root/net/inet/ip.c

/* */

DEFINITIONS

1. ip_print
2. ip_ioctl
3. strict_route
4. loose_route
5. print_ipprot
6. ip_route_check
7. build_options
8. ip_send
9. ip_build_header
10. do_options
11. ip_compute_csum
12. ip_csum
13. ip_send_check
14. ip_forward
15. ip_rcv
16. ip_queue_xmit
17. ip_retransmit

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              The Internet Protocol (IP) module.
   7  *
   8  * Version:     @(#)ip.c        1.0.16  06/02/93
   9  *
  10  * Authors:     Ross Biro, <bir7@leland.Stanford.Edu>
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *
  13  *              This program is free software; you can redistribute it and/or
  14  *              modify it under the terms of the GNU General Public License
  15  *              as published by the Free Software Foundation; either version
  16  *              2 of the License, or (at your option) any later version.
  17  */
  18 #include <asm/segment.h>
  19 #include <asm/system.h>
  20 #include <linux/types.h>
  21 #include <linux/kernel.h>
  22 #include <linux/sched.h>
  23 #include <linux/string.h>
  24 #include <linux/errno.h>
  25 #include <linux/socket.h>
  26 #include <linux/sockios.h>
  27 #include <linux/in.h>
  28 #include "inet.h"
  29 #include "timer.h"
  30 #include "dev.h"
  31 #include "eth.h"
  32 #include "ip.h"
  33 #include "protocol.h"
  34 #include "route.h"
  35 #include "tcp.h"
  36 #include "skbuff.h"
  37 #include "sock.h"
  38 #include "arp.h"
  39 #include "icmp.h"
  40 
  41 
  42 void
  43 ip_print(struct iphdr *ip)
     /*  */
  44 {
  45   unsigned char buff[32];
  46   unsigned char *ptr;
  47   int addr, len, i;
  48 
  49   if (inet_debug != DBG_IP) return;
  50 
  51   /* Dump the IP header. */
  52   printk("IP: ihl=%d, version=%d, tos=%d, tot_len=%d\n",
  53            ip->ihl, ip->version, ip->tos, ntohs(ip->tot_len));
  54   printk("    id=%X, ttl=%d, prot=%d, check=%X\n",
  55            ip->id, ip->ttl, ip->protocol, ip->check);
  56   printk("    frag_off=%d\n", ip->frag_off);
  57   printk("    soucre=%s ", in_ntoa(ip->saddr));
  58   printk("dest=%s\n", in_ntoa(ip->daddr));
  59   printk("    ----\n");
  60 
  61   /* Dump the data. */
  62   ptr = (unsigned char *)(ip + 1);
  63   addr = 0;
  64   len = ntohs(ip->tot_len) - (4 * ip->ihl);
  65   while (len > 0) {
  66         printk("    %04X: ", addr);
  67         for(i = 0; i < 16; i++) {
  68                 if (len > 0) {
  69                         printk("%02X ", (*ptr & 0xFF));
  70                         buff[i] = *ptr++;
  71                         if (buff[i] < 32 || buff[i] > 126) buff[i] = '.';
  72                 } else {
  73                         printk("   ");
  74                         buff[i] = ' ';
  75                 }
  76                 addr++;
  77                 len--;
  78         };
  79         buff[i] = '\0';
  80         printk("  \"%s\"\n", buff);
  81   }
  82   printk("    ----\n\n");
  83 }
  84 
  85 
  86 int
  87 ip_ioctl(struct sock *sk, int cmd, unsigned long arg)
     /*  */
  88 {
  89   switch(cmd) {
  90         case DDIOCSDBG:
  91                 return(dbg_ioctl((void *) arg, DBG_IP));
  92         default:
  93                 return(-EINVAL);
  94   }
  95 }
  96 
  97 
  98 /* these two routines will do routining. */
  99 static void
 100 strict_route(struct iphdr *iph, struct options *opt)
     /*  */
 101 {
 102 }
 103 
 104 
 105 static void
 106 loose_route(struct iphdr *iph, struct options *opt)
     /*  */
 107 {
 108 }
 109 
 110 
 111 static void
 112 print_ipprot(struct inet_protocol *ipprot)
     /*  */
 113 {
 114   DPRINTF((DBG_IP, "handler = %X, protocol = %d, copy=%d \n",
 115            ipprot->handler, ipprot->protocol, ipprot->copy));
 116 }
 117 
 118 
 119 /* This routine will check to see if we have lost a gateway. */
 120 void
 121 ip_route_check(unsigned long daddr)
     /*  */
 122 {
 123 }
 124 
 125 
 126 #if 0
 127 /* this routine puts the options at the end of an ip header. */
 128 static int
 129 build_options(struct iphdr *iph, struct options *opt)
     /*  */
 130 {
 131   unsigned char *ptr;
 132   /* currently we don't support any options. */
 133   ptr = (unsigned char *)(iph+1);
 134   *ptr = 0;
 135   return (4);
 136 }
 137 #endif
 138 
 139 
 140 /* Take an skb, and fill in the MAC header. */
 141 static int
 142 ip_send(struct sk_buff *skb, unsigned long daddr, int len, struct device *dev,
     /*  */
 143         unsigned long saddr)
 144 {
 145   unsigned char *ptr;
 146   int mac;
 147 
 148   ptr = (unsigned char *)(skb + 1);
 149   mac = 0;
 150   skb->arp = 1;
 151   if (dev->hard_header) {
 152         mac = dev->hard_header(ptr, dev, ETH_P_IP, daddr, saddr, len);
 153   }
 154   if (mac < 0) {
 155         mac = -mac;
 156         skb->arp = 0;
 157   }
 158   skb->dev = dev;
 159   return(mac);
 160 }
 161 
 162 
 163 /*
 164  * This routine builds the appropriate hardware/IP headers for
 165  * the routine.  It assumes that if *dev != NULL then the
 166  * protocol knows what it's doing, otherwise it uses the
 167  * routing/ARP tables to select a device struct.
 168  */
 169 int
 170 ip_build_header(struct sk_buff *skb, unsigned long saddr, unsigned long daddr,
     /*  */
 171                 struct device **dev, int type, struct options *opt, int len)
 172 {
 173   static struct options optmem;
 174   struct iphdr *iph;
 175   struct rtable *rt;
 176   unsigned char *buff;
 177   unsigned long raddr;
 178   static int count = 0;
 179   int tmp;
 180 
 181   if (saddr == 0) saddr = my_addr();
 182   DPRINTF((DBG_IP, "ip_build_header (skb=%X, saddr=%X, daddr=%X, *dev=%X,\n"
 183            "                 type=%d, opt=%X, len = %d)\n",
 184            skb, saddr, daddr, *dev, type, opt, len));
 185   buff = (unsigned char *)(skb + 1);
 186 
 187   /* See if we need to look up the device. */
 188   if (*dev == NULL) {
 189         rt = rt_route(daddr, &optmem);
 190         if (rt == NULL) return(-ENETUNREACH);
 191 
 192         *dev = rt->rt_dev;
 193         if (daddr != 0x0100007F) saddr = rt->rt_dev->pa_addr;
 194         raddr = rt->rt_gateway;
 195 
 196         DPRINTF((DBG_IP, "ip_build_header: saddr set to %s\n", in_ntoa(saddr)));
 197         opt = &optmem;
 198   } else {
 199         /* We still need the address of the first hop. */
 200         rt = rt_route(daddr, &optmem);
 201         raddr = (rt == NULL) ? 0 : rt->rt_gateway;
 202   }
 203   if (raddr == 0) raddr = daddr;
 204 
 205   /* Now build the MAC header. */
 206   tmp = ip_send(skb, raddr, len, *dev, saddr);
 207   buff += tmp;
 208   len -= tmp;
 209 
 210   skb->dev = *dev;
 211   skb->saddr = saddr;
 212   if (skb->sk) skb->sk->saddr = saddr;
 213 
 214   /* Now build the IP header. */
 215   iph = (struct iphdr *)buff;
 216   iph->version  = 4;
 217   iph->tos      = 0;
 218   iph->frag_off = 0;
 219   iph->ttl      = 32;
 220   iph->daddr    = daddr;
 221   iph->saddr    = saddr;
 222   iph->protocol = type;
 223   iph->ihl      = 5;
 224   iph->id       = htons(count++);
 225 
 226   /* Setup the IP options. */
 227 #ifdef Not_Yet_Avail
 228   build_options(iph, opt);
 229 #endif
 230 
 231   return(20 + tmp);     /* IP header plus MAC header size */
 232 }
 233 
 234 
 235 static int
 236 do_options(struct iphdr *iph, struct options *opt)
     /*  */
 237 {
 238   unsigned char *buff;
 239   int done = 0;
 240   int i, len = sizeof(struct iphdr);
 241 
 242   /* Zero out the options. */
 243   opt->record_route.route_size = 0;
 244   opt->loose_route.route_size  = 0;
 245   opt->strict_route.route_size = 0;
 246   opt->tstamp.ptr              = 0;
 247   opt->security                = 0;
 248   opt->compartment             = 0;
 249   opt->handling                = 0;
 250   opt->stream                  = 0;
 251   opt->tcc                     = 0;
 252   return(0);
 253 
 254   /* Advance the pointer to start at the options. */
 255   buff = (unsigned char *)(iph + 1);
 256 
 257   /* Now start the processing. */
 258   while (!done && len < iph->ihl*4) switch(*buff) {
 259         case IPOPT_END:
 260                 done = 1;
 261                 break;
 262         case IPOPT_NOOP:
 263                 buff++;
 264                 len++;
 265                 break;
 266         case IPOPT_SEC:
 267                 buff++;
 268                 if (*buff != 11) return(1);
 269                 buff++;
 270                 opt->security = ntohs(*(unsigned short *)buff);
 271                 buff += 2;
 272                 opt->compartment = ntohs(*(unsigned short *)buff);
 273                 buff += 2;
 274                 opt->handling = ntohs(*(unsigned short *)buff);
 275                 buff += 2;
 276                 opt->tcc = ((*buff) << 16) + ntohs(*(unsigned short *)(buff+1));
 277                 buff += 3;
 278                 len += 11;
 279                 break;
 280         case IPOPT_LSRR:
 281                 buff++;
 282                 if ((*buff - 3)% 4 != 0) return(1);
 283                 len += *buff;
 284                 opt->loose_route.route_size = (*buff -3)/4;
 285                 buff++;
 286                 if (*buff % 4 != 0) return(1);
 287                 opt->loose_route.pointer = *buff/4 - 1;
 288                 buff++;
 289                 buff++;
 290                 for (i = 0; i < opt->loose_route.route_size; i++) {
 291                         opt->loose_route.route[i] = *(unsigned long *)buff;
 292                         buff += 4;
 293                 }
 294                 break;
 295         case IPOPT_SSRR:
 296                 buff++;
 297                 if ((*buff - 3)% 4 != 0) return(1);
 298                 len += *buff;
 299                 opt->strict_route.route_size = (*buff -3)/4;
 300                 buff++;
 301                 if (*buff % 4 != 0) return(1);
 302                 opt->strict_route.pointer = *buff/4 - 1;
 303                 buff++;
 304                 buff++;
 305                 for (i = 0; i < opt->strict_route.route_size; i++) {
 306                         opt->strict_route.route[i] = *(unsigned long *)buff;
 307                         buff += 4;
 308                 }
 309                 break;
 310         case IPOPT_RR:
 311                 buff++;
 312                 if ((*buff - 3)% 4 != 0) return(1);
 313                 len += *buff;
 314                 opt->record_route.route_size = (*buff -3)/4;
 315                 buff++;
 316                 if (*buff % 4 != 0) return(1);
 317                 opt->record_route.pointer = *buff/4 - 1;
 318                 buff++;
 319                 buff++;
 320                 for (i = 0; i < opt->record_route.route_size; i++) {
 321                         opt->record_route.route[i] = *(unsigned long *)buff;
 322                         buff += 4;
 323                 }
 324                 break;
 325         case IPOPT_SID:
 326                 len += 4;
 327                 buff +=2;
 328                 opt->stream = *(unsigned short *)buff;
 329                 buff += 2;
 330                 break;
 331         case IPOPT_TIMESTAMP:
 332                 buff++;
 333                 len += *buff;
 334                 if (*buff % 4 != 0) return(1);
 335                 opt->tstamp.len = *buff / 4 - 1;
 336                 buff++;
 337                 if ((*buff - 1) % 4 != 0) return(1);
 338                 opt->tstamp.ptr = (*buff-1)/4;
 339                 buff++;
 340                 opt->tstamp.x.full_char = *buff;
 341                 buff++;
 342                 for (i = 0; i < opt->tstamp.len; i++) {
 343                         opt->tstamp.data[i] = *(unsigned long *)buff;
 344                         buff += 4;
 345                 }
 346                 break;
 347         default:
 348                 return(1);
 349   }
 350 
 351   if (opt->record_route.route_size == 0) {
 352         if (opt->strict_route.route_size != 0) {
 353                 memcpy(&(opt->record_route), &(opt->strict_route),
 354                                              sizeof(opt->record_route));
 355         } else if (opt->loose_route.route_size != 0) {
 356                 memcpy(&(opt->record_route), &(opt->loose_route),
 357                                              sizeof(opt->record_route));
 358         }
 359   }
 360 
 361   if (opt->strict_route.route_size != 0 &&
 362       opt->strict_route.route_size != opt->strict_route.pointer) {
 363         strict_route(iph, opt);
 364         return(0);
 365   }
 366 
 367   if (opt->loose_route.route_size != 0 &&
 368       opt->loose_route.route_size != opt->loose_route.pointer) {
 369         loose_route(iph, opt);
 370         return(0);
 371   }
 372 
 373   return(0);
 374 }
 375 
 376 
 377 /*
 378  * This routine does all the checksum computations that don't
 379  * require anything special (like copying or special headers).
 380  */
 381 unsigned short
 382 ip_compute_csum(unsigned char * buff, int len)
     /*  */
 383 {
 384   unsigned long sum = 0;
 385 
 386   /* Do the first multiple of 4 bytes and convert to 16 bits. */
 387   if (len > 3) {
 388         __asm__("\t clc\n"
 389                 "1:\n"
 390                 "\t lodsl\n"
 391                 "\t adcl %%eax, %%ebx\n"
 392                 "\t loop 1b\n"
 393                 "\t adcl $0, %%ebx\n"
 394                 "\t movl %%ebx, %%eax\n"
 395                 "\t shrl $16, %%eax\n"
 396                 "\t addw %%ax, %%bx\n"
 397                 "\t adcw $0, %%bx\n"
 398                 : "=b" (sum) , "=S" (buff)
 399                 : "0" (sum), "c" (len >> 2) ,"1" (buff)
 400                 : "ax", "cx", "si", "bx" );
 401   }
 402   if (len & 2) {
 403         __asm__("\t lodsw\n"
 404                 "\t addw %%ax, %%bx\n"
 405                 "\t adcw $0, %%bx\n"
 406                 : "=b" (sum), "=S" (buff)
 407                 : "0" (sum), "1" (buff)
 408                 : "bx", "ax", "si");
 409   }
 410   if (len & 1) {
 411         __asm__("\t lodsb\n"
 412                 "\t movb $0, %%ah\n"
 413                 "\t addw %%ax, %%bx\n"
 414                 "\t adcw $0, %%bx\n"
 415                 : "=b" (sum), "=S" (buff)
 416                 : "0" (sum), "1" (buff)
 417                 : "bx", "ax", "si");
 418   }
 419   sum =~sum;
 420   return(sum & 0xffff);
 421 }
 422 
 423 
 424 /* Check the header of an incoming IP datagram. */
 425 static int
 426 ip_csum(struct iphdr *iph)
     /*  */
 427 {
 428   if (iph->check == 0) return(0);
 429   if (ip_compute_csum((unsigned char *)iph, iph->ihl*4) == 0) return(0);
 430   return(1);
 431 }
 432 
 433 
 434 /* Generate a checksym for an outgoing IP datagram. */
 435 static void
 436 ip_send_check(struct iphdr *iph)
     /*  */
 437 {
 438    iph->check = 0;
 439    iph->check = ip_compute_csum((unsigned char *)iph, iph->ihl*4);
 440 }
 441 
 442 
 443 /* Forward an IP datagram to its next destination. */
 444 static void
 445 ip_forward(struct sk_buff *skb, struct device *dev)
     /*  */
 446 {
 447   struct device *dev2;
 448   struct iphdr *iph;
 449   struct sk_buff *skb2;
 450   struct rtable *rt;
 451   unsigned char *ptr;
 452   unsigned long raddr;
 453 
 454   /*
 455    * According to the RFC, we must first decrease the TTL field. If
 456    * that reaches zero, we must reply an ICMP control message telling
 457    * that the packet's lifetime expired.
 458    */
 459   iph = skb->h.iph;
 460   iph->ttl--;
 461   if (iph->ttl <= 0) {
 462         DPRINTF((DBG_IP, "\nIP: *** datagram expired: TTL=0 (ignored) ***\n"));
 463         DPRINTF((DBG_IP, "    SRC = %s   ", in_ntoa(iph->saddr)));
 464         DPRINTF((DBG_IP, "    DST = %s (ignored)\n", in_ntoa(iph->daddr)));
 465 
 466         /* Tell the sender its packet died... */
 467         icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, dev);
 468         return;
 469   }
 470 
 471   /* Re-compute the IP header checksum. */
 472   ip_send_check(iph);
 473 
 474   /*
 475    * OK, the packet is still valid.  Fetch its destination address,
 476    * and give it to the IP sender for further processing.
 477    */
 478   rt = rt_route(iph->daddr, NULL);
 479   if (rt == NULL) {
 480         DPRINTF((DBG_IP, "\nIP: *** routing (phase I) failed ***\n"));
 481 
 482         /* Tell the sender its packet cannot be delivered... */
 483         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_NET_UNREACH, dev);
 484         return;
 485   }
 486 
 487   /*
 488    * Gosh.  Not only is the packet valid; we even know how to
 489    * forward it onto its final destination.  Can we say this
 490    * is being plain lucky?
 491    * If the router told us that there is no GW, use the dest.
 492    * IP address itself- we seem to be connected directly...
 493    */
 494   raddr = rt->rt_gateway;
 495   if (raddr != 0) {
 496         rt = rt_route(raddr, NULL);
 497         if (rt == NULL) {
 498                 DPRINTF((DBG_IP, "\nIP: *** routing (phase II) failed ***\n"));
 499 
 500                 /* Tell the sender its packet cannot be delivered... */
 501                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, dev);
 502                 return;
 503         }
 504         if (rt->rt_gateway != 0) raddr = rt->rt_gateway;
 505   } else raddr = iph->daddr;
 506   dev2 = rt->rt_dev;
 507 
 508   /*
 509    * We now allocate a new buffer, and copy the datagram into it.
 510    * If the indicated interface is up and running, kick it.
 511    */
 512   DPRINTF((DBG_IP, "\nIP: *** fwd %s -> ", in_ntoa(iph->saddr)));
 513   DPRINTF((DBG_IP, "%s (via %s), LEN=%d\n",
 514                         in_ntoa(raddr), dev2->name, skb->len));
 515 
 516   if (dev2->flags & IFF_UP) {
 517         skb2 = kmalloc(sizeof(struct sk_buff) +
 518                        dev2->hard_header_len + skb->len, GFP_ATOMIC);
 519         if (skb2 == NULL) {
 520                 printk("\nIP: No memory available for IP forward\n");
 521                 return;
 522         }
 523         ptr = (unsigned char *)(skb2 + 1);
 524         skb2->lock = 0;
 525         skb2->sk = NULL;
 526         skb2->len = skb->len + dev2->hard_header_len;
 527         skb2->mem_addr = skb2;
 528         skb2->mem_len = sizeof(struct sk_buff) + skb2->len;
 529         skb2->next = NULL;
 530         skb2->h.raw = ptr;
 531 
 532         /* Copy the packet data into the new buffer. */
 533         skb2->h.raw = ptr;
 534         memcpy(ptr + dev2->hard_header_len, skb->h.raw, skb->len);
 535                 
 536         /* Now build the MAC header. */
 537         (void) ip_send(skb2, raddr, skb->len, dev2, dev2->pa_addr);
 538 
 539         dev2->queue_xmit(skb2, dev2, SOPRI_NORMAL);
 540   }
 541 }
 542 
 543 
 544 /* This function receives all incoming IP datagrams. */
 545 int
 546 ip_rcv(struct sk_buff *skb, struct device *dev, struct packet_type *pt)
     /*  */
 547 {
 548   struct iphdr *iph;
 549   unsigned char hash;
 550   unsigned char flag = 0;
 551   struct inet_protocol *ipprot;
 552   static struct options opt; /* since we don't use these yet, and they
 553                                 take up stack space. */
 554   int brd;
 555 
 556   iph = skb->h.iph;
 557   memset((char *) &opt, 0, sizeof(opt));
 558   DPRINTF((DBG_IP, "<<\n"));
 559   ip_print(iph);
 560 
 561   /* Is the datagram acceptable? */
 562   if (ip_csum(iph) || do_options(iph, &opt) || iph->version != 4) {
 563         DPRINTF((DBG_IP, "\nIP: *** datagram error ***\n"));
 564         DPRINTF((DBG_IP, "    SRC = %s   ", in_ntoa(iph->saddr)));
 565         DPRINTF((DBG_IP, "    DST = %s (ignored)\n", in_ntoa(iph->daddr)));
 566         skb->sk = NULL;
 567         kfree_skb(skb, FREE_WRITE);
 568         return(0);
 569   }
 570 
 571   /* Do any IP forwarding required. */
 572   if ((brd = chk_addr(iph->daddr)) == 0) {
 573         ip_forward(skb, dev);
 574         skb->sk = NULL;
 575         kfree_skb(skb, FREE_WRITE);
 576         return(0);
 577   }
 578 
 579   /*
 580    * Deal with fragments: not really...
 581    * Fragmentation is definitely a required part of IP (yeah, guys,
 582    * I read Linux-Activists.NET too :-), but the current "sk_buff"
 583    * allocation stuff doesn't make things simpler.  When we're all
 584    * done cleaning up the mess, we'll add Ross Biro's "mbuf" stuff
 585    * to the code, which will replace the sk_buff stuff completely.
 586    * That will (a) make the code even cleaner, (b) allow me to do
 587    * the DDI (Device Driver Interface) the way I want to, and (c),
 588    * it will allow for easy addition of fragging.  Any takers? -FvK
 589    */
 590   if ((iph->frag_off & 32) || (ntohs(iph->frag_off) & 0x1fff)) {
 591         printk("\nIP: *** datagram fragmentation not yet implemented ***\n");
 592         printk("    SRC = %s   ", in_ntoa(iph->saddr));
 593         printk("    DST = %s (ignored)\n", in_ntoa(iph->daddr));
 594         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, dev);
 595         skb->sk = NULL;
 596         kfree_skb(skb, FREE_WRITE);
 597         return(0);
 598   }
 599 
 600   /* Point into the IP datagram, just past the header. */
 601   skb->h.raw += iph->ihl*4;
 602   hash = iph->protocol & (MAX_INET_PROTOS -1);
 603   for (ipprot = (struct inet_protocol *)inet_protos[hash];
 604        ipprot != NULL;
 605        ipprot=(struct inet_protocol *)ipprot->next)
 606     {
 607        struct sk_buff *skb2;
 608 
 609        if (ipprot->protocol != iph->protocol) continue;
 610        DPRINTF((DBG_IP, "Using protocol = %X:\n", ipprot));
 611        print_ipprot(ipprot);
 612 
 613        /*
 614         * See if we need to make a copy of it.  This will
 615         * only be set if more than one protpocol wants it. 
 616         * and then not for the last one.
 617         */
 618        if (ipprot->copy) {
 619                 skb2 = kmalloc (skb->mem_len, GFP_ATOMIC);
 620                 if (skb2 == NULL) continue;
 621                 memcpy(skb2, skb, skb->mem_len);
 622                 skb2->mem_addr = skb2;
 623                 skb2->lock = 0;
 624                 skb2->h.raw = (void *)(
 625                                 (unsigned long)skb2 +
 626                                 (unsigned long) skb->h.raw -
 627                                 (unsigned long)skb);
 628         } else {
 629                 skb2 = skb;
 630         }
 631         flag = 1;
 632 
 633        /*
 634         * Pass on the datagram to each protocol that wants it,
 635         * based on the datagram protocol.  We should really
 636         * check the protocol handler's return values here...
 637         */
 638         ipprot->handler(skb2, dev, &opt, iph->daddr,
 639                         (ntohs(iph->tot_len) - (iph->ihl * 4)),
 640                         iph->saddr, 0, ipprot);
 641 
 642   }
 643 
 644   /*
 645    * All protocols checked.
 646    * If this packet was a broadcast, we may *not* reply to it, since that
 647    * causes (proven, grin) ARP storms and a leakage of memory (i.e. all
 648    * ICMP reply messages get queued up for transmission...)
 649    */
 650   if (!flag) {
 651         if (brd != IS_BROADCAST)
 652                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, dev);
 653         skb->sk = NULL;
 654         kfree_skb(skb, FREE_WRITE);
 655   }
 656 
 657   return(0);
 658 }
 659 
 660 
 661 /*
 662  * Queues a packet to be sent, and starts the transmitter
 663  * if necessary.  if free = 1 then we free the block after
 664  * transmit, otherwise we don't.
 665  * This routine also needs to put in the total length, and
 666  * compute the checksum.
 667  */
 668 void
 669 ip_queue_xmit(struct sock *sk, struct device *dev, 
     /*  */
 670               struct sk_buff *skb, int free)
 671 {
 672   struct iphdr *iph;
 673   unsigned char *ptr;
 674 
 675   if (sk == NULL) free = 1;
 676   if (dev == NULL) {
 677         printk("IP: ip_queue_xmit dev = NULL\n");
 678         return;
 679   }
 680   skb->free = free;
 681   skb->dev = dev;
 682   skb->when = jiffies;
 683 
 684   DPRINTF((DBG_IP, ">>\n"));
 685   ptr = (unsigned char *)(skb + 1);
 686   ptr += dev->hard_header_len;
 687   iph = (struct iphdr *)ptr;
 688   iph->tot_len = ntohs(skb->len - dev->hard_header_len);
 689   ip_send_check(iph);
 690   ip_print(iph);
 691   skb->next = NULL;
 692 
 693   /* See if this is the one trashing our queue. Ross? */
 694   skb->magic = 1;
 695   if (!free) {
 696         skb->link3 = NULL;
 697         sk->packets_out++;
 698         cli();
 699         if (sk->send_head == NULL) {
 700                 sk->send_tail = skb;
 701                 sk->send_head = skb;
 702         } else {
 703                 /* See if we've got a problem. */
 704                 if (sk->send_tail == NULL) {
 705                         extern void sort_send(volatile struct sock *sk);
 706 
 707                         printk("IP: ***bug sk->send_tail == NULL != sk->send_head\n");
 708                         sort_send(sk);
 709                 } else {
 710                         sk->send_tail->link3 = skb;
 711                         sk->send_tail = skb;
 712                 }
 713         }
 714         sti();
 715         sk->time_wait.len = sk->rtt<<1;
 716         sk->timeout = TIME_WRITE;
 717         reset_timer ((struct timer *)&sk->time_wait);
 718   } else {
 719         skb->sk = sk;
 720   }
 721 
 722   /* If the indicated interface is up and running, kick it. */
 723   if (dev->flags & IFF_UP) {
 724         if (sk != NULL) {
 725                 dev->queue_xmit(skb, dev, sk->priority);
 726         } else {
 727                 dev->queue_xmit(skb, dev, SOPRI_NORMAL);
 728         }
 729   } else {
 730         if (free) kfree_skb(skb, FREE_WRITE);
 731   }
 732 }
 733 
 734 
 735 void
 736 ip_retransmit(struct sock *sk, int all)
     /*  */
 737 {
 738   struct sk_buff * skb;
 739   struct proto *prot;
 740   struct device *dev;
 741 
 742   prot = sk->prot;
 743   skb = sk->send_head;
 744   while (skb != NULL) {
 745         dev = skb->dev;
 746 
 747         /*
 748          * The rebuild_header function sees if the ARP is done.
 749          * If not it sends a new ARP request, and if so it builds
 750          * the header.
 751          */
 752         if (!skb->arp) {
 753                 if (dev->rebuild_header((struct enet_header *)(skb+1),dev)) {
 754                         if (!all) break;
 755                         skb = (struct sk_buff *)skb->link3;
 756                         continue;
 757                 }
 758         }
 759         skb->arp = 1;
 760         skb->when = jiffies;
 761 
 762         /* If the interface is (still) up and running, kick it. */
 763         if (dev->flags & IFF_UP) {
 764                 if (sk) dev->queue_xmit(skb, dev, sk->priority);
 765                   else dev->queue_xmit(skb, dev, SOPRI_NORMAL );
 766         }
 767 
 768         sk->retransmits++;
 769         sk->prot->retransmits ++;
 770         if (!all) break;
 771 
 772         /* This should cut it off before we send too many packets. */
 773         if (sk->retransmits > sk->cong_window) break;
 774         skb = (struct sk_buff *)skb->link3;
 775   }
 776 
 777   /*
 778    * Double the RTT time every time we retransmit. 
 779    * This will cause exponential back off on how hard we try to
 780    * get through again.  Once we get through, the rtt will settle
 781    * back down reasonably quickly.
 782    */
 783   sk->rtt *= 2;
 784   sk->time_wait.len = sk->rtt;
 785   sk->timeout = TIME_WRITE;
 786   reset_timer((struct timer *)&sk->time_wait);
 787 }

/* */