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 #include <net/net_namespace.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 {
  51         unsigned char addr[TR_ALEN];
  52         int iface;
  53         __be16 rcf;
  54         __be16 rseg[8];
  55         struct rif_cache *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 *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", ntohs(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 __be16 tr_type_trans(struct sk_buff *skb, struct net_device *dev)
 191 {
 192
 193         struct trh_hdr *trh;
 194         struct trllc *trllc;
 195         unsigned riflen=0;
 196
 197         skb->dev = dev;
 198         skb_reset_mac_header(skb);
 199         trh = tr_hdr(skb);
 200
 201         if(trh->saddr[0] & TR_RII)
 202                 riflen = (ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8;
 203
 204         trllc = (struct trllc *)(skb->data+sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
 205
 206         skb_pull(skb,sizeof(struct trh_hdr)-TR_MAXRIFLEN+riflen);
 207
 208         if(*trh->daddr & 0x80)
 209         {
 210                 if(!memcmp(trh->daddr,dev->broadcast,TR_ALEN))
 211                         skb->pkt_type=PACKET_BROADCAST;
 212                 else
 213                         skb->pkt_type=PACKET_MULTICAST;
 214         }
 215         else if ( (trh->daddr[0] & 0x01) && (trh->daddr[1] & 0x00) && (trh->daddr[2] & 0x5E))
 216         {
 217                 skb->pkt_type=PACKET_MULTICAST;
 218         }
 219         else if(dev->flags & IFF_PROMISC)
 220         {
 221                 if(memcmp(trh->daddr, dev->dev_addr, TR_ALEN))
 222                         skb->pkt_type=PACKET_OTHERHOST;
 223         }
 224
 225         if ((skb->pkt_type != PACKET_BROADCAST) &&
 226             (skb->pkt_type != PACKET_MULTICAST))
 227                 tr_add_rif_info(trh,dev) ;
 228
 229         /*
 230          * Strip the SNAP header from ARP packets since we don't
 231          * pass them through to the 802.2/SNAP layers.
 232          */
 233
 234         if (trllc->dsap == EXTENDED_SAP &&
 235             (trllc->ethertype == htons(ETH_P_IP) ||
 236              trllc->ethertype == htons(ETH_P_IPV6) ||
 237              trllc->ethertype == htons(ETH_P_ARP)))
 238         {
 239                 skb_pull(skb, sizeof(struct trllc));
 240                 return trllc->ethertype;
 241         }
 242
 243         return htons(ETH_P_TR_802_2);
 244 }
 245
 246 /*
 247  *      We try to do source routing...
 248  */
 249
 250 void tr_source_route(struct sk_buff *skb,struct trh_hdr *trh,struct net_device *dev)
 251 {
 252         int slack;
 253         unsigned int hash;
 254         struct rif_cache *entry;
 255         unsigned char *olddata;
 256         unsigned long flags;
 257         static const unsigned char mcast_func_addr[]
 258                 = {0xC0,0x00,0x00,0x04,0x00,0x00};
 259
 260         spin_lock_irqsave(&rif_lock, flags);
 261
 262         /*
 263          *      Broadcasts are single route as stated in RFC 1042
 264          */
 265         if( (!memcmp(&(trh->daddr[0]),&(dev->broadcast[0]),TR_ALEN)) ||
 266             (!memcmp(&(trh->daddr[0]),&(mcast_func_addr[0]), TR_ALEN))  )
 267         {
 268                 trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
 269                                | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
 270                 trh->saddr[0]|=TR_RII;
 271         }
 272         else
 273         {
 274                 hash = rif_hash(trh->daddr);
 275                 /*
 276                  *      Walk the hash table and look for an entry
 277                  */
 278                 for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->daddr[0]),TR_ALEN);entry=entry->next);
 279
 280                 /*
 281                  *      If we found an entry we can route the frame.
 282                  */
 283                 if(entry)
 284                 {
 285 #if TR_SR_DEBUG
 286 printk("source routing for %02X:%02X:%02X:%02X:%02X:%02X\n",trh->daddr[0],
 287                   trh->daddr[1],trh->daddr[2],trh->daddr[3],trh->daddr[4],trh->daddr[5]);
 288 #endif
 289                         if(!entry->local_ring && (ntohs(entry->rcf) & TR_RCF_LEN_MASK) >> 8)
 290                         {
 291                                 trh->rcf=entry->rcf;
 292                                 memcpy(&trh->rseg[0],&entry->rseg[0],8*sizeof(unsigned short));
 293                                 trh->rcf^=htons(TR_RCF_DIR_BIT);
 294                                 trh->rcf&=htons(0x1fff);        /* Issam Chehab <ichehab@madge1.demon.co.uk> */
 295
 296                                 trh->saddr[0]|=TR_RII;
 297 #if TR_SR_DEBUG
 298                                 printk("entry found with rcf %04x\n", entry->rcf);
 299                         }
 300                         else
 301                         {
 302                                 printk("entry found but without rcf length, local=%02x\n", entry->local_ring);
 303 #endif
 304                         }
 305                         entry->last_used=jiffies;
 306                 }
 307                 else
 308                 {
 309                         /*
 310                          *      Without the information we simply have to shout
 311                          *      on the wire. The replies should rapidly clean this
 312                          *      situation up.
 313                          */
 314                         trh->rcf=htons((((sizeof(trh->rcf)) << 8) & TR_RCF_LEN_MASK)
 315                                        | TR_RCF_FRAME2K | TR_RCF_LIMITED_BROADCAST);
 316                         trh->saddr[0]|=TR_RII;
 317 #if TR_SR_DEBUG
 318                         printk("no entry in rif table found - broadcasting frame\n");
 319 #endif
 320                 }
 321         }
 322
 323         /* Compress the RIF here so we don't have to do it in the driver(s) */
 324         if (!(trh->saddr[0] & 0x80))
 325                 slack = 18;
 326         else
 327                 slack = 18 - ((ntohs(trh->rcf) & TR_RCF_LEN_MASK)>>8);
 328         olddata = skb->data;
 329         spin_unlock_irqrestore(&rif_lock, flags);
 330
 331         skb_pull(skb, slack);
 332         memmove(skb->data, olddata, sizeof(struct trh_hdr) - slack);
 333 }
 334
 335 /*
 336  *      We have learned some new RIF information for our source
 337  *      routing.
 338  */
 339
 340 static void tr_add_rif_info(struct trh_hdr *trh, struct net_device *dev)
 341 {
 342         unsigned int hash, rii_p = 0;
 343         unsigned long flags;
 344         struct rif_cache *entry;
 345         unsigned char saddr0;
 346
 347         spin_lock_irqsave(&rif_lock, flags);
 348         saddr0 = trh->saddr[0];
 349
 350         /*
 351          *      Firstly see if the entry exists
 352          */
 353
 354         if(trh->saddr[0] & TR_RII)
 355         {
 356                 trh->saddr[0]&=0x7f;
 357                 if (((ntohs(trh->rcf) & TR_RCF_LEN_MASK) >> 8) > 2)
 358                 {
 359                         rii_p = 1;
 360                 }
 361         }
 362
 363         hash = rif_hash(trh->saddr);
 364         for(entry=rif_table[hash];entry && memcmp(&(entry->addr[0]),&(trh->saddr[0]),TR_ALEN);entry=entry->next);
 365
 366         if(entry==NULL)
 367         {
 368 #if TR_SR_DEBUG
 369 printk("adding rif_entry: addr:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
 370                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
 371                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
 372                 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:%02X:%02X:%02X:%02X:%02X:%02X rcf:%04X\n",
 418                 trh->saddr[0],trh->saddr[1],trh->saddr[2],
 419                 trh->saddr[3],trh->saddr[4],trh->saddr[5],
 420                 ntohs(trh->rcf));
 421 #endif
 422                             entry->rcf = trh->rcf & htons((unsigned short)~TR_RCF_BROADCAST_MASK);
 423                             memcpy(&(entry->rseg[0]),&(trh->rseg[0]),8*sizeof(unsigned short));
 424                     }
 425                 entry->last_used=jiffies;
 426         }
 427         trh->saddr[0]=saddr0; /* put the routing indicator back for tcpdump */
 428         spin_unlock_irqrestore(&rif_lock, flags);
 429 }
 430
 431 /*
 432  *      Scan the cache with a timer and see what we need to throw out.
 433  */
 434
 435 static void rif_check_expire(unsigned long dummy)
 436 {
 437         int i;
 438         unsigned long flags, next_interval = jiffies + sysctl_tr_rif_timeout/2;
 439
 440         spin_lock_irqsave(&rif_lock, flags);
 441
 442         for(i =0; i < RIF_TABLE_SIZE; i++) {
 443                 struct rif_cache *entry, **pentry;
 444
 445                 pentry = rif_table+i;
 446                 while((entry=*pentry) != NULL) {
 447                         unsigned long expires
 448                                 = entry->last_used + sysctl_tr_rif_timeout;
 449
 450                         if (time_before_eq(expires, jiffies)) {
 451                                 *pentry = entry->next;
 452                                 kfree(entry);
 453                         } else {
 454                                 pentry = &entry->next;
 455
 456                                 if (time_before(expires, next_interval))
 457                                         next_interval = expires;
 458                         }
 459                 }
 460         }
 461
 462         spin_unlock_irqrestore(&rif_lock, flags);
 463
 464         mod_timer(&rif_timer, next_interval);
 465
 466 }
 467
 468 /*
 469  *      Generate the /proc/net information for the token ring RIF
 470  *      routing.
 471  */
 472
 473 #ifdef CONFIG_PROC_FS
 474
 475 static struct rif_cache *rif_get_idx(loff_t pos)
 476 {
 477         int i;
 478         struct rif_cache *entry;
 479         loff_t off = 0;
 480
 481         for(i = 0; i < RIF_TABLE_SIZE; i++)
 482                 for(entry = rif_table[i]; entry; entry = entry->next) {
 483                         if (off == pos)
 484                                 return entry;
 485                         ++off;
 486                 }
 487
 488         return NULL;
 489 }
 490
 491 static void *rif_seq_start(struct seq_file *seq, loff_t *pos)
 492 {
 493         spin_lock_irq(&rif_lock);
 494
 495         return *pos ? rif_get_idx(*pos - 1) : SEQ_START_TOKEN;
 496 }
 497
 498 static void *rif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 499 {
 500         int i;
 501         struct rif_cache *ent = v;
 502
 503         ++*pos;
 504
 505         if (v == SEQ_START_TOKEN) {
 506                 i = -1;
 507                 goto scan;
 508         }
 509
 510         if (ent->next)
 511                 return ent->next;
 512
 513         i = rif_hash(ent->addr);
 514  scan:
 515         while (++i < RIF_TABLE_SIZE) {
 516                 if ((ent = rif_table[i]) != NULL)
 517                         return ent;
 518         }
 519         return NULL;
 520 }
 521
 522 static void rif_seq_stop(struct seq_file *seq, void *v)
 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 %02X:%02X:%02X:%02X:%02X:%02X %7li ",
 541                            dev?dev->name:"?",
 542                            entry->addr[0],entry->addr[1],entry->addr[2],
 543                            entry->addr[3],entry->addr[4],entry->addr[5],
 544                            ttl/HZ);
 545
 546                         if (entry->local_ring)
 547                                 seq_puts(seq, "local\n");
 548                         else {
 549
 550                                 seq_printf(seq, "%04X", ntohs(entry->rcf));
 551                                 rcf_len = ((ntohs(entry->rcf) & TR_RCF_LEN_MASK)>>8)-2;
 552                                 if (rcf_len)
 553                                         rcf_len >>= 1;
 554                                 for(j = 1; j < rcf_len; j++) {
 555                                         if(j==1) {
 556                                                 segment=ntohs(entry->rseg[j-1])>>4;
 557                                                 seq_printf(seq,"  %03X",segment);
 558                                         }
 559
 560                                         segment=ntohs(entry->rseg[j])>>4;
 561                                         brdgnmb=ntohs(entry->rseg[j-1])&0x00f;
 562                                         seq_printf(seq,"-%01X-%03X",brdgnmb,segment);
 563                                 }
 564                                 seq_putc(seq, '\n');
 565                         }
 566                 }
 567         return 0;
 568 }
 569
 570
 571 static const struct seq_operations rif_seq_ops = {
 572         .start = rif_seq_start,
 573         .next  = rif_seq_next,
 574         .stop  = rif_seq_stop,
 575         .show  = rif_seq_show,
 576 };
 577
 578 static int rif_seq_open(struct inode *inode, struct file *file)
 579 {
 580         return seq_open(file, &rif_seq_ops);
 581 }
 582
 583 static const struct file_operations rif_seq_fops = {
 584         .owner   = THIS_MODULE,
 585         .open    = rif_seq_open,
 586         .read    = seq_read,
 587         .llseek  = seq_lseek,
 588         .release = seq_release,
 589 };
 590
 591 #endif
 592
 593 static void tr_setup(struct net_device *dev)
 594 {
 595         /*
 596          *      Configure and register
 597          */
 598
 599         dev->hard_header        = tr_header;
 600         dev->rebuild_header     = tr_rebuild_header;
 601
 602         dev->type               = ARPHRD_IEEE802_TR;
 603         dev->hard_header_len    = TR_HLEN;
 604         dev->mtu                = 2000;
 605         dev->addr_len           = TR_ALEN;
 606         dev->tx_queue_len       = 100;  /* Long queues on tr */
 607
 608         memset(dev->broadcast,0xFF, TR_ALEN);
 609
 610         /* New-style flags. */
 611         dev->flags              = IFF_BROADCAST | IFF_MULTICAST ;
 612 }
 613
 614 /**
 615  * alloc_trdev - Register token ring device
 616  * @sizeof_priv: Size of additional driver-private structure to be allocated
 617  *      for this token ring device
 618  *
 619  * Fill in the fields of the device structure with token ring-generic values.
 620  *
 621  * Constructs a new net device, complete with a private data area of
 622  * size @sizeof_priv.  A 32-byte (not bit) alignment is enforced for
 623  * this private data area.
 624  */
 625 struct net_device *alloc_trdev(int sizeof_priv)
 626 {
 627         return alloc_netdev(sizeof_priv, "tr%d", tr_setup);
 628 }
 629
 630 /*
 631  *      Called during bootup.  We don't actually have to initialise
 632  *      too much for this.
 633  */
 634
 635 static int __init rif_init(void)
 636 {
 637         init_timer(&rif_timer);
 638         rif_timer.expires  = sysctl_tr_rif_timeout;
 639         rif_timer.data     = 0L;
 640         rif_timer.function = rif_check_expire;
 641         add_timer(&rif_timer);
 642
 643         proc_net_fops_create(&init_net, "tr_rif", S_IRUGO, &rif_seq_fops);
 644         return 0;
 645 }
 646
 647 module_init(rif_init);
 648
 649 EXPORT_SYMBOL(tr_type_trans);
 650 EXPORT_SYMBOL(alloc_trdev);