]>
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 | |
a4fd284a PO |
78 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb, |
79 | struct sk_buff *prev_tail, struct net_device *dev); | |
1706d587 | 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); | |
a39aca67 | 85 | struct net *net = q->fqdir->net; |
54db0cc2 | 86 | |
648700f7 | 87 | const struct frag_v4_compare_key *key = a; |
c6fda282 | 88 | |
648700f7 ED |
89 | q->key.v4 = *key; |
90 | qp->ecn = 0; | |
6ce3b4dc | 91 | qp->peer = q->fqdir->max_dist ? |
648700f7 | 92 | inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) : |
192132b9 | 93 | NULL; |
c6fda282 PE |
94 | } |
95 | ||
aa1f731e | 96 | static void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 97 | { |
1e4b8287 PE |
98 | struct ipq *qp; |
99 | ||
100 | qp = container_of(q, struct ipq, q); | |
101 | if (qp->peer) | |
102 | inet_putpeer(qp->peer); | |
1da177e4 LT |
103 | } |
104 | ||
1da177e4 LT |
105 | |
106 | /* Destruction primitives. */ | |
107 | ||
aa1f731e | 108 | static void ipq_put(struct ipq *ipq) |
1da177e4 | 109 | { |
093ba729 | 110 | inet_frag_put(&ipq->q); |
1da177e4 LT |
111 | } |
112 | ||
113 | /* Kill ipq entry. It is not destroyed immediately, | |
114 | * because caller (and someone more) holds reference count. | |
115 | */ | |
116 | static void ipq_kill(struct ipq *ipq) | |
117 | { | |
093ba729 | 118 | inet_frag_kill(&ipq->q); |
1da177e4 LT |
119 | } |
120 | ||
5cf42280 AZ |
121 | static bool frag_expire_skip_icmp(u32 user) |
122 | { | |
123 | return user == IP_DEFRAG_AF_PACKET || | |
124 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN, | |
8bc04864 AZ |
125 | __IP_DEFRAG_CONNTRACK_IN_END) || |
126 | ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN, | |
127 | __IP_DEFRAG_CONNTRACK_BRIDGE_IN); | |
5cf42280 AZ |
128 | } |
129 | ||
1da177e4 LT |
130 | /* |
131 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
132 | */ | |
78802011 | 133 | static void ip_expire(struct timer_list *t) |
1da177e4 | 134 | { |
78802011 | 135 | struct inet_frag_queue *frag = from_timer(frag, t, timer); |
399d1404 | 136 | const struct iphdr *iph; |
fa0f5273 | 137 | struct sk_buff *head = NULL; |
84a3aa00 | 138 | struct net *net; |
399d1404 ED |
139 | struct ipq *qp; |
140 | int err; | |
e521db9d | 141 | |
78802011 | 142 | qp = container_of(frag, struct ipq, q); |
a39aca67 | 143 | net = qp->q.fqdir->net; |
1da177e4 | 144 | |
ec4fbd64 | 145 | rcu_read_lock(); |
d5dd8879 ED |
146 | |
147 | if (qp->q.fqdir->dead) | |
148 | goto out_rcu_unlock; | |
149 | ||
5ab11c98 | 150 | spin_lock(&qp->q.lock); |
1da177e4 | 151 | |
06aa8b8a | 152 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
153 | goto out; |
154 | ||
155 | ipq_kill(qp); | |
b45386ef | 156 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
399d1404 | 157 | __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT); |
2e404f63 | 158 | |
70837ffe | 159 | if (!(qp->q.flags & INET_FRAG_FIRST_IN)) |
399d1404 | 160 | goto out; |
2e404f63 | 161 | |
fa0f5273 PO |
162 | /* sk_buff::dev and sk_buff::rbnode are unionized. So we |
163 | * pull the head out of the tree in order to be able to | |
164 | * deal with head->dev. | |
165 | */ | |
c23f35d1 PO |
166 | head = inet_frag_pull_head(&qp->q); |
167 | if (!head) | |
168 | goto out; | |
399d1404 ED |
169 | head->dev = dev_get_by_index_rcu(net, qp->iif); |
170 | if (!head->dev) | |
171 | goto out; | |
ec4fbd64 | 172 | |
e9017b55 | 173 | |
399d1404 ED |
174 | /* skb has no dst, perform route lookup again */ |
175 | iph = ip_hdr(head); | |
176 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, | |
c6cffba4 | 177 | iph->tos, head->dev); |
399d1404 ED |
178 | if (err) |
179 | goto out; | |
180 | ||
181 | /* Only an end host needs to send an ICMP | |
182 | * "Fragment Reassembly Timeout" message, per RFC792. | |
183 | */ | |
184 | if (frag_expire_skip_icmp(qp->q.key.v4.user) && | |
185 | (skb_rtable(head)->rt_type != RTN_LOCAL)) | |
186 | goto out; | |
187 | ||
1eec5d56 ED |
188 | spin_unlock(&qp->q.lock); |
189 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
1eec5d56 | 190 | goto out_rcu_unlock; |
399d1404 | 191 | |
1da177e4 | 192 | out: |
5ab11c98 | 193 | spin_unlock(&qp->q.lock); |
ec4fbd64 ED |
194 | out_rcu_unlock: |
195 | rcu_read_unlock(); | |
1d08962f | 196 | kfree_skb(head); |
4b6cb5d8 | 197 | ipq_put(qp); |
1da177e4 LT |
198 | } |
199 | ||
abd6523d PE |
200 | /* Find the correct entry in the "incomplete datagrams" queue for |
201 | * this IP datagram, and create new one, if nothing is found. | |
202 | */ | |
9972f134 DA |
203 | static struct ipq *ip_find(struct net *net, struct iphdr *iph, |
204 | u32 user, int vif) | |
1da177e4 | 205 | { |
648700f7 ED |
206 | struct frag_v4_compare_key key = { |
207 | .saddr = iph->saddr, | |
208 | .daddr = iph->daddr, | |
209 | .user = user, | |
210 | .vif = vif, | |
211 | .id = iph->id, | |
212 | .protocol = iph->protocol, | |
213 | }; | |
c6fda282 | 214 | struct inet_frag_queue *q; |
9a375803 | 215 | |
4907abc6 | 216 | q = inet_frag_find(net->ipv4.fqdir, &key); |
2d44ed22 | 217 | if (!q) |
5a3da1fe | 218 | return NULL; |
2d44ed22 | 219 | |
c6fda282 | 220 | return container_of(q, struct ipq, q); |
1da177e4 LT |
221 | } |
222 | ||
89cee8b1 | 223 | /* Is the fragment too far ahead to be part of ipq? */ |
aa1f731e | 224 | static int ip_frag_too_far(struct ipq *qp) |
89cee8b1 HX |
225 | { |
226 | struct inet_peer *peer = qp->peer; | |
6ce3b4dc | 227 | unsigned int max = qp->q.fqdir->max_dist; |
89cee8b1 HX |
228 | unsigned int start, end; |
229 | ||
230 | int rc; | |
231 | ||
232 | if (!peer || !max) | |
233 | return 0; | |
234 | ||
235 | start = qp->rid; | |
236 | end = atomic_inc_return(&peer->rid); | |
237 | qp->rid = end; | |
238 | ||
fa0f5273 | 239 | rc = qp->q.fragments_tail && (end - start) > max; |
89cee8b1 | 240 | |
a39aca67 ED |
241 | if (rc) |
242 | __IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
243 | |
244 | return rc; | |
245 | } | |
246 | ||
247 | static int ip_frag_reinit(struct ipq *qp) | |
248 | { | |
d433673e | 249 | unsigned int sum_truesize = 0; |
89cee8b1 | 250 | |
6ce3b4dc | 251 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) { |
edcb6918 | 252 | refcount_inc(&qp->q.refcnt); |
89cee8b1 HX |
253 | return -ETIMEDOUT; |
254 | } | |
255 | ||
a4fd284a | 256 | sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments); |
6ce3b4dc | 257 | sub_frag_mem_limit(qp->q.fqdir, sum_truesize); |
89cee8b1 | 258 | |
06aa8b8a | 259 | qp->q.flags = 0; |
5ab11c98 PE |
260 | qp->q.len = 0; |
261 | qp->q.meat = 0; | |
fa0f5273 | 262 | qp->q.rb_fragments = RB_ROOT; |
d6bebca9 | 263 | qp->q.fragments_tail = NULL; |
a4fd284a | 264 | qp->q.last_run_head = NULL; |
89cee8b1 | 265 | qp->iif = 0; |
6623e3b2 | 266 | qp->ecn = 0; |
89cee8b1 HX |
267 | |
268 | return 0; | |
269 | } | |
270 | ||
1da177e4 | 271 | /* Add new segment to existing queue. */ |
1706d587 | 272 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 | 273 | { |
a39aca67 | 274 | struct net *net = qp->q.fqdir->net; |
c23f35d1 PO |
275 | int ihl, end, flags, offset; |
276 | struct sk_buff *prev_tail; | |
1706d587 | 277 | struct net_device *dev; |
d6b915e2 | 278 | unsigned int fragsize; |
1706d587 | 279 | int err = -ENOENT; |
6623e3b2 | 280 | u8 ecn; |
1da177e4 | 281 | |
06aa8b8a | 282 | if (qp->q.flags & INET_FRAG_COMPLETE) |
1da177e4 LT |
283 | goto err; |
284 | ||
89cee8b1 | 285 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
286 | unlikely(ip_frag_too_far(qp)) && |
287 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
288 | ipq_kill(qp); |
289 | goto err; | |
290 | } | |
291 | ||
6623e3b2 | 292 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 293 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
294 | flags = offset & ~IP_OFFSET; |
295 | offset &= IP_OFFSET; | |
296 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 297 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
298 | |
299 | /* Determine the position of this fragment. */ | |
0848f642 | 300 | end = offset + skb->len - skb_network_offset(skb) - ihl; |
1706d587 | 301 | err = -EINVAL; |
1da177e4 LT |
302 | |
303 | /* Is this the final fragment? */ | |
304 | if ((flags & IP_MF) == 0) { | |
305 | /* If we already have some bits beyond end | |
42b2aa86 | 306 | * or have different end, the segment is corrupted. |
1da177e4 | 307 | */ |
5ab11c98 | 308 | if (end < qp->q.len || |
06aa8b8a | 309 | ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len)) |
0ff89efb | 310 | goto discard_qp; |
06aa8b8a | 311 | qp->q.flags |= INET_FRAG_LAST_IN; |
5ab11c98 | 312 | qp->q.len = end; |
1da177e4 LT |
313 | } else { |
314 | if (end&7) { | |
315 | end &= ~7; | |
316 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
317 | skb->ip_summed = CHECKSUM_NONE; | |
318 | } | |
5ab11c98 | 319 | if (end > qp->q.len) { |
1da177e4 | 320 | /* Some bits beyond end -> corruption. */ |
06aa8b8a | 321 | if (qp->q.flags & INET_FRAG_LAST_IN) |
0ff89efb | 322 | goto discard_qp; |
5ab11c98 | 323 | qp->q.len = end; |
1da177e4 LT |
324 | } |
325 | } | |
326 | if (end == offset) | |
0ff89efb | 327 | goto discard_qp; |
1da177e4 | 328 | |
1706d587 | 329 | err = -ENOMEM; |
0848f642 | 330 | if (!pskb_pull(skb, skb_network_offset(skb) + ihl)) |
0ff89efb | 331 | goto discard_qp; |
1706d587 HX |
332 | |
333 | err = pskb_trim_rcsum(skb, end - offset); | |
334 | if (err) | |
0ff89efb | 335 | goto discard_qp; |
1da177e4 | 336 | |
fa0f5273 PO |
337 | /* Note : skb->rbnode and skb->dev share the same location. */ |
338 | dev = skb->dev; | |
339 | /* Makes sure compiler wont do silly aliasing games */ | |
340 | barrier(); | |
1da177e4 | 341 | |
a4fd284a | 342 | prev_tail = qp->q.fragments_tail; |
c23f35d1 PO |
343 | err = inet_frag_queue_insert(&qp->q, skb, offset, end); |
344 | if (err) | |
345 | goto insert_error; | |
1da177e4 | 346 | |
bf663371 ED |
347 | if (dev) |
348 | qp->iif = dev->ifindex; | |
1da177e4 | 349 | |
5ab11c98 PE |
350 | qp->q.stamp = skb->tstamp; |
351 | qp->q.meat += skb->len; | |
6623e3b2 | 352 | qp->ecn |= ecn; |
6ce3b4dc | 353 | add_frag_mem_limit(qp->q.fqdir, skb->truesize); |
1da177e4 | 354 | if (offset == 0) |
06aa8b8a | 355 | qp->q.flags |= INET_FRAG_FIRST_IN; |
1da177e4 | 356 | |
d6b915e2 FW |
357 | fragsize = skb->len + ihl; |
358 | ||
359 | if (fragsize > qp->q.max_size) | |
360 | qp->q.max_size = fragsize; | |
361 | ||
5f2d04f1 | 362 | if (ip_hdr(skb)->frag_off & htons(IP_DF) && |
d6b915e2 FW |
363 | fragsize > qp->max_df_size) |
364 | qp->max_df_size = fragsize; | |
5f2d04f1 | 365 | |
06aa8b8a | 366 | if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
97599dc7 ED |
367 | qp->q.meat == qp->q.len) { |
368 | unsigned long orefdst = skb->_skb_refdst; | |
1706d587 | 369 | |
97599dc7 | 370 | skb->_skb_refdst = 0UL; |
a4fd284a | 371 | err = ip_frag_reasm(qp, skb, prev_tail, dev); |
97599dc7 | 372 | skb->_skb_refdst = orefdst; |
0ff89efb PO |
373 | if (err) |
374 | inet_frag_kill(&qp->q); | |
97599dc7 ED |
375 | return err; |
376 | } | |
377 | ||
378 | skb_dst_drop(skb); | |
1706d587 | 379 | return -EINPROGRESS; |
1da177e4 | 380 | |
c23f35d1 PO |
381 | insert_error: |
382 | if (err == IPFRAG_DUP) { | |
383 | kfree_skb(skb); | |
384 | return -EINVAL; | |
385 | } | |
386 | err = -EINVAL; | |
0ff89efb | 387 | __IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS); |
7969e5c4 PO |
388 | discard_qp: |
389 | inet_frag_kill(&qp->q); | |
c23f35d1 | 390 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
391 | err: |
392 | kfree_skb(skb); | |
1706d587 | 393 | return err; |
1da177e4 LT |
394 | } |
395 | ||
891584f4 GN |
396 | static bool ip_frag_coalesce_ok(const struct ipq *qp) |
397 | { | |
398 | return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER; | |
399 | } | |
400 | ||
1da177e4 | 401 | /* Build a new IP datagram from all its fragments. */ |
fa0f5273 | 402 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb, |
a4fd284a | 403 | struct sk_buff *prev_tail, struct net_device *dev) |
1da177e4 | 404 | { |
a39aca67 | 405 | struct net *net = qp->q.fqdir->net; |
1da177e4 | 406 | struct iphdr *iph; |
c23f35d1 PO |
407 | void *reasm_data; |
408 | int len, err; | |
5173cc05 | 409 | u8 ecn; |
1da177e4 LT |
410 | |
411 | ipq_kill(qp); | |
412 | ||
be991971 | 413 | ecn = ip_frag_ecn_table[qp->ecn]; |
5173cc05 ED |
414 | if (unlikely(ecn == 0xff)) { |
415 | err = -EINVAL; | |
416 | goto out_fail; | |
417 | } | |
1706d587 | 418 | |
c23f35d1 PO |
419 | /* Make the one we just received the head. */ |
420 | reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail); | |
421 | if (!reasm_data) | |
422 | goto out_nomem; | |
1da177e4 | 423 | |
c23f35d1 | 424 | len = ip_hdrlen(skb) + qp->q.len; |
1706d587 | 425 | err = -E2BIG; |
132adf54 | 426 | if (len > 65535) |
1da177e4 LT |
427 | goto out_oversize; |
428 | ||
891584f4 GN |
429 | inet_frag_reasm_finish(&qp->q, skb, reasm_data, |
430 | ip_frag_coalesce_ok(qp)); | |
1da177e4 | 431 | |
c23f35d1 PO |
432 | skb->dev = dev; |
433 | IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size); | |
1da177e4 | 434 | |
c23f35d1 | 435 | iph = ip_hdr(skb); |
1da177e4 | 436 | iph->tot_len = htons(len); |
5173cc05 | 437 | iph->tos |= ecn; |
d6b915e2 FW |
438 | |
439 | /* When we set IP_DF on a refragmented skb we must also force a | |
440 | * call to ip_fragment to avoid forwarding a DF-skb of size s while | |
441 | * original sender only sent fragments of size f (where f < s). | |
442 | * | |
443 | * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest | |
444 | * frag seen to avoid sending tiny DF-fragments in case skb was built | |
445 | * from one very small df-fragment and one large non-df frag. | |
446 | */ | |
447 | if (qp->max_df_size == qp->q.max_size) { | |
c23f35d1 | 448 | IPCB(skb)->flags |= IPSKB_FRAG_PMTU; |
d6b915e2 FW |
449 | iph->frag_off = htons(IP_DF); |
450 | } else { | |
451 | iph->frag_off = 0; | |
452 | } | |
453 | ||
0848f642 EHJ |
454 | ip_send_check(iph); |
455 | ||
b45386ef | 456 | __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS); |
fa0f5273 | 457 | qp->q.rb_fragments = RB_ROOT; |
d6bebca9 | 458 | qp->q.fragments_tail = NULL; |
a4fd284a | 459 | qp->q.last_run_head = NULL; |
1706d587 | 460 | return 0; |
1da177e4 LT |
461 | |
462 | out_nomem: | |
ba7a46f1 | 463 | net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp); |
45542479 | 464 | err = -ENOMEM; |
1da177e4 LT |
465 | goto out_fail; |
466 | out_oversize: | |
648700f7 | 467 | net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr); |
1da177e4 | 468 | out_fail: |
b45386ef | 469 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 470 | return err; |
1da177e4 LT |
471 | } |
472 | ||
473 | /* Process an incoming IP datagram fragment. */ | |
19bcf9f2 | 474 | int ip_defrag(struct net *net, struct sk_buff *skb, u32 user) |
1da177e4 | 475 | { |
9972f134 | 476 | struct net_device *dev = skb->dev ? : skb_dst(skb)->dev; |
385add90 | 477 | int vif = l3mdev_master_ifindex_rcu(dev); |
1da177e4 | 478 | struct ipq *qp; |
e905a9ed | 479 | |
b45386ef | 480 | __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS); |
8282f274 | 481 | skb_orphan(skb); |
1da177e4 | 482 | |
1da177e4 | 483 | /* Lookup (or create) queue header */ |
9972f134 | 484 | qp = ip_find(net, ip_hdr(skb), user, vif); |
00db4124 | 485 | if (qp) { |
1706d587 | 486 | int ret; |
1da177e4 | 487 | |
5ab11c98 | 488 | spin_lock(&qp->q.lock); |
1da177e4 | 489 | |
1706d587 | 490 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 491 | |
5ab11c98 | 492 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 493 | ipq_put(qp); |
776c729e | 494 | return ret; |
1da177e4 LT |
495 | } |
496 | ||
b45386ef | 497 | __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 498 | kfree_skb(skb); |
776c729e | 499 | return -ENOMEM; |
1da177e4 | 500 | } |
4bc2f18b | 501 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 502 | |
19bcf9f2 | 503 | struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user) |
bc416d97 | 504 | { |
1bf3751e | 505 | struct iphdr iph; |
3e32e733 | 506 | int netoff; |
bc416d97 ED |
507 | u32 len; |
508 | ||
509 | if (skb->protocol != htons(ETH_P_IP)) | |
510 | return skb; | |
511 | ||
3e32e733 AD |
512 | netoff = skb_network_offset(skb); |
513 | ||
514 | if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0) | |
bc416d97 ED |
515 | return skb; |
516 | ||
1bf3751e | 517 | if (iph.ihl < 5 || iph.version != 4) |
bc416d97 | 518 | return skb; |
1bf3751e JB |
519 | |
520 | len = ntohs(iph.tot_len); | |
3e32e733 | 521 | if (skb->len < netoff + len || len < (iph.ihl * 4)) |
bc416d97 ED |
522 | return skb; |
523 | ||
1bf3751e | 524 | if (ip_is_fragment(&iph)) { |
bc416d97 ED |
525 | skb = skb_share_check(skb, GFP_ATOMIC); |
526 | if (skb) { | |
7de414a9 CW |
527 | if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) { |
528 | kfree_skb(skb); | |
529 | return NULL; | |
530 | } | |
531 | if (pskb_trim_rcsum(skb, netoff + len)) { | |
532 | kfree_skb(skb); | |
533 | return NULL; | |
534 | } | |
bc416d97 | 535 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); |
19bcf9f2 | 536 | if (ip_defrag(net, skb, user)) |
bc416d97 | 537 | return NULL; |
7539fadc | 538 | skb_clear_hash(skb); |
bc416d97 ED |
539 | } |
540 | } | |
541 | return skb; | |
542 | } | |
543 | EXPORT_SYMBOL(ip_check_defrag); | |
544 | ||
8d8354d2 | 545 | #ifdef CONFIG_SYSCTL |
3d234012 | 546 | static int dist_min; |
8d8354d2 | 547 | |
0a64b4b8 | 548 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 549 | { |
8d8354d2 | 550 | .procname = "ipfrag_high_thresh", |
3e67f106 | 551 | .maxlen = sizeof(unsigned long), |
8d8354d2 | 552 | .mode = 0644, |
3e67f106 | 553 | .proc_handler = proc_doulongvec_minmax, |
8d8354d2 PE |
554 | }, |
555 | { | |
8d8354d2 | 556 | .procname = "ipfrag_low_thresh", |
3e67f106 | 557 | .maxlen = sizeof(unsigned long), |
8d8354d2 | 558 | .mode = 0644, |
3e67f106 | 559 | .proc_handler = proc_doulongvec_minmax, |
8d8354d2 PE |
560 | }, |
561 | { | |
8d8354d2 | 562 | .procname = "ipfrag_time", |
8d8354d2 PE |
563 | .maxlen = sizeof(int), |
564 | .mode = 0644, | |
6d9f239a | 565 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 566 | }, |
0fbf4cb2 NB |
567 | { |
568 | .procname = "ipfrag_max_dist", | |
0fbf4cb2 NB |
569 | .maxlen = sizeof(int), |
570 | .mode = 0644, | |
571 | .proc_handler = proc_dointvec_minmax, | |
3d234012 | 572 | .extra1 = &dist_min, |
0fbf4cb2 | 573 | }, |
7d291ebb PE |
574 | { } |
575 | }; | |
576 | ||
e3a57d18 FW |
577 | /* secret interval has been deprecated */ |
578 | static int ip4_frags_secret_interval_unused; | |
7d291ebb | 579 | static struct ctl_table ip4_frags_ctl_table[] = { |
8d8354d2 | 580 | { |
8d8354d2 | 581 | .procname = "ipfrag_secret_interval", |
e3a57d18 | 582 | .data = &ip4_frags_secret_interval_unused, |
8d8354d2 PE |
583 | .maxlen = sizeof(int), |
584 | .mode = 0644, | |
6d9f239a | 585 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 586 | }, |
8d8354d2 PE |
587 | { } |
588 | }; | |
589 | ||
2c8c1e72 | 590 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 591 | { |
e4a2d5c2 | 592 | struct ctl_table *table; |
8d8354d2 PE |
593 | struct ctl_table_header *hdr; |
594 | ||
0a64b4b8 | 595 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 596 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 597 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
51456b29 | 598 | if (!table) |
e4a2d5c2 PE |
599 | goto err_alloc; |
600 | ||
e4a2d5c2 | 601 | } |
4907abc6 ED |
602 | table[0].data = &net->ipv4.fqdir->high_thresh; |
603 | table[0].extra1 = &net->ipv4.fqdir->low_thresh; | |
604 | table[1].data = &net->ipv4.fqdir->low_thresh; | |
605 | table[1].extra2 = &net->ipv4.fqdir->high_thresh; | |
606 | table[2].data = &net->ipv4.fqdir->timeout; | |
607 | table[3].data = &net->ipv4.fqdir->max_dist; | |
e4a2d5c2 | 608 | |
ec8f23ce | 609 | hdr = register_net_sysctl(net, "net/ipv4", table); |
51456b29 | 610 | if (!hdr) |
e4a2d5c2 PE |
611 | goto err_reg; |
612 | ||
613 | net->ipv4.frags_hdr = hdr; | |
614 | return 0; | |
615 | ||
616 | err_reg: | |
09ad9bc7 | 617 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
618 | kfree(table); |
619 | err_alloc: | |
620 | return -ENOMEM; | |
621 | } | |
622 | ||
2c8c1e72 | 623 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
624 | { |
625 | struct ctl_table *table; | |
626 | ||
627 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
628 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
629 | kfree(table); | |
8d8354d2 | 630 | } |
7d291ebb | 631 | |
57a02c39 | 632 | static void __init ip4_frags_ctl_register(void) |
7d291ebb | 633 | { |
43444757 | 634 | register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); |
7d291ebb | 635 | } |
8d8354d2 | 636 | #else |
aa1f731e | 637 | static int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
638 | { |
639 | return 0; | |
640 | } | |
e4a2d5c2 | 641 | |
aa1f731e | 642 | static void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
643 | { |
644 | } | |
7d291ebb | 645 | |
aa1f731e | 646 | static void __init ip4_frags_ctl_register(void) |
7d291ebb PE |
647 | { |
648 | } | |
8d8354d2 PE |
649 | #endif |
650 | ||
2c8c1e72 | 651 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 652 | { |
787bea77 ED |
653 | int res; |
654 | ||
4907abc6 ED |
655 | res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net); |
656 | if (res < 0) | |
657 | return res; | |
c2a93660 JDB |
658 | /* Fragment cache limits. |
659 | * | |
660 | * The fragment memory accounting code, (tries to) account for | |
661 | * the real memory usage, by measuring both the size of frag | |
662 | * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) | |
663 | * and the SKB's truesize. | |
664 | * | |
665 | * A 64K fragment consumes 129736 bytes (44*2944)+200 | |
666 | * (1500 truesize == 2944, sizeof(struct ipq) == 200) | |
667 | * | |
668 | * We will commit 4MB at one time. Should we cross that limit | |
669 | * we will prune down to 3MB, making room for approx 8 big 64K | |
670 | * fragments 8x128k. | |
e31e0bdc | 671 | */ |
4907abc6 ED |
672 | net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024; |
673 | net->ipv4.fqdir->low_thresh = 3 * 1024 * 1024; | |
b2fd5321 PE |
674 | /* |
675 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
676 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
677 | * by TTL. | |
678 | */ | |
4907abc6 | 679 | net->ipv4.fqdir->timeout = IP_FRAG_TIME; |
b2fd5321 | 680 | |
4907abc6 | 681 | net->ipv4.fqdir->max_dist = 64; |
0fbf4cb2 | 682 | |
787bea77 ED |
683 | res = ip4_frags_ns_ctl_register(net); |
684 | if (res < 0) | |
4907abc6 | 685 | fqdir_exit(net->ipv4.fqdir); |
787bea77 | 686 | return res; |
8d8354d2 PE |
687 | } |
688 | ||
d5dd8879 ED |
689 | static void __net_exit ipv4_frags_pre_exit_net(struct net *net) |
690 | { | |
691 | fqdir_pre_exit(net->ipv4.fqdir); | |
692 | } | |
693 | ||
2c8c1e72 | 694 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 695 | { |
0a64b4b8 | 696 | ip4_frags_ns_ctl_unregister(net); |
4907abc6 | 697 | fqdir_exit(net->ipv4.fqdir); |
81566e83 PE |
698 | } |
699 | ||
700 | static struct pernet_operations ip4_frags_ops = { | |
d5dd8879 ED |
701 | .init = ipv4_frags_init_net, |
702 | .pre_exit = ipv4_frags_pre_exit_net, | |
703 | .exit = ipv4_frags_exit_net, | |
81566e83 PE |
704 | }; |
705 | ||
648700f7 ED |
706 | |
707 | static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed) | |
708 | { | |
709 | return jhash2(data, | |
710 | sizeof(struct frag_v4_compare_key) / sizeof(u32), seed); | |
711 | } | |
712 | ||
713 | static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed) | |
714 | { | |
715 | const struct inet_frag_queue *fq = data; | |
716 | ||
717 | return jhash2((const u32 *)&fq->key.v4, | |
718 | sizeof(struct frag_v4_compare_key) / sizeof(u32), seed); | |
719 | } | |
720 | ||
721 | static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr) | |
722 | { | |
723 | const struct frag_v4_compare_key *key = arg->key; | |
724 | const struct inet_frag_queue *fq = ptr; | |
725 | ||
726 | return !!memcmp(&fq->key, key, sizeof(*key)); | |
727 | } | |
728 | ||
729 | static const struct rhashtable_params ip4_rhash_params = { | |
730 | .head_offset = offsetof(struct inet_frag_queue, node), | |
731 | .key_offset = offsetof(struct inet_frag_queue, key), | |
732 | .key_len = sizeof(struct frag_v4_compare_key), | |
733 | .hashfn = ip4_key_hashfn, | |
734 | .obj_hashfn = ip4_obj_hashfn, | |
735 | .obj_cmpfn = ip4_obj_cmpfn, | |
736 | .automatic_shrinking = true, | |
737 | }; | |
738 | ||
b7aa0bf7 | 739 | void __init ipfrag_init(void) |
1da177e4 | 740 | { |
c6fda282 | 741 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 | 742 | ip4_frags.destructor = ip4_frag_free; |
1e4b8287 | 743 | ip4_frags.qsize = sizeof(struct ipq); |
e521db9d | 744 | ip4_frags.frag_expire = ip_expire; |
d4ad4d22 | 745 | ip4_frags.frags_cache_name = ip_frag_cache_name; |
648700f7 | 746 | ip4_frags.rhash_params = ip4_rhash_params; |
d4ad4d22 NA |
747 | if (inet_frags_init(&ip4_frags)) |
748 | panic("IP: failed to allocate ip4_frags cache\n"); | |
483a6e4f ED |
749 | ip4_frags_ctl_register(); |
750 | register_pernet_subsys(&ip4_frags_ops); | |
1da177e4 | 751 | } |