Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
[sfrench/cifs-2.6.git] / net / ipv6 / netfilter / nf_conntrack_reasm.c
1 /*
2  * IPv6 fragment reassembly for connection tracking
3  *
4  * Copyright (C)2004 USAGI/WIDE Project
5  *
6  * Author:
7  *      Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
8  *
9  * Based on: net/ipv6/reassembly.c
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public License
13  * as published by the Free Software Foundation; either version
14  * 2 of the License, or (at your option) any later version.
15  */
16
17 #include <linux/errno.h>
18 #include <linux/types.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/jiffies.h>
23 #include <linux/net.h>
24 #include <linux/list.h>
25 #include <linux/netdevice.h>
26 #include <linux/in6.h>
27 #include <linux/ipv6.h>
28 #include <linux/icmpv6.h>
29 #include <linux/random.h>
30 #include <linux/jhash.h>
31
32 #include <net/sock.h>
33 #include <net/snmp.h>
34 #include <net/inet_frag.h>
35
36 #include <net/ipv6.h>
37 #include <net/protocol.h>
38 #include <net/transp_v6.h>
39 #include <net/rawv6.h>
40 #include <net/ndisc.h>
41 #include <net/addrconf.h>
42 #include <linux/sysctl.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 #include <linux/kernel.h>
46 #include <linux/module.h>
47
48 #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
49 #define NF_CT_FRAG6_LOW_THRESH 196608  /* == 192*1024 */
50 #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
51
52 struct nf_ct_frag6_skb_cb
53 {
54         struct inet6_skb_parm   h;
55         int                     offset;
56         struct sk_buff          *orig;
57 };
58
59 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
60
61 struct nf_ct_frag6_queue
62 {
63         struct inet_frag_queue  q;
64
65         __be32                  id;             /* fragment id          */
66         struct in6_addr         saddr;
67         struct in6_addr         daddr;
68
69         unsigned int            csum;
70         __u16                   nhoffset;
71 };
72
73 struct inet_frags_ctl nf_frags_ctl __read_mostly = {
74         .high_thresh     = 256 * 1024,
75         .low_thresh      = 192 * 1024,
76         .timeout         = IPV6_FRAG_TIMEOUT,
77         .secret_interval = 10 * 60 * HZ,
78 };
79
80 static struct inet_frags nf_frags;
81
82 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
83                                struct in6_addr *daddr)
84 {
85         u32 a, b, c;
86
87         a = (__force u32)saddr->s6_addr32[0];
88         b = (__force u32)saddr->s6_addr32[1];
89         c = (__force u32)saddr->s6_addr32[2];
90
91         a += JHASH_GOLDEN_RATIO;
92         b += JHASH_GOLDEN_RATIO;
93         c += nf_frags.rnd;
94         __jhash_mix(a, b, c);
95
96         a += (__force u32)saddr->s6_addr32[3];
97         b += (__force u32)daddr->s6_addr32[0];
98         c += (__force u32)daddr->s6_addr32[1];
99         __jhash_mix(a, b, c);
100
101         a += (__force u32)daddr->s6_addr32[2];
102         b += (__force u32)daddr->s6_addr32[3];
103         c += (__force u32)id;
104         __jhash_mix(a, b, c);
105
106         return c & (INETFRAGS_HASHSZ - 1);
107 }
108
109 static unsigned int nf_hashfn(struct inet_frag_queue *q)
110 {
111         struct nf_ct_frag6_queue *nq;
112
113         nq = container_of(q, struct nf_ct_frag6_queue, q);
114         return ip6qhashfn(nq->id, &nq->saddr, &nq->daddr);
115 }
116
117 static void nf_skb_free(struct sk_buff *skb)
118 {
119         if (NFCT_FRAG6_CB(skb)->orig)
120                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
121 }
122
123 /* Memory Tracking Functions. */
124 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
125 {
126         if (work)
127                 *work -= skb->truesize;
128         atomic_sub(skb->truesize, &nf_frags.mem);
129         nf_skb_free(skb);
130         kfree_skb(skb);
131 }
132
133 /* Destruction primitives. */
134
135 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
136 {
137         inet_frag_put(&fq->q, &nf_frags);
138 }
139
140 /* Kill fq entry. It is not destroyed immediately,
141  * because caller (and someone more) holds reference count.
142  */
143 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
144 {
145         inet_frag_kill(&fq->q, &nf_frags);
146 }
147
148 static void nf_ct_frag6_evictor(void)
149 {
150         inet_frag_evictor(&nf_frags);
151 }
152
153 static void nf_ct_frag6_expire(unsigned long data)
154 {
155         struct nf_ct_frag6_queue *fq;
156
157         fq = container_of((struct inet_frag_queue *)data,
158                         struct nf_ct_frag6_queue, q);
159
160         spin_lock(&fq->q.lock);
161
162         if (fq->q.last_in & COMPLETE)
163                 goto out;
164
165         fq_kill(fq);
166
167 out:
168         spin_unlock(&fq->q.lock);
169         fq_put(fq);
170 }
171
172 /* Creation primitives. */
173
174 static __inline__ struct nf_ct_frag6_queue *
175 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
176 {
177         struct inet_frag_queue *q;
178         struct ip6_create_arg arg;
179         unsigned int hash;
180
181         arg.id = id;
182         arg.src = src;
183         arg.dst = dst;
184         hash = ip6qhashfn(id, src, dst);
185
186         q = inet_frag_find(&nf_frags, &arg, hash);
187         if (q == NULL)
188                 goto oom;
189
190         return container_of(q, struct nf_ct_frag6_queue, q);
191
192 oom:
193         pr_debug("Can't alloc new queue\n");
194         return NULL;
195 }
196
197
198 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
199                              struct frag_hdr *fhdr, int nhoff)
200 {
201         struct sk_buff *prev, *next;
202         int offset, end;
203
204         if (fq->q.last_in & COMPLETE) {
205                 pr_debug("Allready completed\n");
206                 goto err;
207         }
208
209         offset = ntohs(fhdr->frag_off) & ~0x7;
210         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
211                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
212
213         if ((unsigned int)end > IPV6_MAXPLEN) {
214                 pr_debug("offset is too large.\n");
215                 return -1;
216         }
217
218         if (skb->ip_summed == CHECKSUM_COMPLETE) {
219                 const unsigned char *nh = skb_network_header(skb);
220                 skb->csum = csum_sub(skb->csum,
221                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
222                                                   0));
223         }
224
225         /* Is this the final fragment? */
226         if (!(fhdr->frag_off & htons(IP6_MF))) {
227                 /* If we already have some bits beyond end
228                  * or have different end, the segment is corrupted.
229                  */
230                 if (end < fq->q.len ||
231                     ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
232                         pr_debug("already received last fragment\n");
233                         goto err;
234                 }
235                 fq->q.last_in |= LAST_IN;
236                 fq->q.len = end;
237         } else {
238                 /* Check if the fragment is rounded to 8 bytes.
239                  * Required by the RFC.
240                  */
241                 if (end & 0x7) {
242                         /* RFC2460 says always send parameter problem in
243                          * this case. -DaveM
244                          */
245                         pr_debug("end of fragment not rounded to 8 bytes.\n");
246                         return -1;
247                 }
248                 if (end > fq->q.len) {
249                         /* Some bits beyond end -> corruption. */
250                         if (fq->q.last_in & LAST_IN) {
251                                 pr_debug("last packet already reached.\n");
252                                 goto err;
253                         }
254                         fq->q.len = end;
255                 }
256         }
257
258         if (end == offset)
259                 goto err;
260
261         /* Point into the IP datagram 'data' part. */
262         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
263                 pr_debug("queue: message is too short.\n");
264                 goto err;
265         }
266         if (pskb_trim_rcsum(skb, end - offset)) {
267                 pr_debug("Can't trim\n");
268                 goto err;
269         }
270
271         /* Find out which fragments are in front and at the back of us
272          * in the chain of fragments so far.  We must know where to put
273          * this fragment, right?
274          */
275         prev = NULL;
276         for (next = fq->q.fragments; next != NULL; next = next->next) {
277                 if (NFCT_FRAG6_CB(next)->offset >= offset)
278                         break;  /* bingo! */
279                 prev = next;
280         }
281
282         /* We found where to put this one.  Check for overlap with
283          * preceding fragment, and, if needed, align things so that
284          * any overlaps are eliminated.
285          */
286         if (prev) {
287                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
288
289                 if (i > 0) {
290                         offset += i;
291                         if (end <= offset) {
292                                 pr_debug("overlap\n");
293                                 goto err;
294                         }
295                         if (!pskb_pull(skb, i)) {
296                                 pr_debug("Can't pull\n");
297                                 goto err;
298                         }
299                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
300                                 skb->ip_summed = CHECKSUM_NONE;
301                 }
302         }
303
304         /* Look for overlap with succeeding segments.
305          * If we can merge fragments, do it.
306          */
307         while (next && NFCT_FRAG6_CB(next)->offset < end) {
308                 /* overlap is 'i' bytes */
309                 int i = end - NFCT_FRAG6_CB(next)->offset;
310
311                 if (i < next->len) {
312                         /* Eat head of the next overlapped fragment
313                          * and leave the loop. The next ones cannot overlap.
314                          */
315                         pr_debug("Eat head of the overlapped parts.: %d", i);
316                         if (!pskb_pull(next, i))
317                                 goto err;
318
319                         /* next fragment */
320                         NFCT_FRAG6_CB(next)->offset += i;
321                         fq->q.meat -= i;
322                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
323                                 next->ip_summed = CHECKSUM_NONE;
324                         break;
325                 } else {
326                         struct sk_buff *free_it = next;
327
328                         /* Old fragmnet is completely overridden with
329                          * new one drop it.
330                          */
331                         next = next->next;
332
333                         if (prev)
334                                 prev->next = next;
335                         else
336                                 fq->q.fragments = next;
337
338                         fq->q.meat -= free_it->len;
339                         frag_kfree_skb(free_it, NULL);
340                 }
341         }
342
343         NFCT_FRAG6_CB(skb)->offset = offset;
344
345         /* Insert this fragment in the chain of fragments. */
346         skb->next = next;
347         if (prev)
348                 prev->next = skb;
349         else
350                 fq->q.fragments = skb;
351
352         skb->dev = NULL;
353         fq->q.stamp = skb->tstamp;
354         fq->q.meat += skb->len;
355         atomic_add(skb->truesize, &nf_frags.mem);
356
357         /* The first fragment.
358          * nhoffset is obtained from the first fragment, of course.
359          */
360         if (offset == 0) {
361                 fq->nhoffset = nhoff;
362                 fq->q.last_in |= FIRST_IN;
363         }
364         write_lock(&nf_frags.lock);
365         list_move_tail(&fq->q.lru_list, &nf_frags.lru_list);
366         write_unlock(&nf_frags.lock);
367         return 0;
368
369 err:
370         return -1;
371 }
372
373 /*
374  *      Check if this packet is complete.
375  *      Returns NULL on failure by any reason, and pointer
376  *      to current nexthdr field in reassembled frame.
377  *
378  *      It is called with locked fq, and caller must check that
379  *      queue is eligible for reassembly i.e. it is not COMPLETE,
380  *      the last and the first frames arrived and all the bits are here.
381  */
382 static struct sk_buff *
383 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
384 {
385         struct sk_buff *fp, *op, *head = fq->q.fragments;
386         int    payload_len;
387
388         fq_kill(fq);
389
390         BUG_TRAP(head != NULL);
391         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
392
393         /* Unfragmented part is taken from the first segment. */
394         payload_len = ((head->data - skb_network_header(head)) -
395                        sizeof(struct ipv6hdr) + fq->q.len -
396                        sizeof(struct frag_hdr));
397         if (payload_len > IPV6_MAXPLEN) {
398                 pr_debug("payload len is too large.\n");
399                 goto out_oversize;
400         }
401
402         /* Head of list must not be cloned. */
403         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
404                 pr_debug("skb is cloned but can't expand head");
405                 goto out_oom;
406         }
407
408         /* If the first fragment is fragmented itself, we split
409          * it to two chunks: the first with data and paged part
410          * and the second, holding only fragments. */
411         if (skb_shinfo(head)->frag_list) {
412                 struct sk_buff *clone;
413                 int i, plen = 0;
414
415                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
416                         pr_debug("Can't alloc skb\n");
417                         goto out_oom;
418                 }
419                 clone->next = head->next;
420                 head->next = clone;
421                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
422                 skb_shinfo(head)->frag_list = NULL;
423                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
424                         plen += skb_shinfo(head)->frags[i].size;
425                 clone->len = clone->data_len = head->data_len - plen;
426                 head->data_len -= clone->len;
427                 head->len -= clone->len;
428                 clone->csum = 0;
429                 clone->ip_summed = head->ip_summed;
430
431                 NFCT_FRAG6_CB(clone)->orig = NULL;
432                 atomic_add(clone->truesize, &nf_frags.mem);
433         }
434
435         /* We have to remove fragment header from datagram and to relocate
436          * header in order to calculate ICV correctly. */
437         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
438         memmove(head->head + sizeof(struct frag_hdr), head->head,
439                 (head->data - head->head) - sizeof(struct frag_hdr));
440         head->mac_header += sizeof(struct frag_hdr);
441         head->network_header += sizeof(struct frag_hdr);
442
443         skb_shinfo(head)->frag_list = head->next;
444         skb_reset_transport_header(head);
445         skb_push(head, head->data - skb_network_header(head));
446         atomic_sub(head->truesize, &nf_frags.mem);
447
448         for (fp=head->next; fp; fp = fp->next) {
449                 head->data_len += fp->len;
450                 head->len += fp->len;
451                 if (head->ip_summed != fp->ip_summed)
452                         head->ip_summed = CHECKSUM_NONE;
453                 else if (head->ip_summed == CHECKSUM_COMPLETE)
454                         head->csum = csum_add(head->csum, fp->csum);
455                 head->truesize += fp->truesize;
456                 atomic_sub(fp->truesize, &nf_frags.mem);
457         }
458
459         head->next = NULL;
460         head->dev = dev;
461         head->tstamp = fq->q.stamp;
462         ipv6_hdr(head)->payload_len = htons(payload_len);
463
464         /* Yes, and fold redundant checksum back. 8) */
465         if (head->ip_summed == CHECKSUM_COMPLETE)
466                 head->csum = csum_partial(skb_network_header(head),
467                                           skb_network_header_len(head),
468                                           head->csum);
469
470         fq->q.fragments = NULL;
471
472         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
473         fp = skb_shinfo(head)->frag_list;
474         if (NFCT_FRAG6_CB(fp)->orig == NULL)
475                 /* at above code, head skb is divided into two skbs. */
476                 fp = fp->next;
477
478         op = NFCT_FRAG6_CB(head)->orig;
479         for (; fp; fp = fp->next) {
480                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
481
482                 op->next = orig;
483                 op = orig;
484                 NFCT_FRAG6_CB(fp)->orig = NULL;
485         }
486
487         return head;
488
489 out_oversize:
490         if (net_ratelimit())
491                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
492         goto out_fail;
493 out_oom:
494         if (net_ratelimit())
495                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
496 out_fail:
497         return NULL;
498 }
499
500 /*
501  * find the header just before Fragment Header.
502  *
503  * if success return 0 and set ...
504  * (*prevhdrp): the value of "Next Header Field" in the header
505  *              just before Fragment Header.
506  * (*prevhoff): the offset of "Next Header Field" in the header
507  *              just before Fragment Header.
508  * (*fhoff)   : the offset of Fragment Header.
509  *
510  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
511  *
512  */
513 static int
514 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
515 {
516         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
517         const int netoff = skb_network_offset(skb);
518         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
519         int start = netoff + sizeof(struct ipv6hdr);
520         int len = skb->len - start;
521         u8 prevhdr = NEXTHDR_IPV6;
522
523         while (nexthdr != NEXTHDR_FRAGMENT) {
524                 struct ipv6_opt_hdr hdr;
525                 int hdrlen;
526
527                 if (!ipv6_ext_hdr(nexthdr)) {
528                         return -1;
529                 }
530                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
531                         pr_debug("too short\n");
532                         return -1;
533                 }
534                 if (nexthdr == NEXTHDR_NONE) {
535                         pr_debug("next header is none\n");
536                         return -1;
537                 }
538                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
539                         BUG();
540                 if (nexthdr == NEXTHDR_AUTH)
541                         hdrlen = (hdr.hdrlen+2)<<2;
542                 else
543                         hdrlen = ipv6_optlen(&hdr);
544
545                 prevhdr = nexthdr;
546                 prev_nhoff = start;
547
548                 nexthdr = hdr.nexthdr;
549                 len -= hdrlen;
550                 start += hdrlen;
551         }
552
553         if (len < 0)
554                 return -1;
555
556         *prevhdrp = prevhdr;
557         *prevhoff = prev_nhoff;
558         *fhoff = start;
559
560         return 0;
561 }
562
563 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
564 {
565         struct sk_buff *clone;
566         struct net_device *dev = skb->dev;
567         struct frag_hdr *fhdr;
568         struct nf_ct_frag6_queue *fq;
569         struct ipv6hdr *hdr;
570         int fhoff, nhoff;
571         u8 prevhdr;
572         struct sk_buff *ret_skb = NULL;
573
574         /* Jumbo payload inhibits frag. header */
575         if (ipv6_hdr(skb)->payload_len == 0) {
576                 pr_debug("payload len = 0\n");
577                 return skb;
578         }
579
580         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
581                 return skb;
582
583         clone = skb_clone(skb, GFP_ATOMIC);
584         if (clone == NULL) {
585                 pr_debug("Can't clone skb\n");
586                 return skb;
587         }
588
589         NFCT_FRAG6_CB(clone)->orig = skb;
590
591         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
592                 pr_debug("message is too short.\n");
593                 goto ret_orig;
594         }
595
596         skb_set_transport_header(clone, fhoff);
597         hdr = ipv6_hdr(clone);
598         fhdr = (struct frag_hdr *)skb_transport_header(clone);
599
600         if (!(fhdr->frag_off & htons(0xFFF9))) {
601                 pr_debug("Invalid fragment offset\n");
602                 /* It is not a fragmented frame */
603                 goto ret_orig;
604         }
605
606         if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh)
607                 nf_ct_frag6_evictor();
608
609         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
610         if (fq == NULL) {
611                 pr_debug("Can't find and can't create new queue\n");
612                 goto ret_orig;
613         }
614
615         spin_lock(&fq->q.lock);
616
617         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
618                 spin_unlock(&fq->q.lock);
619                 pr_debug("Can't insert skb to queue\n");
620                 fq_put(fq);
621                 goto ret_orig;
622         }
623
624         if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
625                 ret_skb = nf_ct_frag6_reasm(fq, dev);
626                 if (ret_skb == NULL)
627                         pr_debug("Can't reassemble fragmented packets\n");
628         }
629         spin_unlock(&fq->q.lock);
630
631         fq_put(fq);
632         return ret_skb;
633
634 ret_orig:
635         kfree_skb(clone);
636         return skb;
637 }
638
639 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
640                         struct net_device *in, struct net_device *out,
641                         int (*okfn)(struct sk_buff *))
642 {
643         struct sk_buff *s, *s2;
644
645         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
646                 nf_conntrack_put_reasm(s->nfct_reasm);
647                 nf_conntrack_get_reasm(skb);
648                 s->nfct_reasm = skb;
649
650                 s2 = s->next;
651                 s->next = NULL;
652
653                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
654                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
655                 s = s2;
656         }
657         nf_conntrack_put_reasm(skb);
658 }
659
660 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
661 {
662         struct sk_buff *s, *s2;
663
664         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
665
666                 s2 = s->next;
667                 kfree_skb(s);
668         }
669
670         kfree_skb(skb);
671
672         return 0;
673 }
674
675 int nf_ct_frag6_init(void)
676 {
677         nf_frags.ctl = &nf_frags_ctl;
678         nf_frags.hashfn = nf_hashfn;
679         nf_frags.constructor = ip6_frag_init;
680         nf_frags.destructor = NULL;
681         nf_frags.skb_free = nf_skb_free;
682         nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
683         nf_frags.match = ip6_frag_match;
684         nf_frags.frag_expire = nf_ct_frag6_expire;
685         inet_frags_init(&nf_frags);
686
687         return 0;
688 }
689
690 void nf_ct_frag6_cleanup(void)
691 {
692         inet_frags_fini(&nf_frags);
693
694         nf_frags_ctl.low_thresh = 0;
695         nf_ct_frag6_evictor();
696 }