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 | /* Describe an entry in the "incomplete datagrams" queue. */ |
61 | struct ipq { | |
5ab11c98 PE |
62 | struct inet_frag_queue q; |
63 | ||
6623e3b2 | 64 | u8 ecn; /* RFC3168 support */ |
d6b915e2 | 65 | u16 max_df_size; /* largest frag with DF set seen */ |
89cee8b1 HX |
66 | int iif; |
67 | unsigned int rid; | |
68 | struct inet_peer *peer; | |
1da177e4 LT |
69 | }; |
70 | ||
aa1f731e | 71 | static u8 ip4_frag_ecn(u8 tos) |
6623e3b2 | 72 | { |
5173cc05 | 73 | return 1 << (tos & INET_ECN_MASK); |
6623e3b2 ED |
74 | } |
75 | ||
7eb95156 | 76 | static struct inet_frags ip4_frags; |
1da177e4 | 77 | |
1706d587 HX |
78 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
79 | struct net_device *dev); | |
80 | ||
abd6523d | 81 | |
36c77782 | 82 | static void ip4_frag_init(struct inet_frag_queue *q, const void *a) |
c6fda282 PE |
83 | { |
84 | struct ipq *qp = container_of(q, struct ipq, q); | |
54db0cc2 G |
85 | struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, |
86 | frags); | |
87 | struct net *net = container_of(ipv4, struct net, ipv4); | |
88 | ||
648700f7 | 89 | const struct frag_v4_compare_key *key = a; |
c6fda282 | 90 | |
648700f7 ED |
91 | q->key.v4 = *key; |
92 | qp->ecn = 0; | |
0fbf4cb2 | 93 | qp->peer = q->net->max_dist ? |
648700f7 | 94 | inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) : |
192132b9 | 95 | NULL; |
c6fda282 PE |
96 | } |
97 | ||
aa1f731e | 98 | static void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 99 | { |
1e4b8287 PE |
100 | struct ipq *qp; |
101 | ||
102 | qp = container_of(q, struct ipq, q); | |
103 | if (qp->peer) | |
104 | inet_putpeer(qp->peer); | |
1da177e4 LT |
105 | } |
106 | ||
1da177e4 LT |
107 | |
108 | /* Destruction primitives. */ | |
109 | ||
aa1f731e | 110 | static void ipq_put(struct ipq *ipq) |
1da177e4 | 111 | { |
093ba729 | 112 | inet_frag_put(&ipq->q); |
1da177e4 LT |
113 | } |
114 | ||
115 | /* Kill ipq entry. It is not destroyed immediately, | |
116 | * because caller (and someone more) holds reference count. | |
117 | */ | |
118 | static void ipq_kill(struct ipq *ipq) | |
119 | { | |
093ba729 | 120 | inet_frag_kill(&ipq->q); |
1da177e4 LT |
121 | } |
122 | ||
5cf42280 AZ |
123 | static bool frag_expire_skip_icmp(u32 user) |
124 | { | |
125 | return user == IP_DEFRAG_AF_PACKET || | |
126 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN, | |
8bc04864 AZ |
127 | __IP_DEFRAG_CONNTRACK_IN_END) || |
128 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN, | |
129 | __IP_DEFRAG_CONNTRACK_BRIDGE_IN); | |
5cf42280 AZ |
130 | } |
131 | ||
1da177e4 LT |
132 | /* |
133 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
134 | */ | |
78802011 | 135 | static void ip_expire(struct timer_list *t) |
1da177e4 | 136 | { |
78802011 | 137 | struct inet_frag_queue *frag = from_timer(frag, t, timer); |
399d1404 | 138 | const struct iphdr *iph; |
1eec5d56 | 139 | struct sk_buff *head; |
84a3aa00 | 140 | struct net *net; |
399d1404 ED |
141 | struct ipq *qp; |
142 | int err; | |
e521db9d | 143 | |
78802011 | 144 | qp = container_of(frag, struct ipq, q); |
84a3aa00 | 145 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 146 | |
ec4fbd64 | 147 | rcu_read_lock(); |
5ab11c98 | 148 | spin_lock(&qp->q.lock); |
1da177e4 | 149 | |
06aa8b8a | 150 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
151 | goto out; |
152 | ||
153 | ipq_kill(qp); | |
b45386ef | 154 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 155 | |
399d1404 | 156 | head = qp->q.fragments; |
cb84663e | 157 | |
399d1404 | 158 | __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT); |
2e404f63 | 159 | |
399d1404 ED |
160 | if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !head) |
161 | goto out; | |
2e404f63 | 162 | |
399d1404 ED |
163 | head->dev = dev_get_by_index_rcu(net, qp->iif); |
164 | if (!head->dev) | |
165 | goto out; | |
ec4fbd64 | 166 | |
e9017b55 | 167 | |
399d1404 ED |
168 | /* skb has no dst, perform route lookup again */ |
169 | iph = ip_hdr(head); | |
170 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, | |
c6cffba4 | 171 | iph->tos, head->dev); |
399d1404 ED |
172 | if (err) |
173 | goto out; | |
174 | ||
175 | /* Only an end host needs to send an ICMP | |
176 | * "Fragment Reassembly Timeout" message, per RFC792. | |
177 | */ | |
178 | if (frag_expire_skip_icmp(qp->q.key.v4.user) && | |
179 | (skb_rtable(head)->rt_type != RTN_LOCAL)) | |
180 | goto out; | |
181 | ||
1eec5d56 ED |
182 | skb_get(head); |
183 | spin_unlock(&qp->q.lock); | |
184 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
185 | kfree_skb(head); | |
186 | goto out_rcu_unlock; | |
399d1404 | 187 | |
1da177e4 | 188 | out: |
5ab11c98 | 189 | spin_unlock(&qp->q.lock); |
ec4fbd64 ED |
190 | out_rcu_unlock: |
191 | rcu_read_unlock(); | |
4b6cb5d8 | 192 | ipq_put(qp); |
1da177e4 LT |
193 | } |
194 | ||
abd6523d PE |
195 | /* Find the correct entry in the "incomplete datagrams" queue for |
196 | * this IP datagram, and create new one, if nothing is found. | |
197 | */ | |
9972f134 DA |
198 | static struct ipq *ip_find(struct net *net, struct iphdr *iph, |
199 | u32 user, int vif) | |
1da177e4 | 200 | { |
648700f7 ED |
201 | struct frag_v4_compare_key key = { |
202 | .saddr = iph->saddr, | |
203 | .daddr = iph->daddr, | |
204 | .user = user, | |
205 | .vif = vif, | |
206 | .id = iph->id, | |
207 | .protocol = iph->protocol, | |
208 | }; | |
c6fda282 | 209 | struct inet_frag_queue *q; |
9a375803 | 210 | |
648700f7 | 211 | q = inet_frag_find(&net->ipv4.frags, &key); |
2d44ed22 | 212 | if (!q) |
5a3da1fe | 213 | return NULL; |
2d44ed22 | 214 | |
c6fda282 | 215 | return container_of(q, struct ipq, q); |
1da177e4 LT |
216 | } |
217 | ||
89cee8b1 | 218 | /* Is the fragment too far ahead to be part of ipq? */ |
aa1f731e | 219 | static int ip_frag_too_far(struct ipq *qp) |
89cee8b1 HX |
220 | { |
221 | struct inet_peer *peer = qp->peer; | |
0fbf4cb2 | 222 | unsigned int max = qp->q.net->max_dist; |
89cee8b1 HX |
223 | unsigned int start, end; |
224 | ||
225 | int rc; | |
226 | ||
227 | if (!peer || !max) | |
228 | return 0; | |
229 | ||
230 | start = qp->rid; | |
231 | end = atomic_inc_return(&peer->rid); | |
232 | qp->rid = end; | |
233 | ||
5ab11c98 | 234 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
235 | |
236 | if (rc) { | |
7c73a6fa PE |
237 | struct net *net; |
238 | ||
239 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
b45386ef | 240 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
89cee8b1 HX |
241 | } |
242 | ||
243 | return rc; | |
244 | } | |
245 | ||
246 | static int ip_frag_reinit(struct ipq *qp) | |
247 | { | |
248 | struct sk_buff *fp; | |
d433673e | 249 | unsigned int sum_truesize = 0; |
89cee8b1 | 250 | |
b2fd5321 | 251 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
edcb6918 | 252 | refcount_inc(&qp->q.refcnt); |
89cee8b1 HX |
253 | return -ETIMEDOUT; |
254 | } | |
255 | ||
5ab11c98 | 256 | fp = qp->q.fragments; |
89cee8b1 HX |
257 | do { |
258 | struct sk_buff *xp = fp->next; | |
d433673e JDB |
259 | |
260 | sum_truesize += fp->truesize; | |
261 | kfree_skb(fp); | |
89cee8b1 HX |
262 | fp = xp; |
263 | } while (fp); | |
0e60d245 | 264 | sub_frag_mem_limit(qp->q.net, sum_truesize); |
89cee8b1 | 265 | |
06aa8b8a | 266 | qp->q.flags = 0; |
5ab11c98 PE |
267 | qp->q.len = 0; |
268 | qp->q.meat = 0; | |
269 | qp->q.fragments = NULL; | |
d6bebca9 | 270 | qp->q.fragments_tail = NULL; |
89cee8b1 | 271 | qp->iif = 0; |
6623e3b2 | 272 | qp->ecn = 0; |
89cee8b1 HX |
273 | |
274 | return 0; | |
275 | } | |
276 | ||
1da177e4 | 277 | /* Add new segment to existing queue. */ |
1706d587 | 278 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
279 | { |
280 | struct sk_buff *prev, *next; | |
1706d587 | 281 | struct net_device *dev; |
d6b915e2 | 282 | unsigned int fragsize; |
1da177e4 LT |
283 | int flags, offset; |
284 | int ihl, end; | |
1706d587 | 285 | int err = -ENOENT; |
6623e3b2 | 286 | u8 ecn; |
1da177e4 | 287 | |
06aa8b8a | 288 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
289 | goto err; |
290 | ||
89cee8b1 | 291 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
292 | unlikely(ip_frag_too_far(qp)) && |
293 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
294 | ipq_kill(qp); |
295 | goto err; | |
296 | } | |
297 | ||
6623e3b2 | 298 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 299 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
300 | flags = offset & ~IP_OFFSET; |
301 | offset &= IP_OFFSET; | |
302 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 303 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
304 | |
305 | /* Determine the position of this fragment. */ | |
0848f642 | 306 | end = offset + skb->len - skb_network_offset(skb) - ihl; |
1706d587 | 307 | err = -EINVAL; |
1da177e4 LT |
308 | |
309 | /* Is this the final fragment? */ | |
310 | if ((flags & IP_MF) == 0) { | |
311 | /* If we already have some bits beyond end | |
42b2aa86 | 312 | * or have different end, the segment is corrupted. |
1da177e4 | 313 | */ |
5ab11c98 | 314 | if (end < qp->q.len || |
06aa8b8a | 315 | ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 316 | goto err; |
06aa8b8a | 317 | qp->q.flags |= INET_FRAG_LAST_IN; |
5ab11c98 | 318 | qp->q.len = end; |
1da177e4 LT |
319 | } else { |
320 | if (end&7) { | |
321 | end &= ~7; | |
322 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
323 | skb->ip_summed = CHECKSUM_NONE; | |
324 | } | |
5ab11c98 | 325 | if (end > qp->q.len) { |
1da177e4 | 326 | /* Some bits beyond end -> corruption. */ |
06aa8b8a | 327 | if (qp->q.flags & INET_FRAG_LAST_IN) |
1da177e4 | 328 | goto err; |
5ab11c98 | 329 | qp->q.len = end; |
1da177e4 LT |
330 | } |
331 | } | |
332 | if (end == offset) | |
333 | goto err; | |
334 | ||
1706d587 | 335 | err = -ENOMEM; |
0848f642 | 336 | if (!pskb_pull(skb, skb_network_offset(skb) + ihl)) |
1da177e4 | 337 | goto err; |
1706d587 HX |
338 | |
339 | err = pskb_trim_rcsum(skb, end - offset); | |
340 | if (err) | |
1da177e4 LT |
341 | goto err; |
342 | ||
343 | /* Find out which fragments are in front and at the back of us | |
344 | * in the chain of fragments so far. We must know where to put | |
345 | * this fragment, right? | |
346 | */ | |
d6bebca9 | 347 | prev = qp->q.fragments_tail; |
bf663371 | 348 | if (!prev || prev->ip_defrag_offset < offset) { |
d6bebca9 CG |
349 | next = NULL; |
350 | goto found; | |
351 | } | |
1da177e4 | 352 | prev = NULL; |
5ab11c98 | 353 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
bf663371 | 354 | if (next->ip_defrag_offset >= offset) |
1da177e4 LT |
355 | break; /* bingo! */ |
356 | prev = next; | |
357 | } | |
358 | ||
d6bebca9 | 359 | found: |
1da177e4 LT |
360 | /* We found where to put this one. Check for overlap with |
361 | * preceding fragment, and, if needed, align things so that | |
362 | * any overlaps are eliminated. | |
363 | */ | |
364 | if (prev) { | |
bf663371 | 365 | int i = (prev->ip_defrag_offset + prev->len) - offset; |
1da177e4 LT |
366 | |
367 | if (i > 0) { | |
368 | offset += i; | |
1706d587 | 369 | err = -EINVAL; |
1da177e4 LT |
370 | if (end <= offset) |
371 | goto err; | |
1706d587 | 372 | err = -ENOMEM; |
1da177e4 LT |
373 | if (!pskb_pull(skb, i)) |
374 | goto err; | |
375 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
376 | skb->ip_summed = CHECKSUM_NONE; | |
377 | } | |
378 | } | |
379 | ||
1706d587 HX |
380 | err = -ENOMEM; |
381 | ||
bf663371 ED |
382 | while (next && next->ip_defrag_offset < end) { |
383 | int i = end - next->ip_defrag_offset; /* overlap is 'i' bytes */ | |
1da177e4 LT |
384 | |
385 | if (i < next->len) { | |
386 | /* Eat head of the next overlapped fragment | |
387 | * and leave the loop. The next ones cannot overlap. | |
388 | */ | |
389 | if (!pskb_pull(next, i)) | |
390 | goto err; | |
bf663371 | 391 | next->ip_defrag_offset += i; |
5ab11c98 | 392 | qp->q.meat -= i; |
1da177e4 LT |
393 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
394 | next->ip_summed = CHECKSUM_NONE; | |
395 | break; | |
396 | } else { | |
397 | struct sk_buff *free_it = next; | |
398 | ||
47c6bf77 | 399 | /* Old fragment is completely overridden with |
1da177e4 LT |
400 | * new one drop it. |
401 | */ | |
402 | next = next->next; | |
403 | ||
404 | if (prev) | |
405 | prev->next = next; | |
406 | else | |
5ab11c98 | 407 | qp->q.fragments = next; |
1da177e4 | 408 | |
5ab11c98 | 409 | qp->q.meat -= free_it->len; |
0e60d245 | 410 | sub_frag_mem_limit(qp->q.net, free_it->truesize); |
d433673e | 411 | kfree_skb(free_it); |
1da177e4 LT |
412 | } |
413 | } | |
414 | ||
bf663371 ED |
415 | /* Note : skb->ip_defrag_offset and skb->dev share the same location */ |
416 | dev = skb->dev; | |
417 | if (dev) | |
418 | qp->iif = dev->ifindex; | |
419 | /* Makes sure compiler wont do silly aliasing games */ | |
420 | barrier(); | |
421 | skb->ip_defrag_offset = offset; | |
1da177e4 LT |
422 | |
423 | /* Insert this fragment in the chain of fragments. */ | |
424 | skb->next = next; | |
d6bebca9 CG |
425 | if (!next) |
426 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
427 | if (prev) |
428 | prev->next = skb; | |
429 | else | |
5ab11c98 | 430 | qp->q.fragments = skb; |
1da177e4 | 431 | |
5ab11c98 PE |
432 | qp->q.stamp = skb->tstamp; |
433 | qp->q.meat += skb->len; | |
6623e3b2 | 434 | qp->ecn |= ecn; |
0e60d245 | 435 | add_frag_mem_limit(qp->q.net, skb->truesize); |
1da177e4 | 436 | if (offset == 0) |
06aa8b8a | 437 | qp->q.flags |= INET_FRAG_FIRST_IN; |
1da177e4 | 438 | |
d6b915e2 FW |
439 | fragsize = skb->len + ihl; |
440 | ||
441 | if (fragsize > qp->q.max_size) | |
442 | qp->q.max_size = fragsize; | |
443 | ||
5f2d04f1 | 444 | if (ip_hdr(skb)->frag_off & htons(IP_DF) && |
d6b915e2 FW |
445 | fragsize > qp->max_df_size) |
446 | qp->max_df_size = fragsize; | |
5f2d04f1 | 447 | |
06aa8b8a | 448 | if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
97599dc7 ED |
449 | qp->q.meat == qp->q.len) { |
450 | unsigned long orefdst = skb->_skb_refdst; | |
1706d587 | 451 | |
97599dc7 ED |
452 | skb->_skb_refdst = 0UL; |
453 | err = ip_frag_reasm(qp, prev, dev); | |
454 | skb->_skb_refdst = orefdst; | |
455 | return err; | |
456 | } | |
457 | ||
458 | skb_dst_drop(skb); | |
1706d587 | 459 | return -EINPROGRESS; |
1da177e4 LT |
460 | |
461 | err: | |
462 | kfree_skb(skb); | |
1706d587 | 463 | return err; |
1da177e4 LT |
464 | } |
465 | ||
466 | ||
467 | /* Build a new IP datagram from all its fragments. */ | |
468 | ||
1706d587 HX |
469 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
470 | struct net_device *dev) | |
1da177e4 | 471 | { |
2bad35b7 | 472 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 473 | struct iphdr *iph; |
5ab11c98 | 474 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
475 | int len; |
476 | int ihlen; | |
1706d587 | 477 | int err; |
5173cc05 | 478 | u8 ecn; |
1da177e4 LT |
479 | |
480 | ipq_kill(qp); | |
481 | ||
be991971 | 482 | ecn = ip_frag_ecn_table[qp->ecn]; |
5173cc05 ED |
483 | if (unlikely(ecn == 0xff)) { |
484 | err = -EINVAL; | |
485 | goto out_fail; | |
486 | } | |
1706d587 HX |
487 | /* Make the one we just received the head. */ |
488 | if (prev) { | |
489 | head = prev->next; | |
490 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
491 | if (!fp) |
492 | goto out_nomem; | |
493 | ||
494 | fp->next = head->next; | |
d6bebca9 CG |
495 | if (!fp->next) |
496 | qp->q.fragments_tail = fp; | |
1706d587 HX |
497 | prev->next = fp; |
498 | ||
5ab11c98 PE |
499 | skb_morph(head, qp->q.fragments); |
500 | head->next = qp->q.fragments->next; | |
1706d587 | 501 | |
cbf8f7bb | 502 | consume_skb(qp->q.fragments); |
5ab11c98 | 503 | qp->q.fragments = head; |
1706d587 HX |
504 | } |
505 | ||
51456b29 | 506 | WARN_ON(!head); |
bf663371 | 507 | WARN_ON(head->ip_defrag_offset != 0); |
1da177e4 LT |
508 | |
509 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 510 | ihlen = ip_hdrlen(head); |
5ab11c98 | 511 | len = ihlen + qp->q.len; |
1da177e4 | 512 | |
1706d587 | 513 | err = -E2BIG; |
132adf54 | 514 | if (len > 65535) |
1da177e4 LT |
515 | goto out_oversize; |
516 | ||
517 | /* Head of list must not be cloned. */ | |
14bbd6a5 | 518 | if (skb_unclone(head, GFP_ATOMIC)) |
1da177e4 LT |
519 | goto out_nomem; |
520 | ||
521 | /* If the first fragment is fragmented itself, we split | |
522 | * it to two chunks: the first with data and paged part | |
523 | * and the second, holding only fragments. */ | |
21dc3301 | 524 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
525 | struct sk_buff *clone; |
526 | int i, plen = 0; | |
527 | ||
51456b29 IM |
528 | clone = alloc_skb(0, GFP_ATOMIC); |
529 | if (!clone) | |
1da177e4 LT |
530 | goto out_nomem; |
531 | clone->next = head->next; | |
532 | head->next = clone; | |
533 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 534 | skb_frag_list_init(head); |
9e903e08 ED |
535 | for (i = 0; i < skb_shinfo(head)->nr_frags; i++) |
536 | plen += skb_frag_size(&skb_shinfo(head)->frags[i]); | |
1da177e4 LT |
537 | clone->len = clone->data_len = head->data_len - plen; |
538 | head->data_len -= clone->len; | |
539 | head->len -= clone->len; | |
540 | clone->csum = 0; | |
541 | clone->ip_summed = head->ip_summed; | |
0e60d245 | 542 | add_frag_mem_limit(qp->q.net, clone->truesize); |
1da177e4 LT |
543 | } |
544 | ||
14fe22e3 | 545 | skb_shinfo(head)->frag_list = head->next; |
d56f90a7 | 546 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 | 547 | |
14fe22e3 FW |
548 | for (fp=head->next; fp; fp = fp->next) { |
549 | head->data_len += fp->len; | |
550 | head->len += fp->len; | |
1da177e4 LT |
551 | if (head->ip_summed != fp->ip_summed) |
552 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 553 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 | 554 | head->csum = csum_add(head->csum, fp->csum); |
14fe22e3 | 555 | head->truesize += fp->truesize; |
1da177e4 | 556 | } |
5510b3c2 | 557 | sub_frag_mem_limit(qp->q.net, head->truesize); |
1da177e4 LT |
558 | |
559 | head->next = NULL; | |
560 | head->dev = dev; | |
5ab11c98 | 561 | head->tstamp = qp->q.stamp; |
d6b915e2 | 562 | IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size); |
1da177e4 | 563 | |
eddc9ec5 | 564 | iph = ip_hdr(head); |
1da177e4 | 565 | iph->tot_len = htons(len); |
5173cc05 | 566 | iph->tos |= ecn; |
d6b915e2 FW |
567 | |
568 | /* When we set IP_DF on a refragmented skb we must also force a | |
569 | * call to ip_fragment to avoid forwarding a DF-skb of size s while | |
570 | * original sender only sent fragments of size f (where f < s). | |
571 | * | |
572 | * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest | |
573 | * frag seen to avoid sending tiny DF-fragments in case skb was built | |
574 | * from one very small df-fragment and one large non-df frag. | |
575 | */ | |
576 | if (qp->max_df_size == qp->q.max_size) { | |
577 | IPCB(head)->flags |= IPSKB_FRAG_PMTU; | |
578 | iph->frag_off = htons(IP_DF); | |
579 | } else { | |
580 | iph->frag_off = 0; | |
581 | } | |
582 | ||
0848f642 EHJ |
583 | ip_send_check(iph); |
584 | ||
b45386ef | 585 | __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 586 | qp->q.fragments = NULL; |
d6bebca9 | 587 | qp->q.fragments_tail = NULL; |
1706d587 | 588 | return 0; |
1da177e4 LT |
589 | |
590 | out_nomem: | |
ba7a46f1 | 591 | net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp); |
45542479 | 592 | err = -ENOMEM; |
1da177e4 LT |
593 | goto out_fail; |
594 | out_oversize: | |
648700f7 | 595 | net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr); |
1da177e4 | 596 | out_fail: |
b45386ef | 597 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 598 | return err; |
1da177e4 LT |
599 | } |
600 | ||
601 | /* Process an incoming IP datagram fragment. */ | |
19bcf9f2 | 602 | int ip_defrag(struct net *net, struct sk_buff *skb, u32 user) |
1da177e4 | 603 | { |
9972f134 | 604 | struct net_device *dev = skb->dev ? : skb_dst(skb)->dev; |
385add90 | 605 | int vif = l3mdev_master_ifindex_rcu(dev); |
1da177e4 | 606 | struct ipq *qp; |
e905a9ed | 607 | |
b45386ef | 608 | __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS); |
8282f274 | 609 | skb_orphan(skb); |
1da177e4 | 610 | |
1da177e4 | 611 | /* Lookup (or create) queue header */ |
9972f134 | 612 | qp = ip_find(net, ip_hdr(skb), user, vif); |
00db4124 | 613 | if (qp) { |
1706d587 | 614 | int ret; |
1da177e4 | 615 | |
5ab11c98 | 616 | spin_lock(&qp->q.lock); |
1da177e4 | 617 | |
1706d587 | 618 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 619 | |
5ab11c98 | 620 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 621 | ipq_put(qp); |
776c729e | 622 | return ret; |
1da177e4 LT |
623 | } |
624 | ||
b45386ef | 625 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 626 | kfree_skb(skb); |
776c729e | 627 | return -ENOMEM; |
1da177e4 | 628 | } |
4bc2f18b | 629 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 630 | |
19bcf9f2 | 631 | struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user) |
bc416d97 | 632 | { |
1bf3751e | 633 | struct iphdr iph; |
3e32e733 | 634 | int netoff; |
bc416d97 ED |
635 | u32 len; |
636 | ||
637 | if (skb->protocol != htons(ETH_P_IP)) | |
638 | return skb; | |
639 | ||
3e32e733 AD |
640 | netoff = skb_network_offset(skb); |
641 | ||
642 | if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0) | |
bc416d97 ED |
643 | return skb; |
644 | ||
1bf3751e | 645 | if (iph.ihl < 5 || iph.version != 4) |
bc416d97 | 646 | return skb; |
1bf3751e JB |
647 | |
648 | len = ntohs(iph.tot_len); | |
3e32e733 | 649 | if (skb->len < netoff + len || len < (iph.ihl * 4)) |
bc416d97 ED |
650 | return skb; |
651 | ||
1bf3751e | 652 | if (ip_is_fragment(&iph)) { |
bc416d97 ED |
653 | skb = skb_share_check(skb, GFP_ATOMIC); |
654 | if (skb) { | |
3e32e733 | 655 | if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) |
1bf3751e | 656 | return skb; |
3e32e733 | 657 | if (pskb_trim_rcsum(skb, netoff + len)) |
bc416d97 ED |
658 | return skb; |
659 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
19bcf9f2 | 660 | if (ip_defrag(net, skb, user)) |
bc416d97 | 661 | return NULL; |
7539fadc | 662 | skb_clear_hash(skb); |
bc416d97 ED |
663 | } |
664 | } | |
665 | return skb; | |
666 | } | |
667 | EXPORT_SYMBOL(ip_check_defrag); | |
668 | ||
8d8354d2 | 669 | #ifdef CONFIG_SYSCTL |
3d234012 | 670 | static int dist_min; |
8d8354d2 | 671 | |
0a64b4b8 | 672 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 673 | { |
8d8354d2 | 674 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 675 | .data = &init_net.ipv4.frags.high_thresh, |
3e67f106 | 676 | .maxlen = sizeof(unsigned long), |
8d8354d2 | 677 | .mode = 0644, |
3e67f106 | 678 | .proc_handler = proc_doulongvec_minmax, |
1bab4c75 | 679 | .extra1 = &init_net.ipv4.frags.low_thresh |
8d8354d2 PE |
680 | }, |
681 | { | |
8d8354d2 | 682 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 683 | .data = &init_net.ipv4.frags.low_thresh, |
3e67f106 | 684 | .maxlen = sizeof(unsigned long), |
8d8354d2 | 685 | .mode = 0644, |
3e67f106 | 686 | .proc_handler = proc_doulongvec_minmax, |
1bab4c75 | 687 | .extra2 = &init_net.ipv4.frags.high_thresh |
8d8354d2 PE |
688 | }, |
689 | { | |
8d8354d2 | 690 | .procname = "ipfrag_time", |
b2fd5321 | 691 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
692 | .maxlen = sizeof(int), |
693 | .mode = 0644, | |
6d9f239a | 694 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 695 | }, |
0fbf4cb2 NB |
696 | { |
697 | .procname = "ipfrag_max_dist", | |
698 | .data = &init_net.ipv4.frags.max_dist, | |
699 | .maxlen = sizeof(int), | |
700 | .mode = 0644, | |
701 | .proc_handler = proc_dointvec_minmax, | |
3d234012 | 702 | .extra1 = &dist_min, |
0fbf4cb2 | 703 | }, |
7d291ebb PE |
704 | { } |
705 | }; | |
706 | ||
e3a57d18 FW |
707 | /* secret interval has been deprecated */ |
708 | static int ip4_frags_secret_interval_unused; | |
7d291ebb | 709 | static struct ctl_table ip4_frags_ctl_table[] = { |
8d8354d2 | 710 | { |
8d8354d2 | 711 | .procname = "ipfrag_secret_interval", |
e3a57d18 | 712 | .data = &ip4_frags_secret_interval_unused, |
8d8354d2 PE |
713 | .maxlen = sizeof(int), |
714 | .mode = 0644, | |
6d9f239a | 715 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 716 | }, |
8d8354d2 PE |
717 | { } |
718 | }; | |
719 | ||
2c8c1e72 | 720 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 721 | { |
e4a2d5c2 | 722 | struct ctl_table *table; |
8d8354d2 PE |
723 | struct ctl_table_header *hdr; |
724 | ||
0a64b4b8 | 725 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 726 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 727 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
51456b29 | 728 | if (!table) |
e4a2d5c2 PE |
729 | goto err_alloc; |
730 | ||
e31e0bdc | 731 | table[0].data = &net->ipv4.frags.high_thresh; |
1bab4c75 NA |
732 | table[0].extra1 = &net->ipv4.frags.low_thresh; |
733 | table[0].extra2 = &init_net.ipv4.frags.high_thresh; | |
e31e0bdc | 734 | table[1].data = &net->ipv4.frags.low_thresh; |
1bab4c75 | 735 | table[1].extra2 = &net->ipv4.frags.high_thresh; |
b2fd5321 | 736 | table[2].data = &net->ipv4.frags.timeout; |
0fbf4cb2 | 737 | table[3].data = &net->ipv4.frags.max_dist; |
e4a2d5c2 PE |
738 | } |
739 | ||
ec8f23ce | 740 | hdr = register_net_sysctl(net, "net/ipv4", table); |
51456b29 | 741 | if (!hdr) |
e4a2d5c2 PE |
742 | goto err_reg; |
743 | ||
744 | net->ipv4.frags_hdr = hdr; | |
745 | return 0; | |
746 | ||
747 | err_reg: | |
09ad9bc7 | 748 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
749 | kfree(table); |
750 | err_alloc: | |
751 | return -ENOMEM; | |
752 | } | |
753 | ||
2c8c1e72 | 754 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
755 | { |
756 | struct ctl_table *table; | |
757 | ||
758 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
759 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
760 | kfree(table); | |
8d8354d2 | 761 | } |
7d291ebb | 762 | |
57a02c39 | 763 | static void __init ip4_frags_ctl_register(void) |
7d291ebb | 764 | { |
43444757 | 765 | register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); |
7d291ebb | 766 | } |
8d8354d2 | 767 | #else |
aa1f731e | 768 | static int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
769 | { |
770 | return 0; | |
771 | } | |
e4a2d5c2 | 772 | |
aa1f731e | 773 | static void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
774 | { |
775 | } | |
7d291ebb | 776 | |
aa1f731e | 777 | static void __init ip4_frags_ctl_register(void) |
7d291ebb PE |
778 | { |
779 | } | |
8d8354d2 PE |
780 | #endif |
781 | ||
2c8c1e72 | 782 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 783 | { |
787bea77 ED |
784 | int res; |
785 | ||
c2a93660 JDB |
786 | /* Fragment cache limits. |
787 | * | |
788 | * The fragment memory accounting code, (tries to) account for | |
789 | * the real memory usage, by measuring both the size of frag | |
790 | * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) | |
791 | * and the SKB's truesize. | |
792 | * | |
793 | * A 64K fragment consumes 129736 bytes (44*2944)+200 | |
794 | * (1500 truesize == 2944, sizeof(struct ipq) == 200) | |
795 | * | |
796 | * We will commit 4MB at one time. Should we cross that limit | |
797 | * we will prune down to 3MB, making room for approx 8 big 64K | |
798 | * fragments 8x128k. | |
e31e0bdc | 799 | */ |
c2a93660 JDB |
800 | net->ipv4.frags.high_thresh = 4 * 1024 * 1024; |
801 | net->ipv4.frags.low_thresh = 3 * 1024 * 1024; | |
b2fd5321 PE |
802 | /* |
803 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
804 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
805 | * by TTL. | |
806 | */ | |
807 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
808 | ||
0fbf4cb2 | 809 | net->ipv4.frags.max_dist = 64; |
093ba729 | 810 | net->ipv4.frags.f = &ip4_frags; |
0fbf4cb2 | 811 | |
787bea77 ED |
812 | res = inet_frags_init_net(&net->ipv4.frags); |
813 | if (res < 0) | |
814 | return res; | |
815 | res = ip4_frags_ns_ctl_register(net); | |
816 | if (res < 0) | |
093ba729 | 817 | inet_frags_exit_net(&net->ipv4.frags); |
787bea77 | 818 | return res; |
8d8354d2 PE |
819 | } |
820 | ||
2c8c1e72 | 821 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 822 | { |
0a64b4b8 | 823 | ip4_frags_ns_ctl_unregister(net); |
093ba729 | 824 | inet_frags_exit_net(&net->ipv4.frags); |
81566e83 PE |
825 | } |
826 | ||
827 | static struct pernet_operations ip4_frags_ops = { | |
828 | .init = ipv4_frags_init_net, | |
829 | .exit = ipv4_frags_exit_net, | |
830 | }; | |
831 | ||
648700f7 ED |
832 | |
833 | static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed) | |
834 | { | |
835 | return jhash2(data, | |
836 | sizeof(struct frag_v4_compare_key) / sizeof(u32), seed); | |
837 | } | |
838 | ||
839 | static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed) | |
840 | { | |
841 | const struct inet_frag_queue *fq = data; | |
842 | ||
843 | return jhash2((const u32 *)&fq->key.v4, | |
844 | sizeof(struct frag_v4_compare_key) / sizeof(u32), seed); | |
845 | } | |
846 | ||
847 | static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr) | |
848 | { | |
849 | const struct frag_v4_compare_key *key = arg->key; | |
850 | const struct inet_frag_queue *fq = ptr; | |
851 | ||
852 | return !!memcmp(&fq->key, key, sizeof(*key)); | |
853 | } | |
854 | ||
855 | static const struct rhashtable_params ip4_rhash_params = { | |
856 | .head_offset = offsetof(struct inet_frag_queue, node), | |
857 | .key_offset = offsetof(struct inet_frag_queue, key), | |
858 | .key_len = sizeof(struct frag_v4_compare_key), | |
859 | .hashfn = ip4_key_hashfn, | |
860 | .obj_hashfn = ip4_obj_hashfn, | |
861 | .obj_cmpfn = ip4_obj_cmpfn, | |
862 | .automatic_shrinking = true, | |
863 | }; | |
864 | ||
b7aa0bf7 | 865 | void __init ipfrag_init(void) |
1da177e4 | 866 | { |
c6fda282 | 867 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 | 868 | ip4_frags.destructor = ip4_frag_free; |
1e4b8287 | 869 | ip4_frags.qsize = sizeof(struct ipq); |
e521db9d | 870 | ip4_frags.frag_expire = ip_expire; |
d4ad4d22 | 871 | ip4_frags.frags_cache_name = ip_frag_cache_name; |
648700f7 | 872 | ip4_frags.rhash_params = ip4_rhash_params; |
d4ad4d22 NA |
873 | if (inet_frags_init(&ip4_frags)) |
874 | panic("IP: failed to allocate ip4_frags cache\n"); | |
483a6e4f ED |
875 | ip4_frags_ctl_register(); |
876 | register_pernet_subsys(&ip4_frags_ops); | |
1da177e4 | 877 | } |