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 LT |
8 | * Version: $Id: ip_fragment.c,v 1.59 2002/01/12 07:54:56 davem Exp $ |
9 | * | |
10 | * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG> | |
11 | * Alan Cox <Alan.Cox@linux.org> | |
12 | * | |
13 | * Fixes: | |
14 | * Alan Cox : Split from ip.c , see ip_input.c for history. | |
15 | * David S. Miller : Begin massive cleanup... | |
16 | * Andi Kleen : Add sysctls. | |
17 | * xxxx : Overlapfrag bug. | |
18 | * Ultima : ip_expire() kernel panic. | |
19 | * Bill Hawes : Frag accounting and evictor fixes. | |
20 | * John McDonald : 0 length frag bug. | |
21 | * Alexey Kuznetsov: SMP races, threading, cleanup. | |
22 | * Patrick McHardy : LRU queue of frag heads for evictor. | |
23 | */ | |
24 | ||
89cee8b1 | 25 | #include <linux/compiler.h> |
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/types.h> | |
28 | #include <linux/mm.h> | |
29 | #include <linux/jiffies.h> | |
30 | #include <linux/skbuff.h> | |
31 | #include <linux/list.h> | |
32 | #include <linux/ip.h> | |
33 | #include <linux/icmp.h> | |
34 | #include <linux/netdevice.h> | |
35 | #include <linux/jhash.h> | |
36 | #include <linux/random.h> | |
37 | #include <net/sock.h> | |
38 | #include <net/ip.h> | |
39 | #include <net/icmp.h> | |
40 | #include <net/checksum.h> | |
89cee8b1 | 41 | #include <net/inetpeer.h> |
5ab11c98 | 42 | #include <net/inet_frag.h> |
1da177e4 LT |
43 | #include <linux/tcp.h> |
44 | #include <linux/udp.h> | |
45 | #include <linux/inet.h> | |
46 | #include <linux/netfilter_ipv4.h> | |
47 | ||
48 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
49 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
50 | * as well. Or notify me, at least. --ANK | |
51 | */ | |
52 | ||
8d8354d2 | 53 | static int sysctl_ipfrag_max_dist __read_mostly = 64; |
89cee8b1 | 54 | |
1da177e4 LT |
55 | struct ipfrag_skb_cb |
56 | { | |
57 | struct inet_skb_parm h; | |
58 | int offset; | |
59 | }; | |
60 | ||
61 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb)) | |
62 | ||
63 | /* Describe an entry in the "incomplete datagrams" queue. */ | |
64 | struct ipq { | |
5ab11c98 PE |
65 | struct inet_frag_queue q; |
66 | ||
1da177e4 | 67 | u32 user; |
18277770 AV |
68 | __be32 saddr; |
69 | __be32 daddr; | |
70 | __be16 id; | |
1da177e4 | 71 | u8 protocol; |
89cee8b1 HX |
72 | int iif; |
73 | unsigned int rid; | |
74 | struct inet_peer *peer; | |
1da177e4 LT |
75 | }; |
76 | ||
7eb95156 | 77 | static struct inet_frags ip4_frags; |
1da177e4 | 78 | |
e5a2bb84 | 79 | int ip_frag_nqueues(struct net *net) |
7eb95156 | 80 | { |
e5a2bb84 | 81 | return net->ipv4.frags.nqueues; |
7eb95156 | 82 | } |
1da177e4 | 83 | |
6ddc0822 | 84 | int ip_frag_mem(struct net *net) |
7eb95156 | 85 | { |
6ddc0822 | 86 | return atomic_read(&net->ipv4.frags.mem); |
7eb95156 | 87 | } |
1da177e4 | 88 | |
1706d587 HX |
89 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
90 | struct net_device *dev); | |
91 | ||
c6fda282 PE |
92 | struct ip4_create_arg { |
93 | struct iphdr *iph; | |
94 | u32 user; | |
95 | }; | |
96 | ||
18277770 | 97 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 98 | { |
18277770 AV |
99 | return jhash_3words((__force u32)id << 16 | prot, |
100 | (__force u32)saddr, (__force u32)daddr, | |
7eb95156 | 101 | ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
102 | } |
103 | ||
321a3a99 | 104 | static unsigned int ip4_hashfn(struct inet_frag_queue *q) |
1da177e4 | 105 | { |
321a3a99 | 106 | struct ipq *ipq; |
1da177e4 | 107 | |
321a3a99 PE |
108 | ipq = container_of(q, struct ipq, q); |
109 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
110 | } |
111 | ||
abd6523d PE |
112 | static int ip4_frag_match(struct inet_frag_queue *q, void *a) |
113 | { | |
114 | struct ipq *qp; | |
115 | struct ip4_create_arg *arg = a; | |
116 | ||
117 | qp = container_of(q, struct ipq, q); | |
118 | return (qp->id == arg->iph->id && | |
119 | qp->saddr == arg->iph->saddr && | |
120 | qp->daddr == arg->iph->daddr && | |
121 | qp->protocol == arg->iph->protocol && | |
122 | qp->user == arg->user); | |
123 | } | |
124 | ||
1da177e4 | 125 | /* Memory Tracking Functions. */ |
6ddc0822 PE |
126 | static __inline__ void frag_kfree_skb(struct netns_frags *nf, |
127 | struct sk_buff *skb, int *work) | |
1da177e4 LT |
128 | { |
129 | if (work) | |
130 | *work -= skb->truesize; | |
6ddc0822 | 131 | atomic_sub(skb->truesize, &nf->mem); |
1da177e4 LT |
132 | kfree_skb(skb); |
133 | } | |
134 | ||
c6fda282 PE |
135 | static void ip4_frag_init(struct inet_frag_queue *q, void *a) |
136 | { | |
137 | struct ipq *qp = container_of(q, struct ipq, q); | |
138 | struct ip4_create_arg *arg = a; | |
139 | ||
140 | qp->protocol = arg->iph->protocol; | |
141 | qp->id = arg->iph->id; | |
142 | qp->saddr = arg->iph->saddr; | |
143 | qp->daddr = arg->iph->daddr; | |
144 | qp->user = arg->user; | |
145 | qp->peer = sysctl_ipfrag_max_dist ? | |
146 | inet_getpeer(arg->iph->saddr, 1) : NULL; | |
147 | } | |
148 | ||
1e4b8287 | 149 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 150 | { |
1e4b8287 PE |
151 | struct ipq *qp; |
152 | ||
153 | qp = container_of(q, struct ipq, q); | |
154 | if (qp->peer) | |
155 | inet_putpeer(qp->peer); | |
1da177e4 LT |
156 | } |
157 | ||
1da177e4 LT |
158 | |
159 | /* Destruction primitives. */ | |
160 | ||
4b6cb5d8 | 161 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 162 | { |
762cc408 | 163 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
164 | } |
165 | ||
166 | /* Kill ipq entry. It is not destroyed immediately, | |
167 | * because caller (and someone more) holds reference count. | |
168 | */ | |
169 | static void ipq_kill(struct ipq *ipq) | |
170 | { | |
277e650d | 171 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
172 | } |
173 | ||
e905a9ed | 174 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
175 | * fragment queue until we are back under the threshold. |
176 | */ | |
6ddc0822 | 177 | static void ip_evictor(struct net *net) |
1da177e4 | 178 | { |
8e7999c4 PE |
179 | int evicted; |
180 | ||
6ddc0822 | 181 | evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags); |
8e7999c4 PE |
182 | if (evicted) |
183 | IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted); | |
1da177e4 LT |
184 | } |
185 | ||
186 | /* | |
187 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
188 | */ | |
189 | static void ip_expire(unsigned long arg) | |
190 | { | |
e521db9d PE |
191 | struct ipq *qp; |
192 | ||
193 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
1da177e4 | 194 | |
5ab11c98 | 195 | spin_lock(&qp->q.lock); |
1da177e4 | 196 | |
5ab11c98 | 197 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
198 | goto out; |
199 | ||
200 | ipq_kill(qp); | |
201 | ||
202 | IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT); | |
203 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
204 | ||
5ab11c98 PE |
205 | if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) { |
206 | struct sk_buff *head = qp->q.fragments; | |
cb84663e DL |
207 | struct net *net; |
208 | ||
209 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
1da177e4 | 210 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
cb84663e | 211 | if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) { |
1da177e4 LT |
212 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); |
213 | dev_put(head->dev); | |
214 | } | |
215 | } | |
216 | out: | |
5ab11c98 | 217 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 218 | ipq_put(qp); |
1da177e4 LT |
219 | } |
220 | ||
abd6523d PE |
221 | /* Find the correct entry in the "incomplete datagrams" queue for |
222 | * this IP datagram, and create new one, if nothing is found. | |
223 | */ | |
ac18e750 | 224 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 225 | { |
c6fda282 PE |
226 | struct inet_frag_queue *q; |
227 | struct ip4_create_arg arg; | |
abd6523d | 228 | unsigned int hash; |
1da177e4 | 229 | |
c6fda282 PE |
230 | arg.iph = iph; |
231 | arg.user = user; | |
abd6523d | 232 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 233 | |
ac18e750 | 234 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
c6fda282 PE |
235 | if (q == NULL) |
236 | goto out_nomem; | |
1da177e4 | 237 | |
c6fda282 | 238 | return container_of(q, struct ipq, q); |
1da177e4 LT |
239 | |
240 | out_nomem: | |
64ce2073 | 241 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
242 | return NULL; |
243 | } | |
244 | ||
89cee8b1 HX |
245 | /* Is the fragment too far ahead to be part of ipq? */ |
246 | static inline int ip_frag_too_far(struct ipq *qp) | |
247 | { | |
248 | struct inet_peer *peer = qp->peer; | |
249 | unsigned int max = sysctl_ipfrag_max_dist; | |
250 | unsigned int start, end; | |
251 | ||
252 | int rc; | |
253 | ||
254 | if (!peer || !max) | |
255 | return 0; | |
256 | ||
257 | start = qp->rid; | |
258 | end = atomic_inc_return(&peer->rid); | |
259 | qp->rid = end; | |
260 | ||
5ab11c98 | 261 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
262 | |
263 | if (rc) { | |
264 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
265 | } | |
266 | ||
267 | return rc; | |
268 | } | |
269 | ||
270 | static int ip_frag_reinit(struct ipq *qp) | |
271 | { | |
272 | struct sk_buff *fp; | |
273 | ||
b2fd5321 | 274 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 275 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
276 | return -ETIMEDOUT; |
277 | } | |
278 | ||
5ab11c98 | 279 | fp = qp->q.fragments; |
89cee8b1 HX |
280 | do { |
281 | struct sk_buff *xp = fp->next; | |
6ddc0822 | 282 | frag_kfree_skb(qp->q.net, fp, NULL); |
89cee8b1 HX |
283 | fp = xp; |
284 | } while (fp); | |
285 | ||
5ab11c98 PE |
286 | qp->q.last_in = 0; |
287 | qp->q.len = 0; | |
288 | qp->q.meat = 0; | |
289 | qp->q.fragments = NULL; | |
89cee8b1 HX |
290 | qp->iif = 0; |
291 | ||
292 | return 0; | |
293 | } | |
294 | ||
1da177e4 | 295 | /* Add new segment to existing queue. */ |
1706d587 | 296 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
297 | { |
298 | struct sk_buff *prev, *next; | |
1706d587 | 299 | struct net_device *dev; |
1da177e4 LT |
300 | int flags, offset; |
301 | int ihl, end; | |
1706d587 | 302 | int err = -ENOENT; |
1da177e4 | 303 | |
5ab11c98 | 304 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
305 | goto err; |
306 | ||
89cee8b1 | 307 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
308 | unlikely(ip_frag_too_far(qp)) && |
309 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
310 | ipq_kill(qp); |
311 | goto err; | |
312 | } | |
313 | ||
eddc9ec5 | 314 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
315 | flags = offset & ~IP_OFFSET; |
316 | offset &= IP_OFFSET; | |
317 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 318 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
319 | |
320 | /* Determine the position of this fragment. */ | |
e905a9ed | 321 | end = offset + skb->len - ihl; |
1706d587 | 322 | err = -EINVAL; |
1da177e4 LT |
323 | |
324 | /* Is this the final fragment? */ | |
325 | if ((flags & IP_MF) == 0) { | |
326 | /* If we already have some bits beyond end | |
327 | * or have different end, the segment is corrrupted. | |
328 | */ | |
5ab11c98 PE |
329 | if (end < qp->q.len || |
330 | ((qp->q.last_in & LAST_IN) && end != qp->q.len)) | |
1da177e4 | 331 | goto err; |
5ab11c98 PE |
332 | qp->q.last_in |= LAST_IN; |
333 | qp->q.len = end; | |
1da177e4 LT |
334 | } else { |
335 | if (end&7) { | |
336 | end &= ~7; | |
337 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
338 | skb->ip_summed = CHECKSUM_NONE; | |
339 | } | |
5ab11c98 | 340 | if (end > qp->q.len) { |
1da177e4 | 341 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 342 | if (qp->q.last_in & LAST_IN) |
1da177e4 | 343 | goto err; |
5ab11c98 | 344 | qp->q.len = end; |
1da177e4 LT |
345 | } |
346 | } | |
347 | if (end == offset) | |
348 | goto err; | |
349 | ||
1706d587 | 350 | err = -ENOMEM; |
1da177e4 LT |
351 | if (pskb_pull(skb, ihl) == NULL) |
352 | goto err; | |
1706d587 HX |
353 | |
354 | err = pskb_trim_rcsum(skb, end - offset); | |
355 | if (err) | |
1da177e4 LT |
356 | goto err; |
357 | ||
358 | /* Find out which fragments are in front and at the back of us | |
359 | * in the chain of fragments so far. We must know where to put | |
360 | * this fragment, right? | |
361 | */ | |
362 | prev = NULL; | |
5ab11c98 | 363 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
364 | if (FRAG_CB(next)->offset >= offset) |
365 | break; /* bingo! */ | |
366 | prev = next; | |
367 | } | |
368 | ||
369 | /* We found where to put this one. Check for overlap with | |
370 | * preceding fragment, and, if needed, align things so that | |
371 | * any overlaps are eliminated. | |
372 | */ | |
373 | if (prev) { | |
374 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
375 | ||
376 | if (i > 0) { | |
377 | offset += i; | |
1706d587 | 378 | err = -EINVAL; |
1da177e4 LT |
379 | if (end <= offset) |
380 | goto err; | |
1706d587 | 381 | err = -ENOMEM; |
1da177e4 LT |
382 | if (!pskb_pull(skb, i)) |
383 | goto err; | |
384 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
385 | skb->ip_summed = CHECKSUM_NONE; | |
386 | } | |
387 | } | |
388 | ||
1706d587 HX |
389 | err = -ENOMEM; |
390 | ||
1da177e4 LT |
391 | while (next && FRAG_CB(next)->offset < end) { |
392 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
393 | ||
394 | if (i < next->len) { | |
395 | /* Eat head of the next overlapped fragment | |
396 | * and leave the loop. The next ones cannot overlap. | |
397 | */ | |
398 | if (!pskb_pull(next, i)) | |
399 | goto err; | |
400 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 401 | qp->q.meat -= i; |
1da177e4 LT |
402 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
403 | next->ip_summed = CHECKSUM_NONE; | |
404 | break; | |
405 | } else { | |
406 | struct sk_buff *free_it = next; | |
407 | ||
47c6bf77 | 408 | /* Old fragment is completely overridden with |
1da177e4 LT |
409 | * new one drop it. |
410 | */ | |
411 | next = next->next; | |
412 | ||
413 | if (prev) | |
414 | prev->next = next; | |
415 | else | |
5ab11c98 | 416 | qp->q.fragments = next; |
1da177e4 | 417 | |
5ab11c98 | 418 | qp->q.meat -= free_it->len; |
6ddc0822 | 419 | frag_kfree_skb(qp->q.net, free_it, NULL); |
1da177e4 LT |
420 | } |
421 | } | |
422 | ||
423 | FRAG_CB(skb)->offset = offset; | |
424 | ||
425 | /* Insert this fragment in the chain of fragments. */ | |
426 | skb->next = next; | |
427 | if (prev) | |
428 | prev->next = skb; | |
429 | else | |
5ab11c98 | 430 | qp->q.fragments = skb; |
1da177e4 | 431 | |
1706d587 HX |
432 | dev = skb->dev; |
433 | if (dev) { | |
434 | qp->iif = dev->ifindex; | |
435 | skb->dev = NULL; | |
436 | } | |
5ab11c98 PE |
437 | qp->q.stamp = skb->tstamp; |
438 | qp->q.meat += skb->len; | |
6ddc0822 | 439 | atomic_add(skb->truesize, &qp->q.net->mem); |
1da177e4 | 440 | if (offset == 0) |
5ab11c98 | 441 | qp->q.last_in |= FIRST_IN; |
1da177e4 | 442 | |
5ab11c98 | 443 | if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len) |
1706d587 HX |
444 | return ip_frag_reasm(qp, prev, dev); |
445 | ||
7eb95156 | 446 | write_lock(&ip4_frags.lock); |
3140c25c | 447 | list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list); |
7eb95156 | 448 | write_unlock(&ip4_frags.lock); |
1706d587 | 449 | return -EINPROGRESS; |
1da177e4 LT |
450 | |
451 | err: | |
452 | kfree_skb(skb); | |
1706d587 | 453 | return err; |
1da177e4 LT |
454 | } |
455 | ||
456 | ||
457 | /* Build a new IP datagram from all its fragments. */ | |
458 | ||
1706d587 HX |
459 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
460 | struct net_device *dev) | |
1da177e4 LT |
461 | { |
462 | struct iphdr *iph; | |
5ab11c98 | 463 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
464 | int len; |
465 | int ihlen; | |
1706d587 | 466 | int err; |
1da177e4 LT |
467 | |
468 | ipq_kill(qp); | |
469 | ||
1706d587 HX |
470 | /* Make the one we just received the head. */ |
471 | if (prev) { | |
472 | head = prev->next; | |
473 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
474 | if (!fp) |
475 | goto out_nomem; | |
476 | ||
477 | fp->next = head->next; | |
478 | prev->next = fp; | |
479 | ||
5ab11c98 PE |
480 | skb_morph(head, qp->q.fragments); |
481 | head->next = qp->q.fragments->next; | |
1706d587 | 482 | |
5ab11c98 PE |
483 | kfree_skb(qp->q.fragments); |
484 | qp->q.fragments = head; | |
1706d587 HX |
485 | } |
486 | ||
1da177e4 LT |
487 | BUG_TRAP(head != NULL); |
488 | BUG_TRAP(FRAG_CB(head)->offset == 0); | |
489 | ||
490 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 491 | ihlen = ip_hdrlen(head); |
5ab11c98 | 492 | len = ihlen + qp->q.len; |
1da177e4 | 493 | |
1706d587 | 494 | err = -E2BIG; |
132adf54 | 495 | if (len > 65535) |
1da177e4 LT |
496 | goto out_oversize; |
497 | ||
498 | /* Head of list must not be cloned. */ | |
499 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
500 | goto out_nomem; | |
501 | ||
502 | /* If the first fragment is fragmented itself, we split | |
503 | * it to two chunks: the first with data and paged part | |
504 | * and the second, holding only fragments. */ | |
505 | if (skb_shinfo(head)->frag_list) { | |
506 | struct sk_buff *clone; | |
507 | int i, plen = 0; | |
508 | ||
509 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
510 | goto out_nomem; | |
511 | clone->next = head->next; | |
512 | head->next = clone; | |
513 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
514 | skb_shinfo(head)->frag_list = NULL; | |
515 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
516 | plen += skb_shinfo(head)->frags[i].size; | |
517 | clone->len = clone->data_len = head->data_len - plen; | |
518 | head->data_len -= clone->len; | |
519 | head->len -= clone->len; | |
520 | clone->csum = 0; | |
521 | clone->ip_summed = head->ip_summed; | |
6ddc0822 | 522 | atomic_add(clone->truesize, &qp->q.net->mem); |
1da177e4 LT |
523 | } |
524 | ||
525 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 526 | skb_push(head, head->data - skb_network_header(head)); |
6ddc0822 | 527 | atomic_sub(head->truesize, &qp->q.net->mem); |
1da177e4 LT |
528 | |
529 | for (fp=head->next; fp; fp = fp->next) { | |
530 | head->data_len += fp->len; | |
531 | head->len += fp->len; | |
532 | if (head->ip_summed != fp->ip_summed) | |
533 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 534 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
535 | head->csum = csum_add(head->csum, fp->csum); |
536 | head->truesize += fp->truesize; | |
6ddc0822 | 537 | atomic_sub(fp->truesize, &qp->q.net->mem); |
1da177e4 LT |
538 | } |
539 | ||
540 | head->next = NULL; | |
541 | head->dev = dev; | |
5ab11c98 | 542 | head->tstamp = qp->q.stamp; |
1da177e4 | 543 | |
eddc9ec5 | 544 | iph = ip_hdr(head); |
1da177e4 LT |
545 | iph->frag_off = 0; |
546 | iph->tot_len = htons(len); | |
547 | IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS); | |
5ab11c98 | 548 | qp->q.fragments = NULL; |
1706d587 | 549 | return 0; |
1da177e4 LT |
550 | |
551 | out_nomem: | |
e905a9ed | 552 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 553 | "queue %p\n", qp); |
45542479 | 554 | err = -ENOMEM; |
1da177e4 LT |
555 | goto out_fail; |
556 | out_oversize: | |
557 | if (net_ratelimit()) | |
558 | printk(KERN_INFO | |
559 | "Oversized IP packet from %d.%d.%d.%d.\n", | |
560 | NIPQUAD(qp->saddr)); | |
561 | out_fail: | |
562 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
1706d587 | 563 | return err; |
1da177e4 LT |
564 | } |
565 | ||
566 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 567 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 568 | { |
1da177e4 | 569 | struct ipq *qp; |
ac18e750 | 570 | struct net *net; |
e905a9ed | 571 | |
1da177e4 LT |
572 | IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS); |
573 | ||
c346dca1 | 574 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev); |
1da177e4 | 575 | /* Start by cleaning up the memory. */ |
e31e0bdc | 576 | if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh) |
6ddc0822 | 577 | ip_evictor(net); |
1da177e4 | 578 | |
1da177e4 | 579 | /* Lookup (or create) queue header */ |
ac18e750 | 580 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 581 | int ret; |
1da177e4 | 582 | |
5ab11c98 | 583 | spin_lock(&qp->q.lock); |
1da177e4 | 584 | |
1706d587 | 585 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 586 | |
5ab11c98 | 587 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 588 | ipq_put(qp); |
776c729e | 589 | return ret; |
1da177e4 LT |
590 | } |
591 | ||
592 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
593 | kfree_skb(skb); | |
776c729e | 594 | return -ENOMEM; |
1da177e4 LT |
595 | } |
596 | ||
8d8354d2 PE |
597 | #ifdef CONFIG_SYSCTL |
598 | static int zero; | |
599 | ||
600 | static struct ctl_table ip4_frags_ctl_table[] = { | |
601 | { | |
602 | .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH, | |
603 | .procname = "ipfrag_high_thresh", | |
e31e0bdc | 604 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
605 | .maxlen = sizeof(int), |
606 | .mode = 0644, | |
607 | .proc_handler = &proc_dointvec | |
608 | }, | |
609 | { | |
610 | .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH, | |
611 | .procname = "ipfrag_low_thresh", | |
e31e0bdc | 612 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
613 | .maxlen = sizeof(int), |
614 | .mode = 0644, | |
615 | .proc_handler = &proc_dointvec | |
616 | }, | |
617 | { | |
618 | .ctl_name = NET_IPV4_IPFRAG_TIME, | |
619 | .procname = "ipfrag_time", | |
b2fd5321 | 620 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
621 | .maxlen = sizeof(int), |
622 | .mode = 0644, | |
623 | .proc_handler = &proc_dointvec_jiffies, | |
624 | .strategy = &sysctl_jiffies | |
625 | }, | |
626 | { | |
627 | .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL, | |
628 | .procname = "ipfrag_secret_interval", | |
3b4bc4a2 | 629 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
630 | .maxlen = sizeof(int), |
631 | .mode = 0644, | |
632 | .proc_handler = &proc_dointvec_jiffies, | |
633 | .strategy = &sysctl_jiffies | |
634 | }, | |
635 | { | |
636 | .procname = "ipfrag_max_dist", | |
637 | .data = &sysctl_ipfrag_max_dist, | |
638 | .maxlen = sizeof(int), | |
639 | .mode = 0644, | |
640 | .proc_handler = &proc_dointvec_minmax, | |
641 | .extra1 = &zero | |
642 | }, | |
643 | { } | |
644 | }; | |
645 | ||
646 | static int ip4_frags_ctl_register(struct net *net) | |
647 | { | |
e4a2d5c2 | 648 | struct ctl_table *table; |
8d8354d2 PE |
649 | struct ctl_table_header *hdr; |
650 | ||
e4a2d5c2 PE |
651 | table = ip4_frags_ctl_table; |
652 | if (net != &init_net) { | |
653 | table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL); | |
654 | if (table == NULL) | |
655 | goto err_alloc; | |
656 | ||
e31e0bdc PE |
657 | table[0].data = &net->ipv4.frags.high_thresh; |
658 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 659 | table[2].data = &net->ipv4.frags.timeout; |
e4a2d5c2 PE |
660 | table[3].mode &= ~0222; |
661 | table[4].mode &= ~0222; | |
662 | } | |
663 | ||
664 | hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table); | |
665 | if (hdr == NULL) | |
666 | goto err_reg; | |
667 | ||
668 | net->ipv4.frags_hdr = hdr; | |
669 | return 0; | |
670 | ||
671 | err_reg: | |
672 | if (net != &init_net) | |
673 | kfree(table); | |
674 | err_alloc: | |
675 | return -ENOMEM; | |
676 | } | |
677 | ||
678 | static void ip4_frags_ctl_unregister(struct net *net) | |
679 | { | |
680 | struct ctl_table *table; | |
681 | ||
682 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
683 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
684 | kfree(table); | |
8d8354d2 PE |
685 | } |
686 | #else | |
687 | static inline int ip4_frags_ctl_register(struct net *net) | |
688 | { | |
689 | return 0; | |
690 | } | |
e4a2d5c2 PE |
691 | |
692 | static inline void ip4_frags_ctl_unregister(struct net *net) | |
693 | { | |
694 | } | |
8d8354d2 PE |
695 | #endif |
696 | ||
697 | static int ipv4_frags_init_net(struct net *net) | |
698 | { | |
e31e0bdc PE |
699 | /* |
700 | * Fragment cache limits. We will commit 256K at one time. Should we | |
701 | * cross that limit we will prune down to 192K. This should cope with | |
702 | * even the most extreme cases without allowing an attacker to | |
703 | * measurably harm machine performance. | |
704 | */ | |
705 | net->ipv4.frags.high_thresh = 256 * 1024; | |
706 | net->ipv4.frags.low_thresh = 192 * 1024; | |
b2fd5321 PE |
707 | /* |
708 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
709 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
710 | * by TTL. | |
711 | */ | |
712 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
713 | ||
e5a2bb84 PE |
714 | inet_frags_init_net(&net->ipv4.frags); |
715 | ||
8d8354d2 PE |
716 | return ip4_frags_ctl_register(net); |
717 | } | |
718 | ||
81566e83 PE |
719 | static void ipv4_frags_exit_net(struct net *net) |
720 | { | |
721 | ip4_frags_ctl_unregister(net); | |
722 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); | |
723 | } | |
724 | ||
725 | static struct pernet_operations ip4_frags_ops = { | |
726 | .init = ipv4_frags_init_net, | |
727 | .exit = ipv4_frags_exit_net, | |
728 | }; | |
729 | ||
b7aa0bf7 | 730 | void __init ipfrag_init(void) |
1da177e4 | 731 | { |
81566e83 | 732 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 733 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 734 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
735 | ip4_frags.destructor = ip4_frag_free; |
736 | ip4_frags.skb_free = NULL; | |
737 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 738 | ip4_frags.match = ip4_frag_match; |
e521db9d | 739 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 740 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 741 | inet_frags_init(&ip4_frags); |
1da177e4 LT |
742 | } |
743 | ||
744 | EXPORT_SYMBOL(ip_defrag); |