tproxy: split off ipv6 defragmentation to a separate module
[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_frag6_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
101 static struct ctl_table_header *nf_ct_frag6_sysctl_header;
102 #endif
103
104 static unsigned int nf_hashfn(struct inet_frag_queue *q)
105 {
106         const struct nf_ct_frag6_queue *nq;
107
108         nq = container_of(q, struct nf_ct_frag6_queue, q);
109         return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
110 }
111
112 static void nf_skb_free(struct sk_buff *skb)
113 {
114         if (NFCT_FRAG6_CB(skb)->orig)
115                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
116 }
117
118 /* Destruction primitives. */
119
120 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
121 {
122         inet_frag_put(&fq->q, &nf_frags);
123 }
124
125 /* Kill fq entry. It is not destroyed immediately,
126  * because caller (and someone more) holds reference count.
127  */
128 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
129 {
130         inet_frag_kill(&fq->q, &nf_frags);
131 }
132
133 static void nf_ct_frag6_evictor(void)
134 {
135         local_bh_disable();
136         inet_frag_evictor(&nf_init_frags, &nf_frags);
137         local_bh_enable();
138 }
139
140 static void nf_ct_frag6_expire(unsigned long data)
141 {
142         struct nf_ct_frag6_queue *fq;
143
144         fq = container_of((struct inet_frag_queue *)data,
145                         struct nf_ct_frag6_queue, q);
146
147         spin_lock(&fq->q.lock);
148
149         if (fq->q.last_in & INET_FRAG_COMPLETE)
150                 goto out;
151
152         fq_kill(fq);
153
154 out:
155         spin_unlock(&fq->q.lock);
156         fq_put(fq);
157 }
158
159 /* Creation primitives. */
160
161 static __inline__ struct nf_ct_frag6_queue *
162 fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
163 {
164         struct inet_frag_queue *q;
165         struct ip6_create_arg arg;
166         unsigned int hash;
167
168         arg.id = id;
169         arg.user = user;
170         arg.src = src;
171         arg.dst = dst;
172
173         read_lock_bh(&nf_frags.lock);
174         hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
175
176         q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
177         local_bh_enable();
178         if (q == NULL)
179                 goto oom;
180
181         return container_of(q, struct nf_ct_frag6_queue, q);
182
183 oom:
184         pr_debug("Can't alloc new queue\n");
185         return NULL;
186 }
187
188
189 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
190                              const struct frag_hdr *fhdr, int nhoff)
191 {
192         struct sk_buff *prev, *next;
193         int offset, end;
194
195         if (fq->q.last_in & INET_FRAG_COMPLETE) {
196                 pr_debug("Already completed\n");
197                 goto err;
198         }
199
200         offset = ntohs(fhdr->frag_off) & ~0x7;
201         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
202                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
203
204         if ((unsigned int)end > IPV6_MAXPLEN) {
205                 pr_debug("offset is too large.\n");
206                 return -1;
207         }
208
209         if (skb->ip_summed == CHECKSUM_COMPLETE) {
210                 const unsigned char *nh = skb_network_header(skb);
211                 skb->csum = csum_sub(skb->csum,
212                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
213                                                   0));
214         }
215
216         /* Is this the final fragment? */
217         if (!(fhdr->frag_off & htons(IP6_MF))) {
218                 /* If we already have some bits beyond end
219                  * or have different end, the segment is corrupted.
220                  */
221                 if (end < fq->q.len ||
222                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
223                         pr_debug("already received last fragment\n");
224                         goto err;
225                 }
226                 fq->q.last_in |= INET_FRAG_LAST_IN;
227                 fq->q.len = end;
228         } else {
229                 /* Check if the fragment is rounded to 8 bytes.
230                  * Required by the RFC.
231                  */
232                 if (end & 0x7) {
233                         /* RFC2460 says always send parameter problem in
234                          * this case. -DaveM
235                          */
236                         pr_debug("end of fragment not rounded to 8 bytes.\n");
237                         return -1;
238                 }
239                 if (end > fq->q.len) {
240                         /* Some bits beyond end -> corruption. */
241                         if (fq->q.last_in & INET_FRAG_LAST_IN) {
242                                 pr_debug("last packet already reached.\n");
243                                 goto err;
244                         }
245                         fq->q.len = end;
246                 }
247         }
248
249         if (end == offset)
250                 goto err;
251
252         /* Point into the IP datagram 'data' part. */
253         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
254                 pr_debug("queue: message is too short.\n");
255                 goto err;
256         }
257         if (pskb_trim_rcsum(skb, end - offset)) {
258                 pr_debug("Can't trim\n");
259                 goto err;
260         }
261
262         /* Find out which fragments are in front and at the back of us
263          * in the chain of fragments so far.  We must know where to put
264          * this fragment, right?
265          */
266         prev = fq->q.fragments_tail;
267         if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
268                 next = NULL;
269                 goto found;
270         }
271         prev = NULL;
272         for (next = fq->q.fragments; next != NULL; next = next->next) {
273                 if (NFCT_FRAG6_CB(next)->offset >= offset)
274                         break;  /* bingo! */
275                 prev = next;
276         }
277
278 found:
279         /* RFC5722, Section 4:
280          *                                  When reassembling an IPv6 datagram, if
281          *   one or more its constituent fragments is determined to be an
282          *   overlapping fragment, the entire datagram (and any constituent
283          *   fragments, including those not yet received) MUST be silently
284          *   discarded.
285          */
286
287         /* Check for overlap with preceding fragment. */
288         if (prev &&
289             (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset > 0)
290                 goto discard_fq;
291
292         /* Look for overlap with succeeding segment. */
293         if (next && NFCT_FRAG6_CB(next)->offset < end)
294                 goto discard_fq;
295
296         NFCT_FRAG6_CB(skb)->offset = offset;
297
298         /* Insert this fragment in the chain of fragments. */
299         skb->next = next;
300         if (!next)
301                 fq->q.fragments_tail = skb;
302         if (prev)
303                 prev->next = skb;
304         else
305                 fq->q.fragments = skb;
306
307         skb->dev = NULL;
308         fq->q.stamp = skb->tstamp;
309         fq->q.meat += skb->len;
310         atomic_add(skb->truesize, &nf_init_frags.mem);
311
312         /* The first fragment.
313          * nhoffset is obtained from the first fragment, of course.
314          */
315         if (offset == 0) {
316                 fq->nhoffset = nhoff;
317                 fq->q.last_in |= INET_FRAG_FIRST_IN;
318         }
319         write_lock(&nf_frags.lock);
320         list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
321         write_unlock(&nf_frags.lock);
322         return 0;
323
324 discard_fq:
325         fq_kill(fq);
326 err:
327         return -1;
328 }
329
330 /*
331  *      Check if this packet is complete.
332  *      Returns NULL on failure by any reason, and pointer
333  *      to current nexthdr field in reassembled frame.
334  *
335  *      It is called with locked fq, and caller must check that
336  *      queue is eligible for reassembly i.e. it is not COMPLETE,
337  *      the last and the first frames arrived and all the bits are here.
338  */
339 static struct sk_buff *
340 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
341 {
342         struct sk_buff *fp, *op, *head = fq->q.fragments;
343         int    payload_len;
344
345         fq_kill(fq);
346
347         WARN_ON(head == NULL);
348         WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
349
350         /* Unfragmented part is taken from the first segment. */
351         payload_len = ((head->data - skb_network_header(head)) -
352                        sizeof(struct ipv6hdr) + fq->q.len -
353                        sizeof(struct frag_hdr));
354         if (payload_len > IPV6_MAXPLEN) {
355                 pr_debug("payload len is too large.\n");
356                 goto out_oversize;
357         }
358
359         /* Head of list must not be cloned. */
360         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
361                 pr_debug("skb is cloned but can't expand head");
362                 goto out_oom;
363         }
364
365         /* If the first fragment is fragmented itself, we split
366          * it to two chunks: the first with data and paged part
367          * and the second, holding only fragments. */
368         if (skb_has_frag_list(head)) {
369                 struct sk_buff *clone;
370                 int i, plen = 0;
371
372                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
373                         pr_debug("Can't alloc skb\n");
374                         goto out_oom;
375                 }
376                 clone->next = head->next;
377                 head->next = clone;
378                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
379                 skb_frag_list_init(head);
380                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
381                         plen += skb_shinfo(head)->frags[i].size;
382                 clone->len = clone->data_len = head->data_len - plen;
383                 head->data_len -= clone->len;
384                 head->len -= clone->len;
385                 clone->csum = 0;
386                 clone->ip_summed = head->ip_summed;
387
388                 NFCT_FRAG6_CB(clone)->orig = NULL;
389                 atomic_add(clone->truesize, &nf_init_frags.mem);
390         }
391
392         /* We have to remove fragment header from datagram and to relocate
393          * header in order to calculate ICV correctly. */
394         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
395         memmove(head->head + sizeof(struct frag_hdr), head->head,
396                 (head->data - head->head) - sizeof(struct frag_hdr));
397         head->mac_header += sizeof(struct frag_hdr);
398         head->network_header += sizeof(struct frag_hdr);
399
400         skb_shinfo(head)->frag_list = head->next;
401         skb_reset_transport_header(head);
402         skb_push(head, head->data - skb_network_header(head));
403
404         for (fp=head->next; fp; fp = fp->next) {
405                 head->data_len += fp->len;
406                 head->len += fp->len;
407                 if (head->ip_summed != fp->ip_summed)
408                         head->ip_summed = CHECKSUM_NONE;
409                 else if (head->ip_summed == CHECKSUM_COMPLETE)
410                         head->csum = csum_add(head->csum, fp->csum);
411                 head->truesize += fp->truesize;
412         }
413         atomic_sub(head->truesize, &nf_init_frags.mem);
414
415         head->next = NULL;
416         head->dev = dev;
417         head->tstamp = fq->q.stamp;
418         ipv6_hdr(head)->payload_len = htons(payload_len);
419
420         /* Yes, and fold redundant checksum back. 8) */
421         if (head->ip_summed == CHECKSUM_COMPLETE)
422                 head->csum = csum_partial(skb_network_header(head),
423                                           skb_network_header_len(head),
424                                           head->csum);
425
426         fq->q.fragments = NULL;
427         fq->q.fragments_tail = NULL;
428
429         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
430         fp = skb_shinfo(head)->frag_list;
431         if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
432                 /* at above code, head skb is divided into two skbs. */
433                 fp = fp->next;
434
435         op = NFCT_FRAG6_CB(head)->orig;
436         for (; fp; fp = fp->next) {
437                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
438
439                 op->next = orig;
440                 op = orig;
441                 NFCT_FRAG6_CB(fp)->orig = NULL;
442         }
443
444         return head;
445
446 out_oversize:
447         if (net_ratelimit())
448                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
449         goto out_fail;
450 out_oom:
451         if (net_ratelimit())
452                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
453 out_fail:
454         return NULL;
455 }
456
457 /*
458  * find the header just before Fragment Header.
459  *
460  * if success return 0 and set ...
461  * (*prevhdrp): the value of "Next Header Field" in the header
462  *              just before Fragment Header.
463  * (*prevhoff): the offset of "Next Header Field" in the header
464  *              just before Fragment Header.
465  * (*fhoff)   : the offset of Fragment Header.
466  *
467  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
468  *
469  */
470 static int
471 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
472 {
473         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
474         const int netoff = skb_network_offset(skb);
475         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
476         int start = netoff + sizeof(struct ipv6hdr);
477         int len = skb->len - start;
478         u8 prevhdr = NEXTHDR_IPV6;
479
480         while (nexthdr != NEXTHDR_FRAGMENT) {
481                 struct ipv6_opt_hdr hdr;
482                 int hdrlen;
483
484                 if (!ipv6_ext_hdr(nexthdr)) {
485                         return -1;
486                 }
487                 if (nexthdr == NEXTHDR_NONE) {
488                         pr_debug("next header is none\n");
489                         return -1;
490                 }
491                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
492                         pr_debug("too short\n");
493                         return -1;
494                 }
495                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
496                         BUG();
497                 if (nexthdr == NEXTHDR_AUTH)
498                         hdrlen = (hdr.hdrlen+2)<<2;
499                 else
500                         hdrlen = ipv6_optlen(&hdr);
501
502                 prevhdr = nexthdr;
503                 prev_nhoff = start;
504
505                 nexthdr = hdr.nexthdr;
506                 len -= hdrlen;
507                 start += hdrlen;
508         }
509
510         if (len < 0)
511                 return -1;
512
513         *prevhdrp = prevhdr;
514         *prevhoff = prev_nhoff;
515         *fhoff = start;
516
517         return 0;
518 }
519
520 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
521 {
522         struct sk_buff *clone;
523         struct net_device *dev = skb->dev;
524         struct frag_hdr *fhdr;
525         struct nf_ct_frag6_queue *fq;
526         struct ipv6hdr *hdr;
527         int fhoff, nhoff;
528         u8 prevhdr;
529         struct sk_buff *ret_skb = NULL;
530
531         /* Jumbo payload inhibits frag. header */
532         if (ipv6_hdr(skb)->payload_len == 0) {
533                 pr_debug("payload len = 0\n");
534                 return skb;
535         }
536
537         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
538                 return skb;
539
540         clone = skb_clone(skb, GFP_ATOMIC);
541         if (clone == NULL) {
542                 pr_debug("Can't clone skb\n");
543                 return skb;
544         }
545
546         NFCT_FRAG6_CB(clone)->orig = skb;
547
548         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
549                 pr_debug("message is too short.\n");
550                 goto ret_orig;
551         }
552
553         skb_set_transport_header(clone, fhoff);
554         hdr = ipv6_hdr(clone);
555         fhdr = (struct frag_hdr *)skb_transport_header(clone);
556
557         if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
558                 nf_ct_frag6_evictor();
559
560         fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
561         if (fq == NULL) {
562                 pr_debug("Can't find and can't create new queue\n");
563                 goto ret_orig;
564         }
565
566         spin_lock_bh(&fq->q.lock);
567
568         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
569                 spin_unlock_bh(&fq->q.lock);
570                 pr_debug("Can't insert skb to queue\n");
571                 fq_put(fq);
572                 goto ret_orig;
573         }
574
575         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
576             fq->q.meat == fq->q.len) {
577                 ret_skb = nf_ct_frag6_reasm(fq, dev);
578                 if (ret_skb == NULL)
579                         pr_debug("Can't reassemble fragmented packets\n");
580         }
581         spin_unlock_bh(&fq->q.lock);
582
583         fq_put(fq);
584         return ret_skb;
585
586 ret_orig:
587         kfree_skb(clone);
588         return skb;
589 }
590
591 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
592                         struct net_device *in, struct net_device *out,
593                         int (*okfn)(struct sk_buff *))
594 {
595         struct sk_buff *s, *s2;
596
597         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
598                 nf_conntrack_put_reasm(s->nfct_reasm);
599                 nf_conntrack_get_reasm(skb);
600                 s->nfct_reasm = skb;
601
602                 s2 = s->next;
603                 s->next = NULL;
604
605                 NF_HOOK_THRESH(NFPROTO_IPV6, hooknum, s, in, out, okfn,
606                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
607                 s = s2;
608         }
609         nf_conntrack_put_reasm(skb);
610 }
611
612 int nf_ct_frag6_init(void)
613 {
614         nf_frags.hashfn = nf_hashfn;
615         nf_frags.constructor = ip6_frag_init;
616         nf_frags.destructor = NULL;
617         nf_frags.skb_free = nf_skb_free;
618         nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
619         nf_frags.match = ip6_frag_match;
620         nf_frags.frag_expire = nf_ct_frag6_expire;
621         nf_frags.secret_interval = 10 * 60 * HZ;
622         nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
623         nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
624         nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
625         inet_frags_init_net(&nf_init_frags);
626         inet_frags_init(&nf_frags);
627
628         nf_ct_frag6_sysctl_header = register_sysctl_paths(nf_net_netfilter_sysctl_path,
629                                                           nf_ct_frag6_sysctl_table);
630         if (!nf_ct_frag6_sysctl_header) {
631                 inet_frags_fini(&nf_frags);
632                 return -ENOMEM;
633         }
634
635         return 0;
636 }
637
638 void nf_ct_frag6_cleanup(void)
639 {
640         unregister_sysctl_table(nf_ct_frag6_sysctl_header);
641         nf_ct_frag6_sysctl_header = NULL;
642
643         inet_frags_fini(&nf_frags);
644
645         nf_init_frags.low_thresh = 0;
646         nf_ct_frag6_evictor();
647 }