[INET]: Consolidate frag queues freeing
[safe/jmp/linux-2.6] / net / ipv6 / reassembly.c
1 /*
2  *      IPv6 fragment reassembly
3  *      Linux INET6 implementation
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>
7  *
8  *      $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
9  *
10  *      Based on: net/ipv4/ip_fragment.c
11  *
12  *      This program is free software; you can redistribute it and/or
13  *      modify it under the terms of the GNU General Public License
14  *      as published by the Free Software Foundation; either version
15  *      2 of the License, or (at your option) any later version.
16  */
17
18 /*
19  *      Fixes:
20  *      Andi Kleen      Make it work with multiple hosts.
21  *                      More RFC compliance.
22  *
23  *      Horst von Brand Add missing #include <linux/string.h>
24  *      Alexey Kuznetsov        SMP races, threading, cleanup.
25  *      Patrick McHardy         LRU queue of frag heads for evictor.
26  *      Mitsuru KANDA @USAGI    Register inet6_protocol{}.
27  *      David Stevens and
28  *      YOSHIFUJI,H. @USAGI     Always remove fragment header to
29  *                              calculate ICV correctly.
30  */
31 #include <linux/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/sockios.h>
36 #include <linux/jiffies.h>
37 #include <linux/net.h>
38 #include <linux/list.h>
39 #include <linux/netdevice.h>
40 #include <linux/in6.h>
41 #include <linux/ipv6.h>
42 #include <linux/icmpv6.h>
43 #include <linux/random.h>
44 #include <linux/jhash.h>
45 #include <linux/skbuff.h>
46
47 #include <net/sock.h>
48 #include <net/snmp.h>
49
50 #include <net/ipv6.h>
51 #include <net/ip6_route.h>
52 #include <net/protocol.h>
53 #include <net/transp_v6.h>
54 #include <net/rawv6.h>
55 #include <net/ndisc.h>
56 #include <net/addrconf.h>
57 #include <net/inet_frag.h>
58
59 struct ip6frag_skb_cb
60 {
61         struct inet6_skb_parm   h;
62         int                     offset;
63 };
64
65 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
66
67
68 /*
69  *      Equivalent of ipv4 struct ipq
70  */
71
72 struct frag_queue
73 {
74         struct inet_frag_queue  q;
75
76         __be32                  id;             /* fragment id          */
77         struct in6_addr         saddr;
78         struct in6_addr         daddr;
79
80         int                     iif;
81         unsigned int            csum;
82         __u16                   nhoffset;
83 };
84
85 struct inet_frags_ctl ip6_frags_ctl __read_mostly = {
86         .high_thresh     = 256 * 1024,
87         .low_thresh      = 192 * 1024,
88         .timeout         = IPV6_FRAG_TIMEOUT,
89         .secret_interval = 10 * 60 * HZ,
90 };
91
92 static struct inet_frags ip6_frags;
93
94 int ip6_frag_nqueues(void)
95 {
96         return ip6_frags.nqueues;
97 }
98
99 int ip6_frag_mem(void)
100 {
101         return atomic_read(&ip6_frags.mem);
102 }
103
104 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
105                           struct net_device *dev);
106
107 /*
108  * callers should be careful not to use the hash value outside the ipfrag_lock
109  * as doing so could race with ipfrag_hash_rnd being recalculated.
110  */
111 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
112                                struct in6_addr *daddr)
113 {
114         u32 a, b, c;
115
116         a = (__force u32)saddr->s6_addr32[0];
117         b = (__force u32)saddr->s6_addr32[1];
118         c = (__force u32)saddr->s6_addr32[2];
119
120         a += JHASH_GOLDEN_RATIO;
121         b += JHASH_GOLDEN_RATIO;
122         c += ip6_frags.rnd;
123         __jhash_mix(a, b, c);
124
125         a += (__force u32)saddr->s6_addr32[3];
126         b += (__force u32)daddr->s6_addr32[0];
127         c += (__force u32)daddr->s6_addr32[1];
128         __jhash_mix(a, b, c);
129
130         a += (__force u32)daddr->s6_addr32[2];
131         b += (__force u32)daddr->s6_addr32[3];
132         c += (__force u32)id;
133         __jhash_mix(a, b, c);
134
135         return c & (INETFRAGS_HASHSZ - 1);
136 }
137
138 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
139 {
140         struct frag_queue *fq;
141
142         fq = container_of(q, struct frag_queue, q);
143         return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
144 }
145
146 int ip6_frag_match(struct inet_frag_queue *q, void *a)
147 {
148         struct frag_queue *fq;
149         struct ip6_create_arg *arg = a;
150
151         fq = container_of(q, struct frag_queue, q);
152         return (fq->id == arg->id &&
153                         ipv6_addr_equal(&fq->saddr, arg->src) &&
154                         ipv6_addr_equal(&fq->daddr, arg->dst));
155 }
156 EXPORT_SYMBOL(ip6_frag_match);
157
158 /* Memory Tracking Functions. */
159 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
160 {
161         if (work)
162                 *work -= skb->truesize;
163         atomic_sub(skb->truesize, &ip6_frags.mem);
164         kfree_skb(skb);
165 }
166
167 void ip6_frag_init(struct inet_frag_queue *q, void *a)
168 {
169         struct frag_queue *fq = container_of(q, struct frag_queue, q);
170         struct ip6_create_arg *arg = a;
171
172         fq->id = arg->id;
173         ipv6_addr_copy(&fq->saddr, arg->src);
174         ipv6_addr_copy(&fq->daddr, arg->dst);
175 }
176 EXPORT_SYMBOL(ip6_frag_init);
177
178 /* Destruction primitives. */
179
180 static __inline__ void fq_put(struct frag_queue *fq)
181 {
182         inet_frag_put(&fq->q, &ip6_frags);
183 }
184
185 /* Kill fq entry. It is not destroyed immediately,
186  * because caller (and someone more) holds reference count.
187  */
188 static __inline__ void fq_kill(struct frag_queue *fq)
189 {
190         inet_frag_kill(&fq->q, &ip6_frags);
191 }
192
193 static void ip6_evictor(struct inet6_dev *idev)
194 {
195         int evicted;
196
197         evicted = inet_frag_evictor(&ip6_frags);
198         if (evicted)
199                 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
200 }
201
202 static void ip6_frag_expire(unsigned long data)
203 {
204         struct frag_queue *fq;
205         struct net_device *dev = NULL;
206
207         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
208
209         spin_lock(&fq->q.lock);
210
211         if (fq->q.last_in & COMPLETE)
212                 goto out;
213
214         fq_kill(fq);
215
216         dev = dev_get_by_index(&init_net, fq->iif);
217         if (!dev)
218                 goto out;
219
220         rcu_read_lock();
221         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
222         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
223         rcu_read_unlock();
224
225         /* Don't send error if the first segment did not arrive. */
226         if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
227                 goto out;
228
229         /*
230            But use as source device on which LAST ARRIVED
231            segment was received. And do not use fq->dev
232            pointer directly, device might already disappeared.
233          */
234         fq->q.fragments->dev = dev;
235         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
236 out:
237         if (dev)
238                 dev_put(dev);
239         spin_unlock(&fq->q.lock);
240         fq_put(fq);
241 }
242
243 static __inline__ struct frag_queue *
244 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
245         struct inet6_dev *idev)
246 {
247         struct inet_frag_queue *q;
248         struct ip6_create_arg arg;
249         unsigned int hash;
250
251         arg.id = id;
252         arg.src = src;
253         arg.dst = dst;
254         hash = ip6qhashfn(id, src, dst);
255
256         q = inet_frag_find(&ip6_frags, &arg, hash);
257         if (q == NULL)
258                 goto oom;
259
260         return container_of(q, struct frag_queue, q);
261
262 oom:
263         IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
264         return NULL;
265 }
266
267 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
268                            struct frag_hdr *fhdr, int nhoff)
269 {
270         struct sk_buff *prev, *next;
271         struct net_device *dev;
272         int offset, end;
273
274         if (fq->q.last_in & COMPLETE)
275                 goto err;
276
277         offset = ntohs(fhdr->frag_off) & ~0x7;
278         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
279                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
280
281         if ((unsigned int)end > IPV6_MAXPLEN) {
282                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
283                                  IPSTATS_MIB_INHDRERRORS);
284                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
285                                   ((u8 *)&fhdr->frag_off -
286                                    skb_network_header(skb)));
287                 return -1;
288         }
289
290         if (skb->ip_summed == CHECKSUM_COMPLETE) {
291                 const unsigned char *nh = skb_network_header(skb);
292                 skb->csum = csum_sub(skb->csum,
293                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
294                                                   0));
295         }
296
297         /* Is this the final fragment? */
298         if (!(fhdr->frag_off & htons(IP6_MF))) {
299                 /* If we already have some bits beyond end
300                  * or have different end, the segment is corrupted.
301                  */
302                 if (end < fq->q.len ||
303                     ((fq->q.last_in & LAST_IN) && end != fq->q.len))
304                         goto err;
305                 fq->q.last_in |= LAST_IN;
306                 fq->q.len = end;
307         } else {
308                 /* Check if the fragment is rounded to 8 bytes.
309                  * Required by the RFC.
310                  */
311                 if (end & 0x7) {
312                         /* RFC2460 says always send parameter problem in
313                          * this case. -DaveM
314                          */
315                         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
316                                          IPSTATS_MIB_INHDRERRORS);
317                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
318                                           offsetof(struct ipv6hdr, payload_len));
319                         return -1;
320                 }
321                 if (end > fq->q.len) {
322                         /* Some bits beyond end -> corruption. */
323                         if (fq->q.last_in & LAST_IN)
324                                 goto err;
325                         fq->q.len = end;
326                 }
327         }
328
329         if (end == offset)
330                 goto err;
331
332         /* Point into the IP datagram 'data' part. */
333         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
334                 goto err;
335
336         if (pskb_trim_rcsum(skb, end - offset))
337                 goto err;
338
339         /* Find out which fragments are in front and at the back of us
340          * in the chain of fragments so far.  We must know where to put
341          * this fragment, right?
342          */
343         prev = NULL;
344         for(next = fq->q.fragments; next != NULL; next = next->next) {
345                 if (FRAG6_CB(next)->offset >= offset)
346                         break;  /* bingo! */
347                 prev = next;
348         }
349
350         /* We found where to put this one.  Check for overlap with
351          * preceding fragment, and, if needed, align things so that
352          * any overlaps are eliminated.
353          */
354         if (prev) {
355                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
356
357                 if (i > 0) {
358                         offset += i;
359                         if (end <= offset)
360                                 goto err;
361                         if (!pskb_pull(skb, i))
362                                 goto err;
363                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
364                                 skb->ip_summed = CHECKSUM_NONE;
365                 }
366         }
367
368         /* Look for overlap with succeeding segments.
369          * If we can merge fragments, do it.
370          */
371         while (next && FRAG6_CB(next)->offset < end) {
372                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
373
374                 if (i < next->len) {
375                         /* Eat head of the next overlapped fragment
376                          * and leave the loop. The next ones cannot overlap.
377                          */
378                         if (!pskb_pull(next, i))
379                                 goto err;
380                         FRAG6_CB(next)->offset += i;    /* next fragment */
381                         fq->q.meat -= i;
382                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
383                                 next->ip_summed = CHECKSUM_NONE;
384                         break;
385                 } else {
386                         struct sk_buff *free_it = next;
387
388                         /* Old fragment is completely overridden with
389                          * new one drop it.
390                          */
391                         next = next->next;
392
393                         if (prev)
394                                 prev->next = next;
395                         else
396                                 fq->q.fragments = next;
397
398                         fq->q.meat -= free_it->len;
399                         frag_kfree_skb(free_it, NULL);
400                 }
401         }
402
403         FRAG6_CB(skb)->offset = offset;
404
405         /* Insert this fragment in the chain of fragments. */
406         skb->next = next;
407         if (prev)
408                 prev->next = skb;
409         else
410                 fq->q.fragments = skb;
411
412         dev = skb->dev;
413         if (dev) {
414                 fq->iif = dev->ifindex;
415                 skb->dev = NULL;
416         }
417         fq->q.stamp = skb->tstamp;
418         fq->q.meat += skb->len;
419         atomic_add(skb->truesize, &ip6_frags.mem);
420
421         /* The first fragment.
422          * nhoffset is obtained from the first fragment, of course.
423          */
424         if (offset == 0) {
425                 fq->nhoffset = nhoff;
426                 fq->q.last_in |= FIRST_IN;
427         }
428
429         if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
430                 return ip6_frag_reasm(fq, prev, dev);
431
432         write_lock(&ip6_frags.lock);
433         list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
434         write_unlock(&ip6_frags.lock);
435         return -1;
436
437 err:
438         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
439         kfree_skb(skb);
440         return -1;
441 }
442
443 /*
444  *      Check if this packet is complete.
445  *      Returns NULL on failure by any reason, and pointer
446  *      to current nexthdr field in reassembled frame.
447  *
448  *      It is called with locked fq, and caller must check that
449  *      queue is eligible for reassembly i.e. it is not COMPLETE,
450  *      the last and the first frames arrived and all the bits are here.
451  */
452 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
453                           struct net_device *dev)
454 {
455         struct sk_buff *fp, *head = fq->q.fragments;
456         int    payload_len;
457         unsigned int nhoff;
458
459         fq_kill(fq);
460
461         /* Make the one we just received the head. */
462         if (prev) {
463                 head = prev->next;
464                 fp = skb_clone(head, GFP_ATOMIC);
465
466                 if (!fp)
467                         goto out_oom;
468
469                 fp->next = head->next;
470                 prev->next = fp;
471
472                 skb_morph(head, fq->q.fragments);
473                 head->next = fq->q.fragments->next;
474
475                 kfree_skb(fq->q.fragments);
476                 fq->q.fragments = head;
477         }
478
479         BUG_TRAP(head != NULL);
480         BUG_TRAP(FRAG6_CB(head)->offset == 0);
481
482         /* Unfragmented part is taken from the first segment. */
483         payload_len = ((head->data - skb_network_header(head)) -
484                        sizeof(struct ipv6hdr) + fq->q.len -
485                        sizeof(struct frag_hdr));
486         if (payload_len > IPV6_MAXPLEN)
487                 goto out_oversize;
488
489         /* Head of list must not be cloned. */
490         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
491                 goto out_oom;
492
493         /* If the first fragment is fragmented itself, we split
494          * it to two chunks: the first with data and paged part
495          * and the second, holding only fragments. */
496         if (skb_shinfo(head)->frag_list) {
497                 struct sk_buff *clone;
498                 int i, plen = 0;
499
500                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
501                         goto out_oom;
502                 clone->next = head->next;
503                 head->next = clone;
504                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
505                 skb_shinfo(head)->frag_list = NULL;
506                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
507                         plen += skb_shinfo(head)->frags[i].size;
508                 clone->len = clone->data_len = head->data_len - plen;
509                 head->data_len -= clone->len;
510                 head->len -= clone->len;
511                 clone->csum = 0;
512                 clone->ip_summed = head->ip_summed;
513                 atomic_add(clone->truesize, &ip6_frags.mem);
514         }
515
516         /* We have to remove fragment header from datagram and to relocate
517          * header in order to calculate ICV correctly. */
518         nhoff = fq->nhoffset;
519         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
520         memmove(head->head + sizeof(struct frag_hdr), head->head,
521                 (head->data - head->head) - sizeof(struct frag_hdr));
522         head->mac_header += sizeof(struct frag_hdr);
523         head->network_header += sizeof(struct frag_hdr);
524
525         skb_shinfo(head)->frag_list = head->next;
526         skb_reset_transport_header(head);
527         skb_push(head, head->data - skb_network_header(head));
528         atomic_sub(head->truesize, &ip6_frags.mem);
529
530         for (fp=head->next; fp; fp = fp->next) {
531                 head->data_len += fp->len;
532                 head->len += fp->len;
533                 if (head->ip_summed != fp->ip_summed)
534                         head->ip_summed = CHECKSUM_NONE;
535                 else if (head->ip_summed == CHECKSUM_COMPLETE)
536                         head->csum = csum_add(head->csum, fp->csum);
537                 head->truesize += fp->truesize;
538                 atomic_sub(fp->truesize, &ip6_frags.mem);
539         }
540
541         head->next = NULL;
542         head->dev = dev;
543         head->tstamp = fq->q.stamp;
544         ipv6_hdr(head)->payload_len = htons(payload_len);
545         IP6CB(head)->nhoff = nhoff;
546
547         /* Yes, and fold redundant checksum back. 8) */
548         if (head->ip_summed == CHECKSUM_COMPLETE)
549                 head->csum = csum_partial(skb_network_header(head),
550                                           skb_network_header_len(head),
551                                           head->csum);
552
553         rcu_read_lock();
554         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
555         rcu_read_unlock();
556         fq->q.fragments = NULL;
557         return 1;
558
559 out_oversize:
560         if (net_ratelimit())
561                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
562         goto out_fail;
563 out_oom:
564         if (net_ratelimit())
565                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
566 out_fail:
567         rcu_read_lock();
568         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
569         rcu_read_unlock();
570         return -1;
571 }
572
573 static int ipv6_frag_rcv(struct sk_buff *skb)
574 {
575         struct frag_hdr *fhdr;
576         struct frag_queue *fq;
577         struct ipv6hdr *hdr = ipv6_hdr(skb);
578
579         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
580
581         /* Jumbo payload inhibits frag. header */
582         if (hdr->payload_len==0) {
583                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
584                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
585                                   skb_network_header_len(skb));
586                 return -1;
587         }
588         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
589                                  sizeof(struct frag_hdr)))) {
590                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
591                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
592                                   skb_network_header_len(skb));
593                 return -1;
594         }
595
596         hdr = ipv6_hdr(skb);
597         fhdr = (struct frag_hdr *)skb_transport_header(skb);
598
599         if (!(fhdr->frag_off & htons(0xFFF9))) {
600                 /* It is not a fragmented frame */
601                 skb->transport_header += sizeof(struct frag_hdr);
602                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
603
604                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
605                 return 1;
606         }
607
608         if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
609                 ip6_evictor(ip6_dst_idev(skb->dst));
610
611         if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
612                           ip6_dst_idev(skb->dst))) != NULL) {
613                 int ret;
614
615                 spin_lock(&fq->q.lock);
616
617                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
618
619                 spin_unlock(&fq->q.lock);
620                 fq_put(fq);
621                 return ret;
622         }
623
624         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
625         kfree_skb(skb);
626         return -1;
627 }
628
629 static struct inet6_protocol frag_protocol =
630 {
631         .handler        =       ipv6_frag_rcv,
632         .flags          =       INET6_PROTO_NOPOLICY,
633 };
634
635 void __init ipv6_frag_init(void)
636 {
637         if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
638                 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
639
640         ip6_frags.ctl = &ip6_frags_ctl;
641         ip6_frags.hashfn = ip6_hashfn;
642         ip6_frags.constructor = ip6_frag_init;
643         ip6_frags.destructor = NULL;
644         ip6_frags.skb_free = NULL;
645         ip6_frags.qsize = sizeof(struct frag_queue);
646         ip6_frags.match = ip6_frag_match;
647         ip6_frags.frag_expire = ip6_frag_expire;
648         inet_frags_init(&ip6_frags);
649 }