]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * The IP fragmentation functionality. | |
e905a9ed | 7 | * |
1da177e4 | 8 | * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG> |
113aa838 | 9 | * Alan Cox <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
10 | * |
11 | * Fixes: | |
12 | * Alan Cox : Split from ip.c , see ip_input.c for history. | |
13 | * David S. Miller : Begin massive cleanup... | |
14 | * Andi Kleen : Add sysctls. | |
15 | * xxxx : Overlapfrag bug. | |
16 | * Ultima : ip_expire() kernel panic. | |
17 | * Bill Hawes : Frag accounting and evictor fixes. | |
18 | * John McDonald : 0 length frag bug. | |
19 | * Alexey Kuznetsov: SMP races, threading, cleanup. | |
20 | * Patrick McHardy : LRU queue of frag heads for evictor. | |
21 | */ | |
22 | ||
89cee8b1 | 23 | #include <linux/compiler.h> |
1da177e4 LT |
24 | #include <linux/module.h> |
25 | #include <linux/types.h> | |
26 | #include <linux/mm.h> | |
27 | #include <linux/jiffies.h> | |
28 | #include <linux/skbuff.h> | |
29 | #include <linux/list.h> | |
30 | #include <linux/ip.h> | |
31 | #include <linux/icmp.h> | |
32 | #include <linux/netdevice.h> | |
33 | #include <linux/jhash.h> | |
34 | #include <linux/random.h> | |
5a0e3ad6 | 35 | #include <linux/slab.h> |
e9017b55 SW |
36 | #include <net/route.h> |
37 | #include <net/dst.h> | |
1da177e4 LT |
38 | #include <net/sock.h> |
39 | #include <net/ip.h> | |
40 | #include <net/icmp.h> | |
41 | #include <net/checksum.h> | |
89cee8b1 | 42 | #include <net/inetpeer.h> |
5ab11c98 | 43 | #include <net/inet_frag.h> |
1da177e4 LT |
44 | #include <linux/tcp.h> |
45 | #include <linux/udp.h> | |
46 | #include <linux/inet.h> | |
47 | #include <linux/netfilter_ipv4.h> | |
48 | ||
49 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
50 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
51 | * as well. Or notify me, at least. --ANK | |
52 | */ | |
53 | ||
8d8354d2 | 54 | static int sysctl_ipfrag_max_dist __read_mostly = 64; |
89cee8b1 | 55 | |
1da177e4 LT |
56 | struct ipfrag_skb_cb |
57 | { | |
58 | struct inet_skb_parm h; | |
59 | int offset; | |
60 | }; | |
61 | ||
fd3f8c4c | 62 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb)) |
1da177e4 LT |
63 | |
64 | /* Describe an entry in the "incomplete datagrams" queue. */ | |
65 | struct ipq { | |
5ab11c98 PE |
66 | struct inet_frag_queue q; |
67 | ||
1da177e4 | 68 | u32 user; |
18277770 AV |
69 | __be32 saddr; |
70 | __be32 daddr; | |
71 | __be16 id; | |
1da177e4 | 72 | u8 protocol; |
89cee8b1 HX |
73 | int iif; |
74 | unsigned int rid; | |
75 | struct inet_peer *peer; | |
1da177e4 LT |
76 | }; |
77 | ||
7eb95156 | 78 | static struct inet_frags ip4_frags; |
1da177e4 | 79 | |
e5a2bb84 | 80 | int ip_frag_nqueues(struct net *net) |
7eb95156 | 81 | { |
e5a2bb84 | 82 | return net->ipv4.frags.nqueues; |
7eb95156 | 83 | } |
1da177e4 | 84 | |
6ddc0822 | 85 | int ip_frag_mem(struct net *net) |
7eb95156 | 86 | { |
6ddc0822 | 87 | return atomic_read(&net->ipv4.frags.mem); |
7eb95156 | 88 | } |
1da177e4 | 89 | |
1706d587 HX |
90 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
91 | struct net_device *dev); | |
92 | ||
c6fda282 PE |
93 | struct ip4_create_arg { |
94 | struct iphdr *iph; | |
95 | u32 user; | |
96 | }; | |
97 | ||
18277770 | 98 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 99 | { |
18277770 AV |
100 | return jhash_3words((__force u32)id << 16 | prot, |
101 | (__force u32)saddr, (__force u32)daddr, | |
7eb95156 | 102 | ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
103 | } |
104 | ||
321a3a99 | 105 | static unsigned int ip4_hashfn(struct inet_frag_queue *q) |
1da177e4 | 106 | { |
321a3a99 | 107 | struct ipq *ipq; |
1da177e4 | 108 | |
321a3a99 PE |
109 | ipq = container_of(q, struct ipq, q); |
110 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
111 | } |
112 | ||
abd6523d PE |
113 | static int ip4_frag_match(struct inet_frag_queue *q, void *a) |
114 | { | |
115 | struct ipq *qp; | |
116 | struct ip4_create_arg *arg = a; | |
117 | ||
118 | qp = container_of(q, struct ipq, q); | |
119 | return (qp->id == arg->iph->id && | |
120 | qp->saddr == arg->iph->saddr && | |
121 | qp->daddr == arg->iph->daddr && | |
122 | qp->protocol == arg->iph->protocol && | |
123 | qp->user == arg->user); | |
124 | } | |
125 | ||
1da177e4 | 126 | /* Memory Tracking Functions. */ |
6ddc0822 PE |
127 | static __inline__ void frag_kfree_skb(struct netns_frags *nf, |
128 | struct sk_buff *skb, int *work) | |
1da177e4 LT |
129 | { |
130 | if (work) | |
131 | *work -= skb->truesize; | |
6ddc0822 | 132 | atomic_sub(skb->truesize, &nf->mem); |
1da177e4 LT |
133 | kfree_skb(skb); |
134 | } | |
135 | ||
c6fda282 PE |
136 | static void ip4_frag_init(struct inet_frag_queue *q, void *a) |
137 | { | |
138 | struct ipq *qp = container_of(q, struct ipq, q); | |
139 | struct ip4_create_arg *arg = a; | |
140 | ||
141 | qp->protocol = arg->iph->protocol; | |
142 | qp->id = arg->iph->id; | |
143 | qp->saddr = arg->iph->saddr; | |
144 | qp->daddr = arg->iph->daddr; | |
145 | qp->user = arg->user; | |
146 | qp->peer = sysctl_ipfrag_max_dist ? | |
147 | inet_getpeer(arg->iph->saddr, 1) : NULL; | |
148 | } | |
149 | ||
1e4b8287 | 150 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 151 | { |
1e4b8287 PE |
152 | struct ipq *qp; |
153 | ||
154 | qp = container_of(q, struct ipq, q); | |
155 | if (qp->peer) | |
156 | inet_putpeer(qp->peer); | |
1da177e4 LT |
157 | } |
158 | ||
1da177e4 LT |
159 | |
160 | /* Destruction primitives. */ | |
161 | ||
4b6cb5d8 | 162 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 163 | { |
762cc408 | 164 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
165 | } |
166 | ||
167 | /* Kill ipq entry. It is not destroyed immediately, | |
168 | * because caller (and someone more) holds reference count. | |
169 | */ | |
170 | static void ipq_kill(struct ipq *ipq) | |
171 | { | |
277e650d | 172 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
173 | } |
174 | ||
e905a9ed | 175 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
176 | * fragment queue until we are back under the threshold. |
177 | */ | |
6ddc0822 | 178 | static void ip_evictor(struct net *net) |
1da177e4 | 179 | { |
8e7999c4 PE |
180 | int evicted; |
181 | ||
6ddc0822 | 182 | evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags); |
8e7999c4 | 183 | if (evicted) |
c5346fe3 | 184 | IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted); |
1da177e4 LT |
185 | } |
186 | ||
187 | /* | |
188 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
189 | */ | |
190 | static void ip_expire(unsigned long arg) | |
191 | { | |
e521db9d | 192 | struct ipq *qp; |
84a3aa00 | 193 | struct net *net; |
e521db9d PE |
194 | |
195 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 196 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 197 | |
5ab11c98 | 198 | spin_lock(&qp->q.lock); |
1da177e4 | 199 | |
bc578a54 | 200 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
201 | goto out; |
202 | ||
203 | ipq_kill(qp); | |
204 | ||
7c73a6fa PE |
205 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); |
206 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
1da177e4 | 207 | |
bc578a54 | 208 | if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) { |
5ab11c98 | 209 | struct sk_buff *head = qp->q.fragments; |
cb84663e | 210 | |
69df9d59 ED |
211 | rcu_read_lock(); |
212 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
213 | if (!head->dev) |
214 | goto out_rcu_unlock; | |
215 | ||
216 | /* | |
217 | * Only search router table for the head fragment, | |
218 | * when defraging timeout at PRE_ROUTING HOOK. | |
219 | */ | |
220 | if (qp->user == IP_DEFRAG_CONNTRACK_IN && !skb_dst(head)) { | |
221 | const struct iphdr *iph = ip_hdr(head); | |
222 | int err = ip_route_input(head, iph->daddr, iph->saddr, | |
223 | iph->tos, head->dev); | |
224 | if (unlikely(err)) | |
225 | goto out_rcu_unlock; | |
226 | ||
227 | /* | |
228 | * Only an end host needs to send an ICMP | |
229 | * "Fragment Reassembly Timeout" message, per RFC792. | |
230 | */ | |
231 | if (skb_rtable(head)->rt_type != RTN_LOCAL) | |
232 | goto out_rcu_unlock; | |
233 | ||
234 | } | |
235 | ||
236 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ | |
237 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 238 | out_rcu_unlock: |
d1c9ae6d PM |
239 | rcu_read_unlock(); |
240 | } | |
1da177e4 | 241 | out: |
5ab11c98 | 242 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 243 | ipq_put(qp); |
1da177e4 LT |
244 | } |
245 | ||
abd6523d PE |
246 | /* Find the correct entry in the "incomplete datagrams" queue for |
247 | * this IP datagram, and create new one, if nothing is found. | |
248 | */ | |
ac18e750 | 249 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 250 | { |
c6fda282 PE |
251 | struct inet_frag_queue *q; |
252 | struct ip4_create_arg arg; | |
abd6523d | 253 | unsigned int hash; |
1da177e4 | 254 | |
c6fda282 PE |
255 | arg.iph = iph; |
256 | arg.user = user; | |
9a375803 PE |
257 | |
258 | read_lock(&ip4_frags.lock); | |
abd6523d | 259 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 260 | |
ac18e750 | 261 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
c6fda282 PE |
262 | if (q == NULL) |
263 | goto out_nomem; | |
1da177e4 | 264 | |
c6fda282 | 265 | return container_of(q, struct ipq, q); |
1da177e4 LT |
266 | |
267 | out_nomem: | |
64ce2073 | 268 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
269 | return NULL; |
270 | } | |
271 | ||
89cee8b1 HX |
272 | /* Is the fragment too far ahead to be part of ipq? */ |
273 | static inline int ip_frag_too_far(struct ipq *qp) | |
274 | { | |
275 | struct inet_peer *peer = qp->peer; | |
276 | unsigned int max = sysctl_ipfrag_max_dist; | |
277 | unsigned int start, end; | |
278 | ||
279 | int rc; | |
280 | ||
281 | if (!peer || !max) | |
282 | return 0; | |
283 | ||
284 | start = qp->rid; | |
285 | end = atomic_inc_return(&peer->rid); | |
286 | qp->rid = end; | |
287 | ||
5ab11c98 | 288 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
289 | |
290 | if (rc) { | |
7c73a6fa PE |
291 | struct net *net; |
292 | ||
293 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
294 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
295 | } |
296 | ||
297 | return rc; | |
298 | } | |
299 | ||
300 | static int ip_frag_reinit(struct ipq *qp) | |
301 | { | |
302 | struct sk_buff *fp; | |
303 | ||
b2fd5321 | 304 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 305 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
306 | return -ETIMEDOUT; |
307 | } | |
308 | ||
5ab11c98 | 309 | fp = qp->q.fragments; |
89cee8b1 HX |
310 | do { |
311 | struct sk_buff *xp = fp->next; | |
6ddc0822 | 312 | frag_kfree_skb(qp->q.net, fp, NULL); |
89cee8b1 HX |
313 | fp = xp; |
314 | } while (fp); | |
315 | ||
5ab11c98 PE |
316 | qp->q.last_in = 0; |
317 | qp->q.len = 0; | |
318 | qp->q.meat = 0; | |
319 | qp->q.fragments = NULL; | |
89cee8b1 HX |
320 | qp->iif = 0; |
321 | ||
322 | return 0; | |
323 | } | |
324 | ||
1da177e4 | 325 | /* Add new segment to existing queue. */ |
1706d587 | 326 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
327 | { |
328 | struct sk_buff *prev, *next; | |
1706d587 | 329 | struct net_device *dev; |
1da177e4 LT |
330 | int flags, offset; |
331 | int ihl, end; | |
1706d587 | 332 | int err = -ENOENT; |
1da177e4 | 333 | |
bc578a54 | 334 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
335 | goto err; |
336 | ||
89cee8b1 | 337 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
338 | unlikely(ip_frag_too_far(qp)) && |
339 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
340 | ipq_kill(qp); |
341 | goto err; | |
342 | } | |
343 | ||
eddc9ec5 | 344 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
345 | flags = offset & ~IP_OFFSET; |
346 | offset &= IP_OFFSET; | |
347 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 348 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
349 | |
350 | /* Determine the position of this fragment. */ | |
e905a9ed | 351 | end = offset + skb->len - ihl; |
1706d587 | 352 | err = -EINVAL; |
1da177e4 LT |
353 | |
354 | /* Is this the final fragment? */ | |
355 | if ((flags & IP_MF) == 0) { | |
356 | /* If we already have some bits beyond end | |
357 | * or have different end, the segment is corrrupted. | |
358 | */ | |
5ab11c98 | 359 | if (end < qp->q.len || |
bc578a54 | 360 | ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 361 | goto err; |
bc578a54 | 362 | qp->q.last_in |= INET_FRAG_LAST_IN; |
5ab11c98 | 363 | qp->q.len = end; |
1da177e4 LT |
364 | } else { |
365 | if (end&7) { | |
366 | end &= ~7; | |
367 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
368 | skb->ip_summed = CHECKSUM_NONE; | |
369 | } | |
5ab11c98 | 370 | if (end > qp->q.len) { |
1da177e4 | 371 | /* Some bits beyond end -> corruption. */ |
bc578a54 | 372 | if (qp->q.last_in & INET_FRAG_LAST_IN) |
1da177e4 | 373 | goto err; |
5ab11c98 | 374 | qp->q.len = end; |
1da177e4 LT |
375 | } |
376 | } | |
377 | if (end == offset) | |
378 | goto err; | |
379 | ||
1706d587 | 380 | err = -ENOMEM; |
1da177e4 LT |
381 | if (pskb_pull(skb, ihl) == NULL) |
382 | goto err; | |
1706d587 HX |
383 | |
384 | err = pskb_trim_rcsum(skb, end - offset); | |
385 | if (err) | |
1da177e4 LT |
386 | goto err; |
387 | ||
388 | /* Find out which fragments are in front and at the back of us | |
389 | * in the chain of fragments so far. We must know where to put | |
390 | * this fragment, right? | |
391 | */ | |
392 | prev = NULL; | |
5ab11c98 | 393 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
394 | if (FRAG_CB(next)->offset >= offset) |
395 | break; /* bingo! */ | |
396 | prev = next; | |
397 | } | |
398 | ||
399 | /* We found where to put this one. Check for overlap with | |
400 | * preceding fragment, and, if needed, align things so that | |
401 | * any overlaps are eliminated. | |
402 | */ | |
403 | if (prev) { | |
404 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
405 | ||
406 | if (i > 0) { | |
407 | offset += i; | |
1706d587 | 408 | err = -EINVAL; |
1da177e4 LT |
409 | if (end <= offset) |
410 | goto err; | |
1706d587 | 411 | err = -ENOMEM; |
1da177e4 LT |
412 | if (!pskb_pull(skb, i)) |
413 | goto err; | |
414 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
415 | skb->ip_summed = CHECKSUM_NONE; | |
416 | } | |
417 | } | |
418 | ||
1706d587 HX |
419 | err = -ENOMEM; |
420 | ||
1da177e4 LT |
421 | while (next && FRAG_CB(next)->offset < end) { |
422 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
423 | ||
424 | if (i < next->len) { | |
425 | /* Eat head of the next overlapped fragment | |
426 | * and leave the loop. The next ones cannot overlap. | |
427 | */ | |
428 | if (!pskb_pull(next, i)) | |
429 | goto err; | |
430 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 431 | qp->q.meat -= i; |
1da177e4 LT |
432 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
433 | next->ip_summed = CHECKSUM_NONE; | |
434 | break; | |
435 | } else { | |
436 | struct sk_buff *free_it = next; | |
437 | ||
47c6bf77 | 438 | /* Old fragment is completely overridden with |
1da177e4 LT |
439 | * new one drop it. |
440 | */ | |
441 | next = next->next; | |
442 | ||
443 | if (prev) | |
444 | prev->next = next; | |
445 | else | |
5ab11c98 | 446 | qp->q.fragments = next; |
1da177e4 | 447 | |
5ab11c98 | 448 | qp->q.meat -= free_it->len; |
6ddc0822 | 449 | frag_kfree_skb(qp->q.net, free_it, NULL); |
1da177e4 LT |
450 | } |
451 | } | |
452 | ||
453 | FRAG_CB(skb)->offset = offset; | |
454 | ||
455 | /* Insert this fragment in the chain of fragments. */ | |
456 | skb->next = next; | |
457 | if (prev) | |
458 | prev->next = skb; | |
459 | else | |
5ab11c98 | 460 | qp->q.fragments = skb; |
1da177e4 | 461 | |
1706d587 HX |
462 | dev = skb->dev; |
463 | if (dev) { | |
464 | qp->iif = dev->ifindex; | |
465 | skb->dev = NULL; | |
466 | } | |
5ab11c98 PE |
467 | qp->q.stamp = skb->tstamp; |
468 | qp->q.meat += skb->len; | |
6ddc0822 | 469 | atomic_add(skb->truesize, &qp->q.net->mem); |
1da177e4 | 470 | if (offset == 0) |
bc578a54 | 471 | qp->q.last_in |= INET_FRAG_FIRST_IN; |
1da177e4 | 472 | |
bc578a54 JP |
473 | if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
474 | qp->q.meat == qp->q.len) | |
1706d587 HX |
475 | return ip_frag_reasm(qp, prev, dev); |
476 | ||
7eb95156 | 477 | write_lock(&ip4_frags.lock); |
3140c25c | 478 | list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list); |
7eb95156 | 479 | write_unlock(&ip4_frags.lock); |
1706d587 | 480 | return -EINPROGRESS; |
1da177e4 LT |
481 | |
482 | err: | |
483 | kfree_skb(skb); | |
1706d587 | 484 | return err; |
1da177e4 LT |
485 | } |
486 | ||
487 | ||
488 | /* Build a new IP datagram from all its fragments. */ | |
489 | ||
1706d587 HX |
490 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
491 | struct net_device *dev) | |
1da177e4 | 492 | { |
2bad35b7 | 493 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 494 | struct iphdr *iph; |
5ab11c98 | 495 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
496 | int len; |
497 | int ihlen; | |
1706d587 | 498 | int err; |
1da177e4 LT |
499 | |
500 | ipq_kill(qp); | |
501 | ||
1706d587 HX |
502 | /* Make the one we just received the head. */ |
503 | if (prev) { | |
504 | head = prev->next; | |
505 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
506 | if (!fp) |
507 | goto out_nomem; | |
508 | ||
509 | fp->next = head->next; | |
510 | prev->next = fp; | |
511 | ||
5ab11c98 PE |
512 | skb_morph(head, qp->q.fragments); |
513 | head->next = qp->q.fragments->next; | |
1706d587 | 514 | |
5ab11c98 PE |
515 | kfree_skb(qp->q.fragments); |
516 | qp->q.fragments = head; | |
1706d587 HX |
517 | } |
518 | ||
547b792c IJ |
519 | WARN_ON(head == NULL); |
520 | WARN_ON(FRAG_CB(head)->offset != 0); | |
1da177e4 LT |
521 | |
522 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 523 | ihlen = ip_hdrlen(head); |
5ab11c98 | 524 | len = ihlen + qp->q.len; |
1da177e4 | 525 | |
1706d587 | 526 | err = -E2BIG; |
132adf54 | 527 | if (len > 65535) |
1da177e4 LT |
528 | goto out_oversize; |
529 | ||
530 | /* Head of list must not be cloned. */ | |
531 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
532 | goto out_nomem; | |
533 | ||
534 | /* If the first fragment is fragmented itself, we split | |
535 | * it to two chunks: the first with data and paged part | |
536 | * and the second, holding only fragments. */ | |
d7fcf1a5 | 537 | if (skb_has_frags(head)) { |
1da177e4 LT |
538 | struct sk_buff *clone; |
539 | int i, plen = 0; | |
540 | ||
541 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
542 | goto out_nomem; | |
543 | clone->next = head->next; | |
544 | head->next = clone; | |
545 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 546 | skb_frag_list_init(head); |
1da177e4 LT |
547 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) |
548 | plen += skb_shinfo(head)->frags[i].size; | |
549 | clone->len = clone->data_len = head->data_len - plen; | |
550 | head->data_len -= clone->len; | |
551 | head->len -= clone->len; | |
552 | clone->csum = 0; | |
553 | clone->ip_summed = head->ip_summed; | |
6ddc0822 | 554 | atomic_add(clone->truesize, &qp->q.net->mem); |
1da177e4 LT |
555 | } |
556 | ||
557 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 558 | skb_push(head, head->data - skb_network_header(head)); |
6ddc0822 | 559 | atomic_sub(head->truesize, &qp->q.net->mem); |
1da177e4 LT |
560 | |
561 | for (fp=head->next; fp; fp = fp->next) { | |
562 | head->data_len += fp->len; | |
563 | head->len += fp->len; | |
564 | if (head->ip_summed != fp->ip_summed) | |
565 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 566 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
567 | head->csum = csum_add(head->csum, fp->csum); |
568 | head->truesize += fp->truesize; | |
6ddc0822 | 569 | atomic_sub(fp->truesize, &qp->q.net->mem); |
1da177e4 LT |
570 | } |
571 | ||
572 | head->next = NULL; | |
573 | head->dev = dev; | |
5ab11c98 | 574 | head->tstamp = qp->q.stamp; |
1da177e4 | 575 | |
eddc9ec5 | 576 | iph = ip_hdr(head); |
1da177e4 LT |
577 | iph->frag_off = 0; |
578 | iph->tot_len = htons(len); | |
2bad35b7 | 579 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 580 | qp->q.fragments = NULL; |
1706d587 | 581 | return 0; |
1da177e4 LT |
582 | |
583 | out_nomem: | |
e905a9ed | 584 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 585 | "queue %p\n", qp); |
45542479 | 586 | err = -ENOMEM; |
1da177e4 LT |
587 | goto out_fail; |
588 | out_oversize: | |
589 | if (net_ratelimit()) | |
673d57e7 HH |
590 | printk(KERN_INFO "Oversized IP packet from %pI4.\n", |
591 | &qp->saddr); | |
1da177e4 | 592 | out_fail: |
bbf31bf1 | 593 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 594 | return err; |
1da177e4 LT |
595 | } |
596 | ||
597 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 598 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 599 | { |
1da177e4 | 600 | struct ipq *qp; |
ac18e750 | 601 | struct net *net; |
e905a9ed | 602 | |
adf30907 | 603 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); |
7c73a6fa | 604 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
605 | |
606 | /* Start by cleaning up the memory. */ | |
e31e0bdc | 607 | if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh) |
6ddc0822 | 608 | ip_evictor(net); |
1da177e4 | 609 | |
1da177e4 | 610 | /* Lookup (or create) queue header */ |
ac18e750 | 611 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 612 | int ret; |
1da177e4 | 613 | |
5ab11c98 | 614 | spin_lock(&qp->q.lock); |
1da177e4 | 615 | |
1706d587 | 616 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 617 | |
5ab11c98 | 618 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 619 | ipq_put(qp); |
776c729e | 620 | return ret; |
1da177e4 LT |
621 | } |
622 | ||
7c73a6fa | 623 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 624 | kfree_skb(skb); |
776c729e | 625 | return -ENOMEM; |
1da177e4 LT |
626 | } |
627 | ||
8d8354d2 PE |
628 | #ifdef CONFIG_SYSCTL |
629 | static int zero; | |
630 | ||
0a64b4b8 | 631 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 632 | { |
8d8354d2 | 633 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 634 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
635 | .maxlen = sizeof(int), |
636 | .mode = 0644, | |
6d9f239a | 637 | .proc_handler = proc_dointvec |
8d8354d2 PE |
638 | }, |
639 | { | |
8d8354d2 | 640 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 641 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
642 | .maxlen = sizeof(int), |
643 | .mode = 0644, | |
6d9f239a | 644 | .proc_handler = proc_dointvec |
8d8354d2 PE |
645 | }, |
646 | { | |
8d8354d2 | 647 | .procname = "ipfrag_time", |
b2fd5321 | 648 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
649 | .maxlen = sizeof(int), |
650 | .mode = 0644, | |
6d9f239a | 651 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 652 | }, |
7d291ebb PE |
653 | { } |
654 | }; | |
655 | ||
656 | static struct ctl_table ip4_frags_ctl_table[] = { | |
8d8354d2 | 657 | { |
8d8354d2 | 658 | .procname = "ipfrag_secret_interval", |
3b4bc4a2 | 659 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
660 | .maxlen = sizeof(int), |
661 | .mode = 0644, | |
6d9f239a | 662 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 PE |
663 | }, |
664 | { | |
665 | .procname = "ipfrag_max_dist", | |
666 | .data = &sysctl_ipfrag_max_dist, | |
667 | .maxlen = sizeof(int), | |
668 | .mode = 0644, | |
6d9f239a | 669 | .proc_handler = proc_dointvec_minmax, |
8d8354d2 PE |
670 | .extra1 = &zero |
671 | }, | |
672 | { } | |
673 | }; | |
674 | ||
2c8c1e72 | 675 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 676 | { |
e4a2d5c2 | 677 | struct ctl_table *table; |
8d8354d2 PE |
678 | struct ctl_table_header *hdr; |
679 | ||
0a64b4b8 | 680 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 681 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 682 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
e4a2d5c2 PE |
683 | if (table == NULL) |
684 | goto err_alloc; | |
685 | ||
e31e0bdc PE |
686 | table[0].data = &net->ipv4.frags.high_thresh; |
687 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 688 | table[2].data = &net->ipv4.frags.timeout; |
e4a2d5c2 PE |
689 | } |
690 | ||
691 | hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table); | |
692 | if (hdr == NULL) | |
693 | goto err_reg; | |
694 | ||
695 | net->ipv4.frags_hdr = hdr; | |
696 | return 0; | |
697 | ||
698 | err_reg: | |
09ad9bc7 | 699 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
700 | kfree(table); |
701 | err_alloc: | |
702 | return -ENOMEM; | |
703 | } | |
704 | ||
2c8c1e72 | 705 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
706 | { |
707 | struct ctl_table *table; | |
708 | ||
709 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
710 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
711 | kfree(table); | |
8d8354d2 | 712 | } |
7d291ebb PE |
713 | |
714 | static void ip4_frags_ctl_register(void) | |
715 | { | |
716 | register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table); | |
717 | } | |
8d8354d2 | 718 | #else |
0a64b4b8 | 719 | static inline int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
720 | { |
721 | return 0; | |
722 | } | |
e4a2d5c2 | 723 | |
0a64b4b8 | 724 | static inline void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
725 | { |
726 | } | |
7d291ebb PE |
727 | |
728 | static inline void ip4_frags_ctl_register(void) | |
729 | { | |
730 | } | |
8d8354d2 PE |
731 | #endif |
732 | ||
2c8c1e72 | 733 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 734 | { |
e31e0bdc PE |
735 | /* |
736 | * Fragment cache limits. We will commit 256K at one time. Should we | |
737 | * cross that limit we will prune down to 192K. This should cope with | |
738 | * even the most extreme cases without allowing an attacker to | |
739 | * measurably harm machine performance. | |
740 | */ | |
741 | net->ipv4.frags.high_thresh = 256 * 1024; | |
742 | net->ipv4.frags.low_thresh = 192 * 1024; | |
b2fd5321 PE |
743 | /* |
744 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
745 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
746 | * by TTL. | |
747 | */ | |
748 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
749 | ||
e5a2bb84 PE |
750 | inet_frags_init_net(&net->ipv4.frags); |
751 | ||
0a64b4b8 | 752 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
753 | } |
754 | ||
2c8c1e72 | 755 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 756 | { |
0a64b4b8 | 757 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
758 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
759 | } | |
760 | ||
761 | static struct pernet_operations ip4_frags_ops = { | |
762 | .init = ipv4_frags_init_net, | |
763 | .exit = ipv4_frags_exit_net, | |
764 | }; | |
765 | ||
b7aa0bf7 | 766 | void __init ipfrag_init(void) |
1da177e4 | 767 | { |
7d291ebb | 768 | ip4_frags_ctl_register(); |
81566e83 | 769 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 770 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 771 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
772 | ip4_frags.destructor = ip4_frag_free; |
773 | ip4_frags.skb_free = NULL; | |
774 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 775 | ip4_frags.match = ip4_frag_match; |
e521db9d | 776 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 777 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 778 | inet_frags_init(&ip4_frags); |
1da177e4 LT |
779 | } |
780 | ||
781 | EXPORT_SYMBOL(ip_defrag); |