1c6e596074dff99fe4571ff4df1d40c081186f01
[linux-3.10.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 <linux/slab.h>
40 #include <net/arp.h>
41 #include <net/net_namespace.h>
42
43 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev);
44 static void rif_check_expire(unsigned long dummy);
45
46 #define TR_SR_DEBUG 0
47
48 /*
49  *      Each RIF entry we learn is kept this way
50  */
51
52 struct rif_cache {
53         unsigned char addr[TR_ALEN];
54         int iface;
55         __be16 rcf;
56         __be16 rseg[8];
57         struct rif_cache *next;
58         unsigned long last_used;
59         unsigned char local_ring;
60 };
61
62 #define RIF_TABLE_SIZE 32
63
64 /*
65  *      We hash the RIF cache 32 ways. We do after all have to look it
66  *      up a lot.
67  */
68
69 static struct rif_cache *rif_table[RIF_TABLE_SIZE];
70
71 static DEFINE_SPINLOCK(rif_lock);
72
73
74 /*
75  *      Garbage disposal timer.
76  */
77
78 static struct timer_list rif_timer;
79
80 static int sysctl_tr_rif_timeout = 60*10*HZ;
81
82 static inline unsigned long rif_hash(const unsigned char *addr)
83 {
84         unsigned long x;
85
86         x = addr[0];
87         x = (x << 2) ^ addr[1];
88         x = (x << 2) ^ addr[2];
89         x = (x << 2) ^ addr[3];
90         x = (x << 2) ^ addr[4];
91         x = (x << 2) ^ addr[5];
92
93         x ^= x >> 8;
94
95         return x & (RIF_TABLE_SIZE - 1);
96 }
97
98 /*
99  *      Put the headers on a token ring packet. Token ring source routing
100  *      makes this a little more exciting than on ethernet.
101  */
102
103 static int tr_header(struct sk_buff *skb, struct net_device *dev,
104                      unsigned short type,
105                      const void *daddr, const void *saddr, unsigned len)
106 {
107         struct trh_hdr *trh;
108         int hdr_len;
109
110         /*
111          * Add the 802.2 SNAP header if IP as the IPv4/IPv6 code calls
112          * dev->hard_header directly.
113          */
114         if (type == ETH_P_IP || type == ETH_P_IPV6 || type == ETH_P_ARP)
115         {
116                 struct trllc *trllc;
117
118                 hdr_len = sizeof(struct trh_hdr) + sizeof(struct trllc);
119                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
120                 trllc = (struct trllc *)(trh+1);
121                 trllc->dsap = trllc->ssap = EXTENDED_SAP;
122                 trllc->llc = UI_CMD;
123                 trllc->protid[0] = trllc->protid[1] = trllc->protid[2] = 0x00;
124                 trllc->ethertype = htons(type);
125         }
126         else
127         {
128                 hdr_len = sizeof(struct trh_hdr);
129                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
130         }
131
132         trh->ac=AC;
133         trh->fc=LLC_FRAME;
134
135         if(saddr)
136                 memcpy(trh->saddr,saddr,dev->addr_len);
137         else
138                 memcpy(trh->saddr,dev->dev_addr,dev->addr_len);
139
140         /*
141          *      Build the destination and then source route the frame
142          */
143
144         if(daddr)
145         {
146                 memcpy(trh->daddr,daddr,dev->addr_len);
147                 tr_source_route(skb, trh, dev);
148                 return(hdr_len);
149         }
150
151         return -hdr_len;
152 }
153
154 /*
155  *      A neighbour discovery of some species (eg arp) has completed. We
156  *      can now send the packet.
157  */
158
159 static int tr_rebuild_header(struct sk_buff *skb)
160 {
161         struct trh_hdr *trh=(struct trh_hdr *)skb->data;
162         struct trllc *trllc=(struct trllc *)(skb->data+sizeof(struct trh_hdr));
163         struct net_device *dev = skb->dev;
164
165         /*
166          *      FIXME: We don't yet support IPv6 over token rings
167          */
168
169         if(trllc->ethertype != htons(ETH_P_IP)) {
170                 printk("tr_rebuild_header: Don't know how to resolve type %04X addresses ?\n", ntohs(trllc->ethertype));
171                 return 0;
172         }
173
174 #ifdef CONFIG_INET
175         if(arp_find(trh->daddr, skb)) {
176                         return 1;
177         }
178         else
179 #endif
180         {
181                 tr_source_route(skb,trh,dev);
182                 return 0;
183         }
184 }
185
186 /*
187  *      Some of this is a bit hackish. We intercept RIF information
188  *      used for source routing. We also grab IP directly and don't feed
189  *      it via SNAP.
190  */
191
192 __be16 tr_type_trans(struct sk_buff *skb, struct net_device *dev)
193 {
194
195         struct trh_hdr *trh;
196         struct trllc *trllc;
197         unsigned riflen=0;
198
199         skb->dev = dev;
200         skb_reset_mac_header(skb);
201         trh = tr_hdr(skb);
202
203         if(trh->saddr[0] & TR_RII)
204                 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
205
206         trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
207
208         skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
209
210         if(*trh->daddr & 0x80)
211         {
212                 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))
213                         skb->pkt_type=PACKET_BROADCAST;
214                 else
215                         skb->pkt_type=PACKET_MULTICAST;
216         }
217         else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
218         {
219                 skb->pkt_type=PACKET_MULTICAST;
220         }
221         else if(dev->flags & IFF_PROMISC)
222         {
223                 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
224                         skb->pkt_type=PACKET_OTHERHOST;
225         }
226
227         if ((skb->pkt_type != PACKET_BROADCAST) &&
228             (skb->pkt_type != PACKET_MULTICAST))
229                 tr_add_rif_info(trh,dev) ;
230
231         /*
232          * Strip the SNAP header from ARP packets since we don't
233          * pass them through to the 802.2/SNAP layers.
234          */
235
236         if (trllc->dsap == EXTENDED_SAP &&
237             (trllc->ethertype == htons(ETH_P_IP) ||
238              trllc->ethertype == htons(ETH_P_IPV6) ||
239              trllc->ethertype == htons(ETH_P_ARP)))
240         {
241                 skb_pull(skb, sizeof(struct trllc));
242                 return trllc->ethertype;
243         }
244
245         return htons(ETH_P_TR_802_2);
246 }
247
248 /*
249  *      We try to do source routing...
250  */
251
252 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,
253                      struct net_device *dev)
254 {
255         int slack;
256         unsigned int hash;
257         struct rif_cache *entry;
258         unsigned char *olddata;
259         unsigned long flags;
260         static const unsigned char mcast_func_addr[]
261                 = {0xC0,0x00,0x00,0x04,0x00,0x00};
262
263         spin_lock_irqsave(&rif_lock, flags);
264
265         /*
266          *      Broadcasts are single route as stated in RFC 1042
267          */
268         if( (!memcmp(&(trh->daddr[0]),&(dev->broadcast[0]),TR_ALEN)) ||
269             (!memcmp(&(trh->daddr[0]),&(mcast_func_addr[0]), TR_ALEN))  )
270         {
271                 trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
272                                | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
273                 trh->saddr[0]|=TR_RII;
274         }
275         else
276         {
277                 hash = rif_hash(trh->daddr);
278                 /*
279                  *      Walk the hash table and look for an entry
280                  */
281                 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->daddr[0]),TR_ALEN);entry=entry->next);
282
283                 /*
284                  *      If we found an entry we can route the frame.
285                  */
286                 if(entry)
287                 {
288 #if TR_SR_DEBUG
289 printk("source routing for %pM\n", trh->daddr);
290 #endif
291                         if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
292                         {
293                                 trh->rcf=entry->rcf;
294                                 memcpy(&trh->rseg[0],&entry->rseg[0],8*sizeof(unsigned short));
295                                 trh->rcf^=htons(TR_RCF_DIR_BIT);
296                                 trh->rcf&=htons(0x1fff);        /* Issam Chehab <ichehab@madge1.demon.co.uk> */
297
298                                 trh->saddr[0]|=TR_RII;
299 #if TR_SR_DEBUG
300                                 printk("entry found with rcf %04x\n", entry->rcf);
301                         }
302                         else
303                         {
304                                 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
305 #endif
306                         }
307                         entry->last_used=jiffies;
308                 }
309                 else
310                 {
311                         /*
312                          *      Without the information we simply have to shout
313                          *      on the wire. The replies should rapidly clean this
314                          *      situation up.
315                          */
316                         trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
317                                        | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
318                         trh->saddr[0]|=TR_RII;
319 #if TR_SR_DEBUG
320                         printk("no entry in rif table found - broadcasting frame\n");
321 #endif
322                 }
323         }
324
325         /* Compress the RIF here so we don't have to do it in the driver(s) */
326         if (!(trh->saddr[0] & 0x80))
327                 slack = 18;
328         else
329                 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
330         olddata = skb->data;
331         spin_unlock_irqrestore(&rif_lock, flags);
332
333         skb_pull(skb, slack);
334         memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
335 }
336
337 /*
338  *      We have learned some new RIF information for our source
339  *      routing.
340  */
341
342 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
343 {
344         unsigned int hash, rii_p = 0;
345         unsigned long flags;
346         struct rif_cache *entry;
347         unsigned char saddr0;
348
349         spin_lock_irqsave(&rif_lock, flags);
350         saddr0 = trh->saddr[0];
351
352         /*
353          *      Firstly see if the entry exists
354          */
355
356         if(trh->saddr[0] & TR_RII)
357         {
358                 trh->saddr[0]&=0x7f;
359                 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
360                 {
361                         rii_p = 1;
362                 }
363         }
364
365         hash = rif_hash(trh->saddr);
366         for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);entry=entry->next);
367
368         if(entry==NULL)
369         {
370 #if TR_SR_DEBUG
371                 printk("adding rif_entry: addr:%pM rcf:%04X\n",
372                        trh->saddr, ntohs(trh->rcf));
373 #endif
374                 /*
375                  *      Allocate our new entry. A failure to allocate loses
376                  *      use the information. This is harmless.
377                  *
378                  *      FIXME: We ought to keep some kind of cache size
379                  *      limiting and adjust the timers to suit.
380                  */
381                 entry=kmalloc(sizeof(struct rif_cache),GFP_ATOMIC);
382
383                 if(!entry)
384                 {
385                         printk(KERN_DEBUG "tr.c: Couldn't malloc rif cache entry !\n");
386                         spin_unlock_irqrestore(&rif_lock, flags);
387                         return;
388                 }
389
390                 memcpy(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);
391                 entry->iface = dev->ifindex;
392                 entry->next=rif_table[hash];
393                 entry->last_used=jiffies;
394                 rif_table[hash]=entry;
395
396                 if (rii_p)
397                 {
398                         entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
399                         memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
400                         entry->local_ring = 0;
401                 }
402                 else
403                 {
404                         entry->local_ring = 1;
405                 }
406         }
407         else    /* Y. Tahara added */
408         {
409                 /*
410                  *      Update existing entries
411                  */
412                 if (!entry->local_ring)
413                     if (entry->rcf != (trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK)) &&
414                          !(trh->rcf & htons(TR_RCF_BROADCAST_MASK)))
415                     {
416 #if TR_SR_DEBUG
417 printk("updating rif_entry: addr:%pM rcf:%04X\n",
418                 trh->saddr, 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         trh->saddr[0]=saddr0; /* put the routing indicator back for tcpdump */
426         spin_unlock_irqrestore(&rif_lock, flags);
427 }
428
429 /*
430  *      Scan the cache with a timer and see what we need to throw out.
431  */
432
433 static void rif_check_expire(unsigned long dummy)
434 {
435         int i;
436         unsigned long flags, next_interval = jiffies + sysctl_tr_rif_timeout/2;
437
438         spin_lock_irqsave(&rif_lock, flags);
439
440         for(i =0; i < RIF_TABLE_SIZE; i++) {
441                 struct rif_cache *entry, **pentry;
442
443                 pentry = rif_table+i;
444                 while((entry=*pentry) != NULL) {
445                         unsigned long expires
446                                 = entry->last_used + sysctl_tr_rif_timeout;
447
448                         if (time_before_eq(expires, jiffies)) {
449                                 *pentry = entry->next;
450                                 kfree(entry);
451                         } else {
452                                 pentry = &entry->next;
453
454                                 if (time_before(expires, next_interval))
455                                         next_interval = expires;
456                         }
457                 }
458         }
459
460         spin_unlock_irqrestore(&rif_lock, flags);
461
462         mod_timer(&rif_timer, next_interval);
463
464 }
465
466 /*
467  *      Generate the /proc/net information for the token ring RIF
468  *      routing.
469  */
470
471 #ifdef CONFIG_PROC_FS
472
473 static struct rif_cache *rif_get_idx(loff_t pos)
474 {
475         int i;
476         struct rif_cache *entry;
477         loff_t off = 0;
478
479         for(i = 0; i < RIF_TABLE_SIZE; i++)
480                 for(entry = rif_table[i]; entry; entry = entry->next) {
481                         if (off == pos)
482                                 return entry;
483                         ++off;
484                 }
485
486         return NULL;
487 }
488
489 static void *rif_seq_start(struct seq_file *seq, loff_t *pos)
490         __acquires(&rif_lock)
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         __releases(&rif_lock)
523 {
524         spin_unlock_irq(&rif_lock);
525 }
526
527 static int rif_seq_show(struct seq_file *seq, void *v)
528 {
529         int j, rcf_len, segment, brdgnmb;
530         struct rif_cache *entry = v;
531
532         if (v == SEQ_START_TOKEN)
533                 seq_puts(seq,
534                      "if     TR address       TTL   rcf   routing segments\n");
535         else {
536                 struct net_device *dev = dev_get_by_index(&init_net, entry->iface);
537                 long ttl = (long) (entry->last_used + sysctl_tr_rif_timeout)
538                                 - (long) jiffies;
539
540                 seq_printf(seq, "%s %pM %7li ",
541                            dev?dev->name:"?",
542                            entry->addr,
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
559                                         segment=ntohs(entry->rseg[j])>>4;
560                                         brdgnmb=ntohs(entry->rseg[j-1])&0x00f;
561                                         seq_printf(seq,"-%01X-%03X",brdgnmb,segment);
562                                 }
563                                 seq_putc(seq, '\n');
564                         }
565
566                 if (dev)
567                         dev_put(dev);
568                 }
569         return 0;
570 }
571
572
573 static const struct seq_operations rif_seq_ops = {
574         .start = rif_seq_start,
575         .next  = rif_seq_next,
576         .stop  = rif_seq_stop,
577         .show  = rif_seq_show,
578 };
579
580 static int rif_seq_open(struct inode *inode, struct file *file)
581 {
582         return seq_open(file, &rif_seq_ops);
583 }
584
585 static const struct file_operations rif_seq_fops = {
586         .owner   = THIS_MODULE,
587         .open    = rif_seq_open,
588         .read    = seq_read,
589         .llseek  = seq_lseek,
590         .release = seq_release,
591 };
592
593 #endif
594
595 static const struct header_ops tr_header_ops = {
596         .create = tr_header,
597         .rebuild= tr_rebuild_header,
598 };
599
600 static void tr_setup(struct net_device *dev)
601 {
602         /*
603          *      Configure and register
604          */
605
606         dev->header_ops = &tr_header_ops;
607
608         dev->type               = ARPHRD_IEEE802_TR;
609         dev->hard_header_len    = TR_HLEN;
610         dev->mtu                = 2000;
611         dev->addr_len           = TR_ALEN;
612         dev->tx_queue_len       = 100;  /* Long queues on tr */
613
614         memset(dev->broadcast,0xFF, TR_ALEN);
615
616         /* New-style flags. */
617         dev->flags              = IFF_BROADCAST | IFF_MULTICAST ;
618 }
619
620 /**
621  * alloc_trdev - Register token ring device
622  * @sizeof_priv: Size of additional driver-private structure to be allocated
623  *      for this token ring device
624  *
625  * Fill in the fields of the device structure with token ring-generic values.
626  *
627  * Constructs a new net device, complete with a private data area of
628  * size @sizeof_priv.  A 32-byte (not bit) alignment is enforced for
629  * this private data area.
630  */
631 struct net_device *alloc_trdev(int sizeof_priv)
632 {
633         return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
634 }
635
636 #ifdef CONFIG_SYSCTL
637 static struct ctl_table tr_table[] = {
638         {
639                 .procname       = "rif_timeout",
640                 .data           = &sysctl_tr_rif_timeout,
641                 .maxlen         = sizeof(int),
642                 .mode           = 0644,
643                 .proc_handler   = proc_dointvec
644         },
645         { },
646 };
647
648 static __initdata struct ctl_path tr_path[] = {
649         { .procname = "net", },
650         { .procname = "token-ring", },
651         { }
652 };
653 #endif
654
655 /*
656  *      Called during bootup.  We don't actually have to initialise
657  *      too much for this.
658  */
659
660 static int __init rif_init(void)
661 {
662         rif_timer.expires  = jiffies + sysctl_tr_rif_timeout;
663         setup_timer(&rif_timer, rif_check_expire, 0);
664         add_timer(&rif_timer);
665 #ifdef CONFIG_SYSCTL
666         register_sysctl_paths(tr_path, tr_table);
667 #endif
668         proc_net_fops_create(&init_net, "tr_rif", S_IRUGO, &rif_seq_fops);
669         return 0;
670 }
671
672 module_init(rif_init);
673
674 EXPORT_SYMBOL(tr_type_trans);
675 EXPORT_SYMBOL(alloc_trdev);
676
677 MODULE_LICENSE("GPL");