Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * The IP fragmentation functionality. | |
e905a9ed | 7 | * |
1da177e4 | 8 | * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG> |
113aa838 | 9 | * Alan Cox <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
10 | * |
11 | * Fixes: | |
12 | * Alan Cox : Split from ip.c , see ip_input.c for history. | |
13 | * David S. Miller : Begin massive cleanup... | |
14 | * Andi Kleen : Add sysctls. | |
15 | * xxxx : Overlapfrag bug. | |
16 | * Ultima : ip_expire() kernel panic. | |
17 | * Bill Hawes : Frag accounting and evictor fixes. | |
18 | * John McDonald : 0 length frag bug. | |
19 | * Alexey Kuznetsov: SMP races, threading, cleanup. | |
20 | * Patrick McHardy : LRU queue of frag heads for evictor. | |
21 | */ | |
22 | ||
afd46503 JP |
23 | #define pr_fmt(fmt) "IPv4: " fmt |
24 | ||
89cee8b1 | 25 | #include <linux/compiler.h> |
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/types.h> | |
28 | #include <linux/mm.h> | |
29 | #include <linux/jiffies.h> | |
30 | #include <linux/skbuff.h> | |
31 | #include <linux/list.h> | |
32 | #include <linux/ip.h> | |
33 | #include <linux/icmp.h> | |
34 | #include <linux/netdevice.h> | |
35 | #include <linux/jhash.h> | |
36 | #include <linux/random.h> | |
5a0e3ad6 | 37 | #include <linux/slab.h> |
e9017b55 SW |
38 | #include <net/route.h> |
39 | #include <net/dst.h> | |
1da177e4 LT |
40 | #include <net/sock.h> |
41 | #include <net/ip.h> | |
42 | #include <net/icmp.h> | |
43 | #include <net/checksum.h> | |
89cee8b1 | 44 | #include <net/inetpeer.h> |
5ab11c98 | 45 | #include <net/inet_frag.h> |
1da177e4 LT |
46 | #include <linux/tcp.h> |
47 | #include <linux/udp.h> | |
48 | #include <linux/inet.h> | |
49 | #include <linux/netfilter_ipv4.h> | |
6623e3b2 | 50 | #include <net/inet_ecn.h> |
1da177e4 LT |
51 | |
52 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 | |
53 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c | |
54 | * as well. Or notify me, at least. --ANK | |
55 | */ | |
56 | ||
8d8354d2 | 57 | static int sysctl_ipfrag_max_dist __read_mostly = 64; |
89cee8b1 | 58 | |
1da177e4 LT |
59 | struct ipfrag_skb_cb |
60 | { | |
61 | struct inet_skb_parm h; | |
62 | int offset; | |
63 | }; | |
64 | ||
fd3f8c4c | 65 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb)) |
1da177e4 LT |
66 | |
67 | /* Describe an entry in the "incomplete datagrams" queue. */ | |
68 | struct ipq { | |
5ab11c98 PE |
69 | struct inet_frag_queue q; |
70 | ||
1da177e4 | 71 | u32 user; |
18277770 AV |
72 | __be32 saddr; |
73 | __be32 daddr; | |
74 | __be16 id; | |
1da177e4 | 75 | u8 protocol; |
6623e3b2 | 76 | u8 ecn; /* RFC3168 support */ |
89cee8b1 HX |
77 | int iif; |
78 | unsigned int rid; | |
79 | struct inet_peer *peer; | |
1da177e4 LT |
80 | }; |
81 | ||
5173cc05 ED |
82 | /* RFC 3168 support : |
83 | * We want to check ECN values of all fragments, do detect invalid combinations. | |
84 | * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value. | |
85 | */ | |
1d1652cb DM |
86 | #define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */ |
87 | #define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */ | |
88 | #define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */ | |
89 | #define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */ | |
6623e3b2 ED |
90 | |
91 | static inline u8 ip4_frag_ecn(u8 tos) | |
92 | { | |
5173cc05 | 93 | return 1 << (tos & INET_ECN_MASK); |
6623e3b2 ED |
94 | } |
95 | ||
5173cc05 ED |
96 | /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements |
97 | * Value : 0xff if frame should be dropped. | |
98 | * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field | |
99 | */ | |
100 | static const u8 ip4_frag_ecn_table[16] = { | |
101 | /* at least one fragment had CE, and others ECT_0 or ECT_1 */ | |
102 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE, | |
103 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
104 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
105 | ||
106 | /* invalid combinations : drop frame */ | |
107 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff, | |
108 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff, | |
109 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff, | |
110 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
111 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff, | |
112 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff, | |
113 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
114 | }; | |
115 | ||
7eb95156 | 116 | static struct inet_frags ip4_frags; |
1da177e4 | 117 | |
e5a2bb84 | 118 | int ip_frag_nqueues(struct net *net) |
7eb95156 | 119 | { |
e5a2bb84 | 120 | return net->ipv4.frags.nqueues; |
7eb95156 | 121 | } |
1da177e4 | 122 | |
6ddc0822 | 123 | int ip_frag_mem(struct net *net) |
7eb95156 | 124 | { |
d433673e | 125 | return sum_frag_mem_limit(&net->ipv4.frags); |
7eb95156 | 126 | } |
1da177e4 | 127 | |
1706d587 HX |
128 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
129 | struct net_device *dev); | |
130 | ||
c6fda282 PE |
131 | struct ip4_create_arg { |
132 | struct iphdr *iph; | |
133 | u32 user; | |
134 | }; | |
135 | ||
18277770 | 136 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 137 | { |
18277770 AV |
138 | return jhash_3words((__force u32)id << 16 | prot, |
139 | (__force u32)saddr, (__force u32)daddr, | |
7eb95156 | 140 | ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
141 | } |
142 | ||
321a3a99 | 143 | static unsigned int ip4_hashfn(struct inet_frag_queue *q) |
1da177e4 | 144 | { |
321a3a99 | 145 | struct ipq *ipq; |
1da177e4 | 146 | |
321a3a99 PE |
147 | ipq = container_of(q, struct ipq, q); |
148 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
149 | } |
150 | ||
cbc264ca | 151 | static bool ip4_frag_match(struct inet_frag_queue *q, void *a) |
abd6523d PE |
152 | { |
153 | struct ipq *qp; | |
154 | struct ip4_create_arg *arg = a; | |
155 | ||
156 | qp = container_of(q, struct ipq, q); | |
a02cec21 | 157 | return qp->id == arg->iph->id && |
cbc264ca ED |
158 | qp->saddr == arg->iph->saddr && |
159 | qp->daddr == arg->iph->daddr && | |
160 | qp->protocol == arg->iph->protocol && | |
161 | qp->user == arg->user; | |
abd6523d PE |
162 | } |
163 | ||
c6fda282 PE |
164 | static void ip4_frag_init(struct inet_frag_queue *q, void *a) |
165 | { | |
166 | struct ipq *qp = container_of(q, struct ipq, q); | |
54db0cc2 G |
167 | struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, |
168 | frags); | |
169 | struct net *net = container_of(ipv4, struct net, ipv4); | |
170 | ||
c6fda282 PE |
171 | struct ip4_create_arg *arg = a; |
172 | ||
173 | qp->protocol = arg->iph->protocol; | |
174 | qp->id = arg->iph->id; | |
6623e3b2 | 175 | qp->ecn = ip4_frag_ecn(arg->iph->tos); |
c6fda282 PE |
176 | qp->saddr = arg->iph->saddr; |
177 | qp->daddr = arg->iph->daddr; | |
178 | qp->user = arg->user; | |
179 | qp->peer = sysctl_ipfrag_max_dist ? | |
c0efc887 | 180 | inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL; |
c6fda282 PE |
181 | } |
182 | ||
1e4b8287 | 183 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 184 | { |
1e4b8287 PE |
185 | struct ipq *qp; |
186 | ||
187 | qp = container_of(q, struct ipq, q); | |
188 | if (qp->peer) | |
189 | inet_putpeer(qp->peer); | |
1da177e4 LT |
190 | } |
191 | ||
1da177e4 LT |
192 | |
193 | /* Destruction primitives. */ | |
194 | ||
4b6cb5d8 | 195 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 196 | { |
762cc408 | 197 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
198 | } |
199 | ||
200 | /* Kill ipq entry. It is not destroyed immediately, | |
201 | * because caller (and someone more) holds reference count. | |
202 | */ | |
203 | static void ipq_kill(struct ipq *ipq) | |
204 | { | |
277e650d | 205 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
206 | } |
207 | ||
e905a9ed | 208 | /* Memory limiting on fragments. Evictor trashes the oldest |
1da177e4 LT |
209 | * fragment queue until we are back under the threshold. |
210 | */ | |
6ddc0822 | 211 | static void ip_evictor(struct net *net) |
1da177e4 | 212 | { |
8e7999c4 PE |
213 | int evicted; |
214 | ||
6b102865 | 215 | evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false); |
8e7999c4 | 216 | if (evicted) |
c5346fe3 | 217 | IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted); |
1da177e4 LT |
218 | } |
219 | ||
220 | /* | |
221 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
222 | */ | |
223 | static void ip_expire(unsigned long arg) | |
224 | { | |
e521db9d | 225 | struct ipq *qp; |
84a3aa00 | 226 | struct net *net; |
e521db9d PE |
227 | |
228 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 229 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 230 | |
5ab11c98 | 231 | spin_lock(&qp->q.lock); |
1da177e4 | 232 | |
bc578a54 | 233 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
234 | goto out; |
235 | ||
236 | ipq_kill(qp); | |
237 | ||
7c73a6fa PE |
238 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); |
239 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
1da177e4 | 240 | |
bc578a54 | 241 | if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) { |
5ab11c98 | 242 | struct sk_buff *head = qp->q.fragments; |
64f3b9e2 ED |
243 | const struct iphdr *iph; |
244 | int err; | |
cb84663e | 245 | |
69df9d59 ED |
246 | rcu_read_lock(); |
247 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
248 | if (!head->dev) |
249 | goto out_rcu_unlock; | |
250 | ||
97599dc7 | 251 | /* skb has no dst, perform route lookup again */ |
64f3b9e2 | 252 | iph = ip_hdr(head); |
c6cffba4 DM |
253 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, |
254 | iph->tos, head->dev); | |
64f3b9e2 ED |
255 | if (err) |
256 | goto out_rcu_unlock; | |
257 | ||
e9017b55 | 258 | /* |
64f3b9e2 ED |
259 | * Only an end host needs to send an ICMP |
260 | * "Fragment Reassembly Timeout" message, per RFC792. | |
e9017b55 | 261 | */ |
595fc71b DM |
262 | if (qp->user == IP_DEFRAG_AF_PACKET || |
263 | (qp->user == IP_DEFRAG_CONNTRACK_IN && | |
264 | skb_rtable(head)->rt_type != RTN_LOCAL)) | |
64f3b9e2 ED |
265 | goto out_rcu_unlock; |
266 | ||
e9017b55 SW |
267 | |
268 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ | |
269 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 270 | out_rcu_unlock: |
d1c9ae6d PM |
271 | rcu_read_unlock(); |
272 | } | |
1da177e4 | 273 | out: |
5ab11c98 | 274 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 275 | ipq_put(qp); |
1da177e4 LT |
276 | } |
277 | ||
abd6523d PE |
278 | /* Find the correct entry in the "incomplete datagrams" queue for |
279 | * this IP datagram, and create new one, if nothing is found. | |
280 | */ | |
ac18e750 | 281 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 282 | { |
c6fda282 PE |
283 | struct inet_frag_queue *q; |
284 | struct ip4_create_arg arg; | |
abd6523d | 285 | unsigned int hash; |
1da177e4 | 286 | |
c6fda282 PE |
287 | arg.iph = iph; |
288 | arg.user = user; | |
9a375803 PE |
289 | |
290 | read_lock(&ip4_frags.lock); | |
abd6523d | 291 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 292 | |
ac18e750 | 293 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
5a3da1fe HFS |
294 | if (IS_ERR_OR_NULL(q)) { |
295 | inet_frag_maybe_warn_overflow(q, pr_fmt()); | |
296 | return NULL; | |
297 | } | |
c6fda282 | 298 | return container_of(q, struct ipq, q); |
1da177e4 LT |
299 | } |
300 | ||
89cee8b1 HX |
301 | /* Is the fragment too far ahead to be part of ipq? */ |
302 | static inline int ip_frag_too_far(struct ipq *qp) | |
303 | { | |
304 | struct inet_peer *peer = qp->peer; | |
305 | unsigned int max = sysctl_ipfrag_max_dist; | |
306 | unsigned int start, end; | |
307 | ||
308 | int rc; | |
309 | ||
310 | if (!peer || !max) | |
311 | return 0; | |
312 | ||
313 | start = qp->rid; | |
314 | end = atomic_inc_return(&peer->rid); | |
315 | qp->rid = end; | |
316 | ||
5ab11c98 | 317 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
318 | |
319 | if (rc) { | |
7c73a6fa PE |
320 | struct net *net; |
321 | ||
322 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
323 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
324 | } |
325 | ||
326 | return rc; | |
327 | } | |
328 | ||
329 | static int ip_frag_reinit(struct ipq *qp) | |
330 | { | |
331 | struct sk_buff *fp; | |
d433673e | 332 | unsigned int sum_truesize = 0; |
89cee8b1 | 333 | |
b2fd5321 | 334 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 335 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
336 | return -ETIMEDOUT; |
337 | } | |
338 | ||
5ab11c98 | 339 | fp = qp->q.fragments; |
89cee8b1 HX |
340 | do { |
341 | struct sk_buff *xp = fp->next; | |
d433673e JDB |
342 | |
343 | sum_truesize += fp->truesize; | |
344 | kfree_skb(fp); | |
89cee8b1 HX |
345 | fp = xp; |
346 | } while (fp); | |
d433673e | 347 | sub_frag_mem_limit(&qp->q, sum_truesize); |
89cee8b1 | 348 | |
5ab11c98 PE |
349 | qp->q.last_in = 0; |
350 | qp->q.len = 0; | |
351 | qp->q.meat = 0; | |
352 | qp->q.fragments = NULL; | |
d6bebca9 | 353 | qp->q.fragments_tail = NULL; |
89cee8b1 | 354 | qp->iif = 0; |
6623e3b2 | 355 | qp->ecn = 0; |
89cee8b1 HX |
356 | |
357 | return 0; | |
358 | } | |
359 | ||
1da177e4 | 360 | /* Add new segment to existing queue. */ |
1706d587 | 361 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
362 | { |
363 | struct sk_buff *prev, *next; | |
1706d587 | 364 | struct net_device *dev; |
1da177e4 LT |
365 | int flags, offset; |
366 | int ihl, end; | |
1706d587 | 367 | int err = -ENOENT; |
6623e3b2 | 368 | u8 ecn; |
1da177e4 | 369 | |
bc578a54 | 370 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
371 | goto err; |
372 | ||
89cee8b1 | 373 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
374 | unlikely(ip_frag_too_far(qp)) && |
375 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
376 | ipq_kill(qp); |
377 | goto err; | |
378 | } | |
379 | ||
6623e3b2 | 380 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 381 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
382 | flags = offset & ~IP_OFFSET; |
383 | offset &= IP_OFFSET; | |
384 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 385 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
386 | |
387 | /* Determine the position of this fragment. */ | |
e905a9ed | 388 | end = offset + skb->len - ihl; |
1706d587 | 389 | err = -EINVAL; |
1da177e4 LT |
390 | |
391 | /* Is this the final fragment? */ | |
392 | if ((flags & IP_MF) == 0) { | |
393 | /* If we already have some bits beyond end | |
42b2aa86 | 394 | * or have different end, the segment is corrupted. |
1da177e4 | 395 | */ |
5ab11c98 | 396 | if (end < qp->q.len || |
bc578a54 | 397 | ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 398 | goto err; |
bc578a54 | 399 | qp->q.last_in |= INET_FRAG_LAST_IN; |
5ab11c98 | 400 | qp->q.len = end; |
1da177e4 LT |
401 | } else { |
402 | if (end&7) { | |
403 | end &= ~7; | |
404 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
405 | skb->ip_summed = CHECKSUM_NONE; | |
406 | } | |
5ab11c98 | 407 | if (end > qp->q.len) { |
1da177e4 | 408 | /* Some bits beyond end -> corruption. */ |
bc578a54 | 409 | if (qp->q.last_in & INET_FRAG_LAST_IN) |
1da177e4 | 410 | goto err; |
5ab11c98 | 411 | qp->q.len = end; |
1da177e4 LT |
412 | } |
413 | } | |
414 | if (end == offset) | |
415 | goto err; | |
416 | ||
1706d587 | 417 | err = -ENOMEM; |
1da177e4 LT |
418 | if (pskb_pull(skb, ihl) == NULL) |
419 | goto err; | |
1706d587 HX |
420 | |
421 | err = pskb_trim_rcsum(skb, end - offset); | |
422 | if (err) | |
1da177e4 LT |
423 | goto err; |
424 | ||
425 | /* Find out which fragments are in front and at the back of us | |
426 | * in the chain of fragments so far. We must know where to put | |
427 | * this fragment, right? | |
428 | */ | |
d6bebca9 CG |
429 | prev = qp->q.fragments_tail; |
430 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
431 | next = NULL; | |
432 | goto found; | |
433 | } | |
1da177e4 | 434 | prev = NULL; |
5ab11c98 | 435 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
436 | if (FRAG_CB(next)->offset >= offset) |
437 | break; /* bingo! */ | |
438 | prev = next; | |
439 | } | |
440 | ||
d6bebca9 | 441 | found: |
1da177e4 LT |
442 | /* We found where to put this one. Check for overlap with |
443 | * preceding fragment, and, if needed, align things so that | |
444 | * any overlaps are eliminated. | |
445 | */ | |
446 | if (prev) { | |
447 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
448 | ||
449 | if (i > 0) { | |
450 | offset += i; | |
1706d587 | 451 | err = -EINVAL; |
1da177e4 LT |
452 | if (end <= offset) |
453 | goto err; | |
1706d587 | 454 | err = -ENOMEM; |
1da177e4 LT |
455 | if (!pskb_pull(skb, i)) |
456 | goto err; | |
457 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
458 | skb->ip_summed = CHECKSUM_NONE; | |
459 | } | |
460 | } | |
461 | ||
1706d587 HX |
462 | err = -ENOMEM; |
463 | ||
1da177e4 LT |
464 | while (next && FRAG_CB(next)->offset < end) { |
465 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
466 | ||
467 | if (i < next->len) { | |
468 | /* Eat head of the next overlapped fragment | |
469 | * and leave the loop. The next ones cannot overlap. | |
470 | */ | |
471 | if (!pskb_pull(next, i)) | |
472 | goto err; | |
473 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 474 | qp->q.meat -= i; |
1da177e4 LT |
475 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
476 | next->ip_summed = CHECKSUM_NONE; | |
477 | break; | |
478 | } else { | |
479 | struct sk_buff *free_it = next; | |
480 | ||
47c6bf77 | 481 | /* Old fragment is completely overridden with |
1da177e4 LT |
482 | * new one drop it. |
483 | */ | |
484 | next = next->next; | |
485 | ||
486 | if (prev) | |
487 | prev->next = next; | |
488 | else | |
5ab11c98 | 489 | qp->q.fragments = next; |
1da177e4 | 490 | |
5ab11c98 | 491 | qp->q.meat -= free_it->len; |
d433673e JDB |
492 | sub_frag_mem_limit(&qp->q, free_it->truesize); |
493 | kfree_skb(free_it); | |
1da177e4 LT |
494 | } |
495 | } | |
496 | ||
497 | FRAG_CB(skb)->offset = offset; | |
498 | ||
499 | /* Insert this fragment in the chain of fragments. */ | |
500 | skb->next = next; | |
d6bebca9 CG |
501 | if (!next) |
502 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
503 | if (prev) |
504 | prev->next = skb; | |
505 | else | |
5ab11c98 | 506 | qp->q.fragments = skb; |
1da177e4 | 507 | |
1706d587 HX |
508 | dev = skb->dev; |
509 | if (dev) { | |
510 | qp->iif = dev->ifindex; | |
511 | skb->dev = NULL; | |
512 | } | |
5ab11c98 PE |
513 | qp->q.stamp = skb->tstamp; |
514 | qp->q.meat += skb->len; | |
6623e3b2 | 515 | qp->ecn |= ecn; |
d433673e | 516 | add_frag_mem_limit(&qp->q, skb->truesize); |
1da177e4 | 517 | if (offset == 0) |
bc578a54 | 518 | qp->q.last_in |= INET_FRAG_FIRST_IN; |
1da177e4 | 519 | |
5f2d04f1 PM |
520 | if (ip_hdr(skb)->frag_off & htons(IP_DF) && |
521 | skb->len + ihl > qp->q.max_size) | |
522 | qp->q.max_size = skb->len + ihl; | |
523 | ||
bc578a54 | 524 | if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
97599dc7 ED |
525 | qp->q.meat == qp->q.len) { |
526 | unsigned long orefdst = skb->_skb_refdst; | |
1706d587 | 527 | |
97599dc7 ED |
528 | skb->_skb_refdst = 0UL; |
529 | err = ip_frag_reasm(qp, prev, dev); | |
530 | skb->_skb_refdst = orefdst; | |
531 | return err; | |
532 | } | |
533 | ||
534 | skb_dst_drop(skb); | |
3ef0eb0d | 535 | inet_frag_lru_move(&qp->q); |
1706d587 | 536 | return -EINPROGRESS; |
1da177e4 LT |
537 | |
538 | err: | |
539 | kfree_skb(skb); | |
1706d587 | 540 | return err; |
1da177e4 LT |
541 | } |
542 | ||
543 | ||
544 | /* Build a new IP datagram from all its fragments. */ | |
545 | ||
1706d587 HX |
546 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
547 | struct net_device *dev) | |
1da177e4 | 548 | { |
2bad35b7 | 549 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 550 | struct iphdr *iph; |
5ab11c98 | 551 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
552 | int len; |
553 | int ihlen; | |
1706d587 | 554 | int err; |
3cc49492 | 555 | int sum_truesize; |
5173cc05 | 556 | u8 ecn; |
1da177e4 LT |
557 | |
558 | ipq_kill(qp); | |
559 | ||
5173cc05 ED |
560 | ecn = ip4_frag_ecn_table[qp->ecn]; |
561 | if (unlikely(ecn == 0xff)) { | |
562 | err = -EINVAL; | |
563 | goto out_fail; | |
564 | } | |
1706d587 HX |
565 | /* Make the one we just received the head. */ |
566 | if (prev) { | |
567 | head = prev->next; | |
568 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
569 | if (!fp) |
570 | goto out_nomem; | |
571 | ||
572 | fp->next = head->next; | |
d6bebca9 CG |
573 | if (!fp->next) |
574 | qp->q.fragments_tail = fp; | |
1706d587 HX |
575 | prev->next = fp; |
576 | ||
5ab11c98 PE |
577 | skb_morph(head, qp->q.fragments); |
578 | head->next = qp->q.fragments->next; | |
1706d587 | 579 | |
cbf8f7bb | 580 | consume_skb(qp->q.fragments); |
5ab11c98 | 581 | qp->q.fragments = head; |
1706d587 HX |
582 | } |
583 | ||
547b792c IJ |
584 | WARN_ON(head == NULL); |
585 | WARN_ON(FRAG_CB(head)->offset != 0); | |
1da177e4 LT |
586 | |
587 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 588 | ihlen = ip_hdrlen(head); |
5ab11c98 | 589 | len = ihlen + qp->q.len; |
1da177e4 | 590 | |
1706d587 | 591 | err = -E2BIG; |
132adf54 | 592 | if (len > 65535) |
1da177e4 LT |
593 | goto out_oversize; |
594 | ||
595 | /* Head of list must not be cloned. */ | |
14bbd6a5 | 596 | if (skb_unclone(head, GFP_ATOMIC)) |
1da177e4 LT |
597 | goto out_nomem; |
598 | ||
599 | /* If the first fragment is fragmented itself, we split | |
600 | * it to two chunks: the first with data and paged part | |
601 | * and the second, holding only fragments. */ | |
21dc3301 | 602 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
603 | struct sk_buff *clone; |
604 | int i, plen = 0; | |
605 | ||
606 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
607 | goto out_nomem; | |
608 | clone->next = head->next; | |
609 | head->next = clone; | |
610 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 611 | skb_frag_list_init(head); |
9e903e08 ED |
612 | for (i = 0; i < skb_shinfo(head)->nr_frags; i++) |
613 | plen += skb_frag_size(&skb_shinfo(head)->frags[i]); | |
1da177e4 LT |
614 | clone->len = clone->data_len = head->data_len - plen; |
615 | head->data_len -= clone->len; | |
616 | head->len -= clone->len; | |
617 | clone->csum = 0; | |
618 | clone->ip_summed = head->ip_summed; | |
d433673e | 619 | add_frag_mem_limit(&qp->q, clone->truesize); |
1da177e4 LT |
620 | } |
621 | ||
d56f90a7 | 622 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 | 623 | |
3cc49492 ED |
624 | sum_truesize = head->truesize; |
625 | for (fp = head->next; fp;) { | |
626 | bool headstolen; | |
627 | int delta; | |
628 | struct sk_buff *next = fp->next; | |
629 | ||
630 | sum_truesize += fp->truesize; | |
1da177e4 LT |
631 | if (head->ip_summed != fp->ip_summed) |
632 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 633 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 | 634 | head->csum = csum_add(head->csum, fp->csum); |
3cc49492 ED |
635 | |
636 | if (skb_try_coalesce(head, fp, &headstolen, &delta)) { | |
637 | kfree_skb_partial(fp, headstolen); | |
638 | } else { | |
639 | if (!skb_shinfo(head)->frag_list) | |
640 | skb_shinfo(head)->frag_list = fp; | |
641 | head->data_len += fp->len; | |
642 | head->len += fp->len; | |
643 | head->truesize += fp->truesize; | |
644 | } | |
645 | fp = next; | |
1da177e4 | 646 | } |
d433673e | 647 | sub_frag_mem_limit(&qp->q, sum_truesize); |
1da177e4 LT |
648 | |
649 | head->next = NULL; | |
650 | head->dev = dev; | |
5ab11c98 | 651 | head->tstamp = qp->q.stamp; |
5f2d04f1 | 652 | IPCB(head)->frag_max_size = qp->q.max_size; |
1da177e4 | 653 | |
eddc9ec5 | 654 | iph = ip_hdr(head); |
5f2d04f1 PM |
655 | /* max_size != 0 implies at least one fragment had IP_DF set */ |
656 | iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0; | |
1da177e4 | 657 | iph->tot_len = htons(len); |
5173cc05 | 658 | iph->tos |= ecn; |
2bad35b7 | 659 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 660 | qp->q.fragments = NULL; |
d6bebca9 | 661 | qp->q.fragments_tail = NULL; |
1706d587 | 662 | return 0; |
1da177e4 LT |
663 | |
664 | out_nomem: | |
afd46503 JP |
665 | LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"), |
666 | qp); | |
45542479 | 667 | err = -ENOMEM; |
1da177e4 LT |
668 | goto out_fail; |
669 | out_oversize: | |
e87cc472 | 670 | net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr); |
1da177e4 | 671 | out_fail: |
bbf31bf1 | 672 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 673 | return err; |
1da177e4 LT |
674 | } |
675 | ||
676 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 677 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 678 | { |
1da177e4 | 679 | struct ipq *qp; |
ac18e750 | 680 | struct net *net; |
e905a9ed | 681 | |
adf30907 | 682 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); |
7c73a6fa | 683 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
684 | |
685 | /* Start by cleaning up the memory. */ | |
6b102865 | 686 | ip_evictor(net); |
1da177e4 | 687 | |
1da177e4 | 688 | /* Lookup (or create) queue header */ |
ac18e750 | 689 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 690 | int ret; |
1da177e4 | 691 | |
5ab11c98 | 692 | spin_lock(&qp->q.lock); |
1da177e4 | 693 | |
1706d587 | 694 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 695 | |
5ab11c98 | 696 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 697 | ipq_put(qp); |
776c729e | 698 | return ret; |
1da177e4 LT |
699 | } |
700 | ||
7c73a6fa | 701 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 702 | kfree_skb(skb); |
776c729e | 703 | return -ENOMEM; |
1da177e4 | 704 | } |
4bc2f18b | 705 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 706 | |
bc416d97 ED |
707 | struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user) |
708 | { | |
1bf3751e | 709 | struct iphdr iph; |
bc416d97 ED |
710 | u32 len; |
711 | ||
712 | if (skb->protocol != htons(ETH_P_IP)) | |
713 | return skb; | |
714 | ||
1bf3751e | 715 | if (!skb_copy_bits(skb, 0, &iph, sizeof(iph))) |
bc416d97 ED |
716 | return skb; |
717 | ||
1bf3751e | 718 | if (iph.ihl < 5 || iph.version != 4) |
bc416d97 | 719 | return skb; |
1bf3751e JB |
720 | |
721 | len = ntohs(iph.tot_len); | |
722 | if (skb->len < len || len < (iph.ihl * 4)) | |
bc416d97 ED |
723 | return skb; |
724 | ||
1bf3751e | 725 | if (ip_is_fragment(&iph)) { |
bc416d97 ED |
726 | skb = skb_share_check(skb, GFP_ATOMIC); |
727 | if (skb) { | |
1bf3751e JB |
728 | if (!pskb_may_pull(skb, iph.ihl*4)) |
729 | return skb; | |
bc416d97 ED |
730 | if (pskb_trim_rcsum(skb, len)) |
731 | return skb; | |
732 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
733 | if (ip_defrag(skb, user)) | |
734 | return NULL; | |
735 | skb->rxhash = 0; | |
736 | } | |
737 | } | |
738 | return skb; | |
739 | } | |
740 | EXPORT_SYMBOL(ip_check_defrag); | |
741 | ||
8d8354d2 PE |
742 | #ifdef CONFIG_SYSCTL |
743 | static int zero; | |
744 | ||
0a64b4b8 | 745 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 746 | { |
8d8354d2 | 747 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 748 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
749 | .maxlen = sizeof(int), |
750 | .mode = 0644, | |
6d9f239a | 751 | .proc_handler = proc_dointvec |
8d8354d2 PE |
752 | }, |
753 | { | |
8d8354d2 | 754 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 755 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
756 | .maxlen = sizeof(int), |
757 | .mode = 0644, | |
6d9f239a | 758 | .proc_handler = proc_dointvec |
8d8354d2 PE |
759 | }, |
760 | { | |
8d8354d2 | 761 | .procname = "ipfrag_time", |
b2fd5321 | 762 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
763 | .maxlen = sizeof(int), |
764 | .mode = 0644, | |
6d9f239a | 765 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 766 | }, |
7d291ebb PE |
767 | { } |
768 | }; | |
769 | ||
770 | static struct ctl_table ip4_frags_ctl_table[] = { | |
8d8354d2 | 771 | { |
8d8354d2 | 772 | .procname = "ipfrag_secret_interval", |
3b4bc4a2 | 773 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
774 | .maxlen = sizeof(int), |
775 | .mode = 0644, | |
6d9f239a | 776 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 PE |
777 | }, |
778 | { | |
779 | .procname = "ipfrag_max_dist", | |
780 | .data = &sysctl_ipfrag_max_dist, | |
781 | .maxlen = sizeof(int), | |
782 | .mode = 0644, | |
6d9f239a | 783 | .proc_handler = proc_dointvec_minmax, |
8d8354d2 PE |
784 | .extra1 = &zero |
785 | }, | |
786 | { } | |
787 | }; | |
788 | ||
2c8c1e72 | 789 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 790 | { |
e4a2d5c2 | 791 | struct ctl_table *table; |
8d8354d2 PE |
792 | struct ctl_table_header *hdr; |
793 | ||
0a64b4b8 | 794 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 795 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 796 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
e4a2d5c2 PE |
797 | if (table == NULL) |
798 | goto err_alloc; | |
799 | ||
e31e0bdc PE |
800 | table[0].data = &net->ipv4.frags.high_thresh; |
801 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 802 | table[2].data = &net->ipv4.frags.timeout; |
464dc801 EB |
803 | |
804 | /* Don't export sysctls to unprivileged users */ | |
805 | if (net->user_ns != &init_user_ns) | |
806 | table[0].procname = NULL; | |
e4a2d5c2 PE |
807 | } |
808 | ||
ec8f23ce | 809 | hdr = register_net_sysctl(net, "net/ipv4", table); |
e4a2d5c2 PE |
810 | if (hdr == NULL) |
811 | goto err_reg; | |
812 | ||
813 | net->ipv4.frags_hdr = hdr; | |
814 | return 0; | |
815 | ||
816 | err_reg: | |
09ad9bc7 | 817 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
818 | kfree(table); |
819 | err_alloc: | |
820 | return -ENOMEM; | |
821 | } | |
822 | ||
2c8c1e72 | 823 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
824 | { |
825 | struct ctl_table *table; | |
826 | ||
827 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
828 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
829 | kfree(table); | |
8d8354d2 | 830 | } |
7d291ebb PE |
831 | |
832 | static void ip4_frags_ctl_register(void) | |
833 | { | |
43444757 | 834 | register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); |
7d291ebb | 835 | } |
8d8354d2 | 836 | #else |
0a64b4b8 | 837 | static inline int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
838 | { |
839 | return 0; | |
840 | } | |
e4a2d5c2 | 841 | |
0a64b4b8 | 842 | static inline void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
843 | { |
844 | } | |
7d291ebb PE |
845 | |
846 | static inline void ip4_frags_ctl_register(void) | |
847 | { | |
848 | } | |
8d8354d2 PE |
849 | #endif |
850 | ||
2c8c1e72 | 851 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 852 | { |
c2a93660 JDB |
853 | /* Fragment cache limits. |
854 | * | |
855 | * The fragment memory accounting code, (tries to) account for | |
856 | * the real memory usage, by measuring both the size of frag | |
857 | * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) | |
858 | * and the SKB's truesize. | |
859 | * | |
860 | * A 64K fragment consumes 129736 bytes (44*2944)+200 | |
861 | * (1500 truesize == 2944, sizeof(struct ipq) == 200) | |
862 | * | |
863 | * We will commit 4MB at one time. Should we cross that limit | |
864 | * we will prune down to 3MB, making room for approx 8 big 64K | |
865 | * fragments 8x128k. | |
e31e0bdc | 866 | */ |
c2a93660 JDB |
867 | net->ipv4.frags.high_thresh = 4 * 1024 * 1024; |
868 | net->ipv4.frags.low_thresh = 3 * 1024 * 1024; | |
b2fd5321 PE |
869 | /* |
870 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
871 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
872 | * by TTL. | |
873 | */ | |
874 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
875 | ||
e5a2bb84 PE |
876 | inet_frags_init_net(&net->ipv4.frags); |
877 | ||
0a64b4b8 | 878 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
879 | } |
880 | ||
2c8c1e72 | 881 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 882 | { |
0a64b4b8 | 883 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
884 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
885 | } | |
886 | ||
887 | static struct pernet_operations ip4_frags_ops = { | |
888 | .init = ipv4_frags_init_net, | |
889 | .exit = ipv4_frags_exit_net, | |
890 | }; | |
891 | ||
b7aa0bf7 | 892 | void __init ipfrag_init(void) |
1da177e4 | 893 | { |
7d291ebb | 894 | ip4_frags_ctl_register(); |
81566e83 | 895 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 896 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 897 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
898 | ip4_frags.destructor = ip4_frag_free; |
899 | ip4_frags.skb_free = NULL; | |
900 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 901 | ip4_frags.match = ip4_frag_match; |
e521db9d | 902 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 903 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 904 | inet_frags_init(&ip4_frags); |
1da177e4 | 905 | } |