502		goto errout;
503
504	err = fdb_fill_info(skb, fdb, 0, 0, type, 0);
505	if (err < 0) {
506		/* -EMSGSIZE implies BUG in fdb_nlmsg_size() */
507		WARN_ON(err == -EMSGSIZE);
508		kfree_skb(skb);
509		goto errout;
510	}
511	rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
512	return;
513errout:
514	if (err < 0)
515		rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
516}
517
518/* Dump information about entries, in response to GETNEIGH */
519int br_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb)
520{
521	struct net *net = sock_net(skb->sk);
522	struct net_device *dev;
523	int idx = 0;
524
525	rcu_read_lock();
526	for_each_netdev_rcu(net, dev) {
527		struct net_bridge *br = netdev_priv(dev);
528		int i;
529
530		if (!(dev->priv_flags & IFF_EBRIDGE))
531			continue;
532
533		for (i = 0; i < BR_HASH_SIZE; i++) {
534			struct hlist_node *h;
535			struct net_bridge_fdb_entry *f;
536
537			hlist_for_each_entry_rcu(f, h, &br->hash[i], hlist) {
538				if (idx < cb->args[0])
539					goto skip;
540
541				if (fdb_fill_info(skb, f,
542						  NETLINK_CB(cb->skb).pid,
543						  cb->nlh->nlmsg_seq,
544						  RTM_NEWNEIGH,
545						  NLM_F_MULTI) < 0)
546					break;
547skip:
548				++idx;
549			}
550		}
551	}
552	rcu_read_unlock();
553
554	cb->args[0] = idx;
555
556	return skb->len;
557}
558
559/* Create new static fdb entry */
560static int fdb_add_entry(struct net_bridge_port *source, const __u8 *addr,
561			 __u16 state)
562{
563	struct net_bridge *br = source->br;
564	struct hlist_head *head = &br->hash[br_mac_hash(addr)];
565	struct net_bridge_fdb_entry *fdb;
566
567	fdb = fdb_find(head, addr);
568	if (fdb)
569		return -EEXIST;
570
571	fdb = fdb_create(head, source, addr);
572	if (!fdb)
573		return -ENOMEM;
574
575	if (state & NUD_PERMANENT)
576		fdb->is_local = fdb->is_static = 1;
577	else if (state & NUD_NOARP)
578		fdb->is_static = 1;
579	return 0;
580}
581
582/* Add new permanent fdb entry with RTM_NEWNEIGH */
583int br_fdb_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
584{
585	struct net *net = sock_net(skb->sk);
586	struct ndmsg *ndm;
587	struct nlattr *tb[NDA_MAX+1];
588	struct net_device *dev;
589	struct net_bridge_port *p;
590	const __u8 *addr;
591	int err;
592
593	ASSERT_RTNL();
594	err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
595	if (err < 0)
596		return err;
597
598	ndm = nlmsg_data(nlh);
599	if (ndm->ndm_ifindex == 0) {
600		pr_info("bridge: RTM_NEWNEIGH with invalid ifindex\n");
601		return -EINVAL;
602	}
603
604	dev = __dev_get_by_index(net, ndm->ndm_ifindex);
605	if (dev == NULL) {
606		pr_info("bridge: RTM_NEWNEIGH with unknown ifindex\n");
607		return -ENODEV;
608	}
609
610	if (!tb[NDA_LLADDR] || nla_len(tb[NDA_LLADDR]) != ETH_ALEN) {
611		pr_info("bridge: RTM_NEWNEIGH with invalid address\n");
612		return -EINVAL;
613	}
614
615	addr = nla_data(tb[NDA_LLADDR]);
616	if (!is_valid_ether_addr(addr)) {
617		pr_info("bridge: RTM_NEWNEIGH with invalid ether address\n");
618		return -EINVAL;
619	}
620
621	p = br_port_get_rtnl(dev);
622	if (p == NULL) {
623		pr_info("bridge: RTM_NEWNEIGH %s not a bridge port\n",
624			dev->name);
625		return -EINVAL;
626	}
627
628	spin_lock_bh(&p->br->hash_lock);
629	err = fdb_add_entry(p, addr, ndm->ndm_state);
630	spin_unlock_bh(&p->br->hash_lock);
631
632	return err;
633}
634
635static int fdb_delete_by_addr(struct net_bridge_port *p, const u8 *addr)
636{
637	struct net_bridge *br = p->br;
638	struct hlist_head *head = &br->hash[br_mac_hash(addr)];
639	struct net_bridge_fdb_entry *fdb;
640
641	fdb = fdb_find(head, addr);
642	if (!fdb)
643		return -ENOENT;
644
645	fdb_delete(fdb);
646	return 0;
647}
648
649/* Remove neighbor entry with RTM_DELNEIGH */
650int br_fdb_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
651{
652	struct net *net = sock_net(skb->sk);
653	struct ndmsg *ndm;
654	struct net_bridge_port *p;
655	struct nlattr *llattr;
656	const __u8 *addr;
657	struct net_device *dev;
658	int err;
659
660	ASSERT_RTNL();
661	if (nlmsg_len(nlh) < sizeof(*ndm))
662		return -EINVAL;
663
664	ndm = nlmsg_data(nlh);
665	if (ndm->ndm_ifindex == 0) {
666		pr_info("bridge: RTM_DELNEIGH with invalid ifindex\n");
667		return -EINVAL;
668	}
669
670	dev = __dev_get_by_index(net, ndm->ndm_ifindex);
671	if (dev == NULL) {
672		pr_info("bridge: RTM_DELNEIGH with unknown ifindex\n");
673		return -ENODEV;
674	}
675
676	llattr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_LLADDR);
677	if (llattr == NULL || nla_len(llattr) != ETH_ALEN) {
678		pr_info("bridge: RTM_DELNEIGH with invalid address\n");
679		return -EINVAL;
680	}
681
682	addr = nla_data(llattr);
683
684	p = br_port_get_rtnl(dev);
685	if (p == NULL) {
686		pr_info("bridge: RTM_DELNEIGH %s not a bridge port\n",
687			dev->name);
688		return -EINVAL;
689	}
690
691	spin_lock_bh(&p->br->hash_lock);
692	err = fdb_delete_by_addr(p, addr);
693	spin_unlock_bh(&p->br->hash_lock);
694
695	return err;
696}
697