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=(struct trh_hdr *)skb->data;
 193         struct trllc *trllc;
 194         unsigned riflen=0;
 195
 196         skb->mac.raw = skb->data;
 197
 198         if(trh->saddr[0] & TR_RII)
 199                 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
 200
 201         trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
 202
 203         skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
 204
 205         if(*trh->daddr & 0x80)
 206         {
 207                 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))
 208                         skb->pkt_type=PACKET_BROADCAST;
 209                 else
 210                         skb->pkt_type=PACKET_MULTICAST;
 211         }
 212         else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
 213         {
 214                 skb->pkt_type=PACKET_MULTICAST;
 215         }
 216         else if(dev->flags & IFF_PROMISC)
 217         {
 218                 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
 219                         skb->pkt_type=PACKET_OTHERHOST;
 220         }
 221
 222         if ((skb->pkt_type != PACKET_BROADCAST) &&
 223             (skb->pkt_type != PACKET_MULTICAST))
 224                 tr_add_rif_info(trh,dev) ;
 225
 226         /*
 227          * Strip the SNAP header from ARP packets since we don't
 228          * pass them through to the 802.2/SNAP layers.
 229          */
 230
 231         if (trllc->dsap == EXTENDED_SAP &&
 232             (trllc->ethertype == htons(ETH_P_IP) ||
 233              trllc->ethertype == htons(ETH_P_IPV6) ||
 234              trllc->ethertype == htons(ETH_P_ARP)))
 235         {
 236                 skb_pull(skb, sizeof(struct trllc));
 237                 return trllc->ethertype;
 238         }
 239
 240         return htons(ETH_P_TR_802_2);
 241 }
 242
 243 /*
 244  *      We try to do source routing...
 245  */
 246
 247 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,struct net_device *dev)
 248 {
 249         int slack;
 250         unsigned int hash;
 251         struct rif_cache *entry;
 252         unsigned char *olddata;
 253         unsigned long flags;
 254         static const unsigned char mcast_func_addr[]
 255                 = {0xC0,0x00,0x00,0x04,0x00,0x00};
 256
 257         spin_lock_irqsave(&rif_lock, flags);
 258
 259         /*
 260          *      Broadcasts are single route as stated in RFC 1042
 261          */
 262         if( (!memcmp(&(trh->daddr[0]),&(dev->broadcast[0]),TR_ALEN)) ||
 263             (!memcmp(&(trh->daddr[0]),&(mcast_func_addr[0]), TR_ALEN))  )
 264         {
 265                 trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
 266                                | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
 267                 trh->saddr[0]|=TR_RII;
 268         }
 269         else
 270         {
 271                 hash = rif_hash(trh->daddr);
 272                 /*
 273                  *      Walk the hash table and look for an entry
 274                  */
 275                 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->daddr[0]),TR_ALEN);entry=entry->next);
 276
 277                 /*
 278                  *      If we found an entry we can route the frame.
 279                  */
 280                 if(entry)
 281                 {
 282 #if TR_SR_DEBUG
 283 printk("source routing for %02X:%02X:%02X:%02X:%02X:%02X\n",trh->daddr[0],
 284                   trh->daddr[1],trh->daddr[2],trh->daddr[3],trh->daddr[4],trh->daddr[5]);
 285 #endif
 286                         if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
 287                         {
 288                                 trh->rcf=entry->rcf;
 289                                 memcpy(&trh->rseg[0],&entry->rseg[0],8*sizeof(unsigned short));
 290                                 trh->rcf^=htons(TR_RCF_DIR_BIT);
 291                                 trh->rcf&=htons(0x1fff);        /* Issam Chehab <ichehab@madge1.demon.co.uk> */
 292
 293                                 trh->saddr[0]|=TR_RII;
 294 #if TR_SR_DEBUG
 295                                 printk("entry found with rcf %04x\n", entry->rcf);
 296                         }
 297                         else
 298                         {
 299                                 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
 300 #endif
 301                         }
 302                         entry->last_used=jiffies;
 303                 }
 304                 else
 305                 {
 306                         /*
 307                          *      Without the information we simply have to shout
 308                          *      on the wire. The replies should rapidly clean this
 309                          *      situation up.
 310                          */
 311                         trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
 312                                        | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
 313                         trh->saddr[0]|=TR_RII;
 314 #if TR_SR_DEBUG
 315                         printk("no entry in rif table found - broadcasting frame\n");
 316 #endif
 317                 }
 318         }
 319
 320         /* Compress the RIF here so we don't have to do it in the driver(s) */
 321         if (!(trh->saddr[0] & 0x80))
 322                 slack = 18;
 323         else
 324                 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
 325         olddata = skb->data;
 326         spin_unlock_irqrestore(&rif_lock, flags);
 327
 328         skb_pull(skb, slack);
 329         memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
 330 }
 331
 332 /*
 333  *      We have learned some new RIF information for our source
 334  *      routing.
 335  */
 336
 337 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
 338 {
 339         unsigned int hash, rii_p = 0;
 340         unsigned long flags;
 341         struct rif_cache *entry;
 342         unsigned char saddr0;
 343
 344         spin_lock_irqsave(&rif_lock, flags);
 345         saddr0 = trh->saddr[0];
 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                 }
 399                 else
 400                 {
 401                         entry->local_ring = 1;
 402                 }
 403         }
 404         else    /* Y. Tahara added */
 405         {
 406                 /*
 407                  *      Update existing entries
 408                  */
 409                 if (!entry->local_ring)
 410                     if (entry->rcf != (trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK)) &&
 411                          !(trh->rcf & htons(TR_RCF_BROADCAST_MASK)))
 412                     {
 413 #if TR_SR_DEBUG
 414 printk("updating rif_entry: addr:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
 415                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
 416                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
 417                 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 {
 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 const 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_type_trans);
 646 EXPORT_SYMBOL(alloc_trdev);