Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * The IP fragmentation functionality. | |
e905a9ed | 8 | * |
1da177e4 | 9 | * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG> |
113aa838 | 10 | * Alan Cox <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
11 | * |
12 | * Fixes: | |
13 | * Alan Cox : Split from ip.c , see ip_input.c for history. | |
14 | * David S. Miller : Begin massive cleanup... | |
15 | * Andi Kleen : Add sysctls. | |
16 | * xxxx : Overlapfrag bug. | |
17 | * Ultima : ip_expire() kernel panic. | |
18 | * Bill Hawes : Frag accounting and evictor fixes. | |
19 | * John McDonald : 0 length frag bug. | |
20 | * Alexey Kuznetsov: SMP races, threading, cleanup. | |
21 | * Patrick McHardy : LRU queue of frag heads for evictor. | |
22 | */ | |
23 | ||
afd46503 JP |
24 | #define pr_fmt(fmt) "IPv4: " fmt |
25 | ||
89cee8b1 | 26 | #include <linux/compiler.h> |
1da177e4 LT |
27 | #include <linux/module.h> |
28 | #include <linux/types.h> | |
29 | #include <linux/mm.h> | |
30 | #include <linux/jiffies.h> | |
31 | #include <linux/skbuff.h> | |
32 | #include <linux/list.h> | |
33 | #include <linux/ip.h> | |
34 | #include <linux/icmp.h> | |
35 | #include <linux/netdevice.h> | |
36 | #include <linux/jhash.h> | |
37 | #include <linux/random.h> | |
5a0e3ad6 | 38 | #include <linux/slab.h> |
e9017b55 SW |
39 | #include <net/route.h> |
40 | #include <net/dst.h> | |
1da177e4 LT |
41 | #include <net/sock.h> |
42 | #include <net/ip.h> | |
43 | #include <net/icmp.h> | |
44 | #include <net/checksum.h> | |
89cee8b1 | 45 | #include <net/inetpeer.h> |
5ab11c98 | 46 | #include <net/inet_frag.h> |
1da177e4 LT |
47 | #include <linux/tcp.h> |
48 | #include <linux/udp.h> | |
49 | #include <linux/inet.h> | |
50 | #include <linux/netfilter_ipv4.h> | |
6623e3b2 | 51 | #include <net/inet_ecn.h> |
385add90 | 52 | #include <net/l3mdev.h> |
1da177e4 LT |
53 | |
54 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
55 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
56 | * as well. Or notify me, at least. --ANK | |
57 | */ | |
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 | 79 | int iif; |
385add90 | 80 | int vif; /* L3 master device index */ |
89cee8b1 HX |
81 | unsigned int rid; |
82 | struct inet_peer *peer; | |
1da177e4 LT |
83 | }; |
84 | ||
aa1f731e | 85 | static u8 ip4_frag_ecn(u8 tos) |
6623e3b2 | 86 | { |
5173cc05 | 87 | return 1 << (tos & INET_ECN_MASK); |
6623e3b2 ED |
88 | } |
89 | ||
7eb95156 | 90 | static struct inet_frags ip4_frags; |
1da177e4 | 91 | |
6ddc0822 | 92 | int ip_frag_mem(struct net *net) |
7eb95156 | 93 | { |
d433673e | 94 | return sum_frag_mem_limit(&net->ipv4.frags); |
7eb95156 | 95 | } |
1da177e4 | 96 | |
1706d587 HX |
97 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
98 | struct net_device *dev); | |
99 | ||
c6fda282 PE |
100 | struct ip4_create_arg { |
101 | struct iphdr *iph; | |
102 | u32 user; | |
9972f134 | 103 | int vif; |
c6fda282 PE |
104 | }; |
105 | ||
18277770 | 106 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 107 | { |
e7b519ba | 108 | net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd)); |
18277770 AV |
109 | return jhash_3words((__force u32)id << 16 | prot, |
110 | (__force u32)saddr, (__force u32)daddr, | |
fb3cfe6e | 111 | ip4_frags.rnd); |
1da177e4 LT |
112 | } |
113 | ||
36c77782 | 114 | static unsigned int ip4_hashfn(const struct inet_frag_queue *q) |
1da177e4 | 115 | { |
36c77782 | 116 | const struct ipq *ipq; |
1da177e4 | 117 | |
321a3a99 PE |
118 | ipq = container_of(q, struct ipq, q); |
119 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
120 | } |
121 | ||
36c77782 | 122 | static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a) |
abd6523d | 123 | { |
36c77782 FW |
124 | const struct ipq *qp; |
125 | const struct ip4_create_arg *arg = a; | |
abd6523d PE |
126 | |
127 | qp = container_of(q, struct ipq, q); | |
a02cec21 | 128 | return qp->id == arg->iph->id && |
cbc264ca ED |
129 | qp->saddr == arg->iph->saddr && |
130 | qp->daddr == arg->iph->daddr && | |
131 | qp->protocol == arg->iph->protocol && | |
9972f134 DA |
132 | qp->user == arg->user && |
133 | qp->vif == arg->vif; | |
abd6523d PE |
134 | } |
135 | ||
36c77782 | 136 | static void ip4_frag_init(struct inet_frag_queue *q, const void *a) |
c6fda282 PE |
137 | { |
138 | struct ipq *qp = container_of(q, struct ipq, q); | |
54db0cc2 G |
139 | struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, |
140 | frags); | |
141 | struct net *net = container_of(ipv4, struct net, ipv4); | |
142 | ||
36c77782 | 143 | const struct ip4_create_arg *arg = a; |
c6fda282 PE |
144 | |
145 | qp->protocol = arg->iph->protocol; | |
146 | qp->id = arg->iph->id; | |
6623e3b2 | 147 | qp->ecn = ip4_frag_ecn(arg->iph->tos); |
c6fda282 PE |
148 | qp->saddr = arg->iph->saddr; |
149 | qp->daddr = arg->iph->daddr; | |
9972f134 | 150 | qp->vif = arg->vif; |
c6fda282 | 151 | qp->user = arg->user; |
0fbf4cb2 | 152 | qp->peer = q->net->max_dist ? |
192132b9 DA |
153 | inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) : |
154 | NULL; | |
c6fda282 PE |
155 | } |
156 | ||
aa1f731e | 157 | static void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 158 | { |
1e4b8287 PE |
159 | struct ipq *qp; |
160 | ||
161 | qp = container_of(q, struct ipq, q); | |
162 | if (qp->peer) | |
163 | inet_putpeer(qp->peer); | |
1da177e4 LT |
164 | } |
165 | ||
1da177e4 LT |
166 | |
167 | /* Destruction primitives. */ | |
168 | ||
aa1f731e | 169 | static void ipq_put(struct ipq *ipq) |
1da177e4 | 170 | { |
762cc408 | 171 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
172 | } |
173 | ||
174 | /* Kill ipq entry. It is not destroyed immediately, | |
175 | * because caller (and someone more) holds reference count. | |
176 | */ | |
177 | static void ipq_kill(struct ipq *ipq) | |
178 | { | |
277e650d | 179 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
180 | } |
181 | ||
5cf42280 AZ |
182 | static bool frag_expire_skip_icmp(u32 user) |
183 | { | |
184 | return user == IP_DEFRAG_AF_PACKET || | |
185 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN, | |
8bc04864 AZ |
186 | __IP_DEFRAG_CONNTRACK_IN_END) || |
187 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN, | |
188 | __IP_DEFRAG_CONNTRACK_BRIDGE_IN); | |
5cf42280 AZ |
189 | } |
190 | ||
1da177e4 LT |
191 | /* |
192 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
193 | */ | |
194 | static void ip_expire(unsigned long arg) | |
195 | { | |
e521db9d | 196 | struct ipq *qp; |
84a3aa00 | 197 | struct net *net; |
e521db9d PE |
198 | |
199 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 200 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 201 | |
ec4fbd64 | 202 | rcu_read_lock(); |
5ab11c98 | 203 | spin_lock(&qp->q.lock); |
1da177e4 | 204 | |
06aa8b8a | 205 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
206 | goto out; |
207 | ||
208 | ipq_kill(qp); | |
b45386ef | 209 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 210 | |
caaecdd3 | 211 | if (!inet_frag_evicting(&qp->q)) { |
ec4fbd64 | 212 | struct sk_buff *clone, *head = qp->q.fragments; |
64f3b9e2 ED |
213 | const struct iphdr *iph; |
214 | int err; | |
cb84663e | 215 | |
b45386ef | 216 | __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT); |
2e404f63 NA |
217 | |
218 | if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments) | |
219 | goto out; | |
220 | ||
69df9d59 | 221 | head->dev = dev_get_by_index_rcu(net, qp->iif); |
e9017b55 | 222 | if (!head->dev) |
ec4fbd64 ED |
223 | goto out; |
224 | ||
e9017b55 | 225 | |
97599dc7 | 226 | /* skb has no dst, perform route lookup again */ |
64f3b9e2 | 227 | iph = ip_hdr(head); |
c6cffba4 DM |
228 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, |
229 | iph->tos, head->dev); | |
64f3b9e2 | 230 | if (err) |
ec4fbd64 | 231 | goto out; |
64f3b9e2 | 232 | |
2e404f63 | 233 | /* Only an end host needs to send an ICMP |
64f3b9e2 | 234 | * "Fragment Reassembly Timeout" message, per RFC792. |
e9017b55 | 235 | */ |
5cf42280 AZ |
236 | if (frag_expire_skip_icmp(qp->user) && |
237 | (skb_rtable(head)->rt_type != RTN_LOCAL)) | |
ec4fbd64 ED |
238 | goto out; |
239 | ||
240 | clone = skb_clone(head, GFP_ATOMIC); | |
64f3b9e2 | 241 | |
e9017b55 | 242 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
ec4fbd64 ED |
243 | if (clone) { |
244 | spin_unlock(&qp->q.lock); | |
245 | icmp_send(clone, ICMP_TIME_EXCEEDED, | |
246 | ICMP_EXC_FRAGTIME, 0); | |
247 | consume_skb(clone); | |
248 | goto out_rcu_unlock; | |
249 | } | |
d1c9ae6d | 250 | } |
1da177e4 | 251 | out: |
5ab11c98 | 252 | spin_unlock(&qp->q.lock); |
ec4fbd64 ED |
253 | out_rcu_unlock: |
254 | rcu_read_unlock(); | |
4b6cb5d8 | 255 | ipq_put(qp); |
1da177e4 LT |
256 | } |
257 | ||
abd6523d PE |
258 | /* Find the correct entry in the "incomplete datagrams" queue for |
259 | * this IP datagram, and create new one, if nothing is found. | |
260 | */ | |
9972f134 DA |
261 | static struct ipq *ip_find(struct net *net, struct iphdr *iph, |
262 | u32 user, int vif) | |
1da177e4 | 263 | { |
c6fda282 PE |
264 | struct inet_frag_queue *q; |
265 | struct ip4_create_arg arg; | |
abd6523d | 266 | unsigned int hash; |
1da177e4 | 267 | |
c6fda282 PE |
268 | arg.iph = iph; |
269 | arg.user = user; | |
9972f134 | 270 | arg.vif = vif; |
9a375803 | 271 | |
abd6523d | 272 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 273 | |
ac18e750 | 274 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
5a3da1fe HFS |
275 | if (IS_ERR_OR_NULL(q)) { |
276 | inet_frag_maybe_warn_overflow(q, pr_fmt()); | |
277 | return NULL; | |
278 | } | |
c6fda282 | 279 | return container_of(q, struct ipq, q); |
1da177e4 LT |
280 | } |
281 | ||
89cee8b1 | 282 | /* Is the fragment too far ahead to be part of ipq? */ |
aa1f731e | 283 | static int ip_frag_too_far(struct ipq *qp) |
89cee8b1 HX |
284 | { |
285 | struct inet_peer *peer = qp->peer; | |
0fbf4cb2 | 286 | unsigned int max = qp->q.net->max_dist; |
89cee8b1 HX |
287 | unsigned int start, end; |
288 | ||
289 | int rc; | |
290 | ||
291 | if (!peer || !max) | |
292 | return 0; | |
293 | ||
294 | start = qp->rid; | |
295 | end = atomic_inc_return(&peer->rid); | |
296 | qp->rid = end; | |
297 | ||
5ab11c98 | 298 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
299 | |
300 | if (rc) { | |
7c73a6fa PE |
301 | struct net *net; |
302 | ||
303 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
b45386ef | 304 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
89cee8b1 HX |
305 | } |
306 | ||
307 | return rc; | |
308 | } | |
309 | ||
310 | static int ip_frag_reinit(struct ipq *qp) | |
311 | { | |
312 | struct sk_buff *fp; | |
d433673e | 313 | unsigned int sum_truesize = 0; |
89cee8b1 | 314 | |
b2fd5321 | 315 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
edcb6918 | 316 | refcount_inc(&qp->q.refcnt); |
89cee8b1 HX |
317 | return -ETIMEDOUT; |
318 | } | |
319 | ||
5ab11c98 | 320 | fp = qp->q.fragments; |
89cee8b1 HX |
321 | do { |
322 | struct sk_buff *xp = fp->next; | |
d433673e JDB |
323 | |
324 | sum_truesize += fp->truesize; | |
325 | kfree_skb(fp); | |
89cee8b1 HX |
326 | fp = xp; |
327 | } while (fp); | |
0e60d245 | 328 | sub_frag_mem_limit(qp->q.net, sum_truesize); |
89cee8b1 | 329 | |
06aa8b8a | 330 | qp->q.flags = 0; |
5ab11c98 PE |
331 | qp->q.len = 0; |
332 | qp->q.meat = 0; | |
333 | qp->q.fragments = NULL; | |
d6bebca9 | 334 | qp->q.fragments_tail = NULL; |
89cee8b1 | 335 | qp->iif = 0; |
6623e3b2 | 336 | qp->ecn = 0; |
89cee8b1 HX |
337 | |
338 | return 0; | |
339 | } | |
340 | ||
1da177e4 | 341 | /* Add new segment to existing queue. */ |
1706d587 | 342 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
343 | { |
344 | struct sk_buff *prev, *next; | |
1706d587 | 345 | struct net_device *dev; |
d6b915e2 | 346 | unsigned int fragsize; |
1da177e4 LT |
347 | int flags, offset; |
348 | int ihl, end; | |
1706d587 | 349 | int err = -ENOENT; |
6623e3b2 | 350 | u8 ecn; |
1da177e4 | 351 | |
06aa8b8a | 352 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
353 | goto err; |
354 | ||
89cee8b1 | 355 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
356 | unlikely(ip_frag_too_far(qp)) && |
357 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
358 | ipq_kill(qp); |
359 | goto err; | |
360 | } | |
361 | ||
6623e3b2 | 362 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 363 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
364 | flags = offset & ~IP_OFFSET; |
365 | offset &= IP_OFFSET; | |
366 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 367 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
368 | |
369 | /* Determine the position of this fragment. */ | |
0848f642 | 370 | end = offset + skb->len - skb_network_offset(skb) - ihl; |
1706d587 | 371 | err = -EINVAL; |
1da177e4 LT |
372 | |
373 | /* Is this the final fragment? */ | |
374 | if ((flags & IP_MF) == 0) { | |
375 | /* If we already have some bits beyond end | |
42b2aa86 | 376 | * or have different end, the segment is corrupted. |
1da177e4 | 377 | */ |
5ab11c98 | 378 | if (end < qp->q.len || |
06aa8b8a | 379 | ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 380 | goto err; |
06aa8b8a | 381 | qp->q.flags |= INET_FRAG_LAST_IN; |
5ab11c98 | 382 | qp->q.len = end; |
1da177e4 LT |
383 | } else { |
384 | if (end&7) { | |
385 | end &= ~7; | |
386 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
387 | skb->ip_summed = CHECKSUM_NONE; | |
388 | } | |
5ab11c98 | 389 | if (end > qp->q.len) { |
1da177e4 | 390 | /* Some bits beyond end -> corruption. */ |
06aa8b8a | 391 | if (qp->q.flags & INET_FRAG_LAST_IN) |
1da177e4 | 392 | goto err; |
5ab11c98 | 393 | qp->q.len = end; |
1da177e4 LT |
394 | } |
395 | } | |
396 | if (end == offset) | |
397 | goto err; | |
398 | ||
1706d587 | 399 | err = -ENOMEM; |
0848f642 | 400 | if (!pskb_pull(skb, skb_network_offset(skb) + ihl)) |
1da177e4 | 401 | goto err; |
1706d587 HX |
402 | |
403 | err = pskb_trim_rcsum(skb, end - offset); | |
404 | if (err) | |
1da177e4 LT |
405 | goto err; |
406 | ||
407 | /* Find out which fragments are in front and at the back of us | |
408 | * in the chain of fragments so far. We must know where to put | |
409 | * this fragment, right? | |
410 | */ | |
d6bebca9 CG |
411 | prev = qp->q.fragments_tail; |
412 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
413 | next = NULL; | |
414 | goto found; | |
415 | } | |
1da177e4 | 416 | prev = NULL; |
5ab11c98 | 417 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
418 | if (FRAG_CB(next)->offset >= offset) |
419 | break; /* bingo! */ | |
420 | prev = next; | |
421 | } | |
422 | ||
d6bebca9 | 423 | found: |
1da177e4 LT |
424 | /* We found where to put this one. Check for overlap with |
425 | * preceding fragment, and, if needed, align things so that | |
426 | * any overlaps are eliminated. | |
427 | */ | |
428 | if (prev) { | |
429 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
430 | ||
431 | if (i > 0) { | |
432 | offset += i; | |
1706d587 | 433 | err = -EINVAL; |
1da177e4 LT |
434 | if (end <= offset) |
435 | goto err; | |
1706d587 | 436 | err = -ENOMEM; |
1da177e4 LT |
437 | if (!pskb_pull(skb, i)) |
438 | goto err; | |
439 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
440 | skb->ip_summed = CHECKSUM_NONE; | |
441 | } | |
442 | } | |
443 | ||
1706d587 HX |
444 | err = -ENOMEM; |
445 | ||
1da177e4 LT |
446 | while (next && FRAG_CB(next)->offset < end) { |
447 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
448 | ||
449 | if (i < next->len) { | |
450 | /* Eat head of the next overlapped fragment | |
451 | * and leave the loop. The next ones cannot overlap. | |
452 | */ | |
453 | if (!pskb_pull(next, i)) | |
454 | goto err; | |
455 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 456 | qp->q.meat -= i; |
1da177e4 LT |
457 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
458 | next->ip_summed = CHECKSUM_NONE; | |
459 | break; | |
460 | } else { | |
461 | struct sk_buff *free_it = next; | |
462 | ||
47c6bf77 | 463 | /* Old fragment is completely overridden with |
1da177e4 LT |
464 | * new one drop it. |
465 | */ | |
466 | next = next->next; | |
467 | ||
468 | if (prev) | |
469 | prev->next = next; | |
470 | else | |
5ab11c98 | 471 | qp->q.fragments = next; |
1da177e4 | 472 | |
5ab11c98 | 473 | qp->q.meat -= free_it->len; |
0e60d245 | 474 | sub_frag_mem_limit(qp->q.net, free_it->truesize); |
d433673e | 475 | kfree_skb(free_it); |
1da177e4 LT |
476 | } |
477 | } | |
478 | ||
479 | FRAG_CB(skb)->offset = offset; | |
480 | ||
481 | /* Insert this fragment in the chain of fragments. */ | |
482 | skb->next = next; | |
d6bebca9 CG |
483 | if (!next) |
484 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
485 | if (prev) |
486 | prev->next = skb; | |
487 | else | |
5ab11c98 | 488 | qp->q.fragments = skb; |
1da177e4 | 489 | |
1706d587 HX |
490 | dev = skb->dev; |
491 | if (dev) { | |
492 | qp->iif = dev->ifindex; | |
493 | skb->dev = NULL; | |
494 | } | |
5ab11c98 PE |
495 | qp->q.stamp = skb->tstamp; |
496 | qp->q.meat += skb->len; | |
6623e3b2 | 497 | qp->ecn |= ecn; |
0e60d245 | 498 | add_frag_mem_limit(qp->q.net, skb->truesize); |
1da177e4 | 499 | if (offset == 0) |
06aa8b8a | 500 | qp->q.flags |= INET_FRAG_FIRST_IN; |
1da177e4 | 501 | |
d6b915e2 FW |
502 | fragsize = skb->len + ihl; |
503 | ||
504 | if (fragsize > qp->q.max_size) | |
505 | qp->q.max_size = fragsize; | |
506 | ||
5f2d04f1 | 507 | if (ip_hdr(skb)->frag_off & htons(IP_DF) && |
d6b915e2 FW |
508 | fragsize > qp->max_df_size) |
509 | qp->max_df_size = fragsize; | |
5f2d04f1 | 510 | |
06aa8b8a | 511 | if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
97599dc7 ED |
512 | qp->q.meat == qp->q.len) { |
513 | unsigned long orefdst = skb->_skb_refdst; | |
1706d587 | 514 | |
97599dc7 ED |
515 | skb->_skb_refdst = 0UL; |
516 | err = ip_frag_reasm(qp, prev, dev); | |
517 | skb->_skb_refdst = orefdst; | |
518 | return err; | |
519 | } | |
520 | ||
521 | skb_dst_drop(skb); | |
1706d587 | 522 | return -EINPROGRESS; |
1da177e4 LT |
523 | |
524 | err: | |
525 | kfree_skb(skb); | |
1706d587 | 526 | return err; |
1da177e4 LT |
527 | } |
528 | ||
529 | ||
530 | /* Build a new IP datagram from all its fragments. */ | |
531 | ||
1706d587 HX |
532 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
533 | struct net_device *dev) | |
1da177e4 | 534 | { |
2bad35b7 | 535 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 536 | struct iphdr *iph; |
5ab11c98 | 537 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
538 | int len; |
539 | int ihlen; | |
1706d587 | 540 | int err; |
5173cc05 | 541 | u8 ecn; |
1da177e4 LT |
542 | |
543 | ipq_kill(qp); | |
544 | ||
be991971 | 545 | ecn = ip_frag_ecn_table[qp->ecn]; |
5173cc05 ED |
546 | if (unlikely(ecn == 0xff)) { |
547 | err = -EINVAL; | |
548 | goto out_fail; | |
549 | } | |
1706d587 HX |
550 | /* Make the one we just received the head. */ |
551 | if (prev) { | |
552 | head = prev->next; | |
553 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
554 | if (!fp) |
555 | goto out_nomem; | |
556 | ||
557 | fp->next = head->next; | |
d6bebca9 CG |
558 | if (!fp->next) |
559 | qp->q.fragments_tail = fp; | |
1706d587 HX |
560 | prev->next = fp; |
561 | ||
5ab11c98 PE |
562 | skb_morph(head, qp->q.fragments); |
563 | head->next = qp->q.fragments->next; | |
1706d587 | 564 | |
cbf8f7bb | 565 | consume_skb(qp->q.fragments); |
5ab11c98 | 566 | qp->q.fragments = head; |
1706d587 HX |
567 | } |
568 | ||
51456b29 | 569 | WARN_ON(!head); |
547b792c | 570 | WARN_ON(FRAG_CB(head)->offset != 0); |
1da177e4 LT |
571 | |
572 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 573 | ihlen = ip_hdrlen(head); |
5ab11c98 | 574 | len = ihlen + qp->q.len; |
1da177e4 | 575 | |
1706d587 | 576 | err = -E2BIG; |
132adf54 | 577 | if (len > 65535) |
1da177e4 LT |
578 | goto out_oversize; |
579 | ||
580 | /* Head of list must not be cloned. */ | |
14bbd6a5 | 581 | if (skb_unclone(head, GFP_ATOMIC)) |
1da177e4 LT |
582 | goto out_nomem; |
583 | ||
584 | /* If the first fragment is fragmented itself, we split | |
585 | * it to two chunks: the first with data and paged part | |
586 | * and the second, holding only fragments. */ | |
21dc3301 | 587 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
588 | struct sk_buff *clone; |
589 | int i, plen = 0; | |
590 | ||
51456b29 IM |
591 | clone = alloc_skb(0, GFP_ATOMIC); |
592 | if (!clone) | |
1da177e4 LT |
593 | goto out_nomem; |
594 | clone->next = head->next; | |
595 | head->next = clone; | |
596 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 597 | skb_frag_list_init(head); |
9e903e08 ED |
598 | for (i = 0; i < skb_shinfo(head)->nr_frags; i++) |
599 | plen += skb_frag_size(&skb_shinfo(head)->frags[i]); | |
1da177e4 LT |
600 | clone->len = clone->data_len = head->data_len - plen; |
601 | head->data_len -= clone->len; | |
602 | head->len -= clone->len; | |
603 | clone->csum = 0; | |
604 | clone->ip_summed = head->ip_summed; | |
0e60d245 | 605 | add_frag_mem_limit(qp->q.net, clone->truesize); |
1da177e4 LT |
606 | } |
607 | ||
14fe22e3 | 608 | skb_shinfo(head)->frag_list = head->next; |
d56f90a7 | 609 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 | 610 | |
14fe22e3 FW |
611 | for (fp=head->next; fp; fp = fp->next) { |
612 | head->data_len += fp->len; | |
613 | head->len += fp->len; | |
1da177e4 LT |
614 | if (head->ip_summed != fp->ip_summed) |
615 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 616 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 | 617 | head->csum = csum_add(head->csum, fp->csum); |
14fe22e3 | 618 | head->truesize += fp->truesize; |
1da177e4 | 619 | } |
5510b3c2 | 620 | sub_frag_mem_limit(qp->q.net, head->truesize); |
1da177e4 LT |
621 | |
622 | head->next = NULL; | |
623 | head->dev = dev; | |
5ab11c98 | 624 | head->tstamp = qp->q.stamp; |
d6b915e2 | 625 | IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size); |
1da177e4 | 626 | |
eddc9ec5 | 627 | iph = ip_hdr(head); |
1da177e4 | 628 | iph->tot_len = htons(len); |
5173cc05 | 629 | iph->tos |= ecn; |
d6b915e2 FW |
630 | |
631 | /* When we set IP_DF on a refragmented skb we must also force a | |
632 | * call to ip_fragment to avoid forwarding a DF-skb of size s while | |
633 | * original sender only sent fragments of size f (where f < s). | |
634 | * | |
635 | * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest | |
636 | * frag seen to avoid sending tiny DF-fragments in case skb was built | |
637 | * from one very small df-fragment and one large non-df frag. | |
638 | */ | |
639 | if (qp->max_df_size == qp->q.max_size) { | |
640 | IPCB(head)->flags |= IPSKB_FRAG_PMTU; | |
641 | iph->frag_off = htons(IP_DF); | |
642 | } else { | |
643 | iph->frag_off = 0; | |
644 | } | |
645 | ||
0848f642 EHJ |
646 | ip_send_check(iph); |
647 | ||
b45386ef | 648 | __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 649 | qp->q.fragments = NULL; |
d6bebca9 | 650 | qp->q.fragments_tail = NULL; |
1706d587 | 651 | return 0; |
1da177e4 LT |
652 | |
653 | out_nomem: | |
ba7a46f1 | 654 | net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp); |
45542479 | 655 | err = -ENOMEM; |
1da177e4 LT |
656 | goto out_fail; |
657 | out_oversize: | |
e87cc472 | 658 | net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr); |
1da177e4 | 659 | out_fail: |
b45386ef | 660 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 661 | return err; |
1da177e4 LT |
662 | } |
663 | ||
664 | /* Process an incoming IP datagram fragment. */ | |
19bcf9f2 | 665 | int ip_defrag(struct net *net, struct sk_buff *skb, u32 user) |
1da177e4 | 666 | { |
9972f134 | 667 | struct net_device *dev = skb->dev ? : skb_dst(skb)->dev; |
385add90 | 668 | int vif = l3mdev_master_ifindex_rcu(dev); |
1da177e4 | 669 | struct ipq *qp; |
e905a9ed | 670 | |
b45386ef | 671 | __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS); |
8282f274 | 672 | skb_orphan(skb); |
1da177e4 | 673 | |
1da177e4 | 674 | /* Lookup (or create) queue header */ |
9972f134 | 675 | qp = ip_find(net, ip_hdr(skb), user, vif); |
00db4124 | 676 | if (qp) { |
1706d587 | 677 | int ret; |
1da177e4 | 678 | |
5ab11c98 | 679 | spin_lock(&qp->q.lock); |
1da177e4 | 680 | |
1706d587 | 681 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 682 | |
5ab11c98 | 683 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 684 | ipq_put(qp); |
776c729e | 685 | return ret; |
1da177e4 LT |
686 | } |
687 | ||
b45386ef | 688 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 689 | kfree_skb(skb); |
776c729e | 690 | return -ENOMEM; |
1da177e4 | 691 | } |
4bc2f18b | 692 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 693 | |
19bcf9f2 | 694 | struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user) |
bc416d97 | 695 | { |
1bf3751e | 696 | struct iphdr iph; |
3e32e733 | 697 | int netoff; |
bc416d97 ED |
698 | u32 len; |
699 | ||
700 | if (skb->protocol != htons(ETH_P_IP)) | |
701 | return skb; | |
702 | ||
3e32e733 AD |
703 | netoff = skb_network_offset(skb); |
704 | ||
705 | if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0) | |
bc416d97 ED |
706 | return skb; |
707 | ||
1bf3751e | 708 | if (iph.ihl < 5 || iph.version != 4) |
bc416d97 | 709 | return skb; |
1bf3751e JB |
710 | |
711 | len = ntohs(iph.tot_len); | |
3e32e733 | 712 | if (skb->len < netoff + len || len < (iph.ihl * 4)) |
bc416d97 ED |
713 | return skb; |
714 | ||
1bf3751e | 715 | if (ip_is_fragment(&iph)) { |
bc416d97 ED |
716 | skb = skb_share_check(skb, GFP_ATOMIC); |
717 | if (skb) { | |
3e32e733 | 718 | if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) |
1bf3751e | 719 | return skb; |
3e32e733 | 720 | if (pskb_trim_rcsum(skb, netoff + len)) |
bc416d97 ED |
721 | return skb; |
722 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
19bcf9f2 | 723 | if (ip_defrag(net, skb, user)) |
bc416d97 | 724 | return NULL; |
7539fadc | 725 | skb_clear_hash(skb); |
bc416d97 ED |
726 | } |
727 | } | |
728 | return skb; | |
729 | } | |
730 | EXPORT_SYMBOL(ip_check_defrag); | |
731 | ||
8d8354d2 PE |
732 | #ifdef CONFIG_SYSCTL |
733 | static int zero; | |
734 | ||
0a64b4b8 | 735 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 736 | { |
8d8354d2 | 737 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 738 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
739 | .maxlen = sizeof(int), |
740 | .mode = 0644, | |
1bab4c75 NA |
741 | .proc_handler = proc_dointvec_minmax, |
742 | .extra1 = &init_net.ipv4.frags.low_thresh | |
8d8354d2 PE |
743 | }, |
744 | { | |
8d8354d2 | 745 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 746 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
747 | .maxlen = sizeof(int), |
748 | .mode = 0644, | |
1bab4c75 NA |
749 | .proc_handler = proc_dointvec_minmax, |
750 | .extra1 = &zero, | |
751 | .extra2 = &init_net.ipv4.frags.high_thresh | |
8d8354d2 PE |
752 | }, |
753 | { | |
8d8354d2 | 754 | .procname = "ipfrag_time", |
b2fd5321 | 755 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
756 | .maxlen = sizeof(int), |
757 | .mode = 0644, | |
6d9f239a | 758 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 759 | }, |
0fbf4cb2 NB |
760 | { |
761 | .procname = "ipfrag_max_dist", | |
762 | .data = &init_net.ipv4.frags.max_dist, | |
763 | .maxlen = sizeof(int), | |
764 | .mode = 0644, | |
765 | .proc_handler = proc_dointvec_minmax, | |
766 | .extra1 = &zero | |
767 | }, | |
7d291ebb PE |
768 | { } |
769 | }; | |
770 | ||
e3a57d18 FW |
771 | /* secret interval has been deprecated */ |
772 | static int ip4_frags_secret_interval_unused; | |
7d291ebb | 773 | static struct ctl_table ip4_frags_ctl_table[] = { |
8d8354d2 | 774 | { |
8d8354d2 | 775 | .procname = "ipfrag_secret_interval", |
e3a57d18 | 776 | .data = &ip4_frags_secret_interval_unused, |
8d8354d2 PE |
777 | .maxlen = sizeof(int), |
778 | .mode = 0644, | |
6d9f239a | 779 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 780 | }, |
8d8354d2 PE |
781 | { } |
782 | }; | |
783 | ||
2c8c1e72 | 784 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 785 | { |
e4a2d5c2 | 786 | struct ctl_table *table; |
8d8354d2 PE |
787 | struct ctl_table_header *hdr; |
788 | ||
0a64b4b8 | 789 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 790 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 791 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
51456b29 | 792 | if (!table) |
e4a2d5c2 PE |
793 | goto err_alloc; |
794 | ||
e31e0bdc | 795 | table[0].data = &net->ipv4.frags.high_thresh; |
1bab4c75 NA |
796 | table[0].extra1 = &net->ipv4.frags.low_thresh; |
797 | table[0].extra2 = &init_net.ipv4.frags.high_thresh; | |
e31e0bdc | 798 | table[1].data = &net->ipv4.frags.low_thresh; |
1bab4c75 | 799 | table[1].extra2 = &net->ipv4.frags.high_thresh; |
b2fd5321 | 800 | table[2].data = &net->ipv4.frags.timeout; |
0fbf4cb2 | 801 | table[3].data = &net->ipv4.frags.max_dist; |
e4a2d5c2 PE |
802 | } |
803 | ||
ec8f23ce | 804 | hdr = register_net_sysctl(net, "net/ipv4", table); |
51456b29 | 805 | if (!hdr) |
e4a2d5c2 PE |
806 | goto err_reg; |
807 | ||
808 | net->ipv4.frags_hdr = hdr; | |
809 | return 0; | |
810 | ||
811 | err_reg: | |
09ad9bc7 | 812 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
813 | kfree(table); |
814 | err_alloc: | |
815 | return -ENOMEM; | |
816 | } | |
817 | ||
2c8c1e72 | 818 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
819 | { |
820 | struct ctl_table *table; | |
821 | ||
822 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
823 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
824 | kfree(table); | |
8d8354d2 | 825 | } |
7d291ebb | 826 | |
57a02c39 | 827 | static void __init ip4_frags_ctl_register(void) |
7d291ebb | 828 | { |
43444757 | 829 | register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); |
7d291ebb | 830 | } |
8d8354d2 | 831 | #else |
aa1f731e | 832 | static int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
833 | { |
834 | return 0; | |
835 | } | |
e4a2d5c2 | 836 | |
aa1f731e | 837 | static void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
838 | { |
839 | } | |
7d291ebb | 840 | |
aa1f731e | 841 | static void __init ip4_frags_ctl_register(void) |
7d291ebb PE |
842 | { |
843 | } | |
8d8354d2 PE |
844 | #endif |
845 | ||
2c8c1e72 | 846 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 847 | { |
c2a93660 JDB |
848 | /* Fragment cache limits. |
849 | * | |
850 | * The fragment memory accounting code, (tries to) account for | |
851 | * the real memory usage, by measuring both the size of frag | |
852 | * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) | |
853 | * and the SKB's truesize. | |
854 | * | |
855 | * A 64K fragment consumes 129736 bytes (44*2944)+200 | |
856 | * (1500 truesize == 2944, sizeof(struct ipq) == 200) | |
857 | * | |
858 | * We will commit 4MB at one time. Should we cross that limit | |
859 | * we will prune down to 3MB, making room for approx 8 big 64K | |
860 | * fragments 8x128k. | |
e31e0bdc | 861 | */ |
c2a93660 JDB |
862 | net->ipv4.frags.high_thresh = 4 * 1024 * 1024; |
863 | net->ipv4.frags.low_thresh = 3 * 1024 * 1024; | |
b2fd5321 PE |
864 | /* |
865 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
866 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
867 | * by TTL. | |
868 | */ | |
869 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
870 | ||
0fbf4cb2 NB |
871 | net->ipv4.frags.max_dist = 64; |
872 | ||
5a63643e JDB |
873 | inet_frags_init_net(&net->ipv4.frags); |
874 | ||
875 | return ip4_frags_ns_ctl_register(net); | |
8d8354d2 PE |
876 | } |
877 | ||
2c8c1e72 | 878 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 879 | { |
0a64b4b8 | 880 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
881 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
882 | } | |
883 | ||
884 | static struct pernet_operations ip4_frags_ops = { | |
885 | .init = ipv4_frags_init_net, | |
886 | .exit = ipv4_frags_exit_net, | |
887 | }; | |
888 | ||
b7aa0bf7 | 889 | void __init ipfrag_init(void) |
1da177e4 | 890 | { |
7d291ebb | 891 | ip4_frags_ctl_register(); |
81566e83 | 892 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 893 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 894 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 | 895 | ip4_frags.destructor = ip4_frag_free; |
1e4b8287 | 896 | ip4_frags.qsize = sizeof(struct ipq); |
abd6523d | 897 | ip4_frags.match = ip4_frag_match; |
e521db9d | 898 | ip4_frags.frag_expire = ip_expire; |
d4ad4d22 NA |
899 | ip4_frags.frags_cache_name = ip_frag_cache_name; |
900 | if (inet_frags_init(&ip4_frags)) | |
901 | panic("IP: failed to allocate ip4_frags cache\n"); | |
1da177e4 | 902 | } |