[INET]: Remove no longer needed ->equal callback
[safe/jmp/linux-2.6] / net / ipv6 / reassembly.c
1 /*
2  *      IPv6 fragment reassembly
3  *      Linux INET6 implementation
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>
7  *
8  *      $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
9  *
10  *      Based on: net/ipv4/ip_fragment.c
11  *
12  *      This program is free software; you can redistribute it and/or
13  *      modify it under the terms of the GNU General Public License
14  *      as published by the Free Software Foundation; either version
15  *      2 of the License, or (at your option) any later version.
16  */
17
18 /*
19  *      Fixes:
20  *      Andi Kleen      Make it work with multiple hosts.
21  *                      More RFC compliance.
22  *
23  *      Horst von Brand Add missing #include <linux/string.h>
24  *      Alexey Kuznetsov        SMP races, threading, cleanup.
25  *      Patrick McHardy         LRU queue of frag heads for evictor.
26  *      Mitsuru KANDA @USAGI    Register inet6_protocol{}.
27  *      David Stevens and
28  *      YOSHIFUJI,H. @USAGI     Always remove fragment header to
29  *                              calculate ICV correctly.
30  */
31 #include <linux/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/sockios.h>
36 #include <linux/jiffies.h>
37 #include <linux/net.h>
38 #include <linux/list.h>
39 #include <linux/netdevice.h>
40 #include <linux/in6.h>
41 #include <linux/ipv6.h>
42 #include <linux/icmpv6.h>
43 #include <linux/random.h>
44 #include <linux/jhash.h>
45 #include <linux/skbuff.h>
46
47 #include <net/sock.h>
48 #include <net/snmp.h>
49
50 #include <net/ipv6.h>
51 #include <net/ip6_route.h>
52 #include <net/protocol.h>
53 #include <net/transp_v6.h>
54 #include <net/rawv6.h>
55 #include <net/ndisc.h>
56 #include <net/addrconf.h>
57 #include <net/inet_frag.h>
58
59 struct ip6frag_skb_cb
60 {
61         struct inet6_skb_parm   h;
62         int                     offset;
63 };
64
65 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
66
67
68 /*
69  *      Equivalent of ipv4 struct ipq
70  */
71
72 struct frag_queue
73 {
74         struct inet_frag_queue  q;
75
76         __be32                  id;             /* fragment id          */
77         struct in6_addr         saddr;
78         struct in6_addr         daddr;
79
80         int                     iif;
81         unsigned int            csum;
82         __u16                   nhoffset;
83 };
84
85 struct inet_frags_ctl ip6_frags_ctl __read_mostly = {
86         .high_thresh     = 256 * 1024,
87         .low_thresh      = 192 * 1024,
88         .timeout         = IPV6_FRAG_TIMEOUT,
89         .secret_interval = 10 * 60 * HZ,
90 };
91
92 static struct inet_frags ip6_frags;
93
94 int ip6_frag_nqueues(void)
95 {
96         return ip6_frags.nqueues;
97 }
98
99 int ip6_frag_mem(void)
100 {
101         return atomic_read(&ip6_frags.mem);
102 }
103
104 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
105                           struct net_device *dev);
106
107 /*
108  * callers should be careful not to use the hash value outside the ipfrag_lock
109  * as doing so could race with ipfrag_hash_rnd being recalculated.
110  */
111 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
112                                struct in6_addr *daddr)
113 {
114         u32 a, b, c;
115
116         a = (__force u32)saddr->s6_addr32[0];
117         b = (__force u32)saddr->s6_addr32[1];
118         c = (__force u32)saddr->s6_addr32[2];
119
120         a += JHASH_GOLDEN_RATIO;
121         b += JHASH_GOLDEN_RATIO;
122         c += ip6_frags.rnd;
123         __jhash_mix(a, b, c);
124
125         a += (__force u32)saddr->s6_addr32[3];
126         b += (__force u32)daddr->s6_addr32[0];
127         c += (__force u32)daddr->s6_addr32[1];
128         __jhash_mix(a, b, c);
129
130         a += (__force u32)daddr->s6_addr32[2];
131         b += (__force u32)daddr->s6_addr32[3];
132         c += (__force u32)id;
133         __jhash_mix(a, b, c);
134
135         return c & (INETFRAGS_HASHSZ - 1);
136 }
137
138 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
139 {
140         struct frag_queue *fq;
141
142         fq = container_of(q, struct frag_queue, q);
143         return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
144 }
145
146 int ip6_frag_match(struct inet_frag_queue *q, void *a)
147 {
148         struct frag_queue *fq;
149         struct ip6_create_arg *arg = a;
150
151         fq = container_of(q, struct frag_queue, q);
152         return (fq->id == arg->id &&
153                         ipv6_addr_equal(&fq->saddr, arg->src) &&
154                         ipv6_addr_equal(&fq->daddr, arg->dst));
155 }
156 EXPORT_SYMBOL(ip6_frag_match);
157
158 /* Memory Tracking Functions. */
159 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
160 {
161         if (work)
162                 *work -= skb->truesize;
163         atomic_sub(skb->truesize, &ip6_frags.mem);
164         kfree_skb(skb);
165 }
166
167 void ip6_frag_init(struct inet_frag_queue *q, void *a)
168 {
169         struct frag_queue *fq = container_of(q, struct frag_queue, q);
170         struct ip6_create_arg *arg = a;
171
172         fq->id = arg->id;
173         ipv6_addr_copy(&fq->saddr, arg->src);
174         ipv6_addr_copy(&fq->daddr, arg->dst);
175 }
176 EXPORT_SYMBOL(ip6_frag_init);
177
178 static void ip6_frag_free(struct inet_frag_queue *fq)
179 {
180         kfree(container_of(fq, struct frag_queue, q));
181 }
182
183 /* Destruction primitives. */
184
185 static __inline__ void fq_put(struct frag_queue *fq)
186 {
187         inet_frag_put(&fq->q, &ip6_frags);
188 }
189
190 /* Kill fq entry. It is not destroyed immediately,
191  * because caller (and someone more) holds reference count.
192  */
193 static __inline__ void fq_kill(struct frag_queue *fq)
194 {
195         inet_frag_kill(&fq->q, &ip6_frags);
196 }
197
198 static void ip6_evictor(struct inet6_dev *idev)
199 {
200         int evicted;
201
202         evicted = inet_frag_evictor(&ip6_frags);
203         if (evicted)
204                 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
205 }
206
207 static void ip6_frag_expire(unsigned long data)
208 {
209         struct frag_queue *fq;
210         struct net_device *dev = NULL;
211
212         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
213
214         spin_lock(&fq->q.lock);
215
216         if (fq->q.last_in & COMPLETE)
217                 goto out;
218
219         fq_kill(fq);
220
221         dev = dev_get_by_index(&init_net, fq->iif);
222         if (!dev)
223                 goto out;
224
225         rcu_read_lock();
226         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
227         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
228         rcu_read_unlock();
229
230         /* Don't send error if the first segment did not arrive. */
231         if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
232                 goto out;
233
234         /*
235            But use as source device on which LAST ARRIVED
236            segment was received. And do not use fq->dev
237            pointer directly, device might already disappeared.
238          */
239         fq->q.fragments->dev = dev;
240         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
241 out:
242         if (dev)
243                 dev_put(dev);
244         spin_unlock(&fq->q.lock);
245         fq_put(fq);
246 }
247
248 static __inline__ struct frag_queue *
249 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
250         struct inet6_dev *idev)
251 {
252         struct inet_frag_queue *q;
253         struct ip6_create_arg arg;
254         unsigned int hash;
255
256         arg.id = id;
257         arg.src = src;
258         arg.dst = dst;
259         hash = ip6qhashfn(id, src, dst);
260
261         q = inet_frag_find(&ip6_frags, &arg, hash);
262         if (q == NULL)
263                 goto oom;
264
265         return container_of(q, struct frag_queue, q);
266
267 oom:
268         IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
269         return NULL;
270 }
271
272 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
273                            struct frag_hdr *fhdr, int nhoff)
274 {
275         struct sk_buff *prev, *next;
276         struct net_device *dev;
277         int offset, end;
278
279         if (fq->q.last_in & COMPLETE)
280                 goto err;
281
282         offset = ntohs(fhdr->frag_off) & ~0x7;
283         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
284                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
285
286         if ((unsigned int)end > IPV6_MAXPLEN) {
287                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
288                                  IPSTATS_MIB_INHDRERRORS);
289                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
290                                   ((u8 *)&fhdr->frag_off -
291                                    skb_network_header(skb)));
292                 return -1;
293         }
294
295         if (skb->ip_summed == CHECKSUM_COMPLETE) {
296                 const unsigned char *nh = skb_network_header(skb);
297                 skb->csum = csum_sub(skb->csum,
298                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
299                                                   0));
300         }
301
302         /* Is this the final fragment? */
303         if (!(fhdr->frag_off & htons(IP6_MF))) {
304                 /* If we already have some bits beyond end
305                  * or have different end, the segment is corrupted.
306                  */
307                 if (end < fq->q.len ||
308                     ((fq->q.last_in & LAST_IN) && end != fq->q.len))
309                         goto err;
310                 fq->q.last_in |= LAST_IN;
311                 fq->q.len = end;
312         } else {
313                 /* Check if the fragment is rounded to 8 bytes.
314                  * Required by the RFC.
315                  */
316                 if (end & 0x7) {
317                         /* RFC2460 says always send parameter problem in
318                          * this case. -DaveM
319                          */
320                         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
321                                          IPSTATS_MIB_INHDRERRORS);
322                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
323                                           offsetof(struct ipv6hdr, payload_len));
324                         return -1;
325                 }
326                 if (end > fq->q.len) {
327                         /* Some bits beyond end -> corruption. */
328                         if (fq->q.last_in & LAST_IN)
329                                 goto err;
330                         fq->q.len = end;
331                 }
332         }
333
334         if (end == offset)
335                 goto err;
336
337         /* Point into the IP datagram 'data' part. */
338         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
339                 goto err;
340
341         if (pskb_trim_rcsum(skb, end - offset))
342                 goto err;
343
344         /* Find out which fragments are in front and at the back of us
345          * in the chain of fragments so far.  We must know where to put
346          * this fragment, right?
347          */
348         prev = NULL;
349         for(next = fq->q.fragments; next != NULL; next = next->next) {
350                 if (FRAG6_CB(next)->offset >= offset)
351                         break;  /* bingo! */
352                 prev = next;
353         }
354
355         /* We found where to put this one.  Check for overlap with
356          * preceding fragment, and, if needed, align things so that
357          * any overlaps are eliminated.
358          */
359         if (prev) {
360                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
361
362                 if (i > 0) {
363                         offset += i;
364                         if (end <= offset)
365                                 goto err;
366                         if (!pskb_pull(skb, i))
367                                 goto err;
368                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
369                                 skb->ip_summed = CHECKSUM_NONE;
370                 }
371         }
372
373         /* Look for overlap with succeeding segments.
374          * If we can merge fragments, do it.
375          */
376         while (next && FRAG6_CB(next)->offset < end) {
377                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
378
379                 if (i < next->len) {
380                         /* Eat head of the next overlapped fragment
381                          * and leave the loop. The next ones cannot overlap.
382                          */
383                         if (!pskb_pull(next, i))
384                                 goto err;
385                         FRAG6_CB(next)->offset += i;    /* next fragment */
386                         fq->q.meat -= i;
387                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
388                                 next->ip_summed = CHECKSUM_NONE;
389                         break;
390                 } else {
391                         struct sk_buff *free_it = next;
392
393                         /* Old fragment is completely overridden with
394                          * new one drop it.
395                          */
396                         next = next->next;
397
398                         if (prev)
399                                 prev->next = next;
400                         else
401                                 fq->q.fragments = next;
402
403                         fq->q.meat -= free_it->len;
404                         frag_kfree_skb(free_it, NULL);
405                 }
406         }
407
408         FRAG6_CB(skb)->offset = offset;
409
410         /* Insert this fragment in the chain of fragments. */
411         skb->next = next;
412         if (prev)
413                 prev->next = skb;
414         else
415                 fq->q.fragments = skb;
416
417         dev = skb->dev;
418         if (dev) {
419                 fq->iif = dev->ifindex;
420                 skb->dev = NULL;
421         }
422         fq->q.stamp = skb->tstamp;
423         fq->q.meat += skb->len;
424         atomic_add(skb->truesize, &ip6_frags.mem);
425
426         /* The first fragment.
427          * nhoffset is obtained from the first fragment, of course.
428          */
429         if (offset == 0) {
430                 fq->nhoffset = nhoff;
431                 fq->q.last_in |= FIRST_IN;
432         }
433
434         if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
435                 return ip6_frag_reasm(fq, prev, dev);
436
437         write_lock(&ip6_frags.lock);
438         list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
439         write_unlock(&ip6_frags.lock);
440         return -1;
441
442 err:
443         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
444         kfree_skb(skb);
445         return -1;
446 }
447
448 /*
449  *      Check if this packet is complete.
450  *      Returns NULL on failure by any reason, and pointer
451  *      to current nexthdr field in reassembled frame.
452  *
453  *      It is called with locked fq, and caller must check that
454  *      queue is eligible for reassembly i.e. it is not COMPLETE,
455  *      the last and the first frames arrived and all the bits are here.
456  */
457 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
458                           struct net_device *dev)
459 {
460         struct sk_buff *fp, *head = fq->q.fragments;
461         int    payload_len;
462         unsigned int nhoff;
463
464         fq_kill(fq);
465
466         /* Make the one we just received the head. */
467         if (prev) {
468                 head = prev->next;
469                 fp = skb_clone(head, GFP_ATOMIC);
470
471                 if (!fp)
472                         goto out_oom;
473
474                 fp->next = head->next;
475                 prev->next = fp;
476
477                 skb_morph(head, fq->q.fragments);
478                 head->next = fq->q.fragments->next;
479
480                 kfree_skb(fq->q.fragments);
481                 fq->q.fragments = head;
482         }
483
484         BUG_TRAP(head != NULL);
485         BUG_TRAP(FRAG6_CB(head)->offset == 0);
486
487         /* Unfragmented part is taken from the first segment. */
488         payload_len = ((head->data - skb_network_header(head)) -
489                        sizeof(struct ipv6hdr) + fq->q.len -
490                        sizeof(struct frag_hdr));
491         if (payload_len > IPV6_MAXPLEN)
492                 goto out_oversize;
493
494         /* Head of list must not be cloned. */
495         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
496                 goto out_oom;
497
498         /* If the first fragment is fragmented itself, we split
499          * it to two chunks: the first with data and paged part
500          * and the second, holding only fragments. */
501         if (skb_shinfo(head)->frag_list) {
502                 struct sk_buff *clone;
503                 int i, plen = 0;
504
505                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
506                         goto out_oom;
507                 clone->next = head->next;
508                 head->next = clone;
509                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
510                 skb_shinfo(head)->frag_list = NULL;
511                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
512                         plen += skb_shinfo(head)->frags[i].size;
513                 clone->len = clone->data_len = head->data_len - plen;
514                 head->data_len -= clone->len;
515                 head->len -= clone->len;
516                 clone->csum = 0;
517                 clone->ip_summed = head->ip_summed;
518                 atomic_add(clone->truesize, &ip6_frags.mem);
519         }
520
521         /* We have to remove fragment header from datagram and to relocate
522          * header in order to calculate ICV correctly. */
523         nhoff = fq->nhoffset;
524         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
525         memmove(head->head + sizeof(struct frag_hdr), head->head,
526                 (head->data - head->head) - sizeof(struct frag_hdr));
527         head->mac_header += sizeof(struct frag_hdr);
528         head->network_header += sizeof(struct frag_hdr);
529
530         skb_shinfo(head)->frag_list = head->next;
531         skb_reset_transport_header(head);
532         skb_push(head, head->data - skb_network_header(head));
533         atomic_sub(head->truesize, &ip6_frags.mem);
534
535         for (fp=head->next; fp; fp = fp->next) {
536                 head->data_len += fp->len;
537                 head->len += fp->len;
538                 if (head->ip_summed != fp->ip_summed)
539                         head->ip_summed = CHECKSUM_NONE;
540                 else if (head->ip_summed == CHECKSUM_COMPLETE)
541                         head->csum = csum_add(head->csum, fp->csum);
542                 head->truesize += fp->truesize;
543                 atomic_sub(fp->truesize, &ip6_frags.mem);
544         }
545
546         head->next = NULL;
547         head->dev = dev;
548         head->tstamp = fq->q.stamp;
549         ipv6_hdr(head)->payload_len = htons(payload_len);
550         IP6CB(head)->nhoff = nhoff;
551
552         /* Yes, and fold redundant checksum back. 8) */
553         if (head->ip_summed == CHECKSUM_COMPLETE)
554                 head->csum = csum_partial(skb_network_header(head),
555                                           skb_network_header_len(head),
556                                           head->csum);
557
558         rcu_read_lock();
559         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
560         rcu_read_unlock();
561         fq->q.fragments = NULL;
562         return 1;
563
564 out_oversize:
565         if (net_ratelimit())
566                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
567         goto out_fail;
568 out_oom:
569         if (net_ratelimit())
570                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
571 out_fail:
572         rcu_read_lock();
573         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
574         rcu_read_unlock();
575         return -1;
576 }
577
578 static int ipv6_frag_rcv(struct sk_buff *skb)
579 {
580         struct frag_hdr *fhdr;
581         struct frag_queue *fq;
582         struct ipv6hdr *hdr = ipv6_hdr(skb);
583
584         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
585
586         /* Jumbo payload inhibits frag. header */
587         if (hdr->payload_len==0) {
588                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
589                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
590                                   skb_network_header_len(skb));
591                 return -1;
592         }
593         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
594                                  sizeof(struct frag_hdr)))) {
595                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
596                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
597                                   skb_network_header_len(skb));
598                 return -1;
599         }
600
601         hdr = ipv6_hdr(skb);
602         fhdr = (struct frag_hdr *)skb_transport_header(skb);
603
604         if (!(fhdr->frag_off & htons(0xFFF9))) {
605                 /* It is not a fragmented frame */
606                 skb->transport_header += sizeof(struct frag_hdr);
607                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
608
609                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
610                 return 1;
611         }
612
613         if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
614                 ip6_evictor(ip6_dst_idev(skb->dst));
615
616         if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
617                           ip6_dst_idev(skb->dst))) != NULL) {
618                 int ret;
619
620                 spin_lock(&fq->q.lock);
621
622                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
623
624                 spin_unlock(&fq->q.lock);
625                 fq_put(fq);
626                 return ret;
627         }
628
629         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
630         kfree_skb(skb);
631         return -1;
632 }
633
634 static struct inet6_protocol frag_protocol =
635 {
636         .handler        =       ipv6_frag_rcv,
637         .flags          =       INET6_PROTO_NOPOLICY,
638 };
639
640 void __init ipv6_frag_init(void)
641 {
642         if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
643                 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
644
645         ip6_frags.ctl = &ip6_frags_ctl;
646         ip6_frags.hashfn = ip6_hashfn;
647         ip6_frags.constructor = ip6_frag_init;
648         ip6_frags.destructor = ip6_frag_free;
649         ip6_frags.skb_free = NULL;
650         ip6_frags.qsize = sizeof(struct frag_queue);
651         ip6_frags.match = ip6_frag_match;
652         ip6_frags.frag_expire = ip6_frag_expire;
653         inet_frags_init(&ip6_frags);
654 }