netfilter: nf_conntrack_reasm: properly handle packets fragmented into a single fragment
[pandora-kernel.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
31 #include <net/sock.h>
32 #include <net/snmp.h>
33 #include <net/inet_frag.h>
34
35 #include <net/ipv6.h>
36 #include <net/protocol.h>
37 #include <net/transp_v6.h>
38 #include <net/rawv6.h>
39 #include <net/ndisc.h>
40 #include <net/addrconf.h>
41 #include <net/netfilter/ipv6/nf_conntrack_ipv6.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
49 struct nf_ct_frag6_skb_cb
50 {
51         struct inet6_skb_parm   h;
52         int                     offset;
53         struct sk_buff          *orig;
54 };
55
56 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
57
58 struct nf_ct_frag6_queue
59 {
60         struct inet_frag_queue  q;
61
62         __be32                  id;             /* fragment id          */
63         u32                     user;
64         struct in6_addr         saddr;
65         struct in6_addr         daddr;
66
67         unsigned int            csum;
68         __u16                   nhoffset;
69 };
70
71 static struct inet_frags nf_frags;
72 static struct netns_frags nf_init_frags;
73
74 #ifdef CONFIG_SYSCTL
75 struct ctl_table nf_ct_ipv6_sysctl_table[] = {
76         {
77                 .procname       = "nf_conntrack_frag6_timeout",
78                 .data           = &nf_init_frags.timeout,
79                 .maxlen         = sizeof(unsigned int),
80                 .mode           = 0644,
81                 .proc_handler   = proc_dointvec_jiffies,
82         },
83         {
84                 .procname       = "nf_conntrack_frag6_low_thresh",
85                 .data           = &nf_init_frags.low_thresh,
86                 .maxlen         = sizeof(unsigned int),
87                 .mode           = 0644,
88                 .proc_handler   = proc_dointvec,
89         },
90         {
91                 .procname       = "nf_conntrack_frag6_high_thresh",
92                 .data           = &nf_init_frags.high_thresh,
93                 .maxlen         = sizeof(unsigned int),
94                 .mode           = 0644,
95                 .proc_handler   = proc_dointvec,
96         },
97         { }
98 };
99 #endif
100
101 static unsigned int nf_hashfn(struct inet_frag_queue *q)
102 {
103         const struct nf_ct_frag6_queue *nq;
104
105         nq = container_of(q, struct nf_ct_frag6_queue, q);
106         return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
107 }
108
109 static void nf_skb_free(struct sk_buff *skb)
110 {
111         if (NFCT_FRAG6_CB(skb)->orig)
112                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
113 }
114
115 /* Memory Tracking Functions. */
116 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
117 {
118         if (work)
119                 *work -= skb->truesize;
120         atomic_sub(skb->truesize, &nf_init_frags.mem);
121         nf_skb_free(skb);
122         kfree_skb(skb);
123 }
124
125 /* Destruction primitives. */
126
127 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
128 {
129         inet_frag_put(&fq->q, &nf_frags);
130 }
131
132 /* Kill fq entry. It is not destroyed immediately,
133  * because caller (and someone more) holds reference count.
134  */
135 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
136 {
137         inet_frag_kill(&fq->q, &nf_frags);
138 }
139
140 static void nf_ct_frag6_evictor(void)
141 {
142         local_bh_disable();
143         inet_frag_evictor(&nf_init_frags, &nf_frags);
144         local_bh_enable();
145 }
146
147 static void nf_ct_frag6_expire(unsigned long data)
148 {
149         struct nf_ct_frag6_queue *fq;
150
151         fq = container_of((struct inet_frag_queue *)data,
152                         struct nf_ct_frag6_queue, q);
153
154         spin_lock(&fq->q.lock);
155
156         if (fq->q.last_in & INET_FRAG_COMPLETE)
157                 goto out;
158
159         fq_kill(fq);
160
161 out:
162         spin_unlock(&fq->q.lock);
163         fq_put(fq);
164 }
165
166 /* Creation primitives. */
167
168 static __inline__ struct nf_ct_frag6_queue *
169 fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
170 {
171         struct inet_frag_queue *q;
172         struct ip6_create_arg arg;
173         unsigned int hash;
174
175         arg.id = id;
176         arg.user = user;
177         arg.src = src;
178         arg.dst = dst;
179
180         read_lock_bh(&nf_frags.lock);
181         hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
182
183         q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
184         local_bh_enable();
185         if (q == NULL)
186                 goto oom;
187
188         return container_of(q, struct nf_ct_frag6_queue, q);
189
190 oom:
191         pr_debug("Can't alloc new queue\n");
192         return NULL;
193 }
194
195
196 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
197                              const struct frag_hdr *fhdr, int nhoff)
198 {
199         struct sk_buff *prev, *next;
200         int offset, end;
201
202         if (fq->q.last_in & INET_FRAG_COMPLETE) {
203                 pr_debug("Allready completed\n");
204                 goto err;
205         }
206
207         offset = ntohs(fhdr->frag_off) & ~0x7;
208         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
209                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
210
211         if ((unsigned int)end > IPV6_MAXPLEN) {
212                 pr_debug("offset is too large.\n");
213                 return -1;
214         }
215
216         if (skb->ip_summed == CHECKSUM_COMPLETE) {
217                 const unsigned char *nh = skb_network_header(skb);
218                 skb->csum = csum_sub(skb->csum,
219                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
220                                                   0));
221         }
222
223         /* Is this the final fragment? */
224         if (!(fhdr->frag_off & htons(IP6_MF))) {
225                 /* If we already have some bits beyond end
226                  * or have different end, the segment is corrupted.
227                  */
228                 if (end < fq->q.len ||
229                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
230                         pr_debug("already received last fragment\n");
231                         goto err;
232                 }
233                 fq->q.last_in |= INET_FRAG_LAST_IN;
234                 fq->q.len = end;
235         } else {
236                 /* Check if the fragment is rounded to 8 bytes.
237                  * Required by the RFC.
238                  */
239                 if (end & 0x7) {
240                         /* RFC2460 says always send parameter problem in
241                          * this case. -DaveM
242                          */
243                         pr_debug("end of fragment not rounded to 8 bytes.\n");
244                         return -1;
245                 }
246                 if (end > fq->q.len) {
247                         /* Some bits beyond end -> corruption. */
248                         if (fq->q.last_in & INET_FRAG_LAST_IN) {
249                                 pr_debug("last packet already reached.\n");
250                                 goto err;
251                         }
252                         fq->q.len = end;
253                 }
254         }
255
256         if (end == offset)
257                 goto err;
258
259         /* Point into the IP datagram 'data' part. */
260         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
261                 pr_debug("queue: message is too short.\n");
262                 goto err;
263         }
264         if (pskb_trim_rcsum(skb, end - offset)) {
265                 pr_debug("Can't trim\n");
266                 goto err;
267         }
268
269         /* Find out which fragments are in front and at the back of us
270          * in the chain of fragments so far.  We must know where to put
271          * this fragment, right?
272          */
273         prev = NULL;
274         for (next = fq->q.fragments; next != NULL; next = next->next) {
275                 if (NFCT_FRAG6_CB(next)->offset >= offset)
276                         break;  /* bingo! */
277                 prev = next;
278         }
279
280         /* We found where to put this one.  Check for overlap with
281          * preceding fragment, and, if needed, align things so that
282          * any overlaps are eliminated.
283          */
284         if (prev) {
285                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
286
287                 if (i > 0) {
288                         offset += i;
289                         if (end <= offset) {
290                                 pr_debug("overlap\n");
291                                 goto err;
292                         }
293                         if (!pskb_pull(skb, i)) {
294                                 pr_debug("Can't pull\n");
295                                 goto err;
296                         }
297                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
298                                 skb->ip_summed = CHECKSUM_NONE;
299                 }
300         }
301
302         /* Look for overlap with succeeding segments.
303          * If we can merge fragments, do it.
304          */
305         while (next && NFCT_FRAG6_CB(next)->offset < end) {
306                 /* overlap is 'i' bytes */
307                 int i = end - NFCT_FRAG6_CB(next)->offset;
308
309                 if (i < next->len) {
310                         /* Eat head of the next overlapped fragment
311                          * and leave the loop. The next ones cannot overlap.
312                          */
313                         pr_debug("Eat head of the overlapped parts.: %d", i);
314                         if (!pskb_pull(next, i))
315                                 goto err;
316
317                         /* next fragment */
318                         NFCT_FRAG6_CB(next)->offset += i;
319                         fq->q.meat -= i;
320                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
321                                 next->ip_summed = CHECKSUM_NONE;
322                         break;
323                 } else {
324                         struct sk_buff *free_it = next;
325
326                         /* Old fragmnet is completely overridden with
327                          * new one drop it.
328                          */
329                         next = next->next;
330
331                         if (prev)
332                                 prev->next = next;
333                         else
334                                 fq->q.fragments = next;
335
336                         fq->q.meat -= free_it->len;
337                         frag_kfree_skb(free_it, NULL);
338                 }
339         }
340
341         NFCT_FRAG6_CB(skb)->offset = offset;
342
343         /* Insert this fragment in the chain of fragments. */
344         skb->next = next;
345         if (prev)
346                 prev->next = skb;
347         else
348                 fq->q.fragments = skb;
349
350         skb->dev = NULL;
351         fq->q.stamp = skb->tstamp;
352         fq->q.meat += skb->len;
353         atomic_add(skb->truesize, &nf_init_frags.mem);
354
355         /* The first fragment.
356          * nhoffset is obtained from the first fragment, of course.
357          */
358         if (offset == 0) {
359                 fq->nhoffset = nhoff;
360                 fq->q.last_in |= INET_FRAG_FIRST_IN;
361         }
362         write_lock(&nf_frags.lock);
363         list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
364         write_unlock(&nf_frags.lock);
365         return 0;
366
367 err:
368         return -1;
369 }
370
371 /*
372  *      Check if this packet is complete.
373  *      Returns NULL on failure by any reason, and pointer
374  *      to current nexthdr field in reassembled frame.
375  *
376  *      It is called with locked fq, and caller must check that
377  *      queue is eligible for reassembly i.e. it is not COMPLETE,
378  *      the last and the first frames arrived and all the bits are here.
379  */
380 static struct sk_buff *
381 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
382 {
383         struct sk_buff *fp, *op, *head = fq->q.fragments;
384         int    payload_len;
385
386         fq_kill(fq);
387
388         WARN_ON(head == NULL);
389         WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
390
391         /* Unfragmented part is taken from the first segment. */
392         payload_len = ((head->data - skb_network_header(head)) -
393                        sizeof(struct ipv6hdr) + fq->q.len -
394                        sizeof(struct frag_hdr));
395         if (payload_len > IPV6_MAXPLEN) {
396                 pr_debug("payload len is too large.\n");
397                 goto out_oversize;
398         }
399
400         /* Head of list must not be cloned. */
401         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
402                 pr_debug("skb is cloned but can't expand head");
403                 goto out_oom;
404         }
405
406         /* If the first fragment is fragmented itself, we split
407          * it to two chunks: the first with data and paged part
408          * and the second, holding only fragments. */
409         if (skb_has_frags(head)) {
410                 struct sk_buff *clone;
411                 int i, plen = 0;
412
413                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
414                         pr_debug("Can't alloc skb\n");
415                         goto out_oom;
416                 }
417                 clone->next = head->next;
418                 head->next = clone;
419                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
420                 skb_frag_list_init(head);
421                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
422                         plen += skb_shinfo(head)->frags[i].size;
423                 clone->len = clone->data_len = head->data_len - plen;
424                 head->data_len -= clone->len;
425                 head->len -= clone->len;
426                 clone->csum = 0;
427                 clone->ip_summed = head->ip_summed;
428
429                 NFCT_FRAG6_CB(clone)->orig = NULL;
430                 atomic_add(clone->truesize, &nf_init_frags.mem);
431         }
432
433         /* We have to remove fragment header from datagram and to relocate
434          * header in order to calculate ICV correctly. */
435         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
436         memmove(head->head + sizeof(struct frag_hdr), head->head,
437                 (head->data - head->head) - sizeof(struct frag_hdr));
438         head->mac_header += sizeof(struct frag_hdr);
439         head->network_header += sizeof(struct frag_hdr);
440
441         skb_shinfo(head)->frag_list = head->next;
442         skb_reset_transport_header(head);
443         skb_push(head, head->data - skb_network_header(head));
444         atomic_sub(head->truesize, &nf_init_frags.mem);
445
446         for (fp=head->next; fp; fp = fp->next) {
447                 head->data_len += fp->len;
448                 head->len += fp->len;
449                 if (head->ip_summed != fp->ip_summed)
450                         head->ip_summed = CHECKSUM_NONE;
451                 else if (head->ip_summed == CHECKSUM_COMPLETE)
452                         head->csum = csum_add(head->csum, fp->csum);
453                 head->truesize += fp->truesize;
454                 atomic_sub(fp->truesize, &nf_init_frags.mem);
455         }
456
457         head->next = NULL;
458         head->dev = dev;
459         head->tstamp = fq->q.stamp;
460         ipv6_hdr(head)->payload_len = htons(payload_len);
461
462         /* Yes, and fold redundant checksum back. 8) */
463         if (head->ip_summed == CHECKSUM_COMPLETE)
464                 head->csum = csum_partial(skb_network_header(head),
465                                           skb_network_header_len(head),
466                                           head->csum);
467
468         fq->q.fragments = NULL;
469
470         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
471         fp = skb_shinfo(head)->frag_list;
472         if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
473                 /* at above code, head skb is divided into two skbs. */
474                 fp = fp->next;
475
476         op = NFCT_FRAG6_CB(head)->orig;
477         for (; fp; fp = fp->next) {
478                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
479
480                 op->next = orig;
481                 op = orig;
482                 NFCT_FRAG6_CB(fp)->orig = NULL;
483         }
484
485         return head;
486
487 out_oversize:
488         if (net_ratelimit())
489                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
490         goto out_fail;
491 out_oom:
492         if (net_ratelimit())
493                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
494 out_fail:
495         return NULL;
496 }
497
498 /*
499  * find the header just before Fragment Header.
500  *
501  * if success return 0 and set ...
502  * (*prevhdrp): the value of "Next Header Field" in the header
503  *              just before Fragment Header.
504  * (*prevhoff): the offset of "Next Header Field" in the header
505  *              just before Fragment Header.
506  * (*fhoff)   : the offset of Fragment Header.
507  *
508  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
509  *
510  */
511 static int
512 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
513 {
514         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
515         const int netoff = skb_network_offset(skb);
516         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
517         int start = netoff + sizeof(struct ipv6hdr);
518         int len = skb->len - start;
519         u8 prevhdr = NEXTHDR_IPV6;
520
521         while (nexthdr != NEXTHDR_FRAGMENT) {
522                 struct ipv6_opt_hdr hdr;
523                 int hdrlen;
524
525                 if (!ipv6_ext_hdr(nexthdr)) {
526                         return -1;
527                 }
528                 if (nexthdr == NEXTHDR_NONE) {
529                         pr_debug("next header is none\n");
530                         return -1;
531                 }
532                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
533                         pr_debug("too short\n");
534                         return -1;
535                 }
536                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
537                         BUG();
538                 if (nexthdr == NEXTHDR_AUTH)
539                         hdrlen = (hdr.hdrlen+2)<<2;
540                 else
541                         hdrlen = ipv6_optlen(&hdr);
542
543                 prevhdr = nexthdr;
544                 prev_nhoff = start;
545
546                 nexthdr = hdr.nexthdr;
547                 len -= hdrlen;
548                 start += hdrlen;
549         }
550
551         if (len < 0)
552                 return -1;
553
554         *prevhdrp = prevhdr;
555         *prevhoff = prev_nhoff;
556         *fhoff = start;
557
558         return 0;
559 }
560
561 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
562 {
563         struct sk_buff *clone;
564         struct net_device *dev = skb->dev;
565         struct frag_hdr *fhdr;
566         struct nf_ct_frag6_queue *fq;
567         struct ipv6hdr *hdr;
568         int fhoff, nhoff;
569         u8 prevhdr;
570         struct sk_buff *ret_skb = NULL;
571
572         /* Jumbo payload inhibits frag. header */
573         if (ipv6_hdr(skb)->payload_len == 0) {
574                 pr_debug("payload len = 0\n");
575                 return skb;
576         }
577
578         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
579                 return skb;
580
581         clone = skb_clone(skb, GFP_ATOMIC);
582         if (clone == NULL) {
583                 pr_debug("Can't clone skb\n");
584                 return skb;
585         }
586
587         NFCT_FRAG6_CB(clone)->orig = skb;
588
589         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
590                 pr_debug("message is too short.\n");
591                 goto ret_orig;
592         }
593
594         skb_set_transport_header(clone, fhoff);
595         hdr = ipv6_hdr(clone);
596         fhdr = (struct frag_hdr *)skb_transport_header(clone);
597
598         if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
599                 nf_ct_frag6_evictor();
600
601         fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
602         if (fq == NULL) {
603                 pr_debug("Can't find and can't create new queue\n");
604                 goto ret_orig;
605         }
606
607         spin_lock_bh(&fq->q.lock);
608
609         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
610                 spin_unlock_bh(&fq->q.lock);
611                 pr_debug("Can't insert skb to queue\n");
612                 fq_put(fq);
613                 goto ret_orig;
614         }
615
616         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
617             fq->q.meat == fq->q.len) {
618                 ret_skb = nf_ct_frag6_reasm(fq, dev);
619                 if (ret_skb == NULL)
620                         pr_debug("Can't reassemble fragmented packets\n");
621         }
622         spin_unlock_bh(&fq->q.lock);
623
624         fq_put(fq);
625         return ret_skb;
626
627 ret_orig:
628         kfree_skb(clone);
629         return skb;
630 }
631
632 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
633                         struct net_device *in, struct net_device *out,
634                         int (*okfn)(struct sk_buff *))
635 {
636         struct sk_buff *s, *s2;
637
638         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
639                 nf_conntrack_put_reasm(s->nfct_reasm);
640                 nf_conntrack_get_reasm(skb);
641                 s->nfct_reasm = skb;
642
643                 s2 = s->next;
644                 s->next = NULL;
645
646                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
647                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
648                 s = s2;
649         }
650         nf_conntrack_put_reasm(skb);
651 }
652
653 int nf_ct_frag6_init(void)
654 {
655         nf_frags.hashfn = nf_hashfn;
656         nf_frags.constructor = ip6_frag_init;
657         nf_frags.destructor = NULL;
658         nf_frags.skb_free = nf_skb_free;
659         nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
660         nf_frags.match = ip6_frag_match;
661         nf_frags.frag_expire = nf_ct_frag6_expire;
662         nf_frags.secret_interval = 10 * 60 * HZ;
663         nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
664         nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
665         nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
666         inet_frags_init_net(&nf_init_frags);
667         inet_frags_init(&nf_frags);
668
669         return 0;
670 }
671
672 void nf_ct_frag6_cleanup(void)
673 {
674         inet_frags_fini(&nf_frags);
675
676         nf_init_frags.low_thresh = 0;
677         nf_ct_frag6_evictor();
678 }