![[first]](../icons/first.png){kind=icon}
![[previous]](../icons/n_left.png){kind=icon}
![[next]](../icons/n_right.png){kind=icon}
![[last]](../icons/last.png){kind=icon}
![[top]](../icons/n_top.png){kind=icon}
![[bottom]](../icons/bottom.png){kind=icon}
![[index]](../icons/index.png){kind=icon}
![[help]](../icons/help.png){kind=icon}
*/
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 *
54 * To Fix:
55 * IP option processing is mostly not needed. ip_forward needs to know about routing rules
56 * and time stamp but that's about all.
57 *
58 * This program is free software; you can redistribute it and/or
59 * modify it under the terms of the GNU General Public License
60 * as published by the Free Software Foundation; either version
61 * 2 of the License, or (at your option) any later version.
62 */
63 #include <asm/segment.h>
64 #include <asm/system.h>
65 #include <linux/types.h>
66 #include <linux/kernel.h>
67 #include <linux/sched.h>
68 #include <linux/string.h>
69 #include <linux/errno.h>
70 #include <linux/socket.h>
71 #include <linux/sockios.h>
72 #include <linux/in.h>
73 #include <linux/inet.h>
74 #include <linux/netdevice.h>
75 #include <linux/etherdevice.h>
76 #include "snmp.h"
77 #include "ip.h"
78 #include "protocol.h"
79 #include "route.h"
80 #include "tcp.h"
81 #include <linux/skbuff.h>
82 #include "sock.h"
83 #include "arp.h"
84 #include "icmp.h"
85
86 #define CONFIG_IP_FORWARD
87 #define CONFIG_IP_DEFRAG
88
89 extern int last_retran;
90 extern void sort_send(struct sock *sk);
91
92 #define min(a,b) ((a)<(b)?(a):(b))
93
94 /*
95 * SNMP management statistics
96 */
97
98 struct ip_mib ip_statistics={1,64,}; /* Forwarding=Yes, Default TTL=64 */
99
100 /*
101 * Print an IP packet for debugging purposes.
102 *
103 * This function is exported for the IP
104 * upper layers to use also.
105 */
106
107 void ip_print(const struct iphdr *ip)
/* ![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[top]](../icons/top.png)
*/
108 {
109 unsigned char buff[32];
110 unsigned char *ptr;
111 int addr;
112 int len;
113 int i;
114
115 /* Are we debugging IP frames */
116
117 if (inet_debug != DBG_IP)
118 return;
119
120 /* Dump the IP header. */
121 printk("IP: ihl=%d, version=%d, tos=%d, tot_len=%d\n",
122 ip->ihl, ip->version, ip->tos, ntohs(ip->tot_len));
123 printk(" id=%X, ttl=%d, prot=%d, check=%X\n",
124 ip->id, ip->ttl, ip->protocol, ip->check);
125 printk(" frag_off=%d\n", ip->frag_off);
126 printk(" soucre=%s ", in_ntoa(ip->saddr));
127 printk("dest=%s\n", in_ntoa(ip->daddr));
128 printk(" ----\n");
129
130 /* Dump the data. */
131 ptr = (unsigned char *)(ip + 1);
132 addr = 0;
133 len = ntohs(ip->tot_len) - (4 * ip->ihl);
134
135 while (len > 0)
136 {
137 printk(" %04X: ", addr);
138 for(i = 0; i < 16; i++)
139 {
140 if (len > 0)
141 {
142 printk("%02X ", (*ptr & 0xFF));
143 buff[i] = *ptr++;
144 if (buff[i] < 32 || buff[i] > 126)
145 buff[i] = '.';
146 }
147 else
148 {
149 printk(" ");
150 buff[i] = ' ';
151 }
152 addr++;
153 len--;
154 };
155 buff[i] = '\0';
156 printk(" \"%s\"\n", buff);
157 }
158 printk(" ----\n\n");
159 }
160
161 /*
162 * Handle the issuing of an ioctl() request
163 * for the ip device. This is scheduled to
164 * disappear
165 */
166
167 int ip_ioctl(struct sock *sk, int cmd, unsigned long arg)
/* ![[previous]](../icons/left.png)
*/
168 {
169 switch(cmd)
170 {
171 case DDIOCSDBG:
172 return(dbg_ioctl((void *) arg, DBG_IP));
173 default:
174 return(-EINVAL);
175 }
176 }
177
178
179 /* these two routines will do routing. */
180
181 static void
182 strict_route(struct iphdr *iph, struct options *opt)
/* */
183 {
184 }
185
186
187 static void
188 loose_route(struct iphdr *iph, struct options *opt)
/* */
189 {
190 }
191
192
193 static void
194 print_ipprot(struct inet_protocol *ipprot)
/* */
195 {
196 DPRINTF((DBG_IP, "handler = %X, protocol = %d, copy=%d \n",
197 ipprot->handler, ipprot->protocol, ipprot->copy));
198 }
199
200
201 /* This routine will check to see if we have lost a gateway. */
202 void
203 ip_route_check(unsigned long daddr)
/* */
204 {
205 }
206
207
208 #if 0
209 /* this routine puts the options at the end of an ip header. */
210 static int
211 build_options(struct iphdr *iph, struct options *opt)
/* */
212 {
213 unsigned char *ptr;
214 /* currently we don't support any options. */
215 ptr = (unsigned char *)(iph+1);
216 *ptr = 0;
217 return (4);
218 }
219 #endif
220
221
222 /*
223 * Take an skb, and fill in the MAC header.
224 */
225
226 static int ip_send(struct sk_buff *skb, unsigned long daddr, int len, struct device *dev, unsigned long saddr)
/* */
227 {
228 int mac = 0;
229
230 skb->dev = dev;
231 skb->arp = 1;
232 if (dev->hard_header)
233 {
234 /*
235 * Build a hardware header. Source address is our mac, destination unknown
236 * (rebuild header will sort this out)
237 */
238 mac = dev->hard_header(skb->data, dev, ETH_P_IP, NULL, NULL, len, skb);
239 if (mac < 0)
240 {
241 mac = -mac;
242 skb->arp = 0;
243 skb->raddr = daddr; /* next routing address */
244 }
245 }
246 return mac;
247 }
248
249
250 /*
251 * This routine builds the appropriate hardware/IP headers for
252 * the routine. It assumes that if *dev != NULL then the
253 * protocol knows what it's doing, otherwise it uses the
254 * routing/ARP tables to select a device struct.
255 */
256 int ip_build_header(struct sk_buff *skb, unsigned long saddr, unsigned long daddr,
/* */
257 struct device **dev, int type, struct options *opt, int len, int tos, int ttl)
258 {
259 static struct options optmem;
260 struct iphdr *iph;
261 struct rtable *rt;
262 unsigned char *buff;
263 unsigned long raddr;
264 static int count = 0;
265 int tmp;
266 unsigned long src;
267
268 /*
269 * If there is no 'from' address as yet, then make it our loopback
270 */
271
272 if (saddr == 0)
273 saddr = ip_my_addr();
274
275 DPRINTF((DBG_IP, "ip_build_header (skb=%X, saddr=%X, daddr=%X, *dev=%X,\n"
276 " type=%d, opt=%X, len = %d)\n",
277 skb, saddr, daddr, *dev, type, opt, len));
278
279 buff = skb->data;
280
281 /*
282 * See if we need to look up the device.
283 */
284
285 if (*dev == NULL)
286 {
287 if(skb->localroute)
288 rt = ip_rt_local(daddr, &optmem, &src);
289 else
290 rt = ip_rt_route(daddr, &optmem, &src);
291 if (rt == NULL)
292 {
293 ip_statistics.IpOutNoRoutes++;
294 return(-ENETUNREACH);
295 }
296
297 *dev = rt->rt_dev;
298 /*
299 * If the frame is from us and going off machine it MUST MUST MUST
300 * have the output device ip address and never the loopback
301 */
302 if (saddr == 0x0100007FL && daddr != 0x0100007FL)
303 saddr = src;/*rt->rt_dev->pa_addr;*/
304 raddr = rt->rt_gateway;
305
306 DPRINTF((DBG_IP, "ip_build_header: saddr set to %s\n", in_ntoa(saddr)));
307 opt = &optmem;
308 }
309 else
310 {
311 /*
312 * We still need the address of the first hop.
313 */
314 if(skb->localroute)
315 rt = ip_rt_local(daddr, &optmem, &src);
316 else
317 rt = ip_rt_route(daddr, &optmem, &src);
318 /*
319 * If the frame is from us and going off machine it MUST MUST MUST
320 * have the output device ip address and never the loopback
321 */
322 if (saddr == 0x0100007FL && daddr != 0x0100007FL)
323 saddr = src;/*rt->rt_dev->pa_addr;*/
324
325 raddr = (rt == NULL) ? 0 : rt->rt_gateway;
326 }
327
328 /*
329 * No gateway so aim at the real destination
330 */
331 if (raddr == 0)
332 raddr = daddr;
333
334 /*
335 * Now build the MAC header.
336 */
337
338 tmp = ip_send(skb, raddr, len, *dev, saddr);
339 buff += tmp;
340 len -= tmp;
341
342 /*
343 * Book keeping
344 */
345
346 skb->dev = *dev;
347 skb->saddr = saddr;
348 if (skb->sk)
349 skb->sk->saddr = saddr;
350
351 /*
352 * Now build the IP header.
353 */
354
355 /*
356 * If we are using IPPROTO_RAW, then we don't need an IP header, since
357 * one is being supplied to us by the user
358 */
359
360 if(type == IPPROTO_RAW)
361 return (tmp);
362
363 iph = (struct iphdr *)buff;
364 iph->version = 4;
365 iph->tos = tos;
366 iph->frag_off = 0;
367 iph->ttl = ttl;
368 iph->daddr = daddr;
369 iph->saddr = saddr;
370 iph->protocol = type;
371 iph->ihl = 5;
372 iph->id = htons(count++);
373
374 /* Setup the IP options. */
375 #ifdef Not_Yet_Avail
376 build_options(iph, opt);
377 #endif
378
379 return(20 + tmp); /* IP header plus MAC header size */
380 }
381
382
383 static int
384 do_options(struct iphdr *iph, struct options *opt)
/* */
385 {
386 unsigned char *buff;
387 int done = 0;
388 int i, len = sizeof(struct iphdr);
389
390 /* Zero out the options. */
391 opt->record_route.route_size = 0;
392 opt->loose_route.route_size = 0;
393 opt->strict_route.route_size = 0;
394 opt->tstamp.ptr = 0;
395 opt->security = 0;
396 opt->compartment = 0;
397 opt->handling = 0;
398 opt->stream = 0;
399 opt->tcc = 0;
400 return(0);
401
402 /* Advance the pointer to start at the options. */
403 buff = (unsigned char *)(iph + 1);
404
405 /* Now start the processing. */
406 while (!done && len < iph->ihl*4) switch(*buff) {
407 case IPOPT_END:
408 done = 1;
409 break;
410 case IPOPT_NOOP:
411 buff++;
412 len++;
413 break;
414 case IPOPT_SEC:
415 buff++;
416 if (*buff != 11) return(1);
417 buff++;
418 opt->security = ntohs(*(unsigned short *)buff);
419 buff += 2;
420 opt->compartment = ntohs(*(unsigned short *)buff);
421 buff += 2;
422 opt->handling = ntohs(*(unsigned short *)buff);
423 buff += 2;
424 opt->tcc = ((*buff) << 16) + ntohs(*(unsigned short *)(buff+1));
425 buff += 3;
426 len += 11;
427 break;
428 case IPOPT_LSRR:
429 buff++;
430 if ((*buff - 3)% 4 != 0) return(1);
431 len += *buff;
432 opt->loose_route.route_size = (*buff -3)/4;
433 buff++;
434 if (*buff % 4 != 0) return(1);
435 opt->loose_route.pointer = *buff/4 - 1;
436 buff++;
437 buff++;
438 for (i = 0; i < opt->loose_route.route_size; i++) {
439 if(i>=MAX_ROUTE)
440 return(1);
441 opt->loose_route.route[i] = *(unsigned long *)buff;
442 buff += 4;
443 }
444 break;
445 case IPOPT_SSRR:
446 buff++;
447 if ((*buff - 3)% 4 != 0) return(1);
448 len += *buff;
449 opt->strict_route.route_size = (*buff -3)/4;
450 buff++;
451 if (*buff % 4 != 0) return(1);
452 opt->strict_route.pointer = *buff/4 - 1;
453 buff++;
454 buff++;
455 for (i = 0; i < opt->strict_route.route_size; i++) {
456 if(i>=MAX_ROUTE)
457 return(1);
458 opt->strict_route.route[i] = *(unsigned long *)buff;
459 buff += 4;
460 }
461 break;
462 case IPOPT_RR:
463 buff++;
464 if ((*buff - 3)% 4 != 0) return(1);
465 len += *buff;
466 opt->record_route.route_size = (*buff -3)/4;
467 buff++;
468 if (*buff % 4 != 0) return(1);
469 opt->record_route.pointer = *buff/4 - 1;
470 buff++;
471 buff++;
472 for (i = 0; i < opt->record_route.route_size; i++) {
473 if(i>=MAX_ROUTE)
474 return 1;
475 opt->record_route.route[i] = *(unsigned long *)buff;
476 buff += 4;
477 }
478 break;
479 case IPOPT_SID:
480 len += 4;
481 buff +=2;
482 opt->stream = *(unsigned short *)buff;
483 buff += 2;
484 break;
485 case IPOPT_TIMESTAMP:
486 buff++;
487 len += *buff;
488 if (*buff % 4 != 0) return(1);
489 opt->tstamp.len = *buff / 4 - 1;
490 buff++;
491 if ((*buff - 1) % 4 != 0) return(1);
492 opt->tstamp.ptr = (*buff-1)/4;
493 buff++;
494 opt->tstamp.x.full_char = *buff;
495 buff++;
496 for (i = 0; i < opt->tstamp.len; i++) {
497 opt->tstamp.data[i] = *(unsigned long *)buff;
498 buff += 4;
499 }
500 break;
501 default:
502 return(1);
503 }
504
505 if (opt->record_route.route_size == 0) {
506 if (opt->strict_route.route_size != 0) {
507 memcpy(&(opt->record_route), &(opt->strict_route),
508 sizeof(opt->record_route));
509 } else if (opt->loose_route.route_size != 0) {
510 memcpy(&(opt->record_route), &(opt->loose_route),
511 sizeof(opt->record_route));
512 }
513 }
514
515 if (opt->strict_route.route_size != 0 &&
516 opt->strict_route.route_size != opt->strict_route.pointer) {
517 strict_route(iph, opt);
518 return(0);
519 }
520
521 if (opt->loose_route.route_size != 0 &&
522 opt->loose_route.route_size != opt->loose_route.pointer) {
523 loose_route(iph, opt);
524 return(0);
525 }
526
527 return(0);
528 }
529
530 /*
531 * This is a version of ip_compute_csum() optimized for IP headers, which
532 * always checksum on 4 octet boundaries.
533 */
534
535 static inline unsigned short ip_fast_csum(unsigned char * buff, int wlen)
/* */
536 {
537 unsigned long sum = 0;
538
539 if (wlen)
540 {
541 unsigned long bogus;
542 __asm__("clc\n"
543 "1:\t"
544 "lodsl\n\t"
545 "adcl %3, %0\n\t"
546 "decl %2\n\t"
547 "jne 1b\n\t"
548 "adcl $0, %0\n\t"
549 "movl %0, %3\n\t"
550 "shrl $16, %3\n\t"
551 "addw %w3, %w0\n\t"
552 "adcw $0, %w0"
553 : "=r" (sum), "=S" (buff), "=r" (wlen), "=a" (bogus)
554 : "0" (sum), "1" (buff), "2" (wlen));
555 }
556 return (~sum) & 0xffff;
557 }
558
559 /*
560 * This routine does all the checksum computations that don't
561 * require anything special (like copying or special headers).
562 */
563
564 unsigned short ip_compute_csum(unsigned char * buff, int len)
/* */
565 {
566 unsigned long sum = 0;
567
568 /* Do the first multiple of 4 bytes and convert to 16 bits. */
569 if (len > 3)
570 {
571 __asm__("clc\n"
572 "1:\t"
573 "lodsl\n\t"
574 "adcl %%eax, %%ebx\n\t"
575 "loop 1b\n\t"
576 "adcl $0, %%ebx\n\t"
577 "movl %%ebx, %%eax\n\t"
578 "shrl $16, %%eax\n\t"
579 "addw %%ax, %%bx\n\t"
580 "adcw $0, %%bx"
581 : "=b" (sum) , "=S" (buff)
582 : "0" (sum), "c" (len >> 2) ,"1" (buff)
583 : "ax", "cx", "si", "bx" );
584 }
585 if (len & 2)
586 {
587 __asm__("lodsw\n\t"
588 "addw %%ax, %%bx\n\t"
589 "adcw $0, %%bx"
590 : "=b" (sum), "=S" (buff)
591 : "0" (sum), "1" (buff)
592 : "bx", "ax", "si");
593 }
594 if (len & 1)
595 {
596 __asm__("lodsb\n\t"
597 "movb $0, %%ah\n\t"
598 "addw %%ax, %%bx\n\t"
599 "adcw $0, %%bx"
600 : "=b" (sum), "=S" (buff)
601 : "0" (sum), "1" (buff)
602 : "bx", "ax", "si");
603 }
604 sum =~sum;
605 return(sum & 0xffff);
606 }
607
608 /*
609 * Check the header of an incoming IP datagram. This version is still used in slhc.c.
610 */
611
612 int ip_csum(struct iphdr *iph)
/* */
613 {
614 return ip_fast_csum((unsigned char *)iph, iph->ihl);
615 }
616
617 /*
618 * Generate a checksym for an outgoing IP datagram.
619 */
620
621 static void ip_send_check(struct iphdr *iph)
/* */
622 {
623 iph->check = 0;
624 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
625 }
626
627 /************************ Fragment Handlers From NET2E not yet with tweaks to beat 4K **********************************/
628
629
630 /*
631 * This fragment handler is a bit of a heap. On the other hand it works quite
632 * happily and handles things quite well.
633 */
634
635 static struct ipq *ipqueue = NULL; /* IP fragment queue */
636
637 /*
638 * Create a new fragment entry.
639 */
640
641 static struct ipfrag *ip_frag_create(int offset, int end, struct sk_buff *skb, unsigned char *ptr)
/* */
642 {
643 struct ipfrag *fp;
644
645 fp = (struct ipfrag *) kmalloc(sizeof(struct ipfrag), GFP_ATOMIC);
646 if (fp == NULL)
647 {
648 printk("IP: frag_create: no memory left !\n");
649 return(NULL);
650 }
651 memset(fp, 0, sizeof(struct ipfrag));
652
653 /* Fill in the structure. */
654 fp->offset = offset;
655 fp->end = end;
656 fp->len = end - offset;
657 fp->skb = skb;
658 fp->ptr = ptr;
659
660 return(fp);
661 }
662
663
664 /*
665 * Find the correct entry in the "incomplete datagrams" queue for
666 * this IP datagram, and return the queue entry address if found.
667 */
668
669 static struct ipq *ip_find(struct iphdr *iph)
/* */
670 {
671 struct ipq *qp;
672 struct ipq *qplast;
673
674 cli();
675 qplast = NULL;
676 for(qp = ipqueue; qp != NULL; qplast = qp, qp = qp->next)
677 {
678 if (iph->id== qp->iph->id && iph->saddr == qp->iph->saddr &&
679 iph->daddr == qp->iph->daddr && iph->protocol == qp->iph->protocol)
680 {
681 del_timer(&qp->timer); /* So it doesn't vanish on us. The timer will be reset anyway */
682 sti();
683 return(qp);
684 }
685 }
686 sti();
687 return(NULL);
688 }
689
690
691 /*
692 * Remove an entry from the "incomplete datagrams" queue, either
693 * because we completed, reassembled and processed it, or because
694 * it timed out.
695 */
696
697 static void ip_free(struct ipq *qp)
/* */
698 {
699 struct ipfrag *fp;
700 struct ipfrag *xp;
701
702 /*
703 * Stop the timer for this entry.
704 */
705
706 del_timer(&qp->timer);
707
708 /* Remove this entry from the "incomplete datagrams" queue. */
709 cli();
710 if (qp->prev == NULL)
711 {
712 ipqueue = qp->next;
713 if (ipqueue != NULL)
714 ipqueue->prev = NULL;
715 }
716 else
717 {
718 qp->prev->next = qp->next;
719 if (qp->next != NULL)
720 qp->next->prev = qp->prev;
721 }
722
723 /* Release all fragment data. */
724
725 fp = qp->fragments;
726 while (fp != NULL)
727 {
728 xp = fp->next;
729 IS_SKB(fp->skb);
730 kfree_skb(fp->skb,FREE_READ);
731 kfree_s(fp, sizeof(struct ipfrag));
732 fp = xp;
733 }
734
735 /* Release the MAC header. */
736 kfree_s(qp->mac, qp->maclen);
737
738 /* Release the IP header. */
739 kfree_s(qp->iph, qp->ihlen + 8);
740
741 /* Finally, release the queue descriptor itself. */
742 kfree_s(qp, sizeof(struct ipq));
743 /* printk("ip_free:done\n");*/
744 sti();
745 }
746
747
748 /*
749 * Oops- a fragment queue timed out. Kill it and send an ICMP reply.
750 */
751
752 static void ip_expire(unsigned long arg)
/* */
753 {
754 struct ipq *qp;
755
756 qp = (struct ipq *)arg;
757 DPRINTF((DBG_IP, "IP: queue_expire: fragment queue 0x%X timed out!\n", qp));
758
759 /*
760 * Send an ICMP "Fragment Reassembly Timeout" message.
761 */
762
763 ip_statistics.IpReasmTimeout++;
764 ip_statistics.IpReasmFails++;
765 /* This if is always true... shrug */
766 if(qp->fragments!=NULL)
767 icmp_send(qp->fragments->skb,ICMP_TIME_EXCEEDED,
768 ICMP_EXC_FRAGTIME, qp->dev);
769
770 /*
771 * Nuke the fragment queue.
772 */
773 ip_free(qp);
774 }
775
776
777 /*
778 * Add an entry to the 'ipq' queue for a newly received IP datagram.
779 * We will (hopefully :-) receive all other fragments of this datagram
780 * in time, so we just create a queue for this datagram, in which we
781 * will insert the received fragments at their respective positions.
782 */
783
784 static struct ipq *ip_create(struct sk_buff *skb, struct iphdr *iph, struct device *dev)
/* */
785 {
786 struct ipq *qp;
787 int maclen;
788 int ihlen;
789
790 qp = (struct ipq *) kmalloc(sizeof(struct ipq), GFP_ATOMIC);
791 if (qp == NULL)
792 {
793 printk("IP: create: no memory left !\n");
794 return(NULL);
795 skb->dev = qp->dev;
796 }
797 memset(qp, 0, sizeof(struct ipq));
798
799 /*
800 * Allocate memory for the MAC header.
801 *
802 * FIXME: We have a maximum MAC address size limit and define
803 * elsewhere. We should use it here and avoid the 3 kmalloc() calls
804 */
805
806 maclen = ((unsigned long) iph) - ((unsigned long) skb->data);
807 qp->mac = (unsigned char *) kmalloc(maclen, GFP_ATOMIC);
808 if (qp->mac == NULL)
809 {
810 printk("IP: create: no memory left !\n");
811 kfree_s(qp, sizeof(struct ipq));
812 return(NULL);
813 }
814
815 /*
816 * Allocate memory for the IP header (plus 8 octects for ICMP).
817 */
818
819 ihlen = (iph->ihl * sizeof(unsigned long));
820 qp->iph = (struct iphdr *) kmalloc(ihlen + 8, GFP_ATOMIC);
821 if (qp->iph == NULL)
822 {
823 printk("IP: create: no memory left !\n");
824 kfree_s(qp->mac, maclen);
825 kfree_s(qp, sizeof(struct ipq));
826 return(NULL);
827 }
828
829 /* Fill in the structure. */
830 memcpy(qp->mac, skb->data, maclen);
831 memcpy(qp->iph, iph, ihlen + 8);
832 qp->len = 0;
833 qp->ihlen = ihlen;
834 qp->maclen = maclen;
835 qp->fragments = NULL;
836 qp->dev = dev;
837
838 /* Start a timer for this entry. */
839 qp->timer.expires = IP_FRAG_TIME; /* about 30 seconds */
840 qp->timer.data = (unsigned long) qp; /* pointer to queue */
841 qp->timer.function = ip_expire; /* expire function */
842 add_timer(&qp->timer);
843
844 /* Add this entry to the queue. */
845 qp->prev = NULL;
846 cli();
847 qp->next = ipqueue;
848 if (qp->next != NULL)
849 qp->next->prev = qp;
850 ipqueue = qp;
851 sti();
852 return(qp);
853 }
854
855
856 /*
857 * See if a fragment queue is complete.
858 */
859
860 static int ip_done(struct ipq *qp)
/* */
861 {
862 struct ipfrag *fp;
863 int offset;
864
865 /* Only possible if we received the final fragment. */
866 if (qp->len == 0)
867 return(0);
868
869 /* Check all fragment offsets to see if they connect. */
870 fp = qp->fragments;
871 offset = 0;
872 while (fp != NULL)
873 {
874 if (fp->offset > offset)
875 return(0); /* fragment(s) missing */
876 offset = fp->end;
877 fp = fp->next;
878 }
879
880 /* All fragments are present. */
881 return(1);
882 }
883
884
885 /*
886 * Build a new IP datagram from all its fragments.
887 *
888 * FIXME: We copy here because we lack an effective way of handling lists
889 * of bits on input. Until the new skb data handling is in I'm not going
890 * to touch this with a bargepole. This also causes a 4Kish limit on
891 * packet sizes.
892 */
893
894 static struct sk_buff *ip_glue(struct ipq *qp)
/* */
895 {
896 struct sk_buff *skb;
897 struct iphdr *iph;
898 struct ipfrag *fp;
899 unsigned char *ptr;
900 int count, len;
901
902 /*
903 * Allocate a new buffer for the datagram.
904 */
905
906 len = qp->maclen + qp->ihlen + qp->len;
907
908 if ((skb = alloc_skb(len,GFP_ATOMIC)) == NULL)
909 {
910 ip_statistics.IpReasmFails++;
911 printk("IP: queue_glue: no memory for glueing queue 0x%X\n", (int) qp);
912 ip_free(qp);
913 return(NULL);
914 }
915
916 /* Fill in the basic details. */
917 skb->len = (len - qp->maclen);
918 skb->h.raw = skb->data;
919 skb->free = 1;
920
921 /* Copy the original MAC and IP headers into the new buffer. */
922 ptr = (unsigned char *) skb->h.raw;
923 memcpy(ptr, ((unsigned char *) qp->mac), qp->maclen);
924 ptr += qp->maclen;
925 memcpy(ptr, ((unsigned char *) qp->iph), qp->ihlen);
926 ptr += qp->ihlen;
927 skb->h.raw += qp->maclen;
928
929 count = 0;
930
931 /* Copy the data portions of all fragments into the new buffer. */
932 fp = qp->fragments;
933 while(fp != NULL)
934 {
935 if(count+fp->len>skb->len)
936 {
937 printk("Invalid fragment list: Fragment over size.\n");
938 ip_free(qp);
939 kfree_skb(skb,FREE_WRITE);
940 ip_statistics.IpReasmFails++;
941 return NULL;
942 }
943 memcpy((ptr + fp->offset), fp->ptr, fp->len);
944 count += fp->len;
945 fp = fp->next;
946 }
947
948 /* We glued together all fragments, so remove the queue entry. */
949 ip_free(qp);
950
951 /* Done with all fragments. Fixup the new IP header. */
952 iph = skb->h.iph;
953 iph->frag_off = 0;
954 iph->tot_len = htons((iph->ihl * sizeof(unsigned long)) + count);
955 skb->ip_hdr = iph;
956
957 ip_statistics.IpReasmOKs++;
958 return(skb);
959 }
960
961
962 /*
963 * Process an incoming IP datagram fragment.
964 */
965
966 static struct sk_buff *ip_defrag(struct iphdr *iph, struct sk_buff *skb, struct device *dev)
/* */
967 {
968 struct ipfrag *prev, *next;
969 struct ipfrag *tfp;
970 struct ipq *qp;
971 struct sk_buff *skb2;
972 unsigned char *ptr;
973 int flags, offset;
974 int i, ihl, end;
975
976 ip_statistics.IpReasmReqds++;
977
978 /* Find the entry of this IP datagram in the "incomplete datagrams" queue. */
979 qp = ip_find(iph);
980
981 /* Is this a non-fragmented datagram? */
982 offset = ntohs(iph->frag_off);
983 flags = offset & ~IP_OFFSET;
984 offset &= IP_OFFSET;
985 if (((flags & IP_MF) == 0) && (offset == 0))
986 {
987 if (qp != NULL)
988 ip_free(qp); /* Huh? How could this exist?? */
989 return(skb);
990 }
991
992 offset <<= 3; /* offset is in 8-byte chunks */
993
994 /*
995 * If the queue already existed, keep restarting its timer as long
996 * as we still are receiving fragments. Otherwise, create a fresh
997 * queue entry.
998 */
999
1000 if (qp != NULL)
1001 {
1002 del_timer(&qp->timer);
1003 qp->timer.expires = IP_FRAG_TIME; /* about 30 seconds */
1004 qp->timer.data = (unsigned long) qp; /* pointer to queue */
1005 qp->timer.function = ip_expire; /* expire function */
1006 add_timer(&qp->timer);
1007 }
1008 else
1009 {
1010 /*
1011 * If we failed to create it, then discard the frame
1012 */
1013 if ((qp = ip_create(skb, iph, dev)) == NULL)
1014 {
1015 skb->sk = NULL;
1016 kfree_skb(skb, FREE_READ);
1017 ip_statistics.IpReasmFails++;
1018 return NULL;
1019 }
1020 }
1021
1022 /*
1023 * Determine the position of this fragment.
1024 */
1025
1026 ihl = (iph->ihl * sizeof(unsigned long));
1027 end = offset + ntohs(iph->tot_len) - ihl;
1028
1029 /*
1030 * Point into the IP datagram 'data' part.
1031 */
1032
1033 ptr = skb->data + dev->hard_header_len + ihl;
1034
1035 /*
1036 * Is this the final fragment?
1037 */
1038
1039 if ((flags & IP_MF) == 0)
1040 qp->len = end;
1041
1042 /*
1043 * Find out which fragments are in front and at the back of us
1044 * in the chain of fragments so far. We must know where to put
1045 * this fragment, right?
1046 */
1047
1048 prev = NULL;
1049 for(next = qp->fragments; next != NULL; next = next->next)
1050 {
1051 if (next->offset > offset)
1052 break; /* bingo! */
1053 prev = next;
1054 }
1055
1056 /*
1057 * We found where to put this one.
1058 * Check for overlap with preceeding fragment, and, if needed,
1059 * align things so that any overlaps are eliminated.
1060 */
1061 if (prev != NULL && offset < prev->end)
1062 {
1063 i = prev->end - offset;
1064 offset += i; /* ptr into datagram */
1065 ptr += i; /* ptr into fragment data */
1066 DPRINTF((DBG_IP, "IP: defrag: fixed low overlap %d bytes\n", i));
1067 }
1068
1069 /*
1070 * Look for overlap with succeeding segments.
1071 * If we can merge fragments, do it.
1072 */
1073
1074 for(; next != NULL; next = tfp)
1075 {
1076 tfp = next->next;
1077 if (next->offset >= end)
1078 break; /* no overlaps at all */
1079
1080 i = end - next->offset; /* overlap is 'i' bytes */
1081 next->len -= i; /* so reduce size of */
1082 next->offset += i; /* next fragment */
1083 next->ptr += i;
1084
1085 /*
1086 * If we get a frag size of <= 0, remove it and the packet
1087 * that it goes with.
1088 */
1089 if (next->len <= 0)
1090 {
1091 DPRINTF((DBG_IP, "IP: defrag: removing frag 0x%X (len %d)\n",
1092 next, next->len));
1093 if (next->prev != NULL)
1094 next->prev->next = next->next;
1095 else
1096 qp->fragments = next->next;
1097
1098 if (tfp->next != NULL)
1099 next->next->prev = next->prev;
1100
1101 kfree_skb(next->skb,FREE_READ);
1102 kfree_s(next, sizeof(struct ipfrag));
1103 }
1104 DPRINTF((DBG_IP, "IP: defrag: fixed high overlap %d bytes\n", i));
1105 }
1106
1107 /*
1108 * Insert this fragment in the chain of fragments.
1109 */
1110
1111 tfp = NULL;
1112 tfp = ip_frag_create(offset, end, skb, ptr);
1113
1114 /*
1115 * No memory to save the fragment - so throw the lot
1116 */
1117
1118 if (!tfp)
1119 {
1120 skb->sk = NULL;
1121 kfree_skb(skb, FREE_READ);
1122 return NULL;
1123 }
1124 tfp->prev = prev;
1125 tfp->next = next;
1126 if (prev != NULL)
1127 prev->next = tfp;
1128 else
1129 qp->fragments = tfp;
1130
1131 if (next != NULL)
1132 next->prev = tfp;
1133
1134 /*
1135 * OK, so we inserted this new fragment into the chain.
1136 * Check if we now have a full IP datagram which we can
1137 * bump up to the IP layer...
1138 */
1139
1140 if (ip_done(qp))
1141 {
1142 skb2 = ip_glue(qp); /* glue together the fragments */
1143 return(skb2);
1144 }
1145 return(NULL);
1146 }
1147
1148
1149 /*
1150 * This IP datagram is too large to be sent in one piece. Break it up into
1151 * smaller pieces (each of size equal to the MAC header plus IP header plus
1152 * a block of the data of the original IP data part) that will yet fit in a
1153 * single device frame, and queue such a frame for sending by calling the
1154 * ip_queue_xmit(). Note that this is recursion, and bad things will happen
1155 * if this function causes a loop...
1156 *
1157 * Yes this is inefficient, feel free to submit a quicker one.
1158 *
1159 * **Protocol Violation**
1160 * We copy all the options to each fragment. !FIXME!
1161 */
1162
1163 void ip_fragment(struct sock *sk, struct sk_buff *skb, struct device *dev, int is_frag)
/* */
1164 {
1165 struct iphdr *iph;
1166 unsigned char *raw;
1167 unsigned char *ptr;
1168 struct sk_buff *skb2;
1169 int left, mtu, hlen, len;
1170 int offset;
1171
1172 /*
1173 * Point into the IP datagram header.
1174 */
1175
1176 raw = skb->data;
1177 iph = (struct iphdr *) (raw + dev->hard_header_len);
1178
1179 skb->ip_hdr = iph;
1180
1181 /*
1182 * Setup starting values.
1183 */
1184
1185 hlen = (iph->ihl * sizeof(unsigned long));
1186 left = ntohs(iph->tot_len) - hlen; /* Space per frame */
1187 hlen += dev->hard_header_len; /* Total header size */
1188 mtu = (dev->mtu - hlen); /* Size of data space */
1189 ptr = (raw + hlen); /* Where to start from */
1190
1191 DPRINTF((DBG_IP, "IP: Fragmentation Desired\n"));
1192 DPRINTF((DBG_IP, " DEV=%s, MTU=%d, LEN=%d SRC=%s",
1193 dev->name, dev->mtu, left, in_ntoa(iph->saddr)));
1194 DPRINTF((DBG_IP, " DST=%s\n", in_ntoa(iph->daddr)));
1195
1196 /*
1197 * Check for any "DF" flag. [DF means do not fragment]
1198 */
1199
1200 if (ntohs(iph->frag_off) & IP_DF)
1201 {
1202 DPRINTF((DBG_IP, "IP: Fragmentation Desired, but DF set !\n"));
1203 DPRINTF((DBG_IP, " DEV=%s, MTU=%d, LEN=%d SRC=%s",
1204 dev->name, dev->mtu, left, in_ntoa(iph->saddr)));
1205 DPRINTF((DBG_IP, " DST=%s\n", in_ntoa(iph->daddr)));
1206
1207 ip_statistics.IpFragFails++;
1208 icmp_send(skb,ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, dev);
1209 return;
1210 }
1211
1212 /*
1213 * The protocol doesn't seem to say what to do in the case that the
1214 * frame + options doesn't fit the mtu. As it used to fall down dead
1215 * in this case we were fortunate it didn't happen
1216 */
1217
1218 if(mtu<8)
1219 {
1220 /* It's wrong but its better than nothing */
1221 icmp_send(skb,ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED,dev);
1222 ip_statistics.IpFragFails++;
1223 return;
1224 }
1225
1226 /*
1227 * Fragment the datagram.
1228 */
1229
1230 /*
1231 * The initial offset is 0 for a complete frame. When
1232 * fragmenting fragments its wherever this one starts.
1233 */
1234
1235 if (is_frag & 2)
1236 offset = (ntohs(iph->frag_off) & 0x1fff) << 3;
1237 else
1238 offset = 0;
1239
1240
1241 /*
1242 * Keep copying data until we run out.
1243 */
1244
1245 while(left > 0)
1246 {
1247 len = left;
1248 /* IF: it doesn't fit, use 'mtu' - the data space left */
1249 if (len > mtu)
1250 len = mtu;
1251 /* IF: we are not sending upto and including the packet end
1252 then align the next start on an eight byte boundary */
1253 if (len < left)
1254 {
1255 len/=8;
1256 len*=8;
1257 }
1258 DPRINTF((DBG_IP,"IP: frag: creating fragment of %d bytes (%d total)\n",
1259 len, len + hlen));
1260
1261 /*
1262 * Allocate buffer.
1263 */
1264
1265 if ((skb2 = alloc_skb(len + hlen,GFP_ATOMIC)) == NULL)
1266 {
1267 printk("IP: frag: no memory for new fragment!\n");
1268 ip_statistics.IpFragFails++;
1269 return;
1270 }
1271
1272 /*
1273 * Set up data on packet
1274 */
1275
1276 skb2->arp = 0;/*skb->arp;*/
1277 skb2->free = skb->free;
1278 skb2->len = len + hlen;
1279 skb2->h.raw=(char *) skb2->data;
1280
1281 /*
1282 * Charge the memory for the fragment to any owner
1283 * it might posess
1284 */
1285
1286 if (sk)
1287 sk->wmem_alloc += skb2->mem_len;
1288
1289 /*
1290 * Copy the packet header into the new buffer.
1291 */
1292
1293 memcpy(skb2->h.raw, raw, hlen);
1294
1295 /*
1296 * Copy a block of the IP datagram.
1297 */
1298 memcpy(skb2->h.raw + hlen, ptr, len);
1299 left -= len;
1300
1301 skb2->h.raw+=dev->hard_header_len;
1302
1303 /*
1304 * Fill in the new header fields.
1305 */
1306 iph = (struct iphdr *)(skb2->h.raw/*+dev->hard_header_len*/);
1307 iph->frag_off = htons((offset >> 3));
1308 /*
1309 * Added AC : If we are fragmenting a fragment thats not the
1310 * last fragment then keep MF on each bit
1311 */
1312 if (left > 0 || (is_frag & 1))
1313 iph->frag_off |= htons(IP_MF);
1314 ptr += len;
1315 offset += len;
1316
1317 /*
1318 * Put this fragment into the sending queue.
1319 */
1320
1321 ip_statistics.IpFragCreates++;
1322
1323 ip_queue_xmit(sk, dev, skb2, 1);
1324 }
1325 ip_statistics.IpFragOKs++;
1326 }
1327
1328
1329
1330 #ifdef CONFIG_IP_FORWARD
1331
1332 /*
1333 * Forward an IP datagram to its next destination.
1334 */
1335
1336 static void ip_forward(struct sk_buff *skb, struct device *dev, int is_frag)
/* */
1337 {
1338 struct device *dev2; /* Output device */
1339 struct iphdr *iph; /* Our header */
1340 struct sk_buff *skb2; /* Output packet */
1341 struct rtable *rt; /* Route we use */
1342 unsigned char *ptr; /* Data pointer */
1343 unsigned long raddr; /* Router IP address */
1344
1345 /*
1346 * Only forward packets that were fired at us when we are in promiscuous
1347 * mode. In standard mode we rely on the driver to filter for us.
1348 */
1349
1350 if(dev->flags&IFF_PROMISC)
1351 {
1352 if(memcmp((char *)&skb[1],dev->dev_addr,dev->addr_len))
1353 return;
1354 }
1355
1356
1357
1358 /*
1359 * According to the RFC, we must first decrease the TTL field. If
1360 * that reaches zero, we must reply an ICMP control message telling
1361 * that the packet's lifetime expired.
1362 *
1363 * Exception:
1364 * We may not generate an ICMP for an ICMP. icmp_send does the
1365 * enforcement of this so we can forget it here. It is however
1366 * sometimes VERY important.
1367 */
1368
1369 iph = skb->h.iph;
1370 iph->ttl--;
1371 if (iph->ttl <= 0)
1372 {
1373 DPRINTF((DBG_IP, "\nIP: *** datagram expired: TTL=0 (ignored) ***\n"));
1374 DPRINTF((DBG_IP, " SRC = %s ", in_ntoa(iph->saddr)));
1375 DPRINTF((DBG_IP, " DST = %s (ignored)\n", in_ntoa(iph->daddr)));
1376
1377 /* Tell the sender its packet died... */
1378 icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, dev);
1379 return;
1380 }
1381
1382 /*
1383 * Re-compute the IP header checksum.
1384 * This is inefficient. We know what has happened to the header
1385 * and could thus adjust the checksum as Phil Karn does in KA9Q
1386 */
1387
1388 ip_send_check(iph);
1389
1390 /*
1391 * OK, the packet is still valid. Fetch its destination address,
1392 * and give it to the IP sender for further processing.
1393 */
1394
1395 rt = ip_rt_route(iph->daddr, NULL, NULL);
1396 if (rt == NULL)
1397 {
1398 DPRINTF((DBG_IP, "\nIP: *** routing (phase I) failed ***\n"));
1399
1400 /*
1401 * Tell the sender its packet cannot be delivered. Again
1402 * ICMP is screened later.
1403 */
1404 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_NET_UNREACH, dev);
1405 return;
1406 }
1407
1408
1409 /*
1410 * Gosh. Not only is the packet valid; we even know how to
1411 * forward it onto its final destination. Can we say this
1412 * is being plain lucky?
1413 * If the router told us that there is no GW, use the dest.
1414 * IP address itself- we seem to be connected directly...
1415 */
1416
1417 raddr = rt->rt_gateway;
1418
1419 if (raddr != 0)
1420 {
1421 /*
1422 * There is a gateway so find the correct route for it.
1423 * Gateways cannot in turn be gatewayed.
1424 */
1425 rt = ip_rt_route(raddr, NULL, NULL);
1426 if (rt == NULL)
1427 {
1428 DPRINTF((DBG_IP, "\nIP: *** routing (phase II) failed ***\n"));
1429
1430 /*
1431 * Tell the sender its packet cannot be delivered...
1432 */
1433 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, dev);
1434 return;
1435 }
1436 if (rt->rt_gateway != 0)
1437 raddr = rt->rt_gateway;
1438 }
1439 else
1440 raddr = iph->daddr;
1441
1442 /*
1443 * Having picked a route we can now send the frame out.
1444 */
1445
1446 dev2 = rt->rt_dev;
1447
1448 /*
1449 * In IP you never forward a frame on the interface that it arrived
1450 * upon. We should generate an ICMP HOST REDIRECT giving the route
1451 * we calculated.
1452 * For now just dropping the packet is an acceptable compromise.
1453 */
1454
1455 if (dev == dev2)
1456 return;
1457
1458 /*
1459 * We now allocate a new buffer, and copy the datagram into it.
1460 * If the indicated interface is up and running, kick it.
1461 */
1462
1463 DPRINTF((DBG_IP, "\nIP: *** fwd %s -> ", in_ntoa(iph->saddr)));
1464 DPRINTF((DBG_IP, "%s (via %s), LEN=%d\n",
1465 in_ntoa(raddr), dev2->name, skb->len));
1466
1467 if (dev2->flags & IFF_UP)
1468 {
1469
1470 /*
1471 * Current design decrees we copy the packet. For identical header
1472 * lengths we could avoid it. The new skb code will let us push
1473 * data so the problem goes away then.
1474 */
1475
1476 skb2 = alloc_skb(dev2->hard_header_len + skb->len, GFP_ATOMIC);
1477 /*
1478 * This is rare and since IP is tolerant of network failures
1479 * quite harmless.
1480 */
1481 if (skb2 == NULL)
1482 {
1483 printk("\nIP: No memory available for IP forward\n");
1484 return;
1485 }
1486 ptr = skb2->data;
1487 skb2->free = 1;
1488 skb2->len = skb->len + dev2->hard_header_len;
1489 skb2->h.raw = ptr;
1490
1491 /*
1492 * Copy the packet data into the new buffer.
1493 */
1494 memcpy(ptr + dev2->hard_header_len, skb->h.raw, skb->len);
1495
1496 /* Now build the MAC header. */
1497 (void) ip_send(skb2, raddr, skb->len, dev2, dev2->pa_addr);
1498
1499 ip_statistics.IpForwDatagrams++;
1500
1501 /*
1502 * See if it needs fragmenting. Note in ip_rcv we tagged
1503 * the fragment type. This must be right so that
1504 * the fragmenter does the right thing.
1505 */
1506
1507 if(skb2->len > dev2->mtu)
1508 {
1509 ip_fragment(NULL,skb2,dev2, is_frag);
1510 kfree_skb(skb2,FREE_WRITE);
1511 }
1512 else
1513 {
1514 /*
1515 * Map service types to priority. We lie about
1516 * throughput being low priority, but its a good
1517 * choice to help improve general usage.
1518 */
1519 if(iph->tos & IPTOS_LOWDELAY)
1520 dev_queue_xmit(skb2, dev2, SOPRI_INTERACTIVE);
1521 else if(iph->tos & IPTOS_THROUGHPUT)
1522 dev_queue_xmit(skb2, dev2, SOPRI_BACKGROUND);
1523 else
1524 dev_queue_xmit(skb2, dev2, SOPRI_NORMAL);
1525 }
1526 }
1527 }
1528
1529
1530 #endif
1531
1532 /*
1533 * This function receives all incoming IP datagrams.
1534 */
1535
1536 int ip_rcv(struct sk_buff *skb, struct device *dev, struct packet_type *pt)
/* */
1537 {
1538 struct iphdr *iph = skb->h.iph;
1539 unsigned char hash;
1540 unsigned char flag = 0;
1541 unsigned char opts_p = 0; /* Set iff the packet has options. */
1542 struct inet_protocol *ipprot;
1543 static struct options opt; /* since we don't use these yet, and they
1544 take up stack space. */
1545 int brd;
1546 int is_frag=0;
1547
1548
1549 ip_statistics.IpInReceives++;
1550
1551 DPRINTF((DBG_IP, "<<\n"));
1552
1553 /*
1554 * Tag the ip header of this packet so we can find it
1555 */
1556
1557 skb->ip_hdr = iph;
1558
1559 /*
1560 * Is the datagram acceptable?
1561 *
1562 * 1. Length at least the size of an ip header
1563 * 2. Version of 4
1564 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
1565 * (4. We ought to check for IP multicast addresses and undefined types.. does this matter ?)
1566 */
1567
1568 if (skb->len<sizeof(struct iphdr) || iph->ihl<5 || iph->version != 4 || ip_fast_csum((unsigned char *)iph, iph->ihl) !=0)
1569 {
1570 ip_statistics.IpInHdrErrors++;
1571 DPRINTF((DBG_IP, "\nIP: *** datagram error ***\n"));
1572 DPRINTF((DBG_IP, " SRC = %s ", in_ntoa(iph->saddr)));
1573 DPRINTF((DBG_IP, " DST = %s (ignored)\n", in_ntoa(iph->daddr)));
1574 kfree_skb(skb, FREE_WRITE);
1575 return(0);
1576 }
1577
1578 /*
1579 * Our transport medium may have padded the buffer out. Now we know it
1580 * is IP we can trim to the true length of the frame.
1581 */
1582
1583 skb->len=ntohs(iph->tot_len);
1584
1585 /*
1586 * Next anaylse the packet for options. Studies show under one packet in
1587 * a thousand have options....
1588 */
1589
1590 if (iph->ihl != 5)
1591 { /* Fast path for the typical optionless IP packet. */
1592 ip_print(iph); /* Bogus, only for debugging. */
1593 memset((char *) &opt, 0, sizeof(opt));
1594 if (do_options(iph, &opt) != 0)
1595 return 0;
1596 opts_p = 1;
1597 }
1598
1599 /*
1600 * Remember if the frame is fragmented.
1601 */
1602
1603 if (iph->frag_off & 0x0020)
1604 is_frag|=1;
1605
1606 /*
1607 * Last fragment ?
1608 */
1609
1610 if (ntohs(iph->frag_off) & 0x1fff)
1611 is_frag|=2;
1612
1613 /*
1614 * Do any IP forwarding required. chk_addr() is expensive -- avoid it someday.
1615 *
1616 * This is inefficient. While finding out if it is for us we could also compute
1617 * the routing table entry. This is where the great unified cache theory comes
1618 * in as and when someone impliments it
1619 */
1620
1621 if ((brd = ip_chk_addr(iph->daddr)) == 0)
1622 {
1623
1624 /*
1625 * The packet is for another target. Forward the frame
1626 */
1627
1628 #ifdef CONFIG_IP_FORWARD
1629 ip_forward(skb, dev, is_frag);
1630 #else
1631 printk("Machine %x tried to use us as a forwarder to %x but we have forwarding disabled!\n",
1632 iph->saddr,iph->daddr);
1633 ip_statistics.IpInAddrErrors++;
1634 #endif
1635 /*
1636 * The forwarder is inefficient and copies the packet. We
1637 * free the original now.
1638 */
1639
1640 kfree_skb(skb, FREE_WRITE);
1641 return(0);
1642 }
1643
1644 /*
1645 * Reassemble IP fragments.
1646 */
1647
1648 if(is_frag)
1649 {
1650 #ifdef CONFIG_IP_DEFRAG
1651 /* Defragment. Obtain the complete packet if there is one */
1652 skb=ip_defrag(iph,skb,dev);
1653 if(skb==NULL)
1654 return 0;
1655 iph=skb->h.iph;
1656 #else
1657 printk("\nIP: *** datagram fragmentation not yet implemented ***\n");
1658 printk(" SRC = %s ", in_ntoa(iph->saddr));
1659 printk(" DST = %s (ignored)\n", in_ntoa(iph->daddr));
1660 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, dev);
1661 kfree_skb(skb, FREE_WRITE);
1662 return(0);
1663 #endif
1664 }
1665
1666 /*
1667 * Point into the IP datagram, just past the header.
1668 */
1669
1670 skb->ip_hdr = iph;
1671 skb->h.raw += iph->ihl*4;
1672
1673 /*
1674 * skb->h.raw now points at the protocol beyond the IP header.
1675 */
1676
1677 hash = iph->protocol & (MAX_INET_PROTOS -1);
1678 for (ipprot = (struct inet_protocol *)inet_protos[hash];ipprot != NULL;ipprot=(struct inet_protocol *)ipprot->next)
1679 {
1680 struct sk_buff *skb2;
1681
1682 if (ipprot->protocol != iph->protocol)
1683 continue;
1684 DPRINTF((DBG_IP, "Using protocol = %X:\n", ipprot));
1685 print_ipprot(ipprot);
1686
1687 /*
1688 * See if we need to make a copy of it. This will
1689 * only be set if more than one protocol wants it.
1690 * and then not for the last one.
1691 *
1692 * This is an artifact of poor upper protocol design.
1693 * Because the upper protocols damage the actual packet
1694 * we must do copying. In actual fact it's even worse
1695 * than this as TCP may hold on to the buffer.
1696 */
1697 if (ipprot->copy)
1698 {
1699 #if 0
1700 skb2 = alloc_skb(skb->mem_len-sizeof(struct sk_buff), GFP_ATOMIC);
1701 if (skb2 == NULL)
1702 continue;
1703 memcpy(skb2, skb, skb2->mem_len);
1704 skb2->ip_hdr = (struct iphdr *)(
1705 (unsigned long)skb2 +
1706 (unsigned long) skb->ip_hdr -
1707 (unsigned long)skb);
1708 skb2->h.raw = (unsigned char *)(
1709 (unsigned long)skb2 +
1710 (unsigned long) skb->h.raw -
1711 (unsigned long)skb);
1712 skb2->free=1;
1713 #else
1714 skb2 = skb_clone(skb, GFP_ATOMIC);
1715 if(skb2==NULL)
1716 continue;
1717 #endif
1718 }
1719 else
1720 {
1721 skb2 = skb;
1722 }
1723 flag = 1;
1724
1725 /*
1726 * Pass on the datagram to each protocol that wants it,
1727 * based on the datagram protocol. We should really
1728 * check the protocol handler's return values here...
1729 */
1730 ipprot->handler(skb2, dev, opts_p ? &opt : 0, iph->daddr,
1731 (ntohs(iph->tot_len) - (iph->ihl * 4)),
1732 iph->saddr, 0, ipprot);
1733
1734 }
1735
1736 /*
1737 * All protocols checked.
1738 * If this packet was a broadcast, we may *not* reply to it, since that
1739 * causes (proven, grin) ARP storms and a leakage of memory (i.e. all
1740 * ICMP reply messages get queued up for transmission...)
1741 */
1742
1743 if (!flag)
1744 {
1745 if (brd != IS_BROADCAST)
1746 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, dev);
1747 kfree_skb(skb, FREE_WRITE);
1748 }
1749
1750 return(0);
1751 }
1752
1753
1754 /*
1755 * Queues a packet to be sent, and starts the transmitter
1756 * if necessary. if free = 1 then we free the block after
1757 * transmit, otherwise we don't.
1758 * This routine also needs to put in the total length,
1759 * and compute the checksum
1760 */
1761
1762 void ip_queue_xmit(struct sock *sk, struct device *dev,
/* */
1763 struct sk_buff *skb, int free)
1764 {
1765 struct iphdr *iph;
1766 unsigned char *ptr;
1767
1768 /* All buffers without an owner socket get freed */
1769 if (sk == NULL)
1770 free = 1;
1771
1772 /* Sanity check */
1773 if (dev == NULL)
1774 {
1775 printk("IP: ip_queue_xmit dev = NULL\n");
1776 return;
1777 }
1778
1779 IS_SKB(skb);
1780
1781 /*
1782 * Do some book-keeping in the packet for later
1783 */
1784
1785 skb->free = free;
1786 skb->dev = dev;
1787 skb->when = jiffies;
1788
1789 DPRINTF((DBG_IP, ">>\n"));
1790
1791 /*
1792 * Find the IP header and set the length. This is bad
1793 * but once we get the skb data handling code in the
1794 * hardware will push its header sensibly and we will
1795 * set skb->ip_hdr to avoid this mess and the fixed
1796 * header length problem
1797 */
1798
1799 ptr = skb->data;
1800 ptr += dev->hard_header_len;
1801 iph = (struct iphdr *)ptr;
1802 skb->ip_hdr = iph;
1803 iph->tot_len = ntohs(skb->len-dev->hard_header_len);
1804
1805 /*
1806 * Do we need to fragment. Again this is inefficient.
1807 * We need to somehow lock the original buffer and use
1808 * bits of it.
1809 */
1810
1811 if(skb->len > dev->mtu)
1812 {
1813 ip_fragment(sk,skb,dev,0);
1814 IS_SKB(skb);
1815 kfree_skb(skb,FREE_WRITE);
1816 return;
1817 }
1818
1819 /*
1820 * Add an IP checksum
1821 */
1822
1823 ip_send_check(iph);
1824
1825 /*
1826 * Print the frame when debugging
1827 */
1828 ip_print(iph);
1829
1830 /*
1831 * More debugging. You cannot queue a packet already on a list
1832 * Spot this and moan loudly.
1833 */
1834 if (skb->next != NULL)
1835 {
1836 printk("ip_queue_xmit: next != NULL\n");
1837 skb_unlink(skb);
1838 }
1839
1840 /*
1841 * If a sender wishes the packet to remain unfreed
1842 * we add it to his send queue. This arguably belongs
1843 * in the TCP level since nobody elses uses it. BUT
1844 * remember IPng might change all the rules.
1845 */
1846
1847 if (!free)
1848 {
1849 unsigned long flags;
1850 /* The socket now has more outstanding blocks */
1851
1852 sk->packets_out++;
1853
1854 /* Protect the list for a moment */
1855 save_flags(flags);
1856 cli();
1857
1858 if (skb->link3 != NULL)
1859 {
1860 printk("ip.c: link3 != NULL\n");
1861 skb->link3 = NULL;
1862 }
1863 if (sk->send_head == NULL)
1864 {
1865 sk->send_tail = skb;
1866 sk->send_head = skb;
1867 }
1868 else
1869 {
1870 sk->send_tail->link3 = skb;
1871 sk->send_tail = skb;
1872 }
1873 /* skb->link3 is NULL */
1874
1875 /* Interrupt restore */
1876 restore_flags(flags);
1877 /* Set the IP write timeout to the round trip time for the packet.
1878 If an acknowledge has not arrived by then we may wish to act */
1879 reset_timer(sk, TIME_WRITE, sk->rto);
1880 }
1881 else
1882 /* Remember who owns the buffer */
1883 skb->sk = sk;
1884
1885 /*
1886 * If the indicated interface is up and running, send the packet.
1887 */
1888 ip_statistics.IpOutRequests++;
1889
1890 if (dev->flags & IFF_UP)
1891 {
1892 /*
1893 * If we have an owner use its priority setting,
1894 * otherwise use NORMAL
1895 */
1896
1897 if (sk != NULL)
1898 {
1899 dev_queue_xmit(skb, dev, sk->priority);
1900 }
1901 else
1902 {
1903 dev_queue_xmit(skb, dev, SOPRI_NORMAL);
1904 }
1905 }
1906 else
1907 {
1908 ip_statistics.IpOutDiscards++;
1909 if (free)
1910 kfree_skb(skb, FREE_WRITE);
1911 }
1912 }
1913
1914
1915 /*
1916 * A socket has timed out on its send queue and wants to do a
1917 * little retransmitting. Currently this means TCP.
1918 */
1919
1920 void ip_do_retransmit(struct sock *sk, int all)
/* */
1921 {
1922 struct sk_buff * skb;
1923 struct proto *prot;
1924 struct device *dev;
1925 int retransmits;
1926
1927 prot = sk->prot;
1928 skb = sk->send_head;
1929 retransmits = sk->retransmits;
1930
1931 while (skb != NULL)
1932 {
1933 dev = skb->dev;
1934 IS_SKB(skb);
1935 #if 0
1936 /********** THIS IS NOW DONE BY THE DEVICE LAYER **********/
1937 /*
1938 * The rebuild_header function sees if the ARP is done.
1939 * If not it sends a new ARP request, and if so it builds
1940 * the header. It isn't really needed here, and with the
1941 * new ARP pretty much will not happen.
1942 */
1943
1944 if (!skb->arp)
1945 {
1946 if (dev->rebuild_header(skb->data, dev, skb->raddr, NULL))
1947 {
1948 if (!all)
1949 break;
1950 skb = skb->link3;
1951 continue;
1952 }
1953 }
1954 #endif
1955 skb->when = jiffies;
1956
1957 /*
1958 * If the interface is (still) up and running, kick it.
1959 */
1960
1961 if (dev->flags & IFF_UP)
1962 {
1963 /*
1964 * If the packet is still being sent by the device/protocol
1965 * below then don't retransmit. This is both needed, and good -
1966 * especially with connected mode AX.25 where it stops resends
1967 * occuring of an as yet unsent anyway frame!
1968 * We still add up the counts as the round trip time wants
1969 * adjusting.
1970 */
1971 if (sk && !skb_device_locked(skb))
1972 {
1973 /* Remove it from any existing driver queue first! */
1974 skb_unlink(skb);
1975 /* Now queue it */
1976 ip_statistics.IpOutRequests++;
1977 dev_queue_xmit(skb, dev, sk->priority);
1978 }
1979 }
1980
1981 /*
1982 * Count retransmissions
1983 */
1984 retransmits++;
1985 sk->prot->retransmits ++;
1986
1987 /*
1988 * Only one retransmit requested.
1989 */
1990 if (!all)
1991 break;
1992
1993 /*
1994 * This should cut it off before we send too many packets.
1995 */
1996 if (sk->retransmits > sk->cong_window)
1997 break;
1998 skb = skb->link3;
1999 }
2000 }
2001
2002 /*
2003 * This is the normal code called for timeouts. It does the retransmission
2004 * and then does backoff. ip_do_retransmit is separated out because
2005 * tcp_ack needs to send stuff from the retransmit queue without
2006 * initiating a backoff.
2007 */
2008
2009 void ip_retransmit(struct sock *sk, int all)
/* */
2010 {
2011 ip_do_retransmit(sk, all);
2012
2013 /*
2014 * Increase the timeout each time we retransmit. Note that
2015 * we do not increase the rtt estimate. rto is initialized
2016 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
2017 * that doubling rto each time is the least we can get away with.
2018 * In KA9Q, Karn uses this for the first few times, and then
2019 * goes to quadratic. netBSD doubles, but only goes up to *64,
2020 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
2021 * defined in the protocol as the maximum possible RTT. I guess
2022 * we'll have to use something other than TCP to talk to the
2023 * University of Mars.
2024 */
2025
2026 sk->retransmits++;
2027 sk->backoff++;
2028 sk->rto = min(sk->rto << 1, 120*HZ);
2029 reset_timer(sk, TIME_WRITE, sk->rto);
2030 }
2031
2032 /*
2033 * Socket option code for IP. This is the end of the line after any TCP,UDP etc options on
2034 * an IP socket.
2035 *
2036 * We impliment IP_TOS (type of service), IP_TTL (time to live).
2037 *
2038 * Next release we will sort out IP_OPTIONS since for some people are kind of important.
2039 */
2040
2041 int ip_setsockopt(struct sock *sk, int level, int optname, char *optval, int optlen)
/* */
2042 {
2043 int val,err;
2044
2045 if (optval == NULL)
2046 return(-EINVAL);
2047
2048 err=verify_area(VERIFY_READ, optval, sizeof(int));
2049 if(err)
2050 return err;
2051
2052 val = get_fs_long((unsigned long *)optval);
2053
2054 if(level!=SOL_IP)
2055 return -EOPNOTSUPP;
2056
2057 switch(optname)
2058 {
2059 case IP_TOS:
2060 if(val<0||val>255)
2061 return -EINVAL;
2062 sk->ip_tos=val;
2063 return 0;
2064 case IP_TTL:
2065 if(val<1||val>255)
2066 return -EINVAL;
2067 sk->ip_ttl=val;
2068 return 0;
2069 /* IP_OPTIONS and friends go here eventually */
2070 default:
2071 return(-ENOPROTOOPT);
2072 }
2073 }
2074
2075 /*
2076 * Get the options. Note for future reference. The GET of IP options gets the
2077 * _received_ ones. The set sets the _sent_ ones.
2078 */
2079
2080 int ip_getsockopt(struct sock *sk, int level, int optname, char *optval, int *optlen)
/* */
2081 {
2082 int val,err;
2083
2084 if(level!=SOL_IP)
2085 return -EOPNOTSUPP;
2086
2087 switch(optname)
2088 {
2089 case IP_TOS:
2090 val=sk->ip_tos;
2091 break;
2092 case IP_TTL:
2093 val=sk->ip_ttl;
2094 break;
2095 default:
2096 return(-ENOPROTOOPT);
2097 }
2098 err=verify_area(VERIFY_WRITE, optlen, sizeof(int));
2099 if(err)
2100 return err;
2101 put_fs_long(sizeof(int),(unsigned long *) optlen);
2102
2103 err=verify_area(VERIFY_WRITE, optval, sizeof(int));
2104 if(err)
2105 return err;
2106 put_fs_long(val,(unsigned long *)optval);
2107
2108 return(0);
2109 }
2110
2111 /*
2112 * IP protocol layer initialiser
2113 */
2114
2115 static struct packet_type ip_packet_type =
2116 {
2117 0, /* MUTTER ntohs(ETH_P_IP),*/
2118 0, /* copy */
2119 ip_rcv,
2120 NULL,
2121 NULL,
2122 };
2123
2124
2125 /*
2126 * IP registers the packet type and then calls the subprotocol initialisers
2127 */
2128
2129 void ip_init(void)
/* ![[last]](../icons/n_last.png)
*/
2130 {
2131 ip_packet_type.type=htons(ETH_P_IP);
2132 dev_add_pack(&ip_packet_type);
2133 /* ip_raw_init();
2134 ip_packet_init();
2135 ip_tcp_init();
2136 ip_udp_init();*/
2137 }
![[bottom]](../icons/n_bottom.png){kind=icon}
*/