[SK_BUFF]: Introduce skb_reset_mac_header(skb)
[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 <net/arp.h>
39
40 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev);
41 static void rif_check_expire(unsigned long dummy);
42
43 #define TR_SR_DEBUG 0
44
45 /*
46  *      Each RIF entry we learn is kept this way
47  */
48
49 struct rif_cache {
50         unsigned char addr[TR_ALEN];
51         int iface;
52         __be16 rcf;
53         __be16 rseg[8];
54         struct rif_cache *next;
55         unsigned long last_used;
56         unsigned char local_ring;
57 };
58
59 #define RIF_TABLE_SIZE 32
60
61 /*
62  *      We hash the RIF cache 32 ways. We do after all have to look it
63  *      up a lot.
64  */
65
66 static struct rif_cache *rif_table[RIF_TABLE_SIZE];
67
68 static DEFINE_SPINLOCK(rif_lock);
69
70
71 /*
72  *      Garbage disposal timer.
73  */
74
75 static struct timer_list rif_timer;
76
77 int sysctl_tr_rif_timeout = 60*10*HZ;
78
79 static inline unsigned long rif_hash(const unsigned char *addr)
80 {
81         unsigned long x;
82
83         x = addr[0];
84         x = (x << 2) ^ addr[1];
85         x = (x << 2) ^ addr[2];
86         x = (x << 2) ^ addr[3];
87         x = (x << 2) ^ addr[4];
88         x = (x << 2) ^ addr[5];
89
90         x ^= x >> 8;
91
92         return x & (RIF_TABLE_SIZE - 1);
93 }
94
95 /*
96  *      Put the headers on a token ring packet. Token ring source routing
97  *      makes this a little more exciting than on ethernet.
98  */
99
100 static int tr_header(struct sk_buff *skb, struct net_device *dev,
101                      unsigned short type,
102                      void *daddr, void *saddr, unsigned len)
103 {
104         struct trh_hdr *trh;
105         int hdr_len;
106
107         /*
108          * Add the 802.2 SNAP header if IP as the IPv4/IPv6 code calls
109          * dev->hard_header directly.
110          */
111         if (type == ETH_P_IP || type == ETH_P_IPV6 || type == ETH_P_ARP)
112         {
113                 struct trllc *trllc;
114
115                 hdr_len = sizeof(struct trh_hdr) + sizeof(struct trllc);
116                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
117                 trllc = (struct trllc *)(trh+1);
118                 trllc->dsap = trllc->ssap = EXTENDED_SAP;
119                 trllc->llc = UI_CMD;
120                 trllc->protid[0] = trllc->protid[1] = trllc->protid[2] = 0x00;
121                 trllc->ethertype = htons(type);
122         }
123         else
124         {
125                 hdr_len = sizeof(struct trh_hdr);
126                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
127         }
128
129         trh->ac=AC;
130         trh->fc=LLC_FRAME;
131
132         if(saddr)
133                 memcpy(trh->saddr,saddr,dev->addr_len);
134         else
135                 memcpy(trh->saddr,dev->dev_addr,dev->addr_len);
136
137         /*
138          *      Build the destination and then source route the frame
139          */
140
141         if(daddr)
142         {
143                 memcpy(trh->daddr,daddr,dev->addr_len);
144                 tr_source_route(skb,trh,dev);
145                 return(hdr_len);
146         }
147
148         return -hdr_len;
149 }
150
151 /*
152  *      A neighbour discovery of some species (eg arp) has completed. We
153  *      can now send the packet.
154  */
155
156 static int tr_rebuild_header(struct sk_buff *skb)
157 {
158         struct trh_hdr *trh=(struct trh_hdr *)skb->data;
159         struct trllc *trllc=(struct trllc *)(skb->data+sizeof(struct trh_hdr));
160         struct net_device *dev = skb->dev;
161
162         /*
163          *      FIXME: We don't yet support IPv6 over token rings
164          */
165
166         if(trllc->ethertype != htons(ETH_P_IP)) {
167                 printk("tr_rebuild_header: Don't know how to resolve type %04X addresses ?\n", ntohs(trllc->ethertype));
168                 return 0;
169         }
170
171 #ifdef CONFIG_INET
172         if(arp_find(trh->daddr, skb)) {
173                         return 1;
174         }
175         else
176 #endif
177         {
178                 tr_source_route(skb,trh,dev);
179                 return 0;
180         }
181 }
182
183 /*
184  *      Some of this is a bit hackish. We intercept RIF information
185  *      used for source routing. We also grab IP directly and don't feed
186  *      it via SNAP.
187  */
188
189 __be16 tr_type_trans(struct sk_buff *skb, struct net_device *dev)
190 {
191
192         struct trh_hdr *trh;
193         struct trllc *trllc;
194         unsigned riflen=0;
195
196         skb->dev = dev;
197         skb_reset_mac_header(skb);
198         trh = tr_hdr(skb);
199
200         if(trh->saddr[0] & TR_RII)
201                 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
202
203         trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
204
205         skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
206
207         if(*trh->daddr & 0x80)
208         {
209                 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))
210                         skb->pkt_type=PACKET_BROADCAST;
211                 else
212                         skb->pkt_type=PACKET_MULTICAST;
213         }
214         else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
215         {
216                 skb->pkt_type=PACKET_MULTICAST;
217         }
218         else if(dev->flags & IFF_PROMISC)
219         {
220                 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
221                         skb->pkt_type=PACKET_OTHERHOST;
222         }
223
224         if ((skb->pkt_type != PACKET_BROADCAST) &&
225             (skb->pkt_type != PACKET_MULTICAST))
226                 tr_add_rif_info(trh,dev) ;
227
228         /*
229          * Strip the SNAP header from ARP packets since we don't
230          * pass them through to the 802.2/SNAP layers.
231          */
232
233         if (trllc->dsap == EXTENDED_SAP &&
234             (trllc->ethertype == htons(ETH_P_IP) ||
235              trllc->ethertype == htons(ETH_P_IPV6) ||
236              trllc->ethertype == htons(ETH_P_ARP)))
237         {
238                 skb_pull(skb, sizeof(struct trllc));
239                 return trllc->ethertype;
240         }
241
242         return htons(ETH_P_TR_802_2);
243 }
244
245 /*
246  *      We try to do source routing...
247  */
248
249 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,struct net_device *dev)
250 {
251         int slack;
252         unsigned int hash;
253         struct rif_cache *entry;
254         unsigned char *olddata;
255         unsigned long flags;
256         static const unsigned char mcast_func_addr[]
257                 = {0xC0,0x00,0x00,0x04,0x00,0x00};
258
259         spin_lock_irqsave(&rif_lock, flags);
260
261         /*
262          *      Broadcasts are single route as stated in RFC 1042
263          */
264         if( (!memcmp(&(trh->daddr[0]),&(dev->broadcast[0]),TR_ALEN)) ||
265             (!memcmp(&(trh->daddr[0]),&(mcast_func_addr[0]), TR_ALEN))  )
266         {
267                 trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
268                                | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
269                 trh->saddr[0]|=TR_RII;
270         }
271         else
272         {
273                 hash = rif_hash(trh->daddr);
274                 /*
275                  *      Walk the hash table and look for an entry
276                  */
277                 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->daddr[0]),TR_ALEN);entry=entry->next);
278
279                 /*
280                  *      If we found an entry we can route the frame.
281                  */
282                 if(entry)
283                 {
284 #if TR_SR_DEBUG
285 printk("source routing for %02X:%02X:%02X:%02X:%02X:%02X\n",trh->daddr[0],
286                   trh->daddr[1],trh->daddr[2],trh->daddr[3],trh->daddr[4],trh->daddr[5]);
287 #endif
288                         if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
289                         {
290                                 trh->rcf=entry->rcf;
291                                 memcpy(&trh->rseg[0],&entry->rseg[0],8*sizeof(unsigned short));
292                                 trh->rcf^=htons(TR_RCF_DIR_BIT);
293                                 trh->rcf&=htons(0x1fff);        /* Issam Chehab <ichehab@madge1.demon.co.uk> */
294
295                                 trh->saddr[0]|=TR_RII;
296 #if TR_SR_DEBUG
297                                 printk("entry found with rcf %04x\n", entry->rcf);
298                         }
299                         else
300                         {
301                                 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
302 #endif
303                         }
304                         entry->last_used=jiffies;
305                 }
306                 else
307                 {
308                         /*
309                          *      Without the information we simply have to shout
310                          *      on the wire. The replies should rapidly clean this
311                          *      situation up.
312                          */
313                         trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
314                                        | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
315                         trh->saddr[0]|=TR_RII;
316 #if TR_SR_DEBUG
317                         printk("no entry in rif table found - broadcasting frame\n");
318 #endif
319                 }
320         }
321
322         /* Compress the RIF here so we don't have to do it in the driver(s) */
323         if (!(trh->saddr[0] & 0x80))
324                 slack = 18;
325         else
326                 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
327         olddata = skb->data;
328         spin_unlock_irqrestore(&rif_lock, flags);
329
330         skb_pull(skb, slack);
331         memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
332 }
333
334 /*
335  *      We have learned some new RIF information for our source
336  *      routing.
337  */
338
339 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
340 {
341         unsigned int hash, rii_p = 0;
342         unsigned long flags;
343         struct rif_cache *entry;
344         unsigned char saddr0;
345
346         spin_lock_irqsave(&rif_lock, flags);
347         saddr0 = trh->saddr[0];
348
349         /*
350          *      Firstly see if the entry exists
351          */
352
353         if(trh->saddr[0] & TR_RII)
354         {
355                 trh->saddr[0]&=0x7f;
356                 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
357                 {
358                         rii_p = 1;
359                 }
360         }
361
362         hash = rif_hash(trh->saddr);
363         for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);entry=entry->next);
364
365         if(entry==NULL)
366         {
367 #if TR_SR_DEBUG
368 printk("adding rif_entry: addr:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
369                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
370                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
371                 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:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
417                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
418                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
419                 ntohs(trh->rcf));
420 #endif
421                             entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
422                             memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
423                     }
424                 entry->last_used=jiffies;
425         }
426         trh->saddr[0]=saddr0; /* put the routing indicator back for tcpdump */
427         spin_unlock_irqrestore(&rif_lock, flags);
428 }
429
430 /*
431  *      Scan the cache with a timer and see what we need to throw out.
432  */
433
434 static void rif_check_expire(unsigned long dummy)
435 {
436         int i;
437         unsigned long flags, next_interval = jiffies + sysctl_tr_rif_timeout/2;
438
439         spin_lock_irqsave(&rif_lock, flags);
440
441         for(i =0; i < RIF_TABLE_SIZE; i++) {
442                 struct rif_cache *entry, **pentry;
443
444                 pentry = rif_table+i;
445                 while((entry=*pentry) != NULL) {
446                         unsigned long expires
447                                 = entry->last_used + sysctl_tr_rif_timeout;
448
449                         if (time_before_eq(expires, jiffies)) {
450                                 *pentry = entry->next;
451                                 kfree(entry);
452                         } else {
453                                 pentry = &entry->next;
454
455                                 if (time_before(expires, next_interval))
456                                         next_interval = expires;
457                         }
458                 }
459         }
460
461         spin_unlock_irqrestore(&rif_lock, flags);
462
463         mod_timer(&rif_timer, next_interval);
464
465 }
466
467 /*
468  *      Generate the /proc/net information for the token ring RIF
469  *      routing.
470  */
471
472 #ifdef CONFIG_PROC_FS
473
474 static struct rif_cache *rif_get_idx(loff_t pos)
475 {
476         int i;
477         struct rif_cache *entry;
478         loff_t off = 0;
479
480         for(i = 0; i < RIF_TABLE_SIZE; i++)
481                 for(entry = rif_table[i]; entry; entry = entry->next) {
482                         if (off == pos)
483                                 return entry;
484                         ++off;
485                 }
486
487         return NULL;
488 }
489
490 static void *rif_seq_start(struct seq_file *seq, loff_t *pos)
491 {
492         spin_lock_irq(&rif_lock);
493
494         return *pos ? rif_get_idx(*pos - 1) : SEQ_START_TOKEN;
495 }
496
497 static void *rif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
498 {
499         int i;
500         struct rif_cache *ent = v;
501
502         ++*pos;
503
504         if (v == SEQ_START_TOKEN) {
505                 i = -1;
506                 goto scan;
507         }
508
509         if (ent->next)
510                 return ent->next;
511
512         i = rif_hash(ent->addr);
513  scan:
514         while (++i < RIF_TABLE_SIZE) {
515                 if ((ent = rif_table[i]) != NULL)
516                         return ent;
517         }
518         return NULL;
519 }
520
521 static void rif_seq_stop(struct seq_file *seq, void *v)
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(entry->iface);
536                 long ttl = (long) (entry->last_used + sysctl_tr_rif_timeout)
537                                 - (long) jiffies;
538
539                 seq_printf(seq, "%s %02X:%02X:%02X:%02X:%02X:%02X %7li ",
540                            dev?dev->name:"?",
541                            entry->addr[0],entry->addr[1],entry->addr[2],
542                            entry->addr[3],entry->addr[4],entry->addr[5],
543                            ttl/HZ);
544
545                         if (entry->local_ring)
546                                 seq_puts(seq, "local\n");
547                         else {
548
549                                 seq_printf(seq, "%04X", ntohs(entry->rcf));
550                                 rcf_len = ((ntohs(entry->rcf) & TR_RCF_LEN_MASK)>>8)-2;
551                                 if (rcf_len)
552                                         rcf_len >>= 1;
553                                 for(j = 1; j < rcf_len; j++) {
554                                         if(j==1) {
555                                                 segment=ntohs(entry->rseg[j-1])>>4;
556                                                 seq_printf(seq,"  %03X",segment);
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         return 0;
566 }
567
568
569 static struct seq_operations rif_seq_ops = {
570         .start = rif_seq_start,
571         .next  = rif_seq_next,
572         .stop  = rif_seq_stop,
573         .show  = rif_seq_show,
574 };
575
576 static int rif_seq_open(struct inode *inode, struct file *file)
577 {
578         return seq_open(file, &rif_seq_ops);
579 }
580
581 static const struct file_operations rif_seq_fops = {
582         .owner   = THIS_MODULE,
583         .open    = rif_seq_open,
584         .read    = seq_read,
585         .llseek  = seq_lseek,
586         .release = seq_release,
587 };
588
589 #endif
590
591 static void tr_setup(struct net_device *dev)
592 {
593         /*
594          *      Configure and register
595          */
596
597         dev->hard_header        = tr_header;
598         dev->rebuild_header     = tr_rebuild_header;
599
600         dev->type               = ARPHRD_IEEE802_TR;
601         dev->hard_header_len    = TR_HLEN;
602         dev->mtu                = 2000;
603         dev->addr_len           = TR_ALEN;
604         dev->tx_queue_len       = 100;  /* Long queues on tr */
605
606         memset(dev->broadcast,0xFF, TR_ALEN);
607
608         /* New-style flags. */
609         dev->flags              = IFF_BROADCAST | IFF_MULTICAST ;
610 }
611
612 /**
613  * alloc_trdev - Register token ring device
614  * @sizeof_priv: Size of additional driver-private structure to be allocated
615  *      for this token ring device
616  *
617  * Fill in the fields of the device structure with token ring-generic values.
618  *
619  * Constructs a new net device, complete with a private data area of
620  * size @sizeof_priv.  A 32-byte (not bit) alignment is enforced for
621  * this private data area.
622  */
623 struct net_device *alloc_trdev(int sizeof_priv)
624 {
625         return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
626 }
627
628 /*
629  *      Called during bootup.  We don't actually have to initialise
630  *      too much for this.
631  */
632
633 static int __init rif_init(void)
634 {
635         init_timer(&rif_timer);
636         rif_timer.expires  = sysctl_tr_rif_timeout;
637         rif_timer.data     = 0L;
638         rif_timer.function = rif_check_expire;
639         add_timer(&rif_timer);
640
641         proc_net_fops_create("tr_rif", S_IRUGO, &rif_seq_fops);
642         return 0;
643 }
644
645 module_init(rif_init);
646
647 EXPORT_SYMBOL(tr_type_trans);
648 EXPORT_SYMBOL(alloc_trdev);