]>
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 | ||
8d8354d2 | 77 | static struct inet_frags_ctl ip4_frags_ctl __read_mostly = { |
04128f23 PE |
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 | |
e5a2bb84 | 98 | int ip_frag_nqueues(struct net *net) |
7eb95156 | 99 | { |
e5a2bb84 | 100 | return net->ipv4.frags.nqueues; |
7eb95156 | 101 | } |
1da177e4 | 102 | |
6ddc0822 | 103 | int ip_frag_mem(struct net *net) |
7eb95156 | 104 | { |
6ddc0822 | 105 | return atomic_read(&net->ipv4.frags.mem); |
7eb95156 | 106 | } |
1da177e4 | 107 | |
1706d587 HX |
108 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
109 | struct net_device *dev); | |
110 | ||
c6fda282 PE |
111 | struct ip4_create_arg { |
112 | struct iphdr *iph; | |
113 | u32 user; | |
114 | }; | |
115 | ||
18277770 | 116 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 117 | { |
18277770 AV |
118 | return jhash_3words((__force u32)id << 16 | prot, |
119 | (__force u32)saddr, (__force u32)daddr, | |
7eb95156 | 120 | ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
121 | } |
122 | ||
321a3a99 | 123 | static unsigned int ip4_hashfn(struct inet_frag_queue *q) |
1da177e4 | 124 | { |
321a3a99 | 125 | struct ipq *ipq; |
1da177e4 | 126 | |
321a3a99 PE |
127 | ipq = container_of(q, struct ipq, q); |
128 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
129 | } |
130 | ||
abd6523d PE |
131 | static int ip4_frag_match(struct inet_frag_queue *q, void *a) |
132 | { | |
133 | struct ipq *qp; | |
134 | struct ip4_create_arg *arg = a; | |
135 | ||
136 | qp = container_of(q, struct ipq, q); | |
137 | return (qp->id == arg->iph->id && | |
138 | qp->saddr == arg->iph->saddr && | |
139 | qp->daddr == arg->iph->daddr && | |
140 | qp->protocol == arg->iph->protocol && | |
141 | qp->user == arg->user); | |
142 | } | |
143 | ||
1da177e4 | 144 | /* Memory Tracking Functions. */ |
6ddc0822 PE |
145 | static __inline__ void frag_kfree_skb(struct netns_frags *nf, |
146 | struct sk_buff *skb, int *work) | |
1da177e4 LT |
147 | { |
148 | if (work) | |
149 | *work -= skb->truesize; | |
6ddc0822 | 150 | atomic_sub(skb->truesize, &nf->mem); |
1da177e4 LT |
151 | kfree_skb(skb); |
152 | } | |
153 | ||
c6fda282 PE |
154 | static void ip4_frag_init(struct inet_frag_queue *q, void *a) |
155 | { | |
156 | struct ipq *qp = container_of(q, struct ipq, q); | |
157 | struct ip4_create_arg *arg = a; | |
158 | ||
159 | qp->protocol = arg->iph->protocol; | |
160 | qp->id = arg->iph->id; | |
161 | qp->saddr = arg->iph->saddr; | |
162 | qp->daddr = arg->iph->daddr; | |
163 | qp->user = arg->user; | |
164 | qp->peer = sysctl_ipfrag_max_dist ? | |
165 | inet_getpeer(arg->iph->saddr, 1) : NULL; | |
166 | } | |
167 | ||
1e4b8287 | 168 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 169 | { |
1e4b8287 PE |
170 | struct ipq *qp; |
171 | ||
172 | qp = container_of(q, struct ipq, q); | |
173 | if (qp->peer) | |
174 | inet_putpeer(qp->peer); | |
1da177e4 LT |
175 | } |
176 | ||
1da177e4 LT |
177 | |
178 | /* Destruction primitives. */ | |
179 | ||
4b6cb5d8 | 180 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 181 | { |
762cc408 | 182 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
183 | } |
184 | ||
185 | /* Kill ipq entry. It is not destroyed immediately, | |
186 | * because caller (and someone more) holds reference count. | |
187 | */ | |
188 | static void ipq_kill(struct ipq *ipq) | |
189 | { | |
277e650d | 190 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
191 | } |
192 | ||
e905a9ed | 193 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
194 | * fragment queue until we are back under the threshold. |
195 | */ | |
6ddc0822 | 196 | static void ip_evictor(struct net *net) |
1da177e4 | 197 | { |
8e7999c4 PE |
198 | int evicted; |
199 | ||
6ddc0822 | 200 | evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags); |
8e7999c4 PE |
201 | if (evicted) |
202 | IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted); | |
1da177e4 LT |
203 | } |
204 | ||
205 | /* | |
206 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
207 | */ | |
208 | static void ip_expire(unsigned long arg) | |
209 | { | |
e521db9d PE |
210 | struct ipq *qp; |
211 | ||
212 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
1da177e4 | 213 | |
5ab11c98 | 214 | spin_lock(&qp->q.lock); |
1da177e4 | 215 | |
5ab11c98 | 216 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
217 | goto out; |
218 | ||
219 | ipq_kill(qp); | |
220 | ||
221 | IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT); | |
222 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
223 | ||
5ab11c98 PE |
224 | if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) { |
225 | struct sk_buff *head = qp->q.fragments; | |
1da177e4 | 226 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
881d966b | 227 | if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) { |
1da177e4 LT |
228 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); |
229 | dev_put(head->dev); | |
230 | } | |
231 | } | |
232 | out: | |
5ab11c98 | 233 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 234 | ipq_put(qp); |
1da177e4 LT |
235 | } |
236 | ||
abd6523d PE |
237 | /* Find the correct entry in the "incomplete datagrams" queue for |
238 | * this IP datagram, and create new one, if nothing is found. | |
239 | */ | |
ac18e750 | 240 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 241 | { |
c6fda282 PE |
242 | struct inet_frag_queue *q; |
243 | struct ip4_create_arg arg; | |
abd6523d | 244 | unsigned int hash; |
1da177e4 | 245 | |
c6fda282 PE |
246 | arg.iph = iph; |
247 | arg.user = user; | |
abd6523d | 248 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 249 | |
ac18e750 | 250 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
c6fda282 PE |
251 | if (q == NULL) |
252 | goto out_nomem; | |
1da177e4 | 253 | |
c6fda282 | 254 | return container_of(q, struct ipq, q); |
1da177e4 LT |
255 | |
256 | out_nomem: | |
64ce2073 | 257 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
258 | return NULL; |
259 | } | |
260 | ||
89cee8b1 HX |
261 | /* Is the fragment too far ahead to be part of ipq? */ |
262 | static inline int ip_frag_too_far(struct ipq *qp) | |
263 | { | |
264 | struct inet_peer *peer = qp->peer; | |
265 | unsigned int max = sysctl_ipfrag_max_dist; | |
266 | unsigned int start, end; | |
267 | ||
268 | int rc; | |
269 | ||
270 | if (!peer || !max) | |
271 | return 0; | |
272 | ||
273 | start = qp->rid; | |
274 | end = atomic_inc_return(&peer->rid); | |
275 | qp->rid = end; | |
276 | ||
5ab11c98 | 277 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
278 | |
279 | if (rc) { | |
280 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
281 | } | |
282 | ||
283 | return rc; | |
284 | } | |
285 | ||
286 | static int ip_frag_reinit(struct ipq *qp) | |
287 | { | |
288 | struct sk_buff *fp; | |
289 | ||
04128f23 | 290 | if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) { |
5ab11c98 | 291 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
292 | return -ETIMEDOUT; |
293 | } | |
294 | ||
5ab11c98 | 295 | fp = qp->q.fragments; |
89cee8b1 HX |
296 | do { |
297 | struct sk_buff *xp = fp->next; | |
6ddc0822 | 298 | frag_kfree_skb(qp->q.net, fp, NULL); |
89cee8b1 HX |
299 | fp = xp; |
300 | } while (fp); | |
301 | ||
5ab11c98 PE |
302 | qp->q.last_in = 0; |
303 | qp->q.len = 0; | |
304 | qp->q.meat = 0; | |
305 | qp->q.fragments = NULL; | |
89cee8b1 HX |
306 | qp->iif = 0; |
307 | ||
308 | return 0; | |
309 | } | |
310 | ||
1da177e4 | 311 | /* Add new segment to existing queue. */ |
1706d587 | 312 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
313 | { |
314 | struct sk_buff *prev, *next; | |
1706d587 | 315 | struct net_device *dev; |
1da177e4 LT |
316 | int flags, offset; |
317 | int ihl, end; | |
1706d587 | 318 | int err = -ENOENT; |
1da177e4 | 319 | |
5ab11c98 | 320 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
321 | goto err; |
322 | ||
89cee8b1 | 323 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
324 | unlikely(ip_frag_too_far(qp)) && |
325 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
326 | ipq_kill(qp); |
327 | goto err; | |
328 | } | |
329 | ||
eddc9ec5 | 330 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
331 | flags = offset & ~IP_OFFSET; |
332 | offset &= IP_OFFSET; | |
333 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 334 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
335 | |
336 | /* Determine the position of this fragment. */ | |
e905a9ed | 337 | end = offset + skb->len - ihl; |
1706d587 | 338 | err = -EINVAL; |
1da177e4 LT |
339 | |
340 | /* Is this the final fragment? */ | |
341 | if ((flags & IP_MF) == 0) { | |
342 | /* If we already have some bits beyond end | |
343 | * or have different end, the segment is corrrupted. | |
344 | */ | |
5ab11c98 PE |
345 | if (end < qp->q.len || |
346 | ((qp->q.last_in & LAST_IN) && end != qp->q.len)) | |
1da177e4 | 347 | goto err; |
5ab11c98 PE |
348 | qp->q.last_in |= LAST_IN; |
349 | qp->q.len = end; | |
1da177e4 LT |
350 | } else { |
351 | if (end&7) { | |
352 | end &= ~7; | |
353 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
354 | skb->ip_summed = CHECKSUM_NONE; | |
355 | } | |
5ab11c98 | 356 | if (end > qp->q.len) { |
1da177e4 | 357 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 358 | if (qp->q.last_in & LAST_IN) |
1da177e4 | 359 | goto err; |
5ab11c98 | 360 | qp->q.len = end; |
1da177e4 LT |
361 | } |
362 | } | |
363 | if (end == offset) | |
364 | goto err; | |
365 | ||
1706d587 | 366 | err = -ENOMEM; |
1da177e4 LT |
367 | if (pskb_pull(skb, ihl) == NULL) |
368 | goto err; | |
1706d587 HX |
369 | |
370 | err = pskb_trim_rcsum(skb, end - offset); | |
371 | if (err) | |
1da177e4 LT |
372 | goto err; |
373 | ||
374 | /* Find out which fragments are in front and at the back of us | |
375 | * in the chain of fragments so far. We must know where to put | |
376 | * this fragment, right? | |
377 | */ | |
378 | prev = NULL; | |
5ab11c98 | 379 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
380 | if (FRAG_CB(next)->offset >= offset) |
381 | break; /* bingo! */ | |
382 | prev = next; | |
383 | } | |
384 | ||
385 | /* We found where to put this one. Check for overlap with | |
386 | * preceding fragment, and, if needed, align things so that | |
387 | * any overlaps are eliminated. | |
388 | */ | |
389 | if (prev) { | |
390 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
391 | ||
392 | if (i > 0) { | |
393 | offset += i; | |
1706d587 | 394 | err = -EINVAL; |
1da177e4 LT |
395 | if (end <= offset) |
396 | goto err; | |
1706d587 | 397 | err = -ENOMEM; |
1da177e4 LT |
398 | if (!pskb_pull(skb, i)) |
399 | goto err; | |
400 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
401 | skb->ip_summed = CHECKSUM_NONE; | |
402 | } | |
403 | } | |
404 | ||
1706d587 HX |
405 | err = -ENOMEM; |
406 | ||
1da177e4 LT |
407 | while (next && FRAG_CB(next)->offset < end) { |
408 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
409 | ||
410 | if (i < next->len) { | |
411 | /* Eat head of the next overlapped fragment | |
412 | * and leave the loop. The next ones cannot overlap. | |
413 | */ | |
414 | if (!pskb_pull(next, i)) | |
415 | goto err; | |
416 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 417 | qp->q.meat -= i; |
1da177e4 LT |
418 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
419 | next->ip_summed = CHECKSUM_NONE; | |
420 | break; | |
421 | } else { | |
422 | struct sk_buff *free_it = next; | |
423 | ||
47c6bf77 | 424 | /* Old fragment is completely overridden with |
1da177e4 LT |
425 | * new one drop it. |
426 | */ | |
427 | next = next->next; | |
428 | ||
429 | if (prev) | |
430 | prev->next = next; | |
431 | else | |
5ab11c98 | 432 | qp->q.fragments = next; |
1da177e4 | 433 | |
5ab11c98 | 434 | qp->q.meat -= free_it->len; |
6ddc0822 | 435 | frag_kfree_skb(qp->q.net, free_it, NULL); |
1da177e4 LT |
436 | } |
437 | } | |
438 | ||
439 | FRAG_CB(skb)->offset = offset; | |
440 | ||
441 | /* Insert this fragment in the chain of fragments. */ | |
442 | skb->next = next; | |
443 | if (prev) | |
444 | prev->next = skb; | |
445 | else | |
5ab11c98 | 446 | qp->q.fragments = skb; |
1da177e4 | 447 | |
1706d587 HX |
448 | dev = skb->dev; |
449 | if (dev) { | |
450 | qp->iif = dev->ifindex; | |
451 | skb->dev = NULL; | |
452 | } | |
5ab11c98 PE |
453 | qp->q.stamp = skb->tstamp; |
454 | qp->q.meat += skb->len; | |
6ddc0822 | 455 | atomic_add(skb->truesize, &qp->q.net->mem); |
1da177e4 | 456 | if (offset == 0) |
5ab11c98 | 457 | qp->q.last_in |= FIRST_IN; |
1da177e4 | 458 | |
5ab11c98 | 459 | if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len) |
1706d587 HX |
460 | return ip_frag_reasm(qp, prev, dev); |
461 | ||
7eb95156 PE |
462 | write_lock(&ip4_frags.lock); |
463 | list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list); | |
464 | write_unlock(&ip4_frags.lock); | |
1706d587 | 465 | return -EINPROGRESS; |
1da177e4 LT |
466 | |
467 | err: | |
468 | kfree_skb(skb); | |
1706d587 | 469 | return err; |
1da177e4 LT |
470 | } |
471 | ||
472 | ||
473 | /* Build a new IP datagram from all its fragments. */ | |
474 | ||
1706d587 HX |
475 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
476 | struct net_device *dev) | |
1da177e4 LT |
477 | { |
478 | struct iphdr *iph; | |
5ab11c98 | 479 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
480 | int len; |
481 | int ihlen; | |
1706d587 | 482 | int err; |
1da177e4 LT |
483 | |
484 | ipq_kill(qp); | |
485 | ||
1706d587 HX |
486 | /* Make the one we just received the head. */ |
487 | if (prev) { | |
488 | head = prev->next; | |
489 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
490 | if (!fp) |
491 | goto out_nomem; | |
492 | ||
493 | fp->next = head->next; | |
494 | prev->next = fp; | |
495 | ||
5ab11c98 PE |
496 | skb_morph(head, qp->q.fragments); |
497 | head->next = qp->q.fragments->next; | |
1706d587 | 498 | |
5ab11c98 PE |
499 | kfree_skb(qp->q.fragments); |
500 | qp->q.fragments = head; | |
1706d587 HX |
501 | } |
502 | ||
1da177e4 LT |
503 | BUG_TRAP(head != NULL); |
504 | BUG_TRAP(FRAG_CB(head)->offset == 0); | |
505 | ||
506 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 507 | ihlen = ip_hdrlen(head); |
5ab11c98 | 508 | len = ihlen + qp->q.len; |
1da177e4 | 509 | |
1706d587 | 510 | err = -E2BIG; |
132adf54 | 511 | if (len > 65535) |
1da177e4 LT |
512 | goto out_oversize; |
513 | ||
514 | /* Head of list must not be cloned. */ | |
515 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
516 | goto out_nomem; | |
517 | ||
518 | /* If the first fragment is fragmented itself, we split | |
519 | * it to two chunks: the first with data and paged part | |
520 | * and the second, holding only fragments. */ | |
521 | if (skb_shinfo(head)->frag_list) { | |
522 | struct sk_buff *clone; | |
523 | int i, plen = 0; | |
524 | ||
525 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
526 | goto out_nomem; | |
527 | clone->next = head->next; | |
528 | head->next = clone; | |
529 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
530 | skb_shinfo(head)->frag_list = NULL; | |
531 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
532 | plen += skb_shinfo(head)->frags[i].size; | |
533 | clone->len = clone->data_len = head->data_len - plen; | |
534 | head->data_len -= clone->len; | |
535 | head->len -= clone->len; | |
536 | clone->csum = 0; | |
537 | clone->ip_summed = head->ip_summed; | |
6ddc0822 | 538 | atomic_add(clone->truesize, &qp->q.net->mem); |
1da177e4 LT |
539 | } |
540 | ||
541 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 542 | skb_push(head, head->data - skb_network_header(head)); |
6ddc0822 | 543 | atomic_sub(head->truesize, &qp->q.net->mem); |
1da177e4 LT |
544 | |
545 | for (fp=head->next; fp; fp = fp->next) { | |
546 | head->data_len += fp->len; | |
547 | head->len += fp->len; | |
548 | if (head->ip_summed != fp->ip_summed) | |
549 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 550 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
551 | head->csum = csum_add(head->csum, fp->csum); |
552 | head->truesize += fp->truesize; | |
6ddc0822 | 553 | atomic_sub(fp->truesize, &qp->q.net->mem); |
1da177e4 LT |
554 | } |
555 | ||
556 | head->next = NULL; | |
557 | head->dev = dev; | |
5ab11c98 | 558 | head->tstamp = qp->q.stamp; |
1da177e4 | 559 | |
eddc9ec5 | 560 | iph = ip_hdr(head); |
1da177e4 LT |
561 | iph->frag_off = 0; |
562 | iph->tot_len = htons(len); | |
563 | IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS); | |
5ab11c98 | 564 | qp->q.fragments = NULL; |
1706d587 | 565 | return 0; |
1da177e4 LT |
566 | |
567 | out_nomem: | |
e905a9ed | 568 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 569 | "queue %p\n", qp); |
45542479 | 570 | err = -ENOMEM; |
1da177e4 LT |
571 | goto out_fail; |
572 | out_oversize: | |
573 | if (net_ratelimit()) | |
574 | printk(KERN_INFO | |
575 | "Oversized IP packet from %d.%d.%d.%d.\n", | |
576 | NIPQUAD(qp->saddr)); | |
577 | out_fail: | |
578 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
1706d587 | 579 | return err; |
1da177e4 LT |
580 | } |
581 | ||
582 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 583 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 584 | { |
1da177e4 | 585 | struct ipq *qp; |
ac18e750 | 586 | struct net *net; |
e905a9ed | 587 | |
1da177e4 LT |
588 | IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS); |
589 | ||
ac18e750 | 590 | net = skb->dev->nd_net; |
1da177e4 | 591 | /* Start by cleaning up the memory. */ |
6ddc0822 PE |
592 | if (atomic_read(&net->ipv4.frags.mem) > ip4_frags_ctl.high_thresh) |
593 | ip_evictor(net); | |
1da177e4 | 594 | |
1da177e4 | 595 | /* Lookup (or create) queue header */ |
ac18e750 | 596 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 597 | int ret; |
1da177e4 | 598 | |
5ab11c98 | 599 | spin_lock(&qp->q.lock); |
1da177e4 | 600 | |
1706d587 | 601 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 602 | |
5ab11c98 | 603 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 604 | ipq_put(qp); |
776c729e | 605 | return ret; |
1da177e4 LT |
606 | } |
607 | ||
608 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
609 | kfree_skb(skb); | |
776c729e | 610 | return -ENOMEM; |
1da177e4 LT |
611 | } |
612 | ||
8d8354d2 PE |
613 | #ifdef CONFIG_SYSCTL |
614 | static int zero; | |
615 | ||
616 | static struct ctl_table ip4_frags_ctl_table[] = { | |
617 | { | |
618 | .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH, | |
619 | .procname = "ipfrag_high_thresh", | |
620 | .data = &ip4_frags_ctl.high_thresh, | |
621 | .maxlen = sizeof(int), | |
622 | .mode = 0644, | |
623 | .proc_handler = &proc_dointvec | |
624 | }, | |
625 | { | |
626 | .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH, | |
627 | .procname = "ipfrag_low_thresh", | |
628 | .data = &ip4_frags_ctl.low_thresh, | |
629 | .maxlen = sizeof(int), | |
630 | .mode = 0644, | |
631 | .proc_handler = &proc_dointvec | |
632 | }, | |
633 | { | |
634 | .ctl_name = NET_IPV4_IPFRAG_TIME, | |
635 | .procname = "ipfrag_time", | |
636 | .data = &ip4_frags_ctl.timeout, | |
637 | .maxlen = sizeof(int), | |
638 | .mode = 0644, | |
639 | .proc_handler = &proc_dointvec_jiffies, | |
640 | .strategy = &sysctl_jiffies | |
641 | }, | |
642 | { | |
643 | .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL, | |
644 | .procname = "ipfrag_secret_interval", | |
645 | .data = &ip4_frags_ctl.secret_interval, | |
646 | .maxlen = sizeof(int), | |
647 | .mode = 0644, | |
648 | .proc_handler = &proc_dointvec_jiffies, | |
649 | .strategy = &sysctl_jiffies | |
650 | }, | |
651 | { | |
652 | .procname = "ipfrag_max_dist", | |
653 | .data = &sysctl_ipfrag_max_dist, | |
654 | .maxlen = sizeof(int), | |
655 | .mode = 0644, | |
656 | .proc_handler = &proc_dointvec_minmax, | |
657 | .extra1 = &zero | |
658 | }, | |
659 | { } | |
660 | }; | |
661 | ||
662 | static int ip4_frags_ctl_register(struct net *net) | |
663 | { | |
664 | struct ctl_table_header *hdr; | |
665 | ||
666 | hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, | |
667 | ip4_frags_ctl_table); | |
668 | return hdr == NULL ? -ENOMEM : 0; | |
669 | } | |
670 | #else | |
671 | static inline int ip4_frags_ctl_register(struct net *net) | |
672 | { | |
673 | return 0; | |
674 | } | |
675 | #endif | |
676 | ||
677 | static int ipv4_frags_init_net(struct net *net) | |
678 | { | |
e5a2bb84 PE |
679 | inet_frags_init_net(&net->ipv4.frags); |
680 | ||
8d8354d2 PE |
681 | return ip4_frags_ctl_register(net); |
682 | } | |
683 | ||
b7aa0bf7 | 684 | void __init ipfrag_init(void) |
1da177e4 | 685 | { |
8d8354d2 | 686 | ipv4_frags_init_net(&init_net); |
04128f23 | 687 | ip4_frags.ctl = &ip4_frags_ctl; |
321a3a99 | 688 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 689 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
690 | ip4_frags.destructor = ip4_frag_free; |
691 | ip4_frags.skb_free = NULL; | |
692 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 693 | ip4_frags.match = ip4_frag_match; |
e521db9d | 694 | ip4_frags.frag_expire = ip_expire; |
7eb95156 | 695 | inet_frags_init(&ip4_frags); |
1da177e4 LT |
696 | } |
697 | ||
698 | EXPORT_SYMBOL(ip_defrag); |