138a8b36270694ec96ca4a21f51a55b0c1d62e98
[linux-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/slab.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 <net/netfilter/ipv6/nf_conntrack_ipv6.h>
43 #include <linux/sysctl.h>
44 #include <linux/netfilter.h>
45 #include <linux/netfilter_ipv6.h>
46 #include <linux/kernel.h>
47 #include <linux/module.h>
48
49
50 struct nf_ct_frag6_skb_cb
51 {
52         struct inet6_skb_parm   h;
53         int                     offset;
54         struct sk_buff          *orig;
55 };
56
57 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
58
59 struct nf_ct_frag6_queue
60 {
61         struct inet_frag_queue  q;
62
63         __be32                  id;             /* fragment id          */
64         u32                     user;
65         struct in6_addr         saddr;
66         struct in6_addr         daddr;
67
68         unsigned int            csum;
69         __u16                   nhoffset;
70 };
71
72 static struct inet_frags nf_frags;
73 static struct netns_frags nf_init_frags;
74
75 #ifdef CONFIG_SYSCTL
76 struct ctl_table nf_ct_ipv6_sysctl_table[] = {
77         {
78                 .procname       = "nf_conntrack_frag6_timeout",
79                 .data           = &nf_init_frags.timeout,
80                 .maxlen         = sizeof(unsigned int),
81                 .mode           = 0644,
82                 .proc_handler   = proc_dointvec_jiffies,
83         },
84         {
85                 .procname       = "nf_conntrack_frag6_low_thresh",
86                 .data           = &nf_init_frags.low_thresh,
87                 .maxlen         = sizeof(unsigned int),
88                 .mode           = 0644,
89                 .proc_handler   = proc_dointvec,
90         },
91         {
92                 .procname       = "nf_conntrack_frag6_high_thresh",
93                 .data           = &nf_init_frags.high_thresh,
94                 .maxlen         = sizeof(unsigned int),
95                 .mode           = 0644,
96                 .proc_handler   = proc_dointvec,
97         },
98         { }
99 };
100 #endif
101
102 static unsigned int nf_hashfn(struct inet_frag_queue *q)
103 {
104         const struct nf_ct_frag6_queue *nq;
105
106         nq = container_of(q, struct nf_ct_frag6_queue, q);
107         return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
108 }
109
110 static void nf_skb_free(struct sk_buff *skb)
111 {
112         if (NFCT_FRAG6_CB(skb)->orig)
113                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
114 }
115
116 /* Destruction primitives. */
117
118 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
119 {
120         inet_frag_put(&fq->q, &nf_frags);
121 }
122
123 /* Kill fq entry. It is not destroyed immediately,
124  * because caller (and someone more) holds reference count.
125  */
126 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
127 {
128         inet_frag_kill(&fq->q, &nf_frags);
129 }
130
131 static void nf_ct_frag6_evictor(void)
132 {
133         local_bh_disable();
134         inet_frag_evictor(&nf_init_frags, &nf_frags);
135         local_bh_enable();
136 }
137
138 static void nf_ct_frag6_expire(unsigned long data)
139 {
140         struct nf_ct_frag6_queue *fq;
141
142         fq = container_of((struct inet_frag_queue *)data,
143                         struct nf_ct_frag6_queue, q);
144
145         spin_lock(&fq->q.lock);
146
147         if (fq->q.last_in & INET_FRAG_COMPLETE)
148                 goto out;
149
150         fq_kill(fq);
151
152 out:
153         spin_unlock(&fq->q.lock);
154         fq_put(fq);
155 }
156
157 /* Creation primitives. */
158
159 static __inline__ struct nf_ct_frag6_queue *
160 fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
161 {
162         struct inet_frag_queue *q;
163         struct ip6_create_arg arg;
164         unsigned int hash;
165
166         arg.id = id;
167         arg.user = user;
168         arg.src = src;
169         arg.dst = dst;
170
171         read_lock_bh(&nf_frags.lock);
172         hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
173
174         q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
175         local_bh_enable();
176         if (q == NULL)
177                 goto oom;
178
179         return container_of(q, struct nf_ct_frag6_queue, q);
180
181 oom:
182         pr_debug("Can't alloc new queue\n");
183         return NULL;
184 }
185
186
187 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
188                              const struct frag_hdr *fhdr, int nhoff)
189 {
190         struct sk_buff *prev, *next;
191         int offset, end;
192
193         if (fq->q.last_in & INET_FRAG_COMPLETE) {
194                 pr_debug("Already completed\n");
195                 goto err;
196         }
197
198         offset = ntohs(fhdr->frag_off) & ~0x7;
199         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
200                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
201
202         if ((unsigned int)end > IPV6_MAXPLEN) {
203                 pr_debug("offset is too large.\n");
204                 return -1;
205         }
206
207         if (skb->ip_summed == CHECKSUM_COMPLETE) {
208                 const unsigned char *nh = skb_network_header(skb);
209                 skb->csum = csum_sub(skb->csum,
210                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
211                                                   0));
212         }
213
214         /* Is this the final fragment? */
215         if (!(fhdr->frag_off & htons(IP6_MF))) {
216                 /* If we already have some bits beyond end
217                  * or have different end, the segment is corrupted.
218                  */
219                 if (end < fq->q.len ||
220                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
221                         pr_debug("already received last fragment\n");
222                         goto err;
223                 }
224                 fq->q.last_in |= INET_FRAG_LAST_IN;
225                 fq->q.len = end;
226         } else {
227                 /* Check if the fragment is rounded to 8 bytes.
228                  * Required by the RFC.
229                  */
230                 if (end & 0x7) {
231                         /* RFC2460 says always send parameter problem in
232                          * this case. -DaveM
233                          */
234                         pr_debug("end of fragment not rounded to 8 bytes.\n");
235                         return -1;
236                 }
237                 if (end > fq->q.len) {
238                         /* Some bits beyond end -> corruption. */
239                         if (fq->q.last_in & INET_FRAG_LAST_IN) {
240                                 pr_debug("last packet already reached.\n");
241                                 goto err;
242                         }
243                         fq->q.len = end;
244                 }
245         }
246
247         if (end == offset)
248                 goto err;
249
250         /* Point into the IP datagram 'data' part. */
251         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
252                 pr_debug("queue: message is too short.\n");
253                 goto err;
254         }
255         if (pskb_trim_rcsum(skb, end - offset)) {
256                 pr_debug("Can't trim\n");
257                 goto err;
258         }
259
260         /* Find out which fragments are in front and at the back of us
261          * in the chain of fragments so far.  We must know where to put
262          * this fragment, right?
263          */
264         prev = fq->q.fragments_tail;
265         if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
266                 next = NULL;
267                 goto found;
268         }
269         prev = NULL;
270         for (next = fq->q.fragments; next != NULL; next = next->next) {
271                 if (NFCT_FRAG6_CB(next)->offset >= offset)
272                         break;  /* bingo! */
273                 prev = next;
274         }
275
276 found:
277         /* RFC5722, Section 4:
278          *                                  When reassembling an IPv6 datagram, if
279          *   one or more its constituent fragments is determined to be an
280          *   overlapping fragment, the entire datagram (and any constituent
281          *   fragments, including those not yet received) MUST be silently
282          *   discarded.
283          */
284
285         /* Check for overlap with preceding fragment. */
286         if (prev &&
287             (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset > 0)
288                 goto discard_fq;
289
290         /* Look for overlap with succeeding segment. */
291         if (next && NFCT_FRAG6_CB(next)->offset < end)
292                 goto discard_fq;
293
294         NFCT_FRAG6_CB(skb)->offset = offset;
295
296         /* Insert this fragment in the chain of fragments. */
297         skb->next = next;
298         if (!next)
299                 fq->q.fragments_tail = skb;
300         if (prev)
301                 prev->next = skb;
302         else
303                 fq->q.fragments = skb;
304
305         skb->dev = NULL;
306         fq->q.stamp = skb->tstamp;
307         fq->q.meat += skb->len;
308         atomic_add(skb->truesize, &nf_init_frags.mem);
309
310         /* The first fragment.
311          * nhoffset is obtained from the first fragment, of course.
312          */
313         if (offset == 0) {
314                 fq->nhoffset = nhoff;
315                 fq->q.last_in |= INET_FRAG_FIRST_IN;
316         }
317         write_lock(&nf_frags.lock);
318         list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
319         write_unlock(&nf_frags.lock);
320         return 0;
321
322 discard_fq:
323         fq_kill(fq);
324 err:
325         return -1;
326 }
327
328 /*
329  *      Check if this packet is complete.
330  *      Returns NULL on failure by any reason, and pointer
331  *      to current nexthdr field in reassembled frame.
332  *
333  *      It is called with locked fq, and caller must check that
334  *      queue is eligible for reassembly i.e. it is not COMPLETE,
335  *      the last and the first frames arrived and all the bits are here.
336  */
337 static struct sk_buff *
338 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
339 {
340         struct sk_buff *fp, *op, *head = fq->q.fragments;
341         int    payload_len;
342
343         fq_kill(fq);
344
345         WARN_ON(head == NULL);
346         WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
347
348         /* Unfragmented part is taken from the first segment. */
349         payload_len = ((head->data - skb_network_header(head)) -
350                        sizeof(struct ipv6hdr) + fq->q.len -
351                        sizeof(struct frag_hdr));
352         if (payload_len > IPV6_MAXPLEN) {
353                 pr_debug("payload len is too large.\n");
354                 goto out_oversize;
355         }
356
357         /* Head of list must not be cloned. */
358         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
359                 pr_debug("skb is cloned but can't expand head");
360                 goto out_oom;
361         }
362
363         /* If the first fragment is fragmented itself, we split
364          * it to two chunks: the first with data and paged part
365          * and the second, holding only fragments. */
366         if (skb_has_frag_list(head)) {
367                 struct sk_buff *clone;
368                 int i, plen = 0;
369
370                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
371                         pr_debug("Can't alloc skb\n");
372                         goto out_oom;
373                 }
374                 clone->next = head->next;
375                 head->next = clone;
376                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
377                 skb_frag_list_init(head);
378                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
379                         plen += skb_shinfo(head)->frags[i].size;
380                 clone->len = clone->data_len = head->data_len - plen;
381                 head->data_len -= clone->len;
382                 head->len -= clone->len;
383                 clone->csum = 0;
384                 clone->ip_summed = head->ip_summed;
385
386                 NFCT_FRAG6_CB(clone)->orig = NULL;
387                 atomic_add(clone->truesize, &nf_init_frags.mem);
388         }
389
390         /* We have to remove fragment header from datagram and to relocate
391          * header in order to calculate ICV correctly. */
392         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
393         memmove(head->head + sizeof(struct frag_hdr), head->head,
394                 (head->data - head->head) - sizeof(struct frag_hdr));
395         head->mac_header += sizeof(struct frag_hdr);
396         head->network_header += sizeof(struct frag_hdr);
397
398         skb_shinfo(head)->frag_list = head->next;
399         skb_reset_transport_header(head);
400         skb_push(head, head->data - skb_network_header(head));
401
402         for (fp=head->next; fp; fp = fp->next) {
403                 head->data_len += fp->len;
404                 head->len += fp->len;
405                 if (head->ip_summed != fp->ip_summed)
406                         head->ip_summed = CHECKSUM_NONE;
407                 else if (head->ip_summed == CHECKSUM_COMPLETE)
408                         head->csum = csum_add(head->csum, fp->csum);
409                 head->truesize += fp->truesize;
410         }
411         atomic_sub(head->truesize, &nf_init_frags.mem);
412
413         head->next = NULL;
414         head->dev = dev;
415         head->tstamp = fq->q.stamp;
416         ipv6_hdr(head)->payload_len = htons(payload_len);
417
418         /* Yes, and fold redundant checksum back. 8) */
419         if (head->ip_summed == CHECKSUM_COMPLETE)
420                 head->csum = csum_partial(skb_network_header(head),
421                                           skb_network_header_len(head),
422                                           head->csum);
423
424         fq->q.fragments = NULL;
425         fq->q.fragments_tail = NULL;
426
427         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
428         fp = skb_shinfo(head)->frag_list;
429         if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
430                 /* at above code, head skb is divided into two skbs. */
431                 fp = fp->next;
432
433         op = NFCT_FRAG6_CB(head)->orig;
434         for (; fp; fp = fp->next) {
435                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
436
437                 op->next = orig;
438                 op = orig;
439                 NFCT_FRAG6_CB(fp)->orig = NULL;
440         }
441
442         return head;
443
444 out_oversize:
445         if (net_ratelimit())
446                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
447         goto out_fail;
448 out_oom:
449         if (net_ratelimit())
450                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
451 out_fail:
452         return NULL;
453 }
454
455 /*
456  * find the header just before Fragment Header.
457  *
458  * if success return 0 and set ...
459  * (*prevhdrp): the value of "Next Header Field" in the header
460  *              just before Fragment Header.
461  * (*prevhoff): the offset of "Next Header Field" in the header
462  *              just before Fragment Header.
463  * (*fhoff)   : the offset of Fragment Header.
464  *
465  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
466  *
467  */
468 static int
469 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
470 {
471         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
472         const int netoff = skb_network_offset(skb);
473         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
474         int start = netoff + sizeof(struct ipv6hdr);
475         int len = skb->len - start;
476         u8 prevhdr = NEXTHDR_IPV6;
477
478         while (nexthdr != NEXTHDR_FRAGMENT) {
479                 struct ipv6_opt_hdr hdr;
480                 int hdrlen;
481
482                 if (!ipv6_ext_hdr(nexthdr)) {
483                         return -1;
484                 }
485                 if (nexthdr == NEXTHDR_NONE) {
486                         pr_debug("next header is none\n");
487                         return -1;
488                 }
489                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
490                         pr_debug("too short\n");
491                         return -1;
492                 }
493                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
494                         BUG();
495                 if (nexthdr == NEXTHDR_AUTH)
496                         hdrlen = (hdr.hdrlen+2)<<2;
497                 else
498                         hdrlen = ipv6_optlen(&hdr);
499
500                 prevhdr = nexthdr;
501                 prev_nhoff = start;
502
503                 nexthdr = hdr.nexthdr;
504                 len -= hdrlen;
505                 start += hdrlen;
506         }
507
508         if (len < 0)
509                 return -1;
510
511         *prevhdrp = prevhdr;
512         *prevhoff = prev_nhoff;
513         *fhoff = start;
514
515         return 0;
516 }
517
518 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
519 {
520         struct sk_buff *clone;
521         struct net_device *dev = skb->dev;
522         struct frag_hdr *fhdr;
523         struct nf_ct_frag6_queue *fq;
524         struct ipv6hdr *hdr;
525         int fhoff, nhoff;
526         u8 prevhdr;
527         struct sk_buff *ret_skb = NULL;
528
529         /* Jumbo payload inhibits frag. header */
530         if (ipv6_hdr(skb)->payload_len == 0) {
531                 pr_debug("payload len = 0\n");
532                 return skb;
533         }
534
535         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
536                 return skb;
537
538         clone = skb_clone(skb, GFP_ATOMIC);
539         if (clone == NULL) {
540                 pr_debug("Can't clone skb\n");
541                 return skb;
542         }
543
544         NFCT_FRAG6_CB(clone)->orig = skb;
545
546         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
547                 pr_debug("message is too short.\n");
548                 goto ret_orig;
549         }
550
551         skb_set_transport_header(clone, fhoff);
552         hdr = ipv6_hdr(clone);
553         fhdr = (struct frag_hdr *)skb_transport_header(clone);
554
555         if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
556                 nf_ct_frag6_evictor();
557
558         fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
559         if (fq == NULL) {
560                 pr_debug("Can't find and can't create new queue\n");
561                 goto ret_orig;
562         }
563
564         spin_lock_bh(&fq->q.lock);
565
566         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
567                 spin_unlock_bh(&fq->q.lock);
568                 pr_debug("Can't insert skb to queue\n");
569                 fq_put(fq);
570                 goto ret_orig;
571         }
572
573         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
574             fq->q.meat == fq->q.len) {
575                 ret_skb = nf_ct_frag6_reasm(fq, dev);
576                 if (ret_skb == NULL)
577                         pr_debug("Can't reassemble fragmented packets\n");
578         }
579         spin_unlock_bh(&fq->q.lock);
580
581         fq_put(fq);
582         return ret_skb;
583
584 ret_orig:
585         kfree_skb(clone);
586         return skb;
587 }
588
589 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
590                         struct net_device *in, struct net_device *out,
591                         int (*okfn)(struct sk_buff *))
592 {
593         struct sk_buff *s, *s2;
594
595         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
596                 nf_conntrack_put_reasm(s->nfct_reasm);
597                 nf_conntrack_get_reasm(skb);
598                 s->nfct_reasm = skb;
599
600                 s2 = s->next;
601                 s->next = NULL;
602
603                 NF_HOOK_THRESH(NFPROTO_IPV6, hooknum, s, in, out, okfn,
604                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
605                 s = s2;
606         }
607         nf_conntrack_put_reasm(skb);
608 }
609
610 int nf_ct_frag6_init(void)
611 {
612         nf_frags.hashfn = nf_hashfn;
613         nf_frags.constructor = ip6_frag_init;
614         nf_frags.destructor = NULL;
615         nf_frags.skb_free = nf_skb_free;
616         nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
617         nf_frags.match = ip6_frag_match;
618         nf_frags.frag_expire = nf_ct_frag6_expire;
619         nf_frags.secret_interval = 10 * 60 * HZ;
620         nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
621         nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
622         nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
623         inet_frags_init_net(&nf_init_frags);
624         inet_frags_init(&nf_frags);
625
626         return 0;
627 }
628
629 void nf_ct_frag6_cleanup(void)
630 {
631         inet_frags_fini(&nf_frags);
632
633         nf_init_frags.low_thresh = 0;
634         nf_ct_frag6_evictor();
635 }