]>
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 | ||
89f5f0ae HX |
356 | ip_select_ident_more(iph, &rt->u.dst, sk, |
357 | (skb_shinfo(skb)->tso_segs ?: 1) - 1); | |
1da177e4 LT |
358 | |
359 | /* Add an IP checksum. */ | |
360 | ip_send_check(iph); | |
361 | ||
362 | skb->priority = sk->sk_priority; | |
363 | ||
364 | return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev, | |
365 | dst_output); | |
366 | ||
367 | no_route: | |
368 | IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES); | |
369 | kfree_skb(skb); | |
370 | return -EHOSTUNREACH; | |
371 | } | |
372 | ||
373 | ||
374 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) | |
375 | { | |
376 | to->pkt_type = from->pkt_type; | |
377 | to->priority = from->priority; | |
378 | to->protocol = from->protocol; | |
1da177e4 LT |
379 | dst_release(to->dst); |
380 | to->dst = dst_clone(from->dst); | |
381 | to->dev = from->dev; | |
382 | ||
383 | /* Copy the flags to each fragment. */ | |
384 | IPCB(to)->flags = IPCB(from)->flags; | |
385 | ||
386 | #ifdef CONFIG_NET_SCHED | |
387 | to->tc_index = from->tc_index; | |
388 | #endif | |
389 | #ifdef CONFIG_NETFILTER | |
390 | to->nfmark = from->nfmark; | |
1da177e4 LT |
391 | /* Connection association is same as pre-frag packet */ |
392 | nf_conntrack_put(to->nfct); | |
393 | to->nfct = from->nfct; | |
394 | nf_conntrack_get(to->nfct); | |
395 | to->nfctinfo = from->nfctinfo; | |
c98d80ed JA |
396 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
397 | to->ipvs_property = from->ipvs_property; | |
398 | #endif | |
1da177e4 LT |
399 | #ifdef CONFIG_BRIDGE_NETFILTER |
400 | nf_bridge_put(to->nf_bridge); | |
401 | to->nf_bridge = from->nf_bridge; | |
402 | nf_bridge_get(to->nf_bridge); | |
403 | #endif | |
1da177e4 LT |
404 | #endif |
405 | } | |
406 | ||
407 | /* | |
408 | * This IP datagram is too large to be sent in one piece. Break it up into | |
409 | * smaller pieces (each of size equal to IP header plus | |
410 | * a block of the data of the original IP data part) that will yet fit in a | |
411 | * single device frame, and queue such a frame for sending. | |
412 | */ | |
413 | ||
414 | int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*)) | |
415 | { | |
416 | struct iphdr *iph; | |
417 | int raw = 0; | |
418 | int ptr; | |
419 | struct net_device *dev; | |
420 | struct sk_buff *skb2; | |
421 | unsigned int mtu, hlen, left, len, ll_rs; | |
422 | int offset; | |
423 | int not_last_frag; | |
424 | struct rtable *rt = (struct rtable*)skb->dst; | |
425 | int err = 0; | |
426 | ||
427 | dev = rt->u.dst.dev; | |
428 | ||
429 | /* | |
430 | * Point into the IP datagram header. | |
431 | */ | |
432 | ||
433 | iph = skb->nh.iph; | |
434 | ||
435 | if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) { | |
436 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, | |
437 | htonl(dst_mtu(&rt->u.dst))); | |
438 | kfree_skb(skb); | |
439 | return -EMSGSIZE; | |
440 | } | |
441 | ||
442 | /* | |
443 | * Setup starting values. | |
444 | */ | |
445 | ||
446 | hlen = iph->ihl * 4; | |
447 | mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */ | |
89cee8b1 | 448 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
1da177e4 LT |
449 | |
450 | /* When frag_list is given, use it. First, check its validity: | |
451 | * some transformers could create wrong frag_list or break existing | |
452 | * one, it is not prohibited. In this case fall back to copying. | |
453 | * | |
454 | * LATER: this step can be merged to real generation of fragments, | |
455 | * we can switch to copy when see the first bad fragment. | |
456 | */ | |
457 | if (skb_shinfo(skb)->frag_list) { | |
458 | struct sk_buff *frag; | |
459 | int first_len = skb_pagelen(skb); | |
460 | ||
461 | if (first_len - hlen > mtu || | |
462 | ((first_len - hlen) & 7) || | |
463 | (iph->frag_off & htons(IP_MF|IP_OFFSET)) || | |
464 | skb_cloned(skb)) | |
465 | goto slow_path; | |
466 | ||
467 | for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) { | |
468 | /* Correct geometry. */ | |
469 | if (frag->len > mtu || | |
470 | ((frag->len & 7) && frag->next) || | |
471 | skb_headroom(frag) < hlen) | |
472 | goto slow_path; | |
473 | ||
474 | /* Partially cloned skb? */ | |
475 | if (skb_shared(frag)) | |
476 | goto slow_path; | |
2fdba6b0 HX |
477 | |
478 | BUG_ON(frag->sk); | |
479 | if (skb->sk) { | |
480 | sock_hold(skb->sk); | |
481 | frag->sk = skb->sk; | |
482 | frag->destructor = sock_wfree; | |
483 | skb->truesize -= frag->truesize; | |
484 | } | |
1da177e4 LT |
485 | } |
486 | ||
487 | /* Everything is OK. Generate! */ | |
488 | ||
489 | err = 0; | |
490 | offset = 0; | |
491 | frag = skb_shinfo(skb)->frag_list; | |
492 | skb_shinfo(skb)->frag_list = NULL; | |
493 | skb->data_len = first_len - skb_headlen(skb); | |
494 | skb->len = first_len; | |
495 | iph->tot_len = htons(first_len); | |
496 | iph->frag_off = htons(IP_MF); | |
497 | ip_send_check(iph); | |
498 | ||
499 | for (;;) { | |
500 | /* Prepare header of the next frame, | |
501 | * before previous one went down. */ | |
502 | if (frag) { | |
503 | frag->ip_summed = CHECKSUM_NONE; | |
504 | frag->h.raw = frag->data; | |
505 | frag->nh.raw = __skb_push(frag, hlen); | |
506 | memcpy(frag->nh.raw, iph, hlen); | |
507 | iph = frag->nh.iph; | |
508 | iph->tot_len = htons(frag->len); | |
509 | ip_copy_metadata(frag, skb); | |
510 | if (offset == 0) | |
511 | ip_options_fragment(frag); | |
512 | offset += skb->len - hlen; | |
513 | iph->frag_off = htons(offset>>3); | |
514 | if (frag->next != NULL) | |
515 | iph->frag_off |= htons(IP_MF); | |
516 | /* Ready, complete checksum */ | |
517 | ip_send_check(iph); | |
518 | } | |
519 | ||
520 | err = output(skb); | |
521 | ||
522 | if (err || !frag) | |
523 | break; | |
524 | ||
525 | skb = frag; | |
526 | frag = skb->next; | |
527 | skb->next = NULL; | |
528 | } | |
529 | ||
530 | if (err == 0) { | |
531 | IP_INC_STATS(IPSTATS_MIB_FRAGOKS); | |
532 | return 0; | |
533 | } | |
534 | ||
535 | while (frag) { | |
536 | skb = frag->next; | |
537 | kfree_skb(frag); | |
538 | frag = skb; | |
539 | } | |
540 | IP_INC_STATS(IPSTATS_MIB_FRAGFAILS); | |
541 | return err; | |
542 | } | |
543 | ||
544 | slow_path: | |
545 | left = skb->len - hlen; /* Space per frame */ | |
546 | ptr = raw + hlen; /* Where to start from */ | |
547 | ||
548 | #ifdef CONFIG_BRIDGE_NETFILTER | |
549 | /* for bridged IP traffic encapsulated inside f.e. a vlan header, | |
550 | * we need to make room for the encapsulating header */ | |
551 | ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb)); | |
552 | mtu -= nf_bridge_pad(skb); | |
553 | #else | |
554 | ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev); | |
555 | #endif | |
556 | /* | |
557 | * Fragment the datagram. | |
558 | */ | |
559 | ||
560 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
561 | not_last_frag = iph->frag_off & htons(IP_MF); | |
562 | ||
563 | /* | |
564 | * Keep copying data until we run out. | |
565 | */ | |
566 | ||
567 | while(left > 0) { | |
568 | len = left; | |
569 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
570 | if (len > mtu) | |
571 | len = mtu; | |
572 | /* IF: we are not sending upto and including the packet end | |
573 | then align the next start on an eight byte boundary */ | |
574 | if (len < left) { | |
575 | len &= ~7; | |
576 | } | |
577 | /* | |
578 | * Allocate buffer. | |
579 | */ | |
580 | ||
581 | if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { | |
64ce2073 | 582 | NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); |
1da177e4 LT |
583 | err = -ENOMEM; |
584 | goto fail; | |
585 | } | |
586 | ||
587 | /* | |
588 | * Set up data on packet | |
589 | */ | |
590 | ||
591 | ip_copy_metadata(skb2, skb); | |
592 | skb_reserve(skb2, ll_rs); | |
593 | skb_put(skb2, len + hlen); | |
594 | skb2->nh.raw = skb2->data; | |
595 | skb2->h.raw = skb2->data + hlen; | |
596 | ||
597 | /* | |
598 | * Charge the memory for the fragment to any owner | |
599 | * it might possess | |
600 | */ | |
601 | ||
602 | if (skb->sk) | |
603 | skb_set_owner_w(skb2, skb->sk); | |
604 | ||
605 | /* | |
606 | * Copy the packet header into the new buffer. | |
607 | */ | |
608 | ||
609 | memcpy(skb2->nh.raw, skb->data, hlen); | |
610 | ||
611 | /* | |
612 | * Copy a block of the IP datagram. | |
613 | */ | |
614 | if (skb_copy_bits(skb, ptr, skb2->h.raw, len)) | |
615 | BUG(); | |
616 | left -= len; | |
617 | ||
618 | /* | |
619 | * Fill in the new header fields. | |
620 | */ | |
621 | iph = skb2->nh.iph; | |
622 | iph->frag_off = htons((offset >> 3)); | |
623 | ||
624 | /* ANK: dirty, but effective trick. Upgrade options only if | |
625 | * the segment to be fragmented was THE FIRST (otherwise, | |
626 | * options are already fixed) and make it ONCE | |
627 | * on the initial skb, so that all the following fragments | |
628 | * will inherit fixed options. | |
629 | */ | |
630 | if (offset == 0) | |
631 | ip_options_fragment(skb); | |
632 | ||
633 | /* | |
634 | * Added AC : If we are fragmenting a fragment that's not the | |
635 | * last fragment then keep MF on each bit | |
636 | */ | |
637 | if (left > 0 || not_last_frag) | |
638 | iph->frag_off |= htons(IP_MF); | |
639 | ptr += len; | |
640 | offset += len; | |
641 | ||
642 | /* | |
643 | * Put this fragment into the sending queue. | |
644 | */ | |
645 | ||
646 | IP_INC_STATS(IPSTATS_MIB_FRAGCREATES); | |
647 | ||
648 | iph->tot_len = htons(len + hlen); | |
649 | ||
650 | ip_send_check(iph); | |
651 | ||
652 | err = output(skb2); | |
653 | if (err) | |
654 | goto fail; | |
655 | } | |
656 | kfree_skb(skb); | |
657 | IP_INC_STATS(IPSTATS_MIB_FRAGOKS); | |
658 | return err; | |
659 | ||
660 | fail: | |
661 | kfree_skb(skb); | |
662 | IP_INC_STATS(IPSTATS_MIB_FRAGFAILS); | |
663 | return err; | |
664 | } | |
665 | ||
666 | int | |
667 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
668 | { | |
669 | struct iovec *iov = from; | |
670 | ||
671 | if (skb->ip_summed == CHECKSUM_HW) { | |
672 | if (memcpy_fromiovecend(to, iov, offset, len) < 0) | |
673 | return -EFAULT; | |
674 | } else { | |
675 | unsigned int csum = 0; | |
676 | if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) | |
677 | return -EFAULT; | |
678 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
679 | } | |
680 | return 0; | |
681 | } | |
682 | ||
683 | static inline unsigned int | |
684 | csum_page(struct page *page, int offset, int copy) | |
685 | { | |
686 | char *kaddr; | |
687 | unsigned int csum; | |
688 | kaddr = kmap(page); | |
689 | csum = csum_partial(kaddr + offset, copy, 0); | |
690 | kunmap(page); | |
691 | return csum; | |
692 | } | |
693 | ||
4b30b1c6 | 694 | static inline int ip_ufo_append_data(struct sock *sk, |
e89e9cf5 AR |
695 | int getfrag(void *from, char *to, int offset, int len, |
696 | int odd, struct sk_buff *skb), | |
697 | void *from, int length, int hh_len, int fragheaderlen, | |
698 | int transhdrlen, int mtu,unsigned int flags) | |
699 | { | |
700 | struct sk_buff *skb; | |
701 | int err; | |
702 | ||
703 | /* There is support for UDP fragmentation offload by network | |
704 | * device, so create one single skb packet containing complete | |
705 | * udp datagram | |
706 | */ | |
707 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) { | |
708 | skb = sock_alloc_send_skb(sk, | |
709 | hh_len + fragheaderlen + transhdrlen + 20, | |
710 | (flags & MSG_DONTWAIT), &err); | |
711 | ||
712 | if (skb == NULL) | |
713 | return err; | |
714 | ||
715 | /* reserve space for Hardware header */ | |
716 | skb_reserve(skb, hh_len); | |
717 | ||
718 | /* create space for UDP/IP header */ | |
719 | skb_put(skb,fragheaderlen + transhdrlen); | |
720 | ||
721 | /* initialize network header pointer */ | |
722 | skb->nh.raw = skb->data; | |
723 | ||
724 | /* initialize protocol header pointer */ | |
725 | skb->h.raw = skb->data + fragheaderlen; | |
726 | ||
727 | skb->ip_summed = CHECKSUM_HW; | |
728 | skb->csum = 0; | |
729 | sk->sk_sndmsg_off = 0; | |
730 | } | |
731 | ||
732 | err = skb_append_datato_frags(sk,skb, getfrag, from, | |
733 | (length - transhdrlen)); | |
734 | if (!err) { | |
735 | /* specify the length of each IP datagram fragment*/ | |
736 | skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen); | |
737 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
738 | ||
739 | return 0; | |
740 | } | |
741 | /* There is not enough support do UFO , | |
742 | * so follow normal path | |
743 | */ | |
744 | kfree_skb(skb); | |
745 | return err; | |
746 | } | |
747 | ||
1da177e4 LT |
748 | /* |
749 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
750 | * from many pieces of data. Each pieces will be holded on the socket | |
751 | * until ip_push_pending_frames() is called. Each piece can be a page | |
752 | * or non-page data. | |
753 | * | |
754 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
755 | * this interface potentially. | |
756 | * | |
757 | * LATER: length must be adjusted by pad at tail, when it is required. | |
758 | */ | |
759 | int ip_append_data(struct sock *sk, | |
760 | int getfrag(void *from, char *to, int offset, int len, | |
761 | int odd, struct sk_buff *skb), | |
762 | void *from, int length, int transhdrlen, | |
763 | struct ipcm_cookie *ipc, struct rtable *rt, | |
764 | unsigned int flags) | |
765 | { | |
766 | struct inet_sock *inet = inet_sk(sk); | |
767 | struct sk_buff *skb; | |
768 | ||
769 | struct ip_options *opt = NULL; | |
770 | int hh_len; | |
771 | int exthdrlen; | |
772 | int mtu; | |
773 | int copy; | |
774 | int err; | |
775 | int offset = 0; | |
776 | unsigned int maxfraglen, fragheaderlen; | |
777 | int csummode = CHECKSUM_NONE; | |
778 | ||
779 | if (flags&MSG_PROBE) | |
780 | return 0; | |
781 | ||
782 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
783 | /* | |
784 | * setup for corking. | |
785 | */ | |
786 | opt = ipc->opt; | |
787 | if (opt) { | |
788 | if (inet->cork.opt == NULL) { | |
789 | inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation); | |
790 | if (unlikely(inet->cork.opt == NULL)) | |
791 | return -ENOBUFS; | |
792 | } | |
793 | memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen); | |
794 | inet->cork.flags |= IPCORK_OPT; | |
795 | inet->cork.addr = ipc->addr; | |
796 | } | |
797 | dst_hold(&rt->u.dst); | |
798 | inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path); | |
799 | inet->cork.rt = rt; | |
800 | inet->cork.length = 0; | |
801 | sk->sk_sndmsg_page = NULL; | |
802 | sk->sk_sndmsg_off = 0; | |
803 | if ((exthdrlen = rt->u.dst.header_len) != 0) { | |
804 | length += exthdrlen; | |
805 | transhdrlen += exthdrlen; | |
806 | } | |
807 | } else { | |
808 | rt = inet->cork.rt; | |
809 | if (inet->cork.flags & IPCORK_OPT) | |
810 | opt = inet->cork.opt; | |
811 | ||
812 | transhdrlen = 0; | |
813 | exthdrlen = 0; | |
814 | mtu = inet->cork.fragsize; | |
815 | } | |
816 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
817 | ||
818 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
819 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
820 | ||
821 | if (inet->cork.length + length > 0xFFFF - fragheaderlen) { | |
822 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen); | |
823 | return -EMSGSIZE; | |
824 | } | |
825 | ||
826 | /* | |
827 | * transhdrlen > 0 means that this is the first fragment and we wish | |
828 | * it won't be fragmented in the future. | |
829 | */ | |
830 | if (transhdrlen && | |
831 | length + fragheaderlen <= mtu && | |
832 | rt->u.dst.dev->features&(NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM) && | |
833 | !exthdrlen) | |
834 | csummode = CHECKSUM_HW; | |
835 | ||
836 | inet->cork.length += length; | |
e89e9cf5 AR |
837 | if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) && |
838 | (rt->u.dst.dev->features & NETIF_F_UFO)) { | |
839 | ||
840 | if(ip_ufo_append_data(sk, getfrag, from, length, hh_len, | |
841 | fragheaderlen, transhdrlen, mtu, flags)) | |
842 | goto error; | |
843 | ||
844 | return 0; | |
845 | } | |
1da177e4 LT |
846 | |
847 | /* So, what's going on in the loop below? | |
848 | * | |
849 | * We use calculated fragment length to generate chained skb, | |
850 | * each of segments is IP fragment ready for sending to network after | |
851 | * adding appropriate IP header. | |
852 | */ | |
853 | ||
854 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
855 | goto alloc_new_skb; | |
856 | ||
857 | while (length > 0) { | |
858 | /* Check if the remaining data fits into current packet. */ | |
859 | copy = mtu - skb->len; | |
860 | if (copy < length) | |
861 | copy = maxfraglen - skb->len; | |
862 | if (copy <= 0) { | |
863 | char *data; | |
864 | unsigned int datalen; | |
865 | unsigned int fraglen; | |
866 | unsigned int fraggap; | |
867 | unsigned int alloclen; | |
868 | struct sk_buff *skb_prev; | |
869 | alloc_new_skb: | |
870 | skb_prev = skb; | |
871 | if (skb_prev) | |
872 | fraggap = skb_prev->len - maxfraglen; | |
873 | else | |
874 | fraggap = 0; | |
875 | ||
876 | /* | |
877 | * If remaining data exceeds the mtu, | |
878 | * we know we need more fragment(s). | |
879 | */ | |
880 | datalen = length + fraggap; | |
881 | if (datalen > mtu - fragheaderlen) | |
882 | datalen = maxfraglen - fragheaderlen; | |
883 | fraglen = datalen + fragheaderlen; | |
884 | ||
885 | if ((flags & MSG_MORE) && | |
886 | !(rt->u.dst.dev->features&NETIF_F_SG)) | |
887 | alloclen = mtu; | |
888 | else | |
889 | alloclen = datalen + fragheaderlen; | |
890 | ||
891 | /* The last fragment gets additional space at tail. | |
892 | * Note, with MSG_MORE we overallocate on fragments, | |
893 | * because we have no idea what fragment will be | |
894 | * the last. | |
895 | */ | |
896 | if (datalen == length) | |
897 | alloclen += rt->u.dst.trailer_len; | |
898 | ||
899 | if (transhdrlen) { | |
900 | skb = sock_alloc_send_skb(sk, | |
901 | alloclen + hh_len + 15, | |
902 | (flags & MSG_DONTWAIT), &err); | |
903 | } else { | |
904 | skb = NULL; | |
905 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
906 | 2 * sk->sk_sndbuf) | |
907 | skb = sock_wmalloc(sk, | |
908 | alloclen + hh_len + 15, 1, | |
909 | sk->sk_allocation); | |
910 | if (unlikely(skb == NULL)) | |
911 | err = -ENOBUFS; | |
912 | } | |
913 | if (skb == NULL) | |
914 | goto error; | |
915 | ||
916 | /* | |
917 | * Fill in the control structures | |
918 | */ | |
919 | skb->ip_summed = csummode; | |
920 | skb->csum = 0; | |
921 | skb_reserve(skb, hh_len); | |
922 | ||
923 | /* | |
924 | * Find where to start putting bytes. | |
925 | */ | |
926 | data = skb_put(skb, fraglen); | |
927 | skb->nh.raw = data + exthdrlen; | |
928 | data += fragheaderlen; | |
929 | skb->h.raw = data + exthdrlen; | |
930 | ||
931 | if (fraggap) { | |
932 | skb->csum = skb_copy_and_csum_bits( | |
933 | skb_prev, maxfraglen, | |
934 | data + transhdrlen, fraggap, 0); | |
935 | skb_prev->csum = csum_sub(skb_prev->csum, | |
936 | skb->csum); | |
937 | data += fraggap; | |
938 | skb_trim(skb_prev, maxfraglen); | |
939 | } | |
940 | ||
941 | copy = datalen - transhdrlen - fraggap; | |
942 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
943 | err = -EFAULT; | |
944 | kfree_skb(skb); | |
945 | goto error; | |
946 | } | |
947 | ||
948 | offset += copy; | |
949 | length -= datalen - fraggap; | |
950 | transhdrlen = 0; | |
951 | exthdrlen = 0; | |
952 | csummode = CHECKSUM_NONE; | |
953 | ||
954 | /* | |
955 | * Put the packet on the pending queue. | |
956 | */ | |
957 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
958 | continue; | |
959 | } | |
960 | ||
961 | if (copy > length) | |
962 | copy = length; | |
963 | ||
964 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) { | |
965 | unsigned int off; | |
966 | ||
967 | off = skb->len; | |
968 | if (getfrag(from, skb_put(skb, copy), | |
969 | offset, copy, off, skb) < 0) { | |
970 | __skb_trim(skb, off); | |
971 | err = -EFAULT; | |
972 | goto error; | |
973 | } | |
974 | } else { | |
975 | int i = skb_shinfo(skb)->nr_frags; | |
976 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1]; | |
977 | struct page *page = sk->sk_sndmsg_page; | |
978 | int off = sk->sk_sndmsg_off; | |
979 | unsigned int left; | |
980 | ||
981 | if (page && (left = PAGE_SIZE - off) > 0) { | |
982 | if (copy >= left) | |
983 | copy = left; | |
984 | if (page != frag->page) { | |
985 | if (i == MAX_SKB_FRAGS) { | |
986 | err = -EMSGSIZE; | |
987 | goto error; | |
988 | } | |
989 | get_page(page); | |
990 | skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0); | |
991 | frag = &skb_shinfo(skb)->frags[i]; | |
992 | } | |
993 | } else if (i < MAX_SKB_FRAGS) { | |
994 | if (copy > PAGE_SIZE) | |
995 | copy = PAGE_SIZE; | |
996 | page = alloc_pages(sk->sk_allocation, 0); | |
997 | if (page == NULL) { | |
998 | err = -ENOMEM; | |
999 | goto error; | |
1000 | } | |
1001 | sk->sk_sndmsg_page = page; | |
1002 | sk->sk_sndmsg_off = 0; | |
1003 | ||
1004 | skb_fill_page_desc(skb, i, page, 0, 0); | |
1005 | frag = &skb_shinfo(skb)->frags[i]; | |
1006 | skb->truesize += PAGE_SIZE; | |
1007 | atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); | |
1008 | } else { | |
1009 | err = -EMSGSIZE; | |
1010 | goto error; | |
1011 | } | |
1012 | if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) { | |
1013 | err = -EFAULT; | |
1014 | goto error; | |
1015 | } | |
1016 | sk->sk_sndmsg_off += copy; | |
1017 | frag->size += copy; | |
1018 | skb->len += copy; | |
1019 | skb->data_len += copy; | |
1020 | } | |
1021 | offset += copy; | |
1022 | length -= copy; | |
1023 | } | |
1024 | ||
1025 | return 0; | |
1026 | ||
1027 | error: | |
1028 | inet->cork.length -= length; | |
1029 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
1030 | return err; | |
1031 | } | |
1032 | ||
1033 | ssize_t ip_append_page(struct sock *sk, struct page *page, | |
1034 | int offset, size_t size, int flags) | |
1035 | { | |
1036 | struct inet_sock *inet = inet_sk(sk); | |
1037 | struct sk_buff *skb; | |
1038 | struct rtable *rt; | |
1039 | struct ip_options *opt = NULL; | |
1040 | int hh_len; | |
1041 | int mtu; | |
1042 | int len; | |
1043 | int err; | |
1044 | unsigned int maxfraglen, fragheaderlen, fraggap; | |
1045 | ||
1046 | if (inet->hdrincl) | |
1047 | return -EPERM; | |
1048 | ||
1049 | if (flags&MSG_PROBE) | |
1050 | return 0; | |
1051 | ||
1052 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1053 | return -EINVAL; | |
1054 | ||
1055 | rt = inet->cork.rt; | |
1056 | if (inet->cork.flags & IPCORK_OPT) | |
1057 | opt = inet->cork.opt; | |
1058 | ||
1059 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) | |
1060 | return -EOPNOTSUPP; | |
1061 | ||
1062 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
1063 | mtu = inet->cork.fragsize; | |
1064 | ||
1065 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1066 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
1067 | ||
1068 | if (inet->cork.length + size > 0xFFFF - fragheaderlen) { | |
1069 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu); | |
1070 | return -EMSGSIZE; | |
1071 | } | |
1072 | ||
1073 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
1074 | return -EINVAL; | |
1075 | ||
1076 | inet->cork.length += size; | |
e89e9cf5 AR |
1077 | if ((sk->sk_protocol == IPPROTO_UDP) && |
1078 | (rt->u.dst.dev->features & NETIF_F_UFO)) | |
1079 | skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen); | |
1080 | ||
1da177e4 LT |
1081 | |
1082 | while (size > 0) { | |
1083 | int i; | |
1084 | ||
e89e9cf5 AR |
1085 | if (skb_shinfo(skb)->ufo_size) |
1086 | len = size; | |
1087 | else { | |
1088 | ||
1089 | /* Check if the remaining data fits into current packet. */ | |
1090 | len = mtu - skb->len; | |
1091 | if (len < size) | |
1092 | len = maxfraglen - skb->len; | |
1093 | } | |
1da177e4 LT |
1094 | if (len <= 0) { |
1095 | struct sk_buff *skb_prev; | |
1096 | char *data; | |
1097 | struct iphdr *iph; | |
1098 | int alloclen; | |
1099 | ||
1100 | skb_prev = skb; | |
0d0d2bba | 1101 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1102 | |
1103 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1104 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1105 | if (unlikely(!skb)) { | |
1106 | err = -ENOBUFS; | |
1107 | goto error; | |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * Fill in the control structures | |
1112 | */ | |
1113 | skb->ip_summed = CHECKSUM_NONE; | |
1114 | skb->csum = 0; | |
1115 | skb_reserve(skb, hh_len); | |
1116 | ||
1117 | /* | |
1118 | * Find where to start putting bytes. | |
1119 | */ | |
1120 | data = skb_put(skb, fragheaderlen + fraggap); | |
1121 | skb->nh.iph = iph = (struct iphdr *)data; | |
1122 | data += fragheaderlen; | |
1123 | skb->h.raw = data; | |
1124 | ||
1125 | if (fraggap) { | |
1126 | skb->csum = skb_copy_and_csum_bits( | |
1127 | skb_prev, maxfraglen, | |
1128 | data, fraggap, 0); | |
1129 | skb_prev->csum = csum_sub(skb_prev->csum, | |
1130 | skb->csum); | |
1131 | skb_trim(skb_prev, maxfraglen); | |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * Put the packet on the pending queue. | |
1136 | */ | |
1137 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1138 | continue; | |
1139 | } | |
1140 | ||
1141 | i = skb_shinfo(skb)->nr_frags; | |
1142 | if (len > size) | |
1143 | len = size; | |
1144 | if (skb_can_coalesce(skb, i, page, offset)) { | |
1145 | skb_shinfo(skb)->frags[i-1].size += len; | |
1146 | } else if (i < MAX_SKB_FRAGS) { | |
1147 | get_page(page); | |
1148 | skb_fill_page_desc(skb, i, page, offset, len); | |
1149 | } else { | |
1150 | err = -EMSGSIZE; | |
1151 | goto error; | |
1152 | } | |
1153 | ||
1154 | if (skb->ip_summed == CHECKSUM_NONE) { | |
1155 | unsigned int csum; | |
1156 | csum = csum_page(page, offset, len); | |
1157 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1158 | } | |
1159 | ||
1160 | skb->len += len; | |
1161 | skb->data_len += len; | |
1162 | offset += len; | |
1163 | size -= len; | |
1164 | } | |
1165 | return 0; | |
1166 | ||
1167 | error: | |
1168 | inet->cork.length -= size; | |
1169 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
1170 | return err; | |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * Combined all pending IP fragments on the socket as one IP datagram | |
1175 | * and push them out. | |
1176 | */ | |
1177 | int ip_push_pending_frames(struct sock *sk) | |
1178 | { | |
1179 | struct sk_buff *skb, *tmp_skb; | |
1180 | struct sk_buff **tail_skb; | |
1181 | struct inet_sock *inet = inet_sk(sk); | |
1182 | struct ip_options *opt = NULL; | |
1183 | struct rtable *rt = inet->cork.rt; | |
1184 | struct iphdr *iph; | |
1185 | int df = 0; | |
1186 | __u8 ttl; | |
1187 | int err = 0; | |
1188 | ||
1189 | if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL) | |
1190 | goto out; | |
1191 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1192 | ||
1193 | /* move skb->data to ip header from ext header */ | |
1194 | if (skb->data < skb->nh.raw) | |
1195 | __skb_pull(skb, skb->nh.raw - skb->data); | |
1196 | while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { | |
1197 | __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw); | |
1198 | *tail_skb = tmp_skb; | |
1199 | tail_skb = &(tmp_skb->next); | |
1200 | skb->len += tmp_skb->len; | |
1201 | skb->data_len += tmp_skb->len; | |
1202 | skb->truesize += tmp_skb->truesize; | |
1203 | __sock_put(tmp_skb->sk); | |
1204 | tmp_skb->destructor = NULL; | |
1205 | tmp_skb->sk = NULL; | |
1206 | } | |
1207 | ||
1208 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1209 | * to fragment the frame generated here. No matter, what transforms | |
1210 | * how transforms change size of the packet, it will come out. | |
1211 | */ | |
1212 | if (inet->pmtudisc != IP_PMTUDISC_DO) | |
1213 | skb->local_df = 1; | |
1214 | ||
1215 | /* DF bit is set when we want to see DF on outgoing frames. | |
1216 | * If local_df is set too, we still allow to fragment this frame | |
1217 | * locally. */ | |
1218 | if (inet->pmtudisc == IP_PMTUDISC_DO || | |
1219 | (skb->len <= dst_mtu(&rt->u.dst) && | |
1220 | ip_dont_fragment(sk, &rt->u.dst))) | |
1221 | df = htons(IP_DF); | |
1222 | ||
1223 | if (inet->cork.flags & IPCORK_OPT) | |
1224 | opt = inet->cork.opt; | |
1225 | ||
1226 | if (rt->rt_type == RTN_MULTICAST) | |
1227 | ttl = inet->mc_ttl; | |
1228 | else | |
1229 | ttl = ip_select_ttl(inet, &rt->u.dst); | |
1230 | ||
1231 | iph = (struct iphdr *)skb->data; | |
1232 | iph->version = 4; | |
1233 | iph->ihl = 5; | |
1234 | if (opt) { | |
1235 | iph->ihl += opt->optlen>>2; | |
1236 | ip_options_build(skb, opt, inet->cork.addr, rt, 0); | |
1237 | } | |
1238 | iph->tos = inet->tos; | |
1239 | iph->tot_len = htons(skb->len); | |
1240 | iph->frag_off = df; | |
1241 | if (!df) { | |
1242 | __ip_select_ident(iph, &rt->u.dst, 0); | |
1243 | } else { | |
1244 | iph->id = htons(inet->id++); | |
1245 | } | |
1246 | iph->ttl = ttl; | |
1247 | iph->protocol = sk->sk_protocol; | |
1248 | iph->saddr = rt->rt_src; | |
1249 | iph->daddr = rt->rt_dst; | |
1250 | ip_send_check(iph); | |
1251 | ||
1252 | skb->priority = sk->sk_priority; | |
1253 | skb->dst = dst_clone(&rt->u.dst); | |
1254 | ||
1255 | /* Netfilter gets whole the not fragmented skb. */ | |
1256 | err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, | |
1257 | skb->dst->dev, dst_output); | |
1258 | if (err) { | |
1259 | if (err > 0) | |
1260 | err = inet->recverr ? net_xmit_errno(err) : 0; | |
1261 | if (err) | |
1262 | goto error; | |
1263 | } | |
1264 | ||
1265 | out: | |
1266 | inet->cork.flags &= ~IPCORK_OPT; | |
a51482bd JJ |
1267 | kfree(inet->cork.opt); |
1268 | inet->cork.opt = NULL; | |
1da177e4 LT |
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; | |
a51482bd JJ |
1292 | kfree(inet->cork.opt); |
1293 | inet->cork.opt = NULL; | |
1da177e4 LT |
1294 | if (inet->cork.rt) { |
1295 | ip_rt_put(inet->cork.rt); | |
1296 | inet->cork.rt = NULL; | |
1297 | } | |
1298 | } | |
1299 | ||
1300 | ||
1301 | /* | |
1302 | * Fetch data from kernel space and fill in checksum if needed. | |
1303 | */ | |
1304 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, | |
1305 | int len, int odd, struct sk_buff *skb) | |
1306 | { | |
1307 | unsigned int csum; | |
1308 | ||
1309 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1310 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
1311 | return 0; | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * Generic function to send a packet as reply to another packet. | |
1316 | * Used to send TCP resets so far. ICMP should use this function too. | |
1317 | * | |
1318 | * Should run single threaded per socket because it uses the sock | |
1319 | * structure to pass arguments. | |
1320 | * | |
1321 | * LATER: switch from ip_build_xmit to ip_append_* | |
1322 | */ | |
1323 | void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg, | |
1324 | unsigned int len) | |
1325 | { | |
1326 | struct inet_sock *inet = inet_sk(sk); | |
1327 | struct { | |
1328 | struct ip_options opt; | |
1329 | char data[40]; | |
1330 | } replyopts; | |
1331 | struct ipcm_cookie ipc; | |
1332 | u32 daddr; | |
1333 | struct rtable *rt = (struct rtable*)skb->dst; | |
1334 | ||
1335 | if (ip_options_echo(&replyopts.opt, skb)) | |
1336 | return; | |
1337 | ||
1338 | daddr = ipc.addr = rt->rt_src; | |
1339 | ipc.opt = NULL; | |
1340 | ||
1341 | if (replyopts.opt.optlen) { | |
1342 | ipc.opt = &replyopts.opt; | |
1343 | ||
1344 | if (ipc.opt->srr) | |
1345 | daddr = replyopts.opt.faddr; | |
1346 | } | |
1347 | ||
1348 | { | |
1349 | struct flowi fl = { .nl_u = { .ip4_u = | |
1350 | { .daddr = daddr, | |
1351 | .saddr = rt->rt_spec_dst, | |
1352 | .tos = RT_TOS(skb->nh.iph->tos) } }, | |
1353 | /* Not quite clean, but right. */ | |
1354 | .uli_u = { .ports = | |
1355 | { .sport = skb->h.th->dest, | |
1356 | .dport = skb->h.th->source } }, | |
1357 | .proto = sk->sk_protocol }; | |
1358 | if (ip_route_output_key(&rt, &fl)) | |
1359 | return; | |
1360 | } | |
1361 | ||
1362 | /* And let IP do all the hard work. | |
1363 | ||
1364 | This chunk is not reenterable, hence spinlock. | |
1365 | Note that it uses the fact, that this function is called | |
1366 | with locally disabled BH and that sk cannot be already spinlocked. | |
1367 | */ | |
1368 | bh_lock_sock(sk); | |
1369 | inet->tos = skb->nh.iph->tos; | |
1370 | sk->sk_priority = skb->priority; | |
1371 | sk->sk_protocol = skb->nh.iph->protocol; | |
1372 | ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0, | |
1373 | &ipc, rt, MSG_DONTWAIT); | |
1374 | if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { | |
1375 | if (arg->csumoffset >= 0) | |
1376 | *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum)); | |
1377 | skb->ip_summed = CHECKSUM_NONE; | |
1378 | ip_push_pending_frames(sk); | |
1379 | } | |
1380 | ||
1381 | bh_unlock_sock(sk); | |
1382 | ||
1383 | ip_rt_put(rt); | |
1384 | } | |
1385 | ||
1da177e4 LT |
1386 | void __init ip_init(void) |
1387 | { | |
1da177e4 LT |
1388 | ip_rt_init(); |
1389 | inet_initpeers(); | |
1390 | ||
1391 | #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS) | |
1392 | igmp_mc_proc_init(); | |
1393 | #endif | |
1394 | } | |
1395 | ||
1da177e4 LT |
1396 | EXPORT_SYMBOL(ip_fragment); |
1397 | EXPORT_SYMBOL(ip_generic_getfrag); | |
1398 | EXPORT_SYMBOL(ip_queue_xmit); | |
1399 | EXPORT_SYMBOL(ip_send_check); |