![[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 * Richard Underwood
15 * Stefan Becker, <stefanb@yello.ping.de>
16 * Jorge Cwik, <jorge@laser.satlink.net>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 *
19 *
20 * Fixes:
21 * Alan Cox : Commented a couple of minor bits of surplus code
22 * Alan Cox : Undefining IP_FORWARD doesn't include the code
23 * (just stops a compiler warning).
24 * Alan Cox : Frames with >=MAX_ROUTE record routes, strict routes or loose routes
25 * are junked rather than corrupting things.
26 * Alan Cox : Frames to bad broadcast subnets are dumped
27 * We used to process them non broadcast and
28 * boy could that cause havoc.
29 * Alan Cox : ip_forward sets the free flag on the
30 * new frame it queues. Still crap because
31 * it copies the frame but at least it
32 * doesn't eat memory too.
33 * Alan Cox : Generic queue code and memory fixes.
34 * Fred Van Kempen : IP fragment support (borrowed from NET2E)
35 * Gerhard Koerting: Forward fragmented frames correctly.
36 * Gerhard Koerting: Fixes to my fix of the above 8-).
37 * Gerhard Koerting: IP interface addressing fix.
38 * Linus Torvalds : More robustness checks
39 * Alan Cox : Even more checks: Still not as robust as it ought to be
40 * Alan Cox : Save IP header pointer for later
41 * Alan Cox : ip option setting
42 * Alan Cox : Use ip_tos/ip_ttl settings
43 * Alan Cox : Fragmentation bogosity removed
44 * (Thanks to Mark.Bush@prg.ox.ac.uk)
45 * Dmitry Gorodchanin : Send of a raw packet crash fix.
46 * Alan Cox : Silly ip bug when an overlength
47 * fragment turns up. Now frees the
48 * queue.
49 * Linus Torvalds/ : Memory leakage on fragmentation
50 * Alan Cox : handling.
51 * Gerhard Koerting: Forwarding uses IP priority hints
52 * Teemu Rantanen : Fragment problems.
53 * Alan Cox : General cleanup, comments and reformat
54 * Alan Cox : SNMP statistics
55 * Alan Cox : BSD address rule semantics. Also see
56 * UDP as there is a nasty checksum issue
57 * if you do things the wrong way.
58 * Alan Cox : Always defrag, moved IP_FORWARD to the config.in file
59 * Alan Cox : IP options adjust sk->priority.
60 * Pedro Roque : Fix mtu/length error in ip_forward.
61 * Alan Cox : Avoid ip_chk_addr when possible.
62 * Richard Underwood : IP multicasting.
63 * Alan Cox : Cleaned up multicast handlers.
64 * Alan Cox : RAW sockets demultiplex in the BSD style.
65 * Gunther Mayer : Fix the SNMP reporting typo
66 * Alan Cox : Always in group 224.0.0.1
67 * Pauline Middelink : Fast ip_checksum update when forwarding
68 * Masquerading support.
69 * Alan Cox : Multicast loopback error for 224.0.0.1
70 * Alan Cox : IP_MULTICAST_LOOP option.
71 * Alan Cox : Use notifiers.
72 * Bjorn Ekwall : Removed ip_csum (from slhc.c too)
73 * Bjorn Ekwall : Moved ip_fast_csum to ip.h (inline!)
74 * Stefan Becker : Send out ICMP HOST REDIRECT
75 * Arnt Gulbrandsen : ip_build_xmit
76 * Alan Cox : Per socket routing cache
77 * Alan Cox : Fixed routing cache, added header cache.
78 * Alan Cox : Loopback didnt work right in original ip_build_xmit - fixed it.
79 * Alan Cox : Only send ICMP_REDIRECT if src/dest are the same net.
80 * Alan Cox : Incoming IP option handling.
81 * Alan Cox : Set saddr on raw output frames as per BSD.
82 * Alan Cox : Stopped broadcast source route explosions.
83 * Alan Cox : Can disable source routing
84 * Takeshi Sone : Masquerading didn't work.
85 * Dave Bonn,Alan Cox : Faster IP forwarding whenever possible.
86 * Alan Cox : Memory leaks, tramples, misc debugging.
87 * Alan Cox : Fixed multicast (by popular demand 8))
88 * Alan Cox : Fixed forwarding (by even more popular demand 8))
89 * Alan Cox : Fixed SNMP statistics [I think]
90 * Gerhard Koerting : IP fragmentation forwarding fix
91 *
92 *
93 *
94 * To Fix:
95 * IP option processing is mostly not needed. ip_forward needs to know about routing rules
96 * and time stamp but that's about all. Use the route mtu field here too
97 * IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
98 * and could be made very efficient with the addition of some virtual memory hacks to permit
99 * the allocation of a buffer that can then be 'grown' by twiddling page tables.
100 * Output fragmentation wants updating along with the buffer management to use a single
101 * interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
102 * output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
103 * fragmentation anyway.
104 *
105 * FIXME: copy frag 0 iph to qp->iph
106 *
107 * This program is free software; you can redistribute it and/or
108 * modify it under the terms of the GNU General Public License
109 * as published by the Free Software Foundation; either version
110 * 2 of the License, or (at your option) any later version.
111 */
112
113 #include <asm/segment.h>
114 #include <asm/system.h>
115 #include <linux/types.h>
116 #include <linux/kernel.h>
117 #include <linux/sched.h>
118 #include <linux/mm.h>
119 #include <linux/string.h>
120 #include <linux/errno.h>
121 #include <linux/config.h>
122
123 #include <linux/socket.h>
124 #include <linux/sockios.h>
125 #include <linux/in.h>
126 #include <linux/inet.h>
127 #include <linux/netdevice.h>
128 #include <linux/etherdevice.h>
129 #include <linux/proc_fs.h>
130 #include <linux/stat.h>
131
132 #include <net/snmp.h>
133 #include <net/ip.h>
134 #include <net/protocol.h>
135 #include <net/route.h>
136 #include <net/tcp.h>
137 #include <net/udp.h>
138 #include <linux/skbuff.h>
139 #include <net/sock.h>
140 #include <net/arp.h>
141 #include <net/icmp.h>
142 #include <net/raw.h>
143 #include <net/checksum.h>
144 #include <linux/igmp.h>
145 #include <linux/ip_fw.h>
146
147 #define CONFIG_IP_DEFRAG
148
149 extern int last_retran;
150 extern void sort_send(struct sock *sk);
151
152 #define min(a,b) ((a)<(b)?(a):(b))
153
154 /*
155 * SNMP management statistics
156 */
157
158 #ifdef CONFIG_IP_FORWARD
159 struct ip_mib ip_statistics={1,64,}; /* Forwarding=Yes, Default TTL=64 */
160 #else
161 struct ip_mib ip_statistics={2,64,}; /* Forwarding=No, Default TTL=64 */
162 #endif
163
164 /*
165 * Handle the issuing of an ioctl() request
166 * for the ip device. This is scheduled to
167 * disappear
168 */
169
170 int ip_ioctl(struct sock *sk, int cmd, unsigned long arg)
/* ![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[top]](../icons/top.png)
*/
171 {
172 switch(cmd)
173 {
174 default:
175 return(-EINVAL);
176 }
177 }
178
179
180 /*
181 * Take an skb, and fill in the MAC header.
182 */
183
184 static int ip_send(struct sk_buff *skb, unsigned long daddr, int len, struct device *dev, unsigned long saddr)
/* ![[previous]](../icons/left.png)
*/
185 {
186 int mac = 0;
187
188 skb->dev = dev;
189 skb->arp = 1;
190 if (dev->hard_header)
191 {
192 /*
193 * Build a hardware header. Source address is our mac, destination unknown
194 * (rebuild header will sort this out)
195 */
196 skb_reserve(skb,(dev->hard_header_len+15)&~15); /* 16 byte aligned IP headers are good */
197 mac = dev->hard_header(skb, dev, ETH_P_IP, NULL, NULL, len);
198 if (mac < 0)
199 {
200 mac = -mac;
201 skb->arp = 0;
202 skb->raddr = daddr; /* next routing address */
203 }
204 }
205 return mac;
206 }
207
208 int ip_id_count = 0;
209
210 /*
211 * This routine builds the appropriate hardware/IP headers for
212 * the routine. It assumes that if *dev != NULL then the
213 * protocol knows what it's doing, otherwise it uses the
214 * routing/ARP tables to select a device struct.
215 */
216 int ip_build_header(struct sk_buff *skb, unsigned long saddr, unsigned long daddr,
/* */
217 struct device **dev, int type, struct options *opt, int len, int tos, int ttl)
218 {
219 struct rtable *rt;
220 unsigned long raddr;
221 int tmp;
222 unsigned long src;
223 struct iphdr *iph;
224
225 /*
226 * See if we need to look up the device.
227 */
228
229 #ifdef CONFIG_IP_MULTICAST
230 if(MULTICAST(daddr) && *dev==NULL && skb->sk && *skb->sk->ip_mc_name)
231 *dev=dev_get(skb->sk->ip_mc_name);
232 #endif
233 if (*dev == NULL)
234 {
235 if(skb->localroute)
236 rt = ip_rt_local(daddr, NULL, &src);
237 else
238 rt = ip_rt_route(daddr, NULL, &src);
239 if (rt == NULL)
240 {
241 ip_statistics.IpOutNoRoutes++;
242 return(-ENETUNREACH);
243 }
244
245 *dev = rt->rt_dev;
246 /*
247 * If the frame is from us and going off machine it MUST MUST MUST
248 * have the output device ip address and never the loopback
249 */
250 if (LOOPBACK(saddr) && !LOOPBACK(daddr))
251 saddr = src;/*rt->rt_dev->pa_addr;*/
252 raddr = rt->rt_gateway;
253
254 }
255 else
256 {
257 /*
258 * We still need the address of the first hop.
259 */
260 if(skb->localroute)
261 rt = ip_rt_local(daddr, NULL, &src);
262 else
263 rt = ip_rt_route(daddr, NULL, &src);
264 /*
265 * If the frame is from us and going off machine it MUST MUST MUST
266 * have the output device ip address and never the loopback
267 */
268 if (LOOPBACK(saddr) && !LOOPBACK(daddr))
269 saddr = src;/*rt->rt_dev->pa_addr;*/
270
271 raddr = (rt == NULL) ? 0 : rt->rt_gateway;
272 }
273
274 /*
275 * No source addr so make it our addr
276 */
277 if (saddr == 0)
278 saddr = src;
279
280 /*
281 * No gateway so aim at the real destination
282 */
283 if (raddr == 0)
284 raddr = daddr;
285
286 /*
287 * Now build the MAC header.
288 */
289
290 tmp = ip_send(skb, raddr, len, *dev, saddr);
291
292 /*
293 * Book keeping
294 */
295
296 skb->dev = *dev;
297 skb->saddr = saddr;
298 if (skb->sk)
299 skb->sk->saddr = saddr;
300
301 /*
302 * Now build the IP header.
303 */
304
305 /*
306 * If we are using IPPROTO_RAW, then we don't need an IP header, since
307 * one is being supplied to us by the user
308 */
309
310 if(type == IPPROTO_RAW)
311 return (tmp);
312
313 /*
314 * Build the IP addresses
315 */
316
317 iph=(struct iphdr *)skb_put(skb,sizeof(struct iphdr));
318
319 iph->version = 4;
320 iph->ihl = 5;
321 iph->tos = tos;
322 iph->frag_off = 0;
323 iph->ttl = ttl;
324 iph->daddr = daddr;
325 iph->saddr = saddr;
326 iph->protocol = type;
327 skb->ip_hdr = iph;
328
329 return(20 + tmp); /* IP header plus MAC header size */
330 }
331
332
333 /*
334 * Generate a checksum for an outgoing IP datagram.
335 */
336
337 void ip_send_check(struct iphdr *iph)
/* */
338 {
339 iph->check = 0;
340 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
341 }
342
343 /************************ Fragment Handlers From NET2E **********************************/
344
345
346 /*
347 * This fragment handler is a bit of a heap. On the other hand it works quite
348 * happily and handles things quite well.
349 */
350
351 static struct ipq *ipqueue = NULL; /* IP fragment queue */
352
353 /*
354 * Create a new fragment entry.
355 */
356
357 static struct ipfrag *ip_frag_create(int offset, int end, struct sk_buff *skb, unsigned char *ptr)
/* */
358 {
359 struct ipfrag *fp;
360
361 fp = (struct ipfrag *) kmalloc(sizeof(struct ipfrag), GFP_ATOMIC);
362 if (fp == NULL)
363 {
364 NETDEBUG(printk("IP: frag_create: no memory left !\n"));
365 return(NULL);
366 }
367 memset(fp, 0, sizeof(struct ipfrag));
368
369 /* Fill in the structure. */
370 fp->offset = offset;
371 fp->end = end;
372 fp->len = end - offset;
373 fp->skb = skb;
374 fp->ptr = ptr;
375
376 return(fp);
377 }
378
379
380 /*
381 * Find the correct entry in the "incomplete datagrams" queue for
382 * this IP datagram, and return the queue entry address if found.
383 */
384
385 static struct ipq *ip_find(struct iphdr *iph)
/* */
386 {
387 struct ipq *qp;
388 struct ipq *qplast;
389
390 cli();
391 qplast = NULL;
392 for(qp = ipqueue; qp != NULL; qplast = qp, qp = qp->next)
393 {
394 if (iph->id== qp->iph->id && iph->saddr == qp->iph->saddr &&
395 iph->daddr == qp->iph->daddr && iph->protocol == qp->iph->protocol)
396 {
397 del_timer(&qp->timer); /* So it doesn't vanish on us. The timer will be reset anyway */
398 sti();
399 return(qp);
400 }
401 }
402 sti();
403 return(NULL);
404 }
405
406
407 /*
408 * Remove an entry from the "incomplete datagrams" queue, either
409 * because we completed, reassembled and processed it, or because
410 * it timed out.
411 */
412
413 static void ip_free(struct ipq *qp)
/* */
414 {
415 struct ipfrag *fp;
416 struct ipfrag *xp;
417
418 /*
419 * Stop the timer for this entry.
420 */
421
422 del_timer(&qp->timer);
423
424 /* Remove this entry from the "incomplete datagrams" queue. */
425 cli();
426 if (qp->prev == NULL)
427 {
428 ipqueue = qp->next;
429 if (ipqueue != NULL)
430 ipqueue->prev = NULL;
431 }
432 else
433 {
434 qp->prev->next = qp->next;
435 if (qp->next != NULL)
436 qp->next->prev = qp->prev;
437 }
438
439 /* Release all fragment data. */
440
441 fp = qp->fragments;
442 while (fp != NULL)
443 {
444 xp = fp->next;
445 IS_SKB(fp->skb);
446 kfree_skb(fp->skb,FREE_READ);
447 kfree_s(fp, sizeof(struct ipfrag));
448 fp = xp;
449 }
450
451 /* Release the IP header. */
452 kfree_s(qp->iph, 64 + 8);
453
454 /* Finally, release the queue descriptor itself. */
455 kfree_s(qp, sizeof(struct ipq));
456 sti();
457 }
458
459
460 /*
461 * Oops- a fragment queue timed out. Kill it and send an ICMP reply.
462 */
463
464 static void ip_expire(unsigned long arg)
/* */
465 {
466 struct ipq *qp;
467
468 qp = (struct ipq *)arg;
469
470 /*
471 * Send an ICMP "Fragment Reassembly Timeout" message.
472 */
473
474 ip_statistics.IpReasmTimeout++;
475 ip_statistics.IpReasmFails++;
476 /* This if is always true... shrug */
477 if(qp->fragments!=NULL)
478 icmp_send(qp->fragments->skb,ICMP_TIME_EXCEEDED,
479 ICMP_EXC_FRAGTIME, 0, qp->dev);
480
481 /*
482 * Nuke the fragment queue.
483 */
484 ip_free(qp);
485 }
486
487
488 /*
489 * Add an entry to the 'ipq' queue for a newly received IP datagram.
490 * We will (hopefully :-) receive all other fragments of this datagram
491 * in time, so we just create a queue for this datagram, in which we
492 * will insert the received fragments at their respective positions.
493 */
494
495 static struct ipq *ip_create(struct sk_buff *skb, struct iphdr *iph, struct device *dev)
/* */
496 {
497 struct ipq *qp;
498 int ihlen;
499
500 qp = (struct ipq *) kmalloc(sizeof(struct ipq), GFP_ATOMIC);
501 if (qp == NULL)
502 {
503 NETDEBUG(printk("IP: create: no memory left !\n"));
504 return(NULL);
505 skb->dev = qp->dev;
506 }
507 memset(qp, 0, sizeof(struct ipq));
508
509 /*
510 * Allocate memory for the IP header (plus 8 octets for ICMP).
511 */
512
513 ihlen = iph->ihl * 4;
514 qp->iph = (struct iphdr *) kmalloc(64 + 8, GFP_ATOMIC);
515 if (qp->iph == NULL)
516 {
517 NETDEBUG(printk("IP: create: no memory left !\n"));
518 kfree_s(qp, sizeof(struct ipq));
519 return(NULL);
520 }
521
522 memcpy(qp->iph, iph, ihlen + 8);
523 qp->len = 0;
524 qp->ihlen = ihlen;
525 qp->fragments = NULL;
526 qp->dev = dev;
527
528 /* Start a timer for this entry. */
529 qp->timer.expires = jiffies + IP_FRAG_TIME; /* about 30 seconds */
530 qp->timer.data = (unsigned long) qp; /* pointer to queue */
531 qp->timer.function = ip_expire; /* expire function */
532 add_timer(&qp->timer);
533
534 /* Add this entry to the queue. */
535 qp->prev = NULL;
536 cli();
537 qp->next = ipqueue;
538 if (qp->next != NULL)
539 qp->next->prev = qp;
540 ipqueue = qp;
541 sti();
542 return(qp);
543 }
544
545
546 /*
547 * See if a fragment queue is complete.
548 */
549
550 static int ip_done(struct ipq *qp)
/* */
551 {
552 struct ipfrag *fp;
553 int offset;
554
555 /* Only possible if we received the final fragment. */
556 if (qp->len == 0)
557 return(0);
558
559 /* Check all fragment offsets to see if they connect. */
560 fp = qp->fragments;
561 offset = 0;
562 while (fp != NULL)
563 {
564 if (fp->offset > offset)
565 return(0); /* fragment(s) missing */
566 offset = fp->end;
567 fp = fp->next;
568 }
569
570 /* All fragments are present. */
571 return(1);
572 }
573
574
575 /*
576 * Build a new IP datagram from all its fragments.
577 *
578 * FIXME: We copy here because we lack an effective way of handling lists
579 * of bits on input. Until the new skb data handling is in I'm not going
580 * to touch this with a bargepole.
581 */
582
583 static struct sk_buff *ip_glue(struct ipq *qp)
/* */
584 {
585 struct sk_buff *skb;
586 struct iphdr *iph;
587 struct ipfrag *fp;
588 unsigned char *ptr;
589 int count, len;
590
591 /*
592 * Allocate a new buffer for the datagram.
593 */
594 len = qp->ihlen + qp->len;
595
596 if ((skb = dev_alloc_skb(len)) == NULL)
597 {
598 ip_statistics.IpReasmFails++;
599 NETDEBUG(printk("IP: queue_glue: no memory for gluing queue %p\n", qp));
600 ip_free(qp);
601 return(NULL);
602 }
603
604 /* Fill in the basic details. */
605 skb_put(skb,len);
606 skb->h.raw = skb->data;
607 skb->free = 1;
608
609 /* Copy the original IP headers into the new buffer. */
610 ptr = (unsigned char *) skb->h.raw;
611 memcpy(ptr, ((unsigned char *) qp->iph), qp->ihlen);
612 ptr += qp->ihlen;
613
614 count = 0;
615
616 /* Copy the data portions of all fragments into the new buffer. */
617 fp = qp->fragments;
618 while(fp != NULL)
619 {
620 if(count+fp->len > skb->len)
621 {
622 NETDEBUG(printk("Invalid fragment list: Fragment over size.\n"));
623 ip_free(qp);
624 kfree_skb(skb,FREE_WRITE);
625 ip_statistics.IpReasmFails++;
626 return NULL;
627 }
628 memcpy((ptr + fp->offset), fp->ptr, fp->len);
629 count += fp->len;
630 fp = fp->next;
631 }
632
633 /* We glued together all fragments, so remove the queue entry. */
634 ip_free(qp);
635
636 /* Done with all fragments. Fixup the new IP header. */
637 iph = skb->h.iph;
638 iph->frag_off = 0;
639 iph->tot_len = htons((iph->ihl * 4) + count);
640 skb->ip_hdr = iph;
641
642 ip_statistics.IpReasmOKs++;
643 return(skb);
644 }
645
646
647 /*
648 * Process an incoming IP datagram fragment.
649 */
650
651 static struct sk_buff *ip_defrag(struct iphdr *iph, struct sk_buff *skb, struct device *dev)
/* */
652 {
653 struct ipfrag *prev, *next, *tmp;
654 struct ipfrag *tfp;
655 struct ipq *qp;
656 struct sk_buff *skb2;
657 unsigned char *ptr;
658 int flags, offset;
659 int i, ihl, end;
660
661 ip_statistics.IpReasmReqds++;
662
663 /* Find the entry of this IP datagram in the "incomplete datagrams" queue. */
664 qp = ip_find(iph);
665
666 /* Is this a non-fragmented datagram? */
667 offset = ntohs(iph->frag_off);
668 flags = offset & ~IP_OFFSET;
669 offset &= IP_OFFSET;
670 if (((flags & IP_MF) == 0) && (offset == 0))
671 {
672 if (qp != NULL)
673 ip_free(qp); /* Huh? How could this exist?? */
674 return(skb);
675 }
676
677 offset <<= 3; /* offset is in 8-byte chunks */
678
679 /*
680 * If the queue already existed, keep restarting its timer as long
681 * as we still are receiving fragments. Otherwise, create a fresh
682 * queue entry.
683 */
684
685 if (qp != NULL)
686 {
687 del_timer(&qp->timer);
688 qp->timer.expires = jiffies + IP_FRAG_TIME; /* about 30 seconds */
689 qp->timer.data = (unsigned long) qp; /* pointer to queue */
690 qp->timer.function = ip_expire; /* expire function */
691 add_timer(&qp->timer);
692 }
693 else
694 {
695 /*
696 * If we failed to create it, then discard the frame
697 */
698 if ((qp = ip_create(skb, iph, dev)) == NULL)
699 {
700 skb->sk = NULL;
701 kfree_skb(skb, FREE_READ);
702 ip_statistics.IpReasmFails++;
703 return NULL;
704 }
705 }
706
707 /*
708 * Determine the position of this fragment.
709 */
710
711 ihl = iph->ihl * 4;
712 end = offset + ntohs(iph->tot_len) - ihl;
713
714 /*
715 * Point into the IP datagram 'data' part.
716 */
717
718 ptr = skb->data + ihl;
719
720 /*
721 * Is this the final fragment?
722 */
723
724 if ((flags & IP_MF) == 0)
725 qp->len = end;
726
727 /*
728 * Find out which fragments are in front and at the back of us
729 * in the chain of fragments so far. We must know where to put
730 * this fragment, right?
731 */
732
733 prev = NULL;
734 for(next = qp->fragments; next != NULL; next = next->next)
735 {
736 if (next->offset > offset)
737 break; /* bingo! */
738 prev = next;
739 }
740
741 /*
742 * We found where to put this one.
743 * Check for overlap with preceding fragment, and, if needed,
744 * align things so that any overlaps are eliminated.
745 */
746 if (prev != NULL && offset < prev->end)
747 {
748 i = prev->end - offset;
749 offset += i; /* ptr into datagram */
750 ptr += i; /* ptr into fragment data */
751 }
752
753 /*
754 * Look for overlap with succeeding segments.
755 * If we can merge fragments, do it.
756 */
757
758 for(tmp=next; tmp != NULL; tmp = tfp)
759 {
760 tfp = tmp->next;
761 if (tmp->offset >= end)
762 break; /* no overlaps at all */
763
764 i = end - next->offset; /* overlap is 'i' bytes */
765 tmp->len -= i; /* so reduce size of */
766 tmp->offset += i; /* next fragment */
767 tmp->ptr += i;
768 /*
769 * If we get a frag size of <= 0, remove it and the packet
770 * that it goes with.
771 */
772 if (tmp->len <= 0)
773 {
774 if (tmp->prev != NULL)
775 tmp->prev->next = tmp->next;
776 else
777 qp->fragments = tmp->next;
778
779 if (tfp->next != NULL)
780 tmp->next->prev = tmp->prev;
781
782 next=tfp; /* We have killed the original next frame */
783
784 kfree_skb(tmp->skb,FREE_READ);
785 kfree_s(tmp, sizeof(struct ipfrag));
786 }
787 }
788
789 /*
790 * Insert this fragment in the chain of fragments.
791 */
792
793 tfp = NULL;
794 tfp = ip_frag_create(offset, end, skb, ptr);
795
796 /*
797 * No memory to save the fragment - so throw the lot
798 */
799
800 if (!tfp)
801 {
802 skb->sk = NULL;
803 kfree_skb(skb, FREE_READ);
804 return NULL;
805 }
806 tfp->prev = prev;
807 tfp->next = next;
808 if (prev != NULL)
809 prev->next = tfp;
810 else
811 qp->fragments = tfp;
812
813 if (next != NULL)
814 next->prev = tfp;
815
816 /*
817 * OK, so we inserted this new fragment into the chain.
818 * Check if we now have a full IP datagram which we can
819 * bump up to the IP layer...
820 */
821
822 if (ip_done(qp))
823 {
824 skb2 = ip_glue(qp); /* glue together the fragments */
825 return(skb2);
826 }
827 return(NULL);
828 }
829
830
831 /*
832 * This IP datagram is too large to be sent in one piece. Break it up into
833 * smaller pieces (each of size equal to the MAC header plus IP header plus
834 * a block of the data of the original IP data part) that will yet fit in a
835 * single device frame, and queue such a frame for sending by calling the
836 * ip_queue_xmit(). Note that this is recursion, and bad things will happen
837 * if this function causes a loop...
838 *
839 * Yes this is inefficient, feel free to submit a quicker one.
840 *
841 * **Protocol Violation**
842 * We copy all the options to each fragment. !FIXME!
843 */
844
845 void ip_fragment(struct sock *sk, struct sk_buff *skb, struct device *dev, int is_frag)
/* */
846 {
847 struct iphdr *iph;
848 unsigned char *raw;
849 unsigned char *ptr;
850 struct sk_buff *skb2;
851 int left, mtu, hlen, len;
852 int offset;
853 unsigned long flags;
854
855 /*
856 * Point into the IP datagram header.
857 */
858
859 raw = skb->data;
860 #if 0
861 iph = (struct iphdr *) (raw + dev->hard_header_len);
862 skb->ip_hdr = iph;
863 #else
864 iph = skb->ip_hdr;
865 #endif
866
867 /*
868 * Setup starting values.
869 */
870
871 hlen = iph->ihl * 4;
872 left = ntohs(iph->tot_len) - hlen; /* Space per frame */
873 hlen += dev->hard_header_len; /* Total header size */
874 mtu = (dev->mtu - hlen); /* Size of data space */
875 ptr = (raw + hlen); /* Where to start from */
876
877 /*
878 * Check for any "DF" flag. [DF means do not fragment]
879 */
880
881 if (ntohs(iph->frag_off) & IP_DF)
882 {
883 /*
884 * Reply giving the MTU of the failed hop.
885 */
886 ip_statistics.IpFragFails++;
887 icmp_send(skb,ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, dev->mtu, dev);
888 return;
889 }
890
891 /*
892 * The protocol doesn't seem to say what to do in the case that the
893 * frame + options doesn't fit the mtu. As it used to fall down dead
894 * in this case we were fortunate it didn't happen
895 */
896
897 if(mtu<8)
898 {
899 /* It's wrong but it's better than nothing */
900 icmp_send(skb,ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED,dev->mtu, dev);
901 ip_statistics.IpFragFails++;
902 return;
903 }
904
905 /*
906 * Fragment the datagram.
907 */
908
909 /*
910 * The initial offset is 0 for a complete frame. When
911 * fragmenting fragments it's wherever this one starts.
912 */
913
914 if (is_frag & 2)
915 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
916 else
917 offset = 0;
918
919
920 /*
921 * Keep copying data until we run out.
922 */
923
924 while(left > 0)
925 {
926 len = left;
927 /* IF: it doesn't fit, use 'mtu' - the data space left */
928 if (len > mtu)
929 len = mtu;
930 /* IF: we are not sending upto and including the packet end
931 then align the next start on an eight byte boundary */
932 if (len < left)
933 {
934 len/=8;
935 len*=8;
936 }
937 /*
938 * Allocate buffer.
939 */
940
941 if ((skb2 = alloc_skb(len + hlen+15,GFP_ATOMIC)) == NULL)
942 {
943 NETDEBUG(printk("IP: frag: no memory for new fragment!\n"));
944 ip_statistics.IpFragFails++;
945 return;
946 }
947
948 /*
949 * Set up data on packet
950 */
951
952 skb2->arp = skb->arp;
953 if(skb->free==0)
954 printk("IP fragmenter: BUG free!=1 in fragmenter\n");
955 skb2->free = 1;
956 skb_put(skb2,len + hlen);
957 skb2->h.raw=(char *) skb2->data;
958 /*
959 * Charge the memory for the fragment to any owner
960 * it might possess
961 */
962
963 save_flags(flags);
964 if (sk)
965 {
966 cli();
967 sk->wmem_alloc += skb2->truesize;
968 skb2->sk=sk;
969 }
970 restore_flags(flags);
971 skb2->raddr = skb->raddr; /* For rebuild_header - must be here */
972
973 /*
974 * Copy the packet header into the new buffer.
975 */
976
977 memcpy(skb2->h.raw, raw, hlen);
978
979 /*
980 * Copy a block of the IP datagram.
981 */
982 memcpy(skb2->h.raw + hlen, ptr, len);
983 left -= len;
984
985 skb2->h.raw+=dev->hard_header_len;
986
987 /*
988 * Fill in the new header fields.
989 */
990 iph = (struct iphdr *)(skb2->h.raw/*+dev->hard_header_len*/);
991 iph->frag_off = htons((offset >> 3));
992 skb2->ip_hdr = iph;
993 /*
994 * Added AC : If we are fragmenting a fragment thats not the
995 * last fragment then keep MF on each bit
996 */
997 if (left > 0 || (is_frag & 1))
998 iph->frag_off |= htons(IP_MF);
999 ptr += len;
1000 offset += len;
1001
1002 /*
1003 * Put this fragment into the sending queue.
1004 */
1005
1006 ip_statistics.IpFragCreates++;
1007
1008 ip_queue_xmit(sk, dev, skb2, 2);
1009 }
1010 ip_statistics.IpFragOKs++;
1011 }
1012
1013
1014
1015 #ifdef CONFIG_IP_FORWARD
1016
1017 /*
1018 * Forward an IP datagram to its next destination.
1019 */
1020
1021 int ip_forward(struct sk_buff *skb, struct device *dev, int is_frag, unsigned long target_addr, int target_strict)
/* */
1022 {
1023 struct device *dev2; /* Output device */
1024 struct iphdr *iph; /* Our header */
1025 struct sk_buff *skb2; /* Output packet */
1026 struct rtable *rt; /* Route we use */
1027 unsigned char *ptr; /* Data pointer */
1028 unsigned long raddr; /* Router IP address */
1029 #ifdef CONFIG_IP_FIREWALL
1030 int fw_res = 0; /* Forwarding result */
1031
1032 /*
1033 * See if we are allowed to forward this.
1034 * Note: demasqueraded fragments are always 'back'warded.
1035 */
1036
1037
1038 if(!(is_frag&4))
1039 {
1040 fw_res=ip_fw_chk(skb->h.iph, dev, ip_fw_fwd_chain, ip_fw_fwd_policy, 0);
1041 switch (fw_res) {
1042 case 1:
1043 #ifdef CONFIG_IP_MASQUERADE
1044 case 2:
1045 #endif
1046 break;
1047 case -1:
1048 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, dev);
1049 /* fall thru */
1050 default:
1051 return -1;
1052 }
1053 }
1054 #endif
1055 /*
1056 * According to the RFC, we must first decrease the TTL field. If
1057 * that reaches zero, we must reply an ICMP control message telling
1058 * that the packet's lifetime expired.
1059 *
1060 * Exception:
1061 * We may not generate an ICMP for an ICMP. icmp_send does the
1062 * enforcement of this so we can forget it here. It is however
1063 * sometimes VERY important.
1064 */
1065
1066 iph = skb->h.iph;
1067 iph->ttl--;
1068
1069 /*
1070 * Re-compute the IP header checksum.
1071 * This is inefficient. We know what has happened to the header
1072 * and could thus adjust the checksum as Phil Karn does in KA9Q
1073 */
1074
1075 iph->check = ntohs(iph->check) + 0x0100;
1076 if ((iph->check & 0xFF00) == 0)
1077 iph->check++; /* carry overflow */
1078 iph->check = htons(iph->check);
1079
1080 if (iph->ttl <= 0)
1081 {
1082 /* Tell the sender its packet died... */
1083 icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0, dev);
1084 return -1;
1085 }
1086
1087 /*
1088 * OK, the packet is still valid. Fetch its destination address,
1089 * and give it to the IP sender for further processing.
1090 */
1091
1092 rt = ip_rt_route(target_addr, NULL, NULL);
1093 if (rt == NULL)
1094 {
1095 /*
1096 * Tell the sender its packet cannot be delivered. Again
1097 * ICMP is screened later.
1098 */
1099 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_NET_UNREACH, 0, dev);
1100 return -1;
1101 }
1102
1103
1104 /*
1105 * Gosh. Not only is the packet valid; we even know how to
1106 * forward it onto its final destination. Can we say this
1107 * is being plain lucky?
1108 * If the router told us that there is no GW, use the dest.
1109 * IP address itself- we seem to be connected directly...
1110 */
1111
1112 raddr = rt->rt_gateway;
1113
1114 if (raddr != 0)
1115 {
1116 /*
1117 * Strict routing permits no gatewaying
1118 */
1119
1120 if(target_strict)
1121 {
1122 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0, dev);
1123 return -1;
1124 }
1125
1126 /*
1127 * There is a gateway so find the correct route for it.
1128 * Gateways cannot in turn be gatewayed.
1129 */
1130
1131 rt = ip_rt_route(raddr, NULL, NULL);
1132 if (rt == NULL)
1133 {
1134 /*
1135 * Tell the sender its packet cannot be delivered...
1136 */
1137 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, dev);
1138 return -1;
1139 }
1140 if (rt->rt_gateway != 0)
1141 raddr = rt->rt_gateway;
1142 }
1143 else
1144 raddr = target_addr;
1145
1146 /*
1147 * Having picked a route we can now send the frame out.
1148 */
1149
1150 dev2 = rt->rt_dev;
1151
1152 /*
1153 * In IP you never have to forward a frame on the interface that it
1154 * arrived upon. We now generate an ICMP HOST REDIRECT giving the route
1155 * we calculated.
1156 */
1157 #ifndef CONFIG_IP_NO_ICMP_REDIRECT
1158 if (dev == dev2 && !((iph->saddr^iph->daddr)&dev->pa_mask) && (rt->rt_flags&RTF_MODIFIED))
1159 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, raddr, dev);
1160 #endif
1161
1162 /*
1163 * We now may allocate a new buffer, and copy the datagram into it.
1164 * If the indicated interface is up and running, kick it.
1165 */
1166
1167 if (dev2->flags & IFF_UP)
1168 {
1169 #ifdef CONFIG_IP_MASQUERADE
1170 /*
1171 * If this fragment needs masquerading, make it so...
1172 * (Dont masquerade de-masqueraded fragments)
1173 */
1174 if (!(is_frag&4) && fw_res==2)
1175 ip_fw_masquerade(&skb, dev2);
1176 #endif
1177 IS_SKB(skb);
1178
1179 if(skb_headroom(skb)<dev2->hard_header_len)
1180 {
1181 skb2 = alloc_skb(dev2->hard_header_len + skb->len + 15, GFP_ATOMIC);
1182 IS_SKB(skb2);
1183
1184 /*
1185 * This is rare and since IP is tolerant of network failures
1186 * quite harmless.
1187 */
1188
1189 if (skb2 == NULL)
1190 {
1191 NETDEBUG(printk("\nIP: No memory available for IP forward\n"));
1192 return -1;
1193 }
1194
1195 /*
1196 * Add the physical headers.
1197 */
1198
1199 ip_send(skb2,raddr,skb->len,dev2,dev2->pa_addr);
1200
1201 /*
1202 * We have to copy the bytes over as the new header wouldn't fit
1203 * the old buffer. This should be very rare.
1204 */
1205
1206 ptr = skb_put(skb2,skb->len);
1207 skb2->free = 1;
1208 skb2->h.raw = ptr;
1209
1210 /*
1211 * Copy the packet data into the new buffer.
1212 */
1213 memcpy(ptr, skb->h.raw, skb->len);
1214 }
1215 else
1216 {
1217 /*
1218 * Build a new MAC header.
1219 */
1220
1221 skb2 = skb;
1222 skb2->dev=dev2;
1223 skb->arp=1;
1224 skb->raddr=raddr;
1225 if(dev2->hard_header)
1226 {
1227 if(dev2->hard_header(skb, dev2, ETH_P_IP, NULL, NULL, skb->len)<0)
1228 skb->arp=0;
1229 }
1230 ip_statistics.IpForwDatagrams++;
1231 }
1232 /*
1233 * See if it needs fragmenting. Note in ip_rcv we tagged
1234 * the fragment type. This must be right so that
1235 * the fragmenter does the right thing.
1236 */
1237
1238 if(skb2->len > dev2->mtu + dev2->hard_header_len)
1239 {
1240 ip_fragment(NULL,skb2,dev2, is_frag);
1241 kfree_skb(skb2,FREE_WRITE);
1242 }
1243 else
1244 {
1245 #ifdef CONFIG_IP_ACCT
1246 /*
1247 * Count mapping we shortcut
1248 */
1249
1250 ip_fw_chk(iph,dev,ip_acct_chain,IP_FW_F_ACCEPT,1);
1251 #endif
1252
1253 /*
1254 * Map service types to priority. We lie about
1255 * throughput being low priority, but it's a good
1256 * choice to help improve general usage.
1257 */
1258 if(iph->tos & IPTOS_LOWDELAY)
1259 dev_queue_xmit(skb2, dev2, SOPRI_INTERACTIVE);
1260 else if(iph->tos & IPTOS_THROUGHPUT)
1261 dev_queue_xmit(skb2, dev2, SOPRI_BACKGROUND);
1262 else
1263 dev_queue_xmit(skb2, dev2, SOPRI_NORMAL);
1264 }
1265 }
1266 else
1267 return -1;
1268
1269 /*
1270 * Tell the caller if their buffer is free.
1271 */
1272
1273 if(skb==skb2)
1274 return 0;
1275 return 1;
1276 }
1277
1278
1279 #endif
1280
1281 /*
1282 * This function receives all incoming IP datagrams.
1283 *
1284 * On entry skb->data points to the start of the IP header and
1285 * the MAC header has been removed.
1286 */
1287
1288 int ip_rcv(struct sk_buff *skb, struct device *dev, struct packet_type *pt)
/* */
1289 {
1290 struct iphdr *iph = skb->h.iph;
1291 struct sock *raw_sk=NULL;
1292 unsigned char hash;
1293 unsigned char flag = 0;
1294 struct inet_protocol *ipprot;
1295 int brd=IS_MYADDR;
1296 unsigned long target_addr;
1297 int target_strict=0;
1298 int is_frag=0;
1299 #ifdef CONFIG_IP_FIREWALL
1300 int err;
1301 #endif
1302
1303 ip_statistics.IpInReceives++;
1304
1305 /*
1306 * Tag the ip header of this packet so we can find it
1307 */
1308
1309 skb->ip_hdr = iph;
1310
1311 /*
1312 * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum.
1313 * RFC1122: 3.1.2.3 MUST discard a frame with invalid source address [NEEDS FIXING].
1314 *
1315 * Is the datagram acceptable?
1316 *
1317 * 1. Length at least the size of an ip header
1318 * 2. Version of 4
1319 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
1320 * 4. Doesn't have a bogus length
1321 * (5. We ought to check for IP multicast addresses and undefined types.. does this matter ?)
1322 */
1323
1324 if (skb->len<sizeof(struct iphdr) || iph->ihl<5 || iph->version != 4 || ip_fast_csum((unsigned char *)iph, iph->ihl) !=0
1325 || skb->len < ntohs(iph->tot_len))
1326 {
1327 ip_statistics.IpInHdrErrors++;
1328 kfree_skb(skb, FREE_WRITE);
1329 return(0);
1330 }
1331
1332 /*
1333 * Our transport medium may have padded the buffer out. Now we know it
1334 * is IP we can trim to the true length of the frame.
1335 * Note this now means skb->len holds ntohs(iph->tot_len).
1336 */
1337
1338 skb_trim(skb,ntohs(iph->tot_len));
1339
1340 /*
1341 * See if the firewall wants to dispose of the packet.
1342 */
1343
1344 #ifdef CONFIG_IP_FIREWALL
1345
1346 if ((err=ip_fw_chk(iph,dev,ip_fw_blk_chain,ip_fw_blk_policy, 0))<1)
1347 {
1348 if(err==-1)
1349 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0, dev);
1350 kfree_skb(skb, FREE_WRITE);
1351 return 0;
1352 }
1353
1354 #endif
1355
1356
1357 /*
1358 * Next analyse the packet for options. Studies show under one packet in
1359 * a thousand have options....
1360 */
1361
1362 target_addr = iph->daddr;
1363
1364 if (iph->ihl != 5)
1365 {
1366 /* Humph.. options. Lots of annoying fiddly bits */
1367
1368 /*
1369 * This is straight from the RFC. It might even be right ;)
1370 *
1371 * RFC 1122: 3.2.1.8 STREAMID option is obsolete and MUST be ignored.
1372 * RFC 1122: 3.2.1.8 MUST NOT crash on a zero length option.
1373 * RFC 1122: 3.2.1.8 MUST support acting as final destination of a source route.
1374 */
1375
1376 int opt_space=4*(iph->ihl-5);
1377 int opt_size;
1378 unsigned char *opt_ptr=skb->h.raw+sizeof(struct iphdr);
1379
1380 skb->ip_summed=0; /* Our free checksum is bogus for this case */
1381
1382 while(opt_space>0)
1383 {
1384 if(*opt_ptr==IPOPT_NOOP)
1385 {
1386 opt_ptr++;
1387 opt_space--;
1388 continue;
1389 }
1390 if(*opt_ptr==IPOPT_END)
1391 break; /* Done */
1392 if(opt_space<2 || (opt_size=opt_ptr[1])<2 || opt_ptr[1]>opt_space)
1393 {
1394 /*
1395 * RFC 1122: 3.2.2.5 SHOULD send parameter problem reports.
1396 */
1397 icmp_send(skb, ICMP_PARAMETERPROB, 0, 0, skb->dev);
1398 kfree_skb(skb, FREE_READ);
1399 return -EINVAL;
1400 }
1401 switch(opt_ptr[0])
1402 {
1403 case IPOPT_SEC:
1404 /* Should we drop this ?? */
1405 break;
1406 case IPOPT_SSRR: /* These work almost the same way */
1407 target_strict=1;
1408 /* Fall through */
1409 case IPOPT_LSRR:
1410 #ifdef CONFIG_IP_NOSR
1411 kfree_skb(skb, FREE_READ);
1412 return -EINVAL;
1413 #endif
1414 case IPOPT_RR:
1415 /*
1416 * RFC 1122: 3.2.1.8 Support for RR is OPTIONAL.
1417 */
1418 if (iph->daddr!=skb->dev->pa_addr && (brd = ip_chk_addr(iph->daddr)) == 0)
1419 break;
1420 if((opt_size<3) || ( opt_ptr[0]==IPOPT_RR && opt_ptr[2] > opt_size-4 ))
1421 {
1422 if(ip_chk_addr(iph->daddr))
1423 icmp_send(skb, ICMP_PARAMETERPROB, 0, 0, skb->dev);
1424 kfree_skb(skb, FREE_READ);
1425 return -EINVAL;
1426 }
1427 if(opt_ptr[2] > opt_size-4 )
1428 break;
1429 /* Bytes are [IPOPT_xxRR][Length][EntryPointer][Entry0][Entry1].... */
1430 /* This isn't going to be too portable - FIXME */
1431 if(opt_ptr[0]!=IPOPT_RR)
1432 {
1433 int t;
1434 target_addr=*(u32 *)(&opt_ptr[opt_ptr[2]]); /* Get hop */
1435 t=ip_chk_addr(target_addr);
1436 if(t==IS_MULTICAST||t==IS_BROADCAST)
1437 {
1438 if(ip_chk_addr(iph->daddr))
1439 icmp_send(skb, ICMP_PARAMETERPROB, 0, 0, skb->dev);
1440 kfree_skb(skb,FREE_READ);
1441 return -EINVAL;
1442 }
1443 }
1444 *(u32 *)(&opt_ptr[opt_ptr[2]])=skb->dev->pa_addr; /* Record hop */
1445 break;
1446 case IPOPT_TIMESTAMP:
1447 /*
1448 * RFC 1122: 3.2.1.8 The timestamp option is OPTIONAL but if implemented
1449 * MUST meet various rules (read the spec).
1450 */
1451 NETDEBUG(printk("ICMP: Someone finish the timestamp routine ;)\n"));
1452 break;
1453 default:
1454 break;
1455 }
1456 opt_ptr+=opt_size;
1457 opt_space-=opt_size;
1458 }
1459
1460 }
1461
1462
1463 /*
1464 * Remember if the frame is fragmented.
1465 */
1466
1467 if(iph->frag_off)
1468 {
1469 if (iph->frag_off & htons(IP_MF))
1470 is_frag|=1;
1471 /*
1472 * Last fragment ?
1473 */
1474
1475 if (iph->frag_off & htons(IP_OFFSET))
1476 is_frag|=2;
1477 }
1478
1479 /*
1480 * Do any IP forwarding required. chk_addr() is expensive -- avoid it someday.
1481 *
1482 * This is inefficient. While finding out if it is for us we could also compute
1483 * the routing table entry. This is where the great unified cache theory comes
1484 * in as and when someone implements it
1485 *
1486 * For most hosts over 99% of packets match the first conditional
1487 * and don't go via ip_chk_addr. Note: brd is set to IS_MYADDR at
1488 * function entry.
1489 */
1490
1491 if ( iph->daddr == skb->dev->pa_addr || (brd = ip_chk_addr(iph->daddr)) != 0)
1492 {
1493 #ifdef CONFIG_IP_MULTICAST
1494
1495 if(brd==IS_MULTICAST && iph->daddr!=IGMP_ALL_HOSTS && !(dev->flags&IFF_LOOPBACK))
1496 {
1497 /*
1498 * Check it is for one of our groups
1499 */
1500 struct ip_mc_list *ip_mc=dev->ip_mc_list;
1501 do
1502 {
1503 if(ip_mc==NULL)
1504 {
1505 kfree_skb(skb, FREE_WRITE);
1506 return 0;
1507 }
1508 if(ip_mc->multiaddr==iph->daddr)
1509 break;
1510 ip_mc=ip_mc->next;
1511 }
1512 while(1);
1513 }
1514 #endif
1515
1516 #ifdef CONFIG_IP_MASQUERADE
1517 /*
1518 * Do we need to de-masquerade this fragment?
1519 */
1520 if (ip_fw_demasquerade(skb))
1521 {
1522 struct iphdr *iph=skb->h.iph;
1523 if(ip_forward(skb, dev, is_frag|4, iph->daddr, 0))
1524 kfree_skb(skb, FREE_WRITE);
1525 return(0);
1526 }
1527 #endif
1528
1529 /*
1530 * Account for the packet
1531 */
1532
1533 #ifdef CONFIG_IP_ACCT
1534 ip_fw_chk(iph,dev,ip_acct_chain,IP_FW_F_ACCEPT,1);
1535 #endif
1536
1537 /*
1538 * Reassemble IP fragments.
1539 */
1540
1541 if(is_frag)
1542 {
1543 /* Defragment. Obtain the complete packet if there is one */
1544 skb=ip_defrag(iph,skb,dev);
1545 if(skb==NULL)
1546 return 0;
1547 skb->dev = dev;
1548 iph=skb->h.iph;
1549 }
1550
1551 /*
1552 * Point into the IP datagram, just past the header.
1553 */
1554
1555 skb->ip_hdr = iph;
1556 skb->h.raw += iph->ihl*4;
1557
1558 /*
1559 * Deliver to raw sockets. This is fun as to avoid copies we want to make no surplus copies.
1560 *
1561 * RFC 1122: SHOULD pass TOS value up to the transport layer.
1562 */
1563
1564 hash = iph->protocol & (SOCK_ARRAY_SIZE-1);
1565
1566 /*
1567 * If there maybe a raw socket we must check - if not we don't care less
1568 */
1569
1570 if((raw_sk=raw_prot.sock_array[hash])!=NULL)
1571 {
1572 struct sock *sknext=NULL;
1573 struct sk_buff *skb1;
1574 raw_sk=get_sock_raw(raw_sk, hash, iph->saddr, iph->daddr);
1575 if(raw_sk) /* Any raw sockets */
1576 {
1577 do
1578 {
1579 /* Find the next */
1580 sknext=get_sock_raw(raw_sk->next, hash, iph->saddr, iph->daddr);
1581 if(sknext)
1582 skb1=skb_clone(skb, GFP_ATOMIC);
1583 else
1584 break; /* One pending raw socket left */
1585 if(skb1)
1586 raw_rcv(raw_sk, skb1, dev, iph->saddr,iph->daddr);
1587 raw_sk=sknext;
1588 }
1589 while(raw_sk!=NULL);
1590
1591 /*
1592 * Here either raw_sk is the last raw socket, or NULL if none
1593 */
1594
1595 /*
1596 * We deliver to the last raw socket AFTER the protocol checks as it avoids a surplus copy
1597 */
1598 }
1599 }
1600
1601 /*
1602 * skb->h.raw now points at the protocol beyond the IP header.
1603 */
1604
1605 hash = iph->protocol & (MAX_INET_PROTOS -1);
1606 for (ipprot = (struct inet_protocol *)inet_protos[hash];ipprot != NULL;ipprot=(struct inet_protocol *)ipprot->next)
1607 {
1608 struct sk_buff *skb2;
1609
1610 if (ipprot->protocol != iph->protocol)
1611 continue;
1612 /*
1613 * See if we need to make a copy of it. This will
1614 * only be set if more than one protocol wants it.
1615 * and then not for the last one. If there is a pending
1616 * raw delivery wait for that
1617 */
1618
1619 if (ipprot->copy || raw_sk)
1620 {
1621 skb2 = skb_clone(skb, GFP_ATOMIC);
1622 if(skb2==NULL)
1623 continue;
1624 }
1625 else
1626 {
1627 skb2 = skb;
1628 }
1629 flag = 1;
1630
1631 /*
1632 * Pass on the datagram to each protocol that wants it,
1633 * based on the datagram protocol. We should really
1634 * check the protocol handler's return values here...
1635 */
1636
1637 ipprot->handler(skb2, dev, NULL, iph->daddr,
1638 (ntohs(iph->tot_len) - (iph->ihl * 4)),
1639 iph->saddr, 0, ipprot);
1640
1641 }
1642
1643 /*
1644 * All protocols checked.
1645 * If this packet was a broadcast, we may *not* reply to it, since that
1646 * causes (proven, grin) ARP storms and a leakage of memory (i.e. all
1647 * ICMP reply messages get queued up for transmission...)
1648 */
1649
1650 if(raw_sk!=NULL) /* Shift to last raw user */
1651 raw_rcv(raw_sk, skb, dev, iph->saddr, iph->daddr);
1652 else if (!flag) /* Free and report errors */
1653 {
1654 if (brd != IS_BROADCAST && brd!=IS_MULTICAST)
1655 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, 0, dev);
1656 kfree_skb(skb, FREE_WRITE);
1657 }
1658
1659 return(0);
1660 }
1661
1662 /*
1663 * Do any IP forwarding required.
1664 */
1665
1666 /*
1667 * Don't forward multicast or broadcast frames.
1668 */
1669
1670 if(skb->pkt_type!=PACKET_HOST || brd==IS_BROADCAST)
1671 {
1672 kfree_skb(skb,FREE_WRITE);
1673 return 0;
1674 }
1675
1676 /*
1677 * The packet is for another target. Forward the frame
1678 */
1679
1680 #ifdef CONFIG_IP_FORWARD
1681 if(ip_forward(skb, dev, is_frag, target_addr, target_strict))
1682 kfree_skb(skb, FREE_WRITE);
1683 #else
1684 /* printk("Machine %lx tried to use us as a forwarder to %lx but we have forwarding disabled!\n",
1685 iph->saddr,iph->daddr);*/
1686 ip_statistics.IpInAddrErrors++;
1687 kfree_skb(skb, FREE_WRITE);
1688 #endif
1689 return(0);
1690 }
1691
1692
1693 /*
1694 * Loop a packet back to the sender.
1695 */
1696
1697 static void ip_loopback(struct device *old_dev, struct sk_buff *skb)
/* */
1698 {
1699 struct device *dev=&loopback_dev;
1700 int len=skb->len-old_dev->hard_header_len;
1701 struct sk_buff *newskb=dev_alloc_skb(len+dev->hard_header_len+15);
1702
1703 if(newskb==NULL)
1704 return;
1705
1706 newskb->link3=NULL;
1707 newskb->sk=NULL;
1708 newskb->dev=dev;
1709 newskb->saddr=skb->saddr;
1710 newskb->daddr=skb->daddr;
1711 newskb->raddr=skb->raddr;
1712 newskb->free=1;
1713 newskb->lock=0;
1714 newskb->users=0;
1715 newskb->pkt_type=skb->pkt_type;
1716
1717 /*
1718 * Put a MAC header on the packet
1719 */
1720 ip_send(newskb, skb->ip_hdr->daddr, len, dev, skb->ip_hdr->saddr);
1721 /*
1722 * Add the rest of the data space.
1723 */
1724 newskb->ip_hdr=(struct iphdr *)skb_put(newskb, len);
1725 /*
1726 * Copy the data
1727 */
1728 memcpy(newskb->ip_hdr,skb->ip_hdr,len);
1729
1730 /* Recurse. The device check against IFF_LOOPBACK will stop infinite recursion */
1731
1732 /*printk("Loopback output queued [%lX to %lX].\n", newskb->ip_hdr->saddr,newskb->ip_hdr->daddr);*/
1733 ip_queue_xmit(NULL, dev, newskb, 1);
1734 }
1735
1736
1737 /*
1738 * Queues a packet to be sent, and starts the transmitter
1739 * if necessary. if free = 1 then we free the block after
1740 * transmit, otherwise we don't. If free==2 we not only
1741 * free the block but also don't assign a new ip seq number.
1742 * This routine also needs to put in the total length,
1743 * and compute the checksum
1744 */
1745
1746 void ip_queue_xmit(struct sock *sk, struct device *dev,
/* */
1747 struct sk_buff *skb, int free)
1748 {
1749 struct iphdr *iph;
1750 /* unsigned char *ptr;*/
1751
1752 /* Sanity check */
1753 if (dev == NULL)
1754 {
1755 NETDEBUG(printk("IP: ip_queue_xmit dev = NULL\n"));
1756 return;
1757 }
1758
1759 IS_SKB(skb);
1760
1761 /*
1762 * Do some book-keeping in the packet for later
1763 */
1764
1765
1766 skb->dev = dev;
1767 skb->when = jiffies;
1768
1769 /*
1770 * Find the IP header and set the length. This is bad
1771 * but once we get the skb data handling code in the
1772 * hardware will push its header sensibly and we will
1773 * set skb->ip_hdr to avoid this mess and the fixed
1774 * header length problem
1775 */
1776
1777 #if 0
1778 ptr = skb->data;
1779 ptr += dev->hard_header_len;
1780 iph = (struct iphdr *)ptr;
1781 skb->ip_hdr = iph;
1782 #else
1783 iph = skb->ip_hdr;
1784 #endif
1785 iph->tot_len = ntohs(skb->len-(((unsigned char *)iph)-skb->data));
1786
1787 #ifdef CONFIG_IP_FIREWALL
1788 if(ip_fw_chk(iph, dev, ip_fw_blk_chain, ip_fw_blk_policy, 0) != 1)
1789 /* just don't send this packet */
1790 return;
1791 #endif
1792
1793 /*
1794 * No reassigning numbers to fragments...
1795 */
1796
1797 if(free!=2)
1798 iph->id = htons(ip_id_count++);
1799 else
1800 free=1;
1801
1802 /* All buffers without an owner socket get freed */
1803 if (sk == NULL)
1804 free = 1;
1805
1806 skb->free = free;
1807
1808 /*
1809 * Do we need to fragment. Again this is inefficient.
1810 * We need to somehow lock the original buffer and use
1811 * bits of it.
1812 */
1813
1814 if(ntohs(iph->tot_len)> dev->mtu)
1815 {
1816 ip_fragment(sk,skb,dev,0);
1817 IS_SKB(skb);
1818 kfree_skb(skb,FREE_WRITE);
1819 return;
1820 }
1821
1822 /*
1823 * Add an IP checksum
1824 */
1825
1826 ip_send_check(iph);
1827
1828 /*
1829 * Print the frame when debugging
1830 */
1831
1832 /*
1833 * More debugging. You cannot queue a packet already on a list
1834 * Spot this and moan loudly.
1835 */
1836 if (skb->next != NULL)
1837 {
1838 NETDEBUG(printk("ip_queue_xmit: next != NULL\n"));
1839 skb_unlink(skb);
1840 }
1841
1842 /*
1843 * If a sender wishes the packet to remain unfreed
1844 * we add it to his send queue. This arguably belongs
1845 * in the TCP level since nobody else uses it. BUT
1846 * remember IPng might change all the rules.
1847 */
1848
1849 if (!free)
1850 {
1851 unsigned long flags;
1852 /* The socket now has more outstanding blocks */
1853
1854 sk->packets_out++;
1855
1856 /* Protect the list for a moment */
1857 save_flags(flags);
1858 cli();
1859
1860 if (skb->link3 != NULL)
1861 {
1862 NETDEBUG(printk("ip.c: link3 != NULL\n"));
1863 skb->link3 = NULL;
1864 }
1865 if (sk->send_head == NULL)
1866 {
1867 sk->send_tail = skb;
1868 sk->send_head = skb;
1869 }
1870 else
1871 {
1872 sk->send_tail->link3 = skb;
1873 sk->send_tail = skb;
1874 }
1875 /* skb->link3 is NULL */
1876
1877 /* Interrupt restore */
1878 restore_flags(flags);
1879 }
1880 else
1881 /* Remember who owns the buffer */
1882 skb->sk = sk;
1883
1884 /*
1885 * If the indicated interface is up and running, send the packet.
1886 */
1887
1888 ip_statistics.IpOutRequests++;
1889 #ifdef CONFIG_IP_ACCT
1890 ip_fw_chk(iph,dev,ip_acct_chain,IP_FW_F_ACCEPT,1);
1891 #endif
1892
1893 #ifdef CONFIG_IP_MULTICAST
1894
1895 /*
1896 * Multicasts are looped back for other local users
1897 */
1898
1899 if (MULTICAST(iph->daddr) && !(dev->flags&IFF_LOOPBACK))
1900 {
1901 if(sk==NULL || sk->ip_mc_loop)
1902 {
1903 if(iph->daddr==IGMP_ALL_HOSTS)
1904 ip_loopback(dev,skb);
1905 else
1906 {
1907 struct ip_mc_list *imc=dev->ip_mc_list;
1908 while(imc!=NULL)
1909 {
1910 if(imc->multiaddr==iph->daddr)
1911 {
1912 ip_loopback(dev,skb);
1913 break;
1914 }
1915 imc=imc->next;
1916 }
1917 }
1918 }
1919 /* Multicasts with ttl 0 must not go beyond the host */
1920
1921 if(skb->ip_hdr->ttl==0)
1922 {
1923 kfree_skb(skb, FREE_READ);
1924 return;
1925 }
1926 }
1927 #endif
1928 if((dev->flags&IFF_BROADCAST) && (iph->daddr==dev->pa_brdaddr||iph->daddr==0xFFFFFFFF) && !(dev->flags&IFF_LOOPBACK))
1929 ip_loopback(dev,skb);
1930
1931 if (dev->flags & IFF_UP)
1932 {
1933 /*
1934 * If we have an owner use its priority setting,
1935 * otherwise use NORMAL
1936 */
1937
1938 if (sk != NULL)
1939 {
1940 dev_queue_xmit(skb, dev, sk->priority);
1941 }
1942 else
1943 {
1944 dev_queue_xmit(skb, dev, SOPRI_NORMAL);
1945 }
1946 }
1947 else
1948 {
1949 ip_statistics.IpOutDiscards++;
1950 if (free)
1951 kfree_skb(skb, FREE_WRITE);
1952 }
1953 }
1954
1955
1956
1957 #ifdef CONFIG_IP_MULTICAST
1958
1959 /*
1960 * Write an multicast group list table for the IGMP daemon to
1961 * read.
1962 */
1963
1964 int ip_mc_procinfo(char *buffer, char **start, off_t offset, int length, int dummy)
/* */
1965 {
1966 off_t pos=0, begin=0;
1967 struct ip_mc_list *im;
1968 unsigned long flags;
1969 int len=0;
1970 struct device *dev;
1971
1972 len=sprintf(buffer,"Device : Count\tGroup Users Timer\n");
1973 save_flags(flags);
1974 cli();
1975
1976 for(dev = dev_base; dev; dev = dev->next)
1977 {
1978 if((dev->flags&IFF_UP)&&(dev->flags&IFF_MULTICAST))
1979 {
1980 len+=sprintf(buffer+len,"%-10s: %5d\n",
1981 dev->name, dev->mc_count);
1982 for(im = dev->ip_mc_list; im; im = im->next)
1983 {
1984 len+=sprintf(buffer+len,
1985 "\t\t\t%08lX %5d %d:%08lX\n",
1986 im->multiaddr, im->users,
1987 im->tm_running, im->timer.expires-jiffies);
1988 pos=begin+len;
1989 if(pos<offset)
1990 {
1991 len=0;
1992 begin=pos;
1993 }
1994 if(pos>offset+length)
1995 break;
1996 }
1997 }
1998 }
1999 restore_flags(flags);
2000 *start=buffer+(offset-begin);
2001 len-=(offset-begin);
2002 if(len>length)
2003 len=length;
2004 return len;
2005 }
2006
2007
2008 /*
2009 * Socket option code for IP. This is the end of the line after any TCP,UDP etc options on
2010 * an IP socket.
2011 *
2012 * We implement IP_TOS (type of service), IP_TTL (time to live).
2013 *
2014 * Next release we will sort out IP_OPTIONS since for some people are kind of important.
2015 */
2016
2017 static struct device *ip_mc_find_devfor(unsigned long addr)
/* */
2018 {
2019 struct device *dev;
2020 for(dev = dev_base; dev; dev = dev->next)
2021 {
2022 if((dev->flags&IFF_UP)&&(dev->flags&IFF_MULTICAST)&&
2023 (dev->pa_addr==addr))
2024 return dev;
2025 }
2026
2027 return NULL;
2028 }
2029
2030 #endif
2031
2032 int ip_setsockopt(struct sock *sk, int level, int optname, char *optval, int optlen)
/* */
2033 {
2034 int val,err;
2035 unsigned char ucval;
2036 #if defined(CONFIG_IP_FIREWALL) || defined(CONFIG_IP_ACCT)
2037 struct ip_fw tmp_fw;
2038 #endif
2039 if (optval == NULL)
2040 return(-EINVAL);
2041
2042 err=verify_area(VERIFY_READ, optval, sizeof(int));
2043 if(err)
2044 return err;
2045
2046 val = get_user((int *) optval);
2047 ucval=get_user((unsigned char *) optval);
2048
2049 if(level!=SOL_IP)
2050 return -EOPNOTSUPP;
2051
2052 switch(optname)
2053 {
2054 case IP_TOS:
2055 if(val<0||val>255)
2056 return -EINVAL;
2057 sk->ip_tos=val;
2058 if(val==IPTOS_LOWDELAY)
2059 sk->priority=SOPRI_INTERACTIVE;
2060 if(val==IPTOS_THROUGHPUT)
2061 sk->priority=SOPRI_BACKGROUND;
2062 return 0;
2063 case IP_TTL:
2064 if(val<1||val>255)
2065 return -EINVAL;
2066 sk->ip_ttl=val;
2067 return 0;
2068 #ifdef CONFIG_IP_MULTICAST
2069 case IP_MULTICAST_TTL:
2070 {
2071 sk->ip_mc_ttl=(int)ucval;
2072 return 0;
2073 }
2074 case IP_MULTICAST_LOOP:
2075 {
2076 if(ucval!=0 && ucval!=1)
2077 return -EINVAL;
2078 sk->ip_mc_loop=(int)ucval;
2079 return 0;
2080 }
2081 case IP_MULTICAST_IF:
2082 {
2083 struct in_addr addr;
2084 struct device *dev=NULL;
2085
2086 /*
2087 * Check the arguments are allowable
2088 */
2089
2090 err=verify_area(VERIFY_READ, optval, sizeof(addr));
2091 if(err)
2092 return err;
2093
2094 memcpy_fromfs(&addr,optval,sizeof(addr));
2095
2096
2097 /*
2098 * What address has been requested
2099 */
2100
2101 if(addr.s_addr==INADDR_ANY) /* Default */
2102 {
2103 sk->ip_mc_name[0]=0;
2104 return 0;
2105 }
2106
2107 /*
2108 * Find the device
2109 */
2110
2111 dev=ip_mc_find_devfor(addr.s_addr);
2112
2113 /*
2114 * Did we find one
2115 */
2116
2117 if(dev)
2118 {
2119 strcpy(sk->ip_mc_name,dev->name);
2120 return 0;
2121 }
2122 return -EADDRNOTAVAIL;
2123 }
2124
2125 case IP_ADD_MEMBERSHIP:
2126 {
2127
2128 /*
2129 * FIXME: Add/Del membership should have a semaphore protecting them from re-entry
2130 */
2131 struct ip_mreq mreq;
2132 unsigned long route_src;
2133 struct rtable *rt;
2134 struct device *dev=NULL;
2135
2136 /*
2137 * Check the arguments.
2138 */
2139
2140 err=verify_area(VERIFY_READ, optval, sizeof(mreq));
2141 if(err)
2142 return err;
2143
2144 memcpy_fromfs(&mreq,optval,sizeof(mreq));
2145
2146 /*
2147 * Get device for use later
2148 */
2149
2150 if(mreq.imr_interface.s_addr==INADDR_ANY)
2151 {
2152 /*
2153 * Not set so scan.
2154 */
2155 if((rt=ip_rt_route(mreq.imr_multiaddr.s_addr,NULL, &route_src))!=NULL)
2156 {
2157 dev=rt->rt_dev;
2158 rt->rt_use--;
2159 }
2160 }
2161 else
2162 {
2163 /*
2164 * Find a suitable device.
2165 */
2166
2167 dev=ip_mc_find_devfor(mreq.imr_interface.s_addr);
2168 }
2169
2170 /*
2171 * No device, no cookies.
2172 */
2173
2174 if(!dev)
2175 return -ENODEV;
2176
2177 /*
2178 * Join group.
2179 */
2180
2181 return ip_mc_join_group(sk,dev,mreq.imr_multiaddr.s_addr);
2182 }
2183
2184 case IP_DROP_MEMBERSHIP:
2185 {
2186 struct ip_mreq mreq;
2187 struct rtable *rt;
2188 unsigned long route_src;
2189 struct device *dev=NULL;
2190
2191 /*
2192 * Check the arguments
2193 */
2194
2195 err=verify_area(VERIFY_READ, optval, sizeof(mreq));
2196 if(err)
2197 return err;
2198
2199 memcpy_fromfs(&mreq,optval,sizeof(mreq));
2200
2201 /*
2202 * Get device for use later
2203 */
2204
2205 if(mreq.imr_interface.s_addr==INADDR_ANY)
2206 {
2207 if((rt=ip_rt_route(mreq.imr_multiaddr.s_addr,NULL, &route_src))!=NULL)
2208 {
2209 dev=rt->rt_dev;
2210 rt->rt_use--;
2211 }
2212 }
2213 else
2214 {
2215
2216 dev=ip_mc_find_devfor(mreq.imr_interface.s_addr);
2217 }
2218
2219 /*
2220 * Did we find a suitable device.
2221 */
2222
2223 if(!dev)
2224 return -ENODEV;
2225
2226 /*
2227 * Leave group
2228 */
2229
2230 return ip_mc_leave_group(sk,dev,mreq.imr_multiaddr.s_addr);
2231 }
2232 #endif
2233 #ifdef CONFIG_IP_FIREWALL
2234 case IP_FW_ADD_BLK:
2235 case IP_FW_DEL_BLK:
2236 case IP_FW_ADD_FWD:
2237 case IP_FW_DEL_FWD:
2238 case IP_FW_CHK_BLK:
2239 case IP_FW_CHK_FWD:
2240 case IP_FW_FLUSH_BLK:
2241 case IP_FW_FLUSH_FWD:
2242 case IP_FW_ZERO_BLK:
2243 case IP_FW_ZERO_FWD:
2244 case IP_FW_POLICY_BLK:
2245 case IP_FW_POLICY_FWD:
2246 if(!suser())
2247 return -EPERM;
2248 if(optlen>sizeof(tmp_fw) || optlen<1)
2249 return -EINVAL;
2250 err=verify_area(VERIFY_READ,optval,optlen);
2251 if(err)
2252 return err;
2253 memcpy_fromfs(&tmp_fw,optval,optlen);
2254 err=ip_fw_ctl(optname, &tmp_fw,optlen);
2255 return -err; /* -0 is 0 after all */
2256
2257 #endif
2258 #ifdef CONFIG_IP_ACCT
2259 case IP_ACCT_DEL:
2260 case IP_ACCT_ADD:
2261 case IP_ACCT_FLUSH:
2262 case IP_ACCT_ZERO:
2263 if(!suser())
2264 return -EPERM;
2265 if(optlen>sizeof(tmp_fw) || optlen<1)
2266 return -EINVAL;
2267 err=verify_area(VERIFY_READ,optval,optlen);
2268 if(err)
2269 return err;
2270 memcpy_fromfs(&tmp_fw, optval,optlen);
2271 err=ip_acct_ctl(optname, &tmp_fw,optlen);
2272 return -err; /* -0 is 0 after all */
2273 #endif
2274 /* IP_OPTIONS and friends go here eventually */
2275 default:
2276 return(-ENOPROTOOPT);
2277 }
2278 }
2279
2280 /*
2281 * Get the options. Note for future reference. The GET of IP options gets the
2282 * _received_ ones. The set sets the _sent_ ones.
2283 */
2284
2285 int ip_getsockopt(struct sock *sk, int level, int optname, char *optval, int *optlen)
/* */
2286 {
2287 int val,err;
2288 #ifdef CONFIG_IP_MULTICAST
2289 int len;
2290 #endif
2291
2292 if(level!=SOL_IP)
2293 return -EOPNOTSUPP;
2294
2295 switch(optname)
2296 {
2297 case IP_TOS:
2298 val=sk->ip_tos;
2299 break;
2300 case IP_TTL:
2301 val=sk->ip_ttl;
2302 break;
2303 #ifdef CONFIG_IP_MULTICAST
2304 case IP_MULTICAST_TTL:
2305 val=sk->ip_mc_ttl;
2306 break;
2307 case IP_MULTICAST_LOOP:
2308 val=sk->ip_mc_loop;
2309 break;
2310 case IP_MULTICAST_IF:
2311 err=verify_area(VERIFY_WRITE, optlen, sizeof(int));
2312 if(err)
2313 return err;
2314 len=strlen(sk->ip_mc_name);
2315 err=verify_area(VERIFY_WRITE, optval, len);
2316 if(err)
2317 return err;
2318 put_user(len,(int *) optlen);
2319 memcpy_tofs((void *)optval,sk->ip_mc_name, len);
2320 return 0;
2321 #endif
2322 default:
2323 return(-ENOPROTOOPT);
2324 }
2325 err=verify_area(VERIFY_WRITE, optlen, sizeof(int));
2326 if(err)
2327 return err;
2328 put_user(sizeof(int),(int *) optlen);
2329
2330 err=verify_area(VERIFY_WRITE, optval, sizeof(int));
2331 if(err)
2332 return err;
2333 put_user(val,(int *) optval);
2334
2335 return(0);
2336 }
2337
2338 /*
2339 * Build and send a packet, with as little as one copy
2340 *
2341 * Doesn't care much about ip options... option length can be
2342 * different for fragment at 0 and other fragments.
2343 *
2344 * Note that the fragment at the highest offset is sent first,
2345 * so the getfrag routine can fill in the TCP/UDP checksum header
2346 * field in the last fragment it sends... actually it also helps
2347 * the reassemblers, they can put most packets in at the head of
2348 * the fragment queue, and they know the total size in advance. This
2349 * last feature will measurable improve the Linux fragment handler.
2350 *
2351 * The callback has five args, an arbitrary pointer (copy of frag),
2352 * the source IP address (may depend on the routing table), the
2353 * destination adddress (char *), the offset to copy from, and the
2354 * length to be copied.
2355 *
2356 */
2357
2358 int ip_build_xmit(struct sock *sk,
/* */
2359 void getfrag (const void *,
2360 int,
2361 char *,
2362 unsigned int,
2363 unsigned int),
2364 const void *frag,
2365 unsigned short int length,
2366 int daddr,
2367 int flags,
2368 int type)
2369 {
2370 struct rtable *rt;
2371 unsigned int fraglen, maxfraglen, fragheaderlen;
2372 int offset, mf;
2373 unsigned long saddr;
2374 unsigned short id;
2375 struct iphdr *iph;
2376 int local=0;
2377 struct device *dev;
2378 int nfrags=0;
2379
2380 ip_statistics.IpOutRequests++;
2381
2382
2383 #ifdef CONFIG_IP_MULTICAST
2384 if(sk && MULTICAST(daddr) && *sk->ip_mc_name)
2385 {
2386 dev=dev_get(sk->ip_mc_name);
2387 if(!dev)
2388 return -ENODEV;
2389 rt=NULL;
2390 if (sk->saddr && (!LOOPBACK(sk->saddr) || LOOPBACK(daddr)))
2391 saddr = sk->saddr;
2392 else
2393 saddr = dev->pa_addr;
2394 }
2395 else
2396 {
2397 #endif
2398 /*
2399 * Perform the IP routing decisions
2400 */
2401
2402 if(sk->localroute || flags&MSG_DONTROUTE)
2403 local=1;
2404
2405 rt = sk->ip_route_cache;
2406
2407 /*
2408 * See if the routing cache is outdated. We need to clean this up once we are happy it is reliable
2409 * by doing the invalidation actively in the route change and header change.
2410 */
2411
2412 saddr=sk->ip_route_saddr;
2413 if(!rt || sk->ip_route_stamp != rt_stamp || daddr!=sk->ip_route_daddr || sk->ip_route_local!=local || sk->saddr!=sk->ip_route_saddr)
2414 {
2415 if(local)
2416 rt = ip_rt_local(daddr, NULL, &saddr);
2417 else
2418 rt = ip_rt_route(daddr, NULL, &saddr);
2419 sk->ip_route_local=local;
2420 sk->ip_route_daddr=daddr;
2421 sk->ip_route_saddr=saddr;
2422 sk->ip_route_stamp=rt_stamp;
2423 sk->ip_route_cache=rt;
2424 sk->ip_hcache_ver=NULL;
2425 sk->ip_hcache_state= 0;
2426 }
2427 else if(rt)
2428 {
2429 /*
2430 * Attempt header caches only if the cached route is being reused. Header cache
2431 * is not ultra cheap to set up. This means we only set it up on the second packet,
2432 * so one shot communications are not slowed. We assume (seems reasonable) that 2 is
2433 * probably going to be a stream of data.
2434 */
2435 if(rt->rt_dev->header_cache && sk->ip_hcache_state!= -1)
2436 {
2437 if(sk->ip_hcache_ver==NULL || sk->ip_hcache_stamp!=*sk->ip_hcache_ver)
2438 rt->rt_dev->header_cache(rt->rt_dev,sk,saddr,daddr);
2439 else
2440 /* Can't cache. Remember this */
2441 sk->ip_hcache_state= -1;
2442 }
2443 }
2444
2445 if (rt == NULL)
2446 {
2447 ip_statistics.IpOutNoRoutes++;
2448 return(-ENETUNREACH);
2449 }
2450
2451 if (sk->saddr && (!LOOPBACK(sk->saddr) || LOOPBACK(daddr)))
2452 saddr = sk->saddr;
2453
2454 dev=rt->rt_dev;
2455 #ifdef CONFIG_IP_MULTICAST
2456 }
2457 #endif
2458
2459 /*
2460 * Now compute the buffer space we require
2461 */
2462
2463 /*
2464 * Try the simple case first. This leaves broadcast, multicast, fragmented frames, and by
2465 * choice RAW frames within 20 bytes of maximum size(rare) to the long path
2466 */
2467
2468 if(length+20 <= dev->mtu && !MULTICAST(daddr) && daddr!=0xFFFFFFFF && daddr!=dev->pa_brdaddr)
2469 {
2470 int error;
2471 struct sk_buff *skb=sock_alloc_send_skb(sk, length+20+15+dev->hard_header_len,0,&error);
2472 if(skb==NULL)
2473 {
2474 ip_statistics.IpOutDiscards++;
2475 return error;
2476 }
2477 skb->dev=dev;
2478 skb->free=1;
2479 skb->when=jiffies;
2480 skb->sk=sk;
2481 skb->arp=0;
2482 skb->saddr=saddr;
2483 length+=20; /* We do this twice so the subtract once is quicker */
2484 skb->raddr=(rt&&rt->rt_gateway)?rt->rt_gateway:daddr;
2485 skb_reserve(skb,(dev->hard_header_len+15)&~15);
2486 if(sk->ip_hcache_state>0)
2487 {
2488 memcpy(skb_push(skb,dev->hard_header_len),sk->ip_hcache_data,dev->hard_header_len);
2489 skb->arp=1;
2490 }
2491 else if(dev->hard_header)
2492 {
2493 if(dev->hard_header(skb,dev,ETH_P_IP,NULL,NULL,0)>0)
2494 skb->arp=1;
2495 }
2496 skb->ip_hdr=iph=(struct iphdr *)skb_put(skb,length);
2497 if(type!=IPPROTO_RAW)
2498 {
2499 iph->version=4;
2500 iph->ihl=5;
2501 iph->tos=sk->ip_tos;
2502 iph->tot_len = htons(length);
2503 iph->id=htons(ip_id_count++);
2504 iph->frag_off = 0;
2505 iph->ttl=sk->ip_ttl;
2506 iph->protocol=type;
2507 iph->saddr=saddr;
2508 iph->daddr=daddr;
2509 iph->check=0;
2510 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
2511 getfrag(frag,saddr,(void *)(iph+1),0, length-20);
2512 }
2513 else
2514 getfrag(frag,saddr,(void *)iph,0,length);
2515 #ifdef CONFIG_IP_ACCT
2516 ip_fw_chk((void *)skb->data,dev,ip_acct_chain, IP_FW_F_ACCEPT,1);
2517 #endif
2518 if(dev->flags&IFF_UP)
2519 dev_queue_xmit(skb,dev,sk->priority);
2520 else
2521 {
2522 ip_statistics.IpOutDiscards++;
2523 kfree_skb(skb, FREE_WRITE);
2524 }
2525 return 0;
2526 }
2527
2528
2529 fragheaderlen = dev->hard_header_len;
2530 if(type != IPPROTO_RAW)
2531 fragheaderlen += 20;
2532
2533 /*
2534 * Fragheaderlen is the size of 'overhead' on each buffer. Now work
2535 * out the size of the frames to send.
2536 */
2537
2538 maxfraglen = ((dev->mtu-20) & ~7) + fragheaderlen;
2539
2540 /*
2541 * Start at the end of the frame by handling the remainder.
2542 */
2543
2544 offset = length - (length % (maxfraglen - fragheaderlen));
2545
2546 /*
2547 * Amount of memory to allocate for final fragment.
2548 */
2549
2550 fraglen = length - offset + fragheaderlen;
2551
2552 if(fraglen==0)
2553 {
2554 fraglen = maxfraglen;
2555 offset -= maxfraglen-fragheaderlen;
2556 }
2557
2558
2559 /*
2560 * The last fragment will not have MF (more fragments) set.
2561 */
2562
2563 mf = 0;
2564
2565 /*
2566 * Can't fragment raw packets
2567 */
2568
2569 if (type == IPPROTO_RAW && offset > 0)
2570 return(-EMSGSIZE);
2571
2572 /*
2573 * Get an identifier
2574 */
2575
2576 id = htons(ip_id_count++);
2577
2578 /*
2579 * Being outputting the bytes.
2580 */
2581
2582 do
2583 {
2584 struct sk_buff * skb;
2585 int error;
2586 char *data;
2587
2588 /*
2589 * Get the memory we require with some space left for alignment.
2590 */
2591
2592 skb = sock_alloc_send_skb(sk, fraglen+15, 0, &error);
2593 if (skb == NULL)
2594 {
2595 ip_statistics.IpOutDiscards++;
2596 if(nfrags>1)
2597 ip_statistics.IpFragCreates++;
2598 return(error);
2599 }
2600
2601 /*
2602 * Fill in the control structures
2603 */
2604
2605 skb->next = skb->prev = NULL;
2606 skb->dev = dev;
2607 skb->when = jiffies;
2608 skb->free = 1; /* dubious, this one */
2609 skb->sk = sk;
2610 skb->arp = 0;
2611 skb->saddr = saddr;
2612 skb->raddr = (rt&&rt->rt_gateway) ? rt->rt_gateway : daddr;
2613 skb_reserve(skb,(dev->hard_header_len+15)&~15);
2614 data = skb_put(skb, fraglen-dev->hard_header_len);
2615
2616 /*
2617 * Save us ARP and stuff. In the optimal case we do no route lookup (route cache ok)
2618 * no ARP lookup (arp cache ok) and output. The cache checks are still too slow but
2619 * this can be fixed later. For gateway routes we ought to have a rt->.. header cache
2620 * pointer to speed header cache builds for identical targets.
2621 */
2622
2623 if(sk->ip_hcache_state>0)
2624 {
2625 memcpy(skb_push(skb,dev->hard_header_len),sk->ip_hcache_data, dev->hard_header_len);
2626 skb->arp=1;
2627 }
2628 else if (dev->hard_header)
2629 {
2630 if(dev->hard_header(skb, dev, ETH_P_IP,
2631 NULL, NULL, 0)>0)
2632 skb->arp=1;
2633 }
2634
2635 /*
2636 * Find where to start putting bytes.
2637 */
2638
2639 skb->ip_hdr = iph = (struct iphdr *)data;
2640
2641 /*
2642 * Only write IP header onto non-raw packets
2643 */
2644
2645 if(type != IPPROTO_RAW)
2646 {
2647
2648 iph->version = 4;
2649 iph->ihl = 5; /* ugh */
2650 iph->tos = sk->ip_tos;
2651 iph->tot_len = htons(fraglen - fragheaderlen + iph->ihl*4);
2652 iph->id = id;
2653 iph->frag_off = htons(offset>>3);
2654 iph->frag_off |= mf;
2655 #ifdef CONFIG_IP_MULTICAST
2656 if (MULTICAST(daddr))
2657 iph->ttl = sk->ip_mc_ttl;
2658 else
2659 #endif
2660 iph->ttl = sk->ip_ttl;
2661 iph->protocol = type;
2662 iph->check = 0;
2663 iph->saddr = saddr;
2664 iph->daddr = daddr;
2665 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
2666 data += iph->ihl*4;
2667
2668 /*
2669 * Any further fragments will have MF set.
2670 */
2671
2672 mf = htons(IP_MF);
2673 }
2674
2675 /*
2676 * User data callback
2677 */
2678
2679 getfrag(frag, saddr, data, offset, fraglen-fragheaderlen);
2680
2681 /*
2682 * Account for the fragment.
2683 */
2684
2685 #ifdef CONFIG_IP_ACCT
2686 if(!offset)
2687 ip_fw_chk(iph, dev, ip_acct_chain, IP_FW_F_ACCEPT, 1);
2688 #endif
2689 offset -= (maxfraglen-fragheaderlen);
2690 fraglen = maxfraglen;
2691
2692 #ifdef CONFIG_IP_MULTICAST
2693
2694 /*
2695 * Multicasts are looped back for other local users
2696 */
2697
2698 if (MULTICAST(daddr) && !(dev->flags&IFF_LOOPBACK))
2699 {
2700 /*
2701 * Loop back any frames. The check for IGMP_ALL_HOSTS is because
2702 * you are always magically a member of this group.
2703 */
2704
2705 if(sk==NULL || sk->ip_mc_loop)
2706 {
2707 if(skb->daddr==IGMP_ALL_HOSTS)
2708 ip_loopback(rt?rt->rt_dev:dev,skb);
2709 else
2710 {
2711 struct ip_mc_list *imc=rt?rt->rt_dev->ip_mc_list:dev->ip_mc_list;
2712 while(imc!=NULL)
2713 {
2714 if(imc->multiaddr==daddr)
2715 {
2716 ip_loopback(rt?rt->rt_dev:dev,skb);
2717 break;
2718 }
2719 imc=imc->next;
2720 }
2721 }
2722 }
2723
2724 /*
2725 * Multicasts with ttl 0 must not go beyond the host. Fixme: avoid the
2726 * extra clone.
2727 */
2728
2729 if(skb->ip_hdr->ttl==0)
2730 kfree_skb(skb, FREE_READ);
2731 }
2732 #endif
2733
2734 nfrags++;
2735
2736 /*
2737 * BSD loops broadcasts
2738 */
2739
2740 if((dev->flags&IFF_BROADCAST) && (daddr==0xFFFFFFFF || daddr==dev->pa_brdaddr) && !(dev->flags&IFF_LOOPBACK))
2741 ip_loopback(dev,skb);
2742
2743 /*
2744 * Now queue the bytes into the device.
2745 */
2746
2747 if (dev->flags & IFF_UP)
2748 {
2749 dev_queue_xmit(skb, dev, sk->priority);
2750 }
2751 else
2752 {
2753 /*
2754 * Whoops...
2755 *
2756 * FIXME: There is a small nasty here. During the ip_build_xmit we could
2757 * page fault between the route lookup and device send, the device might be
2758 * removed and unloaded.... We need to add device locks on this.
2759 */
2760
2761 ip_statistics.IpOutDiscards++;
2762 if(nfrags>1)
2763 ip_statistics.IpFragCreates+=nfrags;
2764 kfree_skb(skb, FREE_WRITE);
2765 return(0); /* lose rest of fragments */
2766 }
2767 }
2768 while (offset >= 0);
2769 if(nfrags>1)
2770 ip_statistics.IpFragCreates+=nfrags;
2771 return(0);
2772 }
2773
2774
2775 /*
2776 * IP protocol layer initialiser
2777 */
2778
2779 static struct packet_type ip_packet_type =
2780 {
2781 0, /* MUTTER ntohs(ETH_P_IP),*/
2782 NULL, /* All devices */
2783 ip_rcv,
2784 NULL,
2785 NULL,
2786 };
2787
2788 /*
2789 * Device notifier
2790 */
2791
2792 static int ip_rt_event(unsigned long event, void *ptr)
/* */
2793 {
2794 if(event==NETDEV_DOWN)
2795 ip_rt_flush(ptr);
2796 return NOTIFY_DONE;
2797 }
2798
2799 struct notifier_block ip_rt_notifier={
2800 ip_rt_event,
2801 NULL,
2802 0
2803 };
2804
2805 /*
2806 * IP registers the packet type and then calls the subprotocol initialisers
2807 */
2808
2809 void ip_init(void)
/* ![[last]](../icons/n_last.png)
*/
2810 {
2811 ip_packet_type.type=htons(ETH_P_IP);
2812 dev_add_pack(&ip_packet_type);
2813
2814 /* So we flush routes when a device is downed */
2815 register_netdevice_notifier(&ip_rt_notifier);
2816
2817 /* ip_raw_init();
2818 ip_packet_init();
2819 ip_tcp_init();
2820 ip_udp_init();*/
2821
2822 #ifdef CONFIG_IP_MULTICAST
2823 proc_net_register(&(struct proc_dir_entry) {
2824 PROC_NET_IGMP, 4, "igmp",
2825 S_IFREG | S_IRUGO, 1, 0, 0,
2826 0, &proc_net_inode_operations,
2827 ip_mc_procinfo
2828 });
2829 #endif
2830 }
2831
![[bottom]](../icons/n_bottom.png){kind=icon}
*/