[PATCH] Lost sockfd_put() in routing_ioctl()
[linux-2.6] / 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/config.h>
21 #include <linux/module.h>
22 #include <linux/types.h>
23 #include <linux/kernel.h>
24 #include <linux/jiffies.h>
25 #include <linux/string.h>
26 #include <linux/mm.h>
27 #include <linux/socket.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/trdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/errno.h>
34 #include <linux/timer.h>
35 #include <linux/net.h>
36 #include <linux/proc_fs.h>
37 #include <linux/seq_file.h>
38 #include <linux/init.h>
39 #include <net/arp.h>
40
41 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev);
42 static void rif_check_expire(unsigned long dummy);
43
44 #define TR_SR_DEBUG 0
45
46 /*
47  *      Each RIF entry we learn is kept this way
48  */
49  
50 struct rif_cache {
51         unsigned char addr[TR_ALEN];
52         int iface;
53         __be16 rcf;
54         __be16 rseg[8];
55         struct rif_cache *next;
56         unsigned long last_used;
57         unsigned char local_ring;
58 };
59
60 #define RIF_TABLE_SIZE 32
61
62 /*
63  *      We hash the RIF cache 32 ways. We do after all have to look it
64  *      up a lot.
65  */
66  
67 static struct rif_cache *rif_table[RIF_TABLE_SIZE];
68
69 static DEFINE_SPINLOCK(rif_lock);
70
71
72 /*
73  *      Garbage disposal timer.
74  */
75  
76 static struct timer_list rif_timer;
77
78 int sysctl_tr_rif_timeout = 60*10*HZ;
79
80 static inline unsigned long rif_hash(const unsigned char *addr)
81 {
82         unsigned long x;
83
84         x = addr[0];
85         x = (x << 2) ^ addr[1];
86         x = (x << 2) ^ addr[2];
87         x = (x << 2) ^ addr[3];
88         x = (x << 2) ^ addr[4];
89         x = (x << 2) ^ addr[5];
90
91         x ^= x >> 8;
92
93         return x & (RIF_TABLE_SIZE - 1);
94 }
95
96 /*
97  *      Put the headers on a token ring packet. Token ring source routing
98  *      makes this a little more exciting than on ethernet.
99  */
100  
101 static int tr_header(struct sk_buff *skb, struct net_device *dev,
102                      unsigned short type,
103                      void *daddr, void *saddr, unsigned len) 
104 {
105         struct trh_hdr *trh;
106         int hdr_len;
107
108         /* 
109          * Add the 802.2 SNAP header if IP as the IPv4/IPv6 code calls  
110          * dev->hard_header directly.
111          */
112         if (type == ETH_P_IP || type == ETH_P_IPV6 || type == ETH_P_ARP)
113         {
114                 struct trllc *trllc;
115
116                 hdr_len = sizeof(struct trh_hdr) + sizeof(struct trllc);
117                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
118                 trllc = (struct trllc *)(trh+1);
119                 trllc->dsap = trllc->ssap = EXTENDED_SAP;
120                 trllc->llc = UI_CMD;
121                 trllc->protid[0] = trllc->protid[1] = trllc->protid[2] = 0x00;
122                 trllc->ethertype = htons(type);
123         }
124         else
125         {
126                 hdr_len = sizeof(struct trh_hdr);
127                 trh = (struct trh_hdr *)skb_push(skb, hdr_len); 
128         }
129
130         trh->ac=AC;
131         trh->fc=LLC_FRAME;
132
133         if(saddr)
134                 memcpy(trh->saddr,saddr,dev->addr_len);
135         else
136                 memcpy(trh->saddr,dev->dev_addr,dev->addr_len);
137
138         /*
139          *      Build the destination and then source route the frame
140          */
141          
142         if(daddr) 
143         {
144                 memcpy(trh->daddr,daddr,dev->addr_len);
145                 tr_source_route(skb,trh,dev);
146                 return(hdr_len);
147         }
148
149         return -hdr_len;
150 }
151         
152 /*
153  *      A neighbour discovery of some species (eg arp) has completed. We
154  *      can now send the packet.
155  */
156  
157 static int tr_rebuild_header(struct sk_buff *skb) 
158 {
159         struct trh_hdr *trh=(struct trh_hdr *)skb->data;
160         struct trllc *trllc=(struct trllc *)(skb->data+sizeof(struct trh_hdr));
161         struct net_device *dev = skb->dev;
162
163         /*
164          *      FIXME: We don't yet support IPv6 over token rings
165          */
166          
167         if(trllc->ethertype != htons(ETH_P_IP)) {
168                 printk("tr_rebuild_header: Don't know how to resolve type %04X addresses ?\n",(unsigned int)htons(trllc->ethertype));
169                 return 0;
170         }
171
172 #ifdef CONFIG_INET
173         if(arp_find(trh->daddr, skb)) {
174                         return 1;
175         }
176         else 
177 #endif  
178         {       
179                 tr_source_route(skb,trh,dev); 
180                 return 0;
181         }
182 }
183         
184 /*
185  *      Some of this is a bit hackish. We intercept RIF information
186  *      used for source routing. We also grab IP directly and don't feed
187  *      it via SNAP.
188  */
189  
190 unsigned short tr_type_trans(struct sk_buff *skb, struct net_device *dev) 
191 {
192
193         struct trh_hdr *trh=(struct trh_hdr *)skb->data;
194         struct trllc *trllc;
195         unsigned riflen=0;
196         
197         skb->mac.raw = skb->data;
198         
199         if(trh->saddr[0] & TR_RII)
200                 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
201
202         trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
203
204         skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
205
206         if(*trh->daddr & 0x80) 
207         {
208                 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))  
209                         skb->pkt_type=PACKET_BROADCAST;
210                 else
211                         skb->pkt_type=PACKET_MULTICAST;
212         }
213         else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
214         {
215                 skb->pkt_type=PACKET_MULTICAST;
216         }
217         else if(dev->flags & IFF_PROMISC) 
218         {
219                 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
220                         skb->pkt_type=PACKET_OTHERHOST;
221         }
222
223         if ((skb->pkt_type != PACKET_BROADCAST) &&
224             (skb->pkt_type != PACKET_MULTICAST))
225                 tr_add_rif_info(trh,dev) ; 
226
227         /*
228          * Strip the SNAP header from ARP packets since we don't 
229          * pass them through to the 802.2/SNAP layers.
230          */
231
232         if (trllc->dsap == EXTENDED_SAP &&
233             (trllc->ethertype == ntohs(ETH_P_IP) ||
234              trllc->ethertype == ntohs(ETH_P_IPV6) ||
235              trllc->ethertype == ntohs(ETH_P_ARP)))
236         {
237                 skb_pull(skb, sizeof(struct trllc));
238                 return trllc->ethertype;
239         }
240
241         return ntohs(ETH_P_802_2);
242 }
243
244 /*
245  *      We try to do source routing... 
246  */
247
248 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,struct net_device *dev) 
249 {
250         int slack;
251         unsigned int hash;
252         struct rif_cache *entry;
253         unsigned char *olddata;
254         unsigned long flags;
255         static const unsigned char mcast_func_addr[] 
256                 = {0xC0,0x00,0x00,0x04,0x00,0x00};
257         
258         spin_lock_irqsave(&rif_lock, flags);
259
260         /*
261          *      Broadcasts are single route as stated in RFC 1042 
262          */
263         if( (!memcmp(&(trh->daddr[0]),&(dev->broadcast[0]),TR_ALEN)) ||
264             (!memcmp(&(trh->daddr[0]),&(mcast_func_addr[0]), TR_ALEN))  )
265         {
266                 trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)  
267                                | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
268                 trh->saddr[0]|=TR_RII;
269         }
270         else 
271         {
272                 hash = rif_hash(trh->daddr);
273                 /*
274                  *      Walk the hash table and look for an entry
275                  */
276                 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->daddr[0]),TR_ALEN);entry=entry->next);
277
278                 /*
279                  *      If we found an entry we can route the frame.
280                  */
281                 if(entry) 
282                 {
283 #if TR_SR_DEBUG
284 printk("source routing for %02X:%02X:%02X:%02X:%02X:%02X\n",trh->daddr[0],
285                   trh->daddr[1],trh->daddr[2],trh->daddr[3],trh->daddr[4],trh->daddr[5]);
286 #endif
287                         if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
288                         {
289                                 trh->rcf=entry->rcf;
290                                 memcpy(&trh->rseg[0],&entry->rseg[0],8*sizeof(unsigned short));
291                                 trh->rcf^=htons(TR_RCF_DIR_BIT);        
292                                 trh->rcf&=htons(0x1fff);        /* Issam Chehab <ichehab@madge1.demon.co.uk> */
293
294                                 trh->saddr[0]|=TR_RII;
295 #if TR_SR_DEBUG
296                                 printk("entry found with rcf %04x\n", entry->rcf);
297                         }
298                         else
299                         {
300                                 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
301 #endif
302                         }
303                         entry->last_used=jiffies;
304                 }
305                 else 
306                 {
307                         /*
308                          *      Without the information we simply have to shout
309                          *      on the wire. The replies should rapidly clean this
310                          *      situation up.
311                          */
312                         trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)  
313                                        | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
314                         trh->saddr[0]|=TR_RII;
315 #if TR_SR_DEBUG
316                         printk("no entry in rif table found - broadcasting frame\n");
317 #endif
318                 }
319         }
320
321         /* Compress the RIF here so we don't have to do it in the driver(s) */
322         if (!(trh->saddr[0] & 0x80))
323                 slack = 18;
324         else 
325                 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
326         olddata = skb->data;
327         spin_unlock_irqrestore(&rif_lock, flags);
328
329         skb_pull(skb, slack);
330         memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
331 }
332
333 /*
334  *      We have learned some new RIF information for our source
335  *      routing.
336  */
337  
338 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
339 {
340         unsigned int hash, rii_p = 0;
341         unsigned long flags;
342         struct rif_cache *entry;
343
344
345         spin_lock_irqsave(&rif_lock, flags);
346         
347         /*
348          *      Firstly see if the entry exists
349          */
350
351         if(trh->saddr[0] & TR_RII)
352         {
353                 trh->saddr[0]&=0x7f;
354                 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
355                 {
356                         rii_p = 1;
357                 }
358         }
359
360         hash = rif_hash(trh->saddr);
361         for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);entry=entry->next);
362
363         if(entry==NULL) 
364         {
365 #if TR_SR_DEBUG
366 printk("adding rif_entry: addr:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
367                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
368                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
369                 ntohs(trh->rcf));
370 #endif
371                 /*
372                  *      Allocate our new entry. A failure to allocate loses
373                  *      use the information. This is harmless.
374                  *
375                  *      FIXME: We ought to keep some kind of cache size
376                  *      limiting and adjust the timers to suit.
377                  */
378                 entry=kmalloc(sizeof(struct rif_cache),GFP_ATOMIC);
379
380                 if(!entry) 
381                 {
382                         printk(KERN_DEBUG "tr.c: Couldn't malloc rif cache entry !\n");
383                         spin_unlock_irqrestore(&rif_lock, flags);
384                         return;
385                 }
386
387                 memcpy(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);
388                 entry->iface = dev->ifindex;
389                 entry->next=rif_table[hash];
390                 entry->last_used=jiffies;
391                 rif_table[hash]=entry;
392
393                 if (rii_p)
394                 {
395                         entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
396                         memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
397                         entry->local_ring = 0;
398                         trh->saddr[0]|=TR_RII; /* put the routing indicator back for tcpdump */
399                 }
400                 else
401                 {
402                         entry->local_ring = 1;
403                 }
404         }       
405         else    /* Y. Tahara added */
406         { 
407                 /*
408                  *      Update existing entries
409                  */
410                 if (!entry->local_ring) 
411                     if (entry->rcf != (trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK)) &&
412                          !(trh->rcf & htons(TR_RCF_BROADCAST_MASK)))
413                     {
414 #if TR_SR_DEBUG
415 printk("updating rif_entry: addr:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
416                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
417                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
418                 ntohs(trh->rcf));
419 #endif
420                             entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
421                             memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
422                     }                                         
423                 entry->last_used=jiffies;               
424         }
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 {
490         spin_lock_irq(&rif_lock);
491
492         return *pos ? rif_get_idx(*pos - 1) : SEQ_START_TOKEN;
493 }
494
495 static void *rif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
496 {
497         int i;
498         struct rif_cache *ent = v;
499
500         ++*pos;
501
502         if (v == SEQ_START_TOKEN) {
503                 i = -1;
504                 goto scan;
505         }
506
507         if (ent->next) 
508                 return ent->next;
509
510         i = rif_hash(ent->addr);
511  scan:
512         while (++i < RIF_TABLE_SIZE) {
513                 if ((ent = rif_table[i]) != NULL)
514                         return ent;
515         }
516         return NULL;
517 }
518
519 static void rif_seq_stop(struct seq_file *seq, void *v)
520 {
521         spin_unlock_irq(&rif_lock);
522 }
523
524 static int rif_seq_show(struct seq_file *seq, void *v)
525 {
526         int j, rcf_len, segment, brdgnmb;
527         struct rif_cache *entry = v;
528
529         if (v == SEQ_START_TOKEN)
530                 seq_puts(seq,
531                      "if     TR address       TTL   rcf   routing segments\n");
532         else {
533                 struct net_device *dev = dev_get_by_index(entry->iface);
534                 long ttl = (long) (entry->last_used + sysctl_tr_rif_timeout)
535                                 - (long) jiffies;
536
537                 seq_printf(seq, "%s %02X:%02X:%02X:%02X:%02X:%02X %7li ",
538                            dev?dev->name:"?",
539                            entry->addr[0],entry->addr[1],entry->addr[2],
540                            entry->addr[3],entry->addr[4],entry->addr[5],
541                            ttl/HZ);
542
543                         if (entry->local_ring)
544                                 seq_puts(seq, "local\n");
545                         else {
546
547                                 seq_printf(seq, "%04X", ntohs(entry->rcf));
548                                 rcf_len = ((ntohs(entry->rcf) & TR_RCF_LEN_MASK)>>8)-2; 
549                                 if (rcf_len)
550                                         rcf_len >>= 1;
551                                 for(j = 1; j < rcf_len; j++) {
552                                         if(j==1) {
553                                                 segment=ntohs(entry->rseg[j-1])>>4;
554                                                 seq_printf(seq,"  %03X",segment);
555                                         };
556                                         segment=ntohs(entry->rseg[j])>>4;
557                                         brdgnmb=ntohs(entry->rseg[j-1])&0x00f;
558                                         seq_printf(seq,"-%01X-%03X",brdgnmb,segment);
559                                 }
560                                 seq_putc(seq, '\n');
561                         }
562                 }
563         return 0;
564 }
565
566
567 static 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,
572 };
573
574 static int rif_seq_open(struct inode *inode, struct file *file)
575 {
576         return seq_open(file, &rif_seq_ops);
577 }
578
579 static struct file_operations rif_seq_fops = {
580         .owner   = THIS_MODULE,
581         .open    = rif_seq_open,
582         .read    = seq_read,
583         .llseek  = seq_lseek,
584         .release = seq_release,
585 };
586
587 #endif
588
589 static void tr_setup(struct net_device *dev)
590 {
591         /*
592          *      Configure and register
593          */
594         
595         dev->hard_header        = tr_header;
596         dev->rebuild_header     = tr_rebuild_header;
597
598         dev->type               = ARPHRD_IEEE802_TR;
599         dev->hard_header_len    = TR_HLEN;
600         dev->mtu                = 2000;
601         dev->addr_len           = TR_ALEN;
602         dev->tx_queue_len       = 100;  /* Long queues on tr */
603         
604         memset(dev->broadcast,0xFF, TR_ALEN);
605
606         /* New-style flags. */
607         dev->flags              = IFF_BROADCAST | IFF_MULTICAST ;
608 }
609
610 /**
611  * alloc_trdev - Register token ring device
612  * @sizeof_priv: Size of additional driver-private structure to be allocated
613  *      for this token ring device
614  *
615  * Fill in the fields of the device structure with token ring-generic values.
616  *
617  * Constructs a new net device, complete with a private data area of
618  * size @sizeof_priv.  A 32-byte (not bit) alignment is enforced for
619  * this private data area.
620  */
621 struct net_device *alloc_trdev(int sizeof_priv)
622 {
623         return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
624 }
625
626 /*
627  *      Called during bootup.  We don't actually have to initialise
628  *      too much for this.
629  */
630
631 static int __init rif_init(void)
632 {
633         init_timer(&rif_timer);
634         rif_timer.expires  = sysctl_tr_rif_timeout;
635         rif_timer.data     = 0L;
636         rif_timer.function = rif_check_expire;
637         add_timer(&rif_timer);
638
639         proc_net_fops_create("tr_rif", S_IRUGO, &rif_seq_fops);
640         return 0;
641 }
642
643 module_init(rif_init);
644
645 EXPORT_SYMBOL(tr_source_route);
646 EXPORT_SYMBOL(tr_type_trans);
647 EXPORT_SYMBOL(alloc_trdev);