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> |
1da177e4 LT |
51 | |
52 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
53 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
54 | * as well. Or notify me, at least. --ANK | |
55 | */ | |
56 | ||
8d8354d2 | 57 | static int sysctl_ipfrag_max_dist __read_mostly = 64; |
d4ad4d22 | 58 | static const char ip_frag_cache_name[] = "ip4-frags"; |
89cee8b1 | 59 | |
1da177e4 LT |
60 | struct ipfrag_skb_cb |
61 | { | |
62 | struct inet_skb_parm h; | |
63 | int offset; | |
64 | }; | |
65 | ||
fd3f8c4c | 66 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb)) |
1da177e4 LT |
67 | |
68 | /* Describe an entry in the "incomplete datagrams" queue. */ | |
69 | struct ipq { | |
5ab11c98 PE |
70 | struct inet_frag_queue q; |
71 | ||
1da177e4 | 72 | u32 user; |
18277770 AV |
73 | __be32 saddr; |
74 | __be32 daddr; | |
75 | __be16 id; | |
1da177e4 | 76 | u8 protocol; |
6623e3b2 | 77 | u8 ecn; /* RFC3168 support */ |
d6b915e2 | 78 | u16 max_df_size; /* largest frag with DF set seen */ |
89cee8b1 HX |
79 | int iif; |
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; | |
102 | }; | |
103 | ||
18277770 | 104 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 105 | { |
e7b519ba | 106 | net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd)); |
18277770 AV |
107 | return jhash_3words((__force u32)id << 16 | prot, |
108 | (__force u32)saddr, (__force u32)daddr, | |
fb3cfe6e | 109 | ip4_frags.rnd); |
1da177e4 LT |
110 | } |
111 | ||
36c77782 | 112 | static unsigned int ip4_hashfn(const struct inet_frag_queue *q) |
1da177e4 | 113 | { |
36c77782 | 114 | const struct ipq *ipq; |
1da177e4 | 115 | |
321a3a99 PE |
116 | ipq = container_of(q, struct ipq, q); |
117 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
118 | } |
119 | ||
36c77782 | 120 | static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a) |
abd6523d | 121 | { |
36c77782 FW |
122 | const struct ipq *qp; |
123 | const struct ip4_create_arg *arg = a; | |
abd6523d PE |
124 | |
125 | qp = container_of(q, struct ipq, q); | |
a02cec21 | 126 | return qp->id == arg->iph->id && |
cbc264ca ED |
127 | qp->saddr == arg->iph->saddr && |
128 | qp->daddr == arg->iph->daddr && | |
129 | qp->protocol == arg->iph->protocol && | |
130 | qp->user == arg->user; | |
abd6523d PE |
131 | } |
132 | ||
36c77782 | 133 | static void ip4_frag_init(struct inet_frag_queue *q, const void *a) |
c6fda282 PE |
134 | { |
135 | struct ipq *qp = container_of(q, struct ipq, q); | |
54db0cc2 G |
136 | struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, |
137 | frags); | |
138 | struct net *net = container_of(ipv4, struct net, ipv4); | |
139 | ||
36c77782 | 140 | const struct ip4_create_arg *arg = a; |
c6fda282 PE |
141 | |
142 | qp->protocol = arg->iph->protocol; | |
143 | qp->id = arg->iph->id; | |
6623e3b2 | 144 | qp->ecn = ip4_frag_ecn(arg->iph->tos); |
c6fda282 PE |
145 | qp->saddr = arg->iph->saddr; |
146 | qp->daddr = arg->iph->daddr; | |
147 | qp->user = arg->user; | |
148 | qp->peer = sysctl_ipfrag_max_dist ? | |
c0efc887 | 149 | inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL; |
c6fda282 PE |
150 | } |
151 | ||
aa1f731e | 152 | static void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 153 | { |
1e4b8287 PE |
154 | struct ipq *qp; |
155 | ||
156 | qp = container_of(q, struct ipq, q); | |
157 | if (qp->peer) | |
158 | inet_putpeer(qp->peer); | |
1da177e4 LT |
159 | } |
160 | ||
1da177e4 LT |
161 | |
162 | /* Destruction primitives. */ | |
163 | ||
aa1f731e | 164 | static void ipq_put(struct ipq *ipq) |
1da177e4 | 165 | { |
762cc408 | 166 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
167 | } |
168 | ||
169 | /* Kill ipq entry. It is not destroyed immediately, | |
170 | * because caller (and someone more) holds reference count. | |
171 | */ | |
172 | static void ipq_kill(struct ipq *ipq) | |
173 | { | |
277e650d | 174 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
175 | } |
176 | ||
5cf42280 AZ |
177 | static bool frag_expire_skip_icmp(u32 user) |
178 | { | |
179 | return user == IP_DEFRAG_AF_PACKET || | |
180 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN, | |
8bc04864 AZ |
181 | __IP_DEFRAG_CONNTRACK_IN_END) || |
182 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN, | |
183 | __IP_DEFRAG_CONNTRACK_BRIDGE_IN); | |
5cf42280 AZ |
184 | } |
185 | ||
1da177e4 LT |
186 | /* |
187 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
188 | */ | |
189 | static void ip_expire(unsigned long arg) | |
190 | { | |
e521db9d | 191 | struct ipq *qp; |
84a3aa00 | 192 | struct net *net; |
e521db9d PE |
193 | |
194 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 195 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 196 | |
5ab11c98 | 197 | spin_lock(&qp->q.lock); |
1da177e4 | 198 | |
06aa8b8a | 199 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
200 | goto out; |
201 | ||
202 | ipq_kill(qp); | |
7c73a6fa | 203 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 204 | |
caaecdd3 | 205 | if (!inet_frag_evicting(&qp->q)) { |
5ab11c98 | 206 | struct sk_buff *head = qp->q.fragments; |
64f3b9e2 ED |
207 | const struct iphdr *iph; |
208 | int err; | |
cb84663e | 209 | |
2e404f63 NA |
210 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); |
211 | ||
212 | if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments) | |
213 | goto out; | |
214 | ||
69df9d59 ED |
215 | rcu_read_lock(); |
216 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
217 | if (!head->dev) |
218 | goto out_rcu_unlock; | |
219 | ||
97599dc7 | 220 | /* skb has no dst, perform route lookup again */ |
64f3b9e2 | 221 | iph = ip_hdr(head); |
c6cffba4 DM |
222 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, |
223 | iph->tos, head->dev); | |
64f3b9e2 ED |
224 | if (err) |
225 | goto out_rcu_unlock; | |
226 | ||
2e404f63 | 227 | /* Only an end host needs to send an ICMP |
64f3b9e2 | 228 | * "Fragment Reassembly Timeout" message, per RFC792. |
e9017b55 | 229 | */ |
5cf42280 AZ |
230 | if (frag_expire_skip_icmp(qp->user) && |
231 | (skb_rtable(head)->rt_type != RTN_LOCAL)) | |
64f3b9e2 ED |
232 | goto out_rcu_unlock; |
233 | ||
e9017b55 SW |
234 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
235 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 236 | out_rcu_unlock: |
d1c9ae6d PM |
237 | rcu_read_unlock(); |
238 | } | |
1da177e4 | 239 | out: |
5ab11c98 | 240 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 241 | ipq_put(qp); |
1da177e4 LT |
242 | } |
243 | ||
abd6523d PE |
244 | /* Find the correct entry in the "incomplete datagrams" queue for |
245 | * this IP datagram, and create new one, if nothing is found. | |
246 | */ | |
aa1f731e | 247 | static struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 248 | { |
c6fda282 PE |
249 | struct inet_frag_queue *q; |
250 | struct ip4_create_arg arg; | |
abd6523d | 251 | unsigned int hash; |
1da177e4 | 252 | |
c6fda282 PE |
253 | arg.iph = iph; |
254 | arg.user = user; | |
9a375803 | 255 | |
abd6523d | 256 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 257 | |
ac18e750 | 258 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
5a3da1fe HFS |
259 | if (IS_ERR_OR_NULL(q)) { |
260 | inet_frag_maybe_warn_overflow(q, pr_fmt()); | |
261 | return NULL; | |
262 | } | |
c6fda282 | 263 | return container_of(q, struct ipq, q); |
1da177e4 LT |
264 | } |
265 | ||
89cee8b1 | 266 | /* Is the fragment too far ahead to be part of ipq? */ |
aa1f731e | 267 | static int ip_frag_too_far(struct ipq *qp) |
89cee8b1 HX |
268 | { |
269 | struct inet_peer *peer = qp->peer; | |
270 | unsigned int max = sysctl_ipfrag_max_dist; | |
271 | unsigned int start, end; | |
272 | ||
273 | int rc; | |
274 | ||
275 | if (!peer || !max) | |
276 | return 0; | |
277 | ||
278 | start = qp->rid; | |
279 | end = atomic_inc_return(&peer->rid); | |
280 | qp->rid = end; | |
281 | ||
5ab11c98 | 282 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
283 | |
284 | if (rc) { | |
7c73a6fa PE |
285 | struct net *net; |
286 | ||
287 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
288 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
289 | } |
290 | ||
291 | return rc; | |
292 | } | |
293 | ||
294 | static int ip_frag_reinit(struct ipq *qp) | |
295 | { | |
296 | struct sk_buff *fp; | |
d433673e | 297 | unsigned int sum_truesize = 0; |
89cee8b1 | 298 | |
b2fd5321 | 299 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 300 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
301 | return -ETIMEDOUT; |
302 | } | |
303 | ||
5ab11c98 | 304 | fp = qp->q.fragments; |
89cee8b1 HX |
305 | do { |
306 | struct sk_buff *xp = fp->next; | |
d433673e JDB |
307 | |
308 | sum_truesize += fp->truesize; | |
309 | kfree_skb(fp); | |
89cee8b1 HX |
310 | fp = xp; |
311 | } while (fp); | |
0e60d245 | 312 | sub_frag_mem_limit(qp->q.net, sum_truesize); |
89cee8b1 | 313 | |
06aa8b8a | 314 | qp->q.flags = 0; |
5ab11c98 PE |
315 | qp->q.len = 0; |
316 | qp->q.meat = 0; | |
317 | qp->q.fragments = NULL; | |
d6bebca9 | 318 | qp->q.fragments_tail = NULL; |
89cee8b1 | 319 | qp->iif = 0; |
6623e3b2 | 320 | qp->ecn = 0; |
89cee8b1 HX |
321 | |
322 | return 0; | |
323 | } | |
324 | ||
1da177e4 | 325 | /* Add new segment to existing queue. */ |
1706d587 | 326 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
327 | { |
328 | struct sk_buff *prev, *next; | |
1706d587 | 329 | struct net_device *dev; |
d6b915e2 | 330 | unsigned int fragsize; |
1da177e4 LT |
331 | int flags, offset; |
332 | int ihl, end; | |
1706d587 | 333 | int err = -ENOENT; |
6623e3b2 | 334 | u8 ecn; |
1da177e4 | 335 | |
06aa8b8a | 336 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
337 | goto err; |
338 | ||
89cee8b1 | 339 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
340 | unlikely(ip_frag_too_far(qp)) && |
341 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
342 | ipq_kill(qp); |
343 | goto err; | |
344 | } | |
345 | ||
6623e3b2 | 346 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 347 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
348 | flags = offset & ~IP_OFFSET; |
349 | offset &= IP_OFFSET; | |
350 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 351 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
352 | |
353 | /* Determine the position of this fragment. */ | |
0848f642 | 354 | end = offset + skb->len - skb_network_offset(skb) - ihl; |
1706d587 | 355 | err = -EINVAL; |
1da177e4 LT |
356 | |
357 | /* Is this the final fragment? */ | |
358 | if ((flags & IP_MF) == 0) { | |
359 | /* If we already have some bits beyond end | |
42b2aa86 | 360 | * or have different end, the segment is corrupted. |
1da177e4 | 361 | */ |
5ab11c98 | 362 | if (end < qp->q.len || |
06aa8b8a | 363 | ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 364 | goto err; |
06aa8b8a | 365 | qp->q.flags |= INET_FRAG_LAST_IN; |
5ab11c98 | 366 | qp->q.len = end; |
1da177e4 LT |
367 | } else { |
368 | if (end&7) { | |
369 | end &= ~7; | |
370 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
371 | skb->ip_summed = CHECKSUM_NONE; | |
372 | } | |
5ab11c98 | 373 | if (end > qp->q.len) { |
1da177e4 | 374 | /* Some bits beyond end -> corruption. */ |
06aa8b8a | 375 | if (qp->q.flags & INET_FRAG_LAST_IN) |
1da177e4 | 376 | goto err; |
5ab11c98 | 377 | qp->q.len = end; |
1da177e4 LT |
378 | } |
379 | } | |
380 | if (end == offset) | |
381 | goto err; | |
382 | ||
1706d587 | 383 | err = -ENOMEM; |
0848f642 | 384 | if (!pskb_pull(skb, skb_network_offset(skb) + ihl)) |
1da177e4 | 385 | goto err; |
1706d587 HX |
386 | |
387 | err = pskb_trim_rcsum(skb, end - offset); | |
388 | if (err) | |
1da177e4 LT |
389 | goto err; |
390 | ||
391 | /* Find out which fragments are in front and at the back of us | |
392 | * in the chain of fragments so far. We must know where to put | |
393 | * this fragment, right? | |
394 | */ | |
d6bebca9 CG |
395 | prev = qp->q.fragments_tail; |
396 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
397 | next = NULL; | |
398 | goto found; | |
399 | } | |
1da177e4 | 400 | prev = NULL; |
5ab11c98 | 401 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
402 | if (FRAG_CB(next)->offset >= offset) |
403 | break; /* bingo! */ | |
404 | prev = next; | |
405 | } | |
406 | ||
d6bebca9 | 407 | found: |
1da177e4 LT |
408 | /* We found where to put this one. Check for overlap with |
409 | * preceding fragment, and, if needed, align things so that | |
410 | * any overlaps are eliminated. | |
411 | */ | |
412 | if (prev) { | |
413 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
414 | ||
415 | if (i > 0) { | |
416 | offset += i; | |
1706d587 | 417 | err = -EINVAL; |
1da177e4 LT |
418 | if (end <= offset) |
419 | goto err; | |
1706d587 | 420 | err = -ENOMEM; |
1da177e4 LT |
421 | if (!pskb_pull(skb, i)) |
422 | goto err; | |
423 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
424 | skb->ip_summed = CHECKSUM_NONE; | |
425 | } | |
426 | } | |
427 | ||
1706d587 HX |
428 | err = -ENOMEM; |
429 | ||
1da177e4 LT |
430 | while (next && FRAG_CB(next)->offset < end) { |
431 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
432 | ||
433 | if (i < next->len) { | |
434 | /* Eat head of the next overlapped fragment | |
435 | * and leave the loop. The next ones cannot overlap. | |
436 | */ | |
437 | if (!pskb_pull(next, i)) | |
438 | goto err; | |
439 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 440 | qp->q.meat -= i; |
1da177e4 LT |
441 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
442 | next->ip_summed = CHECKSUM_NONE; | |
443 | break; | |
444 | } else { | |
445 | struct sk_buff *free_it = next; | |
446 | ||
47c6bf77 | 447 | /* Old fragment is completely overridden with |
1da177e4 LT |
448 | * new one drop it. |
449 | */ | |
450 | next = next->next; | |
451 | ||
452 | if (prev) | |
453 | prev->next = next; | |
454 | else | |
5ab11c98 | 455 | qp->q.fragments = next; |
1da177e4 | 456 | |
5ab11c98 | 457 | qp->q.meat -= free_it->len; |
0e60d245 | 458 | sub_frag_mem_limit(qp->q.net, free_it->truesize); |
d433673e | 459 | kfree_skb(free_it); |
1da177e4 LT |
460 | } |
461 | } | |
462 | ||
463 | FRAG_CB(skb)->offset = offset; | |
464 | ||
465 | /* Insert this fragment in the chain of fragments. */ | |
466 | skb->next = next; | |
d6bebca9 CG |
467 | if (!next) |
468 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
469 | if (prev) |
470 | prev->next = skb; | |
471 | else | |
5ab11c98 | 472 | qp->q.fragments = skb; |
1da177e4 | 473 | |
1706d587 HX |
474 | dev = skb->dev; |
475 | if (dev) { | |
476 | qp->iif = dev->ifindex; | |
477 | skb->dev = NULL; | |
478 | } | |
5ab11c98 PE |
479 | qp->q.stamp = skb->tstamp; |
480 | qp->q.meat += skb->len; | |
6623e3b2 | 481 | qp->ecn |= ecn; |
0e60d245 | 482 | add_frag_mem_limit(qp->q.net, skb->truesize); |
1da177e4 | 483 | if (offset == 0) |
06aa8b8a | 484 | qp->q.flags |= INET_FRAG_FIRST_IN; |
1da177e4 | 485 | |
d6b915e2 FW |
486 | fragsize = skb->len + ihl; |
487 | ||
488 | if (fragsize > qp->q.max_size) | |
489 | qp->q.max_size = fragsize; | |
490 | ||
5f2d04f1 | 491 | if (ip_hdr(skb)->frag_off & htons(IP_DF) && |
d6b915e2 FW |
492 | fragsize > qp->max_df_size) |
493 | qp->max_df_size = fragsize; | |
5f2d04f1 | 494 | |
06aa8b8a | 495 | if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
97599dc7 ED |
496 | qp->q.meat == qp->q.len) { |
497 | unsigned long orefdst = skb->_skb_refdst; | |
1706d587 | 498 | |
97599dc7 ED |
499 | skb->_skb_refdst = 0UL; |
500 | err = ip_frag_reasm(qp, prev, dev); | |
501 | skb->_skb_refdst = orefdst; | |
502 | return err; | |
503 | } | |
504 | ||
505 | skb_dst_drop(skb); | |
1706d587 | 506 | return -EINPROGRESS; |
1da177e4 LT |
507 | |
508 | err: | |
509 | kfree_skb(skb); | |
1706d587 | 510 | return err; |
1da177e4 LT |
511 | } |
512 | ||
513 | ||
514 | /* Build a new IP datagram from all its fragments. */ | |
515 | ||
1706d587 HX |
516 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
517 | struct net_device *dev) | |
1da177e4 | 518 | { |
2bad35b7 | 519 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 520 | struct iphdr *iph; |
5ab11c98 | 521 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
522 | int len; |
523 | int ihlen; | |
1706d587 | 524 | int err; |
5173cc05 | 525 | u8 ecn; |
1da177e4 LT |
526 | |
527 | ipq_kill(qp); | |
528 | ||
be991971 | 529 | ecn = ip_frag_ecn_table[qp->ecn]; |
5173cc05 ED |
530 | if (unlikely(ecn == 0xff)) { |
531 | err = -EINVAL; | |
532 | goto out_fail; | |
533 | } | |
1706d587 HX |
534 | /* Make the one we just received the head. */ |
535 | if (prev) { | |
536 | head = prev->next; | |
537 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
538 | if (!fp) |
539 | goto out_nomem; | |
540 | ||
541 | fp->next = head->next; | |
d6bebca9 CG |
542 | if (!fp->next) |
543 | qp->q.fragments_tail = fp; | |
1706d587 HX |
544 | prev->next = fp; |
545 | ||
5ab11c98 PE |
546 | skb_morph(head, qp->q.fragments); |
547 | head->next = qp->q.fragments->next; | |
1706d587 | 548 | |
cbf8f7bb | 549 | consume_skb(qp->q.fragments); |
5ab11c98 | 550 | qp->q.fragments = head; |
1706d587 HX |
551 | } |
552 | ||
51456b29 | 553 | WARN_ON(!head); |
547b792c | 554 | WARN_ON(FRAG_CB(head)->offset != 0); |
1da177e4 LT |
555 | |
556 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 557 | ihlen = ip_hdrlen(head); |
5ab11c98 | 558 | len = ihlen + qp->q.len; |
1da177e4 | 559 | |
1706d587 | 560 | err = -E2BIG; |
132adf54 | 561 | if (len > 65535) |
1da177e4 LT |
562 | goto out_oversize; |
563 | ||
564 | /* Head of list must not be cloned. */ | |
14bbd6a5 | 565 | if (skb_unclone(head, GFP_ATOMIC)) |
1da177e4 LT |
566 | goto out_nomem; |
567 | ||
568 | /* If the first fragment is fragmented itself, we split | |
569 | * it to two chunks: the first with data and paged part | |
570 | * and the second, holding only fragments. */ | |
21dc3301 | 571 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
572 | struct sk_buff *clone; |
573 | int i, plen = 0; | |
574 | ||
51456b29 IM |
575 | clone = alloc_skb(0, GFP_ATOMIC); |
576 | if (!clone) | |
1da177e4 LT |
577 | goto out_nomem; |
578 | clone->next = head->next; | |
579 | head->next = clone; | |
580 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 581 | skb_frag_list_init(head); |
9e903e08 ED |
582 | for (i = 0; i < skb_shinfo(head)->nr_frags; i++) |
583 | plen += skb_frag_size(&skb_shinfo(head)->frags[i]); | |
1da177e4 LT |
584 | clone->len = clone->data_len = head->data_len - plen; |
585 | head->data_len -= clone->len; | |
586 | head->len -= clone->len; | |
587 | clone->csum = 0; | |
588 | clone->ip_summed = head->ip_summed; | |
0e60d245 | 589 | add_frag_mem_limit(qp->q.net, clone->truesize); |
1da177e4 LT |
590 | } |
591 | ||
14fe22e3 | 592 | skb_shinfo(head)->frag_list = head->next; |
d56f90a7 | 593 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 | 594 | |
14fe22e3 FW |
595 | for (fp=head->next; fp; fp = fp->next) { |
596 | head->data_len += fp->len; | |
597 | head->len += fp->len; | |
1da177e4 LT |
598 | if (head->ip_summed != fp->ip_summed) |
599 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 600 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 | 601 | head->csum = csum_add(head->csum, fp->csum); |
14fe22e3 | 602 | head->truesize += fp->truesize; |
1da177e4 | 603 | } |
5510b3c2 | 604 | sub_frag_mem_limit(qp->q.net, head->truesize); |
1da177e4 LT |
605 | |
606 | head->next = NULL; | |
607 | head->dev = dev; | |
5ab11c98 | 608 | head->tstamp = qp->q.stamp; |
d6b915e2 | 609 | IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size); |
1da177e4 | 610 | |
eddc9ec5 | 611 | iph = ip_hdr(head); |
1da177e4 | 612 | iph->tot_len = htons(len); |
5173cc05 | 613 | iph->tos |= ecn; |
d6b915e2 FW |
614 | |
615 | /* When we set IP_DF on a refragmented skb we must also force a | |
616 | * call to ip_fragment to avoid forwarding a DF-skb of size s while | |
617 | * original sender only sent fragments of size f (where f < s). | |
618 | * | |
619 | * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest | |
620 | * frag seen to avoid sending tiny DF-fragments in case skb was built | |
621 | * from one very small df-fragment and one large non-df frag. | |
622 | */ | |
623 | if (qp->max_df_size == qp->q.max_size) { | |
624 | IPCB(head)->flags |= IPSKB_FRAG_PMTU; | |
625 | iph->frag_off = htons(IP_DF); | |
626 | } else { | |
627 | iph->frag_off = 0; | |
628 | } | |
629 | ||
0848f642 EHJ |
630 | ip_send_check(iph); |
631 | ||
2bad35b7 | 632 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 633 | qp->q.fragments = NULL; |
d6bebca9 | 634 | qp->q.fragments_tail = NULL; |
1706d587 | 635 | return 0; |
1da177e4 LT |
636 | |
637 | out_nomem: | |
ba7a46f1 | 638 | net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp); |
45542479 | 639 | err = -ENOMEM; |
1da177e4 LT |
640 | goto out_fail; |
641 | out_oversize: | |
e87cc472 | 642 | net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr); |
1da177e4 | 643 | out_fail: |
bbf31bf1 | 644 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 645 | return err; |
1da177e4 LT |
646 | } |
647 | ||
648 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 649 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 650 | { |
1da177e4 | 651 | struct ipq *qp; |
ac18e750 | 652 | struct net *net; |
e905a9ed | 653 | |
adf30907 | 654 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); |
7c73a6fa | 655 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 | 656 | |
1da177e4 | 657 | /* Lookup (or create) queue header */ |
00db4124 IM |
658 | qp = ip_find(net, ip_hdr(skb), user); |
659 | if (qp) { | |
1706d587 | 660 | int ret; |
1da177e4 | 661 | |
5ab11c98 | 662 | spin_lock(&qp->q.lock); |
1da177e4 | 663 | |
1706d587 | 664 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 665 | |
5ab11c98 | 666 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 667 | ipq_put(qp); |
776c729e | 668 | return ret; |
1da177e4 LT |
669 | } |
670 | ||
7c73a6fa | 671 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 672 | kfree_skb(skb); |
776c729e | 673 | return -ENOMEM; |
1da177e4 | 674 | } |
4bc2f18b | 675 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 676 | |
bc416d97 ED |
677 | struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user) |
678 | { | |
1bf3751e | 679 | struct iphdr iph; |
3e32e733 | 680 | int netoff; |
bc416d97 ED |
681 | u32 len; |
682 | ||
683 | if (skb->protocol != htons(ETH_P_IP)) | |
684 | return skb; | |
685 | ||
3e32e733 AD |
686 | netoff = skb_network_offset(skb); |
687 | ||
688 | if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0) | |
bc416d97 ED |
689 | return skb; |
690 | ||
1bf3751e | 691 | if (iph.ihl < 5 || iph.version != 4) |
bc416d97 | 692 | return skb; |
1bf3751e JB |
693 | |
694 | len = ntohs(iph.tot_len); | |
3e32e733 | 695 | if (skb->len < netoff + len || len < (iph.ihl * 4)) |
bc416d97 ED |
696 | return skb; |
697 | ||
1bf3751e | 698 | if (ip_is_fragment(&iph)) { |
bc416d97 ED |
699 | skb = skb_share_check(skb, GFP_ATOMIC); |
700 | if (skb) { | |
3e32e733 | 701 | if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) |
1bf3751e | 702 | return skb; |
3e32e733 | 703 | if (pskb_trim_rcsum(skb, netoff + len)) |
bc416d97 ED |
704 | return skb; |
705 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
706 | if (ip_defrag(skb, user)) | |
707 | return NULL; | |
7539fadc | 708 | skb_clear_hash(skb); |
bc416d97 ED |
709 | } |
710 | } | |
711 | return skb; | |
712 | } | |
713 | EXPORT_SYMBOL(ip_check_defrag); | |
714 | ||
8d8354d2 PE |
715 | #ifdef CONFIG_SYSCTL |
716 | static int zero; | |
717 | ||
0a64b4b8 | 718 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 719 | { |
8d8354d2 | 720 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 721 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
722 | .maxlen = sizeof(int), |
723 | .mode = 0644, | |
1bab4c75 NA |
724 | .proc_handler = proc_dointvec_minmax, |
725 | .extra1 = &init_net.ipv4.frags.low_thresh | |
8d8354d2 PE |
726 | }, |
727 | { | |
8d8354d2 | 728 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 729 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
730 | .maxlen = sizeof(int), |
731 | .mode = 0644, | |
1bab4c75 NA |
732 | .proc_handler = proc_dointvec_minmax, |
733 | .extra1 = &zero, | |
734 | .extra2 = &init_net.ipv4.frags.high_thresh | |
8d8354d2 PE |
735 | }, |
736 | { | |
8d8354d2 | 737 | .procname = "ipfrag_time", |
b2fd5321 | 738 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
739 | .maxlen = sizeof(int), |
740 | .mode = 0644, | |
6d9f239a | 741 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 742 | }, |
7d291ebb PE |
743 | { } |
744 | }; | |
745 | ||
e3a57d18 FW |
746 | /* secret interval has been deprecated */ |
747 | static int ip4_frags_secret_interval_unused; | |
7d291ebb | 748 | static struct ctl_table ip4_frags_ctl_table[] = { |
8d8354d2 | 749 | { |
8d8354d2 | 750 | .procname = "ipfrag_secret_interval", |
e3a57d18 | 751 | .data = &ip4_frags_secret_interval_unused, |
8d8354d2 PE |
752 | .maxlen = sizeof(int), |
753 | .mode = 0644, | |
6d9f239a | 754 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 PE |
755 | }, |
756 | { | |
757 | .procname = "ipfrag_max_dist", | |
758 | .data = &sysctl_ipfrag_max_dist, | |
759 | .maxlen = sizeof(int), | |
760 | .mode = 0644, | |
6d9f239a | 761 | .proc_handler = proc_dointvec_minmax, |
8d8354d2 PE |
762 | .extra1 = &zero |
763 | }, | |
764 | { } | |
765 | }; | |
766 | ||
2c8c1e72 | 767 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 768 | { |
e4a2d5c2 | 769 | struct ctl_table *table; |
8d8354d2 PE |
770 | struct ctl_table_header *hdr; |
771 | ||
0a64b4b8 | 772 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 773 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 774 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
51456b29 | 775 | if (!table) |
e4a2d5c2 PE |
776 | goto err_alloc; |
777 | ||
e31e0bdc | 778 | table[0].data = &net->ipv4.frags.high_thresh; |
1bab4c75 NA |
779 | table[0].extra1 = &net->ipv4.frags.low_thresh; |
780 | table[0].extra2 = &init_net.ipv4.frags.high_thresh; | |
e31e0bdc | 781 | table[1].data = &net->ipv4.frags.low_thresh; |
1bab4c75 | 782 | table[1].extra2 = &net->ipv4.frags.high_thresh; |
b2fd5321 | 783 | table[2].data = &net->ipv4.frags.timeout; |
464dc801 EB |
784 | |
785 | /* Don't export sysctls to unprivileged users */ | |
786 | if (net->user_ns != &init_user_ns) | |
787 | table[0].procname = NULL; | |
e4a2d5c2 PE |
788 | } |
789 | ||
ec8f23ce | 790 | hdr = register_net_sysctl(net, "net/ipv4", table); |
51456b29 | 791 | if (!hdr) |
e4a2d5c2 PE |
792 | goto err_reg; |
793 | ||
794 | net->ipv4.frags_hdr = hdr; | |
795 | return 0; | |
796 | ||
797 | err_reg: | |
09ad9bc7 | 798 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
799 | kfree(table); |
800 | err_alloc: | |
801 | return -ENOMEM; | |
802 | } | |
803 | ||
2c8c1e72 | 804 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
805 | { |
806 | struct ctl_table *table; | |
807 | ||
808 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
809 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
810 | kfree(table); | |
8d8354d2 | 811 | } |
7d291ebb | 812 | |
57a02c39 | 813 | static void __init ip4_frags_ctl_register(void) |
7d291ebb | 814 | { |
43444757 | 815 | register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); |
7d291ebb | 816 | } |
8d8354d2 | 817 | #else |
aa1f731e | 818 | static int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
819 | { |
820 | return 0; | |
821 | } | |
e4a2d5c2 | 822 | |
aa1f731e | 823 | static void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
824 | { |
825 | } | |
7d291ebb | 826 | |
aa1f731e | 827 | static void __init ip4_frags_ctl_register(void) |
7d291ebb PE |
828 | { |
829 | } | |
8d8354d2 PE |
830 | #endif |
831 | ||
2c8c1e72 | 832 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 833 | { |
c2a93660 JDB |
834 | /* Fragment cache limits. |
835 | * | |
836 | * The fragment memory accounting code, (tries to) account for | |
837 | * the real memory usage, by measuring both the size of frag | |
838 | * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) | |
839 | * and the SKB's truesize. | |
840 | * | |
841 | * A 64K fragment consumes 129736 bytes (44*2944)+200 | |
842 | * (1500 truesize == 2944, sizeof(struct ipq) == 200) | |
843 | * | |
844 | * We will commit 4MB at one time. Should we cross that limit | |
845 | * we will prune down to 3MB, making room for approx 8 big 64K | |
846 | * fragments 8x128k. | |
e31e0bdc | 847 | */ |
c2a93660 JDB |
848 | net->ipv4.frags.high_thresh = 4 * 1024 * 1024; |
849 | net->ipv4.frags.low_thresh = 3 * 1024 * 1024; | |
b2fd5321 PE |
850 | /* |
851 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
852 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
853 | * by TTL. | |
854 | */ | |
855 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
856 | ||
e5a2bb84 PE |
857 | inet_frags_init_net(&net->ipv4.frags); |
858 | ||
0a64b4b8 | 859 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
860 | } |
861 | ||
2c8c1e72 | 862 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 863 | { |
0a64b4b8 | 864 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
865 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
866 | } | |
867 | ||
868 | static struct pernet_operations ip4_frags_ops = { | |
869 | .init = ipv4_frags_init_net, | |
870 | .exit = ipv4_frags_exit_net, | |
871 | }; | |
872 | ||
b7aa0bf7 | 873 | void __init ipfrag_init(void) |
1da177e4 | 874 | { |
7d291ebb | 875 | ip4_frags_ctl_register(); |
81566e83 | 876 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 877 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 878 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
879 | ip4_frags.destructor = ip4_frag_free; |
880 | ip4_frags.skb_free = NULL; | |
881 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 882 | ip4_frags.match = ip4_frag_match; |
e521db9d | 883 | ip4_frags.frag_expire = ip_expire; |
d4ad4d22 NA |
884 | ip4_frags.frags_cache_name = ip_frag_cache_name; |
885 | if (inet_frags_init(&ip4_frags)) | |
886 | panic("IP: failed to allocate ip4_frags cache\n"); | |
1da177e4 | 887 | } |