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> | |
35 | #include <net/sock.h> | |
36 | #include <net/ip.h> | |
37 | #include <net/icmp.h> | |
38 | #include <net/checksum.h> | |
89cee8b1 | 39 | #include <net/inetpeer.h> |
5ab11c98 | 40 | #include <net/inet_frag.h> |
1da177e4 LT |
41 | #include <linux/tcp.h> |
42 | #include <linux/udp.h> | |
43 | #include <linux/inet.h> | |
44 | #include <linux/netfilter_ipv4.h> | |
45 | ||
46 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
47 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
48 | * as well. Or notify me, at least. --ANK | |
49 | */ | |
50 | ||
8d8354d2 | 51 | static int sysctl_ipfrag_max_dist __read_mostly = 64; |
89cee8b1 | 52 | |
1da177e4 LT |
53 | struct ipfrag_skb_cb |
54 | { | |
55 | struct inet_skb_parm h; | |
56 | int offset; | |
57 | }; | |
58 | ||
59 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb)) | |
60 | ||
61 | /* Describe an entry in the "incomplete datagrams" queue. */ | |
62 | struct ipq { | |
5ab11c98 PE |
63 | struct inet_frag_queue q; |
64 | ||
1da177e4 | 65 | u32 user; |
18277770 AV |
66 | __be32 saddr; |
67 | __be32 daddr; | |
68 | __be16 id; | |
1da177e4 | 69 | u8 protocol; |
89cee8b1 HX |
70 | int iif; |
71 | unsigned int rid; | |
72 | struct inet_peer *peer; | |
1da177e4 LT |
73 | }; |
74 | ||
7eb95156 | 75 | static struct inet_frags ip4_frags; |
1da177e4 | 76 | |
e5a2bb84 | 77 | int ip_frag_nqueues(struct net *net) |
7eb95156 | 78 | { |
e5a2bb84 | 79 | return net->ipv4.frags.nqueues; |
7eb95156 | 80 | } |
1da177e4 | 81 | |
6ddc0822 | 82 | int ip_frag_mem(struct net *net) |
7eb95156 | 83 | { |
6ddc0822 | 84 | return atomic_read(&net->ipv4.frags.mem); |
7eb95156 | 85 | } |
1da177e4 | 86 | |
1706d587 HX |
87 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
88 | struct net_device *dev); | |
89 | ||
c6fda282 PE |
90 | struct ip4_create_arg { |
91 | struct iphdr *iph; | |
92 | u32 user; | |
93 | }; | |
94 | ||
18277770 | 95 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 96 | { |
18277770 AV |
97 | return jhash_3words((__force u32)id << 16 | prot, |
98 | (__force u32)saddr, (__force u32)daddr, | |
7eb95156 | 99 | ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
100 | } |
101 | ||
321a3a99 | 102 | static unsigned int ip4_hashfn(struct inet_frag_queue *q) |
1da177e4 | 103 | { |
321a3a99 | 104 | struct ipq *ipq; |
1da177e4 | 105 | |
321a3a99 PE |
106 | ipq = container_of(q, struct ipq, q); |
107 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
108 | } |
109 | ||
abd6523d PE |
110 | static int ip4_frag_match(struct inet_frag_queue *q, void *a) |
111 | { | |
112 | struct ipq *qp; | |
113 | struct ip4_create_arg *arg = a; | |
114 | ||
115 | qp = container_of(q, struct ipq, q); | |
116 | return (qp->id == arg->iph->id && | |
117 | qp->saddr == arg->iph->saddr && | |
118 | qp->daddr == arg->iph->daddr && | |
119 | qp->protocol == arg->iph->protocol && | |
120 | qp->user == arg->user); | |
121 | } | |
122 | ||
1da177e4 | 123 | /* Memory Tracking Functions. */ |
6ddc0822 PE |
124 | static __inline__ void frag_kfree_skb(struct netns_frags *nf, |
125 | struct sk_buff *skb, int *work) | |
1da177e4 LT |
126 | { |
127 | if (work) | |
128 | *work -= skb->truesize; | |
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 | |
1da177e4 | 208 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
cb84663e | 209 | if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) { |
1da177e4 LT |
210 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); |
211 | dev_put(head->dev); | |
212 | } | |
213 | } | |
214 | out: | |
5ab11c98 | 215 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 216 | ipq_put(qp); |
1da177e4 LT |
217 | } |
218 | ||
abd6523d PE |
219 | /* Find the correct entry in the "incomplete datagrams" queue for |
220 | * this IP datagram, and create new one, if nothing is found. | |
221 | */ | |
ac18e750 | 222 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 223 | { |
c6fda282 PE |
224 | struct inet_frag_queue *q; |
225 | struct ip4_create_arg arg; | |
abd6523d | 226 | unsigned int hash; |
1da177e4 | 227 | |
c6fda282 PE |
228 | arg.iph = iph; |
229 | arg.user = user; | |
9a375803 PE |
230 | |
231 | read_lock(&ip4_frags.lock); | |
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) { | |
7c73a6fa PE |
264 | struct net *net; |
265 | ||
266 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
267 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
268 | } |
269 | ||
270 | return rc; | |
271 | } | |
272 | ||
273 | static int ip_frag_reinit(struct ipq *qp) | |
274 | { | |
275 | struct sk_buff *fp; | |
276 | ||
b2fd5321 | 277 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 278 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
279 | return -ETIMEDOUT; |
280 | } | |
281 | ||
5ab11c98 | 282 | fp = qp->q.fragments; |
89cee8b1 HX |
283 | do { |
284 | struct sk_buff *xp = fp->next; | |
6ddc0822 | 285 | frag_kfree_skb(qp->q.net, fp, NULL); |
89cee8b1 HX |
286 | fp = xp; |
287 | } while (fp); | |
288 | ||
5ab11c98 PE |
289 | qp->q.last_in = 0; |
290 | qp->q.len = 0; | |
291 | qp->q.meat = 0; | |
292 | qp->q.fragments = NULL; | |
89cee8b1 HX |
293 | qp->iif = 0; |
294 | ||
295 | return 0; | |
296 | } | |
297 | ||
1da177e4 | 298 | /* Add new segment to existing queue. */ |
1706d587 | 299 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
300 | { |
301 | struct sk_buff *prev, *next; | |
1706d587 | 302 | struct net_device *dev; |
1da177e4 LT |
303 | int flags, offset; |
304 | int ihl, end; | |
1706d587 | 305 | int err = -ENOENT; |
1da177e4 | 306 | |
bc578a54 | 307 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
308 | goto err; |
309 | ||
89cee8b1 | 310 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
311 | unlikely(ip_frag_too_far(qp)) && |
312 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
313 | ipq_kill(qp); |
314 | goto err; | |
315 | } | |
316 | ||
eddc9ec5 | 317 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
318 | flags = offset & ~IP_OFFSET; |
319 | offset &= IP_OFFSET; | |
320 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 321 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
322 | |
323 | /* Determine the position of this fragment. */ | |
e905a9ed | 324 | end = offset + skb->len - ihl; |
1706d587 | 325 | err = -EINVAL; |
1da177e4 LT |
326 | |
327 | /* Is this the final fragment? */ | |
328 | if ((flags & IP_MF) == 0) { | |
329 | /* If we already have some bits beyond end | |
330 | * or have different end, the segment is corrrupted. | |
331 | */ | |
5ab11c98 | 332 | if (end < qp->q.len || |
bc578a54 | 333 | ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 334 | goto err; |
bc578a54 | 335 | qp->q.last_in |= INET_FRAG_LAST_IN; |
5ab11c98 | 336 | qp->q.len = end; |
1da177e4 LT |
337 | } else { |
338 | if (end&7) { | |
339 | end &= ~7; | |
340 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
341 | skb->ip_summed = CHECKSUM_NONE; | |
342 | } | |
5ab11c98 | 343 | if (end > qp->q.len) { |
1da177e4 | 344 | /* Some bits beyond end -> corruption. */ |
bc578a54 | 345 | if (qp->q.last_in & INET_FRAG_LAST_IN) |
1da177e4 | 346 | goto err; |
5ab11c98 | 347 | qp->q.len = end; |
1da177e4 LT |
348 | } |
349 | } | |
350 | if (end == offset) | |
351 | goto err; | |
352 | ||
1706d587 | 353 | err = -ENOMEM; |
1da177e4 LT |
354 | if (pskb_pull(skb, ihl) == NULL) |
355 | goto err; | |
1706d587 HX |
356 | |
357 | err = pskb_trim_rcsum(skb, end - offset); | |
358 | if (err) | |
1da177e4 LT |
359 | goto err; |
360 | ||
361 | /* Find out which fragments are in front and at the back of us | |
362 | * in the chain of fragments so far. We must know where to put | |
363 | * this fragment, right? | |
364 | */ | |
365 | prev = NULL; | |
5ab11c98 | 366 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
367 | if (FRAG_CB(next)->offset >= offset) |
368 | break; /* bingo! */ | |
369 | prev = next; | |
370 | } | |
371 | ||
372 | /* We found where to put this one. Check for overlap with | |
373 | * preceding fragment, and, if needed, align things so that | |
374 | * any overlaps are eliminated. | |
375 | */ | |
376 | if (prev) { | |
377 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
378 | ||
379 | if (i > 0) { | |
380 | offset += i; | |
1706d587 | 381 | err = -EINVAL; |
1da177e4 LT |
382 | if (end <= offset) |
383 | goto err; | |
1706d587 | 384 | err = -ENOMEM; |
1da177e4 LT |
385 | if (!pskb_pull(skb, i)) |
386 | goto err; | |
387 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
388 | skb->ip_summed = CHECKSUM_NONE; | |
389 | } | |
390 | } | |
391 | ||
1706d587 HX |
392 | err = -ENOMEM; |
393 | ||
1da177e4 LT |
394 | while (next && FRAG_CB(next)->offset < end) { |
395 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
396 | ||
397 | if (i < next->len) { | |
398 | /* Eat head of the next overlapped fragment | |
399 | * and leave the loop. The next ones cannot overlap. | |
400 | */ | |
401 | if (!pskb_pull(next, i)) | |
402 | goto err; | |
403 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 404 | qp->q.meat -= i; |
1da177e4 LT |
405 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
406 | next->ip_summed = CHECKSUM_NONE; | |
407 | break; | |
408 | } else { | |
409 | struct sk_buff *free_it = next; | |
410 | ||
47c6bf77 | 411 | /* Old fragment is completely overridden with |
1da177e4 LT |
412 | * new one drop it. |
413 | */ | |
414 | next = next->next; | |
415 | ||
416 | if (prev) | |
417 | prev->next = next; | |
418 | else | |
5ab11c98 | 419 | qp->q.fragments = next; |
1da177e4 | 420 | |
5ab11c98 | 421 | qp->q.meat -= free_it->len; |
6ddc0822 | 422 | frag_kfree_skb(qp->q.net, free_it, NULL); |
1da177e4 LT |
423 | } |
424 | } | |
425 | ||
426 | FRAG_CB(skb)->offset = offset; | |
427 | ||
428 | /* Insert this fragment in the chain of fragments. */ | |
429 | skb->next = next; | |
430 | if (prev) | |
431 | prev->next = skb; | |
432 | else | |
5ab11c98 | 433 | qp->q.fragments = skb; |
1da177e4 | 434 | |
1706d587 HX |
435 | dev = skb->dev; |
436 | if (dev) { | |
437 | qp->iif = dev->ifindex; | |
438 | skb->dev = NULL; | |
439 | } | |
5ab11c98 PE |
440 | qp->q.stamp = skb->tstamp; |
441 | qp->q.meat += skb->len; | |
6ddc0822 | 442 | atomic_add(skb->truesize, &qp->q.net->mem); |
1da177e4 | 443 | if (offset == 0) |
bc578a54 | 444 | qp->q.last_in |= INET_FRAG_FIRST_IN; |
1da177e4 | 445 | |
bc578a54 JP |
446 | if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
447 | qp->q.meat == qp->q.len) | |
1706d587 HX |
448 | return ip_frag_reasm(qp, prev, dev); |
449 | ||
7eb95156 | 450 | write_lock(&ip4_frags.lock); |
3140c25c | 451 | list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list); |
7eb95156 | 452 | write_unlock(&ip4_frags.lock); |
1706d587 | 453 | return -EINPROGRESS; |
1da177e4 LT |
454 | |
455 | err: | |
456 | kfree_skb(skb); | |
1706d587 | 457 | return err; |
1da177e4 LT |
458 | } |
459 | ||
460 | ||
461 | /* Build a new IP datagram from all its fragments. */ | |
462 | ||
1706d587 HX |
463 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
464 | struct net_device *dev) | |
1da177e4 LT |
465 | { |
466 | struct iphdr *iph; | |
5ab11c98 | 467 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
468 | int len; |
469 | int ihlen; | |
1706d587 | 470 | int err; |
1da177e4 LT |
471 | |
472 | ipq_kill(qp); | |
473 | ||
1706d587 HX |
474 | /* Make the one we just received the head. */ |
475 | if (prev) { | |
476 | head = prev->next; | |
477 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
478 | if (!fp) |
479 | goto out_nomem; | |
480 | ||
481 | fp->next = head->next; | |
482 | prev->next = fp; | |
483 | ||
5ab11c98 PE |
484 | skb_morph(head, qp->q.fragments); |
485 | head->next = qp->q.fragments->next; | |
1706d587 | 486 | |
5ab11c98 PE |
487 | kfree_skb(qp->q.fragments); |
488 | qp->q.fragments = head; | |
1706d587 HX |
489 | } |
490 | ||
547b792c IJ |
491 | WARN_ON(head == NULL); |
492 | WARN_ON(FRAG_CB(head)->offset != 0); | |
1da177e4 LT |
493 | |
494 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 495 | ihlen = ip_hdrlen(head); |
5ab11c98 | 496 | len = ihlen + qp->q.len; |
1da177e4 | 497 | |
1706d587 | 498 | err = -E2BIG; |
132adf54 | 499 | if (len > 65535) |
1da177e4 LT |
500 | goto out_oversize; |
501 | ||
502 | /* Head of list must not be cloned. */ | |
503 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
504 | goto out_nomem; | |
505 | ||
506 | /* If the first fragment is fragmented itself, we split | |
507 | * it to two chunks: the first with data and paged part | |
508 | * and the second, holding only fragments. */ | |
509 | if (skb_shinfo(head)->frag_list) { | |
510 | struct sk_buff *clone; | |
511 | int i, plen = 0; | |
512 | ||
513 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
514 | goto out_nomem; | |
515 | clone->next = head->next; | |
516 | head->next = clone; | |
517 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
518 | skb_shinfo(head)->frag_list = NULL; | |
519 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
520 | plen += skb_shinfo(head)->frags[i].size; | |
521 | clone->len = clone->data_len = head->data_len - plen; | |
522 | head->data_len -= clone->len; | |
523 | head->len -= clone->len; | |
524 | clone->csum = 0; | |
525 | clone->ip_summed = head->ip_summed; | |
6ddc0822 | 526 | atomic_add(clone->truesize, &qp->q.net->mem); |
1da177e4 LT |
527 | } |
528 | ||
529 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 530 | skb_push(head, head->data - skb_network_header(head)); |
6ddc0822 | 531 | atomic_sub(head->truesize, &qp->q.net->mem); |
1da177e4 LT |
532 | |
533 | for (fp=head->next; fp; fp = fp->next) { | |
534 | head->data_len += fp->len; | |
535 | head->len += fp->len; | |
536 | if (head->ip_summed != fp->ip_summed) | |
537 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 538 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
539 | head->csum = csum_add(head->csum, fp->csum); |
540 | head->truesize += fp->truesize; | |
6ddc0822 | 541 | atomic_sub(fp->truesize, &qp->q.net->mem); |
1da177e4 LT |
542 | } |
543 | ||
544 | head->next = NULL; | |
545 | head->dev = dev; | |
5ab11c98 | 546 | head->tstamp = qp->q.stamp; |
1da177e4 | 547 | |
eddc9ec5 | 548 | iph = ip_hdr(head); |
1da177e4 LT |
549 | iph->frag_off = 0; |
550 | iph->tot_len = htons(len); | |
7c73a6fa | 551 | IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_REASMOKS); |
5ab11c98 | 552 | qp->q.fragments = NULL; |
1706d587 | 553 | return 0; |
1da177e4 LT |
554 | |
555 | out_nomem: | |
e905a9ed | 556 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 557 | "queue %p\n", qp); |
45542479 | 558 | err = -ENOMEM; |
1da177e4 LT |
559 | goto out_fail; |
560 | out_oversize: | |
561 | if (net_ratelimit()) | |
562 | printk(KERN_INFO | |
a7d632b6 | 563 | "Oversized IP packet from " NIPQUAD_FMT ".\n", |
1da177e4 LT |
564 | NIPQUAD(qp->saddr)); |
565 | out_fail: | |
7c73a6fa | 566 | IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_REASMFAILS); |
1706d587 | 567 | return err; |
1da177e4 LT |
568 | } |
569 | ||
570 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 571 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 572 | { |
1da177e4 | 573 | struct ipq *qp; |
ac18e750 | 574 | struct net *net; |
e905a9ed | 575 | |
84a3aa00 | 576 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev); |
7c73a6fa | 577 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
578 | |
579 | /* Start by cleaning up the memory. */ | |
e31e0bdc | 580 | if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh) |
6ddc0822 | 581 | ip_evictor(net); |
1da177e4 | 582 | |
1da177e4 | 583 | /* Lookup (or create) queue header */ |
ac18e750 | 584 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 585 | int ret; |
1da177e4 | 586 | |
5ab11c98 | 587 | spin_lock(&qp->q.lock); |
1da177e4 | 588 | |
1706d587 | 589 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 590 | |
5ab11c98 | 591 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 592 | ipq_put(qp); |
776c729e | 593 | return ret; |
1da177e4 LT |
594 | } |
595 | ||
7c73a6fa | 596 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 597 | kfree_skb(skb); |
776c729e | 598 | return -ENOMEM; |
1da177e4 LT |
599 | } |
600 | ||
8d8354d2 PE |
601 | #ifdef CONFIG_SYSCTL |
602 | static int zero; | |
603 | ||
0a64b4b8 | 604 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 PE |
605 | { |
606 | .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH, | |
607 | .procname = "ipfrag_high_thresh", | |
e31e0bdc | 608 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
609 | .maxlen = sizeof(int), |
610 | .mode = 0644, | |
611 | .proc_handler = &proc_dointvec | |
612 | }, | |
613 | { | |
614 | .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH, | |
615 | .procname = "ipfrag_low_thresh", | |
e31e0bdc | 616 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
617 | .maxlen = sizeof(int), |
618 | .mode = 0644, | |
619 | .proc_handler = &proc_dointvec | |
620 | }, | |
621 | { | |
622 | .ctl_name = NET_IPV4_IPFRAG_TIME, | |
623 | .procname = "ipfrag_time", | |
b2fd5321 | 624 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
625 | .maxlen = sizeof(int), |
626 | .mode = 0644, | |
627 | .proc_handler = &proc_dointvec_jiffies, | |
628 | .strategy = &sysctl_jiffies | |
629 | }, | |
7d291ebb PE |
630 | { } |
631 | }; | |
632 | ||
633 | static struct ctl_table ip4_frags_ctl_table[] = { | |
8d8354d2 PE |
634 | { |
635 | .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL, | |
636 | .procname = "ipfrag_secret_interval", | |
3b4bc4a2 | 637 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
638 | .maxlen = sizeof(int), |
639 | .mode = 0644, | |
640 | .proc_handler = &proc_dointvec_jiffies, | |
641 | .strategy = &sysctl_jiffies | |
642 | }, | |
643 | { | |
644 | .procname = "ipfrag_max_dist", | |
645 | .data = &sysctl_ipfrag_max_dist, | |
646 | .maxlen = sizeof(int), | |
647 | .mode = 0644, | |
648 | .proc_handler = &proc_dointvec_minmax, | |
649 | .extra1 = &zero | |
650 | }, | |
651 | { } | |
652 | }; | |
653 | ||
0a64b4b8 | 654 | static int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 655 | { |
e4a2d5c2 | 656 | struct ctl_table *table; |
8d8354d2 PE |
657 | struct ctl_table_header *hdr; |
658 | ||
0a64b4b8 | 659 | table = ip4_frags_ns_ctl_table; |
e4a2d5c2 | 660 | if (net != &init_net) { |
0a64b4b8 | 661 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
e4a2d5c2 PE |
662 | if (table == NULL) |
663 | goto err_alloc; | |
664 | ||
e31e0bdc PE |
665 | table[0].data = &net->ipv4.frags.high_thresh; |
666 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 667 | table[2].data = &net->ipv4.frags.timeout; |
e4a2d5c2 PE |
668 | } |
669 | ||
670 | hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table); | |
671 | if (hdr == NULL) | |
672 | goto err_reg; | |
673 | ||
674 | net->ipv4.frags_hdr = hdr; | |
675 | return 0; | |
676 | ||
677 | err_reg: | |
678 | if (net != &init_net) | |
679 | kfree(table); | |
680 | err_alloc: | |
681 | return -ENOMEM; | |
682 | } | |
683 | ||
0a64b4b8 | 684 | static void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
685 | { |
686 | struct ctl_table *table; | |
687 | ||
688 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
689 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
690 | kfree(table); | |
8d8354d2 | 691 | } |
7d291ebb PE |
692 | |
693 | static void ip4_frags_ctl_register(void) | |
694 | { | |
695 | register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table); | |
696 | } | |
8d8354d2 | 697 | #else |
0a64b4b8 | 698 | static inline int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
699 | { |
700 | return 0; | |
701 | } | |
e4a2d5c2 | 702 | |
0a64b4b8 | 703 | static inline void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
704 | { |
705 | } | |
7d291ebb PE |
706 | |
707 | static inline void ip4_frags_ctl_register(void) | |
708 | { | |
709 | } | |
8d8354d2 PE |
710 | #endif |
711 | ||
712 | static int ipv4_frags_init_net(struct net *net) | |
713 | { | |
e31e0bdc PE |
714 | /* |
715 | * Fragment cache limits. We will commit 256K at one time. Should we | |
716 | * cross that limit we will prune down to 192K. This should cope with | |
717 | * even the most extreme cases without allowing an attacker to | |
718 | * measurably harm machine performance. | |
719 | */ | |
720 | net->ipv4.frags.high_thresh = 256 * 1024; | |
721 | net->ipv4.frags.low_thresh = 192 * 1024; | |
b2fd5321 PE |
722 | /* |
723 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
724 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
725 | * by TTL. | |
726 | */ | |
727 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
728 | ||
e5a2bb84 PE |
729 | inet_frags_init_net(&net->ipv4.frags); |
730 | ||
0a64b4b8 | 731 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
732 | } |
733 | ||
81566e83 PE |
734 | static void ipv4_frags_exit_net(struct net *net) |
735 | { | |
0a64b4b8 | 736 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
737 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
738 | } | |
739 | ||
740 | static struct pernet_operations ip4_frags_ops = { | |
741 | .init = ipv4_frags_init_net, | |
742 | .exit = ipv4_frags_exit_net, | |
743 | }; | |
744 | ||
b7aa0bf7 | 745 | void __init ipfrag_init(void) |
1da177e4 | 746 | { |
7d291ebb | 747 | ip4_frags_ctl_register(); |
81566e83 | 748 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 749 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 750 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
751 | ip4_frags.destructor = ip4_frag_free; |
752 | ip4_frags.skb_free = NULL; | |
753 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 754 | ip4_frags.match = ip4_frag_match; |
e521db9d | 755 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 756 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 757 | inet_frags_init(&ip4_frags); |
1da177e4 LT |
758 | } |
759 | ||
760 | EXPORT_SYMBOL(ip_defrag); |