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