]>
Commit | Line | Data |
---|---|---|
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 | * Authors: Ross Biro | |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> | |
10 | * Donald Becker, <becker@super.org> | |
11 | * Alan Cox, <Alan.Cox@linux.org> | |
12 | * Richard Underwood | |
13 | * Stefan Becker, <stefanb@yello.ping.de> | |
14 | * Jorge Cwik, <jorge@laser.satlink.net> | |
15 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
16 | * Hirokazu Takahashi, <taka@valinux.co.jp> | |
17 | * | |
18 | * See ip_input.c for original log | |
19 | * | |
20 | * Fixes: | |
21 | * Alan Cox : Missing nonblock feature in ip_build_xmit. | |
22 | * Mike Kilburn : htons() missing in ip_build_xmit. | |
23 | * Bradford Johnson: Fix faulty handling of some frames when | |
24 | * no route is found. | |
25 | * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit | |
26 | * (in case if packet not accepted by | |
27 | * output firewall rules) | |
28 | * Mike McLagan : Routing by source | |
29 | * Alexey Kuznetsov: use new route cache | |
30 | * Andi Kleen: Fix broken PMTU recovery and remove | |
31 | * some redundant tests. | |
32 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. | |
33 | * Andi Kleen : Replace ip_reply with ip_send_reply. | |
34 | * Andi Kleen : Split fast and slow ip_build_xmit path | |
35 | * for decreased register pressure on x86 | |
36 | * and more readibility. | |
37 | * Marc Boucher : When call_out_firewall returns FW_QUEUE, | |
38 | * silently drop skb instead of failing with -EPERM. | |
39 | * Detlev Wengorz : Copy protocol for fragments. | |
40 | * Hirokazu Takahashi: HW checksumming for outgoing UDP | |
41 | * datagrams. | |
42 | * Hirokazu Takahashi: sendfile() on UDP works now. | |
43 | */ | |
44 | ||
45 | #include <linux/uaccess.h> | |
46 | #include <linux/module.h> | |
47 | #include <linux/types.h> | |
48 | #include <linux/kernel.h> | |
49 | #include <linux/mm.h> | |
50 | #include <linux/string.h> | |
51 | #include <linux/errno.h> | |
52 | #include <linux/highmem.h> | |
53 | #include <linux/slab.h> | |
54 | ||
55 | #include <linux/socket.h> | |
56 | #include <linux/sockios.h> | |
57 | #include <linux/in.h> | |
58 | #include <linux/inet.h> | |
59 | #include <linux/netdevice.h> | |
60 | #include <linux/etherdevice.h> | |
61 | #include <linux/proc_fs.h> | |
62 | #include <linux/stat.h> | |
63 | #include <linux/init.h> | |
64 | ||
65 | #include <net/snmp.h> | |
66 | #include <net/ip.h> | |
67 | #include <net/protocol.h> | |
68 | #include <net/route.h> | |
69 | #include <net/xfrm.h> | |
70 | #include <linux/skbuff.h> | |
71 | #include <net/sock.h> | |
72 | #include <net/arp.h> | |
73 | #include <net/icmp.h> | |
74 | #include <net/checksum.h> | |
75 | #include <net/inetpeer.h> | |
76 | #include <net/lwtunnel.h> | |
77 | #include <linux/bpf-cgroup.h> | |
78 | #include <linux/igmp.h> | |
79 | #include <linux/netfilter_ipv4.h> | |
80 | #include <linux/netfilter_bridge.h> | |
81 | #include <linux/netlink.h> | |
82 | #include <linux/tcp.h> | |
83 | ||
84 | static int | |
85 | ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, | |
86 | unsigned int mtu, | |
87 | int (*output)(struct net *, struct sock *, struct sk_buff *)); | |
88 | ||
89 | /* Generate a checksum for an outgoing IP datagram. */ | |
90 | void ip_send_check(struct iphdr *iph) | |
91 | { | |
92 | iph->check = 0; | |
93 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
94 | } | |
95 | EXPORT_SYMBOL(ip_send_check); | |
96 | ||
97 | int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) | |
98 | { | |
99 | struct iphdr *iph = ip_hdr(skb); | |
100 | ||
101 | iph->tot_len = htons(skb->len); | |
102 | ip_send_check(iph); | |
103 | ||
104 | /* if egress device is enslaved to an L3 master device pass the | |
105 | * skb to its handler for processing | |
106 | */ | |
107 | skb = l3mdev_ip_out(sk, skb); | |
108 | if (unlikely(!skb)) | |
109 | return 0; | |
110 | ||
111 | skb->protocol = htons(ETH_P_IP); | |
112 | ||
113 | return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, | |
114 | net, sk, skb, NULL, skb_dst(skb)->dev, | |
115 | dst_output); | |
116 | } | |
117 | ||
118 | int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) | |
119 | { | |
120 | int err; | |
121 | ||
122 | err = __ip_local_out(net, sk, skb); | |
123 | if (likely(err == 1)) | |
124 | err = dst_output(net, sk, skb); | |
125 | ||
126 | return err; | |
127 | } | |
128 | EXPORT_SYMBOL_GPL(ip_local_out); | |
129 | ||
130 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) | |
131 | { | |
132 | int ttl = inet->uc_ttl; | |
133 | ||
134 | if (ttl < 0) | |
135 | ttl = ip4_dst_hoplimit(dst); | |
136 | return ttl; | |
137 | } | |
138 | ||
139 | /* | |
140 | * Add an ip header to a skbuff and send it out. | |
141 | * | |
142 | */ | |
143 | int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, | |
144 | __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) | |
145 | { | |
146 | struct inet_sock *inet = inet_sk(sk); | |
147 | struct rtable *rt = skb_rtable(skb); | |
148 | struct net *net = sock_net(sk); | |
149 | struct iphdr *iph; | |
150 | ||
151 | /* Build the IP header. */ | |
152 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); | |
153 | skb_reset_network_header(skb); | |
154 | iph = ip_hdr(skb); | |
155 | iph->version = 4; | |
156 | iph->ihl = 5; | |
157 | iph->tos = inet->tos; | |
158 | iph->ttl = ip_select_ttl(inet, &rt->dst); | |
159 | iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); | |
160 | iph->saddr = saddr; | |
161 | iph->protocol = sk->sk_protocol; | |
162 | if (ip_dont_fragment(sk, &rt->dst)) { | |
163 | iph->frag_off = htons(IP_DF); | |
164 | iph->id = 0; | |
165 | } else { | |
166 | iph->frag_off = 0; | |
167 | __ip_select_ident(net, iph, 1); | |
168 | } | |
169 | ||
170 | if (opt && opt->opt.optlen) { | |
171 | iph->ihl += opt->opt.optlen>>2; | |
172 | ip_options_build(skb, &opt->opt, daddr, rt, 0); | |
173 | } | |
174 | ||
175 | skb->priority = sk->sk_priority; | |
176 | if (!skb->mark) | |
177 | skb->mark = sk->sk_mark; | |
178 | ||
179 | /* Send it out. */ | |
180 | return ip_local_out(net, skb->sk, skb); | |
181 | } | |
182 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); | |
183 | ||
184 | static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) | |
185 | { | |
186 | struct dst_entry *dst = skb_dst(skb); | |
187 | struct rtable *rt = (struct rtable *)dst; | |
188 | struct net_device *dev = dst->dev; | |
189 | unsigned int hh_len = LL_RESERVED_SPACE(dev); | |
190 | struct neighbour *neigh; | |
191 | u32 nexthop; | |
192 | ||
193 | if (rt->rt_type == RTN_MULTICAST) { | |
194 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); | |
195 | } else if (rt->rt_type == RTN_BROADCAST) | |
196 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); | |
197 | ||
198 | /* Be paranoid, rather than too clever. */ | |
199 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { | |
200 | struct sk_buff *skb2; | |
201 | ||
202 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
203 | if (!skb2) { | |
204 | kfree_skb(skb); | |
205 | return -ENOMEM; | |
206 | } | |
207 | if (skb->sk) | |
208 | skb_set_owner_w(skb2, skb->sk); | |
209 | consume_skb(skb); | |
210 | skb = skb2; | |
211 | } | |
212 | ||
213 | if (lwtunnel_xmit_redirect(dst->lwtstate)) { | |
214 | int res = lwtunnel_xmit(skb); | |
215 | ||
216 | if (res < 0 || res == LWTUNNEL_XMIT_DONE) | |
217 | return res; | |
218 | } | |
219 | ||
220 | rcu_read_lock_bh(); | |
221 | nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); | |
222 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); | |
223 | if (unlikely(!neigh)) | |
224 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
225 | if (!IS_ERR(neigh)) { | |
226 | int res; | |
227 | ||
228 | sock_confirm_neigh(skb, neigh); | |
229 | res = neigh_output(neigh, skb); | |
230 | ||
231 | rcu_read_unlock_bh(); | |
232 | return res; | |
233 | } | |
234 | rcu_read_unlock_bh(); | |
235 | ||
236 | net_dbg_ratelimited("%s: No header cache and no neighbour!\n", | |
237 | __func__); | |
238 | kfree_skb(skb); | |
239 | return -EINVAL; | |
240 | } | |
241 | ||
242 | static int ip_finish_output_gso(struct net *net, struct sock *sk, | |
243 | struct sk_buff *skb, unsigned int mtu) | |
244 | { | |
245 | netdev_features_t features; | |
246 | struct sk_buff *segs; | |
247 | int ret = 0; | |
248 | ||
249 | /* common case: seglen is <= mtu | |
250 | */ | |
251 | if (skb_gso_validate_network_len(skb, mtu)) | |
252 | return ip_finish_output2(net, sk, skb); | |
253 | ||
254 | /* Slowpath - GSO segment length exceeds the egress MTU. | |
255 | * | |
256 | * This can happen in several cases: | |
257 | * - Forwarding of a TCP GRO skb, when DF flag is not set. | |
258 | * - Forwarding of an skb that arrived on a virtualization interface | |
259 | * (virtio-net/vhost/tap) with TSO/GSO size set by other network | |
260 | * stack. | |
261 | * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an | |
262 | * interface with a smaller MTU. | |
263 | * - Arriving GRO skb (or GSO skb in a virtualized environment) that is | |
264 | * bridged to a NETIF_F_TSO tunnel stacked over an interface with an | |
265 | * insufficent MTU. | |
266 | */ | |
267 | features = netif_skb_features(skb); | |
268 | BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET); | |
269 | segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); | |
270 | if (IS_ERR_OR_NULL(segs)) { | |
271 | kfree_skb(skb); | |
272 | return -ENOMEM; | |
273 | } | |
274 | ||
275 | consume_skb(skb); | |
276 | ||
277 | do { | |
278 | struct sk_buff *nskb = segs->next; | |
279 | int err; | |
280 | ||
281 | segs->next = NULL; | |
282 | err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); | |
283 | ||
284 | if (err && ret == 0) | |
285 | ret = err; | |
286 | segs = nskb; | |
287 | } while (segs); | |
288 | ||
289 | return ret; | |
290 | } | |
291 | ||
292 | static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) | |
293 | { | |
294 | unsigned int mtu; | |
295 | int ret; | |
296 | ||
297 | ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); | |
298 | if (ret) { | |
299 | kfree_skb(skb); | |
300 | return ret; | |
301 | } | |
302 | ||
303 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) | |
304 | /* Policy lookup after SNAT yielded a new policy */ | |
305 | if (skb_dst(skb)->xfrm) { | |
306 | IPCB(skb)->flags |= IPSKB_REROUTED; | |
307 | return dst_output(net, sk, skb); | |
308 | } | |
309 | #endif | |
310 | mtu = ip_skb_dst_mtu(sk, skb); | |
311 | if (skb_is_gso(skb)) | |
312 | return ip_finish_output_gso(net, sk, skb, mtu); | |
313 | ||
314 | if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU)) | |
315 | return ip_fragment(net, sk, skb, mtu, ip_finish_output2); | |
316 | ||
317 | return ip_finish_output2(net, sk, skb); | |
318 | } | |
319 | ||
320 | static int ip_mc_finish_output(struct net *net, struct sock *sk, | |
321 | struct sk_buff *skb) | |
322 | { | |
323 | int ret; | |
324 | ||
325 | ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); | |
326 | if (ret) { | |
327 | kfree_skb(skb); | |
328 | return ret; | |
329 | } | |
330 | ||
331 | return dev_loopback_xmit(net, sk, skb); | |
332 | } | |
333 | ||
334 | int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) | |
335 | { | |
336 | struct rtable *rt = skb_rtable(skb); | |
337 | struct net_device *dev = rt->dst.dev; | |
338 | ||
339 | /* | |
340 | * If the indicated interface is up and running, send the packet. | |
341 | */ | |
342 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); | |
343 | ||
344 | skb->dev = dev; | |
345 | skb->protocol = htons(ETH_P_IP); | |
346 | ||
347 | /* | |
348 | * Multicasts are looped back for other local users | |
349 | */ | |
350 | ||
351 | if (rt->rt_flags&RTCF_MULTICAST) { | |
352 | if (sk_mc_loop(sk) | |
353 | #ifdef CONFIG_IP_MROUTE | |
354 | /* Small optimization: do not loopback not local frames, | |
355 | which returned after forwarding; they will be dropped | |
356 | by ip_mr_input in any case. | |
357 | Note, that local frames are looped back to be delivered | |
358 | to local recipients. | |
359 | ||
360 | This check is duplicated in ip_mr_input at the moment. | |
361 | */ | |
362 | && | |
363 | ((rt->rt_flags & RTCF_LOCAL) || | |
364 | !(IPCB(skb)->flags & IPSKB_FORWARDED)) | |
365 | #endif | |
366 | ) { | |
367 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
368 | if (newskb) | |
369 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, | |
370 | net, sk, newskb, NULL, newskb->dev, | |
371 | ip_mc_finish_output); | |
372 | } | |
373 | ||
374 | /* Multicasts with ttl 0 must not go beyond the host */ | |
375 | ||
376 | if (ip_hdr(skb)->ttl == 0) { | |
377 | kfree_skb(skb); | |
378 | return 0; | |
379 | } | |
380 | } | |
381 | ||
382 | if (rt->rt_flags&RTCF_BROADCAST) { | |
383 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
384 | if (newskb) | |
385 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, | |
386 | net, sk, newskb, NULL, newskb->dev, | |
387 | ip_mc_finish_output); | |
388 | } | |
389 | ||
390 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, | |
391 | net, sk, skb, NULL, skb->dev, | |
392 | ip_finish_output, | |
393 | !(IPCB(skb)->flags & IPSKB_REROUTED)); | |
394 | } | |
395 | ||
396 | int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) | |
397 | { | |
398 | struct net_device *dev = skb_dst(skb)->dev; | |
399 | ||
400 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); | |
401 | ||
402 | skb->dev = dev; | |
403 | skb->protocol = htons(ETH_P_IP); | |
404 | ||
405 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, | |
406 | net, sk, skb, NULL, dev, | |
407 | ip_finish_output, | |
408 | !(IPCB(skb)->flags & IPSKB_REROUTED)); | |
409 | } | |
410 | ||
411 | /* | |
412 | * copy saddr and daddr, possibly using 64bit load/stores | |
413 | * Equivalent to : | |
414 | * iph->saddr = fl4->saddr; | |
415 | * iph->daddr = fl4->daddr; | |
416 | */ | |
417 | static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) | |
418 | { | |
419 | BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != | |
420 | offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); | |
421 | memcpy(&iph->saddr, &fl4->saddr, | |
422 | sizeof(fl4->saddr) + sizeof(fl4->daddr)); | |
423 | } | |
424 | ||
425 | /* Note: skb->sk can be different from sk, in case of tunnels */ | |
426 | int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) | |
427 | { | |
428 | struct inet_sock *inet = inet_sk(sk); | |
429 | struct net *net = sock_net(sk); | |
430 | struct ip_options_rcu *inet_opt; | |
431 | struct flowi4 *fl4; | |
432 | struct rtable *rt; | |
433 | struct iphdr *iph; | |
434 | int res; | |
435 | ||
436 | /* Skip all of this if the packet is already routed, | |
437 | * f.e. by something like SCTP. | |
438 | */ | |
439 | rcu_read_lock(); | |
440 | inet_opt = rcu_dereference(inet->inet_opt); | |
441 | fl4 = &fl->u.ip4; | |
442 | rt = skb_rtable(skb); | |
443 | if (rt) | |
444 | goto packet_routed; | |
445 | ||
446 | /* Make sure we can route this packet. */ | |
447 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
448 | if (!rt) { | |
449 | __be32 daddr; | |
450 | ||
451 | /* Use correct destination address if we have options. */ | |
452 | daddr = inet->inet_daddr; | |
453 | if (inet_opt && inet_opt->opt.srr) | |
454 | daddr = inet_opt->opt.faddr; | |
455 | ||
456 | /* If this fails, retransmit mechanism of transport layer will | |
457 | * keep trying until route appears or the connection times | |
458 | * itself out. | |
459 | */ | |
460 | rt = ip_route_output_ports(net, fl4, sk, | |
461 | daddr, inet->inet_saddr, | |
462 | inet->inet_dport, | |
463 | inet->inet_sport, | |
464 | sk->sk_protocol, | |
465 | RT_CONN_FLAGS(sk), | |
466 | sk->sk_bound_dev_if); | |
467 | if (IS_ERR(rt)) | |
468 | goto no_route; | |
469 | sk_setup_caps(sk, &rt->dst); | |
470 | } | |
471 | skb_dst_set_noref(skb, &rt->dst); | |
472 | ||
473 | packet_routed: | |
474 | if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) | |
475 | goto no_route; | |
476 | ||
477 | /* OK, we know where to send it, allocate and build IP header. */ | |
478 | skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); | |
479 | skb_reset_network_header(skb); | |
480 | iph = ip_hdr(skb); | |
481 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); | |
482 | if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) | |
483 | iph->frag_off = htons(IP_DF); | |
484 | else | |
485 | iph->frag_off = 0; | |
486 | iph->ttl = ip_select_ttl(inet, &rt->dst); | |
487 | iph->protocol = sk->sk_protocol; | |
488 | ip_copy_addrs(iph, fl4); | |
489 | ||
490 | /* Transport layer set skb->h.foo itself. */ | |
491 | ||
492 | if (inet_opt && inet_opt->opt.optlen) { | |
493 | iph->ihl += inet_opt->opt.optlen >> 2; | |
494 | ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); | |
495 | } | |
496 | ||
497 | ip_select_ident_segs(net, skb, sk, | |
498 | skb_shinfo(skb)->gso_segs ?: 1); | |
499 | ||
500 | /* TODO : should we use skb->sk here instead of sk ? */ | |
501 | skb->priority = sk->sk_priority; | |
502 | skb->mark = sk->sk_mark; | |
503 | ||
504 | res = ip_local_out(net, sk, skb); | |
505 | rcu_read_unlock(); | |
506 | return res; | |
507 | ||
508 | no_route: | |
509 | rcu_read_unlock(); | |
510 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); | |
511 | kfree_skb(skb); | |
512 | return -EHOSTUNREACH; | |
513 | } | |
514 | EXPORT_SYMBOL(ip_queue_xmit); | |
515 | ||
516 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) | |
517 | { | |
518 | to->pkt_type = from->pkt_type; | |
519 | to->priority = from->priority; | |
520 | to->protocol = from->protocol; | |
521 | skb_dst_drop(to); | |
522 | skb_dst_copy(to, from); | |
523 | to->dev = from->dev; | |
524 | to->mark = from->mark; | |
525 | ||
526 | /* Copy the flags to each fragment. */ | |
527 | IPCB(to)->flags = IPCB(from)->flags; | |
528 | ||
529 | #ifdef CONFIG_NET_SCHED | |
530 | to->tc_index = from->tc_index; | |
531 | #endif | |
532 | nf_copy(to, from); | |
533 | #if IS_ENABLED(CONFIG_IP_VS) | |
534 | to->ipvs_property = from->ipvs_property; | |
535 | #endif | |
536 | skb_copy_secmark(to, from); | |
537 | } | |
538 | ||
539 | static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, | |
540 | unsigned int mtu, | |
541 | int (*output)(struct net *, struct sock *, struct sk_buff *)) | |
542 | { | |
543 | struct iphdr *iph = ip_hdr(skb); | |
544 | ||
545 | if ((iph->frag_off & htons(IP_DF)) == 0) | |
546 | return ip_do_fragment(net, sk, skb, output); | |
547 | ||
548 | if (unlikely(!skb->ignore_df || | |
549 | (IPCB(skb)->frag_max_size && | |
550 | IPCB(skb)->frag_max_size > mtu))) { | |
551 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); | |
552 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, | |
553 | htonl(mtu)); | |
554 | kfree_skb(skb); | |
555 | return -EMSGSIZE; | |
556 | } | |
557 | ||
558 | return ip_do_fragment(net, sk, skb, output); | |
559 | } | |
560 | ||
561 | /* | |
562 | * This IP datagram is too large to be sent in one piece. Break it up into | |
563 | * smaller pieces (each of size equal to IP header plus | |
564 | * a block of the data of the original IP data part) that will yet fit in a | |
565 | * single device frame, and queue such a frame for sending. | |
566 | */ | |
567 | ||
568 | int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, | |
569 | int (*output)(struct net *, struct sock *, struct sk_buff *)) | |
570 | { | |
571 | struct iphdr *iph; | |
572 | int ptr; | |
573 | struct sk_buff *skb2; | |
574 | unsigned int mtu, hlen, left, len, ll_rs; | |
575 | int offset; | |
576 | __be16 not_last_frag; | |
577 | struct rtable *rt = skb_rtable(skb); | |
578 | int err = 0; | |
579 | ||
580 | /* for offloaded checksums cleanup checksum before fragmentation */ | |
581 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
582 | (err = skb_checksum_help(skb))) | |
583 | goto fail; | |
584 | ||
585 | /* | |
586 | * Point into the IP datagram header. | |
587 | */ | |
588 | ||
589 | iph = ip_hdr(skb); | |
590 | ||
591 | mtu = ip_skb_dst_mtu(sk, skb); | |
592 | if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) | |
593 | mtu = IPCB(skb)->frag_max_size; | |
594 | ||
595 | /* | |
596 | * Setup starting values. | |
597 | */ | |
598 | ||
599 | hlen = iph->ihl * 4; | |
600 | mtu = mtu - hlen; /* Size of data space */ | |
601 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; | |
602 | ll_rs = LL_RESERVED_SPACE(rt->dst.dev); | |
603 | ||
604 | /* When frag_list is given, use it. First, check its validity: | |
605 | * some transformers could create wrong frag_list or break existing | |
606 | * one, it is not prohibited. In this case fall back to copying. | |
607 | * | |
608 | * LATER: this step can be merged to real generation of fragments, | |
609 | * we can switch to copy when see the first bad fragment. | |
610 | */ | |
611 | if (skb_has_frag_list(skb)) { | |
612 | struct sk_buff *frag, *frag2; | |
613 | unsigned int first_len = skb_pagelen(skb); | |
614 | ||
615 | if (first_len - hlen > mtu || | |
616 | ((first_len - hlen) & 7) || | |
617 | ip_is_fragment(iph) || | |
618 | skb_cloned(skb) || | |
619 | skb_headroom(skb) < ll_rs) | |
620 | goto slow_path; | |
621 | ||
622 | skb_walk_frags(skb, frag) { | |
623 | /* Correct geometry. */ | |
624 | if (frag->len > mtu || | |
625 | ((frag->len & 7) && frag->next) || | |
626 | skb_headroom(frag) < hlen + ll_rs) | |
627 | goto slow_path_clean; | |
628 | ||
629 | /* Partially cloned skb? */ | |
630 | if (skb_shared(frag)) | |
631 | goto slow_path_clean; | |
632 | ||
633 | BUG_ON(frag->sk); | |
634 | if (skb->sk) { | |
635 | frag->sk = skb->sk; | |
636 | frag->destructor = sock_wfree; | |
637 | } | |
638 | skb->truesize -= frag->truesize; | |
639 | } | |
640 | ||
641 | /* Everything is OK. Generate! */ | |
642 | ||
643 | err = 0; | |
644 | offset = 0; | |
645 | frag = skb_shinfo(skb)->frag_list; | |
646 | skb_frag_list_init(skb); | |
647 | skb->data_len = first_len - skb_headlen(skb); | |
648 | skb->len = first_len; | |
649 | iph->tot_len = htons(first_len); | |
650 | iph->frag_off = htons(IP_MF); | |
651 | ip_send_check(iph); | |
652 | ||
653 | for (;;) { | |
654 | /* Prepare header of the next frame, | |
655 | * before previous one went down. */ | |
656 | if (frag) { | |
657 | frag->ip_summed = CHECKSUM_NONE; | |
658 | skb_reset_transport_header(frag); | |
659 | __skb_push(frag, hlen); | |
660 | skb_reset_network_header(frag); | |
661 | memcpy(skb_network_header(frag), iph, hlen); | |
662 | iph = ip_hdr(frag); | |
663 | iph->tot_len = htons(frag->len); | |
664 | ip_copy_metadata(frag, skb); | |
665 | if (offset == 0) | |
666 | ip_options_fragment(frag); | |
667 | offset += skb->len - hlen; | |
668 | iph->frag_off = htons(offset>>3); | |
669 | if (frag->next) | |
670 | iph->frag_off |= htons(IP_MF); | |
671 | /* Ready, complete checksum */ | |
672 | ip_send_check(iph); | |
673 | } | |
674 | ||
675 | err = output(net, sk, skb); | |
676 | ||
677 | if (!err) | |
678 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); | |
679 | if (err || !frag) | |
680 | break; | |
681 | ||
682 | skb = frag; | |
683 | frag = skb->next; | |
684 | skb->next = NULL; | |
685 | } | |
686 | ||
687 | if (err == 0) { | |
688 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); | |
689 | return 0; | |
690 | } | |
691 | ||
692 | while (frag) { | |
693 | skb = frag->next; | |
694 | kfree_skb(frag); | |
695 | frag = skb; | |
696 | } | |
697 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); | |
698 | return err; | |
699 | ||
700 | slow_path_clean: | |
701 | skb_walk_frags(skb, frag2) { | |
702 | if (frag2 == frag) | |
703 | break; | |
704 | frag2->sk = NULL; | |
705 | frag2->destructor = NULL; | |
706 | skb->truesize += frag2->truesize; | |
707 | } | |
708 | } | |
709 | ||
710 | slow_path: | |
711 | iph = ip_hdr(skb); | |
712 | ||
713 | left = skb->len - hlen; /* Space per frame */ | |
714 | ptr = hlen; /* Where to start from */ | |
715 | ||
716 | /* | |
717 | * Fragment the datagram. | |
718 | */ | |
719 | ||
720 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
721 | not_last_frag = iph->frag_off & htons(IP_MF); | |
722 | ||
723 | /* | |
724 | * Keep copying data until we run out. | |
725 | */ | |
726 | ||
727 | while (left > 0) { | |
728 | len = left; | |
729 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
730 | if (len > mtu) | |
731 | len = mtu; | |
732 | /* IF: we are not sending up to and including the packet end | |
733 | then align the next start on an eight byte boundary */ | |
734 | if (len < left) { | |
735 | len &= ~7; | |
736 | } | |
737 | ||
738 | /* Allocate buffer */ | |
739 | skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC); | |
740 | if (!skb2) { | |
741 | err = -ENOMEM; | |
742 | goto fail; | |
743 | } | |
744 | ||
745 | /* | |
746 | * Set up data on packet | |
747 | */ | |
748 | ||
749 | ip_copy_metadata(skb2, skb); | |
750 | skb_reserve(skb2, ll_rs); | |
751 | skb_put(skb2, len + hlen); | |
752 | skb_reset_network_header(skb2); | |
753 | skb2->transport_header = skb2->network_header + hlen; | |
754 | ||
755 | /* | |
756 | * Charge the memory for the fragment to any owner | |
757 | * it might possess | |
758 | */ | |
759 | ||
760 | if (skb->sk) | |
761 | skb_set_owner_w(skb2, skb->sk); | |
762 | ||
763 | /* | |
764 | * Copy the packet header into the new buffer. | |
765 | */ | |
766 | ||
767 | skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); | |
768 | ||
769 | /* | |
770 | * Copy a block of the IP datagram. | |
771 | */ | |
772 | if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) | |
773 | BUG(); | |
774 | left -= len; | |
775 | ||
776 | /* | |
777 | * Fill in the new header fields. | |
778 | */ | |
779 | iph = ip_hdr(skb2); | |
780 | iph->frag_off = htons((offset >> 3)); | |
781 | ||
782 | if (IPCB(skb)->flags & IPSKB_FRAG_PMTU) | |
783 | iph->frag_off |= htons(IP_DF); | |
784 | ||
785 | /* ANK: dirty, but effective trick. Upgrade options only if | |
786 | * the segment to be fragmented was THE FIRST (otherwise, | |
787 | * options are already fixed) and make it ONCE | |
788 | * on the initial skb, so that all the following fragments | |
789 | * will inherit fixed options. | |
790 | */ | |
791 | if (offset == 0) | |
792 | ip_options_fragment(skb); | |
793 | ||
794 | /* | |
795 | * Added AC : If we are fragmenting a fragment that's not the | |
796 | * last fragment then keep MF on each bit | |
797 | */ | |
798 | if (left > 0 || not_last_frag) | |
799 | iph->frag_off |= htons(IP_MF); | |
800 | ptr += len; | |
801 | offset += len; | |
802 | ||
803 | /* | |
804 | * Put this fragment into the sending queue. | |
805 | */ | |
806 | iph->tot_len = htons(len + hlen); | |
807 | ||
808 | ip_send_check(iph); | |
809 | ||
810 | err = output(net, sk, skb2); | |
811 | if (err) | |
812 | goto fail; | |
813 | ||
814 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); | |
815 | } | |
816 | consume_skb(skb); | |
817 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); | |
818 | return err; | |
819 | ||
820 | fail: | |
821 | kfree_skb(skb); | |
822 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); | |
823 | return err; | |
824 | } | |
825 | EXPORT_SYMBOL(ip_do_fragment); | |
826 | ||
827 | int | |
828 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
829 | { | |
830 | struct msghdr *msg = from; | |
831 | ||
832 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
833 | if (!copy_from_iter_full(to, len, &msg->msg_iter)) | |
834 | return -EFAULT; | |
835 | } else { | |
836 | __wsum csum = 0; | |
837 | if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter)) | |
838 | return -EFAULT; | |
839 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
840 | } | |
841 | return 0; | |
842 | } | |
843 | EXPORT_SYMBOL(ip_generic_getfrag); | |
844 | ||
845 | static inline __wsum | |
846 | csum_page(struct page *page, int offset, int copy) | |
847 | { | |
848 | char *kaddr; | |
849 | __wsum csum; | |
850 | kaddr = kmap(page); | |
851 | csum = csum_partial(kaddr + offset, copy, 0); | |
852 | kunmap(page); | |
853 | return csum; | |
854 | } | |
855 | ||
856 | static int __ip_append_data(struct sock *sk, | |
857 | struct flowi4 *fl4, | |
858 | struct sk_buff_head *queue, | |
859 | struct inet_cork *cork, | |
860 | struct page_frag *pfrag, | |
861 | int getfrag(void *from, char *to, int offset, | |
862 | int len, int odd, struct sk_buff *skb), | |
863 | void *from, int length, int transhdrlen, | |
864 | unsigned int flags) | |
865 | { | |
866 | struct inet_sock *inet = inet_sk(sk); | |
867 | struct sk_buff *skb; | |
868 | ||
869 | struct ip_options *opt = cork->opt; | |
870 | int hh_len; | |
871 | int exthdrlen; | |
872 | int mtu; | |
873 | int copy; | |
874 | int err; | |
875 | int offset = 0; | |
876 | unsigned int maxfraglen, fragheaderlen, maxnonfragsize; | |
877 | int csummode = CHECKSUM_NONE; | |
878 | struct rtable *rt = (struct rtable *)cork->dst; | |
879 | unsigned int wmem_alloc_delta = 0; | |
880 | u32 tskey = 0; | |
881 | ||
882 | skb = skb_peek_tail(queue); | |
883 | ||
884 | exthdrlen = !skb ? rt->dst.header_len : 0; | |
885 | mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; | |
886 | ||
887 | if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && | |
888 | sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) | |
889 | tskey = sk->sk_tskey++; | |
890 | ||
891 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); | |
892 | ||
893 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
894 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
895 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; | |
896 | ||
897 | if (cork->length + length > maxnonfragsize - fragheaderlen) { | |
898 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, | |
899 | mtu - (opt ? opt->optlen : 0)); | |
900 | return -EMSGSIZE; | |
901 | } | |
902 | ||
903 | /* | |
904 | * transhdrlen > 0 means that this is the first fragment and we wish | |
905 | * it won't be fragmented in the future. | |
906 | */ | |
907 | if (transhdrlen && | |
908 | length + fragheaderlen <= mtu && | |
909 | rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && | |
910 | (!(flags & MSG_MORE) || cork->gso_size) && | |
911 | !exthdrlen) | |
912 | csummode = CHECKSUM_PARTIAL; | |
913 | ||
914 | cork->length += length; | |
915 | ||
916 | /* So, what's going on in the loop below? | |
917 | * | |
918 | * We use calculated fragment length to generate chained skb, | |
919 | * each of segments is IP fragment ready for sending to network after | |
920 | * adding appropriate IP header. | |
921 | */ | |
922 | ||
923 | if (!skb) | |
924 | goto alloc_new_skb; | |
925 | ||
926 | while (length > 0) { | |
927 | /* Check if the remaining data fits into current packet. */ | |
928 | copy = mtu - skb->len; | |
929 | if (copy < length) | |
930 | copy = maxfraglen - skb->len; | |
931 | if (copy <= 0) { | |
932 | char *data; | |
933 | unsigned int datalen; | |
934 | unsigned int fraglen; | |
935 | unsigned int fraggap; | |
936 | unsigned int alloclen; | |
937 | struct sk_buff *skb_prev; | |
938 | alloc_new_skb: | |
939 | skb_prev = skb; | |
940 | if (skb_prev) | |
941 | fraggap = skb_prev->len - maxfraglen; | |
942 | else | |
943 | fraggap = 0; | |
944 | ||
945 | /* | |
946 | * If remaining data exceeds the mtu, | |
947 | * we know we need more fragment(s). | |
948 | */ | |
949 | datalen = length + fraggap; | |
950 | if (datalen > mtu - fragheaderlen) | |
951 | datalen = maxfraglen - fragheaderlen; | |
952 | fraglen = datalen + fragheaderlen; | |
953 | ||
954 | if ((flags & MSG_MORE) && | |
955 | !(rt->dst.dev->features&NETIF_F_SG)) | |
956 | alloclen = mtu; | |
957 | else | |
958 | alloclen = fraglen; | |
959 | ||
960 | alloclen += exthdrlen; | |
961 | ||
962 | /* The last fragment gets additional space at tail. | |
963 | * Note, with MSG_MORE we overallocate on fragments, | |
964 | * because we have no idea what fragment will be | |
965 | * the last. | |
966 | */ | |
967 | if (datalen == length + fraggap) | |
968 | alloclen += rt->dst.trailer_len; | |
969 | ||
970 | if (transhdrlen) { | |
971 | skb = sock_alloc_send_skb(sk, | |
972 | alloclen + hh_len + 15, | |
973 | (flags & MSG_DONTWAIT), &err); | |
974 | } else { | |
975 | skb = NULL; | |
976 | if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <= | |
977 | 2 * sk->sk_sndbuf) | |
978 | skb = alloc_skb(alloclen + hh_len + 15, | |
979 | sk->sk_allocation); | |
980 | if (unlikely(!skb)) | |
981 | err = -ENOBUFS; | |
982 | } | |
983 | if (!skb) | |
984 | goto error; | |
985 | ||
986 | /* | |
987 | * Fill in the control structures | |
988 | */ | |
989 | skb->ip_summed = csummode; | |
990 | skb->csum = 0; | |
991 | skb_reserve(skb, hh_len); | |
992 | ||
993 | /* only the initial fragment is time stamped */ | |
994 | skb_shinfo(skb)->tx_flags = cork->tx_flags; | |
995 | cork->tx_flags = 0; | |
996 | skb_shinfo(skb)->tskey = tskey; | |
997 | tskey = 0; | |
998 | ||
999 | /* | |
1000 | * Find where to start putting bytes. | |
1001 | */ | |
1002 | data = skb_put(skb, fraglen + exthdrlen); | |
1003 | skb_set_network_header(skb, exthdrlen); | |
1004 | skb->transport_header = (skb->network_header + | |
1005 | fragheaderlen); | |
1006 | data += fragheaderlen + exthdrlen; | |
1007 | ||
1008 | if (fraggap) { | |
1009 | skb->csum = skb_copy_and_csum_bits( | |
1010 | skb_prev, maxfraglen, | |
1011 | data + transhdrlen, fraggap, 0); | |
1012 | skb_prev->csum = csum_sub(skb_prev->csum, | |
1013 | skb->csum); | |
1014 | data += fraggap; | |
1015 | pskb_trim_unique(skb_prev, maxfraglen); | |
1016 | } | |
1017 | ||
1018 | copy = datalen - transhdrlen - fraggap; | |
1019 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
1020 | err = -EFAULT; | |
1021 | kfree_skb(skb); | |
1022 | goto error; | |
1023 | } | |
1024 | ||
1025 | offset += copy; | |
1026 | length -= datalen - fraggap; | |
1027 | transhdrlen = 0; | |
1028 | exthdrlen = 0; | |
1029 | csummode = CHECKSUM_NONE; | |
1030 | ||
1031 | if ((flags & MSG_CONFIRM) && !skb_prev) | |
1032 | skb_set_dst_pending_confirm(skb, 1); | |
1033 | ||
1034 | /* | |
1035 | * Put the packet on the pending queue. | |
1036 | */ | |
1037 | if (!skb->destructor) { | |
1038 | skb->destructor = sock_wfree; | |
1039 | skb->sk = sk; | |
1040 | wmem_alloc_delta += skb->truesize; | |
1041 | } | |
1042 | __skb_queue_tail(queue, skb); | |
1043 | continue; | |
1044 | } | |
1045 | ||
1046 | if (copy > length) | |
1047 | copy = length; | |
1048 | ||
1049 | if (!(rt->dst.dev->features&NETIF_F_SG)) { | |
1050 | unsigned int off; | |
1051 | ||
1052 | off = skb->len; | |
1053 | if (getfrag(from, skb_put(skb, copy), | |
1054 | offset, copy, off, skb) < 0) { | |
1055 | __skb_trim(skb, off); | |
1056 | err = -EFAULT; | |
1057 | goto error; | |
1058 | } | |
1059 | } else { | |
1060 | int i = skb_shinfo(skb)->nr_frags; | |
1061 | ||
1062 | err = -ENOMEM; | |
1063 | if (!sk_page_frag_refill(sk, pfrag)) | |
1064 | goto error; | |
1065 | ||
1066 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
1067 | pfrag->offset)) { | |
1068 | err = -EMSGSIZE; | |
1069 | if (i == MAX_SKB_FRAGS) | |
1070 | goto error; | |
1071 | ||
1072 | __skb_fill_page_desc(skb, i, pfrag->page, | |
1073 | pfrag->offset, 0); | |
1074 | skb_shinfo(skb)->nr_frags = ++i; | |
1075 | get_page(pfrag->page); | |
1076 | } | |
1077 | copy = min_t(int, copy, pfrag->size - pfrag->offset); | |
1078 | if (getfrag(from, | |
1079 | page_address(pfrag->page) + pfrag->offset, | |
1080 | offset, copy, skb->len, skb) < 0) | |
1081 | goto error_efault; | |
1082 | ||
1083 | pfrag->offset += copy; | |
1084 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1085 | skb->len += copy; | |
1086 | skb->data_len += copy; | |
1087 | skb->truesize += copy; | |
1088 | wmem_alloc_delta += copy; | |
1089 | } | |
1090 | offset += copy; | |
1091 | length -= copy; | |
1092 | } | |
1093 | ||
1094 | if (wmem_alloc_delta) | |
1095 | refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); | |
1096 | return 0; | |
1097 | ||
1098 | error_efault: | |
1099 | err = -EFAULT; | |
1100 | error: | |
1101 | cork->length -= length; | |
1102 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); | |
1103 | refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); | |
1104 | return err; | |
1105 | } | |
1106 | ||
1107 | static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, | |
1108 | struct ipcm_cookie *ipc, struct rtable **rtp) | |
1109 | { | |
1110 | struct ip_options_rcu *opt; | |
1111 | struct rtable *rt; | |
1112 | ||
1113 | rt = *rtp; | |
1114 | if (unlikely(!rt)) | |
1115 | return -EFAULT; | |
1116 | ||
1117 | /* | |
1118 | * setup for corking. | |
1119 | */ | |
1120 | opt = ipc->opt; | |
1121 | if (opt) { | |
1122 | if (!cork->opt) { | |
1123 | cork->opt = kmalloc(sizeof(struct ip_options) + 40, | |
1124 | sk->sk_allocation); | |
1125 | if (unlikely(!cork->opt)) | |
1126 | return -ENOBUFS; | |
1127 | } | |
1128 | memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); | |
1129 | cork->flags |= IPCORK_OPT; | |
1130 | cork->addr = ipc->addr; | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * We steal reference to this route, caller should not release it | |
1135 | */ | |
1136 | *rtp = NULL; | |
1137 | cork->fragsize = ip_sk_use_pmtu(sk) ? | |
1138 | dst_mtu(&rt->dst) : rt->dst.dev->mtu; | |
1139 | ||
1140 | cork->gso_size = sk->sk_type == SOCK_DGRAM ? ipc->gso_size : 0; | |
1141 | cork->dst = &rt->dst; | |
1142 | cork->length = 0; | |
1143 | cork->ttl = ipc->ttl; | |
1144 | cork->tos = ipc->tos; | |
1145 | cork->priority = ipc->priority; | |
1146 | cork->tx_flags = ipc->tx_flags; | |
1147 | ||
1148 | return 0; | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
1153 | * from many pieces of data. Each pieces will be holded on the socket | |
1154 | * until ip_push_pending_frames() is called. Each piece can be a page | |
1155 | * or non-page data. | |
1156 | * | |
1157 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
1158 | * this interface potentially. | |
1159 | * | |
1160 | * LATER: length must be adjusted by pad at tail, when it is required. | |
1161 | */ | |
1162 | int ip_append_data(struct sock *sk, struct flowi4 *fl4, | |
1163 | int getfrag(void *from, char *to, int offset, int len, | |
1164 | int odd, struct sk_buff *skb), | |
1165 | void *from, int length, int transhdrlen, | |
1166 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1167 | unsigned int flags) | |
1168 | { | |
1169 | struct inet_sock *inet = inet_sk(sk); | |
1170 | int err; | |
1171 | ||
1172 | if (flags&MSG_PROBE) | |
1173 | return 0; | |
1174 | ||
1175 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
1176 | err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); | |
1177 | if (err) | |
1178 | return err; | |
1179 | } else { | |
1180 | transhdrlen = 0; | |
1181 | } | |
1182 | ||
1183 | return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, | |
1184 | sk_page_frag(sk), getfrag, | |
1185 | from, length, transhdrlen, flags); | |
1186 | } | |
1187 | ||
1188 | ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, | |
1189 | int offset, size_t size, int flags) | |
1190 | { | |
1191 | struct inet_sock *inet = inet_sk(sk); | |
1192 | struct sk_buff *skb; | |
1193 | struct rtable *rt; | |
1194 | struct ip_options *opt = NULL; | |
1195 | struct inet_cork *cork; | |
1196 | int hh_len; | |
1197 | int mtu; | |
1198 | int len; | |
1199 | int err; | |
1200 | unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; | |
1201 | ||
1202 | if (inet->hdrincl) | |
1203 | return -EPERM; | |
1204 | ||
1205 | if (flags&MSG_PROBE) | |
1206 | return 0; | |
1207 | ||
1208 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1209 | return -EINVAL; | |
1210 | ||
1211 | cork = &inet->cork.base; | |
1212 | rt = (struct rtable *)cork->dst; | |
1213 | if (cork->flags & IPCORK_OPT) | |
1214 | opt = cork->opt; | |
1215 | ||
1216 | if (!(rt->dst.dev->features&NETIF_F_SG)) | |
1217 | return -EOPNOTSUPP; | |
1218 | ||
1219 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); | |
1220 | mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; | |
1221 | ||
1222 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1223 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
1224 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; | |
1225 | ||
1226 | if (cork->length + size > maxnonfragsize - fragheaderlen) { | |
1227 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, | |
1228 | mtu - (opt ? opt->optlen : 0)); | |
1229 | return -EMSGSIZE; | |
1230 | } | |
1231 | ||
1232 | skb = skb_peek_tail(&sk->sk_write_queue); | |
1233 | if (!skb) | |
1234 | return -EINVAL; | |
1235 | ||
1236 | cork->length += size; | |
1237 | ||
1238 | while (size > 0) { | |
1239 | /* Check if the remaining data fits into current packet. */ | |
1240 | len = mtu - skb->len; | |
1241 | if (len < size) | |
1242 | len = maxfraglen - skb->len; | |
1243 | ||
1244 | if (len <= 0) { | |
1245 | struct sk_buff *skb_prev; | |
1246 | int alloclen; | |
1247 | ||
1248 | skb_prev = skb; | |
1249 | fraggap = skb_prev->len - maxfraglen; | |
1250 | ||
1251 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1252 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1253 | if (unlikely(!skb)) { | |
1254 | err = -ENOBUFS; | |
1255 | goto error; | |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * Fill in the control structures | |
1260 | */ | |
1261 | skb->ip_summed = CHECKSUM_NONE; | |
1262 | skb->csum = 0; | |
1263 | skb_reserve(skb, hh_len); | |
1264 | ||
1265 | /* | |
1266 | * Find where to start putting bytes. | |
1267 | */ | |
1268 | skb_put(skb, fragheaderlen + fraggap); | |
1269 | skb_reset_network_header(skb); | |
1270 | skb->transport_header = (skb->network_header + | |
1271 | fragheaderlen); | |
1272 | if (fraggap) { | |
1273 | skb->csum = skb_copy_and_csum_bits(skb_prev, | |
1274 | maxfraglen, | |
1275 | skb_transport_header(skb), | |
1276 | fraggap, 0); | |
1277 | skb_prev->csum = csum_sub(skb_prev->csum, | |
1278 | skb->csum); | |
1279 | pskb_trim_unique(skb_prev, maxfraglen); | |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * Put the packet on the pending queue. | |
1284 | */ | |
1285 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1286 | continue; | |
1287 | } | |
1288 | ||
1289 | if (len > size) | |
1290 | len = size; | |
1291 | ||
1292 | if (skb_append_pagefrags(skb, page, offset, len)) { | |
1293 | err = -EMSGSIZE; | |
1294 | goto error; | |
1295 | } | |
1296 | ||
1297 | if (skb->ip_summed == CHECKSUM_NONE) { | |
1298 | __wsum csum; | |
1299 | csum = csum_page(page, offset, len); | |
1300 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1301 | } | |
1302 | ||
1303 | skb->len += len; | |
1304 | skb->data_len += len; | |
1305 | skb->truesize += len; | |
1306 | refcount_add(len, &sk->sk_wmem_alloc); | |
1307 | offset += len; | |
1308 | size -= len; | |
1309 | } | |
1310 | return 0; | |
1311 | ||
1312 | error: | |
1313 | cork->length -= size; | |
1314 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); | |
1315 | return err; | |
1316 | } | |
1317 | ||
1318 | static void ip_cork_release(struct inet_cork *cork) | |
1319 | { | |
1320 | cork->flags &= ~IPCORK_OPT; | |
1321 | kfree(cork->opt); | |
1322 | cork->opt = NULL; | |
1323 | dst_release(cork->dst); | |
1324 | cork->dst = NULL; | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * Combined all pending IP fragments on the socket as one IP datagram | |
1329 | * and push them out. | |
1330 | */ | |
1331 | struct sk_buff *__ip_make_skb(struct sock *sk, | |
1332 | struct flowi4 *fl4, | |
1333 | struct sk_buff_head *queue, | |
1334 | struct inet_cork *cork) | |
1335 | { | |
1336 | struct sk_buff *skb, *tmp_skb; | |
1337 | struct sk_buff **tail_skb; | |
1338 | struct inet_sock *inet = inet_sk(sk); | |
1339 | struct net *net = sock_net(sk); | |
1340 | struct ip_options *opt = NULL; | |
1341 | struct rtable *rt = (struct rtable *)cork->dst; | |
1342 | struct iphdr *iph; | |
1343 | __be16 df = 0; | |
1344 | __u8 ttl; | |
1345 | ||
1346 | skb = __skb_dequeue(queue); | |
1347 | if (!skb) | |
1348 | goto out; | |
1349 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1350 | ||
1351 | /* move skb->data to ip header from ext header */ | |
1352 | if (skb->data < skb_network_header(skb)) | |
1353 | __skb_pull(skb, skb_network_offset(skb)); | |
1354 | while ((tmp_skb = __skb_dequeue(queue)) != NULL) { | |
1355 | __skb_pull(tmp_skb, skb_network_header_len(skb)); | |
1356 | *tail_skb = tmp_skb; | |
1357 | tail_skb = &(tmp_skb->next); | |
1358 | skb->len += tmp_skb->len; | |
1359 | skb->data_len += tmp_skb->len; | |
1360 | skb->truesize += tmp_skb->truesize; | |
1361 | tmp_skb->destructor = NULL; | |
1362 | tmp_skb->sk = NULL; | |
1363 | } | |
1364 | ||
1365 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1366 | * to fragment the frame generated here. No matter, what transforms | |
1367 | * how transforms change size of the packet, it will come out. | |
1368 | */ | |
1369 | skb->ignore_df = ip_sk_ignore_df(sk); | |
1370 | ||
1371 | /* DF bit is set when we want to see DF on outgoing frames. | |
1372 | * If ignore_df is set too, we still allow to fragment this frame | |
1373 | * locally. */ | |
1374 | if (inet->pmtudisc == IP_PMTUDISC_DO || | |
1375 | inet->pmtudisc == IP_PMTUDISC_PROBE || | |
1376 | (skb->len <= dst_mtu(&rt->dst) && | |
1377 | ip_dont_fragment(sk, &rt->dst))) | |
1378 | df = htons(IP_DF); | |
1379 | ||
1380 | if (cork->flags & IPCORK_OPT) | |
1381 | opt = cork->opt; | |
1382 | ||
1383 | if (cork->ttl != 0) | |
1384 | ttl = cork->ttl; | |
1385 | else if (rt->rt_type == RTN_MULTICAST) | |
1386 | ttl = inet->mc_ttl; | |
1387 | else | |
1388 | ttl = ip_select_ttl(inet, &rt->dst); | |
1389 | ||
1390 | iph = ip_hdr(skb); | |
1391 | iph->version = 4; | |
1392 | iph->ihl = 5; | |
1393 | iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; | |
1394 | iph->frag_off = df; | |
1395 | iph->ttl = ttl; | |
1396 | iph->protocol = sk->sk_protocol; | |
1397 | ip_copy_addrs(iph, fl4); | |
1398 | ip_select_ident(net, skb, sk); | |
1399 | ||
1400 | if (opt) { | |
1401 | iph->ihl += opt->optlen>>2; | |
1402 | ip_options_build(skb, opt, cork->addr, rt, 0); | |
1403 | } | |
1404 | ||
1405 | skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; | |
1406 | skb->mark = sk->sk_mark; | |
1407 | /* | |
1408 | * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec | |
1409 | * on dst refcount | |
1410 | */ | |
1411 | cork->dst = NULL; | |
1412 | skb_dst_set(skb, &rt->dst); | |
1413 | ||
1414 | if (iph->protocol == IPPROTO_ICMP) | |
1415 | icmp_out_count(net, ((struct icmphdr *) | |
1416 | skb_transport_header(skb))->type); | |
1417 | ||
1418 | ip_cork_release(cork); | |
1419 | out: | |
1420 | return skb; | |
1421 | } | |
1422 | ||
1423 | int ip_send_skb(struct net *net, struct sk_buff *skb) | |
1424 | { | |
1425 | int err; | |
1426 | ||
1427 | err = ip_local_out(net, skb->sk, skb); | |
1428 | if (err) { | |
1429 | if (err > 0) | |
1430 | err = net_xmit_errno(err); | |
1431 | if (err) | |
1432 | IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); | |
1433 | } | |
1434 | ||
1435 | return err; | |
1436 | } | |
1437 | ||
1438 | int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) | |
1439 | { | |
1440 | struct sk_buff *skb; | |
1441 | ||
1442 | skb = ip_finish_skb(sk, fl4); | |
1443 | if (!skb) | |
1444 | return 0; | |
1445 | ||
1446 | /* Netfilter gets whole the not fragmented skb. */ | |
1447 | return ip_send_skb(sock_net(sk), skb); | |
1448 | } | |
1449 | ||
1450 | /* | |
1451 | * Throw away all pending data on the socket. | |
1452 | */ | |
1453 | static void __ip_flush_pending_frames(struct sock *sk, | |
1454 | struct sk_buff_head *queue, | |
1455 | struct inet_cork *cork) | |
1456 | { | |
1457 | struct sk_buff *skb; | |
1458 | ||
1459 | while ((skb = __skb_dequeue_tail(queue)) != NULL) | |
1460 | kfree_skb(skb); | |
1461 | ||
1462 | ip_cork_release(cork); | |
1463 | } | |
1464 | ||
1465 | void ip_flush_pending_frames(struct sock *sk) | |
1466 | { | |
1467 | __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); | |
1468 | } | |
1469 | ||
1470 | struct sk_buff *ip_make_skb(struct sock *sk, | |
1471 | struct flowi4 *fl4, | |
1472 | int getfrag(void *from, char *to, int offset, | |
1473 | int len, int odd, struct sk_buff *skb), | |
1474 | void *from, int length, int transhdrlen, | |
1475 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1476 | struct inet_cork *cork, unsigned int flags) | |
1477 | { | |
1478 | struct sk_buff_head queue; | |
1479 | int err; | |
1480 | ||
1481 | if (flags & MSG_PROBE) | |
1482 | return NULL; | |
1483 | ||
1484 | __skb_queue_head_init(&queue); | |
1485 | ||
1486 | cork->flags = 0; | |
1487 | cork->addr = 0; | |
1488 | cork->opt = NULL; | |
1489 | err = ip_setup_cork(sk, cork, ipc, rtp); | |
1490 | if (err) | |
1491 | return ERR_PTR(err); | |
1492 | ||
1493 | err = __ip_append_data(sk, fl4, &queue, cork, | |
1494 | ¤t->task_frag, getfrag, | |
1495 | from, length, transhdrlen, flags); | |
1496 | if (err) { | |
1497 | __ip_flush_pending_frames(sk, &queue, cork); | |
1498 | return ERR_PTR(err); | |
1499 | } | |
1500 | ||
1501 | return __ip_make_skb(sk, fl4, &queue, cork); | |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * Fetch data from kernel space and fill in checksum if needed. | |
1506 | */ | |
1507 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, | |
1508 | int len, int odd, struct sk_buff *skb) | |
1509 | { | |
1510 | __wsum csum; | |
1511 | ||
1512 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1513 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * Generic function to send a packet as reply to another packet. | |
1519 | * Used to send some TCP resets/acks so far. | |
1520 | */ | |
1521 | void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, | |
1522 | const struct ip_options *sopt, | |
1523 | __be32 daddr, __be32 saddr, | |
1524 | const struct ip_reply_arg *arg, | |
1525 | unsigned int len) | |
1526 | { | |
1527 | struct ip_options_data replyopts; | |
1528 | struct ipcm_cookie ipc; | |
1529 | struct flowi4 fl4; | |
1530 | struct rtable *rt = skb_rtable(skb); | |
1531 | struct net *net = sock_net(sk); | |
1532 | struct sk_buff *nskb; | |
1533 | int err; | |
1534 | int oif; | |
1535 | ||
1536 | if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt)) | |
1537 | return; | |
1538 | ||
1539 | ipc.addr = daddr; | |
1540 | ipc.opt = NULL; | |
1541 | ipc.tx_flags = 0; | |
1542 | ipc.ttl = 0; | |
1543 | ipc.tos = -1; | |
1544 | ||
1545 | if (replyopts.opt.opt.optlen) { | |
1546 | ipc.opt = &replyopts.opt; | |
1547 | ||
1548 | if (replyopts.opt.opt.srr) | |
1549 | daddr = replyopts.opt.opt.faddr; | |
1550 | } | |
1551 | ||
1552 | oif = arg->bound_dev_if; | |
1553 | if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) | |
1554 | oif = skb->skb_iif; | |
1555 | ||
1556 | flowi4_init_output(&fl4, oif, | |
1557 | IP4_REPLY_MARK(net, skb->mark), | |
1558 | RT_TOS(arg->tos), | |
1559 | RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, | |
1560 | ip_reply_arg_flowi_flags(arg), | |
1561 | daddr, saddr, | |
1562 | tcp_hdr(skb)->source, tcp_hdr(skb)->dest, | |
1563 | arg->uid); | |
1564 | security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); | |
1565 | rt = ip_route_output_key(net, &fl4); | |
1566 | if (IS_ERR(rt)) | |
1567 | return; | |
1568 | ||
1569 | inet_sk(sk)->tos = arg->tos; | |
1570 | ||
1571 | sk->sk_priority = skb->priority; | |
1572 | sk->sk_protocol = ip_hdr(skb)->protocol; | |
1573 | sk->sk_bound_dev_if = arg->bound_dev_if; | |
1574 | sk->sk_sndbuf = sysctl_wmem_default; | |
1575 | sk->sk_mark = fl4.flowi4_mark; | |
1576 | err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, | |
1577 | len, 0, &ipc, &rt, MSG_DONTWAIT); | |
1578 | if (unlikely(err)) { | |
1579 | ip_flush_pending_frames(sk); | |
1580 | goto out; | |
1581 | } | |
1582 | ||
1583 | nskb = skb_peek(&sk->sk_write_queue); | |
1584 | if (nskb) { | |
1585 | if (arg->csumoffset >= 0) | |
1586 | *((__sum16 *)skb_transport_header(nskb) + | |
1587 | arg->csumoffset) = csum_fold(csum_add(nskb->csum, | |
1588 | arg->csum)); | |
1589 | nskb->ip_summed = CHECKSUM_NONE; | |
1590 | ip_push_pending_frames(sk, &fl4); | |
1591 | } | |
1592 | out: | |
1593 | ip_rt_put(rt); | |
1594 | } | |
1595 | ||
1596 | void __init ip_init(void) | |
1597 | { | |
1598 | ip_rt_init(); | |
1599 | inet_initpeers(); | |
1600 | ||
1601 | #if defined(CONFIG_IP_MULTICAST) | |
1602 | igmp_mc_init(); | |
1603 | #endif | |
1604 | } |