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 | ||
afd46503 JP |
23 | #define pr_fmt(fmt) "IPv4: " fmt |
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> | |
5a0e3ad6 | 37 | #include <linux/slab.h> |
e9017b55 SW |
38 | #include <net/route.h> |
39 | #include <net/dst.h> | |
1da177e4 LT |
40 | #include <net/sock.h> |
41 | #include <net/ip.h> | |
42 | #include <net/icmp.h> | |
43 | #include <net/checksum.h> | |
89cee8b1 | 44 | #include <net/inetpeer.h> |
5ab11c98 | 45 | #include <net/inet_frag.h> |
1da177e4 LT |
46 | #include <linux/tcp.h> |
47 | #include <linux/udp.h> | |
48 | #include <linux/inet.h> | |
49 | #include <linux/netfilter_ipv4.h> | |
6623e3b2 | 50 | #include <net/inet_ecn.h> |
385add90 | 51 | #include <net/l3mdev.h> |
1da177e4 LT |
52 | |
53 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
54 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
55 | * as well. Or notify me, at least. --ANK | |
56 | */ | |
d4ad4d22 | 57 | static const char ip_frag_cache_name[] = "ip4-frags"; |
89cee8b1 | 58 | |
1da177e4 LT |
59 | struct ipfrag_skb_cb |
60 | { | |
61 | struct inet_skb_parm h; | |
62 | int offset; | |
63 | }; | |
64 | ||
fd3f8c4c | 65 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb)) |
1da177e4 LT |
66 | |
67 | /* Describe an entry in the "incomplete datagrams" queue. */ | |
68 | struct ipq { | |
5ab11c98 PE |
69 | struct inet_frag_queue q; |
70 | ||
1da177e4 | 71 | u32 user; |
18277770 AV |
72 | __be32 saddr; |
73 | __be32 daddr; | |
74 | __be16 id; | |
1da177e4 | 75 | u8 protocol; |
6623e3b2 | 76 | u8 ecn; /* RFC3168 support */ |
d6b915e2 | 77 | u16 max_df_size; /* largest frag with DF set seen */ |
89cee8b1 | 78 | int iif; |
385add90 | 79 | int vif; /* L3 master device index */ |
89cee8b1 HX |
80 | unsigned int rid; |
81 | struct inet_peer *peer; | |
1da177e4 LT |
82 | }; |
83 | ||
aa1f731e | 84 | static u8 ip4_frag_ecn(u8 tos) |
6623e3b2 | 85 | { |
5173cc05 | 86 | return 1 << (tos & INET_ECN_MASK); |
6623e3b2 ED |
87 | } |
88 | ||
7eb95156 | 89 | static struct inet_frags ip4_frags; |
1da177e4 | 90 | |
6ddc0822 | 91 | int ip_frag_mem(struct net *net) |
7eb95156 | 92 | { |
d433673e | 93 | return sum_frag_mem_limit(&net->ipv4.frags); |
7eb95156 | 94 | } |
1da177e4 | 95 | |
1706d587 HX |
96 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
97 | struct net_device *dev); | |
98 | ||
c6fda282 PE |
99 | struct ip4_create_arg { |
100 | struct iphdr *iph; | |
101 | u32 user; | |
9972f134 | 102 | int vif; |
c6fda282 PE |
103 | }; |
104 | ||
18277770 | 105 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 106 | { |
e7b519ba | 107 | net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd)); |
18277770 AV |
108 | return jhash_3words((__force u32)id << 16 | prot, |
109 | (__force u32)saddr, (__force u32)daddr, | |
fb3cfe6e | 110 | ip4_frags.rnd); |
1da177e4 LT |
111 | } |
112 | ||
36c77782 | 113 | static unsigned int ip4_hashfn(const struct inet_frag_queue *q) |
1da177e4 | 114 | { |
36c77782 | 115 | const struct ipq *ipq; |
1da177e4 | 116 | |
321a3a99 PE |
117 | ipq = container_of(q, struct ipq, q); |
118 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
119 | } |
120 | ||
36c77782 | 121 | static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a) |
abd6523d | 122 | { |
36c77782 FW |
123 | const struct ipq *qp; |
124 | const struct ip4_create_arg *arg = a; | |
abd6523d PE |
125 | |
126 | qp = container_of(q, struct ipq, q); | |
a02cec21 | 127 | return qp->id == arg->iph->id && |
cbc264ca ED |
128 | qp->saddr == arg->iph->saddr && |
129 | qp->daddr == arg->iph->daddr && | |
130 | qp->protocol == arg->iph->protocol && | |
9972f134 DA |
131 | qp->user == arg->user && |
132 | qp->vif == arg->vif; | |
abd6523d PE |
133 | } |
134 | ||
36c77782 | 135 | static void ip4_frag_init(struct inet_frag_queue *q, const void *a) |
c6fda282 PE |
136 | { |
137 | struct ipq *qp = container_of(q, struct ipq, q); | |
54db0cc2 G |
138 | struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, |
139 | frags); | |
140 | struct net *net = container_of(ipv4, struct net, ipv4); | |
141 | ||
36c77782 | 142 | const struct ip4_create_arg *arg = a; |
c6fda282 PE |
143 | |
144 | qp->protocol = arg->iph->protocol; | |
145 | qp->id = arg->iph->id; | |
6623e3b2 | 146 | qp->ecn = ip4_frag_ecn(arg->iph->tos); |
c6fda282 PE |
147 | qp->saddr = arg->iph->saddr; |
148 | qp->daddr = arg->iph->daddr; | |
9972f134 | 149 | qp->vif = arg->vif; |
c6fda282 | 150 | qp->user = arg->user; |
0fbf4cb2 | 151 | qp->peer = q->net->max_dist ? |
192132b9 DA |
152 | inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) : |
153 | NULL; | |
c6fda282 PE |
154 | } |
155 | ||
aa1f731e | 156 | static void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 157 | { |
1e4b8287 PE |
158 | struct ipq *qp; |
159 | ||
160 | qp = container_of(q, struct ipq, q); | |
161 | if (qp->peer) | |
162 | inet_putpeer(qp->peer); | |
1da177e4 LT |
163 | } |
164 | ||
1da177e4 LT |
165 | |
166 | /* Destruction primitives. */ | |
167 | ||
aa1f731e | 168 | static void ipq_put(struct ipq *ipq) |
1da177e4 | 169 | { |
762cc408 | 170 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
171 | } |
172 | ||
173 | /* Kill ipq entry. It is not destroyed immediately, | |
174 | * because caller (and someone more) holds reference count. | |
175 | */ | |
176 | static void ipq_kill(struct ipq *ipq) | |
177 | { | |
277e650d | 178 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
179 | } |
180 | ||
5cf42280 AZ |
181 | static bool frag_expire_skip_icmp(u32 user) |
182 | { | |
183 | return user == IP_DEFRAG_AF_PACKET || | |
184 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN, | |
8bc04864 AZ |
185 | __IP_DEFRAG_CONNTRACK_IN_END) || |
186 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN, | |
187 | __IP_DEFRAG_CONNTRACK_BRIDGE_IN); | |
5cf42280 AZ |
188 | } |
189 | ||
1da177e4 LT |
190 | /* |
191 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
192 | */ | |
193 | static void ip_expire(unsigned long arg) | |
194 | { | |
e521db9d | 195 | struct ipq *qp; |
84a3aa00 | 196 | struct net *net; |
e521db9d PE |
197 | |
198 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 199 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 200 | |
5ab11c98 | 201 | spin_lock(&qp->q.lock); |
1da177e4 | 202 | |
06aa8b8a | 203 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
204 | goto out; |
205 | ||
206 | ipq_kill(qp); | |
b45386ef | 207 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 208 | |
caaecdd3 | 209 | if (!inet_frag_evicting(&qp->q)) { |
5ab11c98 | 210 | struct sk_buff *head = qp->q.fragments; |
64f3b9e2 ED |
211 | const struct iphdr *iph; |
212 | int err; | |
cb84663e | 213 | |
b45386ef | 214 | __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT); |
2e404f63 NA |
215 | |
216 | if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments) | |
217 | goto out; | |
218 | ||
69df9d59 ED |
219 | rcu_read_lock(); |
220 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
221 | if (!head->dev) |
222 | goto out_rcu_unlock; | |
223 | ||
97599dc7 | 224 | /* skb has no dst, perform route lookup again */ |
64f3b9e2 | 225 | iph = ip_hdr(head); |
c6cffba4 DM |
226 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, |
227 | iph->tos, head->dev); | |
64f3b9e2 ED |
228 | if (err) |
229 | goto out_rcu_unlock; | |
230 | ||
2e404f63 | 231 | /* Only an end host needs to send an ICMP |
64f3b9e2 | 232 | * "Fragment Reassembly Timeout" message, per RFC792. |
e9017b55 | 233 | */ |
5cf42280 AZ |
234 | if (frag_expire_skip_icmp(qp->user) && |
235 | (skb_rtable(head)->rt_type != RTN_LOCAL)) | |
64f3b9e2 ED |
236 | goto out_rcu_unlock; |
237 | ||
e9017b55 SW |
238 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
239 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 240 | out_rcu_unlock: |
d1c9ae6d PM |
241 | rcu_read_unlock(); |
242 | } | |
1da177e4 | 243 | out: |
5ab11c98 | 244 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 245 | ipq_put(qp); |
1da177e4 LT |
246 | } |
247 | ||
abd6523d PE |
248 | /* Find the correct entry in the "incomplete datagrams" queue for |
249 | * this IP datagram, and create new one, if nothing is found. | |
250 | */ | |
9972f134 DA |
251 | static struct ipq *ip_find(struct net *net, struct iphdr *iph, |
252 | u32 user, int vif) | |
1da177e4 | 253 | { |
c6fda282 PE |
254 | struct inet_frag_queue *q; |
255 | struct ip4_create_arg arg; | |
abd6523d | 256 | unsigned int hash; |
1da177e4 | 257 | |
c6fda282 PE |
258 | arg.iph = iph; |
259 | arg.user = user; | |
9972f134 | 260 | arg.vif = vif; |
9a375803 | 261 | |
abd6523d | 262 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 263 | |
ac18e750 | 264 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
5a3da1fe HFS |
265 | if (IS_ERR_OR_NULL(q)) { |
266 | inet_frag_maybe_warn_overflow(q, pr_fmt()); | |
267 | return NULL; | |
268 | } | |
c6fda282 | 269 | return container_of(q, struct ipq, q); |
1da177e4 LT |
270 | } |
271 | ||
89cee8b1 | 272 | /* Is the fragment too far ahead to be part of ipq? */ |
aa1f731e | 273 | static int ip_frag_too_far(struct ipq *qp) |
89cee8b1 HX |
274 | { |
275 | struct inet_peer *peer = qp->peer; | |
0fbf4cb2 | 276 | unsigned int max = qp->q.net->max_dist; |
89cee8b1 HX |
277 | unsigned int start, end; |
278 | ||
279 | int rc; | |
280 | ||
281 | if (!peer || !max) | |
282 | return 0; | |
283 | ||
284 | start = qp->rid; | |
285 | end = atomic_inc_return(&peer->rid); | |
286 | qp->rid = end; | |
287 | ||
5ab11c98 | 288 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
289 | |
290 | if (rc) { | |
7c73a6fa PE |
291 | struct net *net; |
292 | ||
293 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
b45386ef | 294 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
89cee8b1 HX |
295 | } |
296 | ||
297 | return rc; | |
298 | } | |
299 | ||
300 | static int ip_frag_reinit(struct ipq *qp) | |
301 | { | |
302 | struct sk_buff *fp; | |
d433673e | 303 | unsigned int sum_truesize = 0; |
89cee8b1 | 304 | |
b2fd5321 | 305 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 306 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
307 | return -ETIMEDOUT; |
308 | } | |
309 | ||
5ab11c98 | 310 | fp = qp->q.fragments; |
89cee8b1 HX |
311 | do { |
312 | struct sk_buff *xp = fp->next; | |
d433673e JDB |
313 | |
314 | sum_truesize += fp->truesize; | |
315 | kfree_skb(fp); | |
89cee8b1 HX |
316 | fp = xp; |
317 | } while (fp); | |
0e60d245 | 318 | sub_frag_mem_limit(qp->q.net, sum_truesize); |
89cee8b1 | 319 | |
06aa8b8a | 320 | qp->q.flags = 0; |
5ab11c98 PE |
321 | qp->q.len = 0; |
322 | qp->q.meat = 0; | |
323 | qp->q.fragments = NULL; | |
d6bebca9 | 324 | qp->q.fragments_tail = NULL; |
89cee8b1 | 325 | qp->iif = 0; |
6623e3b2 | 326 | qp->ecn = 0; |
89cee8b1 HX |
327 | |
328 | return 0; | |
329 | } | |
330 | ||
1da177e4 | 331 | /* Add new segment to existing queue. */ |
1706d587 | 332 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
333 | { |
334 | struct sk_buff *prev, *next; | |
1706d587 | 335 | struct net_device *dev; |
d6b915e2 | 336 | unsigned int fragsize; |
1da177e4 LT |
337 | int flags, offset; |
338 | int ihl, end; | |
1706d587 | 339 | int err = -ENOENT; |
6623e3b2 | 340 | u8 ecn; |
1da177e4 | 341 | |
06aa8b8a | 342 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
343 | goto err; |
344 | ||
89cee8b1 | 345 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
346 | unlikely(ip_frag_too_far(qp)) && |
347 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
348 | ipq_kill(qp); |
349 | goto err; | |
350 | } | |
351 | ||
6623e3b2 | 352 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 353 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
354 | flags = offset & ~IP_OFFSET; |
355 | offset &= IP_OFFSET; | |
356 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 357 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
358 | |
359 | /* Determine the position of this fragment. */ | |
0848f642 | 360 | end = offset + skb->len - skb_network_offset(skb) - ihl; |
1706d587 | 361 | err = -EINVAL; |
1da177e4 LT |
362 | |
363 | /* Is this the final fragment? */ | |
364 | if ((flags & IP_MF) == 0) { | |
365 | /* If we already have some bits beyond end | |
42b2aa86 | 366 | * or have different end, the segment is corrupted. |
1da177e4 | 367 | */ |
5ab11c98 | 368 | if (end < qp->q.len || |
06aa8b8a | 369 | ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 370 | goto err; |
06aa8b8a | 371 | qp->q.flags |= INET_FRAG_LAST_IN; |
5ab11c98 | 372 | qp->q.len = end; |
1da177e4 LT |
373 | } else { |
374 | if (end&7) { | |
375 | end &= ~7; | |
376 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
377 | skb->ip_summed = CHECKSUM_NONE; | |
378 | } | |
5ab11c98 | 379 | if (end > qp->q.len) { |
1da177e4 | 380 | /* Some bits beyond end -> corruption. */ |
06aa8b8a | 381 | if (qp->q.flags & INET_FRAG_LAST_IN) |
1da177e4 | 382 | goto err; |
5ab11c98 | 383 | qp->q.len = end; |
1da177e4 LT |
384 | } |
385 | } | |
386 | if (end == offset) | |
387 | goto err; | |
388 | ||
1706d587 | 389 | err = -ENOMEM; |
0848f642 | 390 | if (!pskb_pull(skb, skb_network_offset(skb) + ihl)) |
1da177e4 | 391 | goto err; |
1706d587 HX |
392 | |
393 | err = pskb_trim_rcsum(skb, end - offset); | |
394 | if (err) | |
1da177e4 LT |
395 | goto err; |
396 | ||
397 | /* Find out which fragments are in front and at the back of us | |
398 | * in the chain of fragments so far. We must know where to put | |
399 | * this fragment, right? | |
400 | */ | |
d6bebca9 CG |
401 | prev = qp->q.fragments_tail; |
402 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
403 | next = NULL; | |
404 | goto found; | |
405 | } | |
1da177e4 | 406 | prev = NULL; |
5ab11c98 | 407 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
408 | if (FRAG_CB(next)->offset >= offset) |
409 | break; /* bingo! */ | |
410 | prev = next; | |
411 | } | |
412 | ||
d6bebca9 | 413 | found: |
1da177e4 LT |
414 | /* We found where to put this one. Check for overlap with |
415 | * preceding fragment, and, if needed, align things so that | |
416 | * any overlaps are eliminated. | |
417 | */ | |
418 | if (prev) { | |
419 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
420 | ||
421 | if (i > 0) { | |
422 | offset += i; | |
1706d587 | 423 | err = -EINVAL; |
1da177e4 LT |
424 | if (end <= offset) |
425 | goto err; | |
1706d587 | 426 | err = -ENOMEM; |
1da177e4 LT |
427 | if (!pskb_pull(skb, i)) |
428 | goto err; | |
429 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
430 | skb->ip_summed = CHECKSUM_NONE; | |
431 | } | |
432 | } | |
433 | ||
1706d587 HX |
434 | err = -ENOMEM; |
435 | ||
1da177e4 LT |
436 | while (next && FRAG_CB(next)->offset < end) { |
437 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
438 | ||
439 | if (i < next->len) { | |
440 | /* Eat head of the next overlapped fragment | |
441 | * and leave the loop. The next ones cannot overlap. | |
442 | */ | |
443 | if (!pskb_pull(next, i)) | |
444 | goto err; | |
445 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 446 | qp->q.meat -= i; |
1da177e4 LT |
447 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
448 | next->ip_summed = CHECKSUM_NONE; | |
449 | break; | |
450 | } else { | |
451 | struct sk_buff *free_it = next; | |
452 | ||
47c6bf77 | 453 | /* Old fragment is completely overridden with |
1da177e4 LT |
454 | * new one drop it. |
455 | */ | |
456 | next = next->next; | |
457 | ||
458 | if (prev) | |
459 | prev->next = next; | |
460 | else | |
5ab11c98 | 461 | qp->q.fragments = next; |
1da177e4 | 462 | |
5ab11c98 | 463 | qp->q.meat -= free_it->len; |
0e60d245 | 464 | sub_frag_mem_limit(qp->q.net, free_it->truesize); |
d433673e | 465 | kfree_skb(free_it); |
1da177e4 LT |
466 | } |
467 | } | |
468 | ||
469 | FRAG_CB(skb)->offset = offset; | |
470 | ||
471 | /* Insert this fragment in the chain of fragments. */ | |
472 | skb->next = next; | |
d6bebca9 CG |
473 | if (!next) |
474 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
475 | if (prev) |
476 | prev->next = skb; | |
477 | else | |
5ab11c98 | 478 | qp->q.fragments = skb; |
1da177e4 | 479 | |
1706d587 HX |
480 | dev = skb->dev; |
481 | if (dev) { | |
482 | qp->iif = dev->ifindex; | |
483 | skb->dev = NULL; | |
484 | } | |
5ab11c98 PE |
485 | qp->q.stamp = skb->tstamp; |
486 | qp->q.meat += skb->len; | |
6623e3b2 | 487 | qp->ecn |= ecn; |
0e60d245 | 488 | add_frag_mem_limit(qp->q.net, skb->truesize); |
1da177e4 | 489 | if (offset == 0) |
06aa8b8a | 490 | qp->q.flags |= INET_FRAG_FIRST_IN; |
1da177e4 | 491 | |
d6b915e2 FW |
492 | fragsize = skb->len + ihl; |
493 | ||
494 | if (fragsize > qp->q.max_size) | |
495 | qp->q.max_size = fragsize; | |
496 | ||
5f2d04f1 | 497 | if (ip_hdr(skb)->frag_off & htons(IP_DF) && |
d6b915e2 FW |
498 | fragsize > qp->max_df_size) |
499 | qp->max_df_size = fragsize; | |
5f2d04f1 | 500 | |
06aa8b8a | 501 | if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
97599dc7 ED |
502 | qp->q.meat == qp->q.len) { |
503 | unsigned long orefdst = skb->_skb_refdst; | |
1706d587 | 504 | |
97599dc7 ED |
505 | skb->_skb_refdst = 0UL; |
506 | err = ip_frag_reasm(qp, prev, dev); | |
507 | skb->_skb_refdst = orefdst; | |
508 | return err; | |
509 | } | |
510 | ||
511 | skb_dst_drop(skb); | |
1706d587 | 512 | return -EINPROGRESS; |
1da177e4 LT |
513 | |
514 | err: | |
515 | kfree_skb(skb); | |
1706d587 | 516 | return err; |
1da177e4 LT |
517 | } |
518 | ||
519 | ||
520 | /* Build a new IP datagram from all its fragments. */ | |
521 | ||
1706d587 HX |
522 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
523 | struct net_device *dev) | |
1da177e4 | 524 | { |
2bad35b7 | 525 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 526 | struct iphdr *iph; |
5ab11c98 | 527 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
528 | int len; |
529 | int ihlen; | |
1706d587 | 530 | int err; |
5173cc05 | 531 | u8 ecn; |
1da177e4 LT |
532 | |
533 | ipq_kill(qp); | |
534 | ||
be991971 | 535 | ecn = ip_frag_ecn_table[qp->ecn]; |
5173cc05 ED |
536 | if (unlikely(ecn == 0xff)) { |
537 | err = -EINVAL; | |
538 | goto out_fail; | |
539 | } | |
1706d587 HX |
540 | /* Make the one we just received the head. */ |
541 | if (prev) { | |
542 | head = prev->next; | |
543 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
544 | if (!fp) |
545 | goto out_nomem; | |
546 | ||
547 | fp->next = head->next; | |
d6bebca9 CG |
548 | if (!fp->next) |
549 | qp->q.fragments_tail = fp; | |
1706d587 HX |
550 | prev->next = fp; |
551 | ||
5ab11c98 PE |
552 | skb_morph(head, qp->q.fragments); |
553 | head->next = qp->q.fragments->next; | |
1706d587 | 554 | |
cbf8f7bb | 555 | consume_skb(qp->q.fragments); |
5ab11c98 | 556 | qp->q.fragments = head; |
1706d587 HX |
557 | } |
558 | ||
51456b29 | 559 | WARN_ON(!head); |
547b792c | 560 | WARN_ON(FRAG_CB(head)->offset != 0); |
1da177e4 LT |
561 | |
562 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 563 | ihlen = ip_hdrlen(head); |
5ab11c98 | 564 | len = ihlen + qp->q.len; |
1da177e4 | 565 | |
1706d587 | 566 | err = -E2BIG; |
132adf54 | 567 | if (len > 65535) |
1da177e4 LT |
568 | goto out_oversize; |
569 | ||
570 | /* Head of list must not be cloned. */ | |
14bbd6a5 | 571 | if (skb_unclone(head, GFP_ATOMIC)) |
1da177e4 LT |
572 | goto out_nomem; |
573 | ||
574 | /* If the first fragment is fragmented itself, we split | |
575 | * it to two chunks: the first with data and paged part | |
576 | * and the second, holding only fragments. */ | |
21dc3301 | 577 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
578 | struct sk_buff *clone; |
579 | int i, plen = 0; | |
580 | ||
51456b29 IM |
581 | clone = alloc_skb(0, GFP_ATOMIC); |
582 | if (!clone) | |
1da177e4 LT |
583 | goto out_nomem; |
584 | clone->next = head->next; | |
585 | head->next = clone; | |
586 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 587 | skb_frag_list_init(head); |
9e903e08 ED |
588 | for (i = 0; i < skb_shinfo(head)->nr_frags; i++) |
589 | plen += skb_frag_size(&skb_shinfo(head)->frags[i]); | |
1da177e4 LT |
590 | clone->len = clone->data_len = head->data_len - plen; |
591 | head->data_len -= clone->len; | |
592 | head->len -= clone->len; | |
593 | clone->csum = 0; | |
594 | clone->ip_summed = head->ip_summed; | |
0e60d245 | 595 | add_frag_mem_limit(qp->q.net, clone->truesize); |
1da177e4 LT |
596 | } |
597 | ||
14fe22e3 | 598 | skb_shinfo(head)->frag_list = head->next; |
d56f90a7 | 599 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 | 600 | |
14fe22e3 FW |
601 | for (fp=head->next; fp; fp = fp->next) { |
602 | head->data_len += fp->len; | |
603 | head->len += fp->len; | |
1da177e4 LT |
604 | if (head->ip_summed != fp->ip_summed) |
605 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 606 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 | 607 | head->csum = csum_add(head->csum, fp->csum); |
14fe22e3 | 608 | head->truesize += fp->truesize; |
1da177e4 | 609 | } |
5510b3c2 | 610 | sub_frag_mem_limit(qp->q.net, head->truesize); |
1da177e4 LT |
611 | |
612 | head->next = NULL; | |
613 | head->dev = dev; | |
5ab11c98 | 614 | head->tstamp = qp->q.stamp; |
d6b915e2 | 615 | IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size); |
1da177e4 | 616 | |
eddc9ec5 | 617 | iph = ip_hdr(head); |
1da177e4 | 618 | iph->tot_len = htons(len); |
5173cc05 | 619 | iph->tos |= ecn; |
d6b915e2 FW |
620 | |
621 | /* When we set IP_DF on a refragmented skb we must also force a | |
622 | * call to ip_fragment to avoid forwarding a DF-skb of size s while | |
623 | * original sender only sent fragments of size f (where f < s). | |
624 | * | |
625 | * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest | |
626 | * frag seen to avoid sending tiny DF-fragments in case skb was built | |
627 | * from one very small df-fragment and one large non-df frag. | |
628 | */ | |
629 | if (qp->max_df_size == qp->q.max_size) { | |
630 | IPCB(head)->flags |= IPSKB_FRAG_PMTU; | |
631 | iph->frag_off = htons(IP_DF); | |
632 | } else { | |
633 | iph->frag_off = 0; | |
634 | } | |
635 | ||
0848f642 EHJ |
636 | ip_send_check(iph); |
637 | ||
b45386ef | 638 | __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 639 | qp->q.fragments = NULL; |
d6bebca9 | 640 | qp->q.fragments_tail = NULL; |
1706d587 | 641 | return 0; |
1da177e4 LT |
642 | |
643 | out_nomem: | |
ba7a46f1 | 644 | net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp); |
45542479 | 645 | err = -ENOMEM; |
1da177e4 LT |
646 | goto out_fail; |
647 | out_oversize: | |
e87cc472 | 648 | net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr); |
1da177e4 | 649 | out_fail: |
b45386ef | 650 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 651 | return err; |
1da177e4 LT |
652 | } |
653 | ||
654 | /* Process an incoming IP datagram fragment. */ | |
19bcf9f2 | 655 | int ip_defrag(struct net *net, struct sk_buff *skb, u32 user) |
1da177e4 | 656 | { |
9972f134 | 657 | struct net_device *dev = skb->dev ? : skb_dst(skb)->dev; |
385add90 | 658 | int vif = l3mdev_master_ifindex_rcu(dev); |
1da177e4 | 659 | struct ipq *qp; |
e905a9ed | 660 | |
b45386ef | 661 | __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS); |
8282f274 | 662 | skb_orphan(skb); |
1da177e4 | 663 | |
1da177e4 | 664 | /* Lookup (or create) queue header */ |
9972f134 | 665 | qp = ip_find(net, ip_hdr(skb), user, vif); |
00db4124 | 666 | if (qp) { |
1706d587 | 667 | int ret; |
1da177e4 | 668 | |
5ab11c98 | 669 | spin_lock(&qp->q.lock); |
1da177e4 | 670 | |
1706d587 | 671 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 672 | |
5ab11c98 | 673 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 674 | ipq_put(qp); |
776c729e | 675 | return ret; |
1da177e4 LT |
676 | } |
677 | ||
b45386ef | 678 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 679 | kfree_skb(skb); |
776c729e | 680 | return -ENOMEM; |
1da177e4 | 681 | } |
4bc2f18b | 682 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 683 | |
19bcf9f2 | 684 | struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user) |
bc416d97 | 685 | { |
1bf3751e | 686 | struct iphdr iph; |
3e32e733 | 687 | int netoff; |
bc416d97 ED |
688 | u32 len; |
689 | ||
690 | if (skb->protocol != htons(ETH_P_IP)) | |
691 | return skb; | |
692 | ||
3e32e733 AD |
693 | netoff = skb_network_offset(skb); |
694 | ||
695 | if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0) | |
bc416d97 ED |
696 | return skb; |
697 | ||
1bf3751e | 698 | if (iph.ihl < 5 || iph.version != 4) |
bc416d97 | 699 | return skb; |
1bf3751e JB |
700 | |
701 | len = ntohs(iph.tot_len); | |
3e32e733 | 702 | if (skb->len < netoff + len || len < (iph.ihl * 4)) |
bc416d97 ED |
703 | return skb; |
704 | ||
1bf3751e | 705 | if (ip_is_fragment(&iph)) { |
bc416d97 ED |
706 | skb = skb_share_check(skb, GFP_ATOMIC); |
707 | if (skb) { | |
3e32e733 | 708 | if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) |
1bf3751e | 709 | return skb; |
3e32e733 | 710 | if (pskb_trim_rcsum(skb, netoff + len)) |
bc416d97 ED |
711 | return skb; |
712 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
19bcf9f2 | 713 | if (ip_defrag(net, skb, user)) |
bc416d97 | 714 | return NULL; |
7539fadc | 715 | skb_clear_hash(skb); |
bc416d97 ED |
716 | } |
717 | } | |
718 | return skb; | |
719 | } | |
720 | EXPORT_SYMBOL(ip_check_defrag); | |
721 | ||
8d8354d2 PE |
722 | #ifdef CONFIG_SYSCTL |
723 | static int zero; | |
724 | ||
0a64b4b8 | 725 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 726 | { |
8d8354d2 | 727 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 728 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
729 | .maxlen = sizeof(int), |
730 | .mode = 0644, | |
1bab4c75 NA |
731 | .proc_handler = proc_dointvec_minmax, |
732 | .extra1 = &init_net.ipv4.frags.low_thresh | |
8d8354d2 PE |
733 | }, |
734 | { | |
8d8354d2 | 735 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 736 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
737 | .maxlen = sizeof(int), |
738 | .mode = 0644, | |
1bab4c75 NA |
739 | .proc_handler = proc_dointvec_minmax, |
740 | .extra1 = &zero, | |
741 | .extra2 = &init_net.ipv4.frags.high_thresh | |
8d8354d2 PE |
742 | }, |
743 | { | |
8d8354d2 | 744 | .procname = "ipfrag_time", |
b2fd5321 | 745 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
746 | .maxlen = sizeof(int), |
747 | .mode = 0644, | |
6d9f239a | 748 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 749 | }, |
0fbf4cb2 NB |
750 | { |
751 | .procname = "ipfrag_max_dist", | |
752 | .data = &init_net.ipv4.frags.max_dist, | |
753 | .maxlen = sizeof(int), | |
754 | .mode = 0644, | |
755 | .proc_handler = proc_dointvec_minmax, | |
756 | .extra1 = &zero | |
757 | }, | |
7d291ebb PE |
758 | { } |
759 | }; | |
760 | ||
e3a57d18 FW |
761 | /* secret interval has been deprecated */ |
762 | static int ip4_frags_secret_interval_unused; | |
7d291ebb | 763 | static struct ctl_table ip4_frags_ctl_table[] = { |
8d8354d2 | 764 | { |
8d8354d2 | 765 | .procname = "ipfrag_secret_interval", |
e3a57d18 | 766 | .data = &ip4_frags_secret_interval_unused, |
8d8354d2 PE |
767 | .maxlen = sizeof(int), |
768 | .mode = 0644, | |
6d9f239a | 769 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 770 | }, |
8d8354d2 PE |
771 | { } |
772 | }; | |
773 | ||
2c8c1e72 | 774 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 775 | { |
e4a2d5c2 | 776 | struct ctl_table *table; |
8d8354d2 PE |
777 | struct ctl_table_header *hdr; |
778 | ||
0a64b4b8 | 779 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 780 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 781 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
51456b29 | 782 | if (!table) |
e4a2d5c2 PE |
783 | goto err_alloc; |
784 | ||
e31e0bdc | 785 | table[0].data = &net->ipv4.frags.high_thresh; |
1bab4c75 NA |
786 | table[0].extra1 = &net->ipv4.frags.low_thresh; |
787 | table[0].extra2 = &init_net.ipv4.frags.high_thresh; | |
e31e0bdc | 788 | table[1].data = &net->ipv4.frags.low_thresh; |
1bab4c75 | 789 | table[1].extra2 = &net->ipv4.frags.high_thresh; |
b2fd5321 | 790 | table[2].data = &net->ipv4.frags.timeout; |
0fbf4cb2 | 791 | table[3].data = &net->ipv4.frags.max_dist; |
e4a2d5c2 PE |
792 | } |
793 | ||
ec8f23ce | 794 | hdr = register_net_sysctl(net, "net/ipv4", table); |
51456b29 | 795 | if (!hdr) |
e4a2d5c2 PE |
796 | goto err_reg; |
797 | ||
798 | net->ipv4.frags_hdr = hdr; | |
799 | return 0; | |
800 | ||
801 | err_reg: | |
09ad9bc7 | 802 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
803 | kfree(table); |
804 | err_alloc: | |
805 | return -ENOMEM; | |
806 | } | |
807 | ||
2c8c1e72 | 808 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
809 | { |
810 | struct ctl_table *table; | |
811 | ||
812 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
813 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
814 | kfree(table); | |
8d8354d2 | 815 | } |
7d291ebb | 816 | |
57a02c39 | 817 | static void __init ip4_frags_ctl_register(void) |
7d291ebb | 818 | { |
43444757 | 819 | register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); |
7d291ebb | 820 | } |
8d8354d2 | 821 | #else |
aa1f731e | 822 | static int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
823 | { |
824 | return 0; | |
825 | } | |
e4a2d5c2 | 826 | |
aa1f731e | 827 | static void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
828 | { |
829 | } | |
7d291ebb | 830 | |
aa1f731e | 831 | static void __init ip4_frags_ctl_register(void) |
7d291ebb PE |
832 | { |
833 | } | |
8d8354d2 PE |
834 | #endif |
835 | ||
2c8c1e72 | 836 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 837 | { |
1d6119ba ED |
838 | int res; |
839 | ||
c2a93660 JDB |
840 | /* Fragment cache limits. |
841 | * | |
842 | * The fragment memory accounting code, (tries to) account for | |
843 | * the real memory usage, by measuring both the size of frag | |
844 | * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) | |
845 | * and the SKB's truesize. | |
846 | * | |
847 | * A 64K fragment consumes 129736 bytes (44*2944)+200 | |
848 | * (1500 truesize == 2944, sizeof(struct ipq) == 200) | |
849 | * | |
850 | * We will commit 4MB at one time. Should we cross that limit | |
851 | * we will prune down to 3MB, making room for approx 8 big 64K | |
852 | * fragments 8x128k. | |
e31e0bdc | 853 | */ |
c2a93660 JDB |
854 | net->ipv4.frags.high_thresh = 4 * 1024 * 1024; |
855 | net->ipv4.frags.low_thresh = 3 * 1024 * 1024; | |
b2fd5321 PE |
856 | /* |
857 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
858 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
859 | * by TTL. | |
860 | */ | |
861 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
862 | ||
0fbf4cb2 NB |
863 | net->ipv4.frags.max_dist = 64; |
864 | ||
1d6119ba ED |
865 | res = inet_frags_init_net(&net->ipv4.frags); |
866 | if (res) | |
867 | return res; | |
868 | res = ip4_frags_ns_ctl_register(net); | |
869 | if (res) | |
870 | inet_frags_uninit_net(&net->ipv4.frags); | |
871 | return res; | |
8d8354d2 PE |
872 | } |
873 | ||
2c8c1e72 | 874 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 875 | { |
0a64b4b8 | 876 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
877 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
878 | } | |
879 | ||
880 | static struct pernet_operations ip4_frags_ops = { | |
881 | .init = ipv4_frags_init_net, | |
882 | .exit = ipv4_frags_exit_net, | |
883 | }; | |
884 | ||
b7aa0bf7 | 885 | void __init ipfrag_init(void) |
1da177e4 | 886 | { |
7d291ebb | 887 | ip4_frags_ctl_register(); |
81566e83 | 888 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 889 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 890 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 | 891 | ip4_frags.destructor = ip4_frag_free; |
1e4b8287 | 892 | ip4_frags.qsize = sizeof(struct ipq); |
abd6523d | 893 | ip4_frags.match = ip4_frag_match; |
e521db9d | 894 | ip4_frags.frag_expire = ip_expire; |
d4ad4d22 NA |
895 | ip4_frags.frags_cache_name = ip_frag_cache_name; |
896 | if (inet_frags_init(&ip4_frags)) | |
897 | panic("IP: failed to allocate ip4_frags cache\n"); | |
1da177e4 | 898 | } |