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); | |
a02cec21 | 119 | return qp->id == arg->iph->id && |
abd6523d PE |
120 | qp->saddr == arg->iph->saddr && |
121 | qp->daddr == arg->iph->daddr && | |
122 | qp->protocol == arg->iph->protocol && | |
a02cec21 | 123 | qp->user == arg->user; |
abd6523d PE |
124 | } |
125 | ||
1da177e4 | 126 | /* Memory Tracking Functions. */ |
a95d8c88 | 127 | static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb) |
1da177e4 | 128 | { |
6ddc0822 | 129 | atomic_sub(skb->truesize, &nf->mem); |
1da177e4 LT |
130 | kfree_skb(skb); |
131 | } | |
132 | ||
c6fda282 PE |
133 | static void ip4_frag_init(struct inet_frag_queue *q, void *a) |
134 | { | |
135 | struct ipq *qp = container_of(q, struct ipq, q); | |
136 | struct ip4_create_arg *arg = a; | |
137 | ||
138 | qp->protocol = arg->iph->protocol; | |
139 | qp->id = arg->iph->id; | |
140 | qp->saddr = arg->iph->saddr; | |
141 | qp->daddr = arg->iph->daddr; | |
142 | qp->user = arg->user; | |
143 | qp->peer = sysctl_ipfrag_max_dist ? | |
144 | inet_getpeer(arg->iph->saddr, 1) : NULL; | |
145 | } | |
146 | ||
1e4b8287 | 147 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 148 | { |
1e4b8287 PE |
149 | struct ipq *qp; |
150 | ||
151 | qp = container_of(q, struct ipq, q); | |
152 | if (qp->peer) | |
153 | inet_putpeer(qp->peer); | |
1da177e4 LT |
154 | } |
155 | ||
1da177e4 LT |
156 | |
157 | /* Destruction primitives. */ | |
158 | ||
4b6cb5d8 | 159 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 160 | { |
762cc408 | 161 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
162 | } |
163 | ||
164 | /* Kill ipq entry. It is not destroyed immediately, | |
165 | * because caller (and someone more) holds reference count. | |
166 | */ | |
167 | static void ipq_kill(struct ipq *ipq) | |
168 | { | |
277e650d | 169 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
170 | } |
171 | ||
e905a9ed | 172 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
173 | * fragment queue until we are back under the threshold. |
174 | */ | |
6ddc0822 | 175 | static void ip_evictor(struct net *net) |
1da177e4 | 176 | { |
8e7999c4 PE |
177 | int evicted; |
178 | ||
6ddc0822 | 179 | evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags); |
8e7999c4 | 180 | if (evicted) |
c5346fe3 | 181 | IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted); |
1da177e4 LT |
182 | } |
183 | ||
184 | /* | |
185 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
186 | */ | |
187 | static void ip_expire(unsigned long arg) | |
188 | { | |
e521db9d | 189 | struct ipq *qp; |
84a3aa00 | 190 | struct net *net; |
e521db9d PE |
191 | |
192 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 193 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 194 | |
5ab11c98 | 195 | spin_lock(&qp->q.lock); |
1da177e4 | 196 | |
bc578a54 | 197 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
198 | goto out; |
199 | ||
200 | ipq_kill(qp); | |
201 | ||
7c73a6fa PE |
202 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); |
203 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
1da177e4 | 204 | |
bc578a54 | 205 | if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) { |
5ab11c98 | 206 | struct sk_buff *head = qp->q.fragments; |
cb84663e | 207 | |
69df9d59 ED |
208 | rcu_read_lock(); |
209 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
210 | if (!head->dev) |
211 | goto out_rcu_unlock; | |
212 | ||
213 | /* | |
214 | * Only search router table for the head fragment, | |
215 | * when defraging timeout at PRE_ROUTING HOOK. | |
216 | */ | |
217 | if (qp->user == IP_DEFRAG_CONNTRACK_IN && !skb_dst(head)) { | |
218 | const struct iphdr *iph = ip_hdr(head); | |
219 | int err = ip_route_input(head, iph->daddr, iph->saddr, | |
220 | iph->tos, head->dev); | |
221 | if (unlikely(err)) | |
222 | goto out_rcu_unlock; | |
223 | ||
224 | /* | |
225 | * Only an end host needs to send an ICMP | |
226 | * "Fragment Reassembly Timeout" message, per RFC792. | |
227 | */ | |
228 | if (skb_rtable(head)->rt_type != RTN_LOCAL) | |
229 | goto out_rcu_unlock; | |
230 | ||
231 | } | |
232 | ||
233 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ | |
234 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 235 | out_rcu_unlock: |
d1c9ae6d PM |
236 | rcu_read_unlock(); |
237 | } | |
1da177e4 | 238 | out: |
5ab11c98 | 239 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 240 | ipq_put(qp); |
1da177e4 LT |
241 | } |
242 | ||
abd6523d PE |
243 | /* Find the correct entry in the "incomplete datagrams" queue for |
244 | * this IP datagram, and create new one, if nothing is found. | |
245 | */ | |
ac18e750 | 246 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 247 | { |
c6fda282 PE |
248 | struct inet_frag_queue *q; |
249 | struct ip4_create_arg arg; | |
abd6523d | 250 | unsigned int hash; |
1da177e4 | 251 | |
c6fda282 PE |
252 | arg.iph = iph; |
253 | arg.user = user; | |
9a375803 PE |
254 | |
255 | read_lock(&ip4_frags.lock); | |
abd6523d | 256 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 257 | |
ac18e750 | 258 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
c6fda282 PE |
259 | if (q == NULL) |
260 | goto out_nomem; | |
1da177e4 | 261 | |
c6fda282 | 262 | return container_of(q, struct ipq, q); |
1da177e4 LT |
263 | |
264 | out_nomem: | |
64ce2073 | 265 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
266 | return NULL; |
267 | } | |
268 | ||
89cee8b1 HX |
269 | /* Is the fragment too far ahead to be part of ipq? */ |
270 | static inline int ip_frag_too_far(struct ipq *qp) | |
271 | { | |
272 | struct inet_peer *peer = qp->peer; | |
273 | unsigned int max = sysctl_ipfrag_max_dist; | |
274 | unsigned int start, end; | |
275 | ||
276 | int rc; | |
277 | ||
278 | if (!peer || !max) | |
279 | return 0; | |
280 | ||
281 | start = qp->rid; | |
282 | end = atomic_inc_return(&peer->rid); | |
283 | qp->rid = end; | |
284 | ||
5ab11c98 | 285 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
286 | |
287 | if (rc) { | |
7c73a6fa PE |
288 | struct net *net; |
289 | ||
290 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
291 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
292 | } |
293 | ||
294 | return rc; | |
295 | } | |
296 | ||
297 | static int ip_frag_reinit(struct ipq *qp) | |
298 | { | |
299 | struct sk_buff *fp; | |
300 | ||
b2fd5321 | 301 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 302 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
303 | return -ETIMEDOUT; |
304 | } | |
305 | ||
5ab11c98 | 306 | fp = qp->q.fragments; |
89cee8b1 HX |
307 | do { |
308 | struct sk_buff *xp = fp->next; | |
a95d8c88 | 309 | frag_kfree_skb(qp->q.net, fp); |
89cee8b1 HX |
310 | fp = xp; |
311 | } while (fp); | |
312 | ||
5ab11c98 PE |
313 | qp->q.last_in = 0; |
314 | qp->q.len = 0; | |
315 | qp->q.meat = 0; | |
316 | qp->q.fragments = NULL; | |
d6bebca9 | 317 | qp->q.fragments_tail = NULL; |
89cee8b1 HX |
318 | qp->iif = 0; |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
1da177e4 | 323 | /* Add new segment to existing queue. */ |
1706d587 | 324 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
325 | { |
326 | struct sk_buff *prev, *next; | |
1706d587 | 327 | struct net_device *dev; |
1da177e4 LT |
328 | int flags, offset; |
329 | int ihl, end; | |
1706d587 | 330 | int err = -ENOENT; |
1da177e4 | 331 | |
bc578a54 | 332 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
333 | goto err; |
334 | ||
89cee8b1 | 335 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
336 | unlikely(ip_frag_too_far(qp)) && |
337 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
338 | ipq_kill(qp); |
339 | goto err; | |
340 | } | |
341 | ||
eddc9ec5 | 342 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
343 | flags = offset & ~IP_OFFSET; |
344 | offset &= IP_OFFSET; | |
345 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 346 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
347 | |
348 | /* Determine the position of this fragment. */ | |
e905a9ed | 349 | end = offset + skb->len - ihl; |
1706d587 | 350 | err = -EINVAL; |
1da177e4 LT |
351 | |
352 | /* Is this the final fragment? */ | |
353 | if ((flags & IP_MF) == 0) { | |
354 | /* If we already have some bits beyond end | |
355 | * or have different end, the segment is corrrupted. | |
356 | */ | |
5ab11c98 | 357 | if (end < qp->q.len || |
bc578a54 | 358 | ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 359 | goto err; |
bc578a54 | 360 | qp->q.last_in |= INET_FRAG_LAST_IN; |
5ab11c98 | 361 | qp->q.len = end; |
1da177e4 LT |
362 | } else { |
363 | if (end&7) { | |
364 | end &= ~7; | |
365 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
366 | skb->ip_summed = CHECKSUM_NONE; | |
367 | } | |
5ab11c98 | 368 | if (end > qp->q.len) { |
1da177e4 | 369 | /* Some bits beyond end -> corruption. */ |
bc578a54 | 370 | if (qp->q.last_in & INET_FRAG_LAST_IN) |
1da177e4 | 371 | goto err; |
5ab11c98 | 372 | qp->q.len = end; |
1da177e4 LT |
373 | } |
374 | } | |
375 | if (end == offset) | |
376 | goto err; | |
377 | ||
1706d587 | 378 | err = -ENOMEM; |
1da177e4 LT |
379 | if (pskb_pull(skb, ihl) == NULL) |
380 | goto err; | |
1706d587 HX |
381 | |
382 | err = pskb_trim_rcsum(skb, end - offset); | |
383 | if (err) | |
1da177e4 LT |
384 | goto err; |
385 | ||
386 | /* Find out which fragments are in front and at the back of us | |
387 | * in the chain of fragments so far. We must know where to put | |
388 | * this fragment, right? | |
389 | */ | |
d6bebca9 CG |
390 | prev = qp->q.fragments_tail; |
391 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
392 | next = NULL; | |
393 | goto found; | |
394 | } | |
1da177e4 | 395 | prev = NULL; |
5ab11c98 | 396 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
397 | if (FRAG_CB(next)->offset >= offset) |
398 | break; /* bingo! */ | |
399 | prev = next; | |
400 | } | |
401 | ||
d6bebca9 | 402 | found: |
1da177e4 LT |
403 | /* We found where to put this one. Check for overlap with |
404 | * preceding fragment, and, if needed, align things so that | |
405 | * any overlaps are eliminated. | |
406 | */ | |
407 | if (prev) { | |
408 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
409 | ||
410 | if (i > 0) { | |
411 | offset += i; | |
1706d587 | 412 | err = -EINVAL; |
1da177e4 LT |
413 | if (end <= offset) |
414 | goto err; | |
1706d587 | 415 | err = -ENOMEM; |
1da177e4 LT |
416 | if (!pskb_pull(skb, i)) |
417 | goto err; | |
418 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
419 | skb->ip_summed = CHECKSUM_NONE; | |
420 | } | |
421 | } | |
422 | ||
1706d587 HX |
423 | err = -ENOMEM; |
424 | ||
1da177e4 LT |
425 | while (next && FRAG_CB(next)->offset < end) { |
426 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
427 | ||
428 | if (i < next->len) { | |
429 | /* Eat head of the next overlapped fragment | |
430 | * and leave the loop. The next ones cannot overlap. | |
431 | */ | |
432 | if (!pskb_pull(next, i)) | |
433 | goto err; | |
434 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 435 | qp->q.meat -= i; |
1da177e4 LT |
436 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
437 | next->ip_summed = CHECKSUM_NONE; | |
438 | break; | |
439 | } else { | |
440 | struct sk_buff *free_it = next; | |
441 | ||
47c6bf77 | 442 | /* Old fragment is completely overridden with |
1da177e4 LT |
443 | * new one drop it. |
444 | */ | |
445 | next = next->next; | |
446 | ||
447 | if (prev) | |
448 | prev->next = next; | |
449 | else | |
5ab11c98 | 450 | qp->q.fragments = next; |
1da177e4 | 451 | |
5ab11c98 | 452 | qp->q.meat -= free_it->len; |
a95d8c88 | 453 | frag_kfree_skb(qp->q.net, free_it); |
1da177e4 LT |
454 | } |
455 | } | |
456 | ||
457 | FRAG_CB(skb)->offset = offset; | |
458 | ||
459 | /* Insert this fragment in the chain of fragments. */ | |
460 | skb->next = next; | |
d6bebca9 CG |
461 | if (!next) |
462 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
463 | if (prev) |
464 | prev->next = skb; | |
465 | else | |
5ab11c98 | 466 | qp->q.fragments = skb; |
1da177e4 | 467 | |
1706d587 HX |
468 | dev = skb->dev; |
469 | if (dev) { | |
470 | qp->iif = dev->ifindex; | |
471 | skb->dev = NULL; | |
472 | } | |
5ab11c98 PE |
473 | qp->q.stamp = skb->tstamp; |
474 | qp->q.meat += skb->len; | |
6ddc0822 | 475 | atomic_add(skb->truesize, &qp->q.net->mem); |
1da177e4 | 476 | if (offset == 0) |
bc578a54 | 477 | qp->q.last_in |= INET_FRAG_FIRST_IN; |
1da177e4 | 478 | |
bc578a54 JP |
479 | if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
480 | qp->q.meat == qp->q.len) | |
1706d587 HX |
481 | return ip_frag_reasm(qp, prev, dev); |
482 | ||
7eb95156 | 483 | write_lock(&ip4_frags.lock); |
3140c25c | 484 | list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list); |
7eb95156 | 485 | write_unlock(&ip4_frags.lock); |
1706d587 | 486 | return -EINPROGRESS; |
1da177e4 LT |
487 | |
488 | err: | |
489 | kfree_skb(skb); | |
1706d587 | 490 | return err; |
1da177e4 LT |
491 | } |
492 | ||
493 | ||
494 | /* Build a new IP datagram from all its fragments. */ | |
495 | ||
1706d587 HX |
496 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
497 | struct net_device *dev) | |
1da177e4 | 498 | { |
2bad35b7 | 499 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 500 | struct iphdr *iph; |
5ab11c98 | 501 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
502 | int len; |
503 | int ihlen; | |
1706d587 | 504 | int err; |
1da177e4 LT |
505 | |
506 | ipq_kill(qp); | |
507 | ||
1706d587 HX |
508 | /* Make the one we just received the head. */ |
509 | if (prev) { | |
510 | head = prev->next; | |
511 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
512 | if (!fp) |
513 | goto out_nomem; | |
514 | ||
515 | fp->next = head->next; | |
d6bebca9 CG |
516 | if (!fp->next) |
517 | qp->q.fragments_tail = fp; | |
1706d587 HX |
518 | prev->next = fp; |
519 | ||
5ab11c98 PE |
520 | skb_morph(head, qp->q.fragments); |
521 | head->next = qp->q.fragments->next; | |
1706d587 | 522 | |
5ab11c98 PE |
523 | kfree_skb(qp->q.fragments); |
524 | qp->q.fragments = head; | |
1706d587 HX |
525 | } |
526 | ||
547b792c IJ |
527 | WARN_ON(head == NULL); |
528 | WARN_ON(FRAG_CB(head)->offset != 0); | |
1da177e4 LT |
529 | |
530 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 531 | ihlen = ip_hdrlen(head); |
5ab11c98 | 532 | len = ihlen + qp->q.len; |
1da177e4 | 533 | |
1706d587 | 534 | err = -E2BIG; |
132adf54 | 535 | if (len > 65535) |
1da177e4 LT |
536 | goto out_oversize; |
537 | ||
538 | /* Head of list must not be cloned. */ | |
539 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
540 | goto out_nomem; | |
541 | ||
542 | /* If the first fragment is fragmented itself, we split | |
543 | * it to two chunks: the first with data and paged part | |
544 | * and the second, holding only fragments. */ | |
21dc3301 | 545 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
546 | struct sk_buff *clone; |
547 | int i, plen = 0; | |
548 | ||
549 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
550 | goto out_nomem; | |
551 | clone->next = head->next; | |
552 | head->next = clone; | |
553 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 554 | skb_frag_list_init(head); |
1da177e4 LT |
555 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) |
556 | plen += skb_shinfo(head)->frags[i].size; | |
557 | clone->len = clone->data_len = head->data_len - plen; | |
558 | head->data_len -= clone->len; | |
559 | head->len -= clone->len; | |
560 | clone->csum = 0; | |
561 | clone->ip_summed = head->ip_summed; | |
6ddc0822 | 562 | atomic_add(clone->truesize, &qp->q.net->mem); |
1da177e4 LT |
563 | } |
564 | ||
565 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 566 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 LT |
567 | |
568 | for (fp=head->next; fp; fp = fp->next) { | |
569 | head->data_len += fp->len; | |
570 | head->len += fp->len; | |
571 | if (head->ip_summed != fp->ip_summed) | |
572 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 573 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
574 | head->csum = csum_add(head->csum, fp->csum); |
575 | head->truesize += fp->truesize; | |
1da177e4 | 576 | } |
d27f9b35 | 577 | atomic_sub(head->truesize, &qp->q.net->mem); |
1da177e4 LT |
578 | |
579 | head->next = NULL; | |
580 | head->dev = dev; | |
5ab11c98 | 581 | head->tstamp = qp->q.stamp; |
1da177e4 | 582 | |
eddc9ec5 | 583 | iph = ip_hdr(head); |
1da177e4 LT |
584 | iph->frag_off = 0; |
585 | iph->tot_len = htons(len); | |
2bad35b7 | 586 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 587 | qp->q.fragments = NULL; |
d6bebca9 | 588 | qp->q.fragments_tail = NULL; |
1706d587 | 589 | return 0; |
1da177e4 LT |
590 | |
591 | out_nomem: | |
e905a9ed | 592 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 593 | "queue %p\n", qp); |
45542479 | 594 | err = -ENOMEM; |
1da177e4 LT |
595 | goto out_fail; |
596 | out_oversize: | |
597 | if (net_ratelimit()) | |
673d57e7 HH |
598 | printk(KERN_INFO "Oversized IP packet from %pI4.\n", |
599 | &qp->saddr); | |
1da177e4 | 600 | out_fail: |
bbf31bf1 | 601 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 602 | return err; |
1da177e4 LT |
603 | } |
604 | ||
605 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 606 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 607 | { |
1da177e4 | 608 | struct ipq *qp; |
ac18e750 | 609 | struct net *net; |
e905a9ed | 610 | |
adf30907 | 611 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); |
7c73a6fa | 612 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
613 | |
614 | /* Start by cleaning up the memory. */ | |
e31e0bdc | 615 | if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh) |
6ddc0822 | 616 | ip_evictor(net); |
1da177e4 | 617 | |
1da177e4 | 618 | /* Lookup (or create) queue header */ |
ac18e750 | 619 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 620 | int ret; |
1da177e4 | 621 | |
5ab11c98 | 622 | spin_lock(&qp->q.lock); |
1da177e4 | 623 | |
1706d587 | 624 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 625 | |
5ab11c98 | 626 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 627 | ipq_put(qp); |
776c729e | 628 | return ret; |
1da177e4 LT |
629 | } |
630 | ||
7c73a6fa | 631 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 632 | kfree_skb(skb); |
776c729e | 633 | return -ENOMEM; |
1da177e4 | 634 | } |
4bc2f18b | 635 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 636 | |
8d8354d2 PE |
637 | #ifdef CONFIG_SYSCTL |
638 | static int zero; | |
639 | ||
0a64b4b8 | 640 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 641 | { |
8d8354d2 | 642 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 643 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
644 | .maxlen = sizeof(int), |
645 | .mode = 0644, | |
6d9f239a | 646 | .proc_handler = proc_dointvec |
8d8354d2 PE |
647 | }, |
648 | { | |
8d8354d2 | 649 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 650 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
651 | .maxlen = sizeof(int), |
652 | .mode = 0644, | |
6d9f239a | 653 | .proc_handler = proc_dointvec |
8d8354d2 PE |
654 | }, |
655 | { | |
8d8354d2 | 656 | .procname = "ipfrag_time", |
b2fd5321 | 657 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
658 | .maxlen = sizeof(int), |
659 | .mode = 0644, | |
6d9f239a | 660 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 661 | }, |
7d291ebb PE |
662 | { } |
663 | }; | |
664 | ||
665 | static struct ctl_table ip4_frags_ctl_table[] = { | |
8d8354d2 | 666 | { |
8d8354d2 | 667 | .procname = "ipfrag_secret_interval", |
3b4bc4a2 | 668 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
669 | .maxlen = sizeof(int), |
670 | .mode = 0644, | |
6d9f239a | 671 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 PE |
672 | }, |
673 | { | |
674 | .procname = "ipfrag_max_dist", | |
675 | .data = &sysctl_ipfrag_max_dist, | |
676 | .maxlen = sizeof(int), | |
677 | .mode = 0644, | |
6d9f239a | 678 | .proc_handler = proc_dointvec_minmax, |
8d8354d2 PE |
679 | .extra1 = &zero |
680 | }, | |
681 | { } | |
682 | }; | |
683 | ||
2c8c1e72 | 684 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 685 | { |
e4a2d5c2 | 686 | struct ctl_table *table; |
8d8354d2 PE |
687 | struct ctl_table_header *hdr; |
688 | ||
0a64b4b8 | 689 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 690 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 691 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
e4a2d5c2 PE |
692 | if (table == NULL) |
693 | goto err_alloc; | |
694 | ||
e31e0bdc PE |
695 | table[0].data = &net->ipv4.frags.high_thresh; |
696 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 697 | table[2].data = &net->ipv4.frags.timeout; |
e4a2d5c2 PE |
698 | } |
699 | ||
700 | hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table); | |
701 | if (hdr == NULL) | |
702 | goto err_reg; | |
703 | ||
704 | net->ipv4.frags_hdr = hdr; | |
705 | return 0; | |
706 | ||
707 | err_reg: | |
09ad9bc7 | 708 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
709 | kfree(table); |
710 | err_alloc: | |
711 | return -ENOMEM; | |
712 | } | |
713 | ||
2c8c1e72 | 714 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
715 | { |
716 | struct ctl_table *table; | |
717 | ||
718 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
719 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
720 | kfree(table); | |
8d8354d2 | 721 | } |
7d291ebb PE |
722 | |
723 | static void ip4_frags_ctl_register(void) | |
724 | { | |
725 | register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table); | |
726 | } | |
8d8354d2 | 727 | #else |
0a64b4b8 | 728 | static inline int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
729 | { |
730 | return 0; | |
731 | } | |
e4a2d5c2 | 732 | |
0a64b4b8 | 733 | static inline void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
734 | { |
735 | } | |
7d291ebb PE |
736 | |
737 | static inline void ip4_frags_ctl_register(void) | |
738 | { | |
739 | } | |
8d8354d2 PE |
740 | #endif |
741 | ||
2c8c1e72 | 742 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 743 | { |
e31e0bdc PE |
744 | /* |
745 | * Fragment cache limits. We will commit 256K at one time. Should we | |
746 | * cross that limit we will prune down to 192K. This should cope with | |
747 | * even the most extreme cases without allowing an attacker to | |
748 | * measurably harm machine performance. | |
749 | */ | |
750 | net->ipv4.frags.high_thresh = 256 * 1024; | |
751 | net->ipv4.frags.low_thresh = 192 * 1024; | |
b2fd5321 PE |
752 | /* |
753 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
754 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
755 | * by TTL. | |
756 | */ | |
757 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
758 | ||
e5a2bb84 PE |
759 | inet_frags_init_net(&net->ipv4.frags); |
760 | ||
0a64b4b8 | 761 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
762 | } |
763 | ||
2c8c1e72 | 764 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 765 | { |
0a64b4b8 | 766 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
767 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
768 | } | |
769 | ||
770 | static struct pernet_operations ip4_frags_ops = { | |
771 | .init = ipv4_frags_init_net, | |
772 | .exit = ipv4_frags_exit_net, | |
773 | }; | |
774 | ||
b7aa0bf7 | 775 | void __init ipfrag_init(void) |
1da177e4 | 776 | { |
7d291ebb | 777 | ip4_frags_ctl_register(); |
81566e83 | 778 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 779 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 780 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
781 | ip4_frags.destructor = ip4_frag_free; |
782 | ip4_frags.skb_free = NULL; | |
783 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 784 | ip4_frags.match = ip4_frag_match; |
e521db9d | 785 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 786 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 787 | inet_frags_init(&ip4_frags); |
1da177e4 | 788 | } |