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 | ||
1da177e4 LT |
126 | /* Memory Tracking Functions. */ |
127 | static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work) | |
128 | { | |
129 | if (work) | |
130 | *work -= skb->truesize; | |
7eb95156 | 131 | atomic_sub(skb->truesize, &ip4_frags.mem); |
1da177e4 LT |
132 | kfree_skb(skb); |
133 | } | |
134 | ||
1e4b8287 | 135 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 136 | { |
1e4b8287 PE |
137 | struct ipq *qp; |
138 | ||
139 | qp = container_of(q, struct ipq, q); | |
140 | if (qp->peer) | |
141 | inet_putpeer(qp->peer); | |
1da177e4 LT |
142 | kfree(qp); |
143 | } | |
144 | ||
145 | static __inline__ struct ipq *frag_alloc_queue(void) | |
146 | { | |
147 | struct ipq *qp = kmalloc(sizeof(struct ipq), GFP_ATOMIC); | |
148 | ||
132adf54 | 149 | if (!qp) |
1da177e4 | 150 | return NULL; |
7eb95156 | 151 | atomic_add(sizeof(struct ipq), &ip4_frags.mem); |
1da177e4 LT |
152 | return qp; |
153 | } | |
154 | ||
155 | ||
156 | /* Destruction primitives. */ | |
157 | ||
4b6cb5d8 | 158 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 159 | { |
5ab11c98 | 160 | if (atomic_dec_and_test(&ipq->q.refcnt)) |
4b6cb5d8 | 161 | inet_frag_destroy(&ipq->q, &ip4_frags, NULL); |
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 | */ | |
175 | static void ip_evictor(void) | |
176 | { | |
8e7999c4 PE |
177 | int evicted; |
178 | ||
179 | evicted = inet_frag_evictor(&ip4_frags); | |
180 | if (evicted) | |
181 | IP_ADD_STATS_BH(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 | { | |
189 | struct ipq *qp = (struct ipq *) arg; | |
190 | ||
5ab11c98 | 191 | spin_lock(&qp->q.lock); |
1da177e4 | 192 | |
5ab11c98 | 193 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
194 | goto out; |
195 | ||
196 | ipq_kill(qp); | |
197 | ||
198 | IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT); | |
199 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
200 | ||
5ab11c98 PE |
201 | if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) { |
202 | struct sk_buff *head = qp->q.fragments; | |
1da177e4 | 203 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
881d966b | 204 | if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) { |
1da177e4 LT |
205 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); |
206 | dev_put(head->dev); | |
207 | } | |
208 | } | |
209 | out: | |
5ab11c98 | 210 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 211 | ipq_put(qp); |
1da177e4 LT |
212 | } |
213 | ||
214 | /* Creation primitives. */ | |
215 | ||
55c0022e | 216 | static struct ipq *ip_frag_intern(struct ipq *qp_in) |
1da177e4 LT |
217 | { |
218 | struct ipq *qp; | |
e7c8a41e YK |
219 | #ifdef CONFIG_SMP |
220 | struct hlist_node *n; | |
221 | #endif | |
55c0022e DM |
222 | unsigned int hash; |
223 | ||
7eb95156 | 224 | write_lock(&ip4_frags.lock); |
55c0022e DM |
225 | hash = ipqhashfn(qp_in->id, qp_in->saddr, qp_in->daddr, |
226 | qp_in->protocol); | |
1da177e4 LT |
227 | #ifdef CONFIG_SMP |
228 | /* With SMP race we have to recheck hash table, because | |
229 | * such entry could be created on other cpu, while we | |
230 | * promoted read lock to write lock. | |
231 | */ | |
7eb95156 | 232 | hlist_for_each_entry(qp, n, &ip4_frags.hash[hash], q.list) { |
132adf54 SH |
233 | if (qp->id == qp_in->id && |
234 | qp->saddr == qp_in->saddr && | |
235 | qp->daddr == qp_in->daddr && | |
236 | qp->protocol == qp_in->protocol && | |
237 | qp->user == qp_in->user) { | |
5ab11c98 | 238 | atomic_inc(&qp->q.refcnt); |
7eb95156 | 239 | write_unlock(&ip4_frags.lock); |
5ab11c98 | 240 | qp_in->q.last_in |= COMPLETE; |
4b6cb5d8 | 241 | ipq_put(qp_in); |
1da177e4 LT |
242 | return qp; |
243 | } | |
244 | } | |
245 | #endif | |
246 | qp = qp_in; | |
247 | ||
04128f23 | 248 | if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) |
5ab11c98 | 249 | atomic_inc(&qp->q.refcnt); |
1da177e4 | 250 | |
5ab11c98 | 251 | atomic_inc(&qp->q.refcnt); |
7eb95156 | 252 | hlist_add_head(&qp->q.list, &ip4_frags.hash[hash]); |
5ab11c98 | 253 | INIT_LIST_HEAD(&qp->q.lru_list); |
7eb95156 PE |
254 | list_add_tail(&qp->q.lru_list, &ip4_frags.lru_list); |
255 | ip4_frags.nqueues++; | |
256 | write_unlock(&ip4_frags.lock); | |
1da177e4 LT |
257 | return qp; |
258 | } | |
259 | ||
260 | /* Add an entry to the 'ipq' queue for a newly received IP datagram. */ | |
55c0022e | 261 | static struct ipq *ip_frag_create(struct iphdr *iph, u32 user) |
1da177e4 LT |
262 | { |
263 | struct ipq *qp; | |
264 | ||
265 | if ((qp = frag_alloc_queue()) == NULL) | |
266 | goto out_nomem; | |
267 | ||
268 | qp->protocol = iph->protocol; | |
5ab11c98 | 269 | qp->q.last_in = 0; |
1da177e4 LT |
270 | qp->id = iph->id; |
271 | qp->saddr = iph->saddr; | |
272 | qp->daddr = iph->daddr; | |
273 | qp->user = user; | |
5ab11c98 PE |
274 | qp->q.len = 0; |
275 | qp->q.meat = 0; | |
276 | qp->q.fragments = NULL; | |
1da177e4 | 277 | qp->iif = 0; |
89cee8b1 | 278 | qp->peer = sysctl_ipfrag_max_dist ? inet_getpeer(iph->saddr, 1) : NULL; |
1da177e4 LT |
279 | |
280 | /* Initialize a timer for this entry. */ | |
5ab11c98 PE |
281 | init_timer(&qp->q.timer); |
282 | qp->q.timer.data = (unsigned long) qp; /* pointer to queue */ | |
283 | qp->q.timer.function = ip_expire; /* expire function */ | |
284 | spin_lock_init(&qp->q.lock); | |
285 | atomic_set(&qp->q.refcnt, 1); | |
1da177e4 | 286 | |
55c0022e | 287 | return ip_frag_intern(qp); |
1da177e4 LT |
288 | |
289 | out_nomem: | |
64ce2073 | 290 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
291 | return NULL; |
292 | } | |
293 | ||
294 | /* Find the correct entry in the "incomplete datagrams" queue for | |
295 | * this IP datagram, and create new one, if nothing is found. | |
296 | */ | |
297 | static inline struct ipq *ip_find(struct iphdr *iph, u32 user) | |
298 | { | |
76ab608d | 299 | __be16 id = iph->id; |
18277770 AV |
300 | __be32 saddr = iph->saddr; |
301 | __be32 daddr = iph->daddr; | |
1da177e4 | 302 | __u8 protocol = iph->protocol; |
55c0022e | 303 | unsigned int hash; |
1da177e4 | 304 | struct ipq *qp; |
e7c8a41e | 305 | struct hlist_node *n; |
1da177e4 | 306 | |
7eb95156 | 307 | read_lock(&ip4_frags.lock); |
55c0022e | 308 | hash = ipqhashfn(id, saddr, daddr, protocol); |
7eb95156 | 309 | hlist_for_each_entry(qp, n, &ip4_frags.hash[hash], q.list) { |
132adf54 SH |
310 | if (qp->id == id && |
311 | qp->saddr == saddr && | |
312 | qp->daddr == daddr && | |
313 | qp->protocol == protocol && | |
314 | qp->user == user) { | |
5ab11c98 | 315 | atomic_inc(&qp->q.refcnt); |
7eb95156 | 316 | read_unlock(&ip4_frags.lock); |
1da177e4 LT |
317 | return qp; |
318 | } | |
319 | } | |
7eb95156 | 320 | read_unlock(&ip4_frags.lock); |
1da177e4 | 321 | |
55c0022e | 322 | return ip_frag_create(iph, user); |
1da177e4 LT |
323 | } |
324 | ||
89cee8b1 HX |
325 | /* Is the fragment too far ahead to be part of ipq? */ |
326 | static inline int ip_frag_too_far(struct ipq *qp) | |
327 | { | |
328 | struct inet_peer *peer = qp->peer; | |
329 | unsigned int max = sysctl_ipfrag_max_dist; | |
330 | unsigned int start, end; | |
331 | ||
332 | int rc; | |
333 | ||
334 | if (!peer || !max) | |
335 | return 0; | |
336 | ||
337 | start = qp->rid; | |
338 | end = atomic_inc_return(&peer->rid); | |
339 | qp->rid = end; | |
340 | ||
5ab11c98 | 341 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
342 | |
343 | if (rc) { | |
344 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
345 | } | |
346 | ||
347 | return rc; | |
348 | } | |
349 | ||
350 | static int ip_frag_reinit(struct ipq *qp) | |
351 | { | |
352 | struct sk_buff *fp; | |
353 | ||
04128f23 | 354 | if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) { |
5ab11c98 | 355 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
356 | return -ETIMEDOUT; |
357 | } | |
358 | ||
5ab11c98 | 359 | fp = qp->q.fragments; |
89cee8b1 HX |
360 | do { |
361 | struct sk_buff *xp = fp->next; | |
362 | frag_kfree_skb(fp, NULL); | |
363 | fp = xp; | |
364 | } while (fp); | |
365 | ||
5ab11c98 PE |
366 | qp->q.last_in = 0; |
367 | qp->q.len = 0; | |
368 | qp->q.meat = 0; | |
369 | qp->q.fragments = NULL; | |
89cee8b1 HX |
370 | qp->iif = 0; |
371 | ||
372 | return 0; | |
373 | } | |
374 | ||
1da177e4 | 375 | /* Add new segment to existing queue. */ |
1706d587 | 376 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
377 | { |
378 | struct sk_buff *prev, *next; | |
1706d587 | 379 | struct net_device *dev; |
1da177e4 LT |
380 | int flags, offset; |
381 | int ihl, end; | |
1706d587 | 382 | int err = -ENOENT; |
1da177e4 | 383 | |
5ab11c98 | 384 | if (qp->q.last_in & COMPLETE) |
1da177e4 LT |
385 | goto err; |
386 | ||
89cee8b1 | 387 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
388 | unlikely(ip_frag_too_far(qp)) && |
389 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
390 | ipq_kill(qp); |
391 | goto err; | |
392 | } | |
393 | ||
eddc9ec5 | 394 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
395 | flags = offset & ~IP_OFFSET; |
396 | offset &= IP_OFFSET; | |
397 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 398 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
399 | |
400 | /* Determine the position of this fragment. */ | |
e905a9ed | 401 | end = offset + skb->len - ihl; |
1706d587 | 402 | err = -EINVAL; |
1da177e4 LT |
403 | |
404 | /* Is this the final fragment? */ | |
405 | if ((flags & IP_MF) == 0) { | |
406 | /* If we already have some bits beyond end | |
407 | * or have different end, the segment is corrrupted. | |
408 | */ | |
5ab11c98 PE |
409 | if (end < qp->q.len || |
410 | ((qp->q.last_in & LAST_IN) && end != qp->q.len)) | |
1da177e4 | 411 | goto err; |
5ab11c98 PE |
412 | qp->q.last_in |= LAST_IN; |
413 | qp->q.len = end; | |
1da177e4 LT |
414 | } else { |
415 | if (end&7) { | |
416 | end &= ~7; | |
417 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
418 | skb->ip_summed = CHECKSUM_NONE; | |
419 | } | |
5ab11c98 | 420 | if (end > qp->q.len) { |
1da177e4 | 421 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 422 | if (qp->q.last_in & LAST_IN) |
1da177e4 | 423 | goto err; |
5ab11c98 | 424 | qp->q.len = end; |
1da177e4 LT |
425 | } |
426 | } | |
427 | if (end == offset) | |
428 | goto err; | |
429 | ||
1706d587 | 430 | err = -ENOMEM; |
1da177e4 LT |
431 | if (pskb_pull(skb, ihl) == NULL) |
432 | goto err; | |
1706d587 HX |
433 | |
434 | err = pskb_trim_rcsum(skb, end - offset); | |
435 | if (err) | |
1da177e4 LT |
436 | goto err; |
437 | ||
438 | /* Find out which fragments are in front and at the back of us | |
439 | * in the chain of fragments so far. We must know where to put | |
440 | * this fragment, right? | |
441 | */ | |
442 | prev = NULL; | |
5ab11c98 | 443 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
444 | if (FRAG_CB(next)->offset >= offset) |
445 | break; /* bingo! */ | |
446 | prev = next; | |
447 | } | |
448 | ||
449 | /* We found where to put this one. Check for overlap with | |
450 | * preceding fragment, and, if needed, align things so that | |
451 | * any overlaps are eliminated. | |
452 | */ | |
453 | if (prev) { | |
454 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
455 | ||
456 | if (i > 0) { | |
457 | offset += i; | |
1706d587 | 458 | err = -EINVAL; |
1da177e4 LT |
459 | if (end <= offset) |
460 | goto err; | |
1706d587 | 461 | err = -ENOMEM; |
1da177e4 LT |
462 | if (!pskb_pull(skb, i)) |
463 | goto err; | |
464 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
465 | skb->ip_summed = CHECKSUM_NONE; | |
466 | } | |
467 | } | |
468 | ||
1706d587 HX |
469 | err = -ENOMEM; |
470 | ||
1da177e4 LT |
471 | while (next && FRAG_CB(next)->offset < end) { |
472 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
473 | ||
474 | if (i < next->len) { | |
475 | /* Eat head of the next overlapped fragment | |
476 | * and leave the loop. The next ones cannot overlap. | |
477 | */ | |
478 | if (!pskb_pull(next, i)) | |
479 | goto err; | |
480 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 481 | qp->q.meat -= i; |
1da177e4 LT |
482 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
483 | next->ip_summed = CHECKSUM_NONE; | |
484 | break; | |
485 | } else { | |
486 | struct sk_buff *free_it = next; | |
487 | ||
47c6bf77 | 488 | /* Old fragment is completely overridden with |
1da177e4 LT |
489 | * new one drop it. |
490 | */ | |
491 | next = next->next; | |
492 | ||
493 | if (prev) | |
494 | prev->next = next; | |
495 | else | |
5ab11c98 | 496 | qp->q.fragments = next; |
1da177e4 | 497 | |
5ab11c98 | 498 | qp->q.meat -= free_it->len; |
1da177e4 LT |
499 | frag_kfree_skb(free_it, NULL); |
500 | } | |
501 | } | |
502 | ||
503 | FRAG_CB(skb)->offset = offset; | |
504 | ||
505 | /* Insert this fragment in the chain of fragments. */ | |
506 | skb->next = next; | |
507 | if (prev) | |
508 | prev->next = skb; | |
509 | else | |
5ab11c98 | 510 | qp->q.fragments = skb; |
1da177e4 | 511 | |
1706d587 HX |
512 | dev = skb->dev; |
513 | if (dev) { | |
514 | qp->iif = dev->ifindex; | |
515 | skb->dev = NULL; | |
516 | } | |
5ab11c98 PE |
517 | qp->q.stamp = skb->tstamp; |
518 | qp->q.meat += skb->len; | |
7eb95156 | 519 | atomic_add(skb->truesize, &ip4_frags.mem); |
1da177e4 | 520 | if (offset == 0) |
5ab11c98 | 521 | qp->q.last_in |= FIRST_IN; |
1da177e4 | 522 | |
5ab11c98 | 523 | if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len) |
1706d587 HX |
524 | return ip_frag_reasm(qp, prev, dev); |
525 | ||
7eb95156 PE |
526 | write_lock(&ip4_frags.lock); |
527 | list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list); | |
528 | write_unlock(&ip4_frags.lock); | |
1706d587 | 529 | return -EINPROGRESS; |
1da177e4 LT |
530 | |
531 | err: | |
532 | kfree_skb(skb); | |
1706d587 | 533 | return err; |
1da177e4 LT |
534 | } |
535 | ||
536 | ||
537 | /* Build a new IP datagram from all its fragments. */ | |
538 | ||
1706d587 HX |
539 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
540 | struct net_device *dev) | |
1da177e4 LT |
541 | { |
542 | struct iphdr *iph; | |
5ab11c98 | 543 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
544 | int len; |
545 | int ihlen; | |
1706d587 | 546 | int err; |
1da177e4 LT |
547 | |
548 | ipq_kill(qp); | |
549 | ||
1706d587 HX |
550 | /* Make the one we just received the head. */ |
551 | if (prev) { | |
552 | head = prev->next; | |
553 | fp = skb_clone(head, GFP_ATOMIC); | |
554 | ||
555 | if (!fp) | |
556 | goto out_nomem; | |
557 | ||
558 | fp->next = head->next; | |
559 | prev->next = fp; | |
560 | ||
5ab11c98 PE |
561 | skb_morph(head, qp->q.fragments); |
562 | head->next = qp->q.fragments->next; | |
1706d587 | 563 | |
5ab11c98 PE |
564 | kfree_skb(qp->q.fragments); |
565 | qp->q.fragments = head; | |
1706d587 HX |
566 | } |
567 | ||
1da177e4 LT |
568 | BUG_TRAP(head != NULL); |
569 | BUG_TRAP(FRAG_CB(head)->offset == 0); | |
570 | ||
571 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 572 | ihlen = ip_hdrlen(head); |
5ab11c98 | 573 | len = ihlen + qp->q.len; |
1da177e4 | 574 | |
1706d587 | 575 | err = -E2BIG; |
132adf54 | 576 | if (len > 65535) |
1da177e4 LT |
577 | goto out_oversize; |
578 | ||
579 | /* Head of list must not be cloned. */ | |
1706d587 | 580 | err = -ENOMEM; |
1da177e4 LT |
581 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) |
582 | goto out_nomem; | |
583 | ||
584 | /* If the first fragment is fragmented itself, we split | |
585 | * it to two chunks: the first with data and paged part | |
586 | * and the second, holding only fragments. */ | |
587 | if (skb_shinfo(head)->frag_list) { | |
588 | struct sk_buff *clone; | |
589 | int i, plen = 0; | |
590 | ||
591 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
592 | goto out_nomem; | |
593 | clone->next = head->next; | |
594 | head->next = clone; | |
595 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
596 | skb_shinfo(head)->frag_list = NULL; | |
597 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
598 | plen += skb_shinfo(head)->frags[i].size; | |
599 | clone->len = clone->data_len = head->data_len - plen; | |
600 | head->data_len -= clone->len; | |
601 | head->len -= clone->len; | |
602 | clone->csum = 0; | |
603 | clone->ip_summed = head->ip_summed; | |
7eb95156 | 604 | atomic_add(clone->truesize, &ip4_frags.mem); |
1da177e4 LT |
605 | } |
606 | ||
607 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 608 | skb_push(head, head->data - skb_network_header(head)); |
7eb95156 | 609 | atomic_sub(head->truesize, &ip4_frags.mem); |
1da177e4 LT |
610 | |
611 | for (fp=head->next; fp; fp = fp->next) { | |
612 | head->data_len += fp->len; | |
613 | head->len += fp->len; | |
614 | if (head->ip_summed != fp->ip_summed) | |
615 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 616 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
617 | head->csum = csum_add(head->csum, fp->csum); |
618 | head->truesize += fp->truesize; | |
7eb95156 | 619 | atomic_sub(fp->truesize, &ip4_frags.mem); |
1da177e4 LT |
620 | } |
621 | ||
622 | head->next = NULL; | |
623 | head->dev = dev; | |
5ab11c98 | 624 | head->tstamp = qp->q.stamp; |
1da177e4 | 625 | |
eddc9ec5 | 626 | iph = ip_hdr(head); |
1da177e4 LT |
627 | iph->frag_off = 0; |
628 | iph->tot_len = htons(len); | |
629 | IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS); | |
5ab11c98 | 630 | qp->q.fragments = NULL; |
1706d587 | 631 | return 0; |
1da177e4 LT |
632 | |
633 | out_nomem: | |
e905a9ed | 634 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 635 | "queue %p\n", qp); |
1da177e4 LT |
636 | goto out_fail; |
637 | out_oversize: | |
638 | if (net_ratelimit()) | |
639 | printk(KERN_INFO | |
640 | "Oversized IP packet from %d.%d.%d.%d.\n", | |
641 | NIPQUAD(qp->saddr)); | |
642 | out_fail: | |
643 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
1706d587 | 644 | return err; |
1da177e4 LT |
645 | } |
646 | ||
647 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 648 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 649 | { |
1da177e4 | 650 | struct ipq *qp; |
e905a9ed | 651 | |
1da177e4 LT |
652 | IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS); |
653 | ||
654 | /* Start by cleaning up the memory. */ | |
04128f23 | 655 | if (atomic_read(&ip4_frags.mem) > ip4_frags_ctl.high_thresh) |
1da177e4 LT |
656 | ip_evictor(); |
657 | ||
1da177e4 | 658 | /* Lookup (or create) queue header */ |
eddc9ec5 | 659 | if ((qp = ip_find(ip_hdr(skb), user)) != NULL) { |
1706d587 | 660 | int ret; |
1da177e4 | 661 | |
5ab11c98 | 662 | spin_lock(&qp->q.lock); |
1da177e4 | 663 | |
1706d587 | 664 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 665 | |
5ab11c98 | 666 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 667 | ipq_put(qp); |
776c729e | 668 | return ret; |
1da177e4 LT |
669 | } |
670 | ||
671 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); | |
672 | kfree_skb(skb); | |
776c729e | 673 | return -ENOMEM; |
1da177e4 LT |
674 | } |
675 | ||
b7aa0bf7 | 676 | void __init ipfrag_init(void) |
1da177e4 | 677 | { |
04128f23 | 678 | ip4_frags.ctl = &ip4_frags_ctl; |
321a3a99 | 679 | ip4_frags.hashfn = ip4_hashfn; |
1e4b8287 PE |
680 | ip4_frags.destructor = ip4_frag_free; |
681 | ip4_frags.skb_free = NULL; | |
682 | ip4_frags.qsize = sizeof(struct ipq); | |
7eb95156 | 683 | inet_frags_init(&ip4_frags); |
1da177e4 LT |
684 | } |
685 | ||
686 | EXPORT_SYMBOL(ip_defrag); |