net: Rename skb_has_frags to skb_has_frag_list
[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 /* Memory Tracking Functions. */
117 static void frag_kfree_skb(struct sk_buff *skb)
118 {
119         atomic_sub(skb->truesize, &nf_init_frags.mem);
120         nf_skb_free(skb);
121         kfree_skb(skb);
122 }
123
124 /* Destruction primitives. */
125
126 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
127 {
128         inet_frag_put(&fq->q, &nf_frags);
129 }
130
131 /* Kill fq entry. It is not destroyed immediately,
132  * because caller (and someone more) holds reference count.
133  */
134 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
135 {
136         inet_frag_kill(&fq->q, &nf_frags);
137 }
138
139 static void nf_ct_frag6_evictor(void)
140 {
141         local_bh_disable();
142         inet_frag_evictor(&nf_init_frags, &nf_frags);
143         local_bh_enable();
144 }
145
146 static void nf_ct_frag6_expire(unsigned long data)
147 {
148         struct nf_ct_frag6_queue *fq;
149
150         fq = container_of((struct inet_frag_queue *)data,
151                         struct nf_ct_frag6_queue, q);
152
153         spin_lock(&fq->q.lock);
154
155         if (fq->q.last_in & INET_FRAG_COMPLETE)
156                 goto out;
157
158         fq_kill(fq);
159
160 out:
161         spin_unlock(&fq->q.lock);
162         fq_put(fq);
163 }
164
165 /* Creation primitives. */
166
167 static __inline__ struct nf_ct_frag6_queue *
168 fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
169 {
170         struct inet_frag_queue *q;
171         struct ip6_create_arg arg;
172         unsigned int hash;
173
174         arg.id = id;
175         arg.user = user;
176         arg.src = src;
177         arg.dst = dst;
178
179         read_lock_bh(&nf_frags.lock);
180         hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
181
182         q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
183         local_bh_enable();
184         if (q == NULL)
185                 goto oom;
186
187         return container_of(q, struct nf_ct_frag6_queue, q);
188
189 oom:
190         pr_debug("Can't alloc new queue\n");
191         return NULL;
192 }
193
194
195 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
196                              const struct frag_hdr *fhdr, int nhoff)
197 {
198         struct sk_buff *prev, *next;
199         int offset, end;
200
201         if (fq->q.last_in & INET_FRAG_COMPLETE) {
202                 pr_debug("Already completed\n");
203                 goto err;
204         }
205
206         offset = ntohs(fhdr->frag_off) & ~0x7;
207         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
208                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
209
210         if ((unsigned int)end > IPV6_MAXPLEN) {
211                 pr_debug("offset is too large.\n");
212                 return -1;
213         }
214
215         if (skb->ip_summed == CHECKSUM_COMPLETE) {
216                 const unsigned char *nh = skb_network_header(skb);
217                 skb->csum = csum_sub(skb->csum,
218                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
219                                                   0));
220         }
221
222         /* Is this the final fragment? */
223         if (!(fhdr->frag_off & htons(IP6_MF))) {
224                 /* If we already have some bits beyond end
225                  * or have different end, the segment is corrupted.
226                  */
227                 if (end < fq->q.len ||
228                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
229                         pr_debug("already received last fragment\n");
230                         goto err;
231                 }
232                 fq->q.last_in |= INET_FRAG_LAST_IN;
233                 fq->q.len = end;
234         } else {
235                 /* Check if the fragment is rounded to 8 bytes.
236                  * Required by the RFC.
237                  */
238                 if (end & 0x7) {
239                         /* RFC2460 says always send parameter problem in
240                          * this case. -DaveM
241                          */
242                         pr_debug("end of fragment not rounded to 8 bytes.\n");
243                         return -1;
244                 }
245                 if (end > fq->q.len) {
246                         /* Some bits beyond end -> corruption. */
247                         if (fq->q.last_in & INET_FRAG_LAST_IN) {
248                                 pr_debug("last packet already reached.\n");
249                                 goto err;
250                         }
251                         fq->q.len = end;
252                 }
253         }
254
255         if (end == offset)
256                 goto err;
257
258         /* Point into the IP datagram 'data' part. */
259         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
260                 pr_debug("queue: message is too short.\n");
261                 goto err;
262         }
263         if (pskb_trim_rcsum(skb, end - offset)) {
264                 pr_debug("Can't trim\n");
265                 goto err;
266         }
267
268         /* Find out which fragments are in front and at the back of us
269          * in the chain of fragments so far.  We must know where to put
270          * this fragment, right?
271          */
272         prev = fq->q.fragments_tail;
273         if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
274                 next = NULL;
275                 goto found;
276         }
277         prev = NULL;
278         for (next = fq->q.fragments; next != NULL; next = next->next) {
279                 if (NFCT_FRAG6_CB(next)->offset >= offset)
280                         break;  /* bingo! */
281                 prev = next;
282         }
283
284 found:
285         /* We found where to put this one.  Check for overlap with
286          * preceding fragment, and, if needed, align things so that
287          * any overlaps are eliminated.
288          */
289         if (prev) {
290                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
291
292                 if (i > 0) {
293                         offset += i;
294                         if (end <= offset) {
295                                 pr_debug("overlap\n");
296                                 goto err;
297                         }
298                         if (!pskb_pull(skb, i)) {
299                                 pr_debug("Can't pull\n");
300                                 goto err;
301                         }
302                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
303                                 skb->ip_summed = CHECKSUM_NONE;
304                 }
305         }
306
307         /* Look for overlap with succeeding segments.
308          * If we can merge fragments, do it.
309          */
310         while (next && NFCT_FRAG6_CB(next)->offset < end) {
311                 /* overlap is 'i' bytes */
312                 int i = end - NFCT_FRAG6_CB(next)->offset;
313
314                 if (i < next->len) {
315                         /* Eat head of the next overlapped fragment
316                          * and leave the loop. The next ones cannot overlap.
317                          */
318                         pr_debug("Eat head of the overlapped parts.: %d", i);
319                         if (!pskb_pull(next, i))
320                                 goto err;
321
322                         /* next fragment */
323                         NFCT_FRAG6_CB(next)->offset += i;
324                         fq->q.meat -= i;
325                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
326                                 next->ip_summed = CHECKSUM_NONE;
327                         break;
328                 } else {
329                         struct sk_buff *free_it = next;
330
331                         /* Old fragmnet is completely overridden with
332                          * new one drop it.
333                          */
334                         next = next->next;
335
336                         if (prev)
337                                 prev->next = next;
338                         else
339                                 fq->q.fragments = next;
340
341                         fq->q.meat -= free_it->len;
342                         frag_kfree_skb(free_it);
343                 }
344         }
345
346         NFCT_FRAG6_CB(skb)->offset = offset;
347
348         /* Insert this fragment in the chain of fragments. */
349         skb->next = next;
350         if (!next)
351                 fq->q.fragments_tail = skb;
352         if (prev)
353                 prev->next = skb;
354         else
355                 fq->q.fragments = skb;
356
357         skb->dev = NULL;
358         fq->q.stamp = skb->tstamp;
359         fq->q.meat += skb->len;
360         atomic_add(skb->truesize, &nf_init_frags.mem);
361
362         /* The first fragment.
363          * nhoffset is obtained from the first fragment, of course.
364          */
365         if (offset == 0) {
366                 fq->nhoffset = nhoff;
367                 fq->q.last_in |= INET_FRAG_FIRST_IN;
368         }
369         write_lock(&nf_frags.lock);
370         list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
371         write_unlock(&nf_frags.lock);
372         return 0;
373
374 err:
375         return -1;
376 }
377
378 /*
379  *      Check if this packet is complete.
380  *      Returns NULL on failure by any reason, and pointer
381  *      to current nexthdr field in reassembled frame.
382  *
383  *      It is called with locked fq, and caller must check that
384  *      queue is eligible for reassembly i.e. it is not COMPLETE,
385  *      the last and the first frames arrived and all the bits are here.
386  */
387 static struct sk_buff *
388 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
389 {
390         struct sk_buff *fp, *op, *head = fq->q.fragments;
391         int    payload_len;
392
393         fq_kill(fq);
394
395         WARN_ON(head == NULL);
396         WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
397
398         /* Unfragmented part is taken from the first segment. */
399         payload_len = ((head->data - skb_network_header(head)) -
400                        sizeof(struct ipv6hdr) + fq->q.len -
401                        sizeof(struct frag_hdr));
402         if (payload_len > IPV6_MAXPLEN) {
403                 pr_debug("payload len is too large.\n");
404                 goto out_oversize;
405         }
406
407         /* Head of list must not be cloned. */
408         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
409                 pr_debug("skb is cloned but can't expand head");
410                 goto out_oom;
411         }
412
413         /* If the first fragment is fragmented itself, we split
414          * it to two chunks: the first with data and paged part
415          * and the second, holding only fragments. */
416         if (skb_has_frag_list(head)) {
417                 struct sk_buff *clone;
418                 int i, plen = 0;
419
420                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
421                         pr_debug("Can't alloc skb\n");
422                         goto out_oom;
423                 }
424                 clone->next = head->next;
425                 head->next = clone;
426                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
427                 skb_frag_list_init(head);
428                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
429                         plen += skb_shinfo(head)->frags[i].size;
430                 clone->len = clone->data_len = head->data_len - plen;
431                 head->data_len -= clone->len;
432                 head->len -= clone->len;
433                 clone->csum = 0;
434                 clone->ip_summed = head->ip_summed;
435
436                 NFCT_FRAG6_CB(clone)->orig = NULL;
437                 atomic_add(clone->truesize, &nf_init_frags.mem);
438         }
439
440         /* We have to remove fragment header from datagram and to relocate
441          * header in order to calculate ICV correctly. */
442         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
443         memmove(head->head + sizeof(struct frag_hdr), head->head,
444                 (head->data - head->head) - sizeof(struct frag_hdr));
445         head->mac_header += sizeof(struct frag_hdr);
446         head->network_header += sizeof(struct frag_hdr);
447
448         skb_shinfo(head)->frag_list = head->next;
449         skb_reset_transport_header(head);
450         skb_push(head, head->data - skb_network_header(head));
451
452         for (fp=head->next; fp; fp = fp->next) {
453                 head->data_len += fp->len;
454                 head->len += fp->len;
455                 if (head->ip_summed != fp->ip_summed)
456                         head->ip_summed = CHECKSUM_NONE;
457                 else if (head->ip_summed == CHECKSUM_COMPLETE)
458                         head->csum = csum_add(head->csum, fp->csum);
459                 head->truesize += fp->truesize;
460         }
461         atomic_sub(head->truesize, &nf_init_frags.mem);
462
463         head->next = NULL;
464         head->dev = dev;
465         head->tstamp = fq->q.stamp;
466         ipv6_hdr(head)->payload_len = htons(payload_len);
467
468         /* Yes, and fold redundant checksum back. 8) */
469         if (head->ip_summed == CHECKSUM_COMPLETE)
470                 head->csum = csum_partial(skb_network_header(head),
471                                           skb_network_header_len(head),
472                                           head->csum);
473
474         fq->q.fragments = NULL;
475         fq->q.fragments_tail = NULL;
476
477         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
478         fp = skb_shinfo(head)->frag_list;
479         if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
480                 /* at above code, head skb is divided into two skbs. */
481                 fp = fp->next;
482
483         op = NFCT_FRAG6_CB(head)->orig;
484         for (; fp; fp = fp->next) {
485                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
486
487                 op->next = orig;
488                 op = orig;
489                 NFCT_FRAG6_CB(fp)->orig = NULL;
490         }
491
492         return head;
493
494 out_oversize:
495         if (net_ratelimit())
496                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
497         goto out_fail;
498 out_oom:
499         if (net_ratelimit())
500                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
501 out_fail:
502         return NULL;
503 }
504
505 /*
506  * find the header just before Fragment Header.
507  *
508  * if success return 0 and set ...
509  * (*prevhdrp): the value of "Next Header Field" in the header
510  *              just before Fragment Header.
511  * (*prevhoff): the offset of "Next Header Field" in the header
512  *              just before Fragment Header.
513  * (*fhoff)   : the offset of Fragment Header.
514  *
515  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
516  *
517  */
518 static int
519 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
520 {
521         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
522         const int netoff = skb_network_offset(skb);
523         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
524         int start = netoff + sizeof(struct ipv6hdr);
525         int len = skb->len - start;
526         u8 prevhdr = NEXTHDR_IPV6;
527
528         while (nexthdr != NEXTHDR_FRAGMENT) {
529                 struct ipv6_opt_hdr hdr;
530                 int hdrlen;
531
532                 if (!ipv6_ext_hdr(nexthdr)) {
533                         return -1;
534                 }
535                 if (nexthdr == NEXTHDR_NONE) {
536                         pr_debug("next header is none\n");
537                         return -1;
538                 }
539                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
540                         pr_debug("too short\n");
541                         return -1;
542                 }
543                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
544                         BUG();
545                 if (nexthdr == NEXTHDR_AUTH)
546                         hdrlen = (hdr.hdrlen+2)<<2;
547                 else
548                         hdrlen = ipv6_optlen(&hdr);
549
550                 prevhdr = nexthdr;
551                 prev_nhoff = start;
552
553                 nexthdr = hdr.nexthdr;
554                 len -= hdrlen;
555                 start += hdrlen;
556         }
557
558         if (len < 0)
559                 return -1;
560
561         *prevhdrp = prevhdr;
562         *prevhoff = prev_nhoff;
563         *fhoff = start;
564
565         return 0;
566 }
567
568 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
569 {
570         struct sk_buff *clone;
571         struct net_device *dev = skb->dev;
572         struct frag_hdr *fhdr;
573         struct nf_ct_frag6_queue *fq;
574         struct ipv6hdr *hdr;
575         int fhoff, nhoff;
576         u8 prevhdr;
577         struct sk_buff *ret_skb = NULL;
578
579         /* Jumbo payload inhibits frag. header */
580         if (ipv6_hdr(skb)->payload_len == 0) {
581                 pr_debug("payload len = 0\n");
582                 return skb;
583         }
584
585         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
586                 return skb;
587
588         clone = skb_clone(skb, GFP_ATOMIC);
589         if (clone == NULL) {
590                 pr_debug("Can't clone skb\n");
591                 return skb;
592         }
593
594         NFCT_FRAG6_CB(clone)->orig = skb;
595
596         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
597                 pr_debug("message is too short.\n");
598                 goto ret_orig;
599         }
600
601         skb_set_transport_header(clone, fhoff);
602         hdr = ipv6_hdr(clone);
603         fhdr = (struct frag_hdr *)skb_transport_header(clone);
604
605         if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
606                 nf_ct_frag6_evictor();
607
608         fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
609         if (fq == NULL) {
610                 pr_debug("Can't find and can't create new queue\n");
611                 goto ret_orig;
612         }
613
614         spin_lock_bh(&fq->q.lock);
615
616         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
617                 spin_unlock_bh(&fq->q.lock);
618                 pr_debug("Can't insert skb to queue\n");
619                 fq_put(fq);
620                 goto ret_orig;
621         }
622
623         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
624             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_bh(&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(NFPROTO_IPV6, 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_init(void)
661 {
662         nf_frags.hashfn = nf_hashfn;
663         nf_frags.constructor = ip6_frag_init;
664         nf_frags.destructor = NULL;
665         nf_frags.skb_free = nf_skb_free;
666         nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
667         nf_frags.match = ip6_frag_match;
668         nf_frags.frag_expire = nf_ct_frag6_expire;
669         nf_frags.secret_interval = 10 * 60 * HZ;
670         nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
671         nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
672         nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
673         inet_frags_init_net(&nf_init_frags);
674         inet_frags_init(&nf_frags);
675
676         return 0;
677 }
678
679 void nf_ct_frag6_cleanup(void)
680 {
681         inet_frags_fini(&nf_frags);
682
683         nf_init_frags.low_thresh = 0;
684         nf_ct_frag6_evictor();
685 }