]>
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 | ||
ab32ea5d | 53 | 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 | ||
04128f23 PE |
77 | struct inet_frags_ctl ip4_frags_ctl __read_mostly = { |
78 | /* | |
79 | * Fragment cache limits. We will commit 256K at one time. Should we | |
80 | * cross that limit we will prune down to 192K. This should cope with | |
81 | * even the most extreme cases without allowing an attacker to | |
82 | * measurably harm machine performance. | |
83 | */ | |
84 | .high_thresh = 256 * 1024, | |
85 | .low_thresh = 192 * 1024, | |
86 | ||
87 | /* | |
88 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
89 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
90 | * by TTL. | |
91 | */ | |
92 | .timeout = IP_FRAG_TIME, | |
93 | .secret_interval = 10 * 60 * HZ, | |
94 | }; | |
95 | ||
7eb95156 | 96 | static struct inet_frags ip4_frags; |
1da177e4 | 97 | |
7eb95156 PE |
98 | int ip_frag_nqueues(void) |
99 | { | |
100 | return ip4_frags.nqueues; | |
101 | } | |
1da177e4 | 102 | |
7eb95156 PE |
103 | int ip_frag_mem(void) |
104 | { | |
105 | return atomic_read(&ip4_frags.mem); | |
106 | } | |
1da177e4 | 107 | |
1706d587 HX |
108 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
109 | struct net_device *dev); | |
110 | ||
18277770 | 111 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 112 | { |
18277770 AV |
113 | return jhash_3words((__force u32)id << 16 | prot, |
114 | (__force u32)saddr, (__force u32)daddr, | |
7eb95156 | 115 | ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
116 | } |
117 | ||
321a3a99 | 118 | static unsigned int ip4_hashfn(struct inet_frag_queue *q) |
1da177e4 | 119 | { |
321a3a99 | 120 | struct ipq *ipq; |
1da177e4 | 121 | |
321a3a99 PE |
122 | ipq = container_of(q, struct ipq, q); |
123 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
124 | } |
125 | ||
2588fe1d PE |
126 | static int ip4_frag_equal(struct inet_frag_queue *q1, |
127 | struct inet_frag_queue *q2) | |
128 | { | |
129 | struct ipq *qp1, *qp2; | |
130 | ||
131 | qp1 = container_of(q1, struct ipq, q); | |
132 | qp2 = container_of(q2, struct ipq, q); | |
133 | return (qp1->id == qp2->id && | |
134 | qp1->saddr == qp2->saddr && | |
135 | qp1->daddr == qp2->daddr && | |
136 | qp1->protocol == qp2->protocol && | |
137 | qp1->user == qp2->user); | |
138 | } | |
139 | ||
1da177e4 LT |
140 | /* Memory Tracking Functions. */ |
141 | static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work) | |
142 | { | |
143 | if (work) | |
144 | *work -= skb->truesize; | |
7eb95156 | 145 | atomic_sub(skb->truesize, &ip4_frags.mem); |
1da177e4 LT |
146 | kfree_skb(skb); |
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 | kfree(qp); |
157 | } | |
158 | ||
159 | static __inline__ struct ipq *frag_alloc_queue(void) | |
160 | { | |
e521db9d | 161 | struct inet_frag_queue *q; |
1da177e4 | 162 | |
e521db9d PE |
163 | q = inet_frag_alloc(&ip4_frags); |
164 | return q ? container_of(q, struct ipq, q) : NULL; | |
1da177e4 LT |
165 | } |
166 | ||
167 | ||
168 | /* Destruction primitives. */ | |
169 | ||
4b6cb5d8 | 170 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 171 | { |
762cc408 | 172 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
173 | } |
174 | ||
175 | /* Kill ipq entry. It is not destroyed immediately, | |
176 | * because caller (and someone more) holds reference count. | |
177 | */ | |
178 | static void ipq_kill(struct ipq *ipq) | |
179 | { | |
277e650d | 180 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
181 | } |
182 | ||
e905a9ed | 183 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
184 | * fragment queue until we are back under the threshold. |
185 | */ | |
186 | static void ip_evictor(void) | |
187 | { | |
8e7999c4 PE |
188 | int evicted; |
189 | ||
190 | evicted = inet_frag_evictor(&ip4_frags); | |
191 | if (evicted) | |
192 | IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted); | |
1da177e4 LT |
193 | } |
194 | ||
195 | /* | |
196 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
197 | */ | |
198 | static void ip_expire(unsigned long arg) | |
199 | { | |
e521db9d PE |
200 | struct ipq *qp; |
201 | ||
202 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
1da177e4 | 203 | |
5ab11c98 | 204 | spin_lock(&qp->q.lock); |
1da177e4 | 205 | |
5ab11c98 | 206 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
207 | goto out; |
208 | ||
209 | ipq_kill(qp); | |
210 | ||
211 | IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT); | |
212 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
213 | ||
5ab11c98 PE |
214 | if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) { |
215 | struct sk_buff *head = qp->q.fragments; | |
1da177e4 | 216 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
881d966b | 217 | if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) { |
1da177e4 LT |
218 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); |
219 | dev_put(head->dev); | |
220 | } | |
221 | } | |
222 | out: | |
5ab11c98 | 223 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 224 | ipq_put(qp); |
1da177e4 LT |
225 | } |
226 | ||
227 | /* Creation primitives. */ | |
228 | ||
fd9e6354 | 229 | static struct ipq *ip_frag_intern(struct ipq *qp_in, unsigned int hash) |
1da177e4 | 230 | { |
2588fe1d | 231 | struct inet_frag_queue *q; |
55c0022e | 232 | |
2588fe1d PE |
233 | q = inet_frag_intern(&qp_in->q, &ip4_frags, hash); |
234 | return container_of(q, struct ipq, q); | |
1da177e4 LT |
235 | } |
236 | ||
237 | /* Add an entry to the 'ipq' queue for a newly received IP datagram. */ | |
fd9e6354 | 238 | static struct ipq *ip_frag_create(struct iphdr *iph, u32 user, unsigned int h) |
1da177e4 LT |
239 | { |
240 | struct ipq *qp; | |
241 | ||
242 | if ((qp = frag_alloc_queue()) == NULL) | |
243 | goto out_nomem; | |
244 | ||
245 | qp->protocol = iph->protocol; | |
1da177e4 LT |
246 | qp->id = iph->id; |
247 | qp->saddr = iph->saddr; | |
248 | qp->daddr = iph->daddr; | |
249 | qp->user = user; | |
89cee8b1 | 250 | qp->peer = sysctl_ipfrag_max_dist ? inet_getpeer(iph->saddr, 1) : NULL; |
1da177e4 | 251 | |
fd9e6354 | 252 | return ip_frag_intern(qp, h); |
1da177e4 LT |
253 | |
254 | out_nomem: | |
64ce2073 | 255 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
256 | return NULL; |
257 | } | |
258 | ||
259 | /* Find the correct entry in the "incomplete datagrams" queue for | |
260 | * this IP datagram, and create new one, if nothing is found. | |
261 | */ | |
262 | static inline struct ipq *ip_find(struct iphdr *iph, u32 user) | |
263 | { | |
76ab608d | 264 | __be16 id = iph->id; |
18277770 AV |
265 | __be32 saddr = iph->saddr; |
266 | __be32 daddr = iph->daddr; | |
1da177e4 | 267 | __u8 protocol = iph->protocol; |
55c0022e | 268 | unsigned int hash; |
1da177e4 | 269 | struct ipq *qp; |
e7c8a41e | 270 | struct hlist_node *n; |
1da177e4 | 271 | |
7eb95156 | 272 | read_lock(&ip4_frags.lock); |
55c0022e | 273 | hash = ipqhashfn(id, saddr, daddr, protocol); |
7eb95156 | 274 | hlist_for_each_entry(qp, n, &ip4_frags.hash[hash], q.list) { |
132adf54 SH |
275 | if (qp->id == id && |
276 | qp->saddr == saddr && | |
277 | qp->daddr == daddr && | |
278 | qp->protocol == protocol && | |
279 | qp->user == user) { | |
5ab11c98 | 280 | atomic_inc(&qp->q.refcnt); |
7eb95156 | 281 | read_unlock(&ip4_frags.lock); |
1da177e4 LT |
282 | return qp; |
283 | } | |
284 | } | |
7eb95156 | 285 | read_unlock(&ip4_frags.lock); |
1da177e4 | 286 | |
fd9e6354 | 287 | return ip_frag_create(iph, user, hash); |
1da177e4 LT |
288 | } |
289 | ||
89cee8b1 HX |
290 | /* Is the fragment too far ahead to be part of ipq? */ |
291 | static inline int ip_frag_too_far(struct ipq *qp) | |
292 | { | |
293 | struct inet_peer *peer = qp->peer; | |
294 | unsigned int max = sysctl_ipfrag_max_dist; | |
295 | unsigned int start, end; | |
296 | ||
297 | int rc; | |
298 | ||
299 | if (!peer || !max) | |
300 | return 0; | |
301 | ||
302 | start = qp->rid; | |
303 | end = atomic_inc_return(&peer->rid); | |
304 | qp->rid = end; | |
305 | ||
5ab11c98 | 306 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
307 | |
308 | if (rc) { | |
309 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
310 | } | |
311 | ||
312 | return rc; | |
313 | } | |
314 | ||
315 | static int ip_frag_reinit(struct ipq *qp) | |
316 | { | |
317 | struct sk_buff *fp; | |
318 | ||
04128f23 | 319 | if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) { |
5ab11c98 | 320 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
321 | return -ETIMEDOUT; |
322 | } | |
323 | ||
5ab11c98 | 324 | fp = qp->q.fragments; |
89cee8b1 HX |
325 | do { |
326 | struct sk_buff *xp = fp->next; | |
327 | frag_kfree_skb(fp, NULL); | |
328 | fp = xp; | |
329 | } while (fp); | |
330 | ||
5ab11c98 PE |
331 | qp->q.last_in = 0; |
332 | qp->q.len = 0; | |
333 | qp->q.meat = 0; | |
334 | qp->q.fragments = NULL; | |
89cee8b1 HX |
335 | qp->iif = 0; |
336 | ||
337 | return 0; | |
338 | } | |
339 | ||
1da177e4 | 340 | /* Add new segment to existing queue. */ |
1706d587 | 341 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
342 | { |
343 | struct sk_buff *prev, *next; | |
1706d587 | 344 | struct net_device *dev; |
1da177e4 LT |
345 | int flags, offset; |
346 | int ihl, end; | |
1706d587 | 347 | int err = -ENOENT; |
1da177e4 | 348 | |
5ab11c98 | 349 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
350 | goto err; |
351 | ||
89cee8b1 | 352 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
353 | unlikely(ip_frag_too_far(qp)) && |
354 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
355 | ipq_kill(qp); |
356 | goto err; | |
357 | } | |
358 | ||
eddc9ec5 | 359 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
360 | flags = offset & ~IP_OFFSET; |
361 | offset &= IP_OFFSET; | |
362 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 363 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
364 | |
365 | /* Determine the position of this fragment. */ | |
e905a9ed | 366 | end = offset + skb->len - ihl; |
1706d587 | 367 | err = -EINVAL; |
1da177e4 LT |
368 | |
369 | /* Is this the final fragment? */ | |
370 | if ((flags & IP_MF) == 0) { | |
371 | /* If we already have some bits beyond end | |
372 | * or have different end, the segment is corrrupted. | |
373 | */ | |
5ab11c98 PE |
374 | if (end < qp->q.len || |
375 | ((qp->q.last_in & LAST_IN) && end != qp->q.len)) | |
1da177e4 | 376 | goto err; |
5ab11c98 PE |
377 | qp->q.last_in |= LAST_IN; |
378 | qp->q.len = end; | |
1da177e4 LT |
379 | } else { |
380 | if (end&7) { | |
381 | end &= ~7; | |
382 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
383 | skb->ip_summed = CHECKSUM_NONE; | |
384 | } | |
5ab11c98 | 385 | if (end > qp->q.len) { |
1da177e4 | 386 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 387 | if (qp->q.last_in & LAST_IN) |
1da177e4 | 388 | goto err; |
5ab11c98 | 389 | qp->q.len = end; |
1da177e4 LT |
390 | } |
391 | } | |
392 | if (end == offset) | |
393 | goto err; | |
394 | ||
1706d587 | 395 | err = -ENOMEM; |
1da177e4 LT |
396 | if (pskb_pull(skb, ihl) == NULL) |
397 | goto err; | |
1706d587 HX |
398 | |
399 | err = pskb_trim_rcsum(skb, end - offset); | |
400 | if (err) | |
1da177e4 LT |
401 | goto err; |
402 | ||
403 | /* Find out which fragments are in front and at the back of us | |
404 | * in the chain of fragments so far. We must know where to put | |
405 | * this fragment, right? | |
406 | */ | |
407 | prev = NULL; | |
5ab11c98 | 408 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
409 | if (FRAG_CB(next)->offset >= offset) |
410 | break; /* bingo! */ | |
411 | prev = next; | |
412 | } | |
413 | ||
414 | /* We found where to put this one. Check for overlap with | |
415 | * preceding fragment, and, if needed, align things so that | |
416 | * any overlaps are eliminated. | |
417 | */ | |
418 | if (prev) { | |
419 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
420 | ||
421 | if (i > 0) { | |
422 | offset += i; | |
1706d587 | 423 | err = -EINVAL; |
1da177e4 LT |
424 | if (end <= offset) |
425 | goto err; | |
1706d587 | 426 | err = -ENOMEM; |
1da177e4 LT |
427 | if (!pskb_pull(skb, i)) |
428 | goto err; | |
429 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
430 | skb->ip_summed = CHECKSUM_NONE; | |
431 | } | |
432 | } | |
433 | ||
1706d587 HX |
434 | err = -ENOMEM; |
435 | ||
1da177e4 LT |
436 | while (next && FRAG_CB(next)->offset < end) { |
437 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
438 | ||
439 | if (i < next->len) { | |
440 | /* Eat head of the next overlapped fragment | |
441 | * and leave the loop. The next ones cannot overlap. | |
442 | */ | |
443 | if (!pskb_pull(next, i)) | |
444 | goto err; | |
445 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 446 | qp->q.meat -= i; |
1da177e4 LT |
447 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
448 | next->ip_summed = CHECKSUM_NONE; | |
449 | break; | |
450 | } else { | |
451 | struct sk_buff *free_it = next; | |
452 | ||
47c6bf77 | 453 | /* Old fragment is completely overridden with |
1da177e4 LT |
454 | * new one drop it. |
455 | */ | |
456 | next = next->next; | |
457 | ||
458 | if (prev) | |
459 | prev->next = next; | |
460 | else | |
5ab11c98 | 461 | qp->q.fragments = next; |
1da177e4 | 462 | |
5ab11c98 | 463 | qp->q.meat -= free_it->len; |
1da177e4 LT |
464 | frag_kfree_skb(free_it, NULL); |
465 | } | |
466 | } | |
467 | ||
468 | FRAG_CB(skb)->offset = offset; | |
469 | ||
470 | /* Insert this fragment in the chain of fragments. */ | |
471 | skb->next = next; | |
472 | if (prev) | |
473 | prev->next = skb; | |
474 | else | |
5ab11c98 | 475 | qp->q.fragments = skb; |
1da177e4 | 476 | |
1706d587 HX |
477 | dev = skb->dev; |
478 | if (dev) { | |
479 | qp->iif = dev->ifindex; | |
480 | skb->dev = NULL; | |
481 | } | |
5ab11c98 PE |
482 | qp->q.stamp = skb->tstamp; |
483 | qp->q.meat += skb->len; | |
7eb95156 | 484 | atomic_add(skb->truesize, &ip4_frags.mem); |
1da177e4 | 485 | if (offset == 0) |
5ab11c98 | 486 | qp->q.last_in |= FIRST_IN; |
1da177e4 | 487 | |
5ab11c98 | 488 | if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len) |
1706d587 HX |
489 | return ip_frag_reasm(qp, prev, dev); |
490 | ||
7eb95156 PE |
491 | write_lock(&ip4_frags.lock); |
492 | list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list); | |
493 | write_unlock(&ip4_frags.lock); | |
1706d587 | 494 | return -EINPROGRESS; |
1da177e4 LT |
495 | |
496 | err: | |
497 | kfree_skb(skb); | |
1706d587 | 498 | return err; |
1da177e4 LT |
499 | } |
500 | ||
501 | ||
502 | /* Build a new IP datagram from all its fragments. */ | |
503 | ||
1706d587 HX |
504 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
505 | struct net_device *dev) | |
1da177e4 LT |
506 | { |
507 | struct iphdr *iph; | |
5ab11c98 | 508 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
509 | int len; |
510 | int ihlen; | |
1706d587 | 511 | int err; |
1da177e4 LT |
512 | |
513 | ipq_kill(qp); | |
514 | ||
1706d587 HX |
515 | /* Make the one we just received the head. */ |
516 | if (prev) { | |
517 | head = prev->next; | |
518 | fp = skb_clone(head, GFP_ATOMIC); | |
519 | ||
520 | if (!fp) | |
521 | goto out_nomem; | |
522 | ||
523 | fp->next = head->next; | |
524 | prev->next = fp; | |
525 | ||
5ab11c98 PE |
526 | skb_morph(head, qp->q.fragments); |
527 | head->next = qp->q.fragments->next; | |
1706d587 | 528 | |
5ab11c98 PE |
529 | kfree_skb(qp->q.fragments); |
530 | qp->q.fragments = head; | |
1706d587 HX |
531 | } |
532 | ||
1da177e4 LT |
533 | BUG_TRAP(head != NULL); |
534 | BUG_TRAP(FRAG_CB(head)->offset == 0); | |
535 | ||
536 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 537 | ihlen = ip_hdrlen(head); |
5ab11c98 | 538 | len = ihlen + qp->q.len; |
1da177e4 | 539 | |
1706d587 | 540 | err = -E2BIG; |
132adf54 | 541 | if (len > 65535) |
1da177e4 LT |
542 | goto out_oversize; |
543 | ||
544 | /* Head of list must not be cloned. */ | |
1706d587 | 545 | err = -ENOMEM; |
1da177e4 LT |
546 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) |
547 | goto out_nomem; | |
548 | ||
549 | /* If the first fragment is fragmented itself, we split | |
550 | * it to two chunks: the first with data and paged part | |
551 | * and the second, holding only fragments. */ | |
552 | if (skb_shinfo(head)->frag_list) { | |
553 | struct sk_buff *clone; | |
554 | int i, plen = 0; | |
555 | ||
556 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
557 | goto out_nomem; | |
558 | clone->next = head->next; | |
559 | head->next = clone; | |
560 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
561 | skb_shinfo(head)->frag_list = NULL; | |
562 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
563 | plen += skb_shinfo(head)->frags[i].size; | |
564 | clone->len = clone->data_len = head->data_len - plen; | |
565 | head->data_len -= clone->len; | |
566 | head->len -= clone->len; | |
567 | clone->csum = 0; | |
568 | clone->ip_summed = head->ip_summed; | |
7eb95156 | 569 | atomic_add(clone->truesize, &ip4_frags.mem); |
1da177e4 LT |
570 | } |
571 | ||
572 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 573 | skb_push(head, head->data - skb_network_header(head)); |
7eb95156 | 574 | atomic_sub(head->truesize, &ip4_frags.mem); |
1da177e4 LT |
575 | |
576 | for (fp=head->next; fp; fp = fp->next) { | |
577 | head->data_len += fp->len; | |
578 | head->len += fp->len; | |
579 | if (head->ip_summed != fp->ip_summed) | |
580 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 581 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
582 | head->csum = csum_add(head->csum, fp->csum); |
583 | head->truesize += fp->truesize; | |
7eb95156 | 584 | atomic_sub(fp->truesize, &ip4_frags.mem); |
1da177e4 LT |
585 | } |
586 | ||
587 | head->next = NULL; | |
588 | head->dev = dev; | |
5ab11c98 | 589 | head->tstamp = qp->q.stamp; |
1da177e4 | 590 | |
eddc9ec5 | 591 | iph = ip_hdr(head); |
1da177e4 LT |
592 | iph->frag_off = 0; |
593 | iph->tot_len = htons(len); | |
594 | IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS); | |
5ab11c98 | 595 | qp->q.fragments = NULL; |
1706d587 | 596 | return 0; |
1da177e4 LT |
597 | |
598 | out_nomem: | |
e905a9ed | 599 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 600 | "queue %p\n", qp); |
1da177e4 LT |
601 | goto out_fail; |
602 | out_oversize: | |
603 | if (net_ratelimit()) | |
604 | printk(KERN_INFO | |
605 | "Oversized IP packet from %d.%d.%d.%d.\n", | |
606 | NIPQUAD(qp->saddr)); | |
607 | out_fail: | |
608 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
1706d587 | 609 | return err; |
1da177e4 LT |
610 | } |
611 | ||
612 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 613 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 614 | { |
1da177e4 | 615 | struct ipq *qp; |
e905a9ed | 616 | |
1da177e4 LT |
617 | IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS); |
618 | ||
619 | /* Start by cleaning up the memory. */ | |
04128f23 | 620 | if (atomic_read(&ip4_frags.mem) > ip4_frags_ctl.high_thresh) |
1da177e4 LT |
621 | ip_evictor(); |
622 | ||
1da177e4 | 623 | /* Lookup (or create) queue header */ |
eddc9ec5 | 624 | if ((qp = ip_find(ip_hdr(skb), user)) != NULL) { |
1706d587 | 625 | int ret; |
1da177e4 | 626 | |
5ab11c98 | 627 | spin_lock(&qp->q.lock); |
1da177e4 | 628 | |
1706d587 | 629 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 630 | |
5ab11c98 | 631 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 632 | ipq_put(qp); |
776c729e | 633 | return ret; |
1da177e4 LT |
634 | } |
635 | ||
636 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
637 | kfree_skb(skb); | |
776c729e | 638 | return -ENOMEM; |
1da177e4 LT |
639 | } |
640 | ||
b7aa0bf7 | 641 | void __init ipfrag_init(void) |
1da177e4 | 642 | { |
04128f23 | 643 | ip4_frags.ctl = &ip4_frags_ctl; |
321a3a99 | 644 | ip4_frags.hashfn = ip4_hashfn; |
1e4b8287 PE |
645 | ip4_frags.destructor = ip4_frag_free; |
646 | ip4_frags.skb_free = NULL; | |
647 | ip4_frags.qsize = sizeof(struct ipq); | |
2588fe1d | 648 | ip4_frags.equal = ip4_frag_equal; |
e521db9d | 649 | ip4_frags.frag_expire = ip_expire; |
7eb95156 | 650 | inet_frags_init(&ip4_frags); |
1da177e4 LT |
651 | } |
652 | ||
653 | EXPORT_SYMBOL(ip_defrag); |