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 | ||
5173cc05 ED |
80 | /* RFC 3168 support : |
81 | * We want to check ECN values of all fragments, do detect invalid combinations. | |
82 | * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value. | |
83 | */ | |
1d1652cb DM |
84 | #define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */ |
85 | #define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */ | |
86 | #define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */ | |
87 | #define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */ | |
6623e3b2 ED |
88 | |
89 | static inline u8 ip4_frag_ecn(u8 tos) | |
90 | { | |
5173cc05 | 91 | return 1 << (tos & INET_ECN_MASK); |
6623e3b2 ED |
92 | } |
93 | ||
5173cc05 ED |
94 | /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements |
95 | * Value : 0xff if frame should be dropped. | |
96 | * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field | |
97 | */ | |
98 | static const u8 ip4_frag_ecn_table[16] = { | |
99 | /* at least one fragment had CE, and others ECT_0 or ECT_1 */ | |
100 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE, | |
101 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
102 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
103 | ||
104 | /* invalid combinations : drop frame */ | |
105 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff, | |
106 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff, | |
107 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff, | |
108 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
109 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff, | |
110 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff, | |
111 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
112 | }; | |
113 | ||
7eb95156 | 114 | static struct inet_frags ip4_frags; |
1da177e4 | 115 | |
e5a2bb84 | 116 | int ip_frag_nqueues(struct net *net) |
7eb95156 | 117 | { |
e5a2bb84 | 118 | return net->ipv4.frags.nqueues; |
7eb95156 | 119 | } |
1da177e4 | 120 | |
6ddc0822 | 121 | int ip_frag_mem(struct net *net) |
7eb95156 | 122 | { |
6ddc0822 | 123 | return atomic_read(&net->ipv4.frags.mem); |
7eb95156 | 124 | } |
1da177e4 | 125 | |
1706d587 HX |
126 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
127 | struct net_device *dev); | |
128 | ||
c6fda282 PE |
129 | struct ip4_create_arg { |
130 | struct iphdr *iph; | |
131 | u32 user; | |
132 | }; | |
133 | ||
18277770 | 134 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 135 | { |
18277770 AV |
136 | return jhash_3words((__force u32)id << 16 | prot, |
137 | (__force u32)saddr, (__force u32)daddr, | |
7eb95156 | 138 | ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
139 | } |
140 | ||
321a3a99 | 141 | static unsigned int ip4_hashfn(struct inet_frag_queue *q) |
1da177e4 | 142 | { |
321a3a99 | 143 | struct ipq *ipq; |
1da177e4 | 144 | |
321a3a99 PE |
145 | ipq = container_of(q, struct ipq, q); |
146 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
147 | } |
148 | ||
abd6523d PE |
149 | static int ip4_frag_match(struct inet_frag_queue *q, void *a) |
150 | { | |
151 | struct ipq *qp; | |
152 | struct ip4_create_arg *arg = a; | |
153 | ||
154 | qp = container_of(q, struct ipq, q); | |
a02cec21 | 155 | return qp->id == arg->iph->id && |
abd6523d PE |
156 | qp->saddr == arg->iph->saddr && |
157 | qp->daddr == arg->iph->daddr && | |
158 | qp->protocol == arg->iph->protocol && | |
a02cec21 | 159 | qp->user == arg->user; |
abd6523d PE |
160 | } |
161 | ||
1da177e4 | 162 | /* Memory Tracking Functions. */ |
a95d8c88 | 163 | static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb) |
1da177e4 | 164 | { |
6ddc0822 | 165 | atomic_sub(skb->truesize, &nf->mem); |
1da177e4 LT |
166 | kfree_skb(skb); |
167 | } | |
168 | ||
c6fda282 PE |
169 | static void ip4_frag_init(struct inet_frag_queue *q, void *a) |
170 | { | |
171 | struct ipq *qp = container_of(q, struct ipq, q); | |
172 | struct ip4_create_arg *arg = a; | |
173 | ||
174 | qp->protocol = arg->iph->protocol; | |
175 | qp->id = arg->iph->id; | |
6623e3b2 | 176 | qp->ecn = ip4_frag_ecn(arg->iph->tos); |
c6fda282 PE |
177 | qp->saddr = arg->iph->saddr; |
178 | qp->daddr = arg->iph->daddr; | |
179 | qp->user = arg->user; | |
180 | qp->peer = sysctl_ipfrag_max_dist ? | |
b534ecf1 | 181 | inet_getpeer_v4(arg->iph->saddr, 1) : NULL; |
c6fda282 PE |
182 | } |
183 | ||
1e4b8287 | 184 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 185 | { |
1e4b8287 PE |
186 | struct ipq *qp; |
187 | ||
188 | qp = container_of(q, struct ipq, q); | |
189 | if (qp->peer) | |
190 | inet_putpeer(qp->peer); | |
1da177e4 LT |
191 | } |
192 | ||
1da177e4 LT |
193 | |
194 | /* Destruction primitives. */ | |
195 | ||
4b6cb5d8 | 196 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 197 | { |
762cc408 | 198 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
199 | } |
200 | ||
201 | /* Kill ipq entry. It is not destroyed immediately, | |
202 | * because caller (and someone more) holds reference count. | |
203 | */ | |
204 | static void ipq_kill(struct ipq *ipq) | |
205 | { | |
277e650d | 206 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
207 | } |
208 | ||
e905a9ed | 209 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
210 | * fragment queue until we are back under the threshold. |
211 | */ | |
6ddc0822 | 212 | static void ip_evictor(struct net *net) |
1da177e4 | 213 | { |
8e7999c4 PE |
214 | int evicted; |
215 | ||
6ddc0822 | 216 | evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags); |
8e7999c4 | 217 | if (evicted) |
c5346fe3 | 218 | IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted); |
1da177e4 LT |
219 | } |
220 | ||
221 | /* | |
222 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
223 | */ | |
224 | static void ip_expire(unsigned long arg) | |
225 | { | |
e521db9d | 226 | struct ipq *qp; |
84a3aa00 | 227 | struct net *net; |
e521db9d PE |
228 | |
229 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 230 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 231 | |
5ab11c98 | 232 | spin_lock(&qp->q.lock); |
1da177e4 | 233 | |
bc578a54 | 234 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
235 | goto out; |
236 | ||
237 | ipq_kill(qp); | |
238 | ||
7c73a6fa PE |
239 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); |
240 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
1da177e4 | 241 | |
bc578a54 | 242 | if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) { |
5ab11c98 | 243 | struct sk_buff *head = qp->q.fragments; |
64f3b9e2 ED |
244 | const struct iphdr *iph; |
245 | int err; | |
cb84663e | 246 | |
69df9d59 ED |
247 | rcu_read_lock(); |
248 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
249 | if (!head->dev) |
250 | goto out_rcu_unlock; | |
251 | ||
64f3b9e2 ED |
252 | /* skb dst is stale, drop it, and perform route lookup again */ |
253 | skb_dst_drop(head); | |
254 | iph = ip_hdr(head); | |
255 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, | |
256 | iph->tos, head->dev); | |
257 | if (err) | |
258 | goto out_rcu_unlock; | |
259 | ||
e9017b55 | 260 | /* |
64f3b9e2 ED |
261 | * Only an end host needs to send an ICMP |
262 | * "Fragment Reassembly Timeout" message, per RFC792. | |
e9017b55 | 263 | */ |
595fc71b DM |
264 | if (qp->user == IP_DEFRAG_AF_PACKET || |
265 | (qp->user == IP_DEFRAG_CONNTRACK_IN && | |
266 | skb_rtable(head)->rt_type != RTN_LOCAL)) | |
64f3b9e2 ED |
267 | goto out_rcu_unlock; |
268 | ||
e9017b55 SW |
269 | |
270 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ | |
271 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 272 | out_rcu_unlock: |
d1c9ae6d PM |
273 | rcu_read_unlock(); |
274 | } | |
1da177e4 | 275 | out: |
5ab11c98 | 276 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 277 | ipq_put(qp); |
1da177e4 LT |
278 | } |
279 | ||
abd6523d PE |
280 | /* Find the correct entry in the "incomplete datagrams" queue for |
281 | * this IP datagram, and create new one, if nothing is found. | |
282 | */ | |
ac18e750 | 283 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 284 | { |
c6fda282 PE |
285 | struct inet_frag_queue *q; |
286 | struct ip4_create_arg arg; | |
abd6523d | 287 | unsigned int hash; |
1da177e4 | 288 | |
c6fda282 PE |
289 | arg.iph = iph; |
290 | arg.user = user; | |
9a375803 PE |
291 | |
292 | read_lock(&ip4_frags.lock); | |
abd6523d | 293 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 294 | |
ac18e750 | 295 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
c6fda282 PE |
296 | if (q == NULL) |
297 | goto out_nomem; | |
1da177e4 | 298 | |
c6fda282 | 299 | return container_of(q, struct ipq, q); |
1da177e4 LT |
300 | |
301 | out_nomem: | |
64ce2073 | 302 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
1da177e4 LT |
303 | return NULL; |
304 | } | |
305 | ||
89cee8b1 HX |
306 | /* Is the fragment too far ahead to be part of ipq? */ |
307 | static inline int ip_frag_too_far(struct ipq *qp) | |
308 | { | |
309 | struct inet_peer *peer = qp->peer; | |
310 | unsigned int max = sysctl_ipfrag_max_dist; | |
311 | unsigned int start, end; | |
312 | ||
313 | int rc; | |
314 | ||
315 | if (!peer || !max) | |
316 | return 0; | |
317 | ||
318 | start = qp->rid; | |
319 | end = atomic_inc_return(&peer->rid); | |
320 | qp->rid = end; | |
321 | ||
5ab11c98 | 322 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
323 | |
324 | if (rc) { | |
7c73a6fa PE |
325 | struct net *net; |
326 | ||
327 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
328 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
329 | } |
330 | ||
331 | return rc; | |
332 | } | |
333 | ||
334 | static int ip_frag_reinit(struct ipq *qp) | |
335 | { | |
336 | struct sk_buff *fp; | |
337 | ||
b2fd5321 | 338 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 339 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
340 | return -ETIMEDOUT; |
341 | } | |
342 | ||
5ab11c98 | 343 | fp = qp->q.fragments; |
89cee8b1 HX |
344 | do { |
345 | struct sk_buff *xp = fp->next; | |
a95d8c88 | 346 | frag_kfree_skb(qp->q.net, fp); |
89cee8b1 HX |
347 | fp = xp; |
348 | } while (fp); | |
349 | ||
5ab11c98 PE |
350 | qp->q.last_in = 0; |
351 | qp->q.len = 0; | |
352 | qp->q.meat = 0; | |
353 | qp->q.fragments = NULL; | |
d6bebca9 | 354 | qp->q.fragments_tail = NULL; |
89cee8b1 | 355 | qp->iif = 0; |
6623e3b2 | 356 | qp->ecn = 0; |
89cee8b1 HX |
357 | |
358 | return 0; | |
359 | } | |
360 | ||
1da177e4 | 361 | /* Add new segment to existing queue. */ |
1706d587 | 362 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
363 | { |
364 | struct sk_buff *prev, *next; | |
1706d587 | 365 | struct net_device *dev; |
1da177e4 LT |
366 | int flags, offset; |
367 | int ihl, end; | |
1706d587 | 368 | int err = -ENOENT; |
6623e3b2 | 369 | u8 ecn; |
1da177e4 | 370 | |
bc578a54 | 371 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
372 | goto err; |
373 | ||
89cee8b1 | 374 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
375 | unlikely(ip_frag_too_far(qp)) && |
376 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
377 | ipq_kill(qp); |
378 | goto err; | |
379 | } | |
380 | ||
6623e3b2 | 381 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 382 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
383 | flags = offset & ~IP_OFFSET; |
384 | offset &= IP_OFFSET; | |
385 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 386 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
387 | |
388 | /* Determine the position of this fragment. */ | |
e905a9ed | 389 | end = offset + skb->len - ihl; |
1706d587 | 390 | err = -EINVAL; |
1da177e4 LT |
391 | |
392 | /* Is this the final fragment? */ | |
393 | if ((flags & IP_MF) == 0) { | |
394 | /* If we already have some bits beyond end | |
395 | * or have different end, the segment is corrrupted. | |
396 | */ | |
5ab11c98 | 397 | if (end < qp->q.len || |
bc578a54 | 398 | ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 399 | goto err; |
bc578a54 | 400 | qp->q.last_in |= INET_FRAG_LAST_IN; |
5ab11c98 | 401 | qp->q.len = end; |
1da177e4 LT |
402 | } else { |
403 | if (end&7) { | |
404 | end &= ~7; | |
405 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
406 | skb->ip_summed = CHECKSUM_NONE; | |
407 | } | |
5ab11c98 | 408 | if (end > qp->q.len) { |
1da177e4 | 409 | /* Some bits beyond end -> corruption. */ |
bc578a54 | 410 | if (qp->q.last_in & INET_FRAG_LAST_IN) |
1da177e4 | 411 | goto err; |
5ab11c98 | 412 | qp->q.len = end; |
1da177e4 LT |
413 | } |
414 | } | |
415 | if (end == offset) | |
416 | goto err; | |
417 | ||
1706d587 | 418 | err = -ENOMEM; |
1da177e4 LT |
419 | if (pskb_pull(skb, ihl) == NULL) |
420 | goto err; | |
1706d587 HX |
421 | |
422 | err = pskb_trim_rcsum(skb, end - offset); | |
423 | if (err) | |
1da177e4 LT |
424 | goto err; |
425 | ||
426 | /* Find out which fragments are in front and at the back of us | |
427 | * in the chain of fragments so far. We must know where to put | |
428 | * this fragment, right? | |
429 | */ | |
d6bebca9 CG |
430 | prev = qp->q.fragments_tail; |
431 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
432 | next = NULL; | |
433 | goto found; | |
434 | } | |
1da177e4 | 435 | prev = NULL; |
5ab11c98 | 436 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
437 | if (FRAG_CB(next)->offset >= offset) |
438 | break; /* bingo! */ | |
439 | prev = next; | |
440 | } | |
441 | ||
d6bebca9 | 442 | found: |
1da177e4 LT |
443 | /* We found where to put this one. Check for overlap with |
444 | * preceding fragment, and, if needed, align things so that | |
445 | * any overlaps are eliminated. | |
446 | */ | |
447 | if (prev) { | |
448 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
449 | ||
450 | if (i > 0) { | |
451 | offset += i; | |
1706d587 | 452 | err = -EINVAL; |
1da177e4 LT |
453 | if (end <= offset) |
454 | goto err; | |
1706d587 | 455 | err = -ENOMEM; |
1da177e4 LT |
456 | if (!pskb_pull(skb, i)) |
457 | goto err; | |
458 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
459 | skb->ip_summed = CHECKSUM_NONE; | |
460 | } | |
461 | } | |
462 | ||
1706d587 HX |
463 | err = -ENOMEM; |
464 | ||
1da177e4 LT |
465 | while (next && FRAG_CB(next)->offset < end) { |
466 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
467 | ||
468 | if (i < next->len) { | |
469 | /* Eat head of the next overlapped fragment | |
470 | * and leave the loop. The next ones cannot overlap. | |
471 | */ | |
472 | if (!pskb_pull(next, i)) | |
473 | goto err; | |
474 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 475 | qp->q.meat -= i; |
1da177e4 LT |
476 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
477 | next->ip_summed = CHECKSUM_NONE; | |
478 | break; | |
479 | } else { | |
480 | struct sk_buff *free_it = next; | |
481 | ||
47c6bf77 | 482 | /* Old fragment is completely overridden with |
1da177e4 LT |
483 | * new one drop it. |
484 | */ | |
485 | next = next->next; | |
486 | ||
487 | if (prev) | |
488 | prev->next = next; | |
489 | else | |
5ab11c98 | 490 | qp->q.fragments = next; |
1da177e4 | 491 | |
5ab11c98 | 492 | qp->q.meat -= free_it->len; |
a95d8c88 | 493 | frag_kfree_skb(qp->q.net, free_it); |
1da177e4 LT |
494 | } |
495 | } | |
496 | ||
497 | FRAG_CB(skb)->offset = offset; | |
498 | ||
499 | /* Insert this fragment in the chain of fragments. */ | |
500 | skb->next = next; | |
d6bebca9 CG |
501 | if (!next) |
502 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
503 | if (prev) |
504 | prev->next = skb; | |
505 | else | |
5ab11c98 | 506 | qp->q.fragments = skb; |
1da177e4 | 507 | |
1706d587 HX |
508 | dev = skb->dev; |
509 | if (dev) { | |
510 | qp->iif = dev->ifindex; | |
511 | skb->dev = NULL; | |
512 | } | |
5ab11c98 PE |
513 | qp->q.stamp = skb->tstamp; |
514 | qp->q.meat += skb->len; | |
6623e3b2 | 515 | qp->ecn |= ecn; |
6ddc0822 | 516 | atomic_add(skb->truesize, &qp->q.net->mem); |
1da177e4 | 517 | if (offset == 0) |
bc578a54 | 518 | qp->q.last_in |= INET_FRAG_FIRST_IN; |
1da177e4 | 519 | |
bc578a54 JP |
520 | if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
521 | qp->q.meat == qp->q.len) | |
1706d587 HX |
522 | return ip_frag_reasm(qp, prev, dev); |
523 | ||
7eb95156 | 524 | write_lock(&ip4_frags.lock); |
3140c25c | 525 | list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list); |
7eb95156 | 526 | write_unlock(&ip4_frags.lock); |
1706d587 | 527 | return -EINPROGRESS; |
1da177e4 LT |
528 | |
529 | err: | |
530 | kfree_skb(skb); | |
1706d587 | 531 | return err; |
1da177e4 LT |
532 | } |
533 | ||
534 | ||
535 | /* Build a new IP datagram from all its fragments. */ | |
536 | ||
1706d587 HX |
537 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
538 | struct net_device *dev) | |
1da177e4 | 539 | { |
2bad35b7 | 540 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 541 | struct iphdr *iph; |
5ab11c98 | 542 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
543 | int len; |
544 | int ihlen; | |
1706d587 | 545 | int err; |
5173cc05 | 546 | u8 ecn; |
1da177e4 LT |
547 | |
548 | ipq_kill(qp); | |
549 | ||
5173cc05 ED |
550 | ecn = ip4_frag_ecn_table[qp->ecn]; |
551 | if (unlikely(ecn == 0xff)) { | |
552 | err = -EINVAL; | |
553 | goto out_fail; | |
554 | } | |
1706d587 HX |
555 | /* Make the one we just received the head. */ |
556 | if (prev) { | |
557 | head = prev->next; | |
558 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
559 | if (!fp) |
560 | goto out_nomem; | |
561 | ||
562 | fp->next = head->next; | |
d6bebca9 CG |
563 | if (!fp->next) |
564 | qp->q.fragments_tail = fp; | |
1706d587 HX |
565 | prev->next = fp; |
566 | ||
5ab11c98 PE |
567 | skb_morph(head, qp->q.fragments); |
568 | head->next = qp->q.fragments->next; | |
1706d587 | 569 | |
5ab11c98 PE |
570 | kfree_skb(qp->q.fragments); |
571 | qp->q.fragments = head; | |
1706d587 HX |
572 | } |
573 | ||
547b792c IJ |
574 | WARN_ON(head == NULL); |
575 | WARN_ON(FRAG_CB(head)->offset != 0); | |
1da177e4 LT |
576 | |
577 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 578 | ihlen = ip_hdrlen(head); |
5ab11c98 | 579 | len = ihlen + qp->q.len; |
1da177e4 | 580 | |
1706d587 | 581 | err = -E2BIG; |
132adf54 | 582 | if (len > 65535) |
1da177e4 LT |
583 | goto out_oversize; |
584 | ||
585 | /* Head of list must not be cloned. */ | |
586 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
587 | goto out_nomem; | |
588 | ||
589 | /* If the first fragment is fragmented itself, we split | |
590 | * it to two chunks: the first with data and paged part | |
591 | * and the second, holding only fragments. */ | |
21dc3301 | 592 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
593 | struct sk_buff *clone; |
594 | int i, plen = 0; | |
595 | ||
596 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
597 | goto out_nomem; | |
598 | clone->next = head->next; | |
599 | head->next = clone; | |
600 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 601 | skb_frag_list_init(head); |
9e903e08 ED |
602 | for (i = 0; i < skb_shinfo(head)->nr_frags; i++) |
603 | plen += skb_frag_size(&skb_shinfo(head)->frags[i]); | |
1da177e4 LT |
604 | clone->len = clone->data_len = head->data_len - plen; |
605 | head->data_len -= clone->len; | |
606 | head->len -= clone->len; | |
607 | clone->csum = 0; | |
608 | clone->ip_summed = head->ip_summed; | |
6ddc0822 | 609 | atomic_add(clone->truesize, &qp->q.net->mem); |
1da177e4 LT |
610 | } |
611 | ||
612 | skb_shinfo(head)->frag_list = head->next; | |
d56f90a7 | 613 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 LT |
614 | |
615 | for (fp=head->next; fp; fp = fp->next) { | |
616 | head->data_len += fp->len; | |
617 | head->len += fp->len; | |
618 | if (head->ip_summed != fp->ip_summed) | |
619 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 620 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
621 | head->csum = csum_add(head->csum, fp->csum); |
622 | head->truesize += fp->truesize; | |
1da177e4 | 623 | } |
d27f9b35 | 624 | atomic_sub(head->truesize, &qp->q.net->mem); |
1da177e4 LT |
625 | |
626 | head->next = NULL; | |
627 | head->dev = dev; | |
5ab11c98 | 628 | head->tstamp = qp->q.stamp; |
1da177e4 | 629 | |
eddc9ec5 | 630 | iph = ip_hdr(head); |
1da177e4 LT |
631 | iph->frag_off = 0; |
632 | iph->tot_len = htons(len); | |
5173cc05 | 633 | iph->tos |= ecn; |
2bad35b7 | 634 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 635 | qp->q.fragments = NULL; |
d6bebca9 | 636 | qp->q.fragments_tail = NULL; |
1706d587 | 637 | return 0; |
1da177e4 LT |
638 | |
639 | out_nomem: | |
e905a9ed | 640 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
64ce2073 | 641 | "queue %p\n", qp); |
45542479 | 642 | err = -ENOMEM; |
1da177e4 LT |
643 | goto out_fail; |
644 | out_oversize: | |
645 | if (net_ratelimit()) | |
673d57e7 HH |
646 | printk(KERN_INFO "Oversized IP packet from %pI4.\n", |
647 | &qp->saddr); | |
1da177e4 | 648 | out_fail: |
bbf31bf1 | 649 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 650 | return err; |
1da177e4 LT |
651 | } |
652 | ||
653 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 654 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 655 | { |
1da177e4 | 656 | struct ipq *qp; |
ac18e750 | 657 | struct net *net; |
e905a9ed | 658 | |
adf30907 | 659 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); |
7c73a6fa | 660 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
661 | |
662 | /* Start by cleaning up the memory. */ | |
e31e0bdc | 663 | if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh) |
6ddc0822 | 664 | ip_evictor(net); |
1da177e4 | 665 | |
1da177e4 | 666 | /* Lookup (or create) queue header */ |
ac18e750 | 667 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 668 | int ret; |
1da177e4 | 669 | |
5ab11c98 | 670 | spin_lock(&qp->q.lock); |
1da177e4 | 671 | |
1706d587 | 672 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 673 | |
5ab11c98 | 674 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 675 | ipq_put(qp); |
776c729e | 676 | return ret; |
1da177e4 LT |
677 | } |
678 | ||
7c73a6fa | 679 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 680 | kfree_skb(skb); |
776c729e | 681 | return -ENOMEM; |
1da177e4 | 682 | } |
4bc2f18b | 683 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 684 | |
bc416d97 ED |
685 | struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user) |
686 | { | |
687 | const struct iphdr *iph; | |
688 | u32 len; | |
689 | ||
690 | if (skb->protocol != htons(ETH_P_IP)) | |
691 | return skb; | |
692 | ||
693 | if (!pskb_may_pull(skb, sizeof(struct iphdr))) | |
694 | return skb; | |
695 | ||
696 | iph = ip_hdr(skb); | |
697 | if (iph->ihl < 5 || iph->version != 4) | |
698 | return skb; | |
699 | if (!pskb_may_pull(skb, iph->ihl*4)) | |
700 | return skb; | |
701 | iph = ip_hdr(skb); | |
702 | len = ntohs(iph->tot_len); | |
703 | if (skb->len < len || len < (iph->ihl * 4)) | |
704 | return skb; | |
705 | ||
706 | if (ip_is_fragment(ip_hdr(skb))) { | |
707 | skb = skb_share_check(skb, GFP_ATOMIC); | |
708 | if (skb) { | |
709 | if (pskb_trim_rcsum(skb, len)) | |
710 | return skb; | |
711 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
712 | if (ip_defrag(skb, user)) | |
713 | return NULL; | |
714 | skb->rxhash = 0; | |
715 | } | |
716 | } | |
717 | return skb; | |
718 | } | |
719 | EXPORT_SYMBOL(ip_check_defrag); | |
720 | ||
8d8354d2 PE |
721 | #ifdef CONFIG_SYSCTL |
722 | static int zero; | |
723 | ||
0a64b4b8 | 724 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 725 | { |
8d8354d2 | 726 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 727 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
728 | .maxlen = sizeof(int), |
729 | .mode = 0644, | |
6d9f239a | 730 | .proc_handler = proc_dointvec |
8d8354d2 PE |
731 | }, |
732 | { | |
8d8354d2 | 733 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 734 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
735 | .maxlen = sizeof(int), |
736 | .mode = 0644, | |
6d9f239a | 737 | .proc_handler = proc_dointvec |
8d8354d2 PE |
738 | }, |
739 | { | |
8d8354d2 | 740 | .procname = "ipfrag_time", |
b2fd5321 | 741 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
742 | .maxlen = sizeof(int), |
743 | .mode = 0644, | |
6d9f239a | 744 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 745 | }, |
7d291ebb PE |
746 | { } |
747 | }; | |
748 | ||
749 | static struct ctl_table ip4_frags_ctl_table[] = { | |
8d8354d2 | 750 | { |
8d8354d2 | 751 | .procname = "ipfrag_secret_interval", |
3b4bc4a2 | 752 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
753 | .maxlen = sizeof(int), |
754 | .mode = 0644, | |
6d9f239a | 755 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 PE |
756 | }, |
757 | { | |
758 | .procname = "ipfrag_max_dist", | |
759 | .data = &sysctl_ipfrag_max_dist, | |
760 | .maxlen = sizeof(int), | |
761 | .mode = 0644, | |
6d9f239a | 762 | .proc_handler = proc_dointvec_minmax, |
8d8354d2 PE |
763 | .extra1 = &zero |
764 | }, | |
765 | { } | |
766 | }; | |
767 | ||
2c8c1e72 | 768 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 769 | { |
e4a2d5c2 | 770 | struct ctl_table *table; |
8d8354d2 PE |
771 | struct ctl_table_header *hdr; |
772 | ||
0a64b4b8 | 773 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 774 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 775 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
e4a2d5c2 PE |
776 | if (table == NULL) |
777 | goto err_alloc; | |
778 | ||
e31e0bdc PE |
779 | table[0].data = &net->ipv4.frags.high_thresh; |
780 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 781 | table[2].data = &net->ipv4.frags.timeout; |
e4a2d5c2 PE |
782 | } |
783 | ||
784 | hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table); | |
785 | if (hdr == NULL) | |
786 | goto err_reg; | |
787 | ||
788 | net->ipv4.frags_hdr = hdr; | |
789 | return 0; | |
790 | ||
791 | err_reg: | |
09ad9bc7 | 792 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
793 | kfree(table); |
794 | err_alloc: | |
795 | return -ENOMEM; | |
796 | } | |
797 | ||
2c8c1e72 | 798 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
799 | { |
800 | struct ctl_table *table; | |
801 | ||
802 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
803 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
804 | kfree(table); | |
8d8354d2 | 805 | } |
7d291ebb PE |
806 | |
807 | static void ip4_frags_ctl_register(void) | |
808 | { | |
809 | register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table); | |
810 | } | |
8d8354d2 | 811 | #else |
0a64b4b8 | 812 | static inline int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
813 | { |
814 | return 0; | |
815 | } | |
e4a2d5c2 | 816 | |
0a64b4b8 | 817 | static inline void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
818 | { |
819 | } | |
7d291ebb PE |
820 | |
821 | static inline void ip4_frags_ctl_register(void) | |
822 | { | |
823 | } | |
8d8354d2 PE |
824 | #endif |
825 | ||
2c8c1e72 | 826 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 827 | { |
e31e0bdc PE |
828 | /* |
829 | * Fragment cache limits. We will commit 256K at one time. Should we | |
830 | * cross that limit we will prune down to 192K. This should cope with | |
831 | * even the most extreme cases without allowing an attacker to | |
832 | * measurably harm machine performance. | |
833 | */ | |
834 | net->ipv4.frags.high_thresh = 256 * 1024; | |
835 | net->ipv4.frags.low_thresh = 192 * 1024; | |
b2fd5321 PE |
836 | /* |
837 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
838 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
839 | * by TTL. | |
840 | */ | |
841 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
842 | ||
e5a2bb84 PE |
843 | inet_frags_init_net(&net->ipv4.frags); |
844 | ||
0a64b4b8 | 845 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
846 | } |
847 | ||
2c8c1e72 | 848 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 849 | { |
0a64b4b8 | 850 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
851 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
852 | } | |
853 | ||
854 | static struct pernet_operations ip4_frags_ops = { | |
855 | .init = ipv4_frags_init_net, | |
856 | .exit = ipv4_frags_exit_net, | |
857 | }; | |
858 | ||
b7aa0bf7 | 859 | void __init ipfrag_init(void) |
1da177e4 | 860 | { |
7d291ebb | 861 | ip4_frags_ctl_register(); |
81566e83 | 862 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 863 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 864 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
865 | ip4_frags.destructor = ip4_frag_free; |
866 | ip4_frags.skb_free = NULL; | |
867 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 868 | ip4_frags.match = ip4_frag_match; |
e521db9d | 869 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 870 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 871 | inet_frags_init(&ip4_frags); |
1da177e4 | 872 | } |