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> | |
5a0e3ad6 | 35 | #include <linux/slab.h> |
e9017b55 SW |
36 | #include <net/route.h> |
37 | #include <net/dst.h> | |
1da177e4 LT |
38 | #include <net/sock.h> |
39 | #include <net/ip.h> | |
40 | #include <net/icmp.h> | |
41 | #include <net/checksum.h> | |
89cee8b1 | 42 | #include <net/inetpeer.h> |
5ab11c98 | 43 | #include <net/inet_frag.h> |
1da177e4 LT |
44 | #include <linux/tcp.h> |
45 | #include <linux/udp.h> | |
46 | #include <linux/inet.h> | |
47 | #include <linux/netfilter_ipv4.h> | |
6623e3b2 | 48 | #include <net/inet_ecn.h> |
1da177e4 LT |
49 | |
50 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
51 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
52 | * as well. Or notify me, at least. --ANK | |
53 | */ | |
54 | ||
8d8354d2 | 55 | static int sysctl_ipfrag_max_dist __read_mostly = 64; |
89cee8b1 | 56 | |
1da177e4 LT |
57 | struct ipfrag_skb_cb |
58 | { | |
59 | struct inet_skb_parm h; | |
60 | int offset; | |
61 | }; | |
62 | ||
fd3f8c4c | 63 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb)) |
1da177e4 LT |
64 | |
65 | /* Describe an entry in the "incomplete datagrams" queue. */ | |
66 | struct ipq { | |
5ab11c98 PE |
67 | struct inet_frag_queue q; |
68 | ||
1da177e4 | 69 | u32 user; |
18277770 AV |
70 | __be32 saddr; |
71 | __be32 daddr; | |
72 | __be16 id; | |
1da177e4 | 73 | u8 protocol; |
6623e3b2 | 74 | u8 ecn; /* RFC3168 support */ |
89cee8b1 HX |
75 | int iif; |
76 | unsigned int rid; | |
77 | struct inet_peer *peer; | |
1da177e4 LT |
78 | }; |
79 | ||
6623e3b2 ED |
80 | #define IPFRAG_ECN_CLEAR 0x01 /* one frag had INET_ECN_NOT_ECT */ |
81 | #define IPFRAG_ECN_SET_CE 0x04 /* one frag had INET_ECN_CE */ | |
82 | ||
83 | static inline u8 ip4_frag_ecn(u8 tos) | |
84 | { | |
85 | tos = (tos & INET_ECN_MASK) + 1; | |
86 | /* | |
87 | * After the last operation we have (in binary): | |
88 | * INET_ECN_NOT_ECT => 001 | |
89 | * INET_ECN_ECT_1 => 010 | |
90 | * INET_ECN_ECT_0 => 011 | |
91 | * INET_ECN_CE => 100 | |
92 | */ | |
93 | return (tos & 2) ? 0 : tos; | |
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); | |
a02cec21 | 137 | return qp->id == arg->iph->id && |
abd6523d PE |
138 | qp->saddr == arg->iph->saddr && |
139 | qp->daddr == arg->iph->daddr && | |
140 | qp->protocol == arg->iph->protocol && | |
a02cec21 | 141 | qp->user == arg->user; |
abd6523d PE |
142 | } |
143 | ||
1da177e4 | 144 | /* Memory Tracking Functions. */ |
a95d8c88 | 145 | static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb) |
1da177e4 | 146 | { |
6ddc0822 | 147 | atomic_sub(skb->truesize, &nf->mem); |
1da177e4 LT |
148 | kfree_skb(skb); |
149 | } | |
150 | ||
c6fda282 PE |
151 | static void ip4_frag_init(struct inet_frag_queue *q, void *a) |
152 | { | |
153 | struct ipq *qp = container_of(q, struct ipq, q); | |
154 | struct ip4_create_arg *arg = a; | |
155 | ||
156 | qp->protocol = arg->iph->protocol; | |
157 | qp->id = arg->iph->id; | |
6623e3b2 | 158 | qp->ecn = ip4_frag_ecn(arg->iph->tos); |
c6fda282 PE |
159 | qp->saddr = arg->iph->saddr; |
160 | qp->daddr = arg->iph->daddr; | |
161 | qp->user = arg->user; | |
162 | qp->peer = sysctl_ipfrag_max_dist ? | |
b534ecf1 | 163 | inet_getpeer_v4(arg->iph->saddr, 1) : NULL; |
c6fda282 PE |
164 | } |
165 | ||
1e4b8287 | 166 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 167 | { |
1e4b8287 PE |
168 | struct ipq *qp; |
169 | ||
170 | qp = container_of(q, struct ipq, q); | |
171 | if (qp->peer) | |
172 | inet_putpeer(qp->peer); | |
1da177e4 LT |
173 | } |
174 | ||
1da177e4 LT |
175 | |
176 | /* Destruction primitives. */ | |
177 | ||
4b6cb5d8 | 178 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 179 | { |
762cc408 | 180 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
181 | } |
182 | ||
183 | /* Kill ipq entry. It is not destroyed immediately, | |
184 | * because caller (and someone more) holds reference count. | |
185 | */ | |
186 | static void ipq_kill(struct ipq *ipq) | |
187 | { | |
277e650d | 188 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
189 | } |
190 | ||
e905a9ed | 191 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
192 | * fragment queue until we are back under the threshold. |
193 | */ | |
6ddc0822 | 194 | static void ip_evictor(struct net *net) |
1da177e4 | 195 | { |
8e7999c4 PE |
196 | int evicted; |
197 | ||
6ddc0822 | 198 | evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags); |
8e7999c4 | 199 | if (evicted) |
c5346fe3 | 200 | IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted); |
1da177e4 LT |
201 | } |
202 | ||
203 | /* | |
204 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
205 | */ | |
206 | static void ip_expire(unsigned long arg) | |
207 | { | |
e521db9d | 208 | struct ipq *qp; |
84a3aa00 | 209 | struct net *net; |
e521db9d PE |
210 | |
211 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 212 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 213 | |
5ab11c98 | 214 | spin_lock(&qp->q.lock); |
1da177e4 | 215 | |
bc578a54 | 216 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
217 | goto out; |
218 | ||
219 | ipq_kill(qp); | |
220 | ||
7c73a6fa PE |
221 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); |
222 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
1da177e4 | 223 | |
bc578a54 | 224 | if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) { |
5ab11c98 | 225 | struct sk_buff *head = qp->q.fragments; |
cb84663e | 226 | |
69df9d59 ED |
227 | rcu_read_lock(); |
228 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
229 | if (!head->dev) |
230 | goto out_rcu_unlock; | |
231 | ||
232 | /* | |
233 | * Only search router table for the head fragment, | |
234 | * when defraging timeout at PRE_ROUTING HOOK. | |
235 | */ | |
236 | if (qp->user == IP_DEFRAG_CONNTRACK_IN && !skb_dst(head)) { | |
237 | const struct iphdr *iph = ip_hdr(head); | |
238 | int err = ip_route_input(head, iph->daddr, iph->saddr, | |
239 | iph->tos, head->dev); | |
240 | if (unlikely(err)) | |
241 | goto out_rcu_unlock; | |
242 | ||
243 | /* | |
244 | * Only an end host needs to send an ICMP | |
245 | * "Fragment Reassembly Timeout" message, per RFC792. | |
246 | */ | |
247 | if (skb_rtable(head)->rt_type != RTN_LOCAL) | |
248 | goto out_rcu_unlock; | |
249 | ||
250 | } | |
251 | ||
252 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ | |
253 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 254 | out_rcu_unlock: |
d1c9ae6d PM |
255 | rcu_read_unlock(); |
256 | } | |
1da177e4 | 257 | out: |
5ab11c98 | 258 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 259 | ipq_put(qp); |
1da177e4 LT |
260 | } |
261 | ||
abd6523d PE |
262 | /* Find the correct entry in the "incomplete datagrams" queue for |
263 | * this IP datagram, and create new one, if nothing is found. | |
264 | */ | |
ac18e750 | 265 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 266 | { |
c6fda282 PE |
267 | struct inet_frag_queue *q; |
268 | struct ip4_create_arg arg; | |
abd6523d | 269 | unsigned int hash; |
1da177e4 | 270 | |
c6fda282 PE |
271 | arg.iph = iph; |
272 | arg.user = user; | |
9a375803 PE |
273 | |
274 | read_lock(&ip4_frags.lock); | |
abd6523d | 275 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 276 | |
ac18e750 | 277 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
c6fda282 PE |
278 | if (q == NULL) |
279 | goto out_nomem; | |
1da177e4 | 280 | |
c6fda282 | 281 | return container_of(q, struct ipq, q); |
1da177e4 LT |
282 | |
283 | out_nomem: | |
64ce2073 | 284 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
285 | return NULL; |
286 | } | |
287 | ||
89cee8b1 HX |
288 | /* Is the fragment too far ahead to be part of ipq? */ |
289 | static inline int ip_frag_too_far(struct ipq *qp) | |
290 | { | |
291 | struct inet_peer *peer = qp->peer; | |
292 | unsigned int max = sysctl_ipfrag_max_dist; | |
293 | unsigned int start, end; | |
294 | ||
295 | int rc; | |
296 | ||
297 | if (!peer || !max) | |
298 | return 0; | |
299 | ||
300 | start = qp->rid; | |
301 | end = atomic_inc_return(&peer->rid); | |
302 | qp->rid = end; | |
303 | ||
5ab11c98 | 304 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
305 | |
306 | if (rc) { | |
7c73a6fa PE |
307 | struct net *net; |
308 | ||
309 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
310 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
311 | } |
312 | ||
313 | return rc; | |
314 | } | |
315 | ||
316 | static int ip_frag_reinit(struct ipq *qp) | |
317 | { | |
318 | struct sk_buff *fp; | |
319 | ||
b2fd5321 | 320 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 321 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
322 | return -ETIMEDOUT; |
323 | } | |
324 | ||
5ab11c98 | 325 | fp = qp->q.fragments; |
89cee8b1 HX |
326 | do { |
327 | struct sk_buff *xp = fp->next; | |
a95d8c88 | 328 | frag_kfree_skb(qp->q.net, fp); |
89cee8b1 HX |
329 | fp = xp; |
330 | } while (fp); | |
331 | ||
5ab11c98 PE |
332 | qp->q.last_in = 0; |
333 | qp->q.len = 0; | |
334 | qp->q.meat = 0; | |
335 | qp->q.fragments = NULL; | |
d6bebca9 | 336 | qp->q.fragments_tail = NULL; |
89cee8b1 | 337 | qp->iif = 0; |
6623e3b2 | 338 | qp->ecn = 0; |
89cee8b1 HX |
339 | |
340 | return 0; | |
341 | } | |
342 | ||
1da177e4 | 343 | /* Add new segment to existing queue. */ |
1706d587 | 344 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
345 | { |
346 | struct sk_buff *prev, *next; | |
1706d587 | 347 | struct net_device *dev; |
1da177e4 LT |
348 | int flags, offset; |
349 | int ihl, end; | |
1706d587 | 350 | int err = -ENOENT; |
6623e3b2 | 351 | u8 ecn; |
1da177e4 | 352 | |
bc578a54 | 353 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
354 | goto err; |
355 | ||
89cee8b1 | 356 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
357 | unlikely(ip_frag_too_far(qp)) && |
358 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
359 | ipq_kill(qp); |
360 | goto err; | |
361 | } | |
362 | ||
6623e3b2 | 363 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 364 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
365 | flags = offset & ~IP_OFFSET; |
366 | offset &= IP_OFFSET; | |
367 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 368 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
369 | |
370 | /* Determine the position of this fragment. */ | |
e905a9ed | 371 | end = offset + skb->len - ihl; |
1706d587 | 372 | err = -EINVAL; |
1da177e4 LT |
373 | |
374 | /* Is this the final fragment? */ | |
375 | if ((flags & IP_MF) == 0) { | |
376 | /* If we already have some bits beyond end | |
377 | * or have different end, the segment is corrrupted. | |
378 | */ | |
5ab11c98 | 379 | if (end < qp->q.len || |
bc578a54 | 380 | ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 381 | goto err; |
bc578a54 | 382 | qp->q.last_in |= INET_FRAG_LAST_IN; |
5ab11c98 | 383 | qp->q.len = end; |
1da177e4 LT |
384 | } else { |
385 | if (end&7) { | |
386 | end &= ~7; | |
387 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
388 | skb->ip_summed = CHECKSUM_NONE; | |
389 | } | |
5ab11c98 | 390 | if (end > qp->q.len) { |
1da177e4 | 391 | /* Some bits beyond end -> corruption. */ |
bc578a54 | 392 | if (qp->q.last_in & INET_FRAG_LAST_IN) |
1da177e4 | 393 | goto err; |
5ab11c98 | 394 | qp->q.len = end; |
1da177e4 LT |
395 | } |
396 | } | |
397 | if (end == offset) | |
398 | goto err; | |
399 | ||
1706d587 | 400 | err = -ENOMEM; |
1da177e4 LT |
401 | if (pskb_pull(skb, ihl) == NULL) |
402 | goto err; | |
1706d587 HX |
403 | |
404 | err = pskb_trim_rcsum(skb, end - offset); | |
405 | if (err) | |
1da177e4 LT |
406 | goto err; |
407 | ||
408 | /* Find out which fragments are in front and at the back of us | |
409 | * in the chain of fragments so far. We must know where to put | |
410 | * this fragment, right? | |
411 | */ | |
d6bebca9 CG |
412 | prev = qp->q.fragments_tail; |
413 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
414 | next = NULL; | |
415 | goto found; | |
416 | } | |
1da177e4 | 417 | prev = NULL; |
5ab11c98 | 418 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
419 | if (FRAG_CB(next)->offset >= offset) |
420 | break; /* bingo! */ | |
421 | prev = next; | |
422 | } | |
423 | ||
d6bebca9 | 424 | found: |
1da177e4 LT |
425 | /* We found where to put this one. Check for overlap with |
426 | * preceding fragment, and, if needed, align things so that | |
427 | * any overlaps are eliminated. | |
428 | */ | |
429 | if (prev) { | |
430 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
431 | ||
432 | if (i > 0) { | |
433 | offset += i; | |
1706d587 | 434 | err = -EINVAL; |
1da177e4 LT |
435 | if (end <= offset) |
436 | goto err; | |
1706d587 | 437 | err = -ENOMEM; |
1da177e4 LT |
438 | if (!pskb_pull(skb, i)) |
439 | goto err; | |
440 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
441 | skb->ip_summed = CHECKSUM_NONE; | |
442 | } | |
443 | } | |
444 | ||
1706d587 HX |
445 | err = -ENOMEM; |
446 | ||
1da177e4 LT |
447 | while (next && FRAG_CB(next)->offset < end) { |
448 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
449 | ||
450 | if (i < next->len) { | |
451 | /* Eat head of the next overlapped fragment | |
452 | * and leave the loop. The next ones cannot overlap. | |
453 | */ | |
454 | if (!pskb_pull(next, i)) | |
455 | goto err; | |
456 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 457 | qp->q.meat -= i; |
1da177e4 LT |
458 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
459 | next->ip_summed = CHECKSUM_NONE; | |
460 | break; | |
461 | } else { | |
462 | struct sk_buff *free_it = next; | |
463 | ||
47c6bf77 | 464 | /* Old fragment is completely overridden with |
1da177e4 LT |
465 | * new one drop it. |
466 | */ | |
467 | next = next->next; | |
468 | ||
469 | if (prev) | |
470 | prev->next = next; | |
471 | else | |
5ab11c98 | 472 | qp->q.fragments = next; |
1da177e4 | 473 | |
5ab11c98 | 474 | qp->q.meat -= free_it->len; |
a95d8c88 | 475 | frag_kfree_skb(qp->q.net, free_it); |
1da177e4 LT |
476 | } |
477 | } | |
478 | ||
479 | FRAG_CB(skb)->offset = offset; | |
480 | ||
481 | /* Insert this fragment in the chain of fragments. */ | |
482 | skb->next = next; | |
d6bebca9 CG |
483 | if (!next) |
484 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
485 | if (prev) |
486 | prev->next = skb; | |
487 | else | |
5ab11c98 | 488 | qp->q.fragments = skb; |
1da177e4 | 489 | |
1706d587 HX |
490 | dev = skb->dev; |
491 | if (dev) { | |
492 | qp->iif = dev->ifindex; | |
493 | skb->dev = NULL; | |
494 | } | |
5ab11c98 PE |
495 | qp->q.stamp = skb->tstamp; |
496 | qp->q.meat += skb->len; | |
6623e3b2 | 497 | qp->ecn |= ecn; |
6ddc0822 | 498 | atomic_add(skb->truesize, &qp->q.net->mem); |
1da177e4 | 499 | if (offset == 0) |
bc578a54 | 500 | qp->q.last_in |= INET_FRAG_FIRST_IN; |
1da177e4 | 501 | |
bc578a54 JP |
502 | if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
503 | qp->q.meat == qp->q.len) | |
1706d587 HX |
504 | return ip_frag_reasm(qp, prev, dev); |
505 | ||
7eb95156 | 506 | write_lock(&ip4_frags.lock); |
3140c25c | 507 | list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list); |
7eb95156 | 508 | write_unlock(&ip4_frags.lock); |
1706d587 | 509 | return -EINPROGRESS; |
1da177e4 LT |
510 | |
511 | err: | |
512 | kfree_skb(skb); | |
1706d587 | 513 | return err; |
1da177e4 LT |
514 | } |
515 | ||
516 | ||
517 | /* Build a new IP datagram from all its fragments. */ | |
518 | ||
1706d587 HX |
519 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
520 | struct net_device *dev) | |
1da177e4 | 521 | { |
2bad35b7 | 522 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 523 | struct iphdr *iph; |
5ab11c98 | 524 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
525 | int len; |
526 | int ihlen; | |
1706d587 | 527 | int err; |
1da177e4 LT |
528 | |
529 | ipq_kill(qp); | |
530 | ||
1706d587 HX |
531 | /* Make the one we just received the head. */ |
532 | if (prev) { | |
533 | head = prev->next; | |
534 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
535 | if (!fp) |
536 | goto out_nomem; | |
537 | ||
538 | fp->next = head->next; | |
d6bebca9 CG |
539 | if (!fp->next) |
540 | qp->q.fragments_tail = fp; | |
1706d587 HX |
541 | prev->next = fp; |
542 | ||
5ab11c98 PE |
543 | skb_morph(head, qp->q.fragments); |
544 | head->next = qp->q.fragments->next; | |
1706d587 | 545 | |
5ab11c98 PE |
546 | kfree_skb(qp->q.fragments); |
547 | qp->q.fragments = head; | |
1706d587 HX |
548 | } |
549 | ||
547b792c IJ |
550 | WARN_ON(head == NULL); |
551 | WARN_ON(FRAG_CB(head)->offset != 0); | |
1da177e4 LT |
552 | |
553 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 554 | ihlen = ip_hdrlen(head); |
5ab11c98 | 555 | len = ihlen + qp->q.len; |
1da177e4 | 556 | |
1706d587 | 557 | err = -E2BIG; |
132adf54 | 558 | if (len > 65535) |
1da177e4 LT |
559 | goto out_oversize; |
560 | ||
561 | /* Head of list must not be cloned. */ | |
562 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
563 | goto out_nomem; | |
564 | ||
565 | /* If the first fragment is fragmented itself, we split | |
566 | * it to two chunks: the first with data and paged part | |
567 | * and the second, holding only fragments. */ | |
21dc3301 | 568 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
569 | struct sk_buff *clone; |
570 | int i, plen = 0; | |
571 | ||
572 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
573 | goto out_nomem; | |
574 | clone->next = head->next; | |
575 | head->next = clone; | |
576 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 577 | skb_frag_list_init(head); |
1da177e4 LT |
578 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) |
579 | plen += skb_shinfo(head)->frags[i].size; | |
580 | clone->len = clone->data_len = head->data_len - plen; | |
581 | head->data_len -= clone->len; | |
582 | head->len -= clone->len; | |
583 | clone->csum = 0; | |
584 | clone->ip_summed = head->ip_summed; | |
6ddc0822 | 585 | atomic_add(clone->truesize, &qp->q.net->mem); |
1da177e4 LT |
586 | } |
587 | ||
588 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 589 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 LT |
590 | |
591 | for (fp=head->next; fp; fp = fp->next) { | |
592 | head->data_len += fp->len; | |
593 | head->len += fp->len; | |
594 | if (head->ip_summed != fp->ip_summed) | |
595 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 596 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
597 | head->csum = csum_add(head->csum, fp->csum); |
598 | head->truesize += fp->truesize; | |
1da177e4 | 599 | } |
d27f9b35 | 600 | atomic_sub(head->truesize, &qp->q.net->mem); |
1da177e4 LT |
601 | |
602 | head->next = NULL; | |
603 | head->dev = dev; | |
5ab11c98 | 604 | head->tstamp = qp->q.stamp; |
1da177e4 | 605 | |
eddc9ec5 | 606 | iph = ip_hdr(head); |
1da177e4 LT |
607 | iph->frag_off = 0; |
608 | iph->tot_len = htons(len); | |
6623e3b2 ED |
609 | /* RFC3168 5.3 Fragmentation support |
610 | * If one fragment had INET_ECN_NOT_ECT, | |
611 | * reassembled frame also has INET_ECN_NOT_ECT | |
612 | * Elif one fragment had INET_ECN_CE | |
613 | * reassembled frame also has INET_ECN_CE | |
614 | */ | |
615 | if (qp->ecn & IPFRAG_ECN_CLEAR) | |
616 | iph->tos &= ~INET_ECN_MASK; | |
617 | else if (qp->ecn & IPFRAG_ECN_SET_CE) | |
618 | iph->tos |= INET_ECN_CE; | |
619 | ||
2bad35b7 | 620 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 621 | qp->q.fragments = NULL; |
d6bebca9 | 622 | qp->q.fragments_tail = NULL; |
1706d587 | 623 | return 0; |
1da177e4 LT |
624 | |
625 | out_nomem: | |
e905a9ed | 626 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 627 | "queue %p\n", qp); |
45542479 | 628 | err = -ENOMEM; |
1da177e4 LT |
629 | goto out_fail; |
630 | out_oversize: | |
631 | if (net_ratelimit()) | |
673d57e7 HH |
632 | printk(KERN_INFO "Oversized IP packet from %pI4.\n", |
633 | &qp->saddr); | |
1da177e4 | 634 | out_fail: |
bbf31bf1 | 635 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 636 | return err; |
1da177e4 LT |
637 | } |
638 | ||
639 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 640 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 641 | { |
1da177e4 | 642 | struct ipq *qp; |
ac18e750 | 643 | struct net *net; |
e905a9ed | 644 | |
adf30907 | 645 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); |
7c73a6fa | 646 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
647 | |
648 | /* Start by cleaning up the memory. */ | |
e31e0bdc | 649 | if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh) |
6ddc0822 | 650 | ip_evictor(net); |
1da177e4 | 651 | |
1da177e4 | 652 | /* Lookup (or create) queue header */ |
ac18e750 | 653 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 654 | int ret; |
1da177e4 | 655 | |
5ab11c98 | 656 | spin_lock(&qp->q.lock); |
1da177e4 | 657 | |
1706d587 | 658 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 659 | |
5ab11c98 | 660 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 661 | ipq_put(qp); |
776c729e | 662 | return ret; |
1da177e4 LT |
663 | } |
664 | ||
7c73a6fa | 665 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 666 | kfree_skb(skb); |
776c729e | 667 | return -ENOMEM; |
1da177e4 | 668 | } |
4bc2f18b | 669 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 670 | |
8d8354d2 PE |
671 | #ifdef CONFIG_SYSCTL |
672 | static int zero; | |
673 | ||
0a64b4b8 | 674 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 675 | { |
8d8354d2 | 676 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 677 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
678 | .maxlen = sizeof(int), |
679 | .mode = 0644, | |
6d9f239a | 680 | .proc_handler = proc_dointvec |
8d8354d2 PE |
681 | }, |
682 | { | |
8d8354d2 | 683 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 684 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
685 | .maxlen = sizeof(int), |
686 | .mode = 0644, | |
6d9f239a | 687 | .proc_handler = proc_dointvec |
8d8354d2 PE |
688 | }, |
689 | { | |
8d8354d2 | 690 | .procname = "ipfrag_time", |
b2fd5321 | 691 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
692 | .maxlen = sizeof(int), |
693 | .mode = 0644, | |
6d9f239a | 694 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 695 | }, |
7d291ebb PE |
696 | { } |
697 | }; | |
698 | ||
699 | static struct ctl_table ip4_frags_ctl_table[] = { | |
8d8354d2 | 700 | { |
8d8354d2 | 701 | .procname = "ipfrag_secret_interval", |
3b4bc4a2 | 702 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
703 | .maxlen = sizeof(int), |
704 | .mode = 0644, | |
6d9f239a | 705 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 PE |
706 | }, |
707 | { | |
708 | .procname = "ipfrag_max_dist", | |
709 | .data = &sysctl_ipfrag_max_dist, | |
710 | .maxlen = sizeof(int), | |
711 | .mode = 0644, | |
6d9f239a | 712 | .proc_handler = proc_dointvec_minmax, |
8d8354d2 PE |
713 | .extra1 = &zero |
714 | }, | |
715 | { } | |
716 | }; | |
717 | ||
2c8c1e72 | 718 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 719 | { |
e4a2d5c2 | 720 | struct ctl_table *table; |
8d8354d2 PE |
721 | struct ctl_table_header *hdr; |
722 | ||
0a64b4b8 | 723 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 724 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 725 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
e4a2d5c2 PE |
726 | if (table == NULL) |
727 | goto err_alloc; | |
728 | ||
e31e0bdc PE |
729 | table[0].data = &net->ipv4.frags.high_thresh; |
730 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 731 | table[2].data = &net->ipv4.frags.timeout; |
e4a2d5c2 PE |
732 | } |
733 | ||
734 | hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table); | |
735 | if (hdr == NULL) | |
736 | goto err_reg; | |
737 | ||
738 | net->ipv4.frags_hdr = hdr; | |
739 | return 0; | |
740 | ||
741 | err_reg: | |
09ad9bc7 | 742 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
743 | kfree(table); |
744 | err_alloc: | |
745 | return -ENOMEM; | |
746 | } | |
747 | ||
2c8c1e72 | 748 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
749 | { |
750 | struct ctl_table *table; | |
751 | ||
752 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
753 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
754 | kfree(table); | |
8d8354d2 | 755 | } |
7d291ebb PE |
756 | |
757 | static void ip4_frags_ctl_register(void) | |
758 | { | |
759 | register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table); | |
760 | } | |
8d8354d2 | 761 | #else |
0a64b4b8 | 762 | static inline int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
763 | { |
764 | return 0; | |
765 | } | |
e4a2d5c2 | 766 | |
0a64b4b8 | 767 | static inline void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
768 | { |
769 | } | |
7d291ebb PE |
770 | |
771 | static inline void ip4_frags_ctl_register(void) | |
772 | { | |
773 | } | |
8d8354d2 PE |
774 | #endif |
775 | ||
2c8c1e72 | 776 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 777 | { |
e31e0bdc PE |
778 | /* |
779 | * Fragment cache limits. We will commit 256K at one time. Should we | |
780 | * cross that limit we will prune down to 192K. This should cope with | |
781 | * even the most extreme cases without allowing an attacker to | |
782 | * measurably harm machine performance. | |
783 | */ | |
784 | net->ipv4.frags.high_thresh = 256 * 1024; | |
785 | net->ipv4.frags.low_thresh = 192 * 1024; | |
b2fd5321 PE |
786 | /* |
787 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
788 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
789 | * by TTL. | |
790 | */ | |
791 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
792 | ||
e5a2bb84 PE |
793 | inet_frags_init_net(&net->ipv4.frags); |
794 | ||
0a64b4b8 | 795 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
796 | } |
797 | ||
2c8c1e72 | 798 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 799 | { |
0a64b4b8 | 800 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
801 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
802 | } | |
803 | ||
804 | static struct pernet_operations ip4_frags_ops = { | |
805 | .init = ipv4_frags_init_net, | |
806 | .exit = ipv4_frags_exit_net, | |
807 | }; | |
808 | ||
b7aa0bf7 | 809 | void __init ipfrag_init(void) |
1da177e4 | 810 | { |
7d291ebb | 811 | ip4_frags_ctl_register(); |
81566e83 | 812 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 813 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 814 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
815 | ip4_frags.destructor = ip4_frag_free; |
816 | ip4_frags.skb_free = NULL; | |
817 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 818 | ip4_frags.match = ip4_frag_match; |
e521db9d | 819 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 820 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 821 | inet_frags_init(&ip4_frags); |
1da177e4 | 822 | } |