2 * NET3: Token ring device handling subroutines
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.
9 * Fixes: 3 Feb 97 Paul Norton <pnorton@cts.com> Minor routing fixes.
10 * Added rif table to /proc/net/tr_rif and rif timeout to
11 * /proc/sys/net/token-ring/rif_timeout.
12 * 22 Jun 98 Paul Norton <p.norton@computer.org> Rearranged
13 * tr_header and tr_type_trans to handle passing IPX SNAP and
14 * 802.2 through the correct layers. Eliminated tr_reformat.
18 #include <asm/uaccess.h>
19 #include <asm/system.h>
20 #include <linux/module.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/jiffies.h>
24 #include <linux/string.h>
26 #include <linux/socket.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/trdevice.h>
31 #include <linux/skbuff.h>
32 #include <linux/errno.h>
33 #include <linux/timer.h>
34 #include <linux/net.h>
35 #include <linux/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/init.h>
38 #include <linux/sysctl.h>
40 #include <net/net_namespace.h>
42 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev);
43 static void rif_check_expire(unsigned long dummy);
48 * Each RIF entry we learn is kept this way
52 unsigned char addr[TR_ALEN];
56 struct rif_cache *next;
57 unsigned long last_used;
58 unsigned char local_ring;
61 #define RIF_TABLE_SIZE 32
64 * We hash the RIF cache 32 ways. We do after all have to look it
68 static struct rif_cache *rif_table[RIF_TABLE_SIZE];
70 static DEFINE_SPINLOCK(rif_lock);
74 * Garbage disposal timer.
77 static struct timer_list rif_timer;
79 static int sysctl_tr_rif_timeout = 60*10*HZ;
81 static inline unsigned long rif_hash(const unsigned char *addr)
86 x = (x << 2) ^ addr[1];
87 x = (x << 2) ^ addr[2];
88 x = (x << 2) ^ addr[3];
89 x = (x << 2) ^ addr[4];
90 x = (x << 2) ^ addr[5];
94 return x & (RIF_TABLE_SIZE - 1);
98 * Put the headers on a token ring packet. Token ring source routing
99 * makes this a little more exciting than on ethernet.
102 static int tr_header(struct sk_buff *skb, struct net_device *dev,
104 const void *daddr, const void *saddr, unsigned len)
110 * Add the 802.2 SNAP header if IP as the IPv4/IPv6 code calls
111 * dev->hard_header directly.
113 if (type == ETH_P_IP || type == ETH_P_IPV6 || type == ETH_P_ARP)
117 hdr_len = sizeof(struct trh_hdr) + sizeof(struct trllc);
118 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
119 trllc = (struct trllc *)(trh+1);
120 trllc->dsap = trllc->ssap = EXTENDED_SAP;
122 trllc->protid[0] = trllc->protid[1] = trllc->protid[2] = 0x00;
123 trllc->ethertype = htons(type);
127 hdr_len = sizeof(struct trh_hdr);
128 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
135 memcpy(trh->saddr,saddr,dev->addr_len);
137 memcpy(trh->saddr,dev->dev_addr,dev->addr_len);
140 * Build the destination and then source route the frame
145 memcpy(trh->daddr,daddr,dev->addr_len);
146 tr_source_route(skb, trh, dev);
154 * A neighbour discovery of some species (eg arp) has completed. We
155 * can now send the packet.
158 static int tr_rebuild_header(struct sk_buff *skb)
160 struct trh_hdr *trh=(struct trh_hdr *)skb->data;
161 struct trllc *trllc=(struct trllc *)(skb->data+sizeof(struct trh_hdr));
162 struct net_device *dev = skb->dev;
165 * FIXME: We don't yet support IPv6 over token rings
168 if(trllc->ethertype != htons(ETH_P_IP)) {
169 printk("tr_rebuild_header: Don't know how to resolve type %04X addresses ?\n", ntohs(trllc->ethertype));
174 if(arp_find(trh->daddr, skb)) {
180 tr_source_route(skb,trh,dev);
186 * Some of this is a bit hackish. We intercept RIF information
187 * used for source routing. We also grab IP directly and don't feed
191 __be16 tr_type_trans(struct sk_buff *skb, struct net_device *dev)
199 skb_reset_mac_header(skb);
202 if(trh->saddr[0] & TR_RII)
203 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
205 trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
207 skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
209 if(*trh->daddr & 0x80)
211 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))
212 skb->pkt_type=PACKET_BROADCAST;
214 skb->pkt_type=PACKET_MULTICAST;
216 else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
218 skb->pkt_type=PACKET_MULTICAST;
220 else if(dev->flags & IFF_PROMISC)
222 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
223 skb->pkt_type=PACKET_OTHERHOST;
226 if ((skb->pkt_type != PACKET_BROADCAST) &&
227 (skb->pkt_type != PACKET_MULTICAST))
228 tr_add_rif_info(trh,dev) ;
231 * Strip the SNAP header from ARP packets since we don't
232 * pass them through to the 802.2/SNAP layers.
235 if (trllc->dsap == EXTENDED_SAP &&
236 (trllc->ethertype == htons(ETH_P_IP) ||
237 trllc->ethertype == htons(ETH_P_IPV6) ||
238 trllc->ethertype == htons(ETH_P_ARP)))
240 skb_pull(skb, sizeof(struct trllc));
241 return trllc->ethertype;
244 return htons(ETH_P_TR_802_2);
248 * We try to do source routing...
251 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,
252 struct net_device *dev)
256 struct rif_cache *entry;
257 unsigned char *olddata;
259 static const unsigned char mcast_func_addr[]
260 = {0xC0,0x00,0x00,0x04,0x00,0x00};
262 spin_lock_irqsave(&rif_lock, flags);
265 * Broadcasts are single route as stated in RFC 1042
267 if( (!memcmp(&(trh->daddr[0]),&(dev->broadcast[0]),TR_ALEN)) ||
268 (!memcmp(&(trh->daddr[0]),&(mcast_func_addr[0]), TR_ALEN)) )
270 trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
271 | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
272 trh->saddr[0]|=TR_RII;
276 hash = rif_hash(trh->daddr);
278 * Walk the hash table and look for an entry
280 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->daddr[0]),TR_ALEN);entry=entry->next);
283 * If we found an entry we can route the frame.
288 printk("source routing for %pM\n", trh->daddr);
290 if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
293 memcpy(&trh->rseg[0],&entry->rseg[0],8*sizeof(unsigned short));
294 trh->rcf^=htons(TR_RCF_DIR_BIT);
295 trh->rcf&=htons(0x1fff); /* Issam Chehab <ichehab@madge1.demon.co.uk> */
297 trh->saddr[0]|=TR_RII;
299 printk("entry found with rcf %04x\n", entry->rcf);
303 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
306 entry->last_used=jiffies;
311 * Without the information we simply have to shout
312 * on the wire. The replies should rapidly clean this
315 trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
316 | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
317 trh->saddr[0]|=TR_RII;
319 printk("no entry in rif table found - broadcasting frame\n");
324 /* Compress the RIF here so we don't have to do it in the driver(s) */
325 if (!(trh->saddr[0] & 0x80))
328 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
330 spin_unlock_irqrestore(&rif_lock, flags);
332 skb_pull(skb, slack);
333 memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
337 * We have learned some new RIF information for our source
341 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
343 unsigned int hash, rii_p = 0;
345 struct rif_cache *entry;
346 unsigned char saddr0;
348 spin_lock_irqsave(&rif_lock, flags);
349 saddr0 = trh->saddr[0];
352 * Firstly see if the entry exists
355 if(trh->saddr[0] & TR_RII)
358 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
364 hash = rif_hash(trh->saddr);
365 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);entry=entry->next);
370 printk("adding rif_entry: addr:%pM rcf:%04X\n",
371 trh->saddr, ntohs(trh->rcf));
374 * Allocate our new entry. A failure to allocate loses
375 * use the information. This is harmless.
377 * FIXME: We ought to keep some kind of cache size
378 * limiting and adjust the timers to suit.
380 entry=kmalloc(sizeof(struct rif_cache),GFP_ATOMIC);
384 printk(KERN_DEBUG "tr.c: Couldn't malloc rif cache entry !\n");
385 spin_unlock_irqrestore(&rif_lock, flags);
389 memcpy(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);
390 entry->iface = dev->ifindex;
391 entry->next=rif_table[hash];
392 entry->last_used=jiffies;
393 rif_table[hash]=entry;
397 entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
398 memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
399 entry->local_ring = 0;
403 entry->local_ring = 1;
406 else /* Y. Tahara added */
409 * Update existing entries
411 if (!entry->local_ring)
412 if (entry->rcf != (trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK)) &&
413 !(trh->rcf & htons(TR_RCF_BROADCAST_MASK)))
416 printk("updating rif_entry: addr:%pM rcf:%04X\n",
417 trh->saddr, ntohs(trh->rcf));
419 entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
420 memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
422 entry->last_used=jiffies;
424 trh->saddr[0]=saddr0; /* put the routing indicator back for tcpdump */
425 spin_unlock_irqrestore(&rif_lock, flags);
429 * Scan the cache with a timer and see what we need to throw out.
432 static void rif_check_expire(unsigned long dummy)
435 unsigned long flags, next_interval = jiffies + sysctl_tr_rif_timeout/2;
437 spin_lock_irqsave(&rif_lock, flags);
439 for(i =0; i < RIF_TABLE_SIZE; i++) {
440 struct rif_cache *entry, **pentry;
442 pentry = rif_table+i;
443 while((entry=*pentry) != NULL) {
444 unsigned long expires
445 = entry->last_used + sysctl_tr_rif_timeout;
447 if (time_before_eq(expires, jiffies)) {
448 *pentry = entry->next;
451 pentry = &entry->next;
453 if (time_before(expires, next_interval))
454 next_interval = expires;
459 spin_unlock_irqrestore(&rif_lock, flags);
461 mod_timer(&rif_timer, next_interval);
466 * Generate the /proc/net information for the token ring RIF
470 #ifdef CONFIG_PROC_FS
472 static struct rif_cache *rif_get_idx(loff_t pos)
475 struct rif_cache *entry;
478 for(i = 0; i < RIF_TABLE_SIZE; i++)
479 for(entry = rif_table[i]; entry; entry = entry->next) {
488 static void *rif_seq_start(struct seq_file *seq, loff_t *pos)
490 spin_lock_irq(&rif_lock);
492 return *pos ? rif_get_idx(*pos - 1) : SEQ_START_TOKEN;
495 static void *rif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
498 struct rif_cache *ent = v;
502 if (v == SEQ_START_TOKEN) {
510 i = rif_hash(ent->addr);
512 while (++i < RIF_TABLE_SIZE) {
513 if ((ent = rif_table[i]) != NULL)
519 static void rif_seq_stop(struct seq_file *seq, void *v)
521 spin_unlock_irq(&rif_lock);
524 static int rif_seq_show(struct seq_file *seq, void *v)
526 int j, rcf_len, segment, brdgnmb;
527 struct rif_cache *entry = v;
529 if (v == SEQ_START_TOKEN)
531 "if TR address TTL rcf routing segments\n");
533 struct net_device *dev = dev_get_by_index(&init_net, entry->iface);
534 long ttl = (long) (entry->last_used + sysctl_tr_rif_timeout)
537 seq_printf(seq, "%s %pM %7li ",
542 if (entry->local_ring)
543 seq_puts(seq, "local\n");
546 seq_printf(seq, "%04X", ntohs(entry->rcf));
547 rcf_len = ((ntohs(entry->rcf) & TR_RCF_LEN_MASK)>>8)-2;
550 for(j = 1; j < rcf_len; j++) {
552 segment=ntohs(entry->rseg[j-1])>>4;
553 seq_printf(seq," %03X",segment);
556 segment=ntohs(entry->rseg[j])>>4;
557 brdgnmb=ntohs(entry->rseg[j-1])&0x00f;
558 seq_printf(seq,"-%01X-%03X",brdgnmb,segment);
567 static const struct seq_operations rif_seq_ops = {
568 .start = rif_seq_start,
569 .next = rif_seq_next,
570 .stop = rif_seq_stop,
571 .show = rif_seq_show,
574 static int rif_seq_open(struct inode *inode, struct file *file)
576 return seq_open(file, &rif_seq_ops);
579 static const struct file_operations rif_seq_fops = {
580 .owner = THIS_MODULE,
581 .open = rif_seq_open,
584 .release = seq_release,
589 static const struct header_ops tr_header_ops = {
591 .rebuild= tr_rebuild_header,
594 static void tr_setup(struct net_device *dev)
597 * Configure and register
600 dev->header_ops = &tr_header_ops;
602 dev->type = ARPHRD_IEEE802_TR;
603 dev->hard_header_len = TR_HLEN;
605 dev->addr_len = TR_ALEN;
606 dev->tx_queue_len = 100; /* Long queues on tr */
608 memset(dev->broadcast,0xFF, TR_ALEN);
610 /* New-style flags. */
611 dev->flags = IFF_BROADCAST | IFF_MULTICAST ;
615 * alloc_trdev - Register token ring device
616 * @sizeof_priv: Size of additional driver-private structure to be allocated
617 * for this token ring device
619 * Fill in the fields of the device structure with token ring-generic values.
621 * Constructs a new net device, complete with a private data area of
622 * size @sizeof_priv. A 32-byte (not bit) alignment is enforced for
623 * this private data area.
625 struct net_device *alloc_trdev(int sizeof_priv)
627 return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
631 static struct ctl_table tr_table[] = {
633 .ctl_name = NET_TR_RIF_TIMEOUT,
634 .procname = "rif_timeout",
635 .data = &sysctl_tr_rif_timeout,
636 .maxlen = sizeof(int),
638 .proc_handler = proc_dointvec
643 static __initdata struct ctl_path tr_path[] = {
644 { .procname = "net", .ctl_name = CTL_NET, },
645 { .procname = "token-ring", .ctl_name = NET_TR, },
651 * Called during bootup. We don't actually have to initialise
655 static int __init rif_init(void)
657 rif_timer.expires = jiffies + sysctl_tr_rif_timeout;
658 setup_timer(&rif_timer, rif_check_expire, 0);
659 add_timer(&rif_timer);
661 register_sysctl_paths(tr_path, tr_table);
663 proc_net_fops_create(&init_net, "tr_rif", S_IRUGO, &rif_seq_fops);
667 module_init(rif_init);
669 EXPORT_SYMBOL(tr_type_trans);
670 EXPORT_SYMBOL(alloc_trdev);