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 1.0.16b 9/1/93 9 * 10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu> 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Donald Becker, <becker@super.org> 13 * Alan Cox, <gw4pts@gw4pts.ampr.org> 14 * 15 * Fixes: 16 * Alan Cox : Commented a couple of minor bits of surplus code 17 * Alan Cox : Undefining IP_FORWARD doesn't include the code 18 * (just stops a compiler warning). 19 * Alan Cox : Frames with >=MAX_ROUTE record routes, strict routes or loose routes 20 * are junked rather than corrupting things. 21 * Alan Cox : Frames to bad broadcast subnets are dumped 22 * We used to process them non broadcast and 23 * boy could that cause havoc. 24 * Alan Cox : ip_forward sets the free flag on the 25 * new frame it queues. Still crap because 26 * it copies the frame but at least it 27 * doesn't eat memory too. 28 * Alan Cox : Generic queue code and memory fixes. 29 * Fred Van Kempen : IP fragment support (borrowed from NET2E) 30 * Gerhard Koerting: Forward fragmented frames correctly. 31 * Gerhard Koerting: Fixes to my fix of the above 8-). 32 * Gerhard Koerting: IP interface addressing fix. 33 * Linus Torvalds : More robustness checks 34 * Alan Cox : Even more checks: Still not as robust as it ought to be 35 * Alan Cox : Save IP header pointer for later 36 * Alan Cox : ip option setting 37 * Alan Cox : Use ip_tos/ip_ttl settings 38 * Alan Cox : Fragmentation bogosity removed 39 * (Thanks to Mark.Bush@prg.ox.ac.uk) 40 * Dmitry Gorodchanin : Send of a raw packet crash fix. 41 * Alan Cox : Silly ip bug when an overlength 42 * fragment turns up. Now frees the 43 * queue. 44 * Linus Torvalds/ : Memory leakage on fragmentation 45 * Alan Cox : handling. 46 * Gerhard Koerting: Forwarding uses IP priority hints 47 * Teemu Rantanen : Fragment problems. 48 * Alan Cox : General cleanup, comments and reformat 49 * Alan Cox : SNMP statistics 50 * Alan Cox : BSD address rule semantics. Also see 51 * UDP as there is a nasty checksum issue 52 * if you do things the wrong way. 53 * Alan Cox : Always defrag, moved IP_FORWARD to the config.in file 54 * Alan Cox : IP options adjust sk->priority. 55 * 56 * To Fix: 57 * IP option processing is mostly not needed. ip_forward needs to know about routing rules 58 * and time stamp but that's about all. Use the route mtu field here too 59 * 60 * This program is free software; you can redistribute it and/or 61 * modify it under the terms of the GNU General Public License 62 * as published by the Free Software Foundation; either version 63 * 2 of the License, or (at your option) any later version. 64 */ 65 #include <asm/segment.h>
66 #include <asm/system.h>
67 #include <linux/types.h>
68 #include <linux/kernel.h>
69 #include <linux/sched.h>
70 #include <linux/string.h>
71 #include <linux/errno.h>
72 #include <linux/socket.h>
73 #include <linux/sockios.h>
74 #include <linux/in.h>
75 #include <linux/inet.h>
76 #include <linux/netdevice.h>
77 #include <linux/etherdevice.h>
78 #include "snmp.h"
79 #include "ip.h"
80 #include "protocol.h"
81 #include "route.h"
82 #include "tcp.h"
83 #include <linux/skbuff.h>
84 #include "sock.h"
85 #include "arp.h"
86 #include "icmp.h"
87
88 #define CONFIG_IP_DEFRAG
89
90 externintlast_retran;
91 externvoidsort_send(structsock *sk);
92
93 #definemin(a,b) ((a)<(b)?(a):(b))
94 #defineLOOPBACK(x) (((x) & htonl(0xff000000)) == htonl(0x7f000000))
95
96 /* 97 * SNMP management statistics 98 */ 99
100 structip_mibip_statistics={1,64,}; /* Forwarding=Yes, Default TTL=64 */ 101
102 /* 103 * Handle the issuing of an ioctl() request 104 * for the ip device. This is scheduled to 105 * disappear 106 */ 107
108 intip_ioctl(structsock *sk, intcmd, unsignedlongarg)
/* */ 109 { 110 switch(cmd)
111 { 112 default:
113 return(-EINVAL);
114 } 115 } 116
117
118 /* these two routines will do routing. */ 119
120 staticvoid 121 strict_route(structiphdr *iph, structoptions *opt)
/* */ 122 { 123 } 124
125
126 staticvoid 127 loose_route(structiphdr *iph, structoptions *opt)
/* */ 128 { 129 } 130
131
132
133
134 /* This routine will check to see if we have lost a gateway. */ 135 void 136 ip_route_check(unsignedlongdaddr)
/* */ 137 { 138 } 139
140
141 #if 0
142 /* this routine puts the options at the end of an ip header. */ 143 staticint 144 build_options(structiphdr *iph, structoptions *opt)
/* */ 145 { 146 unsignedchar *ptr;
147 /* currently we don't support any options. */ 148 ptr = (unsignedchar *)(iph+1);
149 *ptr = 0;
150 return (4);
151 } 152 #endif 153
154
155 /* 156 * Take an skb, and fill in the MAC header. 157 */ 158
159 staticintip_send(structsk_buff *skb, unsignedlongdaddr, intlen, structdevice *dev, unsignedlongsaddr)
/* */ 160 { 161 intmac = 0;
162
163 skb->dev = dev;
164 skb->arp = 1;
165 if (dev->hard_header)
166 { 167 /* 168 * Build a hardware header. Source address is our mac, destination unknown 169 * (rebuild header will sort this out) 170 */ 171 mac = dev->hard_header(skb->data, dev, ETH_P_IP, NULL, NULL, len, skb);
172 if (mac < 0)
173 { 174 mac = -mac;
175 skb->arp = 0;
176 skb->raddr = daddr; /* next routing address */ 177 } 178 } 179 returnmac;
180 } 181
182 intip_id_count = 0;
183
184 /* 185 * This routine builds the appropriate hardware/IP headers for 186 * the routine. It assumes that if *dev != NULL then the 187 * protocol knows what it's doing, otherwise it uses the 188 * routing/ARP tables to select a device struct. 189 */ 190 intip_build_header(structsk_buff *skb, unsignedlongsaddr, unsignedlongdaddr,
/* */ 191 structdevice **dev, inttype, structoptions *opt, intlen, inttos, intttl)
192 { 193 staticstructoptionsoptmem;
194 structiphdr *iph;
195 structrtable *rt;
196 unsignedchar *buff;
197 unsignedlongraddr;
198 inttmp;
199 unsignedlongsrc;
200
201 /* 202 * If there is no 'from' address as yet, then make it our loopback 203 */ 204
205 if (saddr == 0)
206 saddr = ip_my_addr();
207
208 buff = skb->data;
209
210 /* 211 * See if we need to look up the device. 212 */ 213
214 if (*dev == NULL)
215 { 216 if(skb->localroute)
217 rt = ip_rt_local(daddr, &optmem, &src);
218 else 219 rt = ip_rt_route(daddr, &optmem, &src);
220 if (rt == NULL)
221 { 222 ip_statistics.IpOutNoRoutes++;
223 return(-ENETUNREACH);
224 } 225
226 *dev = rt->rt_dev;
227 /* 228 * If the frame is from us and going off machine it MUST MUST MUST 229 * have the output device ip address and never the loopback 230 */ 231 if (LOOPBACK(saddr) && !LOOPBACK(daddr))
232 saddr = src;/*rt->rt_dev->pa_addr;*/ 233 raddr = rt->rt_gateway;
234
235 opt = &optmem;
236 } 237 else 238 { 239 /* 240 * We still need the address of the first hop. 241 */ 242 if(skb->localroute)
243 rt = ip_rt_local(daddr, &optmem, &src);
244 else 245 rt = ip_rt_route(daddr, &optmem, &src);
246 /* 247 * If the frame is from us and going off machine it MUST MUST MUST 248 * have the output device ip address and never the loopback 249 */ 250 if (LOOPBACK(saddr) && !LOOPBACK(daddr))
251 saddr = src;/*rt->rt_dev->pa_addr;*/ 252
253 raddr = (rt == NULL) ? 0 : rt->rt_gateway;
254 } 255
256 /* 257 * No gateway so aim at the real destination 258 */ 259 if (raddr == 0)
260 raddr = daddr;
261
262 /* 263 * Now build the MAC header. 264 */ 265
266 tmp = ip_send(skb, raddr, len, *dev, saddr);
267 buff += tmp;
268 len -= tmp;
269
270 /* 271 * Book keeping 272 */ 273
274 skb->dev = *dev;
275 skb->saddr = saddr;
276 if (skb->sk)
277 skb->sk->saddr = saddr;
278
279 /* 280 * Now build the IP header. 281 */ 282
283 /* 284 * If we are using IPPROTO_RAW, then we don't need an IP header, since 285 * one is being supplied to us by the user 286 */ 287
288 if(type == IPPROTO_RAW)
289 return (tmp);
290
291 iph = (structiphdr *)buff;
292 iph->version = 4;
293 iph->tos = tos;
294 iph->frag_off = 0;
295 iph->ttl = ttl;
296 iph->daddr = daddr;
297 iph->saddr = saddr;
298 iph->protocol = type;
299 iph->ihl = 5;
300
301 /* Setup the IP options. */ 302 #ifdef Not_Yet_Avail
303 build_options(iph, opt);
304 #endif 305
306 return(20 + tmp); /* IP header plus MAC header size */ 307 } 308
309
310 staticint 311 do_options(structiphdr *iph, structoptions *opt)
/* */ 312 { 313 unsignedchar *buff;
314 intdone = 0;
315 inti, len = sizeof(structiphdr);
316
317 /* Zero out the options. */ 318 opt->record_route.route_size = 0;
319 opt->loose_route.route_size = 0;
320 opt->strict_route.route_size = 0;
321 opt->tstamp.ptr = 0;
322 opt->security = 0;
323 opt->compartment = 0;
324 opt->handling = 0;
325 opt->stream = 0;
326 opt->tcc = 0;
327 return(0);
328
329 /* Advance the pointer to start at the options. */ 330 buff = (unsignedchar *)(iph + 1);
331
332 /* Now start the processing. */ 333 while (!done && len < iph->ihl*4) switch(*buff) { 334 caseIPOPT_END:
335 done = 1;
336 break;
337 caseIPOPT_NOOP:
338 buff++;
339 len++;
340 break;
341 caseIPOPT_SEC:
342 buff++;
343 if (*buff != 11) return(1);
344 buff++;
345 opt->security = ntohs(*(unsignedshort *)buff);
346 buff += 2;
347 opt->compartment = ntohs(*(unsignedshort *)buff);
348 buff += 2;
349 opt->handling = ntohs(*(unsignedshort *)buff);
350 buff += 2;
351 opt->tcc = ((*buff) << 16) + ntohs(*(unsignedshort *)(buff+1));
352 buff += 3;
353 len += 11;
354 break;
355 caseIPOPT_LSRR:
356 buff++;
357 if ((*buff - 3)% 4 != 0) return(1);
358 len += *buff;
359 opt->loose_route.route_size = (*buff -3)/4;
360 buff++;
361 if (*buff % 4 != 0) return(1);
362 opt->loose_route.pointer = *buff/4 - 1;
363 buff++;
364 buff++;
365 for (i = 0; i < opt->loose_route.route_size; i++) { 366 if(i>=MAX_ROUTE)
367 return(1);
368 opt->loose_route.route[i] = *(unsignedlong *)buff;
369 buff += 4;
370 } 371 break;
372 caseIPOPT_SSRR:
373 buff++;
374 if ((*buff - 3)% 4 != 0) return(1);
375 len += *buff;
376 opt->strict_route.route_size = (*buff -3)/4;
377 buff++;
378 if (*buff % 4 != 0) return(1);
379 opt->strict_route.pointer = *buff/4 - 1;
380 buff++;
381 buff++;
382 for (i = 0; i < opt->strict_route.route_size; i++) { 383 if(i>=MAX_ROUTE)
384 return(1);
385 opt->strict_route.route[i] = *(unsignedlong *)buff;
386 buff += 4;
387 } 388 break;
389 caseIPOPT_RR:
390 buff++;
391 if ((*buff - 3)% 4 != 0) return(1);
392 len += *buff;
393 opt->record_route.route_size = (*buff -3)/4;
394 buff++;
395 if (*buff % 4 != 0) return(1);
396 opt->record_route.pointer = *buff/4 - 1;
397 buff++;
398 buff++;
399 for (i = 0; i < opt->record_route.route_size; i++) { 400 if(i>=MAX_ROUTE)
401 return 1;
402 opt->record_route.route[i] = *(unsignedlong *)buff;
403 buff += 4;
404 } 405 break;
406 caseIPOPT_SID:
407 len += 4;
408 buff +=2;
409 opt->stream = *(unsignedshort *)buff;
410 buff += 2;
411 break;
412 caseIPOPT_TIMESTAMP:
413 buff++;
414 len += *buff;
415 if (*buff % 4 != 0) return(1);
416 opt->tstamp.len = *buff / 4 - 1;
417 buff++;
418 if ((*buff - 1) % 4 != 0) return(1);
419 opt->tstamp.ptr = (*buff-1)/4;
420 buff++;
421 opt->tstamp.x.full_char = *buff;
422 buff++;
423 for (i = 0; i < opt->tstamp.len; i++) { 424 opt->tstamp.data[i] = *(unsignedlong *)buff;
425 buff += 4;
426 } 427 break;
428 default:
429 return(1);
430 } 431
432 if (opt->record_route.route_size == 0) { 433 if (opt->strict_route.route_size != 0) { 434 memcpy(&(opt->record_route), &(opt->strict_route),
435 sizeof(opt->record_route));
436 }elseif (opt->loose_route.route_size != 0) { 437 memcpy(&(opt->record_route), &(opt->loose_route),
438 sizeof(opt->record_route));
439 } 440 } 441
442 if (opt->strict_route.route_size != 0 &&
443 opt->strict_route.route_size != opt->strict_route.pointer) { 444 strict_route(iph, opt);
445 return(0);
446 } 447
448 if (opt->loose_route.route_size != 0 &&
449 opt->loose_route.route_size != opt->loose_route.pointer) { 450 loose_route(iph, opt);
451 return(0);
452 } 453
454 return(0);
455 } 456
457 /* 458 * This is a version of ip_compute_csum() optimized for IP headers, which 459 * always checksum on 4 octet boundaries. 460 */ 461
462 staticinlineunsignedshortip_fast_csum(unsignedchar * buff, intwlen)
/* */ 463 { 464 unsignedlongsum = 0;
465
466 if (wlen)
467 { 468 unsignedlongbogus;
469 __asm__("clc\n"
470 "1:\t"
471 "lodsl\n\t"
472 "adcl %3, %0\n\t"
473 "decl %2\n\t"
474 "jne 1b\n\t"
475 "adcl $0, %0\n\t"
476 "movl %0, %3\n\t"
477 "shrl $16, %3\n\t"
478 "addw %w3, %w0\n\t"
479 "adcw $0, %w0"
480 : "=r" (sum), "=S" (buff), "=r" (wlen), "=a" (bogus)
481 : "0" (sum), "1" (buff), "2" (wlen));
482 } 483 return (~sum) & 0xffff;
484 } 485
486 /* 487 * This routine does all the checksum computations that don't 488 * require anything special (like copying or special headers). 489 */ 490
491 unsignedshortip_compute_csum(unsignedchar * buff, intlen)
/* */ 492 { 493 unsignedlongsum = 0;
494
495 /* Do the first multiple of 4 bytes and convert to 16 bits. */ 496 if (len > 3)
497 { 498 __asm__("clc\n"
499 "1:\t"
500 "lodsl\n\t"
501 "adcl %%eax, %%ebx\n\t"
502 "loop 1b\n\t"
503 "adcl $0, %%ebx\n\t"
504 "movl %%ebx, %%eax\n\t"
505 "shrl $16, %%eax\n\t"
506 "addw %%ax, %%bx\n\t"
507 "adcw $0, %%bx"
508 : "=b" (sum) , "=S" (buff)
509 : "0" (sum), "c" (len >> 2) ,"1" (buff)
510 : "ax", "cx", "si", "bx" );
511 } 512 if (len & 2)
513 { 514 __asm__("lodsw\n\t"
515 "addw %%ax, %%bx\n\t"
516 "adcw $0, %%bx"
517 : "=b" (sum), "=S" (buff)
518 : "0" (sum), "1" (buff)
519 : "bx", "ax", "si");
520 } 521 if (len & 1)
522 { 523 __asm__("lodsb\n\t"
524 "movb $0, %%ah\n\t"
525 "addw %%ax, %%bx\n\t"
526 "adcw $0, %%bx"
527 : "=b" (sum), "=S" (buff)
528 : "0" (sum), "1" (buff)
529 : "bx", "ax", "si");
530 } 531 sum =~sum;
532 return(sum & 0xffff);
533 } 534
535 /* 536 * Check the header of an incoming IP datagram. This version is still used in slhc.c. 537 */ 538
539 intip_csum(structiphdr *iph)
/* */ 540 { 541 returnip_fast_csum((unsignedchar *)iph, iph->ihl);
542 } 543
544 /* 545 * Generate a checksym for an outgoing IP datagram. 546 */ 547
548 staticvoidip_send_check(structiphdr *iph)
/* */ 549 { 550 iph->check = 0;
551 iph->check = ip_fast_csum((unsignedchar *)iph, iph->ihl);
552 } 553
554 /************************ Fragment Handlers From NET2E not yet with tweaks to beat 4K **********************************/ 555
556
557 /* 558 * This fragment handler is a bit of a heap. On the other hand it works quite 559 * happily and handles things quite well. 560 */ 561
562 staticstructipq *ipqueue = NULL; /* IP fragment queue */ 563
564 /* 565 * Create a new fragment entry. 566 */ 567
568 staticstructipfrag *ip_frag_create(intoffset, intend, structsk_buff *skb, unsignedchar *ptr)
/* */ 569 { 570 structipfrag *fp;
571
572 fp = (structipfrag *) kmalloc(sizeof(structipfrag), GFP_ATOMIC);
573 if (fp == NULL)
574 { 575 printk("IP: frag_create: no memory left !\n");
576 return(NULL);
577 } 578 memset(fp, 0, sizeof(structipfrag));
579
580 /* Fill in the structure. */ 581 fp->offset = offset;
582 fp->end = end;
583 fp->len = end - offset;
584 fp->skb = skb;
585 fp->ptr = ptr;
586
587 return(fp);
588 } 589
590
591 /* 592 * Find the correct entry in the "incomplete datagrams" queue for 593 * this IP datagram, and return the queue entry address if found. 594 */ 595
596 staticstructipq *ip_find(structiphdr *iph)
/* */ 597 { 598 structipq *qp;
599 structipq *qplast;
600
601 cli();
602 qplast = NULL;
603 for(qp = ipqueue; qp != NULL; qplast = qp, qp = qp->next)
604 { 605 if (iph->id== qp->iph->id && iph->saddr == qp->iph->saddr &&
606 iph->daddr == qp->iph->daddr && iph->protocol == qp->iph->protocol)
607 { 608 del_timer(&qp->timer); /* So it doesn't vanish on us. The timer will be reset anyway */ 609 sti();
610 return(qp);
611 } 612 } 613 sti();
614 return(NULL);
615 } 616
617
618 /* 619 * Remove an entry from the "incomplete datagrams" queue, either 620 * because we completed, reassembled and processed it, or because 621 * it timed out. 622 */ 623
624 staticvoidip_free(structipq *qp)
/* */ 625 { 626 structipfrag *fp;
627 structipfrag *xp;
628
629 /* 630 * Stop the timer for this entry. 631 */ 632
633 del_timer(&qp->timer);
634
635 /* Remove this entry from the "incomplete datagrams" queue. */ 636 cli();
637 if (qp->prev == NULL)
638 { 639 ipqueue = qp->next;
640 if (ipqueue != NULL)
641 ipqueue->prev = NULL;
642 } 643 else 644 { 645 qp->prev->next = qp->next;
646 if (qp->next != NULL)
647 qp->next->prev = qp->prev;
648 } 649
650 /* Release all fragment data. */ 651
652 fp = qp->fragments;
653 while (fp != NULL)
654 { 655 xp = fp->next;
656 IS_SKB(fp->skb);
657 kfree_skb(fp->skb,FREE_READ);
658 kfree_s(fp, sizeof(structipfrag));
659 fp = xp;
660 } 661
662 /* Release the MAC header. */ 663 kfree_s(qp->mac, qp->maclen);
664
665 /* Release the IP header. */ 666 kfree_s(qp->iph, qp->ihlen + 8);
667
668 /* Finally, release the queue descriptor itself. */ 669 kfree_s(qp, sizeof(structipq));
670 sti();
671 } 672
673
674 /* 675 * Oops- a fragment queue timed out. Kill it and send an ICMP reply. 676 */ 677
678 staticvoidip_expire(unsignedlongarg)
/* */ 679 { 680 structipq *qp;
681
682 qp = (structipq *)arg;
683
684 /* 685 * Send an ICMP "Fragment Reassembly Timeout" message. 686 */ 687
688 ip_statistics.IpReasmTimeout++;
689 ip_statistics.IpReasmFails++;
690 /* This if is always true... shrug */ 691 if(qp->fragments!=NULL)
692 icmp_send(qp->fragments->skb,ICMP_TIME_EXCEEDED,
693 ICMP_EXC_FRAGTIME, qp->dev);
694
695 /* 696 * Nuke the fragment queue. 697 */ 698 ip_free(qp);
699 } 700
701
702 /* 703 * Add an entry to the 'ipq' queue for a newly received IP datagram. 704 * We will (hopefully :-) receive all other fragments of this datagram 705 * in time, so we just create a queue for this datagram, in which we 706 * will insert the received fragments at their respective positions. 707 */ 708
709 staticstructipq *ip_create(structsk_buff *skb, structiphdr *iph, structdevice *dev)
/* */ 710 { 711 structipq *qp;
712 intmaclen;
713 intihlen;
714
715 qp = (structipq *) kmalloc(sizeof(structipq), GFP_ATOMIC);
716 if (qp == NULL)
717 { 718 printk("IP: create: no memory left !\n");
719 return(NULL);
720 skb->dev = qp->dev;
721 } 722 memset(qp, 0, sizeof(structipq));
723
724 /* 725 * Allocate memory for the MAC header. 726 * 727 * FIXME: We have a maximum MAC address size limit and define 728 * elsewhere. We should use it here and avoid the 3 kmalloc() calls 729 */ 730
731 maclen = ((unsignedlong) iph) - ((unsignedlong) skb->data);
732 qp->mac = (unsignedchar *) kmalloc(maclen, GFP_ATOMIC);
733 if (qp->mac == NULL)
734 { 735 printk("IP: create: no memory left !\n");
736 kfree_s(qp, sizeof(structipq));
737 return(NULL);
738 } 739
740 /* 741 * Allocate memory for the IP header (plus 8 octects for ICMP). 742 */ 743
744 ihlen = (iph->ihl * sizeof(unsignedlong));
745 qp->iph = (structiphdr *) kmalloc(ihlen + 8, GFP_ATOMIC);
746 if (qp->iph == NULL)
747 { 748 printk("IP: create: no memory left !\n");
749 kfree_s(qp->mac, maclen);
750 kfree_s(qp, sizeof(structipq));
751 return(NULL);
752 } 753
754 /* Fill in the structure. */ 755 memcpy(qp->mac, skb->data, maclen);
756 memcpy(qp->iph, iph, ihlen + 8);
757 qp->len = 0;
758 qp->ihlen = ihlen;
759 qp->maclen = maclen;
760 qp->fragments = NULL;
761 qp->dev = dev;
762
763 /* Start a timer for this entry. */ 764 qp->timer.expires = IP_FRAG_TIME; /* about 30 seconds */ 765 qp->timer.data = (unsignedlong) qp; /* pointer to queue */ 766 qp->timer.function = ip_expire; /* expire function */ 767 add_timer(&qp->timer);
768
769 /* Add this entry to the queue. */ 770 qp->prev = NULL;
771 cli();
772 qp->next = ipqueue;
773 if (qp->next != NULL)
774 qp->next->prev = qp;
775 ipqueue = qp;
776 sti();
777 return(qp);
778 } 779
780
781 /* 782 * See if a fragment queue is complete. 783 */ 784
785 staticintip_done(structipq *qp)
/* */ 786 { 787 structipfrag *fp;
788 intoffset;
789
790 /* Only possible if we received the final fragment. */ 791 if (qp->len == 0)
792 return(0);
793
794 /* Check all fragment offsets to see if they connect. */ 795 fp = qp->fragments;
796 offset = 0;
797 while (fp != NULL)
798 { 799 if (fp->offset > offset)
800 return(0); /* fragment(s) missing */ 801 offset = fp->end;
802 fp = fp->next;
803 } 804
805 /* All fragments are present. */ 806 return(1);
807 } 808
809
810 /* 811 * Build a new IP datagram from all its fragments. 812 * 813 * FIXME: We copy here because we lack an effective way of handling lists 814 * of bits on input. Until the new skb data handling is in I'm not going 815 * to touch this with a bargepole. This also causes a 4Kish limit on 816 * packet sizes. 817 */ 818
819 staticstructsk_buff *ip_glue(structipq *qp)
/* */ 820 { 821 structsk_buff *skb;
822 structiphdr *iph;
823 structipfrag *fp;
824 unsignedchar *ptr;
825 intcount, len;
826
827 /* 828 * Allocate a new buffer for the datagram. 829 */ 830
831 len = qp->maclen + qp->ihlen + qp->len;
832
833 if ((skb = alloc_skb(len,GFP_ATOMIC)) == NULL)
834 { 835 ip_statistics.IpReasmFails++;
836 printk("IP: queue_glue: no memory for glueing queue 0x%X\n", (int) qp);
837 ip_free(qp);
838 return(NULL);
839 } 840
841 /* Fill in the basic details. */ 842 skb->len = (len - qp->maclen);
843 skb->h.raw = skb->data;
844 skb->free = 1;
845
846 /* Copy the original MAC and IP headers into the new buffer. */ 847 ptr = (unsignedchar *) skb->h.raw;
848 memcpy(ptr, ((unsignedchar *) qp->mac), qp->maclen);
849 ptr += qp->maclen;
850 memcpy(ptr, ((unsignedchar *) qp->iph), qp->ihlen);
851 ptr += qp->ihlen;
852 skb->h.raw += qp->maclen;
853
854 count = 0;
855
856 /* Copy the data portions of all fragments into the new buffer. */ 857 fp = qp->fragments;
858 while(fp != NULL)
859 { 860 if(count+fp->len>skb->len)
861 { 862 printk("Invalid fragment list: Fragment over size.\n");
863 ip_free(qp);
864 kfree_skb(skb,FREE_WRITE);
865 ip_statistics.IpReasmFails++;
866 returnNULL;
867 } 868 memcpy((ptr + fp->offset), fp->ptr, fp->len);
869 count += fp->len;
870 fp = fp->next;
871 } 872
873 /* We glued together all fragments, so remove the queue entry. */ 874 ip_free(qp);
875
876 /* Done with all fragments. Fixup the new IP header. */ 877 iph = skb->h.iph;
878 iph->frag_off = 0;
879 iph->tot_len = htons((iph->ihl * sizeof(unsignedlong)) + count);
880 skb->ip_hdr = iph;
881
882 ip_statistics.IpReasmOKs++;
883 return(skb);
884 } 885
886
887 /* 888 * Process an incoming IP datagram fragment. 889 */ 890
891 staticstructsk_buff *ip_defrag(structiphdr *iph, structsk_buff *skb, structdevice *dev)
/* */ 892 { 893 structipfrag *prev, *next;
894 structipfrag *tfp;
895 structipq *qp;
896 structsk_buff *skb2;
897 unsignedchar *ptr;
898 intflags, offset;
899 inti, ihl, end;
900
901 ip_statistics.IpReasmReqds++;
902
903 /* Find the entry of this IP datagram in the "incomplete datagrams" queue. */ 904 qp = ip_find(iph);
905
906 /* Is this a non-fragmented datagram? */ 907 offset = ntohs(iph->frag_off);
908 flags = offset & ~IP_OFFSET;
909 offset &= IP_OFFSET;
910 if (((flags & IP_MF) == 0) && (offset == 0))
911 { 912 if (qp != NULL)
913 ip_free(qp); /* Huh? How could this exist?? */ 914 return(skb);
915 } 916
917 offset <<= 3; /* offset is in 8-byte chunks */ 918
919 /* 920 * If the queue already existed, keep restarting its timer as long 921 * as we still are receiving fragments. Otherwise, create a fresh 922 * queue entry. 923 */ 924
925 if (qp != NULL)
926 { 927 del_timer(&qp->timer);
928 qp->timer.expires = IP_FRAG_TIME; /* about 30 seconds */ 929 qp->timer.data = (unsignedlong) qp; /* pointer to queue */ 930 qp->timer.function = ip_expire; /* expire function */ 931 add_timer(&qp->timer);
932 } 933 else 934 { 935 /* 936 * If we failed to create it, then discard the frame 937 */ 938 if ((qp = ip_create(skb, iph, dev)) == NULL)
939 { 940 skb->sk = NULL;
941 kfree_skb(skb, FREE_READ);
942 ip_statistics.IpReasmFails++;
943 returnNULL;
944 } 945 } 946
947 /* 948 * Determine the position of this fragment. 949 */ 950
951 ihl = (iph->ihl * sizeof(unsignedlong));
952 end = offset + ntohs(iph->tot_len) - ihl;
953
954 /* 955 * Point into the IP datagram 'data' part. 956 */ 957
958 ptr = skb->data + dev->hard_header_len + ihl;
959
960 /* 961 * Is this the final fragment? 962 */ 963
964 if ((flags & IP_MF) == 0)
965 qp->len = end;
966
967 /* 968 * Find out which fragments are in front and at the back of us 969 * in the chain of fragments so far. We must know where to put 970 * this fragment, right? 971 */ 972
973 prev = NULL;
974 for(next = qp->fragments; next != NULL; next = next->next)
975 { 976 if (next->offset > offset)
977 break; /* bingo! */ 978 prev = next;
979 } 980
981 /* 982 * We found where to put this one. 983 * Check for overlap with preceeding fragment, and, if needed, 984 * align things so that any overlaps are eliminated. 985 */ 986 if (prev != NULL && offset < prev->end)
987 { 988 i = prev->end - offset;
989 offset += i; /* ptr into datagram */ 990 ptr += i; /* ptr into fragment data */ 991 } 992
993 /* 994 * Look for overlap with succeeding segments. 995 * If we can merge fragments, do it. 996 */ 997
998 for(; next != NULL; next = tfp)
999 {1000 tfp = next->next;
1001 if (next->offset >= end)
1002 break; /* no overlaps at all */1003
1004 i = end - next->offset; /* overlap is 'i' bytes */1005 next->len -= i; /* so reduce size of */1006 next->offset += i; /* next fragment */1007 next->ptr += i;
1008
1009 /* 1010 * If we get a frag size of <= 0, remove it and the packet1011 * that it goes with.1012 */1013 if (next->len <= 0)
1014 {1015 if (next->prev != NULL)
1016 next->prev->next = next->next;
1017 else1018 qp->fragments = next->next;
1019
1020 if (tfp->next != NULL)
1021 next->next->prev = next->prev;
1022
1023 kfree_skb(next->skb,FREE_READ);
1024 kfree_s(next, sizeof(structipfrag));
1025 }1026 }1027
1028 /* 1029 * Insert this fragment in the chain of fragments. 1030 */1031
1032 tfp = NULL;
1033 tfp = ip_frag_create(offset, end, skb, ptr);
1034
1035 /*1036 * No memory to save the fragment - so throw the lot1037 */1038
1039 if (!tfp)
1040 {1041 skb->sk = NULL;
1042 kfree_skb(skb, FREE_READ);
1043 returnNULL;
1044 }1045 tfp->prev = prev;
1046 tfp->next = next;
1047 if (prev != NULL)
1048 prev->next = tfp;
1049 else1050 qp->fragments = tfp;
1051
1052 if (next != NULL)
1053 next->prev = tfp;
1054
1055 /*1056 * OK, so we inserted this new fragment into the chain.1057 * Check if we now have a full IP datagram which we can1058 * bump up to the IP layer...1059 */1060
1061 if (ip_done(qp))
1062 {1063 skb2 = ip_glue(qp); /* glue together the fragments */1064 return(skb2);
1065 }1066 return(NULL);
1067 }1068
1069
1070 /*1071 * This IP datagram is too large to be sent in one piece. Break it up into1072 * smaller pieces (each of size equal to the MAC header plus IP header plus1073 * a block of the data of the original IP data part) that will yet fit in a1074 * single device frame, and queue such a frame for sending by calling the1075 * ip_queue_xmit(). Note that this is recursion, and bad things will happen1076 * if this function causes a loop...1077 *1078 * Yes this is inefficient, feel free to submit a quicker one.1079 *1080 * **Protocol Violation**1081 * We copy all the options to each fragment. !FIXME!1082 */1083
1084 voidip_fragment(structsock *sk, structsk_buff *skb, structdevice *dev, intis_frag)
/* */1085 {1086 structiphdr *iph;
1087 unsignedchar *raw;
1088 unsignedchar *ptr;
1089 structsk_buff *skb2;
1090 intleft, mtu, hlen, len;
1091 intoffset;
1092 unsignedlongflags;
1093
1094 /* 1095 * Point into the IP datagram header. 1096 */1097
1098 raw = skb->data;
1099 iph = (structiphdr *) (raw + dev->hard_header_len);
1100
1101 skb->ip_hdr = iph;
1102
1103 /* 1104 * Setup starting values. 1105 */1106
1107 hlen = (iph->ihl * sizeof(unsignedlong));
1108 left = ntohs(iph->tot_len) - hlen; /* Space per frame */1109 hlen += dev->hard_header_len; /* Total header size */1110 mtu = (dev->mtu - hlen); /* Size of data space */1111 ptr = (raw + hlen); /* Where to start from */1112
1113 /*1114 * Check for any "DF" flag. [DF means do not fragment]1115 */1116
1117 if (ntohs(iph->frag_off) & IP_DF)
1118 {1119 ip_statistics.IpFragFails++;
1120 icmp_send(skb,ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, dev);
1121 return;
1122 }1123
1124 /*1125 * The protocol doesn't seem to say what to do in the case that the1126 * frame + options doesn't fit the mtu. As it used to fall down dead1127 * in this case we were fortunate it didn't happen1128 */1129
1130 if(mtu<8)
1131 {1132 /* It's wrong but its better than nothing */1133 icmp_send(skb,ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED,dev);
1134 ip_statistics.IpFragFails++;
1135 return;
1136 }1137
1138 /* 1139 * Fragment the datagram. 1140 */1141
1142 /*1143 * The initial offset is 0 for a complete frame. When1144 * fragmenting fragments its wherever this one starts.1145 */1146
1147 if (is_frag & 2)
1148 offset = (ntohs(iph->frag_off) & 0x1fff) << 3;
1149 else1150 offset = 0;
1151
1152
1153 /*1154 * Keep copying data until we run out.1155 */1156
1157 while(left > 0)
1158 {1159 len = left;
1160 /* IF: it doesn't fit, use 'mtu' - the data space left */1161 if (len > mtu)
1162 len = mtu;
1163 /* IF: we are not sending upto and including the packet end1164 then align the next start on an eight byte boundary */1165 if (len < left)
1166 {1167 len/=8;
1168 len*=8;
1169 }1170 /*1171 * Allocate buffer. 1172 */1173
1174 if ((skb2 = alloc_skb(len + hlen,GFP_ATOMIC)) == NULL)
1175 {1176 printk("IP: frag: no memory for new fragment!\n");
1177 ip_statistics.IpFragFails++;
1178 return;
1179 }1180
1181 /*1182 * Set up data on packet1183 */1184
1185 skb2->arp = skb->arp;
1186 if(skb->free==0)
1187 printk("IP fragmenter: BUG free!=1 in fragmenter\n");
1188 skb2->free = 1;
1189 skb2->len = len + hlen;
1190 skb2->h.raw=(char *) skb2->data;
1191 /*1192 * Charge the memory for the fragment to any owner1193 * it might posess1194 */1195
1196 save_flags(flags);
1197 if (sk)
1198 {1199 cli();
1200 sk->wmem_alloc += skb2->mem_len;
1201 skb2->sk=sk;
1202 }1203 restore_flags(flags);
1204 skb2->raddr = skb->raddr; /* For rebuild_header - must be here */1205
1206 /* 1207 * Copy the packet header into the new buffer. 1208 */1209
1210 memcpy(skb2->h.raw, raw, hlen);
1211
1212 /*1213 * Copy a block of the IP datagram. 1214 */1215 memcpy(skb2->h.raw + hlen, ptr, len);
1216 left -= len;
1217
1218 skb2->h.raw+=dev->hard_header_len;
1219
1220 /*1221 * Fill in the new header fields. 1222 */1223 iph = (structiphdr *)(skb2->h.raw/*+dev->hard_header_len*/);
1224 iph->frag_off = htons((offset >> 3));
1225 /* 1226 * Added AC : If we are fragmenting a fragment thats not the1227 * last fragment then keep MF on each bit 1228 */1229 if (left > 0 || (is_frag & 1))
1230 iph->frag_off |= htons(IP_MF);
1231 ptr += len;
1232 offset += len;
1233
1234 /* 1235 * Put this fragment into the sending queue. 1236 */1237
1238 ip_statistics.IpFragCreates++;
1239
1240 ip_queue_xmit(sk, dev, skb2, 2);
1241 }1242 ip_statistics.IpFragOKs++;
1243 }1244
1245
1246
1247 #ifdefCONFIG_IP_FORWARD1248
1249 /* 1250 * Forward an IP datagram to its next destination. 1251 */1252
1253 staticvoidip_forward(structsk_buff *skb, structdevice *dev, intis_frag)
/* */1254 {1255 structdevice *dev2; /* Output device */1256 structiphdr *iph; /* Our header */1257 structsk_buff *skb2; /* Output packet */1258 structrtable *rt; /* Route we use */1259 unsignedchar *ptr; /* Data pointer */1260 unsignedlongraddr; /* Router IP address */1261
1262 /*1263 * According to the RFC, we must first decrease the TTL field. If1264 * that reaches zero, we must reply an ICMP control message telling1265 * that the packet's lifetime expired.1266 *1267 * Exception:1268 * We may not generate an ICMP for an ICMP. icmp_send does the1269 * enforcement of this so we can forget it here. It is however1270 * sometimes VERY important.1271 */1272
1273 iph = skb->h.iph;
1274 iph->ttl--;
1275 if (iph->ttl <= 0)
1276 {1277 /* Tell the sender its packet died... */1278 icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, dev);
1279 return;
1280 }1281
1282 /* 1283 * Re-compute the IP header checksum. 1284 * This is inefficient. We know what has happened to the header1285 * and could thus adjust the checksum as Phil Karn does in KA9Q1286 */1287
1288 ip_send_check(iph);
1289
1290 /*1291 * OK, the packet is still valid. Fetch its destination address,1292 * and give it to the IP sender for further processing.1293 */1294
1295 rt = ip_rt_route(iph->daddr, NULL, NULL);
1296 if (rt == NULL)
1297 {1298 /*1299 * Tell the sender its packet cannot be delivered. Again1300 * ICMP is screened later.1301 */1302 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_NET_UNREACH, dev);
1303 return;
1304 }1305
1306
1307 /*1308 * Gosh. Not only is the packet valid; we even know how to1309 * forward it onto its final destination. Can we say this1310 * is being plain lucky?1311 * If the router told us that there is no GW, use the dest.1312 * IP address itself- we seem to be connected directly...1313 */1314
1315 raddr = rt->rt_gateway;
1316
1317 if (raddr != 0)
1318 {1319 /*1320 * There is a gateway so find the correct route for it.1321 * Gateways cannot in turn be gatewayed.1322 */1323 rt = ip_rt_route(raddr, NULL, NULL);
1324 if (rt == NULL)
1325 {1326 /* 1327 * Tell the sender its packet cannot be delivered... 1328 */1329 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, dev);
1330 return;
1331 }1332 if (rt->rt_gateway != 0)
1333 raddr = rt->rt_gateway;
1334 }1335 else1336 raddr = iph->daddr;
1337
1338 /*1339 * Having picked a route we can now send the frame out.1340 */1341
1342 dev2 = rt->rt_dev;
1343
1344 /*1345 * In IP you never forward a frame on the interface that it arrived1346 * upon. We should generate an ICMP HOST REDIRECT giving the route1347 * we calculated.1348 * For now just dropping the packet is an acceptable compromise.1349 */1350
1351 if (dev == dev2)
1352 return;
1353
1354 /*1355 * We now allocate a new buffer, and copy the datagram into it.1356 * If the indicated interface is up and running, kick it.1357 */1358
1359 if (dev2->flags & IFF_UP)
1360 {1361
1362 /*1363 * Current design decrees we copy the packet. For identical header1364 * lengths we could avoid it. The new skb code will let us push1365 * data so the problem goes away then.1366 */1367
1368 skb2 = alloc_skb(dev2->hard_header_len + skb->len, GFP_ATOMIC);
1369 /*1370 * This is rare and since IP is tolerant of network failures1371 * quite harmless.1372 */1373 if (skb2 == NULL)
1374 {1375 printk("\nIP: No memory available for IP forward\n");
1376 return;
1377 }1378 ptr = skb2->data;
1379 skb2->free = 1;
1380 skb2->len = skb->len + dev2->hard_header_len;
1381 skb2->h.raw = ptr;
1382
1383 /* 1384 * Copy the packet data into the new buffer. 1385 */1386 memcpy(ptr + dev2->hard_header_len, skb->h.raw, skb->len);
1387
1388 /* Now build the MAC header. */1389 (void) ip_send(skb2, raddr, skb->len, dev2, dev2->pa_addr);
1390
1391 ip_statistics.IpForwDatagrams++;
1392
1393 /*1394 * See if it needs fragmenting. Note in ip_rcv we tagged1395 * the fragment type. This must be right so that1396 * the fragmenter does the right thing.1397 */1398
1399 if(skb2->len > dev2->mtu)
1400 {1401 ip_fragment(NULL,skb2,dev2, is_frag);
1402 kfree_skb(skb2,FREE_WRITE);
1403 }1404 else1405 {1406 /*1407 * Map service types to priority. We lie about1408 * throughput being low priority, but its a good1409 * choice to help improve general usage.1410 */1411 if(iph->tos & IPTOS_LOWDELAY)
1412 dev_queue_xmit(skb2, dev2, SOPRI_INTERACTIVE);
1413 elseif(iph->tos & IPTOS_THROUGHPUT)
1414 dev_queue_xmit(skb2, dev2, SOPRI_BACKGROUND);
1415 else1416 dev_queue_xmit(skb2, dev2, SOPRI_NORMAL);
1417 }1418 }1419 }1420
1421
1422 #endif1423
1424 /*1425 * This function receives all incoming IP datagrams. 1426 */1427
1428 intip_rcv(structsk_buff *skb, structdevice *dev, structpacket_type *pt)
/* */1429 {1430 structiphdr *iph = skb->h.iph;
1431 unsignedcharhash;
1432 unsignedcharflag = 0;
1433 unsignedcharopts_p = 0; /* Set iff the packet has options. */1434 structinet_protocol *ipprot;
1435 staticstructoptionsopt; /* since we don't use these yet, and they1436 take up stack space. */1437 intbrd;
1438 intis_frag=0;
1439
1440
1441 ip_statistics.IpInReceives++;
1442
1443 /*1444 * Tag the ip header of this packet so we can find it1445 */1446
1447 skb->ip_hdr = iph;
1448
1449 /*1450 * Is the datagram acceptable? 1451 *1452 * 1. Length at least the size of an ip header1453 * 2. Version of 41454 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]1455 * (4. We ought to check for IP multicast addresses and undefined types.. does this matter ?)1456 */1457
1458 if (skb->len<sizeof(structiphdr) || iph->ihl<5 || iph->version != 4 || ip_fast_csum((unsignedchar *)iph, iph->ihl) !=0)
1459 {1460 ip_statistics.IpInHdrErrors++;
1461 kfree_skb(skb, FREE_WRITE);
1462 return(0);
1463 }1464
1465 /*1466 * Our transport medium may have padded the buffer out. Now we know it1467 * is IP we can trim to the true length of the frame.1468 */1469
1470 skb->len=ntohs(iph->tot_len);
1471
1472 /*1473 * Next anaylse the packet for options. Studies show under one packet in1474 * a thousand have options....1475 */1476
1477 if (iph->ihl != 5)
1478 {/* Fast path for the typical optionless IP packet. */1479 memset((char *) &opt, 0, sizeof(opt));
1480 if (do_options(iph, &opt) != 0)
1481 return 0;
1482 opts_p = 1;
1483 }1484
1485 /*1486 * Remember if the frame is fragmented.1487 */1488
1489 if (iph->frag_off & 0x0020)
1490 is_frag|=1;
1491
1492 /*1493 * Last fragment ?1494 */1495
1496 if (ntohs(iph->frag_off) & 0x1fff)
1497 is_frag|=2;
1498
1499 /* 1500 * Do any IP forwarding required. chk_addr() is expensive -- avoid it someday. 1501 *1502 * This is inefficient. While finding out if it is for us we could also compute1503 * the routing table entry. This is where the great unified cache theory comes1504 * in as and when someone impliments it1505 */1506
1507 if ((brd = ip_chk_addr(iph->daddr)) == 0)
1508 {1509 /*1510 * Don't forward multicast or broadcast frames.1511 */1512
1513 if(skb->pkt_type!=PACKET_HOST)
1514 {1515 kfree_skb(skb,FREE_WRITE);
1516 return 0;
1517 }1518
1519 /*1520 * The packet is for another target. Forward the frame1521 */1522
1523 #ifdefCONFIG_IP_FORWARD1524 ip_forward(skb, dev, is_frag);
1525 #else1526 /* printk("Machine %lx tried to use us as a forwarder to %lx but we have forwarding disabled!\n",1527 iph->saddr,iph->daddr);*/1528 ip_statistics.IpInAddrErrors++;
1529 #endif1530 /*1531 * The forwarder is inefficient and copies the packet. We 1532 * free the original now.1533 */1534
1535 kfree_skb(skb, FREE_WRITE);
1536 return(0);
1537 }1538
1539 /*1540 * Reassemble IP fragments. 1541 */1542
1543 if(is_frag)
1544 {1545 /* Defragment. Obtain the complete packet if there is one */1546 skb=ip_defrag(iph,skb,dev);
1547 if(skb==NULL)
1548 return 0;
1549 iph=skb->h.iph;
1550 }1551
1552 /*1553 * Point into the IP datagram, just past the header. 1554 */1555
1556 skb->ip_hdr = iph;
1557 skb->h.raw += iph->ihl*4;
1558
1559 /*1560 * skb->h.raw now points at the protocol beyond the IP header.1561 */1562
1563 hash = iph->protocol & (MAX_INET_PROTOS -1);
1564 for (ipprot = (structinet_protocol *)inet_protos[hash];ipprot != NULL;ipprot=(structinet_protocol *)ipprot->next)
1565 {1566 structsk_buff *skb2;
1567
1568 if (ipprot->protocol != iph->protocol)
1569 continue;
1570 /*1571 * See if we need to make a copy of it. This will1572 * only be set if more than one protocol wants it. 1573 * and then not for the last one.1574 *1575 * This is an artifact of poor upper protocol design. 1576 * Because the upper protocols damage the actual packet1577 * we must do copying. In actual fact it's even worse1578 * than this as TCP may hold on to the buffer.1579 */1580 if (ipprot->copy)
1581 {1582 skb2 = skb_clone(skb, GFP_ATOMIC);
1583 if(skb2==NULL)
1584 continue;
1585 }1586 else1587 {1588 skb2 = skb;
1589 }1590 flag = 1;
1591
1592 /*1593 * Pass on the datagram to each protocol that wants it,1594 * based on the datagram protocol. We should really1595 * check the protocol handler's return values here...1596 */1597 ipprot->handler(skb2, dev, opts_p ? &opt : 0, iph->daddr,
1598 (ntohs(iph->tot_len) - (iph->ihl * 4)),
1599 iph->saddr, 0, ipprot);
1600
1601 }1602
1603 /*1604 * All protocols checked.1605 * If this packet was a broadcast, we may *not* reply to it, since that1606 * causes (proven, grin) ARP storms and a leakage of memory (i.e. all1607 * ICMP reply messages get queued up for transmission...)1608 */1609
1610 if (!flag)
1611 {1612 if (brd != IS_BROADCAST)
1613 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, dev);
1614 kfree_skb(skb, FREE_WRITE);
1615 }1616
1617 return(0);
1618 }1619
1620
1621 /*1622 * Queues a packet to be sent, and starts the transmitter1623 * if necessary. if free = 1 then we free the block after1624 * transmit, otherwise we don't. If free==2 we not only1625 * free the block but also dont assign a new ip seq number.1626 * This routine also needs to put in the total length,1627 * and compute the checksum1628 */1629
1630 voidip_queue_xmit(structsock *sk, structdevice *dev,
/* */1631 structsk_buff *skb, intfree)
1632 {1633 structiphdr *iph;
1634 unsignedchar *ptr;
1635
1636 /* Sanity check */1637 if (dev == NULL)
1638 {1639 printk("IP: ip_queue_xmit dev = NULL\n");
1640 return;
1641 }1642
1643 IS_SKB(skb);
1644
1645 /*1646 * Do some book-keeping in the packet for later1647 */1648
1649
1650 skb->dev = dev;
1651 skb->when = jiffies;
1652
1653 /*1654 * Find the IP header and set the length. This is bad1655 * but once we get the skb data handling code in the1656 * hardware will push its header sensibly and we will1657 * set skb->ip_hdr to avoid this mess and the fixed1658 * header length problem1659 */1660
1661 ptr = skb->data;
1662 ptr += dev->hard_header_len;
1663 iph = (structiphdr *)ptr;
1664 skb->ip_hdr = iph;
1665 iph->tot_len = ntohs(skb->len-dev->hard_header_len);
1666
1667 /*1668 * No reassigning numbers to fragments...1669 */1670
1671 if(free!=2)
1672 iph->id = htons(ip_id_count++);
1673 else1674 free=1;
1675
1676 /* All buffers without an owner socket get freed */1677 if (sk == NULL)
1678 free = 1;
1679
1680 skb->free = free;
1681
1682 /*1683 * Do we need to fragment. Again this is inefficient. 1684 * We need to somehow lock the original buffer and use1685 * bits of it.1686 */1687
1688 if(skb->len > dev->mtu)
1689 {1690 ip_fragment(sk,skb,dev,0);
1691 IS_SKB(skb);
1692 kfree_skb(skb,FREE_WRITE);
1693 return;
1694 }1695
1696 /*1697 * Add an IP checksum1698 */1699
1700 ip_send_check(iph);
1701
1702 /*1703 * Print the frame when debugging1704 */1705
1706 /*1707 * More debugging. You cannot queue a packet already on a list1708 * Spot this and moan loudly.1709 */1710 if (skb->next != NULL)
1711 {1712 printk("ip_queue_xmit: next != NULL\n");
1713 skb_unlink(skb);
1714 }1715
1716 /*1717 * If a sender wishes the packet to remain unfreed1718 * we add it to his send queue. This arguably belongs1719 * in the TCP level since nobody elses uses it. BUT1720 * remember IPng might change all the rules.1721 */1722
1723 if (!free)
1724 {1725 unsignedlongflags;
1726 /* The socket now has more outstanding blocks */1727
1728 sk->packets_out++;
1729
1730 /* Protect the list for a moment */1731 save_flags(flags);
1732 cli();
1733
1734 if (skb->link3 != NULL)
1735 {1736 printk("ip.c: link3 != NULL\n");
1737 skb->link3 = NULL;
1738 }1739 if (sk->send_head == NULL)
1740 {1741 sk->send_tail = skb;
1742 sk->send_head = skb;
1743 }1744 else1745 {1746 sk->send_tail->link3 = skb;
1747 sk->send_tail = skb;
1748 }1749 /* skb->link3 is NULL */1750
1751 /* Interrupt restore */1752 restore_flags(flags);
1753 /* Set the IP write timeout to the round trip time for the packet.1754 If an acknowledge has not arrived by then we may wish to act */1755 reset_timer(sk, TIME_WRITE, sk->rto);
1756 }1757 else1758 /* Remember who owns the buffer */1759 skb->sk = sk;
1760
1761 /*1762 * If the indicated interface is up and running, send the packet. 1763 */1764 ip_statistics.IpOutRequests++;
1765
1766 if (dev->flags & IFF_UP)
1767 {1768 /* 1769 * If we have an owner use its priority setting,1770 * otherwise use NORMAL1771 */1772
1773 if (sk != NULL)
1774 {1775 dev_queue_xmit(skb, dev, sk->priority);
1776 }1777 else1778 {1779 dev_queue_xmit(skb, dev, SOPRI_NORMAL);
1780 }1781 }1782 else1783 {1784 ip_statistics.IpOutDiscards++;
1785 if (free)
1786 kfree_skb(skb, FREE_WRITE);
1787 }1788 }1789
1790
1791 /*1792 * A socket has timed out on its send queue and wants to do a1793 * little retransmitting. Currently this means TCP.1794 */1795
1796 voidip_do_retransmit(structsock *sk, intall)
/* */1797 {1798 structsk_buff * skb;
1799 structproto *prot;
1800 structdevice *dev;
1801 intretransmits;
1802
1803 prot = sk->prot;
1804 skb = sk->send_head;
1805 retransmits = sk->retransmits;
1806
1807 while (skb != NULL)
1808 {1809 dev = skb->dev;
1810 IS_SKB(skb);
1811 skb->when = jiffies;
1812
1813 /* 1814 * In general it's OK just to use the old packet. However we1815 * need to use the current ack and window fields. Urg and 1816 * urg_ptr could possibly stand to be updated as well, but we 1817 * don't keep the necessary data. That shouldn't be a problem,1818 * if the other end is doing the right thing. Since we're 1819 * changing the packet, we have to issue a new IP identifier.1820 */1821
1822 /* this check may be unnecessary - retransmit only for TCP */1823 if (sk->protocol == IPPROTO_TCP) {1824 structtcphdr *th;
1825 structiphdr *iph;
1826 intsize;
1827
1828 iph = (structiphdr *)(skb->data + dev->hard_header_len);
1829 th = (structtcphdr *)(((char *)iph) + (iph->ihl << 2));
1830 size = skb->len - (((unsignedchar *) th) - skb->data);
1831
1832 iph->id = htons(ip_id_count++);
1833 ip_send_check(iph);
1834
1835 th->ack_seq = ntohl(sk->acked_seq);
1836 th->window = ntohs(tcp_select_window(sk));
1837 tcp_send_check(th, sk->saddr, sk->daddr, size, sk);
1838 }1839
1840 /* 1841 * If the interface is (still) up and running, kick it. 1842 */1843
1844 if (dev->flags & IFF_UP)
1845 {1846 /*1847 * If the packet is still being sent by the device/protocol1848 * below then don't retransmit. This is both needed, and good -1849 * especially with connected mode AX.25 where it stops resends1850 * occuring of an as yet unsent anyway frame!1851 * We still add up the counts as the round trip time wants1852 * adjusting.1853 */1854 if (sk && !skb_device_locked(skb))
1855 {1856 /* Remove it from any existing driver queue first! */1857 skb_unlink(skb);
1858 /* Now queue it */1859 ip_statistics.IpOutRequests++;
1860 dev_queue_xmit(skb, dev, sk->priority);
1861 }1862 }1863
1864 /*1865 * Count retransmissions1866 */1867 retransmits++;
1868 sk->prot->retransmits ++;
1869
1870 /*1871 * Only one retransmit requested.1872 */1873 if (!all)
1874 break;
1875
1876 /*1877 * This should cut it off before we send too many packets. 1878 */1879 if (sk->retransmits > sk->cong_window)
1880 break;
1881 skb = skb->link3;
1882 }1883 }1884
1885 /*1886 * This is the normal code called for timeouts. It does the retransmission1887 * and then does backoff. ip_do_retransmit is separated out because1888 * tcp_ack needs to send stuff from the retransmit queue without1889 * initiating a backoff.1890 */1891
1892 voidip_retransmit(structsock *sk, intall)
/* */1893 {1894 ip_do_retransmit(sk, all);
1895
1896 /*1897 * Increase the timeout each time we retransmit. Note that1898 * we do not increase the rtt estimate. rto is initialized1899 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests1900 * that doubling rto each time is the least we can get away with.1901 * In KA9Q, Karn uses this for the first few times, and then1902 * goes to quadratic. netBSD doubles, but only goes up to *64,1903 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is1904 * defined in the protocol as the maximum possible RTT. I guess1905 * we'll have to use something other than TCP to talk to the1906 * University of Mars.1907 */1908
1909 sk->retransmits++;
1910 sk->backoff++;
1911 sk->rto = min(sk->rto << 1, 120*HZ);
1912 reset_timer(sk, TIME_WRITE, sk->rto);
1913 }1914
1915 /*1916 * Socket option code for IP. This is the end of the line after any TCP,UDP etc options on1917 * an IP socket.1918 *1919 * We impliment IP_TOS (type of service), IP_TTL (time to live).1920 *1921 * Next release we will sort out IP_OPTIONS since for some people are kind of important.1922 */1923
1924 intip_setsockopt(structsock *sk, intlevel, intoptname, char *optval, intoptlen)
/* */1925 {1926 intval,err;
1927
1928 if (optval == NULL)
1929 return(-EINVAL);
1930
1931 err=verify_area(VERIFY_READ, optval, sizeof(int));
1932 if(err)
1933 returnerr;
1934
1935 val = get_fs_long((unsignedlong *)optval);
1936
1937 if(level!=SOL_IP)
1938 return -EOPNOTSUPP;
1939
1940 switch(optname)
1941 {1942 caseIP_TOS:
1943 if(val<0||val>255)
1944 return -EINVAL;
1945 sk->ip_tos=val;
1946 if(val==IPTOS_LOWDELAY)
1947 sk->priority=SOPRI_INTERACTIVE;
1948 if(val==IPTOS_THROUGHPUT)
1949 sk->priority=SOPRI_BACKGROUND;
1950 return 0;
1951 caseIP_TTL:
1952 if(val<1||val>255)
1953 return -EINVAL;
1954 sk->ip_ttl=val;
1955 return 0;
1956 /* IP_OPTIONS and friends go here eventually */1957 default:
1958 return(-ENOPROTOOPT);
1959 }1960 }1961
1962 /*1963 * Get the options. Note for future reference. The GET of IP options gets the1964 * _received_ ones. The set sets the _sent_ ones.1965 */1966
1967 intip_getsockopt(structsock *sk, intlevel, intoptname, char *optval, int *optlen)
/* */1968 {1969 intval,err;
1970
1971 if(level!=SOL_IP)
1972 return -EOPNOTSUPP;
1973
1974 switch(optname)
1975 {1976 caseIP_TOS:
1977 val=sk->ip_tos;
1978 break;
1979 caseIP_TTL:
1980 val=sk->ip_ttl;
1981 break;
1982 default:
1983 return(-ENOPROTOOPT);
1984 }1985 err=verify_area(VERIFY_WRITE, optlen, sizeof(int));
1986 if(err)
1987 returnerr;
1988 put_fs_long(sizeof(int),(unsignedlong *) optlen);
1989
1990 err=verify_area(VERIFY_WRITE, optval, sizeof(int));
1991 if(err)
1992 returnerr;
1993 put_fs_long(val,(unsignedlong *)optval);
1994
1995 return(0);
1996 }1997
1998 /*1999 * IP protocol layer initialiser2000 */2001
2002 staticstructpacket_typeip_packet_type =
2003 {2004 0, /* MUTTER ntohs(ETH_P_IP),*/2005 0, /* copy */2006 ip_rcv,
2007 NULL,
2008 NULL,
2009 };
2010
2011
2012 /*2013 * IP registers the packet type and then calls the subprotocol initialisers2014 */2015
2016 voidip_init(void)
/* */2017 {2018 ip_packet_type.type=htons(ETH_P_IP);
2019 dev_add_pack(&ip_packet_type);
2020 /* ip_raw_init();2021 ip_packet_init();2022 ip_tcp_init();2023 ip_udp_init();*/2024 }