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 | ||
6623e3b2 ED |
82 | static inline u8 ip4_frag_ecn(u8 tos) |
83 | { | |
5173cc05 | 84 | return 1 << (tos & INET_ECN_MASK); |
6623e3b2 ED |
85 | } |
86 | ||
7eb95156 | 87 | static struct inet_frags ip4_frags; |
1da177e4 | 88 | |
6ddc0822 | 89 | int ip_frag_mem(struct net *net) |
7eb95156 | 90 | { |
d433673e | 91 | return sum_frag_mem_limit(&net->ipv4.frags); |
7eb95156 | 92 | } |
1da177e4 | 93 | |
1706d587 HX |
94 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
95 | struct net_device *dev); | |
96 | ||
c6fda282 PE |
97 | struct ip4_create_arg { |
98 | struct iphdr *iph; | |
99 | u32 user; | |
100 | }; | |
101 | ||
18277770 | 102 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
1da177e4 | 103 | { |
e7b519ba | 104 | net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd)); |
18277770 AV |
105 | return jhash_3words((__force u32)id << 16 | prot, |
106 | (__force u32)saddr, (__force u32)daddr, | |
fb3cfe6e | 107 | ip4_frags.rnd); |
1da177e4 LT |
108 | } |
109 | ||
36c77782 | 110 | static unsigned int ip4_hashfn(const struct inet_frag_queue *q) |
1da177e4 | 111 | { |
36c77782 | 112 | const struct ipq *ipq; |
1da177e4 | 113 | |
321a3a99 PE |
114 | ipq = container_of(q, struct ipq, q); |
115 | return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); | |
1da177e4 LT |
116 | } |
117 | ||
36c77782 | 118 | static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a) |
abd6523d | 119 | { |
36c77782 FW |
120 | const struct ipq *qp; |
121 | const struct ip4_create_arg *arg = a; | |
abd6523d PE |
122 | |
123 | qp = container_of(q, struct ipq, q); | |
a02cec21 | 124 | return qp->id == arg->iph->id && |
cbc264ca ED |
125 | qp->saddr == arg->iph->saddr && |
126 | qp->daddr == arg->iph->daddr && | |
127 | qp->protocol == arg->iph->protocol && | |
128 | qp->user == arg->user; | |
abd6523d PE |
129 | } |
130 | ||
36c77782 | 131 | static void ip4_frag_init(struct inet_frag_queue *q, const void *a) |
c6fda282 PE |
132 | { |
133 | struct ipq *qp = container_of(q, struct ipq, q); | |
54db0cc2 G |
134 | struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, |
135 | frags); | |
136 | struct net *net = container_of(ipv4, struct net, ipv4); | |
137 | ||
36c77782 | 138 | const struct ip4_create_arg *arg = a; |
c6fda282 PE |
139 | |
140 | qp->protocol = arg->iph->protocol; | |
141 | qp->id = arg->iph->id; | |
6623e3b2 | 142 | qp->ecn = ip4_frag_ecn(arg->iph->tos); |
c6fda282 PE |
143 | qp->saddr = arg->iph->saddr; |
144 | qp->daddr = arg->iph->daddr; | |
145 | qp->user = arg->user; | |
146 | qp->peer = sysctl_ipfrag_max_dist ? | |
c0efc887 | 147 | inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL; |
c6fda282 PE |
148 | } |
149 | ||
1e4b8287 | 150 | static __inline__ void ip4_frag_free(struct inet_frag_queue *q) |
1da177e4 | 151 | { |
1e4b8287 PE |
152 | struct ipq *qp; |
153 | ||
154 | qp = container_of(q, struct ipq, q); | |
155 | if (qp->peer) | |
156 | inet_putpeer(qp->peer); | |
1da177e4 LT |
157 | } |
158 | ||
1da177e4 LT |
159 | |
160 | /* Destruction primitives. */ | |
161 | ||
4b6cb5d8 | 162 | static __inline__ void ipq_put(struct ipq *ipq) |
1da177e4 | 163 | { |
762cc408 | 164 | inet_frag_put(&ipq->q, &ip4_frags); |
1da177e4 LT |
165 | } |
166 | ||
167 | /* Kill ipq entry. It is not destroyed immediately, | |
168 | * because caller (and someone more) holds reference count. | |
169 | */ | |
170 | static void ipq_kill(struct ipq *ipq) | |
171 | { | |
277e650d | 172 | inet_frag_kill(&ipq->q, &ip4_frags); |
1da177e4 LT |
173 | } |
174 | ||
1da177e4 LT |
175 | /* |
176 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. | |
177 | */ | |
178 | static void ip_expire(unsigned long arg) | |
179 | { | |
e521db9d | 180 | struct ipq *qp; |
84a3aa00 | 181 | struct net *net; |
e521db9d PE |
182 | |
183 | qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); | |
84a3aa00 | 184 | net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 185 | |
5ab11c98 | 186 | spin_lock(&qp->q.lock); |
1da177e4 | 187 | |
bc578a54 | 188 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
189 | goto out; |
190 | ||
191 | ipq_kill(qp); | |
192 | ||
b13d3cbf FW |
193 | if (!(qp->q.last_in & INET_FRAG_EVICTED)) |
194 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); | |
7c73a6fa | 195 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 196 | |
bc578a54 | 197 | if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) { |
5ab11c98 | 198 | struct sk_buff *head = qp->q.fragments; |
64f3b9e2 ED |
199 | const struct iphdr *iph; |
200 | int err; | |
cb84663e | 201 | |
69df9d59 ED |
202 | rcu_read_lock(); |
203 | head->dev = dev_get_by_index_rcu(net, qp->iif); | |
e9017b55 SW |
204 | if (!head->dev) |
205 | goto out_rcu_unlock; | |
206 | ||
97599dc7 | 207 | /* skb has no dst, perform route lookup again */ |
64f3b9e2 | 208 | iph = ip_hdr(head); |
c6cffba4 DM |
209 | err = ip_route_input_noref(head, iph->daddr, iph->saddr, |
210 | iph->tos, head->dev); | |
64f3b9e2 ED |
211 | if (err) |
212 | goto out_rcu_unlock; | |
213 | ||
e9017b55 | 214 | /* |
64f3b9e2 ED |
215 | * Only an end host needs to send an ICMP |
216 | * "Fragment Reassembly Timeout" message, per RFC792. | |
e9017b55 | 217 | */ |
595fc71b | 218 | if (qp->user == IP_DEFRAG_AF_PACKET || |
7c3d5ab1 VA |
219 | ((qp->user >= IP_DEFRAG_CONNTRACK_IN) && |
220 | (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) && | |
221 | (skb_rtable(head)->rt_type != RTN_LOCAL))) | |
64f3b9e2 ED |
222 | goto out_rcu_unlock; |
223 | ||
e9017b55 SW |
224 | |
225 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ | |
226 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); | |
e9017b55 | 227 | out_rcu_unlock: |
d1c9ae6d PM |
228 | rcu_read_unlock(); |
229 | } | |
1da177e4 | 230 | out: |
5ab11c98 | 231 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 232 | ipq_put(qp); |
1da177e4 LT |
233 | } |
234 | ||
abd6523d PE |
235 | /* Find the correct entry in the "incomplete datagrams" queue for |
236 | * this IP datagram, and create new one, if nothing is found. | |
237 | */ | |
ac18e750 | 238 | static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) |
1da177e4 | 239 | { |
c6fda282 PE |
240 | struct inet_frag_queue *q; |
241 | struct ip4_create_arg arg; | |
abd6523d | 242 | unsigned int hash; |
1da177e4 | 243 | |
c6fda282 PE |
244 | arg.iph = iph; |
245 | arg.user = user; | |
9a375803 PE |
246 | |
247 | read_lock(&ip4_frags.lock); | |
abd6523d | 248 | hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); |
1da177e4 | 249 | |
ac18e750 | 250 | q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); |
5a3da1fe HFS |
251 | if (IS_ERR_OR_NULL(q)) { |
252 | inet_frag_maybe_warn_overflow(q, pr_fmt()); | |
253 | return NULL; | |
254 | } | |
c6fda282 | 255 | return container_of(q, struct ipq, q); |
1da177e4 LT |
256 | } |
257 | ||
89cee8b1 HX |
258 | /* Is the fragment too far ahead to be part of ipq? */ |
259 | static inline int ip_frag_too_far(struct ipq *qp) | |
260 | { | |
261 | struct inet_peer *peer = qp->peer; | |
262 | unsigned int max = sysctl_ipfrag_max_dist; | |
263 | unsigned int start, end; | |
264 | ||
265 | int rc; | |
266 | ||
267 | if (!peer || !max) | |
268 | return 0; | |
269 | ||
270 | start = qp->rid; | |
271 | end = atomic_inc_return(&peer->rid); | |
272 | qp->rid = end; | |
273 | ||
5ab11c98 | 274 | rc = qp->q.fragments && (end - start) > max; |
89cee8b1 HX |
275 | |
276 | if (rc) { | |
7c73a6fa PE |
277 | struct net *net; |
278 | ||
279 | net = container_of(qp->q.net, struct net, ipv4.frags); | |
280 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); | |
89cee8b1 HX |
281 | } |
282 | ||
283 | return rc; | |
284 | } | |
285 | ||
286 | static int ip_frag_reinit(struct ipq *qp) | |
287 | { | |
288 | struct sk_buff *fp; | |
d433673e | 289 | unsigned int sum_truesize = 0; |
89cee8b1 | 290 | |
b2fd5321 | 291 | if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { |
5ab11c98 | 292 | atomic_inc(&qp->q.refcnt); |
89cee8b1 HX |
293 | return -ETIMEDOUT; |
294 | } | |
295 | ||
5ab11c98 | 296 | fp = qp->q.fragments; |
89cee8b1 HX |
297 | do { |
298 | struct sk_buff *xp = fp->next; | |
d433673e JDB |
299 | |
300 | sum_truesize += fp->truesize; | |
301 | kfree_skb(fp); | |
89cee8b1 HX |
302 | fp = xp; |
303 | } while (fp); | |
d433673e | 304 | sub_frag_mem_limit(&qp->q, sum_truesize); |
89cee8b1 | 305 | |
5ab11c98 PE |
306 | qp->q.last_in = 0; |
307 | qp->q.len = 0; | |
308 | qp->q.meat = 0; | |
309 | qp->q.fragments = NULL; | |
d6bebca9 | 310 | qp->q.fragments_tail = NULL; |
89cee8b1 | 311 | qp->iif = 0; |
6623e3b2 | 312 | qp->ecn = 0; |
89cee8b1 HX |
313 | |
314 | return 0; | |
315 | } | |
316 | ||
1da177e4 | 317 | /* Add new segment to existing queue. */ |
1706d587 | 318 | static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
1da177e4 LT |
319 | { |
320 | struct sk_buff *prev, *next; | |
1706d587 | 321 | struct net_device *dev; |
1da177e4 LT |
322 | int flags, offset; |
323 | int ihl, end; | |
1706d587 | 324 | int err = -ENOENT; |
6623e3b2 | 325 | u8 ecn; |
1da177e4 | 326 | |
bc578a54 | 327 | if (qp->q.last_in & INET_FRAG_COMPLETE) |
1da177e4 LT |
328 | goto err; |
329 | ||
89cee8b1 | 330 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
1706d587 HX |
331 | unlikely(ip_frag_too_far(qp)) && |
332 | unlikely(err = ip_frag_reinit(qp))) { | |
89cee8b1 HX |
333 | ipq_kill(qp); |
334 | goto err; | |
335 | } | |
336 | ||
6623e3b2 | 337 | ecn = ip4_frag_ecn(ip_hdr(skb)->tos); |
eddc9ec5 | 338 | offset = ntohs(ip_hdr(skb)->frag_off); |
1da177e4 LT |
339 | flags = offset & ~IP_OFFSET; |
340 | offset &= IP_OFFSET; | |
341 | offset <<= 3; /* offset is in 8-byte chunks */ | |
c9bdd4b5 | 342 | ihl = ip_hdrlen(skb); |
1da177e4 LT |
343 | |
344 | /* Determine the position of this fragment. */ | |
e905a9ed | 345 | end = offset + skb->len - ihl; |
1706d587 | 346 | err = -EINVAL; |
1da177e4 LT |
347 | |
348 | /* Is this the final fragment? */ | |
349 | if ((flags & IP_MF) == 0) { | |
350 | /* If we already have some bits beyond end | |
42b2aa86 | 351 | * or have different end, the segment is corrupted. |
1da177e4 | 352 | */ |
5ab11c98 | 353 | if (end < qp->q.len || |
bc578a54 | 354 | ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) |
1da177e4 | 355 | goto err; |
bc578a54 | 356 | qp->q.last_in |= INET_FRAG_LAST_IN; |
5ab11c98 | 357 | qp->q.len = end; |
1da177e4 LT |
358 | } else { |
359 | if (end&7) { | |
360 | end &= ~7; | |
361 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
362 | skb->ip_summed = CHECKSUM_NONE; | |
363 | } | |
5ab11c98 | 364 | if (end > qp->q.len) { |
1da177e4 | 365 | /* Some bits beyond end -> corruption. */ |
bc578a54 | 366 | if (qp->q.last_in & INET_FRAG_LAST_IN) |
1da177e4 | 367 | goto err; |
5ab11c98 | 368 | qp->q.len = end; |
1da177e4 LT |
369 | } |
370 | } | |
371 | if (end == offset) | |
372 | goto err; | |
373 | ||
1706d587 | 374 | err = -ENOMEM; |
1da177e4 LT |
375 | if (pskb_pull(skb, ihl) == NULL) |
376 | goto err; | |
1706d587 HX |
377 | |
378 | err = pskb_trim_rcsum(skb, end - offset); | |
379 | if (err) | |
1da177e4 LT |
380 | goto err; |
381 | ||
382 | /* Find out which fragments are in front and at the back of us | |
383 | * in the chain of fragments so far. We must know where to put | |
384 | * this fragment, right? | |
385 | */ | |
d6bebca9 CG |
386 | prev = qp->q.fragments_tail; |
387 | if (!prev || FRAG_CB(prev)->offset < offset) { | |
388 | next = NULL; | |
389 | goto found; | |
390 | } | |
1da177e4 | 391 | prev = NULL; |
5ab11c98 | 392 | for (next = qp->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
393 | if (FRAG_CB(next)->offset >= offset) |
394 | break; /* bingo! */ | |
395 | prev = next; | |
396 | } | |
397 | ||
d6bebca9 | 398 | found: |
1da177e4 LT |
399 | /* We found where to put this one. Check for overlap with |
400 | * preceding fragment, and, if needed, align things so that | |
401 | * any overlaps are eliminated. | |
402 | */ | |
403 | if (prev) { | |
404 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; | |
405 | ||
406 | if (i > 0) { | |
407 | offset += i; | |
1706d587 | 408 | err = -EINVAL; |
1da177e4 LT |
409 | if (end <= offset) |
410 | goto err; | |
1706d587 | 411 | err = -ENOMEM; |
1da177e4 LT |
412 | if (!pskb_pull(skb, i)) |
413 | goto err; | |
414 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
415 | skb->ip_summed = CHECKSUM_NONE; | |
416 | } | |
417 | } | |
418 | ||
1706d587 HX |
419 | err = -ENOMEM; |
420 | ||
1da177e4 LT |
421 | while (next && FRAG_CB(next)->offset < end) { |
422 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ | |
423 | ||
424 | if (i < next->len) { | |
425 | /* Eat head of the next overlapped fragment | |
426 | * and leave the loop. The next ones cannot overlap. | |
427 | */ | |
428 | if (!pskb_pull(next, i)) | |
429 | goto err; | |
430 | FRAG_CB(next)->offset += i; | |
5ab11c98 | 431 | qp->q.meat -= i; |
1da177e4 LT |
432 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
433 | next->ip_summed = CHECKSUM_NONE; | |
434 | break; | |
435 | } else { | |
436 | struct sk_buff *free_it = next; | |
437 | ||
47c6bf77 | 438 | /* Old fragment is completely overridden with |
1da177e4 LT |
439 | * new one drop it. |
440 | */ | |
441 | next = next->next; | |
442 | ||
443 | if (prev) | |
444 | prev->next = next; | |
445 | else | |
5ab11c98 | 446 | qp->q.fragments = next; |
1da177e4 | 447 | |
5ab11c98 | 448 | qp->q.meat -= free_it->len; |
d433673e JDB |
449 | sub_frag_mem_limit(&qp->q, free_it->truesize); |
450 | kfree_skb(free_it); | |
1da177e4 LT |
451 | } |
452 | } | |
453 | ||
454 | FRAG_CB(skb)->offset = offset; | |
455 | ||
456 | /* Insert this fragment in the chain of fragments. */ | |
457 | skb->next = next; | |
d6bebca9 CG |
458 | if (!next) |
459 | qp->q.fragments_tail = skb; | |
1da177e4 LT |
460 | if (prev) |
461 | prev->next = skb; | |
462 | else | |
5ab11c98 | 463 | qp->q.fragments = skb; |
1da177e4 | 464 | |
1706d587 HX |
465 | dev = skb->dev; |
466 | if (dev) { | |
467 | qp->iif = dev->ifindex; | |
468 | skb->dev = NULL; | |
469 | } | |
5ab11c98 PE |
470 | qp->q.stamp = skb->tstamp; |
471 | qp->q.meat += skb->len; | |
6623e3b2 | 472 | qp->ecn |= ecn; |
d433673e | 473 | add_frag_mem_limit(&qp->q, skb->truesize); |
1da177e4 | 474 | if (offset == 0) |
bc578a54 | 475 | qp->q.last_in |= INET_FRAG_FIRST_IN; |
1da177e4 | 476 | |
5f2d04f1 PM |
477 | if (ip_hdr(skb)->frag_off & htons(IP_DF) && |
478 | skb->len + ihl > qp->q.max_size) | |
479 | qp->q.max_size = skb->len + ihl; | |
480 | ||
bc578a54 | 481 | if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && |
97599dc7 ED |
482 | qp->q.meat == qp->q.len) { |
483 | unsigned long orefdst = skb->_skb_refdst; | |
1706d587 | 484 | |
97599dc7 ED |
485 | skb->_skb_refdst = 0UL; |
486 | err = ip_frag_reasm(qp, prev, dev); | |
487 | skb->_skb_refdst = orefdst; | |
488 | return err; | |
489 | } | |
490 | ||
491 | skb_dst_drop(skb); | |
3ef0eb0d | 492 | inet_frag_lru_move(&qp->q); |
1706d587 | 493 | return -EINPROGRESS; |
1da177e4 LT |
494 | |
495 | err: | |
496 | kfree_skb(skb); | |
1706d587 | 497 | return err; |
1da177e4 LT |
498 | } |
499 | ||
500 | ||
501 | /* Build a new IP datagram from all its fragments. */ | |
502 | ||
1706d587 HX |
503 | static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, |
504 | struct net_device *dev) | |
1da177e4 | 505 | { |
2bad35b7 | 506 | struct net *net = container_of(qp->q.net, struct net, ipv4.frags); |
1da177e4 | 507 | struct iphdr *iph; |
5ab11c98 | 508 | struct sk_buff *fp, *head = qp->q.fragments; |
1da177e4 LT |
509 | int len; |
510 | int ihlen; | |
1706d587 | 511 | int err; |
3cc49492 | 512 | int sum_truesize; |
5173cc05 | 513 | u8 ecn; |
1da177e4 LT |
514 | |
515 | ipq_kill(qp); | |
516 | ||
be991971 | 517 | ecn = ip_frag_ecn_table[qp->ecn]; |
5173cc05 ED |
518 | if (unlikely(ecn == 0xff)) { |
519 | err = -EINVAL; | |
520 | goto out_fail; | |
521 | } | |
1706d587 HX |
522 | /* Make the one we just received the head. */ |
523 | if (prev) { | |
524 | head = prev->next; | |
525 | fp = skb_clone(head, GFP_ATOMIC); | |
1706d587 HX |
526 | if (!fp) |
527 | goto out_nomem; | |
528 | ||
529 | fp->next = head->next; | |
d6bebca9 CG |
530 | if (!fp->next) |
531 | qp->q.fragments_tail = fp; | |
1706d587 HX |
532 | prev->next = fp; |
533 | ||
5ab11c98 PE |
534 | skb_morph(head, qp->q.fragments); |
535 | head->next = qp->q.fragments->next; | |
1706d587 | 536 | |
cbf8f7bb | 537 | consume_skb(qp->q.fragments); |
5ab11c98 | 538 | qp->q.fragments = head; |
1706d587 HX |
539 | } |
540 | ||
547b792c IJ |
541 | WARN_ON(head == NULL); |
542 | WARN_ON(FRAG_CB(head)->offset != 0); | |
1da177e4 LT |
543 | |
544 | /* Allocate a new buffer for the datagram. */ | |
c9bdd4b5 | 545 | ihlen = ip_hdrlen(head); |
5ab11c98 | 546 | len = ihlen + qp->q.len; |
1da177e4 | 547 | |
1706d587 | 548 | err = -E2BIG; |
132adf54 | 549 | if (len > 65535) |
1da177e4 LT |
550 | goto out_oversize; |
551 | ||
552 | /* Head of list must not be cloned. */ | |
14bbd6a5 | 553 | if (skb_unclone(head, GFP_ATOMIC)) |
1da177e4 LT |
554 | goto out_nomem; |
555 | ||
556 | /* If the first fragment is fragmented itself, we split | |
557 | * it to two chunks: the first with data and paged part | |
558 | * and the second, holding only fragments. */ | |
21dc3301 | 559 | if (skb_has_frag_list(head)) { |
1da177e4 LT |
560 | struct sk_buff *clone; |
561 | int i, plen = 0; | |
562 | ||
563 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
564 | goto out_nomem; | |
565 | clone->next = head->next; | |
566 | head->next = clone; | |
567 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
d7fcf1a5 | 568 | skb_frag_list_init(head); |
9e903e08 ED |
569 | for (i = 0; i < skb_shinfo(head)->nr_frags; i++) |
570 | plen += skb_frag_size(&skb_shinfo(head)->frags[i]); | |
1da177e4 LT |
571 | clone->len = clone->data_len = head->data_len - plen; |
572 | head->data_len -= clone->len; | |
573 | head->len -= clone->len; | |
574 | clone->csum = 0; | |
575 | clone->ip_summed = head->ip_summed; | |
d433673e | 576 | add_frag_mem_limit(&qp->q, clone->truesize); |
1da177e4 LT |
577 | } |
578 | ||
d56f90a7 | 579 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 | 580 | |
3cc49492 ED |
581 | sum_truesize = head->truesize; |
582 | for (fp = head->next; fp;) { | |
583 | bool headstolen; | |
584 | int delta; | |
585 | struct sk_buff *next = fp->next; | |
586 | ||
587 | sum_truesize += fp->truesize; | |
1da177e4 LT |
588 | if (head->ip_summed != fp->ip_summed) |
589 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 590 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 | 591 | head->csum = csum_add(head->csum, fp->csum); |
3cc49492 ED |
592 | |
593 | if (skb_try_coalesce(head, fp, &headstolen, &delta)) { | |
594 | kfree_skb_partial(fp, headstolen); | |
595 | } else { | |
596 | if (!skb_shinfo(head)->frag_list) | |
597 | skb_shinfo(head)->frag_list = fp; | |
598 | head->data_len += fp->len; | |
599 | head->len += fp->len; | |
600 | head->truesize += fp->truesize; | |
601 | } | |
602 | fp = next; | |
1da177e4 | 603 | } |
d433673e | 604 | sub_frag_mem_limit(&qp->q, sum_truesize); |
1da177e4 LT |
605 | |
606 | head->next = NULL; | |
607 | head->dev = dev; | |
5ab11c98 | 608 | head->tstamp = qp->q.stamp; |
5f2d04f1 | 609 | IPCB(head)->frag_max_size = qp->q.max_size; |
1da177e4 | 610 | |
eddc9ec5 | 611 | iph = ip_hdr(head); |
5f2d04f1 PM |
612 | /* max_size != 0 implies at least one fragment had IP_DF set */ |
613 | iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0; | |
1da177e4 | 614 | iph->tot_len = htons(len); |
5173cc05 | 615 | iph->tos |= ecn; |
2bad35b7 | 616 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); |
5ab11c98 | 617 | qp->q.fragments = NULL; |
d6bebca9 | 618 | qp->q.fragments_tail = NULL; |
1706d587 | 619 | return 0; |
1da177e4 LT |
620 | |
621 | out_nomem: | |
afd46503 JP |
622 | LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"), |
623 | qp); | |
45542479 | 624 | err = -ENOMEM; |
1da177e4 LT |
625 | goto out_fail; |
626 | out_oversize: | |
e87cc472 | 627 | net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr); |
1da177e4 | 628 | out_fail: |
bbf31bf1 | 629 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1706d587 | 630 | return err; |
1da177e4 LT |
631 | } |
632 | ||
633 | /* Process an incoming IP datagram fragment. */ | |
776c729e | 634 | int ip_defrag(struct sk_buff *skb, u32 user) |
1da177e4 | 635 | { |
1da177e4 | 636 | struct ipq *qp; |
ac18e750 | 637 | struct net *net; |
e905a9ed | 638 | |
adf30907 | 639 | net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); |
7c73a6fa | 640 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); |
1da177e4 | 641 | |
1da177e4 | 642 | /* Lookup (or create) queue header */ |
ac18e750 | 643 | if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { |
1706d587 | 644 | int ret; |
1da177e4 | 645 | |
5ab11c98 | 646 | spin_lock(&qp->q.lock); |
1da177e4 | 647 | |
1706d587 | 648 | ret = ip_frag_queue(qp, skb); |
1da177e4 | 649 | |
5ab11c98 | 650 | spin_unlock(&qp->q.lock); |
4b6cb5d8 | 651 | ipq_put(qp); |
776c729e | 652 | return ret; |
1da177e4 LT |
653 | } |
654 | ||
7c73a6fa | 655 | IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); |
1da177e4 | 656 | kfree_skb(skb); |
776c729e | 657 | return -ENOMEM; |
1da177e4 | 658 | } |
4bc2f18b | 659 | EXPORT_SYMBOL(ip_defrag); |
1da177e4 | 660 | |
bc416d97 ED |
661 | struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user) |
662 | { | |
1bf3751e | 663 | struct iphdr iph; |
bc416d97 ED |
664 | u32 len; |
665 | ||
666 | if (skb->protocol != htons(ETH_P_IP)) | |
667 | return skb; | |
668 | ||
1bf3751e | 669 | if (!skb_copy_bits(skb, 0, &iph, sizeof(iph))) |
bc416d97 ED |
670 | return skb; |
671 | ||
1bf3751e | 672 | if (iph.ihl < 5 || iph.version != 4) |
bc416d97 | 673 | return skb; |
1bf3751e JB |
674 | |
675 | len = ntohs(iph.tot_len); | |
676 | if (skb->len < len || len < (iph.ihl * 4)) | |
bc416d97 ED |
677 | return skb; |
678 | ||
1bf3751e | 679 | if (ip_is_fragment(&iph)) { |
bc416d97 ED |
680 | skb = skb_share_check(skb, GFP_ATOMIC); |
681 | if (skb) { | |
1bf3751e JB |
682 | if (!pskb_may_pull(skb, iph.ihl*4)) |
683 | return skb; | |
bc416d97 ED |
684 | if (pskb_trim_rcsum(skb, len)) |
685 | return skb; | |
686 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
687 | if (ip_defrag(skb, user)) | |
688 | return NULL; | |
7539fadc | 689 | skb_clear_hash(skb); |
bc416d97 ED |
690 | } |
691 | } | |
692 | return skb; | |
693 | } | |
694 | EXPORT_SYMBOL(ip_check_defrag); | |
695 | ||
8d8354d2 PE |
696 | #ifdef CONFIG_SYSCTL |
697 | static int zero; | |
698 | ||
0a64b4b8 | 699 | static struct ctl_table ip4_frags_ns_ctl_table[] = { |
8d8354d2 | 700 | { |
8d8354d2 | 701 | .procname = "ipfrag_high_thresh", |
e31e0bdc | 702 | .data = &init_net.ipv4.frags.high_thresh, |
8d8354d2 PE |
703 | .maxlen = sizeof(int), |
704 | .mode = 0644, | |
6d9f239a | 705 | .proc_handler = proc_dointvec |
8d8354d2 PE |
706 | }, |
707 | { | |
8d8354d2 | 708 | .procname = "ipfrag_low_thresh", |
e31e0bdc | 709 | .data = &init_net.ipv4.frags.low_thresh, |
8d8354d2 PE |
710 | .maxlen = sizeof(int), |
711 | .mode = 0644, | |
6d9f239a | 712 | .proc_handler = proc_dointvec |
8d8354d2 PE |
713 | }, |
714 | { | |
8d8354d2 | 715 | .procname = "ipfrag_time", |
b2fd5321 | 716 | .data = &init_net.ipv4.frags.timeout, |
8d8354d2 PE |
717 | .maxlen = sizeof(int), |
718 | .mode = 0644, | |
6d9f239a | 719 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 | 720 | }, |
7d291ebb PE |
721 | { } |
722 | }; | |
723 | ||
724 | static struct ctl_table ip4_frags_ctl_table[] = { | |
8d8354d2 | 725 | { |
8d8354d2 | 726 | .procname = "ipfrag_secret_interval", |
3b4bc4a2 | 727 | .data = &ip4_frags.secret_interval, |
8d8354d2 PE |
728 | .maxlen = sizeof(int), |
729 | .mode = 0644, | |
6d9f239a | 730 | .proc_handler = proc_dointvec_jiffies, |
8d8354d2 PE |
731 | }, |
732 | { | |
733 | .procname = "ipfrag_max_dist", | |
734 | .data = &sysctl_ipfrag_max_dist, | |
735 | .maxlen = sizeof(int), | |
736 | .mode = 0644, | |
6d9f239a | 737 | .proc_handler = proc_dointvec_minmax, |
8d8354d2 PE |
738 | .extra1 = &zero |
739 | }, | |
740 | { } | |
741 | }; | |
742 | ||
2c8c1e72 | 743 | static int __net_init ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 | 744 | { |
e4a2d5c2 | 745 | struct ctl_table *table; |
8d8354d2 PE |
746 | struct ctl_table_header *hdr; |
747 | ||
0a64b4b8 | 748 | table = ip4_frags_ns_ctl_table; |
09ad9bc7 | 749 | if (!net_eq(net, &init_net)) { |
0a64b4b8 | 750 | table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); |
e4a2d5c2 PE |
751 | if (table == NULL) |
752 | goto err_alloc; | |
753 | ||
e31e0bdc PE |
754 | table[0].data = &net->ipv4.frags.high_thresh; |
755 | table[1].data = &net->ipv4.frags.low_thresh; | |
b2fd5321 | 756 | table[2].data = &net->ipv4.frags.timeout; |
464dc801 EB |
757 | |
758 | /* Don't export sysctls to unprivileged users */ | |
759 | if (net->user_ns != &init_user_ns) | |
760 | table[0].procname = NULL; | |
e4a2d5c2 PE |
761 | } |
762 | ||
ec8f23ce | 763 | hdr = register_net_sysctl(net, "net/ipv4", table); |
e4a2d5c2 PE |
764 | if (hdr == NULL) |
765 | goto err_reg; | |
766 | ||
767 | net->ipv4.frags_hdr = hdr; | |
768 | return 0; | |
769 | ||
770 | err_reg: | |
09ad9bc7 | 771 | if (!net_eq(net, &init_net)) |
e4a2d5c2 PE |
772 | kfree(table); |
773 | err_alloc: | |
774 | return -ENOMEM; | |
775 | } | |
776 | ||
2c8c1e72 | 777 | static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
778 | { |
779 | struct ctl_table *table; | |
780 | ||
781 | table = net->ipv4.frags_hdr->ctl_table_arg; | |
782 | unregister_net_sysctl_table(net->ipv4.frags_hdr); | |
783 | kfree(table); | |
8d8354d2 | 784 | } |
7d291ebb PE |
785 | |
786 | static void ip4_frags_ctl_register(void) | |
787 | { | |
43444757 | 788 | register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); |
7d291ebb | 789 | } |
8d8354d2 | 790 | #else |
0a64b4b8 | 791 | static inline int ip4_frags_ns_ctl_register(struct net *net) |
8d8354d2 PE |
792 | { |
793 | return 0; | |
794 | } | |
e4a2d5c2 | 795 | |
0a64b4b8 | 796 | static inline void ip4_frags_ns_ctl_unregister(struct net *net) |
e4a2d5c2 PE |
797 | { |
798 | } | |
7d291ebb PE |
799 | |
800 | static inline void ip4_frags_ctl_register(void) | |
801 | { | |
802 | } | |
8d8354d2 PE |
803 | #endif |
804 | ||
2c8c1e72 | 805 | static int __net_init ipv4_frags_init_net(struct net *net) |
8d8354d2 | 806 | { |
c2a93660 JDB |
807 | /* Fragment cache limits. |
808 | * | |
809 | * The fragment memory accounting code, (tries to) account for | |
810 | * the real memory usage, by measuring both the size of frag | |
811 | * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) | |
812 | * and the SKB's truesize. | |
813 | * | |
814 | * A 64K fragment consumes 129736 bytes (44*2944)+200 | |
815 | * (1500 truesize == 2944, sizeof(struct ipq) == 200) | |
816 | * | |
817 | * We will commit 4MB at one time. Should we cross that limit | |
818 | * we will prune down to 3MB, making room for approx 8 big 64K | |
819 | * fragments 8x128k. | |
e31e0bdc | 820 | */ |
c2a93660 JDB |
821 | net->ipv4.frags.high_thresh = 4 * 1024 * 1024; |
822 | net->ipv4.frags.low_thresh = 3 * 1024 * 1024; | |
b2fd5321 PE |
823 | /* |
824 | * Important NOTE! Fragment queue must be destroyed before MSL expires. | |
825 | * RFC791 is wrong proposing to prolongate timer each fragment arrival | |
826 | * by TTL. | |
827 | */ | |
828 | net->ipv4.frags.timeout = IP_FRAG_TIME; | |
829 | ||
e5a2bb84 PE |
830 | inet_frags_init_net(&net->ipv4.frags); |
831 | ||
0a64b4b8 | 832 | return ip4_frags_ns_ctl_register(net); |
8d8354d2 PE |
833 | } |
834 | ||
2c8c1e72 | 835 | static void __net_exit ipv4_frags_exit_net(struct net *net) |
81566e83 | 836 | { |
0a64b4b8 | 837 | ip4_frags_ns_ctl_unregister(net); |
81566e83 PE |
838 | inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); |
839 | } | |
840 | ||
841 | static struct pernet_operations ip4_frags_ops = { | |
842 | .init = ipv4_frags_init_net, | |
843 | .exit = ipv4_frags_exit_net, | |
844 | }; | |
845 | ||
b7aa0bf7 | 846 | void __init ipfrag_init(void) |
1da177e4 | 847 | { |
7d291ebb | 848 | ip4_frags_ctl_register(); |
81566e83 | 849 | register_pernet_subsys(&ip4_frags_ops); |
321a3a99 | 850 | ip4_frags.hashfn = ip4_hashfn; |
c6fda282 | 851 | ip4_frags.constructor = ip4_frag_init; |
1e4b8287 PE |
852 | ip4_frags.destructor = ip4_frag_free; |
853 | ip4_frags.skb_free = NULL; | |
854 | ip4_frags.qsize = sizeof(struct ipq); | |
abd6523d | 855 | ip4_frags.match = ip4_frag_match; |
e521db9d | 856 | ip4_frags.frag_expire = ip_expire; |
3b4bc4a2 | 857 | ip4_frags.secret_interval = 10 * 60 * HZ; |
7eb95156 | 858 | inet_frags_init(&ip4_frags); |
1da177e4 | 859 | } |