Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6.git] / net / 802 / tr.c
1 /*
2  * NET3:        Token ring device handling subroutines
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  * 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.
15  *
16  */
17
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>
25 #include <linux/mm.h>
26 #include <linux/socket.h>
27 #include <linux/in.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>
39 #include <net/arp.h>
40 #include <net/net_namespace.h>
41
42 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev);
43 static void rif_check_expire(unsigned long dummy);
44
45 #define TR_SR_DEBUG 0
46
47 /*
48  *      Each RIF entry we learn is kept this way
49  */
50
51 struct rif_cache {
52         unsigned char addr[TR_ALEN];
53         int iface;
54         __be16 rcf;
55         __be16 rseg[8];
56         struct rif_cache *next;
57         unsigned long last_used;
58         unsigned char local_ring;
59 };
60
61 #define RIF_TABLE_SIZE 32
62
63 /*
64  *      We hash the RIF cache 32 ways. We do after all have to look it
65  *      up a lot.
66  */
67
68 static struct rif_cache *rif_table[RIF_TABLE_SIZE];
69
70 static DEFINE_SPINLOCK(rif_lock);
71
72
73 /*
74  *      Garbage disposal timer.
75  */
76
77 static struct timer_list rif_timer;
78
79 static int sysctl_tr_rif_timeout = 60*10*HZ;
80
81 static inline unsigned long rif_hash(const unsigned char *addr)
82 {
83         unsigned long x;
84
85         x = addr[0];
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];
91
92         x ^= x >> 8;
93
94         return x & (RIF_TABLE_SIZE - 1);
95 }
96
97 /*
98  *      Put the headers on a token ring packet. Token ring source routing
99  *      makes this a little more exciting than on ethernet.
100  */
101
102 static int tr_header(struct sk_buff *skb, struct net_device *dev,
103                      unsigned short type,
104                      const void *daddr, const void *saddr, unsigned len)
105 {
106         struct trh_hdr *trh;
107         int hdr_len;
108
109         /*
110          * Add the 802.2 SNAP header if IP as the IPv4/IPv6 code calls
111          * dev->hard_header directly.
112          */
113         if (type == ETH_P_IP || type == ETH_P_IPV6 || type == ETH_P_ARP)
114         {
115                 struct trllc *trllc;
116
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;
121                 trllc->llc = UI_CMD;
122                 trllc->protid[0] = trllc->protid[1] = trllc->protid[2] = 0x00;
123                 trllc->ethertype = htons(type);
124         }
125         else
126         {
127                 hdr_len = sizeof(struct trh_hdr);
128                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
129         }
130
131         trh->ac=AC;
132         trh->fc=LLC_FRAME;
133
134         if(saddr)
135                 memcpy(trh->saddr,saddr,dev->addr_len);
136         else
137                 memcpy(trh->saddr,dev->dev_addr,dev->addr_len);
138
139         /*
140          *      Build the destination and then source route the frame
141          */
142
143         if(daddr)
144         {
145                 memcpy(trh->daddr,daddr,dev->addr_len);
146                 tr_source_route(skb, trh, dev);
147                 return(hdr_len);
148         }
149
150         return -hdr_len;
151 }
152
153 /*
154  *      A neighbour discovery of some species (eg arp) has completed. We
155  *      can now send the packet.
156  */
157
158 static int tr_rebuild_header(struct sk_buff *skb)
159 {
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;
163
164         /*
165          *      FIXME: We don't yet support IPv6 over token rings
166          */
167
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));
170                 return 0;
171         }
172
173 #ifdef CONFIG_INET
174         if(arp_find(trh->daddr, skb)) {
175                         return 1;
176         }
177         else
178 #endif
179         {
180                 tr_source_route(skb,trh,dev);
181                 return 0;
182         }
183 }
184
185 /*
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
188  *      it via SNAP.
189  */
190
191 __be16 tr_type_trans(struct sk_buff *skb, struct net_device *dev)
192 {
193
194         struct trh_hdr *trh;
195         struct trllc *trllc;
196         unsigned riflen=0;
197
198         skb->dev = dev;
199         skb_reset_mac_header(skb);
200         trh = tr_hdr(skb);
201
202         if(trh->saddr[0] & TR_RII)
203                 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
204
205         trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
206
207         skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
208
209         if(*trh->daddr & 0x80)
210         {
211                 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))
212                         skb->pkt_type=PACKET_BROADCAST;
213                 else
214                         skb->pkt_type=PACKET_MULTICAST;
215         }
216         else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
217         {
218                 skb->pkt_type=PACKET_MULTICAST;
219         }
220         else if(dev->flags & IFF_PROMISC)
221         {
222                 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
223                         skb->pkt_type=PACKET_OTHERHOST;
224         }
225
226         if ((skb->pkt_type != PACKET_BROADCAST) &&
227             (skb->pkt_type != PACKET_MULTICAST))
228                 tr_add_rif_info(trh,dev) ;
229
230         /*
231          * Strip the SNAP header from ARP packets since we don't
232          * pass them through to the 802.2/SNAP layers.
233          */
234
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)))
239         {
240                 skb_pull(skb, sizeof(struct trllc));
241                 return trllc->ethertype;
242         }
243
244         return htons(ETH_P_TR_802_2);
245 }
246
247 /*
248  *      We try to do source routing...
249  */
250
251 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,
252                      struct net_device *dev)
253 {
254         int slack;
255         unsigned int hash;
256         struct rif_cache *entry;
257         unsigned char *olddata;
258         unsigned long flags;
259         static const unsigned char mcast_func_addr[]
260                 = {0xC0,0x00,0x00,0x04,0x00,0x00};
261
262         spin_lock_irqsave(&rif_lock, flags);
263
264         /*
265          *      Broadcasts are single route as stated in RFC 1042
266          */
267         if( (!memcmp(&(trh->daddr[0]),&(dev->broadcast[0]),TR_ALEN)) ||
268             (!memcmp(&(trh->daddr[0]),&(mcast_func_addr[0]), TR_ALEN))  )
269         {
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;
273         }
274         else
275         {
276                 hash = rif_hash(trh->daddr);
277                 /*
278                  *      Walk the hash table and look for an entry
279                  */
280                 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->daddr[0]),TR_ALEN);entry=entry->next);
281
282                 /*
283                  *      If we found an entry we can route the frame.
284                  */
285                 if(entry)
286                 {
287 #if TR_SR_DEBUG
288 printk("source routing for %pM\n", trh->daddr);
289 #endif
290                         if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
291                         {
292                                 trh->rcf=entry->rcf;
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> */
296
297                                 trh->saddr[0]|=TR_RII;
298 #if TR_SR_DEBUG
299                                 printk("entry found with rcf %04x\n", entry->rcf);
300                         }
301                         else
302                         {
303                                 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
304 #endif
305                         }
306                         entry->last_used=jiffies;
307                 }
308                 else
309                 {
310                         /*
311                          *      Without the information we simply have to shout
312                          *      on the wire. The replies should rapidly clean this
313                          *      situation up.
314                          */
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;
318 #if TR_SR_DEBUG
319                         printk("no entry in rif table found - broadcasting frame\n");
320 #endif
321                 }
322         }
323
324         /* Compress the RIF here so we don't have to do it in the driver(s) */
325         if (!(trh->saddr[0] & 0x80))
326                 slack = 18;
327         else
328                 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
329         olddata = skb->data;
330         spin_unlock_irqrestore(&rif_lock, flags);
331
332         skb_pull(skb, slack);
333         memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
334 }
335
336 /*
337  *      We have learned some new RIF information for our source
338  *      routing.
339  */
340
341 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
342 {
343         unsigned int hash, rii_p = 0;
344         unsigned long flags;
345         struct rif_cache *entry;
346         unsigned char saddr0;
347
348         spin_lock_irqsave(&rif_lock, flags);
349         saddr0 = trh->saddr[0];
350
351         /*
352          *      Firstly see if the entry exists
353          */
354
355         if(trh->saddr[0] & TR_RII)
356         {
357                 trh->saddr[0]&=0x7f;
358                 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
359                 {
360                         rii_p = 1;
361                 }
362         }
363
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);
366
367         if(entry==NULL)
368         {
369 #if TR_SR_DEBUG
370                 printk("adding rif_entry: addr:%pM rcf:%04X\n",
371                        trh->saddr, ntohs(trh->rcf));
372 #endif
373                 /*
374                  *      Allocate our new entry. A failure to allocate loses
375                  *      use the information. This is harmless.
376                  *
377                  *      FIXME: We ought to keep some kind of cache size
378                  *      limiting and adjust the timers to suit.
379                  */
380                 entry=kmalloc(sizeof(struct rif_cache),GFP_ATOMIC);
381
382                 if(!entry)
383                 {
384                         printk(KERN_DEBUG "tr.c: Couldn't malloc rif cache entry !\n");
385                         spin_unlock_irqrestore(&rif_lock, flags);
386                         return;
387                 }
388
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;
394
395                 if (rii_p)
396                 {
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;
400                 }
401                 else
402                 {
403                         entry->local_ring = 1;
404                 }
405         }
406         else    /* Y. Tahara added */
407         {
408                 /*
409                  *      Update existing entries
410                  */
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)))
414                     {
415 #if TR_SR_DEBUG
416 printk("updating rif_entry: addr:%pM rcf:%04X\n",
417                 trh->saddr, ntohs(trh->rcf));
418 #endif
419                             entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
420                             memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
421                     }
422                 entry->last_used=jiffies;
423         }
424         trh->saddr[0]=saddr0; /* put the routing indicator back for tcpdump */
425         spin_unlock_irqrestore(&rif_lock, flags);
426 }
427
428 /*
429  *      Scan the cache with a timer and see what we need to throw out.
430  */
431
432 static void rif_check_expire(unsigned long dummy)
433 {
434         int i;
435         unsigned long flags, next_interval = jiffies + sysctl_tr_rif_timeout/2;
436
437         spin_lock_irqsave(&rif_lock, flags);
438
439         for(i =0; i < RIF_TABLE_SIZE; i++) {
440                 struct rif_cache *entry, **pentry;
441
442                 pentry = rif_table+i;
443                 while((entry=*pentry) != NULL) {
444                         unsigned long expires
445                                 = entry->last_used + sysctl_tr_rif_timeout;
446
447                         if (time_before_eq(expires, jiffies)) {
448                                 *pentry = entry->next;
449                                 kfree(entry);
450                         } else {
451                                 pentry = &entry->next;
452
453                                 if (time_before(expires, next_interval))
454                                         next_interval = expires;
455                         }
456                 }
457         }
458
459         spin_unlock_irqrestore(&rif_lock, flags);
460
461         mod_timer(&rif_timer, next_interval);
462
463 }
464
465 /*
466  *      Generate the /proc/net information for the token ring RIF
467  *      routing.
468  */
469
470 #ifdef CONFIG_PROC_FS
471
472 static struct rif_cache *rif_get_idx(loff_t pos)
473 {
474         int i;
475         struct rif_cache *entry;
476         loff_t off = 0;
477
478         for(i = 0; i < RIF_TABLE_SIZE; i++)
479                 for(entry = rif_table[i]; entry; entry = entry->next) {
480                         if (off == pos)
481                                 return entry;
482                         ++off;
483                 }
484
485         return NULL;
486 }
487
488 static void *rif_seq_start(struct seq_file *seq, loff_t *pos)
489         __acquires(&rif_lock)
490 {
491         spin_lock_irq(&rif_lock);
492
493         return *pos ? rif_get_idx(*pos - 1) : SEQ_START_TOKEN;
494 }
495
496 static void *rif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
497 {
498         int i;
499         struct rif_cache *ent = v;
500
501         ++*pos;
502
503         if (v == SEQ_START_TOKEN) {
504                 i = -1;
505                 goto scan;
506         }
507
508         if (ent->next)
509                 return ent->next;
510
511         i = rif_hash(ent->addr);
512  scan:
513         while (++i < RIF_TABLE_SIZE) {
514                 if ((ent = rif_table[i]) != NULL)
515                         return ent;
516         }
517         return NULL;
518 }
519
520 static void rif_seq_stop(struct seq_file *seq, void *v)
521         __releases(&rif_lock)
522 {
523         spin_unlock_irq(&rif_lock);
524 }
525
526 static int rif_seq_show(struct seq_file *seq, void *v)
527 {
528         int j, rcf_len, segment, brdgnmb;
529         struct rif_cache *entry = v;
530
531         if (v == SEQ_START_TOKEN)
532                 seq_puts(seq,
533                      "if     TR address       TTL   rcf   routing segments\n");
534         else {
535                 struct net_device *dev = dev_get_by_index(&init_net, entry->iface);
536                 long ttl = (long) (entry->last_used + sysctl_tr_rif_timeout)
537                                 - (long) jiffies;
538
539                 seq_printf(seq, "%s %pM %7li ",
540                            dev?dev->name:"?",
541                            entry->addr,
542                            ttl/HZ);
543
544                         if (entry->local_ring)
545                                 seq_puts(seq, "local\n");
546                         else {
547
548                                 seq_printf(seq, "%04X", ntohs(entry->rcf));
549                                 rcf_len = ((ntohs(entry->rcf) & TR_RCF_LEN_MASK)>>8)-2;
550                                 if (rcf_len)
551                                         rcf_len >>= 1;
552                                 for(j = 1; j < rcf_len; j++) {
553                                         if(j==1) {
554                                                 segment=ntohs(entry->rseg[j-1])>>4;
555                                                 seq_printf(seq,"  %03X",segment);
556                                         }
557
558                                         segment=ntohs(entry->rseg[j])>>4;
559                                         brdgnmb=ntohs(entry->rseg[j-1])&0x00f;
560                                         seq_printf(seq,"-%01X-%03X",brdgnmb,segment);
561                                 }
562                                 seq_putc(seq, '\n');
563                         }
564
565                 if (dev)
566                         dev_put(dev);
567                 }
568         return 0;
569 }
570
571
572 static const struct seq_operations rif_seq_ops = {
573         .start = rif_seq_start,
574         .next  = rif_seq_next,
575         .stop  = rif_seq_stop,
576         .show  = rif_seq_show,
577 };
578
579 static int rif_seq_open(struct inode *inode, struct file *file)
580 {
581         return seq_open(file, &rif_seq_ops);
582 }
583
584 static const struct file_operations rif_seq_fops = {
585         .owner   = THIS_MODULE,
586         .open    = rif_seq_open,
587         .read    = seq_read,
588         .llseek  = seq_lseek,
589         .release = seq_release,
590 };
591
592 #endif
593
594 static const struct header_ops tr_header_ops = {
595         .create = tr_header,
596         .rebuild= tr_rebuild_header,
597 };
598
599 static void tr_setup(struct net_device *dev)
600 {
601         /*
602          *      Configure and register
603          */
604
605         dev->header_ops = &tr_header_ops;
606
607         dev->type               = ARPHRD_IEEE802_TR;
608         dev->hard_header_len    = TR_HLEN;
609         dev->mtu                = 2000;
610         dev->addr_len           = TR_ALEN;
611         dev->tx_queue_len       = 100;  /* Long queues on tr */
612
613         memset(dev->broadcast,0xFF, TR_ALEN);
614
615         /* New-style flags. */
616         dev->flags              = IFF_BROADCAST | IFF_MULTICAST ;
617 }
618
619 /**
620  * alloc_trdev - Register token ring device
621  * @sizeof_priv: Size of additional driver-private structure to be allocated
622  *      for this token ring device
623  *
624  * Fill in the fields of the device structure with token ring-generic values.
625  *
626  * Constructs a new net device, complete with a private data area of
627  * size @sizeof_priv.  A 32-byte (not bit) alignment is enforced for
628  * this private data area.
629  */
630 struct net_device *alloc_trdev(int sizeof_priv)
631 {
632         return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
633 }
634
635 #ifdef CONFIG_SYSCTL
636 static struct ctl_table tr_table[] = {
637         {
638                 .procname       = "rif_timeout",
639                 .data           = &sysctl_tr_rif_timeout,
640                 .maxlen         = sizeof(int),
641                 .mode           = 0644,
642                 .proc_handler   = proc_dointvec
643         },
644         { },
645 };
646
647 static __initdata struct ctl_path tr_path[] = {
648         { .procname = "net", },
649         { .procname = "token-ring", },
650         { }
651 };
652 #endif
653
654 /*
655  *      Called during bootup.  We don't actually have to initialise
656  *      too much for this.
657  */
658
659 static int __init rif_init(void)
660 {
661         rif_timer.expires  = jiffies + sysctl_tr_rif_timeout;
662         setup_timer(&rif_timer, rif_check_expire, 0);
663         add_timer(&rif_timer);
664 #ifdef CONFIG_SYSCTL
665         register_sysctl_paths(tr_path, tr_table);
666 #endif
667         proc_net_fops_create(&init_net, "tr_rif", S_IRUGO, &rif_seq_fops);
668         return 0;
669 }
670
671 module_init(rif_init);
672
673 EXPORT_SYMBOL(tr_type_trans);
674 EXPORT_SYMBOL(alloc_trdev);
675
676 MODULE_LICENSE("GPL");