![[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 * ROUTE - implementation of the IP router.
7 *
8 * Version: @(#)route.c 1.0.14 05/31/93
9 *
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
14 *
15 * Fixes:
16 * Alan Cox : Verify area fixes.
17 * Alan Cox : cli() protects routing changes
18 * Rui Oliveira : ICMP routing table updates
19 * (rco@di.uminho.pt) Routing table insertion and update
20 * Linus Torvalds : Rewrote bits to be sensible
21 * Alan Cox : Added BSD route gw semantics
22 * Alan Cox : Super /proc >4K
23 * Alan Cox : MTU in route table
24 * Alan Cox : MSS actually. Also added the window
25 * clamper.
26 * Sam Lantinga : Fixed route matching in rt_del()
27 *
28 * This program is free software; you can redistribute it and/or
29 * modify it under the terms of the GNU General Public License
30 * as published by the Free Software Foundation; either version
31 * 2 of the License, or (at your option) any later version.
32 */
33
34 #include <asm/segment.h>
35 #include <asm/system.h>
36 #include <linux/types.h>
37 #include <linux/kernel.h>
38 #include <linux/sched.h>
39 #include <linux/mm.h>
40 #include <linux/string.h>
41 #include <linux/socket.h>
42 #include <linux/sockios.h>
43 #include <linux/errno.h>
44 #include <linux/in.h>
45 #include <linux/inet.h>
46 #include <linux/netdevice.h>
47 #include "ip.h"
48 #include "protocol.h"
49 #include "route.h"
50 #include "tcp.h"
51 #include <linux/skbuff.h>
52 #include "sock.h"
53 #include "icmp.h"
54
55 /*
56 * The routing table list
57 */
58
59 static struct rtable *rt_base = NULL;
60
61 /*
62 * Pointer to the loopback route
63 */
64
65 static struct rtable *rt_loopback = NULL;
66
67 /*
68 * Remove a routing table entry.
69 */
70
71 static void rt_del(unsigned long dst, char *devname)
/* ![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[top]](../icons/top.png)
*/
72 {
73 struct rtable *r, **rp;
74 unsigned long flags;
75
76 rp = &rt_base;
77
78 /*
79 * This must be done with interrupts off because we could take
80 * an ICMP_REDIRECT.
81 */
82
83 save_flags(flags);
84 cli();
85 while((r = *rp) != NULL)
86 {
87 /* Make sure both the destination and the device match */
88 if ( r->rt_dst != dst ||
89 (devname != NULL && strcmp((r->rt_dev)->name,devname) != 0) )
90 {
91 rp = &r->rt_next;
92 continue;
93 }
94 *rp = r->rt_next;
95
96 /*
97 * If we delete the loopback route update its pointer.
98 */
99
100 if (rt_loopback == r)
101 rt_loopback = NULL;
102 kfree_s(r, sizeof(struct rtable));
103 }
104 restore_flags(flags);
105 }
106
107
108 /*
109 * Remove all routing table entries for a device. This is called when
110 * a device is downed.
111 */
112
113 void ip_rt_flush(struct device *dev)
/* ![[previous]](../icons/left.png)
*/
114 {
115 struct rtable *r;
116 struct rtable **rp;
117 unsigned long flags;
118
119 rp = &rt_base;
120 save_flags(flags);
121 cli();
122 while ((r = *rp) != NULL) {
123 if (r->rt_dev != dev) {
124 rp = &r->rt_next;
125 continue;
126 }
127 *rp = r->rt_next;
128 if (rt_loopback == r)
129 rt_loopback = NULL;
130 kfree_s(r, sizeof(struct rtable));
131 }
132 restore_flags(flags);
133 }
134
135 /*
136 * Used by 'rt_add()' when we can't get the netmask any other way..
137 *
138 * If the lower byte or two are zero, we guess the mask based on the
139 * number of zero 8-bit net numbers, otherwise we use the "default"
140 * masks judging by the destination address and our device netmask.
141 */
142
143 static inline unsigned long default_mask(unsigned long dst)
/* */
144 {
145 dst = ntohl(dst);
146 if (IN_CLASSA(dst))
147 return htonl(IN_CLASSA_NET);
148 if (IN_CLASSB(dst))
149 return htonl(IN_CLASSB_NET);
150 return htonl(IN_CLASSC_NET);
151 }
152
153
154 /*
155 * If no mask is specified then generate a default entry.
156 */
157
158 static unsigned long guess_mask(unsigned long dst, struct device * dev)
/* */
159 {
160 unsigned long mask;
161
162 if (!dst)
163 return 0;
164 mask = default_mask(dst);
165 if ((dst ^ dev->pa_addr) & mask)
166 return mask;
167 return dev->pa_mask;
168 }
169
170
171 /*
172 * Find the route entry through which our gateway will be reached
173 */
174
175 static inline struct device * get_gw_dev(unsigned long gw)
/* */
176 {
177 struct rtable * rt;
178
179 for (rt = rt_base ; ; rt = rt->rt_next)
180 {
181 if (!rt)
182 return NULL;
183 if ((gw ^ rt->rt_dst) & rt->rt_mask)
184 continue;
185 /*
186 * Gateways behind gateways are a no-no
187 */
188
189 if (rt->rt_flags & RTF_GATEWAY)
190 return NULL;
191 return rt->rt_dev;
192 }
193 }
194
195 /*
196 * Rewrote rt_add(), as the old one was weird - Linus
197 *
198 * This routine is used to update the IP routing table, either
199 * from the kernel (ICMP_REDIRECT) or via an ioctl call issued
200 * by the superuser.
201 */
202
203 void ip_rt_add(short flags, unsigned long dst, unsigned long mask,
/* */
204 unsigned long gw, struct device *dev, unsigned short mtu, unsigned long window)
205 {
206 struct rtable *r, *rt;
207 struct rtable **rp;
208 unsigned long cpuflags;
209
210 /*
211 * A host is a unique machine and has no network bits.
212 */
213
214 if (flags & RTF_HOST)
215 {
216 mask = 0xffffffff;
217 }
218
219 /*
220 * Calculate the network mask
221 */
222
223 else if (!mask)
224 {
225 if (!((dst ^ dev->pa_addr) & dev->pa_mask))
226 {
227 mask = dev->pa_mask;
228 flags &= ~RTF_GATEWAY;
229 if (flags & RTF_DYNAMIC)
230 {
231 /*printk("Dynamic route to my own net rejected\n");*/
232 return;
233 }
234 }
235 else
236 mask = guess_mask(dst, dev);
237 dst &= mask;
238 }
239
240 /*
241 * A gateway must be reachable and not a local address
242 */
243
244 if (gw == dev->pa_addr)
245 flags &= ~RTF_GATEWAY;
246
247 if (flags & RTF_GATEWAY)
248 {
249 /*
250 * Don't try to add a gateway we can't reach..
251 */
252
253 if (dev != get_gw_dev(gw))
254 return;
255
256 flags |= RTF_GATEWAY;
257 }
258 else
259 gw = 0;
260
261 /*
262 * Allocate an entry and fill it in.
263 */
264
265 rt = (struct rtable *) kmalloc(sizeof(struct rtable), GFP_ATOMIC);
266 if (rt == NULL)
267 {
268 return;
269 }
270 memset(rt, 0, sizeof(struct rtable));
271 rt->rt_flags = flags | RTF_UP;
272 rt->rt_dst = dst;
273 rt->rt_dev = dev;
274 rt->rt_gateway = gw;
275 rt->rt_mask = mask;
276 rt->rt_mss = dev->mtu - HEADER_SIZE;
277 rt->rt_window = 0; /* Default is no clamping */
278
279 /* Are the MSS/Window valid ? */
280
281 if(rt->rt_flags & RTF_MSS)
282 rt->rt_mss = mtu;
283
284 if(rt->rt_flags & RTF_WINDOW)
285 rt->rt_window = window;
286
287 /*
288 * What we have to do is loop though this until we have
289 * found the first address which has a higher generality than
290 * the one in rt. Then we can put rt in right before it.
291 * The interrupts must be off for this process.
292 */
293
294 save_flags(cpuflags);
295 cli();
296
297 /*
298 * Remove old route if we are getting a duplicate.
299 */
300
301 rp = &rt_base;
302 while ((r = *rp) != NULL)
303 {
304 if (r->rt_dst != dst ||
305 r->rt_mask != mask)
306 {
307 rp = &r->rt_next;
308 continue;
309 }
310 *rp = r->rt_next;
311 if (rt_loopback == r)
312 rt_loopback = NULL;
313 kfree_s(r, sizeof(struct rtable));
314 }
315
316 /*
317 * Add the new route
318 */
319
320 rp = &rt_base;
321 while ((r = *rp) != NULL) {
322 if ((r->rt_mask & mask) != mask)
323 break;
324 rp = &r->rt_next;
325 }
326 rt->rt_next = r;
327 *rp = rt;
328
329 /*
330 * Update the loopback route
331 */
332
333 if ((rt->rt_dev->flags & IFF_LOOPBACK) && !rt_loopback)
334 rt_loopback = rt;
335
336 /*
337 * Restore the interrupts and return
338 */
339
340 restore_flags(cpuflags);
341 return;
342 }
343
344
345 /*
346 * Check if a mask is acceptable.
347 */
348
349 static inline int bad_mask(unsigned long mask, unsigned long addr)
/* */
350 {
351 if (addr & (mask = ~mask))
352 return 1;
353 mask = ntohl(mask);
354 if (mask & (mask+1))
355 return 1;
356 return 0;
357 }
358
359 /*
360 * Process a route add request from the user
361 */
362
363 static int rt_new(struct rtentry *r)
/* */
364 {
365 int err;
366 char * devname;
367 struct device * dev = NULL;
368 unsigned long flags, daddr, mask, gw;
369
370 /*
371 * If a device is specified find it.
372 */
373
374 if ((devname = r->rt_dev) != NULL)
375 {
376 err = getname(devname, &devname);
377 if (err)
378 return err;
379 dev = dev_get(devname);
380 putname(devname);
381 if (!dev)
382 return -EINVAL;
383 }
384
385 /*
386 * If the device isn't INET, don't allow it
387 */
388
389 if (r->rt_dst.sa_family != AF_INET)
390 return -EAFNOSUPPORT;
391
392 /*
393 * Make local copies of the important bits
394 */
395
396 flags = r->rt_flags;
397 daddr = ((struct sockaddr_in *) &r->rt_dst)->sin_addr.s_addr;
398 mask = ((struct sockaddr_in *) &r->rt_genmask)->sin_addr.s_addr;
399 gw = ((struct sockaddr_in *) &r->rt_gateway)->sin_addr.s_addr;
400
401
402 /*
403 * BSD emulation: Permits route add someroute gw one-of-my-addresses
404 * to indicate which iface. Not as clean as the nice Linux dev technique
405 * but people keep using it...
406 */
407
408 if (!dev && (flags & RTF_GATEWAY))
409 {
410 struct device *dev2;
411 for (dev2 = dev_base ; dev2 != NULL ; dev2 = dev2->next)
412 {
413 if ((dev2->flags & IFF_UP) && dev2->pa_addr == gw)
414 {
415 flags &= ~RTF_GATEWAY;
416 dev = dev2;
417 break;
418 }
419 }
420 }
421
422 /*
423 * Ignore faulty masks
424 */
425
426 if (bad_mask(mask, daddr))
427 mask = 0;
428
429 /*
430 * Set the mask to nothing for host routes.
431 */
432
433 if (flags & RTF_HOST)
434 mask = 0xffffffff;
435 else if (mask && r->rt_genmask.sa_family != AF_INET)
436 return -EAFNOSUPPORT;
437
438 /*
439 * You can only gateway IP via IP..
440 */
441
442 if (flags & RTF_GATEWAY)
443 {
444 if (r->rt_gateway.sa_family != AF_INET)
445 return -EAFNOSUPPORT;
446 if (!dev)
447 dev = get_gw_dev(gw);
448 }
449 else if (!dev)
450 dev = ip_dev_check(daddr);
451
452 /*
453 * Unknown device.
454 */
455
456 if (dev == NULL)
457 return -ENETUNREACH;
458
459 /*
460 * Add the route
461 */
462
463 ip_rt_add(flags, daddr, mask, gw, dev, r->rt_mss, r->rt_window);
464 return 0;
465 }
466
467
468 /*
469 * Remove a route, as requested by the user.
470 */
471
472 static int rt_kill(struct rtentry *r)
/* */
473 {
474 struct sockaddr_in *trg;
475 char *devname;
476 int err;
477
478 trg = (struct sockaddr_in *) &r->rt_dst;
479 if ((devname = r->rt_dev) != NULL)
480 {
481 err = getname(devname, &devname);
482 if (err)
483 return err;
484 }
485 rt_del(trg->sin_addr.s_addr, devname);
486 if ( devname != NULL )
487 putname(devname);
488 return 0;
489 }
490
491
492 /*
493 * Called from the PROCfs module. This outputs /proc/net/route.
494 */
495
496 int rt_get_info(char *buffer, char **start, off_t offset, int length)
/* */
497 {
498 struct rtable *r;
499 int len=0;
500 off_t pos=0;
501 off_t begin=0;
502 int size;
503
504 len += sprintf(buffer,
505 "Iface\tDestination\tGateway \tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU\tWindow\n");
506 pos=len;
507
508 /*
509 * This isn't quite right -- r->rt_dst is a struct!
510 */
511
512 for (r = rt_base; r != NULL; r = r->rt_next)
513 {
514 size = sprintf(buffer+len, "%s\t%08lX\t%08lX\t%02X\t%d\t%lu\t%d\t%08lX\t%d\t%lu\n",
515 r->rt_dev->name, r->rt_dst, r->rt_gateway,
516 r->rt_flags, r->rt_refcnt, r->rt_use, r->rt_metric,
517 r->rt_mask, (int)r->rt_mss, r->rt_window);
518 len+=size;
519 pos+=size;
520 if(pos<offset)
521 {
522 len=0;
523 begin=pos;
524 }
525 if(pos>offset+length)
526 break;
527 }
528
529 *start=buffer+(offset-begin);
530 len-=(offset-begin);
531 if(len>length)
532 len=length;
533 return len;
534 }
535
536 /*
537 * This is hackish, but results in better code. Use "-S" to see why.
538 */
539
540 #define early_out ({ goto no_route; 1; })
541
542 /*
543 * Route a packet. This needs to be fairly quick. Florian & Co.
544 * suggested a unified ARP and IP routing cache. Done right its
545 * probably a brilliant idea. I'd actually suggest a unified
546 * ARP/IP routing/Socket pointer cache. Volunteers welcome
547 */
548
549 struct rtable * ip_rt_route(unsigned long daddr, struct options *opt, unsigned long *src_addr)
/* */
550 {
551 struct rtable *rt;
552
553 for (rt = rt_base; rt != NULL || early_out ; rt = rt->rt_next)
554 {
555 if (!((rt->rt_dst ^ daddr) & rt->rt_mask))
556 break;
557 /*
558 * broadcast addresses can be special cases..
559 */
560 if (rt->rt_flags & RTF_GATEWAY)
561 continue;
562 if ((rt->rt_dev->flags & IFF_BROADCAST) &&
563 (rt->rt_dev->pa_brdaddr == daddr))
564 break;
565 }
566
567 if(src_addr!=NULL)
568 *src_addr= rt->rt_dev->pa_addr;
569
570 if (daddr == rt->rt_dev->pa_addr) {
571 if ((rt = rt_loopback) == NULL)
572 goto no_route;
573 }
574 rt->rt_use++;
575 return rt;
576 no_route:
577 return NULL;
578 }
579
580 struct rtable * ip_rt_local(unsigned long daddr, struct options *opt, unsigned long *src_addr)
/* */
581 {
582 struct rtable *rt;
583
584 for (rt = rt_base; rt != NULL || early_out ; rt = rt->rt_next)
585 {
586 /*
587 * No routed addressing.
588 */
589 if (rt->rt_flags&RTF_GATEWAY)
590 continue;
591
592 if (!((rt->rt_dst ^ daddr) & rt->rt_mask))
593 break;
594 /*
595 * broadcast addresses can be special cases..
596 */
597
598 if ((rt->rt_dev->flags & IFF_BROADCAST) &&
599 rt->rt_dev->pa_brdaddr == daddr)
600 break;
601 }
602
603 if(src_addr!=NULL)
604 *src_addr= rt->rt_dev->pa_addr;
605
606 if (daddr == rt->rt_dev->pa_addr) {
607 if ((rt = rt_loopback) == NULL)
608 goto no_route;
609 }
610 rt->rt_use++;
611 return rt;
612 no_route:
613 return NULL;
614 }
615
616 /*
617 * Backwards compatibility
618 */
619
620 static int ip_get_old_rtent(struct old_rtentry * src, struct rtentry * rt)
/* */
621 {
622 int err;
623 struct old_rtentry tmp;
624
625 err=verify_area(VERIFY_READ, src, sizeof(*src));
626 if (err)
627 return err;
628 memcpy_fromfs(&tmp, src, sizeof(*src));
629 memset(rt, 0, sizeof(*rt));
630 rt->rt_dst = tmp.rt_dst;
631 rt->rt_gateway = tmp.rt_gateway;
632 rt->rt_genmask.sa_family = AF_INET;
633 ((struct sockaddr_in *) &rt->rt_genmask)->sin_addr.s_addr = tmp.rt_genmask;
634 rt->rt_flags = tmp.rt_flags;
635 rt->rt_dev = tmp.rt_dev;
636 printk("Warning: obsolete routing request made.\n");
637 return 0;
638 }
639
640 /*
641 * Handle IP routing ioctl calls. These are used to manipulate the routing tables
642 */
643
644 int ip_rt_ioctl(unsigned int cmd, void *arg)
/* ![[last]](../icons/n_last.png)
*/
645 {
646 int err;
647 struct rtentry rt;
648
649 switch(cmd)
650 {
651 case SIOCADDRTOLD: /* Old style add route */
652 case SIOCDELRTOLD: /* Old style delete route */
653 if (!suser())
654 return -EPERM;
655 err = ip_get_old_rtent((struct old_rtentry *) arg, &rt);
656 if (err)
657 return err;
658 return (cmd == SIOCDELRTOLD) ? rt_kill(&rt) : rt_new(&rt);
659
660 case SIOCADDRT: /* Add a route */
661 case SIOCDELRT: /* Delete a route */
662 if (!suser())
663 return -EPERM;
664 err=verify_area(VERIFY_READ, arg, sizeof(struct rtentry));
665 if (err)
666 return err;
667 memcpy_fromfs(&rt, arg, sizeof(struct rtentry));
668 return (cmd == SIOCDELRT) ? rt_kill(&rt) : rt_new(&rt);
669 }
670
671 return -EINVAL;
672 }
![[bottom]](../icons/n_bottom.png){kind=icon}
*/