net/802/tr.c

   1 /*
   2  * NET3:        Token ring device handling subroutines
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Fixes:       3 Feb 97 Paul Norton <pnorton@cts.com> Minor routing fixes.
  10  *              Added rif table to /proc/net/tr_rif and rif timeout to
  11  *              /proc/sys/net/token-ring/rif_timeout.
  12  *              22 Jun 98 Paul Norton <p.norton@computer.org> Rearranged
  13  *              tr_header and tr_type_trans to handle passing IPX SNAP and
  14  *              802.2 through the correct layers. Eliminated tr_reformat.
  15  *
  16  */
  17
  18 #include <asm/uaccess.h>
  19 #include <asm/system.h>
  20 #include <linux/config.h>
  21 #include <linux/module.h>
  22 #include <linux/types.h>
  23 #include <linux/kernel.h>
  24 #include <linux/jiffies.h>
  25 #include <linux/string.h>
  26 #include <linux/mm.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/inet.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/trdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/errno.h>
  34 #include <linux/timer.h>
  35 #include <linux/net.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <net/arp.h>
  40
  41 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev);
  42 static void rif_check_expire(unsigned long dummy);
  43
  44 #define TR_SR_DEBUG 0
  45
  46 /*
  47  *      Each RIF entry we learn is kept this way
  48  */
  49
  50 struct rif_cache_s {
  51         unsigned char addr[TR_ALEN];
  52         int iface;
  53         __u16 rcf;
  54         __u16 rseg[8];
  55         struct rif_cache_s *next;
  56         unsigned long last_used;
  57         unsigned char local_ring;
  58 };
  59
  60 #define RIF_TABLE_SIZE 32
  61
  62 /*
  63  *      We hash the RIF cache 32 ways. We do after all have to look it
  64  *      up a lot.
  65  */
  66
  67 static struct rif_cache_s *rif_table[RIF_TABLE_SIZE];
  68
  69 static DEFINE_SPINLOCK(rif_lock);
  70
  71
  72 /*
  73  *      Garbage disposal timer.
  74  */
  75
  76 static struct timer_list rif_timer;
  77
  78 int sysctl_tr_rif_timeout = 60*10*HZ;
  79
  80 static inline unsigned long rif_hash(const unsigned char *addr)
  81 {
  82         unsigned long x;
  83
  84         x = addr[0];
  85         x = (x << 2) ^ addr[1];
  86         x = (x << 2) ^ addr[2];
  87         x = (x << 2) ^ addr[3];
  88         x = (x << 2) ^ addr[4];
  89         x = (x << 2) ^ addr[5];
  90
  91         x ^= x >> 8;
  92
  93         return x & (RIF_TABLE_SIZE - 1);
  94 }
  95
  96 /*
  97  *      Put the headers on a token ring packet. Token ring source routing
  98  *      makes this a little more exciting than on ethernet.
  99  */
 100
 101 static int tr_header(struct sk_buff *skb, struct net_device *dev,
 102                      unsigned short type,
 103                      void *daddr, void *saddr, unsigned len)
 104 {
 105         struct trh_hdr *trh;
 106         int hdr_len;
 107
 108         /*
 109          * Add the 802.2 SNAP header if IP as the IPv4/IPv6 code calls
 110          * dev->hard_header directly.
 111          */
 112         if (type == ETH_P_IP || type == ETH_P_IPV6 || type == ETH_P_ARP)
 113         {
 114                 struct trllc *trllc;
 115
 116                 hdr_len = sizeof(struct trh_hdr) + sizeof(struct trllc);
 117                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
 118                 trllc = (struct trllc *)(trh+1);
 119                 trllc->dsap = trllc->ssap = EXTENDED_SAP;
 120                 trllc->llc = UI_CMD;
 121                 trllc->protid[0] = trllc->protid[1] = trllc->protid[2] = 0x00;
 122                 trllc->ethertype = htons(type);
 123         }
 124         else
 125         {
 126                 hdr_len = sizeof(struct trh_hdr);
 127                 trh = (struct trh_hdr *)skb_push(skb, hdr_len);
 128         }
 129
 130         trh->ac=AC;
 131         trh->fc=LLC_FRAME;
 132
 133         if(saddr)
 134                 memcpy(trh->saddr,saddr,dev->addr_len);
 135         else
 136                 memcpy(trh->saddr,dev->dev_addr,dev->addr_len);
 137
 138         /*
 139          *      Build the destination and then source route the frame
 140          */
 141
 142         if(daddr)
 143         {
 144                 memcpy(trh->daddr,daddr,dev->addr_len);
 145                 tr_source_route(skb,trh,dev);
 146                 return(hdr_len);
 147         }
 148
 149         return -hdr_len;
 150 }
 151
 152 /*
 153  *      A neighbour discovery of some species (eg arp) has completed. We
 154  *      can now send the packet.
 155  */
 156
 157 static int tr_rebuild_header(struct sk_buff *skb)
 158 {
 159         struct trh_hdr *trh=(struct trh_hdr *)skb->data;
 160         struct trllc *trllc=(struct trllc *)(skb->data+sizeof(struct trh_hdr));
 161         struct net_device *dev = skb->dev;
 162
 163         /*
 164          *      FIXME: We don't yet support IPv6 over token rings
 165          */
 166
 167         if(trllc->ethertype != htons(ETH_P_IP)) {
 168                 printk("tr_rebuild_header: Don't know how to resolve type %04X addresses ?\n",(unsigned int)htons(trllc->ethertype));
 169                 return 0;
 170         }
 171
 172 #ifdef CONFIG_INET
 173         if(arp_find(trh->daddr, skb)) {
 174                         return 1;
 175         }
 176         else
 177 #endif
 178         {
 179                 tr_source_route(skb,trh,dev);
 180                 return 0;
 181         }
 182 }
 183
 184 /*
 185  *      Some of this is a bit hackish. We intercept RIF information
 186  *      used for source routing. We also grab IP directly and don't feed
 187  *      it via SNAP.
 188  */
 189
 190 unsigned short tr_type_trans(struct sk_buff *skb, struct net_device *dev)
 191 {
 192
 193         struct trh_hdr *trh=(struct trh_hdr *)skb->data;
 194         struct trllc *trllc;
 195         unsigned riflen=0;
 196
 197         skb->mac.raw = skb->data;
 198
 199         if(trh->saddr[0] & TR_RII)
 200                 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
 201
 202         trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
 203
 204         skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
 205
 206         if(*trh->daddr & 0x80)
 207         {
 208                 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))
 209                         skb->pkt_type=PACKET_BROADCAST;
 210                 else
 211                         skb->pkt_type=PACKET_MULTICAST;
 212         }
 213         else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
 214         {
 215                 skb->pkt_type=PACKET_MULTICAST;
 216         }
 217         else if(dev->flags & IFF_PROMISC)
 218         {
 219                 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
 220                         skb->pkt_type=PACKET_OTHERHOST;
 221         }
 222
 223         if ((skb->pkt_type != PACKET_BROADCAST) &&
 224             (skb->pkt_type != PACKET_MULTICAST))
 225                 tr_add_rif_info(trh,dev) ;
 226
 227         /*
 228          * Strip the SNAP header from ARP packets since we don't
 229          * pass them through to the 802.2/SNAP layers.
 230          */
 231
 232         if (trllc->dsap == EXTENDED_SAP &&
 233             (trllc->ethertype == ntohs(ETH_P_IP) ||
 234              trllc->ethertype == ntohs(ETH_P_IPV6) ||
 235              trllc->ethertype == ntohs(ETH_P_ARP)))
 236         {
 237                 skb_pull(skb, sizeof(struct trllc));
 238                 return trllc->ethertype;
 239         }
 240
 241         return ntohs(ETH_P_802_2);
 242 }
 243
 244 /*
 245  *      We try to do source routing...
 246  */
 247
 248 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,struct net_device *dev)
 249 {
 250         int slack;
 251         unsigned int hash;
 252         struct rif_cache_s *entry;
 253         unsigned char *olddata;
 254         static const unsigned char mcast_func_addr[]
 255                 = {0xC0,0x00,0x00,0x04,0x00,0x00};
 256
 257         spin_lock_bh(&rif_lock);
 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_bh(&rif_lock);
 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         struct rif_cache_s *entry;
 341
 342
 343         spin_lock_bh(&rif_lock);
 344
 345         /*
 346          *      Firstly see if the entry exists
 347          */
 348
 349         if(trh->saddr[0] & TR_RII)
 350         {
 351                 trh->saddr[0]&=0x7f;
 352                 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
 353                 {
 354                         rii_p = 1;
 355                 }
 356         }
 357
 358         hash = rif_hash(trh->saddr);
 359         for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);entry=entry->next);
 360
 361         if(entry==NULL)
 362         {
 363 #if TR_SR_DEBUG
 364 printk("adding rif_entry: addr:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
 365                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
 366                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
 367                 ntohs(trh->rcf));
 368 #endif
 369                 /*
 370                  *      Allocate our new entry. A failure to allocate loses
 371                  *      use the information. This is harmless.
 372                  *
 373                  *      FIXME: We ought to keep some kind of cache size
 374                  *      limiting and adjust the timers to suit.
 375                  */
 376                 entry=kmalloc(sizeof(struct rif_cache_s),GFP_ATOMIC);
 377
 378                 if(!entry)
 379                 {
 380                         printk(KERN_DEBUG "tr.c: Couldn't malloc rif cache entry !\n");
 381                         spin_unlock_bh(&rif_lock);
 382                         return;
 383                 }
 384
 385                 memcpy(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);
 386                 entry->iface = dev->ifindex;
 387                 entry->next=rif_table[hash];
 388                 entry->last_used=jiffies;
 389                 rif_table[hash]=entry;
 390
 391                 if (rii_p)
 392                 {
 393                         entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
 394                         memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
 395                         entry->local_ring = 0;
 396                         trh->saddr[0]|=TR_RII; /* put the routing indicator back for tcpdump */
 397                 }
 398                 else
 399                 {
 400                         entry->local_ring = 1;
 401                 }
 402         }
 403         else    /* Y. Tahara added */
 404         {
 405                 /*
 406                  *      Update existing entries
 407                  */
 408                 if (!entry->local_ring)
 409                     if (entry->rcf != (trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK)) &&
 410                          !(trh->rcf & htons(TR_RCF_BROADCAST_MASK)))
 411                     {
 412 #if TR_SR_DEBUG
 413 printk("updating rif_entry: addr:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
 414                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
 415                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
 416                 ntohs(trh->rcf));
 417 #endif
 418                             entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
 419                             memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
 420                     }
 421                 entry->last_used=jiffies;
 422         }
 423         spin_unlock_bh(&rif_lock);
 424 }
 425
 426 /*
 427  *      Scan the cache with a timer and see what we need to throw out.
 428  */
 429
 430 static void rif_check_expire(unsigned long dummy)
 431 {
 432         int i;
 433         unsigned long next_interval = jiffies + sysctl_tr_rif_timeout/2;
 434
 435         spin_lock_bh(&rif_lock);
 436
 437         for(i =0; i < RIF_TABLE_SIZE; i++) {
 438                 struct rif_cache_s *entry, **pentry;
 439
 440                 pentry = rif_table+i;
 441                 while((entry=*pentry) != NULL) {
 442                         unsigned long expires
 443                                 = entry->last_used + sysctl_tr_rif_timeout;
 444
 445                         if (time_before_eq(expires, jiffies)) {
 446                                 *pentry = entry->next;
 447                                 kfree(entry);
 448                         } else {
 449                                 pentry = &entry->next;
 450
 451                                 if (time_before(expires, next_interval))
 452                                         next_interval = expires;
 453                         }
 454                 }
 455         }
 456
 457         spin_unlock_bh(&rif_lock);
 458
 459         mod_timer(&rif_timer, next_interval);
 460
 461 }
 462
 463 /*
 464  *      Generate the /proc/net information for the token ring RIF
 465  *      routing.
 466  */
 467
 468 #ifdef CONFIG_PROC_FS
 469
 470 static struct rif_cache_s *rif_get_idx(loff_t pos)
 471 {
 472         int i;
 473         struct rif_cache_s *entry;
 474         loff_t off = 0;
 475
 476         for(i = 0; i < RIF_TABLE_SIZE; i++)
 477                 for(entry = rif_table[i]; entry; entry = entry->next) {
 478                         if (off == pos)
 479                                 return entry;
 480                         ++off;
 481                 }
 482
 483         return NULL;
 484 }
 485
 486 static void *rif_seq_start(struct seq_file *seq, loff_t *pos)
 487 {
 488         spin_lock_bh(&rif_lock);
 489
 490         return *pos ? rif_get_idx(*pos - 1) : SEQ_START_TOKEN;
 491 }
 492
 493 static void *rif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 494 {
 495         int i;
 496         struct rif_cache_s *ent = v;
 497
 498         ++*pos;
 499
 500         if (v == SEQ_START_TOKEN) {
 501                 i = -1;
 502                 goto scan;
 503         }
 504
 505         if (ent->next)
 506                 return ent->next;
 507
 508         i = rif_hash(ent->addr);
 509  scan:
 510         while (++i < RIF_TABLE_SIZE) {
 511                 if ((ent = rif_table[i]) != NULL)
 512                         return ent;
 513         }
 514         return NULL;
 515 }
 516
 517 static void rif_seq_stop(struct seq_file *seq, void *v)
 518 {
 519         spin_unlock_bh(&rif_lock);
 520 }
 521
 522 static int rif_seq_show(struct seq_file *seq, void *v)
 523 {
 524         int j, rcf_len, segment, brdgnmb;
 525         struct rif_cache_s *entry = v;
 526
 527         if (v == SEQ_START_TOKEN)
 528                 seq_puts(seq,
 529                      "if     TR address       TTL   rcf   routing segments\n");
 530         else {
 531                 struct net_device *dev = dev_get_by_index(entry->iface);
 532                 long ttl = (long) (entry->last_used + sysctl_tr_rif_timeout)
 533                                 - (long) jiffies;
 534
 535                 seq_printf(seq, "%s %02X:%02X:%02X:%02X:%02X:%02X %7li ",
 536                            dev?dev->name:"?",
 537                            entry->addr[0],entry->addr[1],entry->addr[2],
 538                            entry->addr[3],entry->addr[4],entry->addr[5],
 539                            ttl/HZ);
 540
 541                         if (entry->local_ring)
 542                                 seq_puts(seq, "local\n");
 543                         else {
 544
 545                                 seq_printf(seq, "%04X", ntohs(entry->rcf));
 546                                 rcf_len = ((ntohs(entry->rcf) & TR_RCF_LEN_MASK)>>8)-2;
 547                                 if (rcf_len)
 548                                         rcf_len >>= 1;
 549                                 for(j = 1; j < rcf_len; j++) {
 550                                         if(j==1) {
 551                                                 segment=ntohs(entry->rseg[j-1])>>4;
 552                                                 seq_printf(seq,"  %03X",segment);
 553                                         };
 554                                         segment=ntohs(entry->rseg[j])>>4;
 555                                         brdgnmb=ntohs(entry->rseg[j-1])&0x00f;
 556                                         seq_printf(seq,"-%01X-%03X",brdgnmb,segment);
 557                                 }
 558                                 seq_putc(seq, '\n');
 559                         }
 560                 }
 561         return 0;
 562 }
 563
 564
 565 static struct seq_operations rif_seq_ops = {
 566         .start = rif_seq_start,
 567         .next  = rif_seq_next,
 568         .stop  = rif_seq_stop,
 569         .show  = rif_seq_show,
 570 };
 571
 572 static int rif_seq_open(struct inode *inode, struct file *file)
 573 {
 574         return seq_open(file, &rif_seq_ops);
 575 }
 576
 577 static struct file_operations rif_seq_fops = {
 578         .owner   = THIS_MODULE,
 579         .open    = rif_seq_open,
 580         .read    = seq_read,
 581         .llseek  = seq_lseek,
 582         .release = seq_release,
 583 };
 584
 585 #endif
 586
 587 static void tr_setup(struct net_device *dev)
 588 {
 589         /*
 590          *      Configure and register
 591          */
 592
 593         dev->hard_header        = tr_header;
 594         dev->rebuild_header     = tr_rebuild_header;
 595
 596         dev->type               = ARPHRD_IEEE802_TR;
 597         dev->hard_header_len    = TR_HLEN;
 598         dev->mtu                = 2000;
 599         dev->addr_len           = TR_ALEN;
 600         dev->tx_queue_len       = 100;  /* Long queues on tr */
 601
 602         memset(dev->broadcast,0xFF, TR_ALEN);
 603
 604         /* New-style flags. */
 605         dev->flags              = IFF_BROADCAST | IFF_MULTICAST ;
 606 }
 607
 608 /**
 609  * alloc_trdev - Register token ring device
 610  * @sizeof_priv: Size of additional driver-private structure to be allocated
 611  *      for this token ring device
 612  *
 613  * Fill in the fields of the device structure with token ring-generic values.
 614  *
 615  * Constructs a new net device, complete with a private data area of
 616  * size @sizeof_priv.  A 32-byte (not bit) alignment is enforced for
 617  * this private data area.
 618  */
 619 struct net_device *alloc_trdev(int sizeof_priv)
 620 {
 621         return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
 622 }
 623
 624 /*
 625  *      Called during bootup.  We don't actually have to initialise
 626  *      too much for this.
 627  */
 628
 629 static int __init rif_init(void)
 630 {
 631         init_timer(&rif_timer);
 632         rif_timer.expires  = sysctl_tr_rif_timeout;
 633         rif_timer.data     = 0L;
 634         rif_timer.function = rif_check_expire;
 635         add_timer(&rif_timer);
 636
 637         proc_net_fops_create("tr_rif", S_IRUGO, &rif_seq_fops);
 638         return 0;
 639 }
 640
 641 module_init(rif_init);
 642
 643 EXPORT_SYMBOL(tr_source_route);
 644 EXPORT_SYMBOL(tr_type_trans);
 645 EXPORT_SYMBOL(alloc_trdev);