[TR]: Use ctl paths to register net/token-ring/ table
[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 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 {
289 DECLARE_MAC_BUF(mac);
290 printk("source routing for %s\n",print_mac(mac, trh->daddr));
291 }
292 #endif
293                         if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
294                         {
295                                 trh->rcf=entry->rcf;
296                                 memcpy(&trh->rseg[0],&entry->rseg[0],8*sizeof(unsigned short));
297                                 trh->rcf^=htons(TR_RCF_DIR_BIT);
298                                 trh->rcf&=htons(0x1fff);        /* Issam Chehab <ichehab@madge1.demon.co.uk> */
299
300                                 trh->saddr[0]|=TR_RII;
301 #if TR_SR_DEBUG
302                                 printk("entry found with rcf %04x\n", entry->rcf);
303                         }
304                         else
305                         {
306                                 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
307 #endif
308                         }
309                         entry->last_used=jiffies;
310                 }
311                 else
312                 {
313                         /*
314                          *      Without the information we simply have to shout
315                          *      on the wire. The replies should rapidly clean this
316                          *      situation up.
317                          */
318                         trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
319                                        | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
320                         trh->saddr[0]|=TR_RII;
321 #if TR_SR_DEBUG
322                         printk("no entry in rif table found - broadcasting frame\n");
323 #endif
324                 }
325         }
326
327         /* Compress the RIF here so we don't have to do it in the driver(s) */
328         if (!(trh->saddr[0] & 0x80))
329                 slack = 18;
330         else
331                 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
332         olddata = skb->data;
333         spin_unlock_irqrestore(&rif_lock, flags);
334
335         skb_pull(skb, slack);
336         memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
337 }
338
339 /*
340  *      We have learned some new RIF information for our source
341  *      routing.
342  */
343
344 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
345 {
346         unsigned int hash, rii_p = 0;
347         unsigned long flags;
348         struct rif_cache *entry;
349         unsigned char saddr0;
350
351         spin_lock_irqsave(&rif_lock, flags);
352         saddr0 = trh->saddr[0];
353
354         /*
355          *      Firstly see if the entry exists
356          */
357
358         if(trh->saddr[0] & TR_RII)
359         {
360                 trh->saddr[0]&=0x7f;
361                 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
362                 {
363                         rii_p = 1;
364                 }
365         }
366
367         hash = rif_hash(trh->saddr);
368         for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);entry=entry->next);
369
370         if(entry==NULL)
371         {
372 #if TR_SR_DEBUG
373                 DECLARE_MAC_BUF(mac);
374                 printk("adding rif_entry: addr:%s rcf:%04X\n",
375                        print_mac(mac, trh->saddr), ntohs(trh->rcf));
376 #endif
377                 /*
378                  *      Allocate our new entry. A failure to allocate loses
379                  *      use the information. This is harmless.
380                  *
381                  *      FIXME: We ought to keep some kind of cache size
382                  *      limiting and adjust the timers to suit.
383                  */
384                 entry=kmalloc(sizeof(struct rif_cache),GFP_ATOMIC);
385
386                 if(!entry)
387                 {
388                         printk(KERN_DEBUG "tr.c: Couldn't malloc rif cache entry !\n");
389                         spin_unlock_irqrestore(&rif_lock, flags);
390                         return;
391                 }
392
393                 memcpy(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);
394                 entry->iface = dev->ifindex;
395                 entry->next=rif_table[hash];
396                 entry->last_used=jiffies;
397                 rif_table[hash]=entry;
398
399                 if (rii_p)
400                 {
401                         entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
402                         memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
403                         entry->local_ring = 0;
404                 }
405                 else
406                 {
407                         entry->local_ring = 1;
408                 }
409         }
410         else    /* Y. Tahara added */
411         {
412                 /*
413                  *      Update existing entries
414                  */
415                 if (!entry->local_ring)
416                     if (entry->rcf != (trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK)) &&
417                          !(trh->rcf & htons(TR_RCF_BROADCAST_MASK)))
418                     {
419 #if TR_SR_DEBUG
420 {
421 DECLARE_MAC_BUF(mac);
422 printk("updating rif_entry: addr:%s rcf:%04X\n",
423                 print_mac(mac, trh->saddr), ntohs(trh->rcf));
424 }
425 #endif
426                             entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
427                             memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
428                     }
429                 entry->last_used=jiffies;
430         }
431         trh->saddr[0]=saddr0; /* put the routing indicator back for tcpdump */
432         spin_unlock_irqrestore(&rif_lock, flags);
433 }
434
435 /*
436  *      Scan the cache with a timer and see what we need to throw out.
437  */
438
439 static void rif_check_expire(unsigned long dummy)
440 {
441         int i;
442         unsigned long flags, next_interval = jiffies + sysctl_tr_rif_timeout/2;
443
444         spin_lock_irqsave(&rif_lock, flags);
445
446         for(i =0; i < RIF_TABLE_SIZE; i++) {
447                 struct rif_cache *entry, **pentry;
448
449                 pentry = rif_table+i;
450                 while((entry=*pentry) != NULL) {
451                         unsigned long expires
452                                 = entry->last_used + sysctl_tr_rif_timeout;
453
454                         if (time_before_eq(expires, jiffies)) {
455                                 *pentry = entry->next;
456                                 kfree(entry);
457                         } else {
458                                 pentry = &entry->next;
459
460                                 if (time_before(expires, next_interval))
461                                         next_interval = expires;
462                         }
463                 }
464         }
465
466         spin_unlock_irqrestore(&rif_lock, flags);
467
468         mod_timer(&rif_timer, next_interval);
469
470 }
471
472 /*
473  *      Generate the /proc/net information for the token ring RIF
474  *      routing.
475  */
476
477 #ifdef CONFIG_PROC_FS
478
479 static struct rif_cache *rif_get_idx(loff_t pos)
480 {
481         int i;
482         struct rif_cache *entry;
483         loff_t off = 0;
484
485         for(i = 0; i < RIF_TABLE_SIZE; i++)
486                 for(entry = rif_table[i]; entry; entry = entry->next) {
487                         if (off == pos)
488                                 return entry;
489                         ++off;
490                 }
491
492         return NULL;
493 }
494
495 static void *rif_seq_start(struct seq_file *seq, loff_t *pos)
496 {
497         spin_lock_irq(&rif_lock);
498
499         return *pos ? rif_get_idx(*pos - 1) : SEQ_START_TOKEN;
500 }
501
502 static void *rif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
503 {
504         int i;
505         struct rif_cache *ent = v;
506
507         ++*pos;
508
509         if (v == SEQ_START_TOKEN) {
510                 i = -1;
511                 goto scan;
512         }
513
514         if (ent->next)
515                 return ent->next;
516
517         i = rif_hash(ent->addr);
518  scan:
519         while (++i < RIF_TABLE_SIZE) {
520                 if ((ent = rif_table[i]) != NULL)
521                         return ent;
522         }
523         return NULL;
524 }
525
526 static void rif_seq_stop(struct seq_file *seq, void *v)
527 {
528         spin_unlock_irq(&rif_lock);
529 }
530
531 static int rif_seq_show(struct seq_file *seq, void *v)
532 {
533         int j, rcf_len, segment, brdgnmb;
534         struct rif_cache *entry = v;
535         DECLARE_MAC_BUF(mac);
536
537         if (v == SEQ_START_TOKEN)
538                 seq_puts(seq,
539                      "if     TR address       TTL   rcf   routing segments\n");
540         else {
541                 struct net_device *dev = dev_get_by_index(&init_net, entry->iface);
542                 long ttl = (long) (entry->last_used + sysctl_tr_rif_timeout)
543                                 - (long) jiffies;
544
545                 seq_printf(seq, "%s %s %7li ",
546                            dev?dev->name:"?",
547                            print_mac(mac, entry->addr),
548                            ttl/HZ);
549
550                         if (entry->local_ring)
551                                 seq_puts(seq, "local\n");
552                         else {
553
554                                 seq_printf(seq, "%04X", ntohs(entry->rcf));
555                                 rcf_len = ((ntohs(entry->rcf) & TR_RCF_LEN_MASK)>>8)-2;
556                                 if (rcf_len)
557                                         rcf_len >>= 1;
558                                 for(j = 1; j < rcf_len; j++) {
559                                         if(j==1) {
560                                                 segment=ntohs(entry->rseg[j-1])>>4;
561                                                 seq_printf(seq,"  %03X",segment);
562                                         }
563
564                                         segment=ntohs(entry->rseg[j])>>4;
565                                         brdgnmb=ntohs(entry->rseg[j-1])&0x00f;
566                                         seq_printf(seq,"-%01X-%03X",brdgnmb,segment);
567                                 }
568                                 seq_putc(seq, '\n');
569                         }
570                 }
571         return 0;
572 }
573
574
575 static const struct seq_operations rif_seq_ops = {
576         .start = rif_seq_start,
577         .next  = rif_seq_next,
578         .stop  = rif_seq_stop,
579         .show  = rif_seq_show,
580 };
581
582 static int rif_seq_open(struct inode *inode, struct file *file)
583 {
584         return seq_open(file, &rif_seq_ops);
585 }
586
587 static const struct file_operations rif_seq_fops = {
588         .owner   = THIS_MODULE,
589         .open    = rif_seq_open,
590         .read    = seq_read,
591         .llseek  = seq_lseek,
592         .release = seq_release,
593 };
594
595 #endif
596
597 static const struct header_ops tr_header_ops = {
598         .create = tr_header,
599         .rebuild= tr_rebuild_header,
600 };
601
602 static void tr_setup(struct net_device *dev)
603 {
604         /*
605          *      Configure and register
606          */
607
608         dev->header_ops = &tr_header_ops;
609
610         dev->type               = ARPHRD_IEEE802_TR;
611         dev->hard_header_len    = TR_HLEN;
612         dev->mtu                = 2000;
613         dev->addr_len           = TR_ALEN;
614         dev->tx_queue_len       = 100;  /* Long queues on tr */
615
616         memset(dev->broadcast,0xFF, TR_ALEN);
617
618         /* New-style flags. */
619         dev->flags              = IFF_BROADCAST | IFF_MULTICAST ;
620 }
621
622 /**
623  * alloc_trdev - Register token ring device
624  * @sizeof_priv: Size of additional driver-private structure to be allocated
625  *      for this token ring device
626  *
627  * Fill in the fields of the device structure with token ring-generic values.
628  *
629  * Constructs a new net device, complete with a private data area of
630  * size @sizeof_priv.  A 32-byte (not bit) alignment is enforced for
631  * this private data area.
632  */
633 struct net_device *alloc_trdev(int sizeof_priv)
634 {
635         return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
636 }
637
638 #ifdef CONFIG_SYSCTL
639 static struct ctl_table tr_table[] = {
640         {
641                 .ctl_name       = NET_TR_RIF_TIMEOUT,
642                 .procname       = "rif_timeout",
643                 .data           = &sysctl_tr_rif_timeout,
644                 .maxlen         = sizeof(int),
645                 .mode           = 0644,
646                 .proc_handler   = &proc_dointvec
647         },
648         { 0 },
649 };
650
651 static __initdata struct ctl_path tr_path[] = {
652         { .procname = "net", .ctl_name = CTL_NET, },
653         { .procname = "token-ring", .ctl_name = NET_TR, },
654         { }
655 };
656 #endif
657
658 /*
659  *      Called during bootup.  We don't actually have to initialise
660  *      too much for this.
661  */
662
663 static int __init rif_init(void)
664 {
665         rif_timer.expires  = jiffies + sysctl_tr_rif_timeout;
666         setup_timer(&rif_timer, rif_check_expire, 0);
667         add_timer(&rif_timer);
668 #ifdef CONFIG_SYSCTL
669         register_sysctl_paths(tr_path, tr_table);
670 #endif
671         proc_net_fops_create(&init_net, "tr_rif", S_IRUGO, &rif_seq_fops);
672         return 0;
673 }
674
675 module_init(rif_init);
676
677 EXPORT_SYMBOL(tr_type_trans);
678 EXPORT_SYMBOL(alloc_trdev);