]> Pileus Git - ~andy/linux/blob - net/ipv6/netfilter/nf_conntrack_reasm.c
net: Add skb_unclone() helper function.
[~andy/linux] / 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 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
49
50
51 struct nf_ct_frag6_skb_cb
52 {
53         struct inet6_skb_parm   h;
54         int                     offset;
55         struct sk_buff          *orig;
56 };
57
58 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
59
60 static struct inet_frags nf_frags;
61
62 #ifdef CONFIG_SYSCTL
63 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
64         {
65                 .procname       = "nf_conntrack_frag6_timeout",
66                 .data           = &init_net.nf_frag.frags.timeout,
67                 .maxlen         = sizeof(unsigned int),
68                 .mode           = 0644,
69                 .proc_handler   = proc_dointvec_jiffies,
70         },
71         {
72                 .procname       = "nf_conntrack_frag6_low_thresh",
73                 .data           = &init_net.nf_frag.frags.low_thresh,
74                 .maxlen         = sizeof(unsigned int),
75                 .mode           = 0644,
76                 .proc_handler   = proc_dointvec,
77         },
78         {
79                 .procname       = "nf_conntrack_frag6_high_thresh",
80                 .data           = &init_net.nf_frag.frags.high_thresh,
81                 .maxlen         = sizeof(unsigned int),
82                 .mode           = 0644,
83                 .proc_handler   = proc_dointvec,
84         },
85         { }
86 };
87
88 static int nf_ct_frag6_sysctl_register(struct net *net)
89 {
90         struct ctl_table *table;
91         struct ctl_table_header *hdr;
92
93         table = nf_ct_frag6_sysctl_table;
94         if (!net_eq(net, &init_net)) {
95                 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
96                                 GFP_KERNEL);
97                 if (table == NULL)
98                         goto err_alloc;
99
100                 table[0].data = &net->ipv6.frags.high_thresh;
101                 table[1].data = &net->ipv6.frags.low_thresh;
102                 table[2].data = &net->ipv6.frags.timeout;
103         }
104
105         hdr = register_net_sysctl(net, "net/netfilter", table);
106         if (hdr == NULL)
107                 goto err_reg;
108
109         net->nf_frag.sysctl.frags_hdr = hdr;
110         return 0;
111
112 err_reg:
113         if (!net_eq(net, &init_net))
114                 kfree(table);
115 err_alloc:
116         return -ENOMEM;
117 }
118
119 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
120 {
121         struct ctl_table *table;
122
123         table = net->nf_frag.sysctl.frags_hdr->ctl_table_arg;
124         unregister_net_sysctl_table(net->nf_frag.sysctl.frags_hdr);
125         if (!net_eq(net, &init_net))
126                 kfree(table);
127 }
128
129 #else
130 static int nf_ct_frag6_sysctl_register(struct net *net)
131 {
132         return 0;
133 }
134 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
135 {
136 }
137 #endif
138
139 static unsigned int nf_hashfn(struct inet_frag_queue *q)
140 {
141         const struct frag_queue *nq;
142
143         nq = container_of(q, struct frag_queue, q);
144         return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
145 }
146
147 static void nf_skb_free(struct sk_buff *skb)
148 {
149         if (NFCT_FRAG6_CB(skb)->orig)
150                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
151 }
152
153 static void nf_ct_frag6_expire(unsigned long data)
154 {
155         struct frag_queue *fq;
156         struct net *net;
157
158         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
159         net = container_of(fq->q.net, struct net, nf_frag.frags);
160
161         ip6_expire_frag_queue(net, fq, &nf_frags);
162 }
163
164 /* Creation primitives. */
165 static inline struct frag_queue *fq_find(struct net *net, __be32 id,
166                                          u32 user, struct in6_addr *src,
167                                          struct in6_addr *dst)
168 {
169         struct inet_frag_queue *q;
170         struct ip6_create_arg arg;
171         unsigned int hash;
172
173         arg.id = id;
174         arg.user = user;
175         arg.src = src;
176         arg.dst = dst;
177
178         read_lock_bh(&nf_frags.lock);
179         hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
180
181         q = inet_frag_find(&net->nf_frag.frags, &nf_frags, &arg, hash);
182         local_bh_enable();
183         if (q == NULL)
184                 goto oom;
185
186         return container_of(q, struct frag_queue, q);
187
188 oom:
189         return NULL;
190 }
191
192
193 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
194                              const struct frag_hdr *fhdr, int nhoff)
195 {
196         struct sk_buff *prev, *next;
197         unsigned int payload_len;
198         int offset, end;
199
200         if (fq->q.last_in & INET_FRAG_COMPLETE) {
201                 pr_debug("Already completed\n");
202                 goto err;
203         }
204
205         payload_len = ntohs(ipv6_hdr(skb)->payload_len);
206
207         offset = ntohs(fhdr->frag_off) & ~0x7;
208         end = offset + (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 = fq->q.fragments_tail;
274         if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
275                 next = NULL;
276                 goto found;
277         }
278         prev = NULL;
279         for (next = fq->q.fragments; next != NULL; next = next->next) {
280                 if (NFCT_FRAG6_CB(next)->offset >= offset)
281                         break;  /* bingo! */
282                 prev = next;
283         }
284
285 found:
286         /* RFC5722, Section 4:
287          *                                  When reassembling an IPv6 datagram, if
288          *   one or more its constituent fragments is determined to be an
289          *   overlapping fragment, the entire datagram (and any constituent
290          *   fragments, including those not yet received) MUST be silently
291          *   discarded.
292          */
293
294         /* Check for overlap with preceding fragment. */
295         if (prev &&
296             (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
297                 goto discard_fq;
298
299         /* Look for overlap with succeeding segment. */
300         if (next && NFCT_FRAG6_CB(next)->offset < end)
301                 goto discard_fq;
302
303         NFCT_FRAG6_CB(skb)->offset = offset;
304
305         /* Insert this fragment in the chain of fragments. */
306         skb->next = next;
307         if (!next)
308                 fq->q.fragments_tail = skb;
309         if (prev)
310                 prev->next = skb;
311         else
312                 fq->q.fragments = skb;
313
314         if (skb->dev) {
315                 fq->iif = skb->dev->ifindex;
316                 skb->dev = NULL;
317         }
318         fq->q.stamp = skb->tstamp;
319         fq->q.meat += skb->len;
320         if (payload_len > fq->q.max_size)
321                 fq->q.max_size = payload_len;
322         add_frag_mem_limit(&fq->q, skb->truesize);
323
324         /* The first fragment.
325          * nhoffset is obtained from the first fragment, of course.
326          */
327         if (offset == 0) {
328                 fq->nhoffset = nhoff;
329                 fq->q.last_in |= INET_FRAG_FIRST_IN;
330         }
331
332         inet_frag_lru_move(&fq->q);
333         return 0;
334
335 discard_fq:
336         inet_frag_kill(&fq->q, &nf_frags);
337 err:
338         return -1;
339 }
340
341 /*
342  *      Check if this packet is complete.
343  *      Returns NULL on failure by any reason, and pointer
344  *      to current nexthdr field in reassembled frame.
345  *
346  *      It is called with locked fq, and caller must check that
347  *      queue is eligible for reassembly i.e. it is not COMPLETE,
348  *      the last and the first frames arrived and all the bits are here.
349  */
350 static struct sk_buff *
351 nf_ct_frag6_reasm(struct frag_queue *fq, struct net_device *dev)
352 {
353         struct sk_buff *fp, *op, *head = fq->q.fragments;
354         int    payload_len;
355
356         inet_frag_kill(&fq->q, &nf_frags);
357
358         WARN_ON(head == NULL);
359         WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
360
361         /* Unfragmented part is taken from the first segment. */
362         payload_len = ((head->data - skb_network_header(head)) -
363                        sizeof(struct ipv6hdr) + fq->q.len -
364                        sizeof(struct frag_hdr));
365         if (payload_len > IPV6_MAXPLEN) {
366                 pr_debug("payload len is too large.\n");
367                 goto out_oversize;
368         }
369
370         /* Head of list must not be cloned. */
371         if (skb_unclone(head, GFP_ATOMIC)) {
372                 pr_debug("skb is cloned but can't expand head");
373                 goto out_oom;
374         }
375
376         /* If the first fragment is fragmented itself, we split
377          * it to two chunks: the first with data and paged part
378          * and the second, holding only fragments. */
379         if (skb_has_frag_list(head)) {
380                 struct sk_buff *clone;
381                 int i, plen = 0;
382
383                 clone = alloc_skb(0, GFP_ATOMIC);
384                 if (clone == NULL)
385                         goto out_oom;
386
387                 clone->next = head->next;
388                 head->next = clone;
389                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
390                 skb_frag_list_init(head);
391                 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
392                         plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
393                 clone->len = clone->data_len = head->data_len - plen;
394                 head->data_len -= clone->len;
395                 head->len -= clone->len;
396                 clone->csum = 0;
397                 clone->ip_summed = head->ip_summed;
398
399                 NFCT_FRAG6_CB(clone)->orig = NULL;
400                 add_frag_mem_limit(&fq->q, clone->truesize);
401         }
402
403         /* We have to remove fragment header from datagram and to relocate
404          * header in order to calculate ICV correctly. */
405         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
406         memmove(head->head + sizeof(struct frag_hdr), head->head,
407                 (head->data - head->head) - sizeof(struct frag_hdr));
408         head->mac_header += sizeof(struct frag_hdr);
409         head->network_header += sizeof(struct frag_hdr);
410
411         skb_shinfo(head)->frag_list = head->next;
412         skb_reset_transport_header(head);
413         skb_push(head, head->data - skb_network_header(head));
414
415         for (fp=head->next; fp; fp = fp->next) {
416                 head->data_len += fp->len;
417                 head->len += fp->len;
418                 if (head->ip_summed != fp->ip_summed)
419                         head->ip_summed = CHECKSUM_NONE;
420                 else if (head->ip_summed == CHECKSUM_COMPLETE)
421                         head->csum = csum_add(head->csum, fp->csum);
422                 head->truesize += fp->truesize;
423         }
424         sub_frag_mem_limit(&fq->q, head->truesize);
425
426         head->local_df = 1;
427         head->next = NULL;
428         head->dev = dev;
429         head->tstamp = fq->q.stamp;
430         ipv6_hdr(head)->payload_len = htons(payload_len);
431         IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
432
433         /* Yes, and fold redundant checksum back. 8) */
434         if (head->ip_summed == CHECKSUM_COMPLETE)
435                 head->csum = csum_partial(skb_network_header(head),
436                                           skb_network_header_len(head),
437                                           head->csum);
438
439         fq->q.fragments = NULL;
440         fq->q.fragments_tail = NULL;
441
442         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
443         fp = skb_shinfo(head)->frag_list;
444         if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
445                 /* at above code, head skb is divided into two skbs. */
446                 fp = fp->next;
447
448         op = NFCT_FRAG6_CB(head)->orig;
449         for (; fp; fp = fp->next) {
450                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
451
452                 op->next = orig;
453                 op = orig;
454                 NFCT_FRAG6_CB(fp)->orig = NULL;
455         }
456
457         return head;
458
459 out_oversize:
460         net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
461                             payload_len);
462         goto out_fail;
463 out_oom:
464         net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n");
465 out_fail:
466         return NULL;
467 }
468
469 /*
470  * find the header just before Fragment Header.
471  *
472  * if success return 0 and set ...
473  * (*prevhdrp): the value of "Next Header Field" in the header
474  *              just before Fragment Header.
475  * (*prevhoff): the offset of "Next Header Field" in the header
476  *              just before Fragment Header.
477  * (*fhoff)   : the offset of Fragment Header.
478  *
479  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
480  *
481  */
482 static int
483 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
484 {
485         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
486         const int netoff = skb_network_offset(skb);
487         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
488         int start = netoff + sizeof(struct ipv6hdr);
489         int len = skb->len - start;
490         u8 prevhdr = NEXTHDR_IPV6;
491
492         while (nexthdr != NEXTHDR_FRAGMENT) {
493                 struct ipv6_opt_hdr hdr;
494                 int hdrlen;
495
496                 if (!ipv6_ext_hdr(nexthdr)) {
497                         return -1;
498                 }
499                 if (nexthdr == NEXTHDR_NONE) {
500                         pr_debug("next header is none\n");
501                         return -1;
502                 }
503                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
504                         pr_debug("too short\n");
505                         return -1;
506                 }
507                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
508                         BUG();
509                 if (nexthdr == NEXTHDR_AUTH)
510                         hdrlen = (hdr.hdrlen+2)<<2;
511                 else
512                         hdrlen = ipv6_optlen(&hdr);
513
514                 prevhdr = nexthdr;
515                 prev_nhoff = start;
516
517                 nexthdr = hdr.nexthdr;
518                 len -= hdrlen;
519                 start += hdrlen;
520         }
521
522         if (len < 0)
523                 return -1;
524
525         *prevhdrp = prevhdr;
526         *prevhoff = prev_nhoff;
527         *fhoff = start;
528
529         return 0;
530 }
531
532 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
533 {
534         struct sk_buff *clone;
535         struct net_device *dev = skb->dev;
536         struct net *net = skb_dst(skb) ? dev_net(skb_dst(skb)->dev)
537                                        : dev_net(skb->dev);
538         struct frag_hdr *fhdr;
539         struct frag_queue *fq;
540         struct ipv6hdr *hdr;
541         int fhoff, nhoff;
542         u8 prevhdr;
543         struct sk_buff *ret_skb = NULL;
544
545         /* Jumbo payload inhibits frag. header */
546         if (ipv6_hdr(skb)->payload_len == 0) {
547                 pr_debug("payload len = 0\n");
548                 return skb;
549         }
550
551         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
552                 return skb;
553
554         clone = skb_clone(skb, GFP_ATOMIC);
555         if (clone == NULL) {
556                 pr_debug("Can't clone skb\n");
557                 return skb;
558         }
559
560         NFCT_FRAG6_CB(clone)->orig = skb;
561
562         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
563                 pr_debug("message is too short.\n");
564                 goto ret_orig;
565         }
566
567         skb_set_transport_header(clone, fhoff);
568         hdr = ipv6_hdr(clone);
569         fhdr = (struct frag_hdr *)skb_transport_header(clone);
570
571         local_bh_disable();
572         inet_frag_evictor(&net->nf_frag.frags, &nf_frags, false);
573         local_bh_enable();
574
575         fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr);
576         if (fq == NULL) {
577                 pr_debug("Can't find and can't create new queue\n");
578                 goto ret_orig;
579         }
580
581         spin_lock_bh(&fq->q.lock);
582
583         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
584                 spin_unlock_bh(&fq->q.lock);
585                 pr_debug("Can't insert skb to queue\n");
586                 inet_frag_put(&fq->q, &nf_frags);
587                 goto ret_orig;
588         }
589
590         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
591             fq->q.meat == fq->q.len) {
592                 ret_skb = nf_ct_frag6_reasm(fq, dev);
593                 if (ret_skb == NULL)
594                         pr_debug("Can't reassemble fragmented packets\n");
595         }
596         spin_unlock_bh(&fq->q.lock);
597
598         inet_frag_put(&fq->q, &nf_frags);
599         return ret_skb;
600
601 ret_orig:
602         kfree_skb(clone);
603         return skb;
604 }
605
606 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
607                         struct net_device *in, struct net_device *out,
608                         int (*okfn)(struct sk_buff *))
609 {
610         struct sk_buff *s, *s2;
611         unsigned int ret = 0;
612
613         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
614                 nf_conntrack_put_reasm(s->nfct_reasm);
615                 nf_conntrack_get_reasm(skb);
616                 s->nfct_reasm = skb;
617
618                 s2 = s->next;
619                 s->next = NULL;
620
621                 if (ret != -ECANCELED)
622                         ret = NF_HOOK_THRESH(NFPROTO_IPV6, hooknum, s,
623                                              in, out, okfn,
624                                              NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
625                 else
626                         kfree_skb(s);
627
628                 s = s2;
629         }
630         nf_conntrack_put_reasm(skb);
631 }
632
633 static int nf_ct_net_init(struct net *net)
634 {
635         net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
636         net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
637         net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
638         inet_frags_init_net(&net->nf_frag.frags);
639
640         return nf_ct_frag6_sysctl_register(net);
641 }
642
643 static void nf_ct_net_exit(struct net *net)
644 {
645         nf_ct_frags6_sysctl_unregister(net);
646         inet_frags_exit_net(&net->nf_frag.frags, &nf_frags);
647 }
648
649 static struct pernet_operations nf_ct_net_ops = {
650         .init = nf_ct_net_init,
651         .exit = nf_ct_net_exit,
652 };
653
654 int nf_ct_frag6_init(void)
655 {
656         int ret = 0;
657
658         nf_frags.hashfn = nf_hashfn;
659         nf_frags.constructor = ip6_frag_init;
660         nf_frags.destructor = NULL;
661         nf_frags.skb_free = nf_skb_free;
662         nf_frags.qsize = sizeof(struct frag_queue);
663         nf_frags.match = ip6_frag_match;
664         nf_frags.frag_expire = nf_ct_frag6_expire;
665         nf_frags.secret_interval = 10 * 60 * HZ;
666         inet_frags_init(&nf_frags);
667
668         ret = register_pernet_subsys(&nf_ct_net_ops);
669         if (ret)
670                 inet_frags_fini(&nf_frags);
671
672         return ret;
673 }
674
675 void nf_ct_frag6_cleanup(void)
676 {
677         unregister_pernet_subsys(&nf_ct_net_ops);
678         inet_frags_fini(&nf_frags);
679 }