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 Internet Protocol (IP) output module. | |
7 | * | |
8 | * Version: $Id: ip_output.c,v 1.100 2002/02/01 22:01:03 davem Exp $ | |
9 | * | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Donald Becker, <becker@super.org> | |
13 | * Alan Cox, <Alan.Cox@linux.org> | |
14 | * Richard Underwood | |
15 | * Stefan Becker, <stefanb@yello.ping.de> | |
16 | * Jorge Cwik, <jorge@laser.satlink.net> | |
17 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
18 | * Hirokazu Takahashi, <taka@valinux.co.jp> | |
19 | * | |
20 | * See ip_input.c for original log | |
21 | * | |
22 | * Fixes: | |
23 | * Alan Cox : Missing nonblock feature in ip_build_xmit. | |
24 | * Mike Kilburn : htons() missing in ip_build_xmit. | |
25 | * Bradford Johnson: Fix faulty handling of some frames when | |
26 | * no route is found. | |
27 | * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit | |
28 | * (in case if packet not accepted by | |
29 | * output firewall rules) | |
30 | * Mike McLagan : Routing by source | |
31 | * Alexey Kuznetsov: use new route cache | |
32 | * Andi Kleen: Fix broken PMTU recovery and remove | |
33 | * some redundant tests. | |
34 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. | |
35 | * Andi Kleen : Replace ip_reply with ip_send_reply. | |
36 | * Andi Kleen : Split fast and slow ip_build_xmit path | |
37 | * for decreased register pressure on x86 | |
38 | * and more readibility. | |
39 | * Marc Boucher : When call_out_firewall returns FW_QUEUE, | |
40 | * silently drop skb instead of failing with -EPERM. | |
41 | * Detlev Wengorz : Copy protocol for fragments. | |
42 | * Hirokazu Takahashi: HW checksumming for outgoing UDP | |
43 | * datagrams. | |
44 | * Hirokazu Takahashi: sendfile() on UDP works now. | |
45 | */ | |
46 | ||
47 | #include <asm/uaccess.h> | |
48 | #include <asm/system.h> | |
49 | #include <linux/module.h> | |
50 | #include <linux/types.h> | |
51 | #include <linux/kernel.h> | |
52 | #include <linux/sched.h> | |
53 | #include <linux/mm.h> | |
54 | #include <linux/string.h> | |
55 | #include <linux/errno.h> | |
56 | #include <linux/config.h> | |
57 | ||
58 | #include <linux/socket.h> | |
59 | #include <linux/sockios.h> | |
60 | #include <linux/in.h> | |
61 | #include <linux/inet.h> | |
62 | #include <linux/netdevice.h> | |
63 | #include <linux/etherdevice.h> | |
64 | #include <linux/proc_fs.h> | |
65 | #include <linux/stat.h> | |
66 | #include <linux/init.h> | |
67 | ||
68 | #include <net/snmp.h> | |
69 | #include <net/ip.h> | |
70 | #include <net/protocol.h> | |
71 | #include <net/route.h> | |
1da177e4 LT |
72 | #include <linux/skbuff.h> |
73 | #include <net/sock.h> | |
74 | #include <net/arp.h> | |
75 | #include <net/icmp.h> | |
1da177e4 LT |
76 | #include <net/checksum.h> |
77 | #include <net/inetpeer.h> | |
78 | #include <net/checksum.h> | |
79 | #include <linux/igmp.h> | |
80 | #include <linux/netfilter_ipv4.h> | |
81 | #include <linux/netfilter_bridge.h> | |
82 | #include <linux/mroute.h> | |
83 | #include <linux/netlink.h> | |
6cbb0df7 | 84 | #include <linux/tcp.h> |
1da177e4 | 85 | |
1da177e4 LT |
86 | int sysctl_ip_default_ttl = IPDEFTTL; |
87 | ||
88 | /* Generate a checksum for an outgoing IP datagram. */ | |
89 | __inline__ void ip_send_check(struct iphdr *iph) | |
90 | { | |
91 | iph->check = 0; | |
92 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
93 | } | |
94 | ||
95 | /* dev_loopback_xmit for use with netfilter. */ | |
96 | static int ip_dev_loopback_xmit(struct sk_buff *newskb) | |
97 | { | |
98 | newskb->mac.raw = newskb->data; | |
99 | __skb_pull(newskb, newskb->nh.raw - newskb->data); | |
100 | newskb->pkt_type = PACKET_LOOPBACK; | |
101 | newskb->ip_summed = CHECKSUM_UNNECESSARY; | |
102 | BUG_TRAP(newskb->dst); | |
1da177e4 LT |
103 | netif_rx(newskb); |
104 | return 0; | |
105 | } | |
106 | ||
107 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) | |
108 | { | |
109 | int ttl = inet->uc_ttl; | |
110 | ||
111 | if (ttl < 0) | |
112 | ttl = dst_metric(dst, RTAX_HOPLIMIT); | |
113 | return ttl; | |
114 | } | |
115 | ||
116 | /* | |
117 | * Add an ip header to a skbuff and send it out. | |
118 | * | |
119 | */ | |
120 | int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, | |
121 | u32 saddr, u32 daddr, struct ip_options *opt) | |
122 | { | |
123 | struct inet_sock *inet = inet_sk(sk); | |
124 | struct rtable *rt = (struct rtable *)skb->dst; | |
125 | struct iphdr *iph; | |
126 | ||
127 | /* Build the IP header. */ | |
128 | if (opt) | |
129 | iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr) + opt->optlen); | |
130 | else | |
131 | iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr)); | |
132 | ||
133 | iph->version = 4; | |
134 | iph->ihl = 5; | |
135 | iph->tos = inet->tos; | |
136 | if (ip_dont_fragment(sk, &rt->u.dst)) | |
137 | iph->frag_off = htons(IP_DF); | |
138 | else | |
139 | iph->frag_off = 0; | |
140 | iph->ttl = ip_select_ttl(inet, &rt->u.dst); | |
141 | iph->daddr = rt->rt_dst; | |
142 | iph->saddr = rt->rt_src; | |
143 | iph->protocol = sk->sk_protocol; | |
144 | iph->tot_len = htons(skb->len); | |
145 | ip_select_ident(iph, &rt->u.dst, sk); | |
146 | skb->nh.iph = iph; | |
147 | ||
148 | if (opt && opt->optlen) { | |
149 | iph->ihl += opt->optlen>>2; | |
150 | ip_options_build(skb, opt, daddr, rt, 0); | |
151 | } | |
152 | ip_send_check(iph); | |
153 | ||
154 | skb->priority = sk->sk_priority; | |
155 | ||
156 | /* Send it out. */ | |
157 | return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev, | |
158 | dst_output); | |
159 | } | |
160 | ||
d8c97a94 ACM |
161 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
162 | ||
1da177e4 LT |
163 | static inline int ip_finish_output2(struct sk_buff *skb) |
164 | { | |
165 | struct dst_entry *dst = skb->dst; | |
166 | struct hh_cache *hh = dst->hh; | |
167 | struct net_device *dev = dst->dev; | |
168 | int hh_len = LL_RESERVED_SPACE(dev); | |
169 | ||
170 | /* Be paranoid, rather than too clever. */ | |
171 | if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) { | |
172 | struct sk_buff *skb2; | |
173 | ||
174 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
175 | if (skb2 == NULL) { | |
176 | kfree_skb(skb); | |
177 | return -ENOMEM; | |
178 | } | |
179 | if (skb->sk) | |
180 | skb_set_owner_w(skb2, skb->sk); | |
181 | kfree_skb(skb); | |
182 | skb = skb2; | |
183 | } | |
184 | ||
1da177e4 LT |
185 | if (hh) { |
186 | int hh_alen; | |
187 | ||
188 | read_lock_bh(&hh->hh_lock); | |
189 | hh_alen = HH_DATA_ALIGN(hh->hh_len); | |
190 | memcpy(skb->data - hh_alen, hh->hh_data, hh_alen); | |
191 | read_unlock_bh(&hh->hh_lock); | |
192 | skb_push(skb, hh->hh_len); | |
193 | return hh->hh_output(skb); | |
194 | } else if (dst->neighbour) | |
195 | return dst->neighbour->output(skb); | |
196 | ||
197 | if (net_ratelimit()) | |
198 | printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n"); | |
199 | kfree_skb(skb); | |
200 | return -EINVAL; | |
201 | } | |
202 | ||
33d043d6 | 203 | static inline int ip_finish_output(struct sk_buff *skb) |
1da177e4 LT |
204 | { |
205 | struct net_device *dev = skb->dst->dev; | |
206 | ||
207 | skb->dev = dev; | |
208 | skb->protocol = htons(ETH_P_IP); | |
209 | ||
210 | return NF_HOOK(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev, | |
211 | ip_finish_output2); | |
212 | } | |
213 | ||
214 | int ip_mc_output(struct sk_buff *skb) | |
215 | { | |
216 | struct sock *sk = skb->sk; | |
217 | struct rtable *rt = (struct rtable*)skb->dst; | |
218 | struct net_device *dev = rt->u.dst.dev; | |
219 | ||
220 | /* | |
221 | * If the indicated interface is up and running, send the packet. | |
222 | */ | |
223 | IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS); | |
224 | ||
225 | skb->dev = dev; | |
226 | skb->protocol = htons(ETH_P_IP); | |
227 | ||
228 | /* | |
229 | * Multicasts are looped back for other local users | |
230 | */ | |
231 | ||
232 | if (rt->rt_flags&RTCF_MULTICAST) { | |
233 | if ((!sk || inet_sk(sk)->mc_loop) | |
234 | #ifdef CONFIG_IP_MROUTE | |
235 | /* Small optimization: do not loopback not local frames, | |
236 | which returned after forwarding; they will be dropped | |
237 | by ip_mr_input in any case. | |
238 | Note, that local frames are looped back to be delivered | |
239 | to local recipients. | |
240 | ||
241 | This check is duplicated in ip_mr_input at the moment. | |
242 | */ | |
243 | && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED)) | |
244 | #endif | |
245 | ) { | |
246 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
247 | if (newskb) | |
248 | NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL, | |
249 | newskb->dev, | |
250 | ip_dev_loopback_xmit); | |
251 | } | |
252 | ||
253 | /* Multicasts with ttl 0 must not go beyond the host */ | |
254 | ||
255 | if (skb->nh.iph->ttl == 0) { | |
256 | kfree_skb(skb); | |
257 | return 0; | |
258 | } | |
259 | } | |
260 | ||
261 | if (rt->rt_flags&RTCF_BROADCAST) { | |
262 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
263 | if (newskb) | |
264 | NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL, | |
265 | newskb->dev, ip_dev_loopback_xmit); | |
266 | } | |
267 | ||
268 | if (skb->len > dst_mtu(&rt->u.dst)) | |
269 | return ip_fragment(skb, ip_finish_output); | |
270 | else | |
271 | return ip_finish_output(skb); | |
272 | } | |
273 | ||
274 | int ip_output(struct sk_buff *skb) | |
275 | { | |
276 | IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS); | |
277 | ||
e89e9cf5 AR |
278 | if (skb->len > dst_mtu(skb->dst) && |
279 | !(skb_shinfo(skb)->ufo_size || skb_shinfo(skb)->tso_size)) | |
1da177e4 LT |
280 | return ip_fragment(skb, ip_finish_output); |
281 | else | |
282 | return ip_finish_output(skb); | |
283 | } | |
284 | ||
285 | int ip_queue_xmit(struct sk_buff *skb, int ipfragok) | |
286 | { | |
287 | struct sock *sk = skb->sk; | |
288 | struct inet_sock *inet = inet_sk(sk); | |
289 | struct ip_options *opt = inet->opt; | |
290 | struct rtable *rt; | |
291 | struct iphdr *iph; | |
292 | ||
293 | /* Skip all of this if the packet is already routed, | |
294 | * f.e. by something like SCTP. | |
295 | */ | |
296 | rt = (struct rtable *) skb->dst; | |
297 | if (rt != NULL) | |
298 | goto packet_routed; | |
299 | ||
300 | /* Make sure we can route this packet. */ | |
301 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
302 | if (rt == NULL) { | |
303 | u32 daddr; | |
304 | ||
305 | /* Use correct destination address if we have options. */ | |
306 | daddr = inet->daddr; | |
307 | if(opt && opt->srr) | |
308 | daddr = opt->faddr; | |
309 | ||
310 | { | |
311 | struct flowi fl = { .oif = sk->sk_bound_dev_if, | |
312 | .nl_u = { .ip4_u = | |
313 | { .daddr = daddr, | |
314 | .saddr = inet->saddr, | |
315 | .tos = RT_CONN_FLAGS(sk) } }, | |
316 | .proto = sk->sk_protocol, | |
317 | .uli_u = { .ports = | |
318 | { .sport = inet->sport, | |
319 | .dport = inet->dport } } }; | |
320 | ||
321 | /* If this fails, retransmit mechanism of transport layer will | |
322 | * keep trying until route appears or the connection times | |
323 | * itself out. | |
324 | */ | |
325 | if (ip_route_output_flow(&rt, &fl, sk, 0)) | |
326 | goto no_route; | |
327 | } | |
6cbb0df7 | 328 | sk_setup_caps(sk, &rt->u.dst); |
1da177e4 LT |
329 | } |
330 | skb->dst = dst_clone(&rt->u.dst); | |
331 | ||
332 | packet_routed: | |
333 | if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) | |
334 | goto no_route; | |
335 | ||
336 | /* OK, we know where to send it, allocate and build IP header. */ | |
337 | iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0)); | |
338 | *((__u16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); | |
339 | iph->tot_len = htons(skb->len); | |
340 | if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok) | |
341 | iph->frag_off = htons(IP_DF); | |
342 | else | |
343 | iph->frag_off = 0; | |
344 | iph->ttl = ip_select_ttl(inet, &rt->u.dst); | |
345 | iph->protocol = sk->sk_protocol; | |
346 | iph->saddr = rt->rt_src; | |
347 | iph->daddr = rt->rt_dst; | |
348 | skb->nh.iph = iph; | |
349 | /* Transport layer set skb->h.foo itself. */ | |
350 | ||
351 | if (opt && opt->optlen) { | |
352 | iph->ihl += opt->optlen >> 2; | |
353 | ip_options_build(skb, opt, inet->daddr, rt, 0); | |
354 | } | |
355 | ||
356 | ip_select_ident_more(iph, &rt->u.dst, sk, skb_shinfo(skb)->tso_segs); | |
357 | ||
358 | /* Add an IP checksum. */ | |
359 | ip_send_check(iph); | |
360 | ||
361 | skb->priority = sk->sk_priority; | |
362 | ||
363 | return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev, | |
364 | dst_output); | |
365 | ||
366 | no_route: | |
367 | IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES); | |
368 | kfree_skb(skb); | |
369 | return -EHOSTUNREACH; | |
370 | } | |
371 | ||
372 | ||
373 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) | |
374 | { | |
375 | to->pkt_type = from->pkt_type; | |
376 | to->priority = from->priority; | |
377 | to->protocol = from->protocol; | |
1da177e4 LT |
378 | dst_release(to->dst); |
379 | to->dst = dst_clone(from->dst); | |
380 | to->dev = from->dev; | |
381 | ||
382 | /* Copy the flags to each fragment. */ | |
383 | IPCB(to)->flags = IPCB(from)->flags; | |
384 | ||
385 | #ifdef CONFIG_NET_SCHED | |
386 | to->tc_index = from->tc_index; | |
387 | #endif | |
388 | #ifdef CONFIG_NETFILTER | |
389 | to->nfmark = from->nfmark; | |
1da177e4 LT |
390 | /* Connection association is same as pre-frag packet */ |
391 | nf_conntrack_put(to->nfct); | |
392 | to->nfct = from->nfct; | |
393 | nf_conntrack_get(to->nfct); | |
394 | to->nfctinfo = from->nfctinfo; | |
c98d80ed JA |
395 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
396 | to->ipvs_property = from->ipvs_property; | |
397 | #endif | |
1da177e4 LT |
398 | #ifdef CONFIG_BRIDGE_NETFILTER |
399 | nf_bridge_put(to->nf_bridge); | |
400 | to->nf_bridge = from->nf_bridge; | |
401 | nf_bridge_get(to->nf_bridge); | |
402 | #endif | |
1da177e4 LT |
403 | #endif |
404 | } | |
405 | ||
406 | /* | |
407 | * This IP datagram is too large to be sent in one piece. Break it up into | |
408 | * smaller pieces (each of size equal to IP header plus | |
409 | * a block of the data of the original IP data part) that will yet fit in a | |
410 | * single device frame, and queue such a frame for sending. | |
411 | */ | |
412 | ||
413 | int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*)) | |
414 | { | |
415 | struct iphdr *iph; | |
416 | int raw = 0; | |
417 | int ptr; | |
418 | struct net_device *dev; | |
419 | struct sk_buff *skb2; | |
420 | unsigned int mtu, hlen, left, len, ll_rs; | |
421 | int offset; | |
422 | int not_last_frag; | |
423 | struct rtable *rt = (struct rtable*)skb->dst; | |
424 | int err = 0; | |
425 | ||
426 | dev = rt->u.dst.dev; | |
427 | ||
428 | /* | |
429 | * Point into the IP datagram header. | |
430 | */ | |
431 | ||
432 | iph = skb->nh.iph; | |
433 | ||
434 | if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) { | |
435 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, | |
436 | htonl(dst_mtu(&rt->u.dst))); | |
437 | kfree_skb(skb); | |
438 | return -EMSGSIZE; | |
439 | } | |
440 | ||
441 | /* | |
442 | * Setup starting values. | |
443 | */ | |
444 | ||
445 | hlen = iph->ihl * 4; | |
446 | mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */ | |
447 | ||
448 | /* When frag_list is given, use it. First, check its validity: | |
449 | * some transformers could create wrong frag_list or break existing | |
450 | * one, it is not prohibited. In this case fall back to copying. | |
451 | * | |
452 | * LATER: this step can be merged to real generation of fragments, | |
453 | * we can switch to copy when see the first bad fragment. | |
454 | */ | |
455 | if (skb_shinfo(skb)->frag_list) { | |
456 | struct sk_buff *frag; | |
457 | int first_len = skb_pagelen(skb); | |
458 | ||
459 | if (first_len - hlen > mtu || | |
460 | ((first_len - hlen) & 7) || | |
461 | (iph->frag_off & htons(IP_MF|IP_OFFSET)) || | |
462 | skb_cloned(skb)) | |
463 | goto slow_path; | |
464 | ||
465 | for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) { | |
466 | /* Correct geometry. */ | |
467 | if (frag->len > mtu || | |
468 | ((frag->len & 7) && frag->next) || | |
469 | skb_headroom(frag) < hlen) | |
470 | goto slow_path; | |
471 | ||
472 | /* Partially cloned skb? */ | |
473 | if (skb_shared(frag)) | |
474 | goto slow_path; | |
2fdba6b0 HX |
475 | |
476 | BUG_ON(frag->sk); | |
477 | if (skb->sk) { | |
478 | sock_hold(skb->sk); | |
479 | frag->sk = skb->sk; | |
480 | frag->destructor = sock_wfree; | |
481 | skb->truesize -= frag->truesize; | |
482 | } | |
1da177e4 LT |
483 | } |
484 | ||
485 | /* Everything is OK. Generate! */ | |
486 | ||
487 | err = 0; | |
488 | offset = 0; | |
489 | frag = skb_shinfo(skb)->frag_list; | |
490 | skb_shinfo(skb)->frag_list = NULL; | |
491 | skb->data_len = first_len - skb_headlen(skb); | |
492 | skb->len = first_len; | |
493 | iph->tot_len = htons(first_len); | |
494 | iph->frag_off = htons(IP_MF); | |
495 | ip_send_check(iph); | |
496 | ||
497 | for (;;) { | |
498 | /* Prepare header of the next frame, | |
499 | * before previous one went down. */ | |
500 | if (frag) { | |
501 | frag->ip_summed = CHECKSUM_NONE; | |
502 | frag->h.raw = frag->data; | |
503 | frag->nh.raw = __skb_push(frag, hlen); | |
504 | memcpy(frag->nh.raw, iph, hlen); | |
505 | iph = frag->nh.iph; | |
506 | iph->tot_len = htons(frag->len); | |
507 | ip_copy_metadata(frag, skb); | |
508 | if (offset == 0) | |
509 | ip_options_fragment(frag); | |
510 | offset += skb->len - hlen; | |
511 | iph->frag_off = htons(offset>>3); | |
512 | if (frag->next != NULL) | |
513 | iph->frag_off |= htons(IP_MF); | |
514 | /* Ready, complete checksum */ | |
515 | ip_send_check(iph); | |
516 | } | |
517 | ||
518 | err = output(skb); | |
519 | ||
520 | if (err || !frag) | |
521 | break; | |
522 | ||
523 | skb = frag; | |
524 | frag = skb->next; | |
525 | skb->next = NULL; | |
526 | } | |
527 | ||
528 | if (err == 0) { | |
529 | IP_INC_STATS(IPSTATS_MIB_FRAGOKS); | |
530 | return 0; | |
531 | } | |
532 | ||
533 | while (frag) { | |
534 | skb = frag->next; | |
535 | kfree_skb(frag); | |
536 | frag = skb; | |
537 | } | |
538 | IP_INC_STATS(IPSTATS_MIB_FRAGFAILS); | |
539 | return err; | |
540 | } | |
541 | ||
542 | slow_path: | |
543 | left = skb->len - hlen; /* Space per frame */ | |
544 | ptr = raw + hlen; /* Where to start from */ | |
545 | ||
546 | #ifdef CONFIG_BRIDGE_NETFILTER | |
547 | /* for bridged IP traffic encapsulated inside f.e. a vlan header, | |
548 | * we need to make room for the encapsulating header */ | |
549 | ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb)); | |
550 | mtu -= nf_bridge_pad(skb); | |
551 | #else | |
552 | ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev); | |
553 | #endif | |
554 | /* | |
555 | * Fragment the datagram. | |
556 | */ | |
557 | ||
558 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
559 | not_last_frag = iph->frag_off & htons(IP_MF); | |
560 | ||
561 | /* | |
562 | * Keep copying data until we run out. | |
563 | */ | |
564 | ||
565 | while(left > 0) { | |
566 | len = left; | |
567 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
568 | if (len > mtu) | |
569 | len = mtu; | |
570 | /* IF: we are not sending upto and including the packet end | |
571 | then align the next start on an eight byte boundary */ | |
572 | if (len < left) { | |
573 | len &= ~7; | |
574 | } | |
575 | /* | |
576 | * Allocate buffer. | |
577 | */ | |
578 | ||
579 | if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { | |
64ce2073 | 580 | NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); |
1da177e4 LT |
581 | err = -ENOMEM; |
582 | goto fail; | |
583 | } | |
584 | ||
585 | /* | |
586 | * Set up data on packet | |
587 | */ | |
588 | ||
589 | ip_copy_metadata(skb2, skb); | |
590 | skb_reserve(skb2, ll_rs); | |
591 | skb_put(skb2, len + hlen); | |
592 | skb2->nh.raw = skb2->data; | |
593 | skb2->h.raw = skb2->data + hlen; | |
594 | ||
595 | /* | |
596 | * Charge the memory for the fragment to any owner | |
597 | * it might possess | |
598 | */ | |
599 | ||
600 | if (skb->sk) | |
601 | skb_set_owner_w(skb2, skb->sk); | |
602 | ||
603 | /* | |
604 | * Copy the packet header into the new buffer. | |
605 | */ | |
606 | ||
607 | memcpy(skb2->nh.raw, skb->data, hlen); | |
608 | ||
609 | /* | |
610 | * Copy a block of the IP datagram. | |
611 | */ | |
612 | if (skb_copy_bits(skb, ptr, skb2->h.raw, len)) | |
613 | BUG(); | |
614 | left -= len; | |
615 | ||
616 | /* | |
617 | * Fill in the new header fields. | |
618 | */ | |
619 | iph = skb2->nh.iph; | |
620 | iph->frag_off = htons((offset >> 3)); | |
621 | ||
622 | /* ANK: dirty, but effective trick. Upgrade options only if | |
623 | * the segment to be fragmented was THE FIRST (otherwise, | |
624 | * options are already fixed) and make it ONCE | |
625 | * on the initial skb, so that all the following fragments | |
626 | * will inherit fixed options. | |
627 | */ | |
628 | if (offset == 0) | |
629 | ip_options_fragment(skb); | |
630 | ||
631 | /* | |
632 | * Added AC : If we are fragmenting a fragment that's not the | |
633 | * last fragment then keep MF on each bit | |
634 | */ | |
635 | if (left > 0 || not_last_frag) | |
636 | iph->frag_off |= htons(IP_MF); | |
637 | ptr += len; | |
638 | offset += len; | |
639 | ||
640 | /* | |
641 | * Put this fragment into the sending queue. | |
642 | */ | |
643 | ||
644 | IP_INC_STATS(IPSTATS_MIB_FRAGCREATES); | |
645 | ||
646 | iph->tot_len = htons(len + hlen); | |
647 | ||
648 | ip_send_check(iph); | |
649 | ||
650 | err = output(skb2); | |
651 | if (err) | |
652 | goto fail; | |
653 | } | |
654 | kfree_skb(skb); | |
655 | IP_INC_STATS(IPSTATS_MIB_FRAGOKS); | |
656 | return err; | |
657 | ||
658 | fail: | |
659 | kfree_skb(skb); | |
660 | IP_INC_STATS(IPSTATS_MIB_FRAGFAILS); | |
661 | return err; | |
662 | } | |
663 | ||
664 | int | |
665 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
666 | { | |
667 | struct iovec *iov = from; | |
668 | ||
669 | if (skb->ip_summed == CHECKSUM_HW) { | |
670 | if (memcpy_fromiovecend(to, iov, offset, len) < 0) | |
671 | return -EFAULT; | |
672 | } else { | |
673 | unsigned int csum = 0; | |
674 | if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) | |
675 | return -EFAULT; | |
676 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
677 | } | |
678 | return 0; | |
679 | } | |
680 | ||
681 | static inline unsigned int | |
682 | csum_page(struct page *page, int offset, int copy) | |
683 | { | |
684 | char *kaddr; | |
685 | unsigned int csum; | |
686 | kaddr = kmap(page); | |
687 | csum = csum_partial(kaddr + offset, copy, 0); | |
688 | kunmap(page); | |
689 | return csum; | |
690 | } | |
691 | ||
e89e9cf5 AR |
692 | inline int ip_ufo_append_data(struct sock *sk, |
693 | int getfrag(void *from, char *to, int offset, int len, | |
694 | int odd, struct sk_buff *skb), | |
695 | void *from, int length, int hh_len, int fragheaderlen, | |
696 | int transhdrlen, int mtu,unsigned int flags) | |
697 | { | |
698 | struct sk_buff *skb; | |
699 | int err; | |
700 | ||
701 | /* There is support for UDP fragmentation offload by network | |
702 | * device, so create one single skb packet containing complete | |
703 | * udp datagram | |
704 | */ | |
705 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) { | |
706 | skb = sock_alloc_send_skb(sk, | |
707 | hh_len + fragheaderlen + transhdrlen + 20, | |
708 | (flags & MSG_DONTWAIT), &err); | |
709 | ||
710 | if (skb == NULL) | |
711 | return err; | |
712 | ||
713 | /* reserve space for Hardware header */ | |
714 | skb_reserve(skb, hh_len); | |
715 | ||
716 | /* create space for UDP/IP header */ | |
717 | skb_put(skb,fragheaderlen + transhdrlen); | |
718 | ||
719 | /* initialize network header pointer */ | |
720 | skb->nh.raw = skb->data; | |
721 | ||
722 | /* initialize protocol header pointer */ | |
723 | skb->h.raw = skb->data + fragheaderlen; | |
724 | ||
725 | skb->ip_summed = CHECKSUM_HW; | |
726 | skb->csum = 0; | |
727 | sk->sk_sndmsg_off = 0; | |
728 | } | |
729 | ||
730 | err = skb_append_datato_frags(sk,skb, getfrag, from, | |
731 | (length - transhdrlen)); | |
732 | if (!err) { | |
733 | /* specify the length of each IP datagram fragment*/ | |
734 | skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen); | |
735 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
736 | ||
737 | return 0; | |
738 | } | |
739 | /* There is not enough support do UFO , | |
740 | * so follow normal path | |
741 | */ | |
742 | kfree_skb(skb); | |
743 | return err; | |
744 | } | |
745 | ||
1da177e4 LT |
746 | /* |
747 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
748 | * from many pieces of data. Each pieces will be holded on the socket | |
749 | * until ip_push_pending_frames() is called. Each piece can be a page | |
750 | * or non-page data. | |
751 | * | |
752 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
753 | * this interface potentially. | |
754 | * | |
755 | * LATER: length must be adjusted by pad at tail, when it is required. | |
756 | */ | |
757 | int ip_append_data(struct sock *sk, | |
758 | int getfrag(void *from, char *to, int offset, int len, | |
759 | int odd, struct sk_buff *skb), | |
760 | void *from, int length, int transhdrlen, | |
761 | struct ipcm_cookie *ipc, struct rtable *rt, | |
762 | unsigned int flags) | |
763 | { | |
764 | struct inet_sock *inet = inet_sk(sk); | |
765 | struct sk_buff *skb; | |
766 | ||
767 | struct ip_options *opt = NULL; | |
768 | int hh_len; | |
769 | int exthdrlen; | |
770 | int mtu; | |
771 | int copy; | |
772 | int err; | |
773 | int offset = 0; | |
774 | unsigned int maxfraglen, fragheaderlen; | |
775 | int csummode = CHECKSUM_NONE; | |
776 | ||
777 | if (flags&MSG_PROBE) | |
778 | return 0; | |
779 | ||
780 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
781 | /* | |
782 | * setup for corking. | |
783 | */ | |
784 | opt = ipc->opt; | |
785 | if (opt) { | |
786 | if (inet->cork.opt == NULL) { | |
787 | inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation); | |
788 | if (unlikely(inet->cork.opt == NULL)) | |
789 | return -ENOBUFS; | |
790 | } | |
791 | memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen); | |
792 | inet->cork.flags |= IPCORK_OPT; | |
793 | inet->cork.addr = ipc->addr; | |
794 | } | |
795 | dst_hold(&rt->u.dst); | |
796 | inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path); | |
797 | inet->cork.rt = rt; | |
798 | inet->cork.length = 0; | |
799 | sk->sk_sndmsg_page = NULL; | |
800 | sk->sk_sndmsg_off = 0; | |
801 | if ((exthdrlen = rt->u.dst.header_len) != 0) { | |
802 | length += exthdrlen; | |
803 | transhdrlen += exthdrlen; | |
804 | } | |
805 | } else { | |
806 | rt = inet->cork.rt; | |
807 | if (inet->cork.flags & IPCORK_OPT) | |
808 | opt = inet->cork.opt; | |
809 | ||
810 | transhdrlen = 0; | |
811 | exthdrlen = 0; | |
812 | mtu = inet->cork.fragsize; | |
813 | } | |
814 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
815 | ||
816 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
817 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
818 | ||
819 | if (inet->cork.length + length > 0xFFFF - fragheaderlen) { | |
820 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen); | |
821 | return -EMSGSIZE; | |
822 | } | |
823 | ||
824 | /* | |
825 | * transhdrlen > 0 means that this is the first fragment and we wish | |
826 | * it won't be fragmented in the future. | |
827 | */ | |
828 | if (transhdrlen && | |
829 | length + fragheaderlen <= mtu && | |
830 | rt->u.dst.dev->features&(NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM) && | |
831 | !exthdrlen) | |
832 | csummode = CHECKSUM_HW; | |
833 | ||
834 | inet->cork.length += length; | |
e89e9cf5 AR |
835 | if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) && |
836 | (rt->u.dst.dev->features & NETIF_F_UFO)) { | |
837 | ||
838 | if(ip_ufo_append_data(sk, getfrag, from, length, hh_len, | |
839 | fragheaderlen, transhdrlen, mtu, flags)) | |
840 | goto error; | |
841 | ||
842 | return 0; | |
843 | } | |
1da177e4 LT |
844 | |
845 | /* So, what's going on in the loop below? | |
846 | * | |
847 | * We use calculated fragment length to generate chained skb, | |
848 | * each of segments is IP fragment ready for sending to network after | |
849 | * adding appropriate IP header. | |
850 | */ | |
851 | ||
852 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
853 | goto alloc_new_skb; | |
854 | ||
855 | while (length > 0) { | |
856 | /* Check if the remaining data fits into current packet. */ | |
857 | copy = mtu - skb->len; | |
858 | if (copy < length) | |
859 | copy = maxfraglen - skb->len; | |
860 | if (copy <= 0) { | |
861 | char *data; | |
862 | unsigned int datalen; | |
863 | unsigned int fraglen; | |
864 | unsigned int fraggap; | |
865 | unsigned int alloclen; | |
866 | struct sk_buff *skb_prev; | |
867 | alloc_new_skb: | |
868 | skb_prev = skb; | |
869 | if (skb_prev) | |
870 | fraggap = skb_prev->len - maxfraglen; | |
871 | else | |
872 | fraggap = 0; | |
873 | ||
874 | /* | |
875 | * If remaining data exceeds the mtu, | |
876 | * we know we need more fragment(s). | |
877 | */ | |
878 | datalen = length + fraggap; | |
879 | if (datalen > mtu - fragheaderlen) | |
880 | datalen = maxfraglen - fragheaderlen; | |
881 | fraglen = datalen + fragheaderlen; | |
882 | ||
883 | if ((flags & MSG_MORE) && | |
884 | !(rt->u.dst.dev->features&NETIF_F_SG)) | |
885 | alloclen = mtu; | |
886 | else | |
887 | alloclen = datalen + fragheaderlen; | |
888 | ||
889 | /* The last fragment gets additional space at tail. | |
890 | * Note, with MSG_MORE we overallocate on fragments, | |
891 | * because we have no idea what fragment will be | |
892 | * the last. | |
893 | */ | |
894 | if (datalen == length) | |
895 | alloclen += rt->u.dst.trailer_len; | |
896 | ||
897 | if (transhdrlen) { | |
898 | skb = sock_alloc_send_skb(sk, | |
899 | alloclen + hh_len + 15, | |
900 | (flags & MSG_DONTWAIT), &err); | |
901 | } else { | |
902 | skb = NULL; | |
903 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
904 | 2 * sk->sk_sndbuf) | |
905 | skb = sock_wmalloc(sk, | |
906 | alloclen + hh_len + 15, 1, | |
907 | sk->sk_allocation); | |
908 | if (unlikely(skb == NULL)) | |
909 | err = -ENOBUFS; | |
910 | } | |
911 | if (skb == NULL) | |
912 | goto error; | |
913 | ||
914 | /* | |
915 | * Fill in the control structures | |
916 | */ | |
917 | skb->ip_summed = csummode; | |
918 | skb->csum = 0; | |
919 | skb_reserve(skb, hh_len); | |
920 | ||
921 | /* | |
922 | * Find where to start putting bytes. | |
923 | */ | |
924 | data = skb_put(skb, fraglen); | |
925 | skb->nh.raw = data + exthdrlen; | |
926 | data += fragheaderlen; | |
927 | skb->h.raw = data + exthdrlen; | |
928 | ||
929 | if (fraggap) { | |
930 | skb->csum = skb_copy_and_csum_bits( | |
931 | skb_prev, maxfraglen, | |
932 | data + transhdrlen, fraggap, 0); | |
933 | skb_prev->csum = csum_sub(skb_prev->csum, | |
934 | skb->csum); | |
935 | data += fraggap; | |
936 | skb_trim(skb_prev, maxfraglen); | |
937 | } | |
938 | ||
939 | copy = datalen - transhdrlen - fraggap; | |
940 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
941 | err = -EFAULT; | |
942 | kfree_skb(skb); | |
943 | goto error; | |
944 | } | |
945 | ||
946 | offset += copy; | |
947 | length -= datalen - fraggap; | |
948 | transhdrlen = 0; | |
949 | exthdrlen = 0; | |
950 | csummode = CHECKSUM_NONE; | |
951 | ||
952 | /* | |
953 | * Put the packet on the pending queue. | |
954 | */ | |
955 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
956 | continue; | |
957 | } | |
958 | ||
959 | if (copy > length) | |
960 | copy = length; | |
961 | ||
962 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) { | |
963 | unsigned int off; | |
964 | ||
965 | off = skb->len; | |
966 | if (getfrag(from, skb_put(skb, copy), | |
967 | offset, copy, off, skb) < 0) { | |
968 | __skb_trim(skb, off); | |
969 | err = -EFAULT; | |
970 | goto error; | |
971 | } | |
972 | } else { | |
973 | int i = skb_shinfo(skb)->nr_frags; | |
974 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1]; | |
975 | struct page *page = sk->sk_sndmsg_page; | |
976 | int off = sk->sk_sndmsg_off; | |
977 | unsigned int left; | |
978 | ||
979 | if (page && (left = PAGE_SIZE - off) > 0) { | |
980 | if (copy >= left) | |
981 | copy = left; | |
982 | if (page != frag->page) { | |
983 | if (i == MAX_SKB_FRAGS) { | |
984 | err = -EMSGSIZE; | |
985 | goto error; | |
986 | } | |
987 | get_page(page); | |
988 | skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0); | |
989 | frag = &skb_shinfo(skb)->frags[i]; | |
990 | } | |
991 | } else if (i < MAX_SKB_FRAGS) { | |
992 | if (copy > PAGE_SIZE) | |
993 | copy = PAGE_SIZE; | |
994 | page = alloc_pages(sk->sk_allocation, 0); | |
995 | if (page == NULL) { | |
996 | err = -ENOMEM; | |
997 | goto error; | |
998 | } | |
999 | sk->sk_sndmsg_page = page; | |
1000 | sk->sk_sndmsg_off = 0; | |
1001 | ||
1002 | skb_fill_page_desc(skb, i, page, 0, 0); | |
1003 | frag = &skb_shinfo(skb)->frags[i]; | |
1004 | skb->truesize += PAGE_SIZE; | |
1005 | atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); | |
1006 | } else { | |
1007 | err = -EMSGSIZE; | |
1008 | goto error; | |
1009 | } | |
1010 | if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) { | |
1011 | err = -EFAULT; | |
1012 | goto error; | |
1013 | } | |
1014 | sk->sk_sndmsg_off += copy; | |
1015 | frag->size += copy; | |
1016 | skb->len += copy; | |
1017 | skb->data_len += copy; | |
1018 | } | |
1019 | offset += copy; | |
1020 | length -= copy; | |
1021 | } | |
1022 | ||
1023 | return 0; | |
1024 | ||
1025 | error: | |
1026 | inet->cork.length -= length; | |
1027 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
1028 | return err; | |
1029 | } | |
1030 | ||
1031 | ssize_t ip_append_page(struct sock *sk, struct page *page, | |
1032 | int offset, size_t size, int flags) | |
1033 | { | |
1034 | struct inet_sock *inet = inet_sk(sk); | |
1035 | struct sk_buff *skb; | |
1036 | struct rtable *rt; | |
1037 | struct ip_options *opt = NULL; | |
1038 | int hh_len; | |
1039 | int mtu; | |
1040 | int len; | |
1041 | int err; | |
1042 | unsigned int maxfraglen, fragheaderlen, fraggap; | |
1043 | ||
1044 | if (inet->hdrincl) | |
1045 | return -EPERM; | |
1046 | ||
1047 | if (flags&MSG_PROBE) | |
1048 | return 0; | |
1049 | ||
1050 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1051 | return -EINVAL; | |
1052 | ||
1053 | rt = inet->cork.rt; | |
1054 | if (inet->cork.flags & IPCORK_OPT) | |
1055 | opt = inet->cork.opt; | |
1056 | ||
1057 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) | |
1058 | return -EOPNOTSUPP; | |
1059 | ||
1060 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
1061 | mtu = inet->cork.fragsize; | |
1062 | ||
1063 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1064 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
1065 | ||
1066 | if (inet->cork.length + size > 0xFFFF - fragheaderlen) { | |
1067 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu); | |
1068 | return -EMSGSIZE; | |
1069 | } | |
1070 | ||
1071 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
1072 | return -EINVAL; | |
1073 | ||
1074 | inet->cork.length += size; | |
e89e9cf5 AR |
1075 | if ((sk->sk_protocol == IPPROTO_UDP) && |
1076 | (rt->u.dst.dev->features & NETIF_F_UFO)) | |
1077 | skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen); | |
1078 | ||
1da177e4 LT |
1079 | |
1080 | while (size > 0) { | |
1081 | int i; | |
1082 | ||
e89e9cf5 AR |
1083 | if (skb_shinfo(skb)->ufo_size) |
1084 | len = size; | |
1085 | else { | |
1086 | ||
1087 | /* Check if the remaining data fits into current packet. */ | |
1088 | len = mtu - skb->len; | |
1089 | if (len < size) | |
1090 | len = maxfraglen - skb->len; | |
1091 | } | |
1da177e4 LT |
1092 | if (len <= 0) { |
1093 | struct sk_buff *skb_prev; | |
1094 | char *data; | |
1095 | struct iphdr *iph; | |
1096 | int alloclen; | |
1097 | ||
1098 | skb_prev = skb; | |
0d0d2bba | 1099 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1100 | |
1101 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1102 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1103 | if (unlikely(!skb)) { | |
1104 | err = -ENOBUFS; | |
1105 | goto error; | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Fill in the control structures | |
1110 | */ | |
1111 | skb->ip_summed = CHECKSUM_NONE; | |
1112 | skb->csum = 0; | |
1113 | skb_reserve(skb, hh_len); | |
1114 | ||
1115 | /* | |
1116 | * Find where to start putting bytes. | |
1117 | */ | |
1118 | data = skb_put(skb, fragheaderlen + fraggap); | |
1119 | skb->nh.iph = iph = (struct iphdr *)data; | |
1120 | data += fragheaderlen; | |
1121 | skb->h.raw = data; | |
1122 | ||
1123 | if (fraggap) { | |
1124 | skb->csum = skb_copy_and_csum_bits( | |
1125 | skb_prev, maxfraglen, | |
1126 | data, fraggap, 0); | |
1127 | skb_prev->csum = csum_sub(skb_prev->csum, | |
1128 | skb->csum); | |
1129 | skb_trim(skb_prev, maxfraglen); | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * Put the packet on the pending queue. | |
1134 | */ | |
1135 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1136 | continue; | |
1137 | } | |
1138 | ||
1139 | i = skb_shinfo(skb)->nr_frags; | |
1140 | if (len > size) | |
1141 | len = size; | |
1142 | if (skb_can_coalesce(skb, i, page, offset)) { | |
1143 | skb_shinfo(skb)->frags[i-1].size += len; | |
1144 | } else if (i < MAX_SKB_FRAGS) { | |
1145 | get_page(page); | |
1146 | skb_fill_page_desc(skb, i, page, offset, len); | |
1147 | } else { | |
1148 | err = -EMSGSIZE; | |
1149 | goto error; | |
1150 | } | |
1151 | ||
1152 | if (skb->ip_summed == CHECKSUM_NONE) { | |
1153 | unsigned int csum; | |
1154 | csum = csum_page(page, offset, len); | |
1155 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1156 | } | |
1157 | ||
1158 | skb->len += len; | |
1159 | skb->data_len += len; | |
1160 | offset += len; | |
1161 | size -= len; | |
1162 | } | |
1163 | return 0; | |
1164 | ||
1165 | error: | |
1166 | inet->cork.length -= size; | |
1167 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
1168 | return err; | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * Combined all pending IP fragments on the socket as one IP datagram | |
1173 | * and push them out. | |
1174 | */ | |
1175 | int ip_push_pending_frames(struct sock *sk) | |
1176 | { | |
1177 | struct sk_buff *skb, *tmp_skb; | |
1178 | struct sk_buff **tail_skb; | |
1179 | struct inet_sock *inet = inet_sk(sk); | |
1180 | struct ip_options *opt = NULL; | |
1181 | struct rtable *rt = inet->cork.rt; | |
1182 | struct iphdr *iph; | |
1183 | int df = 0; | |
1184 | __u8 ttl; | |
1185 | int err = 0; | |
1186 | ||
1187 | if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL) | |
1188 | goto out; | |
1189 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1190 | ||
1191 | /* move skb->data to ip header from ext header */ | |
1192 | if (skb->data < skb->nh.raw) | |
1193 | __skb_pull(skb, skb->nh.raw - skb->data); | |
1194 | while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { | |
1195 | __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw); | |
1196 | *tail_skb = tmp_skb; | |
1197 | tail_skb = &(tmp_skb->next); | |
1198 | skb->len += tmp_skb->len; | |
1199 | skb->data_len += tmp_skb->len; | |
1200 | skb->truesize += tmp_skb->truesize; | |
1201 | __sock_put(tmp_skb->sk); | |
1202 | tmp_skb->destructor = NULL; | |
1203 | tmp_skb->sk = NULL; | |
1204 | } | |
1205 | ||
1206 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1207 | * to fragment the frame generated here. No matter, what transforms | |
1208 | * how transforms change size of the packet, it will come out. | |
1209 | */ | |
1210 | if (inet->pmtudisc != IP_PMTUDISC_DO) | |
1211 | skb->local_df = 1; | |
1212 | ||
1213 | /* DF bit is set when we want to see DF on outgoing frames. | |
1214 | * If local_df is set too, we still allow to fragment this frame | |
1215 | * locally. */ | |
1216 | if (inet->pmtudisc == IP_PMTUDISC_DO || | |
1217 | (skb->len <= dst_mtu(&rt->u.dst) && | |
1218 | ip_dont_fragment(sk, &rt->u.dst))) | |
1219 | df = htons(IP_DF); | |
1220 | ||
1221 | if (inet->cork.flags & IPCORK_OPT) | |
1222 | opt = inet->cork.opt; | |
1223 | ||
1224 | if (rt->rt_type == RTN_MULTICAST) | |
1225 | ttl = inet->mc_ttl; | |
1226 | else | |
1227 | ttl = ip_select_ttl(inet, &rt->u.dst); | |
1228 | ||
1229 | iph = (struct iphdr *)skb->data; | |
1230 | iph->version = 4; | |
1231 | iph->ihl = 5; | |
1232 | if (opt) { | |
1233 | iph->ihl += opt->optlen>>2; | |
1234 | ip_options_build(skb, opt, inet->cork.addr, rt, 0); | |
1235 | } | |
1236 | iph->tos = inet->tos; | |
1237 | iph->tot_len = htons(skb->len); | |
1238 | iph->frag_off = df; | |
1239 | if (!df) { | |
1240 | __ip_select_ident(iph, &rt->u.dst, 0); | |
1241 | } else { | |
1242 | iph->id = htons(inet->id++); | |
1243 | } | |
1244 | iph->ttl = ttl; | |
1245 | iph->protocol = sk->sk_protocol; | |
1246 | iph->saddr = rt->rt_src; | |
1247 | iph->daddr = rt->rt_dst; | |
1248 | ip_send_check(iph); | |
1249 | ||
1250 | skb->priority = sk->sk_priority; | |
1251 | skb->dst = dst_clone(&rt->u.dst); | |
1252 | ||
1253 | /* Netfilter gets whole the not fragmented skb. */ | |
1254 | err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, | |
1255 | skb->dst->dev, dst_output); | |
1256 | if (err) { | |
1257 | if (err > 0) | |
1258 | err = inet->recverr ? net_xmit_errno(err) : 0; | |
1259 | if (err) | |
1260 | goto error; | |
1261 | } | |
1262 | ||
1263 | out: | |
1264 | inet->cork.flags &= ~IPCORK_OPT; | |
1265 | if (inet->cork.opt) { | |
1266 | kfree(inet->cork.opt); | |
1267 | inet->cork.opt = NULL; | |
1268 | } | |
1269 | if (inet->cork.rt) { | |
1270 | ip_rt_put(inet->cork.rt); | |
1271 | inet->cork.rt = NULL; | |
1272 | } | |
1273 | return err; | |
1274 | ||
1275 | error: | |
1276 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
1277 | goto out; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * Throw away all pending data on the socket. | |
1282 | */ | |
1283 | void ip_flush_pending_frames(struct sock *sk) | |
1284 | { | |
1285 | struct inet_sock *inet = inet_sk(sk); | |
1286 | struct sk_buff *skb; | |
1287 | ||
1288 | while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) | |
1289 | kfree_skb(skb); | |
1290 | ||
1291 | inet->cork.flags &= ~IPCORK_OPT; | |
1292 | if (inet->cork.opt) { | |
1293 | kfree(inet->cork.opt); | |
1294 | inet->cork.opt = NULL; | |
1295 | } | |
1296 | if (inet->cork.rt) { | |
1297 | ip_rt_put(inet->cork.rt); | |
1298 | inet->cork.rt = NULL; | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | ||
1303 | /* | |
1304 | * Fetch data from kernel space and fill in checksum if needed. | |
1305 | */ | |
1306 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, | |
1307 | int len, int odd, struct sk_buff *skb) | |
1308 | { | |
1309 | unsigned int csum; | |
1310 | ||
1311 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1312 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * Generic function to send a packet as reply to another packet. | |
1318 | * Used to send TCP resets so far. ICMP should use this function too. | |
1319 | * | |
1320 | * Should run single threaded per socket because it uses the sock | |
1321 | * structure to pass arguments. | |
1322 | * | |
1323 | * LATER: switch from ip_build_xmit to ip_append_* | |
1324 | */ | |
1325 | void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg, | |
1326 | unsigned int len) | |
1327 | { | |
1328 | struct inet_sock *inet = inet_sk(sk); | |
1329 | struct { | |
1330 | struct ip_options opt; | |
1331 | char data[40]; | |
1332 | } replyopts; | |
1333 | struct ipcm_cookie ipc; | |
1334 | u32 daddr; | |
1335 | struct rtable *rt = (struct rtable*)skb->dst; | |
1336 | ||
1337 | if (ip_options_echo(&replyopts.opt, skb)) | |
1338 | return; | |
1339 | ||
1340 | daddr = ipc.addr = rt->rt_src; | |
1341 | ipc.opt = NULL; | |
1342 | ||
1343 | if (replyopts.opt.optlen) { | |
1344 | ipc.opt = &replyopts.opt; | |
1345 | ||
1346 | if (ipc.opt->srr) | |
1347 | daddr = replyopts.opt.faddr; | |
1348 | } | |
1349 | ||
1350 | { | |
1351 | struct flowi fl = { .nl_u = { .ip4_u = | |
1352 | { .daddr = daddr, | |
1353 | .saddr = rt->rt_spec_dst, | |
1354 | .tos = RT_TOS(skb->nh.iph->tos) } }, | |
1355 | /* Not quite clean, but right. */ | |
1356 | .uli_u = { .ports = | |
1357 | { .sport = skb->h.th->dest, | |
1358 | .dport = skb->h.th->source } }, | |
1359 | .proto = sk->sk_protocol }; | |
1360 | if (ip_route_output_key(&rt, &fl)) | |
1361 | return; | |
1362 | } | |
1363 | ||
1364 | /* And let IP do all the hard work. | |
1365 | ||
1366 | This chunk is not reenterable, hence spinlock. | |
1367 | Note that it uses the fact, that this function is called | |
1368 | with locally disabled BH and that sk cannot be already spinlocked. | |
1369 | */ | |
1370 | bh_lock_sock(sk); | |
1371 | inet->tos = skb->nh.iph->tos; | |
1372 | sk->sk_priority = skb->priority; | |
1373 | sk->sk_protocol = skb->nh.iph->protocol; | |
1374 | ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0, | |
1375 | &ipc, rt, MSG_DONTWAIT); | |
1376 | if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { | |
1377 | if (arg->csumoffset >= 0) | |
1378 | *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum)); | |
1379 | skb->ip_summed = CHECKSUM_NONE; | |
1380 | ip_push_pending_frames(sk); | |
1381 | } | |
1382 | ||
1383 | bh_unlock_sock(sk); | |
1384 | ||
1385 | ip_rt_put(rt); | |
1386 | } | |
1387 | ||
1da177e4 LT |
1388 | void __init ip_init(void) |
1389 | { | |
1da177e4 LT |
1390 | ip_rt_init(); |
1391 | inet_initpeers(); | |
1392 | ||
1393 | #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS) | |
1394 | igmp_mc_proc_init(); | |
1395 | #endif | |
1396 | } | |
1397 | ||
1da177e4 LT |
1398 | EXPORT_SYMBOL(ip_fragment); |
1399 | EXPORT_SYMBOL(ip_generic_getfrag); | |
1400 | EXPORT_SYMBOL(ip_queue_xmit); | |
1401 | EXPORT_SYMBOL(ip_send_check); |