Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * The Internet Protocol (IP) output module. | |
8 | * | |
02c30a84 | 9 | * Authors: Ross Biro |
1da177e4 LT |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Donald Becker, <becker@super.org> | |
12 | * Alan Cox, <Alan.Cox@linux.org> | |
13 | * Richard Underwood | |
14 | * Stefan Becker, <stefanb@yello.ping.de> | |
15 | * Jorge Cwik, <jorge@laser.satlink.net> | |
16 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
17 | * Hirokazu Takahashi, <taka@valinux.co.jp> | |
18 | * | |
19 | * See ip_input.c for original log | |
20 | * | |
21 | * Fixes: | |
22 | * Alan Cox : Missing nonblock feature in ip_build_xmit. | |
23 | * Mike Kilburn : htons() missing in ip_build_xmit. | |
e905a9ed | 24 | * Bradford Johnson: Fix faulty handling of some frames when |
1da177e4 LT |
25 | * no route is found. |
26 | * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit | |
27 | * (in case if packet not accepted by | |
28 | * output firewall rules) | |
29 | * Mike McLagan : Routing by source | |
30 | * Alexey Kuznetsov: use new route cache | |
31 | * Andi Kleen: Fix broken PMTU recovery and remove | |
32 | * some redundant tests. | |
33 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. | |
34 | * Andi Kleen : Replace ip_reply with ip_send_reply. | |
e905a9ed YH |
35 | * Andi Kleen : Split fast and slow ip_build_xmit path |
36 | * for decreased register pressure on x86 | |
a66e04ce | 37 | * and more readability. |
1da177e4 LT |
38 | * Marc Boucher : When call_out_firewall returns FW_QUEUE, |
39 | * silently drop skb instead of failing with -EPERM. | |
40 | * Detlev Wengorz : Copy protocol for fragments. | |
41 | * Hirokazu Takahashi: HW checksumming for outgoing UDP | |
42 | * datagrams. | |
43 | * Hirokazu Takahashi: sendfile() on UDP works now. | |
44 | */ | |
45 | ||
7c0f6ba6 | 46 | #include <linux/uaccess.h> |
1da177e4 LT |
47 | #include <linux/module.h> |
48 | #include <linux/types.h> | |
49 | #include <linux/kernel.h> | |
1da177e4 LT |
50 | #include <linux/mm.h> |
51 | #include <linux/string.h> | |
52 | #include <linux/errno.h> | |
a1f8e7f7 | 53 | #include <linux/highmem.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
1da177e4 LT |
55 | |
56 | #include <linux/socket.h> | |
57 | #include <linux/sockios.h> | |
58 | #include <linux/in.h> | |
59 | #include <linux/inet.h> | |
60 | #include <linux/netdevice.h> | |
61 | #include <linux/etherdevice.h> | |
62 | #include <linux/proc_fs.h> | |
63 | #include <linux/stat.h> | |
64 | #include <linux/init.h> | |
65 | ||
66 | #include <net/snmp.h> | |
67 | #include <net/ip.h> | |
68 | #include <net/protocol.h> | |
69 | #include <net/route.h> | |
cfacb057 | 70 | #include <net/xfrm.h> |
1da177e4 LT |
71 | #include <linux/skbuff.h> |
72 | #include <net/sock.h> | |
73 | #include <net/arp.h> | |
74 | #include <net/icmp.h> | |
1da177e4 LT |
75 | #include <net/checksum.h> |
76 | #include <net/inetpeer.h> | |
ba9e04a7 | 77 | #include <net/inet_ecn.h> |
14972cbd | 78 | #include <net/lwtunnel.h> |
33b48679 | 79 | #include <linux/bpf-cgroup.h> |
1da177e4 LT |
80 | #include <linux/igmp.h> |
81 | #include <linux/netfilter_ipv4.h> | |
82 | #include <linux/netfilter_bridge.h> | |
1da177e4 | 83 | #include <linux/netlink.h> |
6cbb0df7 | 84 | #include <linux/tcp.h> |
1da177e4 | 85 | |
694869b3 EB |
86 | static int |
87 | ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, | |
88 | unsigned int mtu, | |
89 | int (*output)(struct net *, struct sock *, struct sk_buff *)); | |
49d16b23 | 90 | |
1da177e4 | 91 | /* Generate a checksum for an outgoing IP datagram. */ |
2fbd9679 | 92 | void ip_send_check(struct iphdr *iph) |
1da177e4 LT |
93 | { |
94 | iph->check = 0; | |
95 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
96 | } | |
4bc2f18b | 97 | EXPORT_SYMBOL(ip_send_check); |
1da177e4 | 98 | |
cf91a99d | 99 | int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) |
c439cb2e HX |
100 | { |
101 | struct iphdr *iph = ip_hdr(skb); | |
102 | ||
103 | iph->tot_len = htons(skb->len); | |
104 | ip_send_check(iph); | |
a8e3e1a9 DA |
105 | |
106 | /* if egress device is enslaved to an L3 master device pass the | |
107 | * skb to its handler for processing | |
108 | */ | |
109 | skb = l3mdev_ip_out(sk, skb); | |
110 | if (unlikely(!skb)) | |
111 | return 0; | |
112 | ||
f4180439 EC |
113 | skb->protocol = htons(ETH_P_IP); |
114 | ||
29a26a56 EB |
115 | return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, |
116 | net, sk, skb, NULL, skb_dst(skb)->dev, | |
13206b6b | 117 | dst_output); |
7026b1dd DM |
118 | } |
119 | ||
33224b16 | 120 | int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) |
c439cb2e HX |
121 | { |
122 | int err; | |
123 | ||
cf91a99d | 124 | err = __ip_local_out(net, sk, skb); |
c439cb2e | 125 | if (likely(err == 1)) |
13206b6b | 126 | err = dst_output(net, sk, skb); |
c439cb2e HX |
127 | |
128 | return err; | |
129 | } | |
e2cb77db | 130 | EXPORT_SYMBOL_GPL(ip_local_out); |
c439cb2e | 131 | |
1da177e4 LT |
132 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) |
133 | { | |
134 | int ttl = inet->uc_ttl; | |
135 | ||
136 | if (ttl < 0) | |
323e126f | 137 | ttl = ip4_dst_hoplimit(dst); |
1da177e4 LT |
138 | return ttl; |
139 | } | |
140 | ||
e905a9ed | 141 | /* |
1da177e4 LT |
142 | * Add an ip header to a skbuff and send it out. |
143 | * | |
144 | */ | |
cfe673b0 | 145 | int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, |
de033b7d WW |
146 | __be32 saddr, __be32 daddr, struct ip_options_rcu *opt, |
147 | u8 tos) | |
1da177e4 LT |
148 | { |
149 | struct inet_sock *inet = inet_sk(sk); | |
511c3f92 | 150 | struct rtable *rt = skb_rtable(skb); |
77589ce0 | 151 | struct net *net = sock_net(sk); |
1da177e4 LT |
152 | struct iphdr *iph; |
153 | ||
154 | /* Build the IP header. */ | |
f6d8bd05 | 155 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); |
8856dfa3 | 156 | skb_reset_network_header(skb); |
eddc9ec5 | 157 | iph = ip_hdr(skb); |
1da177e4 LT |
158 | iph->version = 4; |
159 | iph->ihl = 5; | |
de033b7d | 160 | iph->tos = tos; |
d8d1f30b | 161 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
dd927a26 DM |
162 | iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); |
163 | iph->saddr = saddr; | |
1da177e4 | 164 | iph->protocol = sk->sk_protocol; |
cfe673b0 ED |
165 | if (ip_dont_fragment(sk, &rt->dst)) { |
166 | iph->frag_off = htons(IP_DF); | |
167 | iph->id = 0; | |
168 | } else { | |
169 | iph->frag_off = 0; | |
77589ce0 | 170 | __ip_select_ident(net, iph, 1); |
cfe673b0 | 171 | } |
1da177e4 | 172 | |
f6d8bd05 ED |
173 | if (opt && opt->opt.optlen) { |
174 | iph->ihl += opt->opt.optlen>>2; | |
175 | ip_options_build(skb, &opt->opt, daddr, rt, 0); | |
1da177e4 | 176 | } |
1da177e4 LT |
177 | |
178 | skb->priority = sk->sk_priority; | |
e05a90ec JHS |
179 | if (!skb->mark) |
180 | skb->mark = sk->sk_mark; | |
1da177e4 LT |
181 | |
182 | /* Send it out. */ | |
33224b16 | 183 | return ip_local_out(net, skb->sk, skb); |
1da177e4 | 184 | } |
d8c97a94 ACM |
185 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
186 | ||
694869b3 | 187 | static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 188 | { |
adf30907 | 189 | struct dst_entry *dst = skb_dst(skb); |
80787ebc | 190 | struct rtable *rt = (struct rtable *)dst; |
1da177e4 | 191 | struct net_device *dev = dst->dev; |
c2636b4d | 192 | unsigned int hh_len = LL_RESERVED_SPACE(dev); |
f6b72b62 | 193 | struct neighbour *neigh; |
5c9f7c1d | 194 | bool is_v6gw = false; |
1da177e4 | 195 | |
edf391ff | 196 | if (rt->rt_type == RTN_MULTICAST) { |
4ba1bf42 | 197 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); |
edf391ff | 198 | } else if (rt->rt_type == RTN_BROADCAST) |
4ba1bf42 | 199 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); |
80787ebc | 200 | |
1da177e4 | 201 | /* Be paranoid, rather than too clever. */ |
3b04ddde | 202 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { |
1da177e4 LT |
203 | struct sk_buff *skb2; |
204 | ||
205 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
51456b29 | 206 | if (!skb2) { |
1da177e4 LT |
207 | kfree_skb(skb); |
208 | return -ENOMEM; | |
209 | } | |
210 | if (skb->sk) | |
211 | skb_set_owner_w(skb2, skb->sk); | |
5d0ba55b | 212 | consume_skb(skb); |
1da177e4 LT |
213 | skb = skb2; |
214 | } | |
215 | ||
14972cbd RP |
216 | if (lwtunnel_xmit_redirect(dst->lwtstate)) { |
217 | int res = lwtunnel_xmit(skb); | |
218 | ||
219 | if (res < 0 || res == LWTUNNEL_XMIT_DONE) | |
220 | return res; | |
221 | } | |
222 | ||
a263b309 | 223 | rcu_read_lock_bh(); |
5c9f7c1d | 224 | neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); |
9871f1ad | 225 | if (!IS_ERR(neigh)) { |
4ff06203 JA |
226 | int res; |
227 | ||
228 | sock_confirm_neigh(skb, neigh); | |
5c9f7c1d DA |
229 | /* if crossing protocols, can not use the cached header */ |
230 | res = neigh_output(neigh, skb, is_v6gw); | |
a263b309 | 231 | rcu_read_unlock_bh(); |
f2c31e32 ED |
232 | return res; |
233 | } | |
a263b309 | 234 | rcu_read_unlock_bh(); |
05e3aa09 | 235 | |
e87cc472 JP |
236 | net_dbg_ratelimited("%s: No header cache and no neighbour!\n", |
237 | __func__); | |
1da177e4 LT |
238 | kfree_skb(skb); |
239 | return -EINVAL; | |
240 | } | |
241 | ||
694869b3 EB |
242 | static int ip_finish_output_gso(struct net *net, struct sock *sk, |
243 | struct sk_buff *skb, unsigned int mtu) | |
c7ba65d7 | 244 | { |
88bebdf5 | 245 | struct sk_buff *segs, *nskb; |
c7ba65d7 | 246 | netdev_features_t features; |
c7ba65d7 FW |
247 | int ret = 0; |
248 | ||
9ee6c5dc | 249 | /* common case: seglen is <= mtu |
359ebda2 | 250 | */ |
779b7931 | 251 | if (skb_gso_validate_network_len(skb, mtu)) |
694869b3 | 252 | return ip_finish_output2(net, sk, skb); |
c7ba65d7 | 253 | |
0ace81ec | 254 | /* Slowpath - GSO segment length exceeds the egress MTU. |
c7ba65d7 | 255 | * |
0ace81ec LR |
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 | |
a66e04ce | 265 | * insufficient MTU. |
c7ba65d7 FW |
266 | */ |
267 | features = netif_skb_features(skb); | |
a08e7fd9 | 268 | BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET); |
c7ba65d7 | 269 | segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); |
330966e5 | 270 | if (IS_ERR_OR_NULL(segs)) { |
c7ba65d7 FW |
271 | kfree_skb(skb); |
272 | return -ENOMEM; | |
273 | } | |
274 | ||
275 | consume_skb(skb); | |
276 | ||
88bebdf5 | 277 | skb_list_walk_safe(segs, segs, nskb) { |
c7ba65d7 FW |
278 | int err; |
279 | ||
a8305bff | 280 | skb_mark_not_on_list(segs); |
694869b3 | 281 | err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); |
c7ba65d7 FW |
282 | |
283 | if (err && ret == 0) | |
284 | ret = err; | |
88bebdf5 | 285 | } |
c7ba65d7 FW |
286 | |
287 | return ret; | |
288 | } | |
289 | ||
956fe219 | 290 | static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 291 | { |
c5501eb3 FW |
292 | unsigned int mtu; |
293 | ||
5c901daa PM |
294 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) |
295 | /* Policy lookup after SNAT yielded a new policy */ | |
00db4124 | 296 | if (skb_dst(skb)->xfrm) { |
48d5cad8 | 297 | IPCB(skb)->flags |= IPSKB_REROUTED; |
13206b6b | 298 | return dst_output(net, sk, skb); |
48d5cad8 | 299 | } |
5c901daa | 300 | #endif |
fedbb6b4 | 301 | mtu = ip_skb_dst_mtu(sk, skb); |
c7ba65d7 | 302 | if (skb_is_gso(skb)) |
694869b3 | 303 | return ip_finish_output_gso(net, sk, skb, mtu); |
c7ba65d7 | 304 | |
bb4cc1a1 | 305 | if (skb->len > mtu || IPCB(skb)->frag_max_size) |
694869b3 | 306 | return ip_fragment(net, sk, skb, mtu, ip_finish_output2); |
c7ba65d7 | 307 | |
694869b3 | 308 | return ip_finish_output2(net, sk, skb); |
1da177e4 LT |
309 | } |
310 | ||
956fe219 | 311 | static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
312 | { | |
313 | int ret; | |
314 | ||
315 | ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); | |
316 | switch (ret) { | |
317 | case NET_XMIT_SUCCESS: | |
318 | return __ip_finish_output(net, sk, skb); | |
319 | case NET_XMIT_CN: | |
320 | return __ip_finish_output(net, sk, skb) ? : ret; | |
321 | default: | |
322 | kfree_skb(skb); | |
323 | return ret; | |
324 | } | |
325 | } | |
326 | ||
33b48679 DM |
327 | static int ip_mc_finish_output(struct net *net, struct sock *sk, |
328 | struct sk_buff *skb) | |
329 | { | |
5b18f128 | 330 | struct rtable *new_rt; |
d96ff269 DM |
331 | bool do_cn = false; |
332 | int ret, err; | |
33b48679 DM |
333 | |
334 | ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); | |
956fe219 | 335 | switch (ret) { |
956fe219 | 336 | case NET_XMIT_CN: |
d96ff269 | 337 | do_cn = true; |
a8eceea8 | 338 | fallthrough; |
d96ff269 DM |
339 | case NET_XMIT_SUCCESS: |
340 | break; | |
956fe219 | 341 | default: |
33b48679 DM |
342 | kfree_skb(skb); |
343 | return ret; | |
344 | } | |
345 | ||
5b18f128 SS |
346 | /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting |
347 | * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten, | |
348 | * see ipv4_pktinfo_prepare(). | |
349 | */ | |
350 | new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb)); | |
351 | if (new_rt) { | |
352 | new_rt->rt_iif = 0; | |
353 | skb_dst_drop(skb); | |
354 | skb_dst_set(skb, &new_rt->dst); | |
355 | } | |
356 | ||
d96ff269 DM |
357 | err = dev_loopback_xmit(net, sk, skb); |
358 | return (do_cn && err) ? ret : err; | |
33b48679 DM |
359 | } |
360 | ||
ede2059d | 361 | int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 362 | { |
511c3f92 | 363 | struct rtable *rt = skb_rtable(skb); |
d8d1f30b | 364 | struct net_device *dev = rt->dst.dev; |
1da177e4 LT |
365 | |
366 | /* | |
367 | * If the indicated interface is up and running, send the packet. | |
368 | */ | |
88f5cc24 | 369 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
1da177e4 LT |
370 | |
371 | skb->dev = dev; | |
372 | skb->protocol = htons(ETH_P_IP); | |
373 | ||
374 | /* | |
375 | * Multicasts are looped back for other local users | |
376 | */ | |
377 | ||
378 | if (rt->rt_flags&RTCF_MULTICAST) { | |
7ad6848c | 379 | if (sk_mc_loop(sk) |
1da177e4 LT |
380 | #ifdef CONFIG_IP_MROUTE |
381 | /* Small optimization: do not loopback not local frames, | |
382 | which returned after forwarding; they will be dropped | |
383 | by ip_mr_input in any case. | |
384 | Note, that local frames are looped back to be delivered | |
385 | to local recipients. | |
386 | ||
387 | This check is duplicated in ip_mr_input at the moment. | |
388 | */ | |
9d4fb27d JP |
389 | && |
390 | ((rt->rt_flags & RTCF_LOCAL) || | |
391 | !(IPCB(skb)->flags & IPSKB_FORWARDED)) | |
1da177e4 | 392 | #endif |
9d4fb27d | 393 | ) { |
1da177e4 LT |
394 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); |
395 | if (newskb) | |
9bbc768a | 396 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
29a26a56 | 397 | net, sk, newskb, NULL, newskb->dev, |
33b48679 | 398 | ip_mc_finish_output); |
1da177e4 LT |
399 | } |
400 | ||
401 | /* Multicasts with ttl 0 must not go beyond the host */ | |
402 | ||
eddc9ec5 | 403 | if (ip_hdr(skb)->ttl == 0) { |
1da177e4 LT |
404 | kfree_skb(skb); |
405 | return 0; | |
406 | } | |
407 | } | |
408 | ||
409 | if (rt->rt_flags&RTCF_BROADCAST) { | |
410 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
411 | if (newskb) | |
29a26a56 EB |
412 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
413 | net, sk, newskb, NULL, newskb->dev, | |
33b48679 | 414 | ip_mc_finish_output); |
1da177e4 LT |
415 | } |
416 | ||
29a26a56 EB |
417 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
418 | net, sk, skb, NULL, skb->dev, | |
419 | ip_finish_output, | |
48d5cad8 | 420 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
421 | } |
422 | ||
ede2059d | 423 | int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 424 | { |
28f8bfd1 | 425 | struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev; |
1bd9bef6 | 426 | |
88f5cc24 | 427 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
1da177e4 | 428 | |
1bd9bef6 PM |
429 | skb->dev = dev; |
430 | skb->protocol = htons(ETH_P_IP); | |
431 | ||
29a26a56 | 432 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
28f8bfd1 | 433 | net, sk, skb, indev, dev, |
e905a9ed | 434 | ip_finish_output, |
48d5cad8 | 435 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 | 436 | } |
6585d7dc | 437 | EXPORT_SYMBOL(ip_output); |
1da177e4 | 438 | |
84f9307c ED |
439 | /* |
440 | * copy saddr and daddr, possibly using 64bit load/stores | |
441 | * Equivalent to : | |
442 | * iph->saddr = fl4->saddr; | |
443 | * iph->daddr = fl4->daddr; | |
444 | */ | |
445 | static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) | |
446 | { | |
447 | BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != | |
448 | offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); | |
449 | memcpy(&iph->saddr, &fl4->saddr, | |
450 | sizeof(fl4->saddr) + sizeof(fl4->daddr)); | |
451 | } | |
452 | ||
b0270e91 | 453 | /* Note: skb->sk can be different from sk, in case of tunnels */ |
69b9e1e0 XL |
454 | int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl, |
455 | __u8 tos) | |
1da177e4 | 456 | { |
1da177e4 | 457 | struct inet_sock *inet = inet_sk(sk); |
77589ce0 | 458 | struct net *net = sock_net(sk); |
f6d8bd05 | 459 | struct ip_options_rcu *inet_opt; |
b57ae01a | 460 | struct flowi4 *fl4; |
1da177e4 LT |
461 | struct rtable *rt; |
462 | struct iphdr *iph; | |
ab6e3feb | 463 | int res; |
1da177e4 LT |
464 | |
465 | /* Skip all of this if the packet is already routed, | |
466 | * f.e. by something like SCTP. | |
467 | */ | |
ab6e3feb | 468 | rcu_read_lock(); |
f6d8bd05 | 469 | inet_opt = rcu_dereference(inet->inet_opt); |
ea4fc0d6 | 470 | fl4 = &fl->u.ip4; |
511c3f92 | 471 | rt = skb_rtable(skb); |
00db4124 | 472 | if (rt) |
1da177e4 LT |
473 | goto packet_routed; |
474 | ||
475 | /* Make sure we can route this packet. */ | |
476 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
51456b29 | 477 | if (!rt) { |
3ca3c68e | 478 | __be32 daddr; |
1da177e4 LT |
479 | |
480 | /* Use correct destination address if we have options. */ | |
c720c7e8 | 481 | daddr = inet->inet_daddr; |
f6d8bd05 ED |
482 | if (inet_opt && inet_opt->opt.srr) |
483 | daddr = inet_opt->opt.faddr; | |
1da177e4 | 484 | |
78fbfd8a DM |
485 | /* If this fails, retransmit mechanism of transport layer will |
486 | * keep trying until route appears or the connection times | |
487 | * itself out. | |
488 | */ | |
77589ce0 | 489 | rt = ip_route_output_ports(net, fl4, sk, |
78fbfd8a DM |
490 | daddr, inet->inet_saddr, |
491 | inet->inet_dport, | |
492 | inet->inet_sport, | |
493 | sk->sk_protocol, | |
69b9e1e0 | 494 | RT_CONN_FLAGS_TOS(sk, tos), |
78fbfd8a DM |
495 | sk->sk_bound_dev_if); |
496 | if (IS_ERR(rt)) | |
497 | goto no_route; | |
d8d1f30b | 498 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 499 | } |
d8d1f30b | 500 | skb_dst_set_noref(skb, &rt->dst); |
1da177e4 LT |
501 | |
502 | packet_routed: | |
77d5bc7e | 503 | if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) |
1da177e4 LT |
504 | goto no_route; |
505 | ||
506 | /* OK, we know where to send it, allocate and build IP header. */ | |
f6d8bd05 | 507 | skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); |
8856dfa3 | 508 | skb_reset_network_header(skb); |
eddc9ec5 | 509 | iph = ip_hdr(skb); |
69b9e1e0 | 510 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff)); |
60ff7467 | 511 | if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) |
1da177e4 LT |
512 | iph->frag_off = htons(IP_DF); |
513 | else | |
514 | iph->frag_off = 0; | |
d8d1f30b | 515 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 516 | iph->protocol = sk->sk_protocol; |
84f9307c ED |
517 | ip_copy_addrs(iph, fl4); |
518 | ||
1da177e4 LT |
519 | /* Transport layer set skb->h.foo itself. */ |
520 | ||
f6d8bd05 ED |
521 | if (inet_opt && inet_opt->opt.optlen) { |
522 | iph->ihl += inet_opt->opt.optlen >> 2; | |
523 | ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); | |
1da177e4 LT |
524 | } |
525 | ||
77589ce0 | 526 | ip_select_ident_segs(net, skb, sk, |
b6a7719a | 527 | skb_shinfo(skb)->gso_segs ?: 1); |
1da177e4 | 528 | |
b0270e91 | 529 | /* TODO : should we use skb->sk here instead of sk ? */ |
1da177e4 | 530 | skb->priority = sk->sk_priority; |
4a19ec58 | 531 | skb->mark = sk->sk_mark; |
1da177e4 | 532 | |
33224b16 | 533 | res = ip_local_out(net, sk, skb); |
ab6e3feb ED |
534 | rcu_read_unlock(); |
535 | return res; | |
1da177e4 LT |
536 | |
537 | no_route: | |
ab6e3feb | 538 | rcu_read_unlock(); |
77589ce0 | 539 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
1da177e4 LT |
540 | kfree_skb(skb); |
541 | return -EHOSTUNREACH; | |
542 | } | |
69b9e1e0 | 543 | EXPORT_SYMBOL(__ip_queue_xmit); |
1da177e4 | 544 | |
05e22e83 ED |
545 | int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) |
546 | { | |
547 | return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos); | |
548 | } | |
549 | EXPORT_SYMBOL(ip_queue_xmit); | |
550 | ||
1da177e4 LT |
551 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) |
552 | { | |
553 | to->pkt_type = from->pkt_type; | |
554 | to->priority = from->priority; | |
555 | to->protocol = from->protocol; | |
d2f0c961 | 556 | to->skb_iif = from->skb_iif; |
adf30907 | 557 | skb_dst_drop(to); |
fe76cda3 | 558 | skb_dst_copy(to, from); |
1da177e4 | 559 | to->dev = from->dev; |
82e91ffe | 560 | to->mark = from->mark; |
1da177e4 | 561 | |
3dd1c9a1 PA |
562 | skb_copy_hash(to, from); |
563 | ||
1da177e4 LT |
564 | #ifdef CONFIG_NET_SCHED |
565 | to->tc_index = from->tc_index; | |
566 | #endif | |
e7ac05f3 | 567 | nf_copy(to, from); |
df5042f4 | 568 | skb_ext_copy(to, from); |
6ca40d4e | 569 | #if IS_ENABLED(CONFIG_IP_VS) |
c98d80ed | 570 | to->ipvs_property = from->ipvs_property; |
1da177e4 | 571 | #endif |
984bc16c | 572 | skb_copy_secmark(to, from); |
1da177e4 LT |
573 | } |
574 | ||
694869b3 | 575 | static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, |
c5501eb3 | 576 | unsigned int mtu, |
694869b3 | 577 | int (*output)(struct net *, struct sock *, struct sk_buff *)) |
49d16b23 AZ |
578 | { |
579 | struct iphdr *iph = ip_hdr(skb); | |
49d16b23 | 580 | |
d6b915e2 | 581 | if ((iph->frag_off & htons(IP_DF)) == 0) |
694869b3 | 582 | return ip_do_fragment(net, sk, skb, output); |
d6b915e2 FW |
583 | |
584 | if (unlikely(!skb->ignore_df || | |
49d16b23 AZ |
585 | (IPCB(skb)->frag_max_size && |
586 | IPCB(skb)->frag_max_size > mtu))) { | |
9479b0af | 587 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
49d16b23 AZ |
588 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, |
589 | htonl(mtu)); | |
590 | kfree_skb(skb); | |
591 | return -EMSGSIZE; | |
592 | } | |
593 | ||
694869b3 | 594 | return ip_do_fragment(net, sk, skb, output); |
49d16b23 AZ |
595 | } |
596 | ||
c8b17be0 PNA |
597 | void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph, |
598 | unsigned int hlen, struct ip_fraglist_iter *iter) | |
599 | { | |
600 | unsigned int first_len = skb_pagelen(skb); | |
601 | ||
b7034146 | 602 | iter->frag = skb_shinfo(skb)->frag_list; |
c8b17be0 PNA |
603 | skb_frag_list_init(skb); |
604 | ||
605 | iter->offset = 0; | |
606 | iter->iph = iph; | |
607 | iter->hlen = hlen; | |
608 | ||
609 | skb->data_len = first_len - skb_headlen(skb); | |
610 | skb->len = first_len; | |
611 | iph->tot_len = htons(first_len); | |
612 | iph->frag_off = htons(IP_MF); | |
613 | ip_send_check(iph); | |
614 | } | |
615 | EXPORT_SYMBOL(ip_fraglist_init); | |
616 | ||
19c3401a PNA |
617 | static void ip_fraglist_ipcb_prepare(struct sk_buff *skb, |
618 | struct ip_fraglist_iter *iter) | |
619 | { | |
620 | struct sk_buff *to = iter->frag; | |
621 | ||
622 | /* Copy the flags to each fragment. */ | |
623 | IPCB(to)->flags = IPCB(skb)->flags; | |
624 | ||
625 | if (iter->offset == 0) | |
626 | ip_options_fragment(to); | |
627 | } | |
628 | ||
c8b17be0 PNA |
629 | void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter) |
630 | { | |
631 | unsigned int hlen = iter->hlen; | |
632 | struct iphdr *iph = iter->iph; | |
633 | struct sk_buff *frag; | |
634 | ||
635 | frag = iter->frag; | |
636 | frag->ip_summed = CHECKSUM_NONE; | |
637 | skb_reset_transport_header(frag); | |
638 | __skb_push(frag, hlen); | |
639 | skb_reset_network_header(frag); | |
640 | memcpy(skb_network_header(frag), iph, hlen); | |
641 | iter->iph = ip_hdr(frag); | |
642 | iph = iter->iph; | |
643 | iph->tot_len = htons(frag->len); | |
644 | ip_copy_metadata(frag, skb); | |
c8b17be0 PNA |
645 | iter->offset += skb->len - hlen; |
646 | iph->frag_off = htons(iter->offset >> 3); | |
647 | if (frag->next) | |
648 | iph->frag_off |= htons(IP_MF); | |
649 | /* Ready, complete checksum */ | |
650 | ip_send_check(iph); | |
651 | } | |
652 | EXPORT_SYMBOL(ip_fraglist_prepare); | |
653 | ||
065ff79f | 654 | void ip_frag_init(struct sk_buff *skb, unsigned int hlen, |
e7a409c3 | 655 | unsigned int ll_rs, unsigned int mtu, bool DF, |
065ff79f PNA |
656 | struct ip_frag_state *state) |
657 | { | |
658 | struct iphdr *iph = ip_hdr(skb); | |
659 | ||
e7a409c3 | 660 | state->DF = DF; |
065ff79f PNA |
661 | state->hlen = hlen; |
662 | state->ll_rs = ll_rs; | |
663 | state->mtu = mtu; | |
664 | ||
665 | state->left = skb->len - hlen; /* Space per frame */ | |
666 | state->ptr = hlen; /* Where to start from */ | |
667 | ||
668 | state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
669 | state->not_last_frag = iph->frag_off & htons(IP_MF); | |
670 | } | |
671 | EXPORT_SYMBOL(ip_frag_init); | |
672 | ||
19c3401a PNA |
673 | static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to, |
674 | bool first_frag, struct ip_frag_state *state) | |
675 | { | |
676 | /* Copy the flags to each fragment. */ | |
677 | IPCB(to)->flags = IPCB(from)->flags; | |
678 | ||
19c3401a PNA |
679 | /* ANK: dirty, but effective trick. Upgrade options only if |
680 | * the segment to be fragmented was THE FIRST (otherwise, | |
681 | * options are already fixed) and make it ONCE | |
682 | * on the initial skb, so that all the following fragments | |
683 | * will inherit fixed options. | |
684 | */ | |
685 | if (first_frag) | |
686 | ip_options_fragment(from); | |
687 | } | |
688 | ||
065ff79f PNA |
689 | struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state) |
690 | { | |
691 | unsigned int len = state->left; | |
692 | struct sk_buff *skb2; | |
693 | struct iphdr *iph; | |
694 | ||
695 | len = state->left; | |
696 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
697 | if (len > state->mtu) | |
698 | len = state->mtu; | |
699 | /* IF: we are not sending up to and including the packet end | |
700 | then align the next start on an eight byte boundary */ | |
701 | if (len < state->left) { | |
702 | len &= ~7; | |
703 | } | |
704 | ||
705 | /* Allocate buffer */ | |
706 | skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC); | |
707 | if (!skb2) | |
708 | return ERR_PTR(-ENOMEM); | |
709 | ||
710 | /* | |
711 | * Set up data on packet | |
712 | */ | |
713 | ||
714 | ip_copy_metadata(skb2, skb); | |
715 | skb_reserve(skb2, state->ll_rs); | |
716 | skb_put(skb2, len + state->hlen); | |
717 | skb_reset_network_header(skb2); | |
718 | skb2->transport_header = skb2->network_header + state->hlen; | |
719 | ||
720 | /* | |
721 | * Charge the memory for the fragment to any owner | |
722 | * it might possess | |
723 | */ | |
724 | ||
725 | if (skb->sk) | |
726 | skb_set_owner_w(skb2, skb->sk); | |
727 | ||
728 | /* | |
729 | * Copy the packet header into the new buffer. | |
730 | */ | |
731 | ||
732 | skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen); | |
733 | ||
734 | /* | |
735 | * Copy a block of the IP datagram. | |
736 | */ | |
737 | if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len)) | |
738 | BUG(); | |
739 | state->left -= len; | |
740 | ||
741 | /* | |
742 | * Fill in the new header fields. | |
743 | */ | |
744 | iph = ip_hdr(skb2); | |
745 | iph->frag_off = htons((state->offset >> 3)); | |
e7a409c3 ED |
746 | if (state->DF) |
747 | iph->frag_off |= htons(IP_DF); | |
065ff79f | 748 | |
065ff79f PNA |
749 | /* |
750 | * Added AC : If we are fragmenting a fragment that's not the | |
751 | * last fragment then keep MF on each bit | |
752 | */ | |
753 | if (state->left > 0 || state->not_last_frag) | |
754 | iph->frag_off |= htons(IP_MF); | |
755 | state->ptr += len; | |
756 | state->offset += len; | |
757 | ||
758 | iph->tot_len = htons(len + state->hlen); | |
759 | ||
760 | ip_send_check(iph); | |
761 | ||
762 | return skb2; | |
763 | } | |
764 | EXPORT_SYMBOL(ip_frag_next); | |
765 | ||
1da177e4 LT |
766 | /* |
767 | * This IP datagram is too large to be sent in one piece. Break it up into | |
768 | * smaller pieces (each of size equal to IP header plus | |
769 | * a block of the data of the original IP data part) that will yet fit in a | |
770 | * single device frame, and queue such a frame for sending. | |
771 | */ | |
772 | ||
694869b3 EB |
773 | int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, |
774 | int (*output)(struct net *, struct sock *, struct sk_buff *)) | |
1da177e4 LT |
775 | { |
776 | struct iphdr *iph; | |
1da177e4 | 777 | struct sk_buff *skb2; |
511c3f92 | 778 | struct rtable *rt = skb_rtable(skb); |
065ff79f | 779 | unsigned int mtu, hlen, ll_rs; |
c8b17be0 | 780 | struct ip_fraglist_iter iter; |
9669fffc | 781 | ktime_t tstamp = skb->tstamp; |
065ff79f | 782 | struct ip_frag_state state; |
1da177e4 LT |
783 | int err = 0; |
784 | ||
dbd3393c HFS |
785 | /* for offloaded checksums cleanup checksum before fragmentation */ |
786 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
787 | (err = skb_checksum_help(skb))) | |
788 | goto fail; | |
789 | ||
1da177e4 LT |
790 | /* |
791 | * Point into the IP datagram header. | |
792 | */ | |
793 | ||
eddc9ec5 | 794 | iph = ip_hdr(skb); |
1da177e4 | 795 | |
fedbb6b4 | 796 | mtu = ip_skb_dst_mtu(sk, skb); |
d6b915e2 FW |
797 | if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) |
798 | mtu = IPCB(skb)->frag_max_size; | |
1da177e4 LT |
799 | |
800 | /* | |
801 | * Setup starting values. | |
802 | */ | |
803 | ||
804 | hlen = iph->ihl * 4; | |
f87c10a8 | 805 | mtu = mtu - hlen; /* Size of data space */ |
89cee8b1 | 806 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
254d900b | 807 | ll_rs = LL_RESERVED_SPACE(rt->dst.dev); |
1da177e4 LT |
808 | |
809 | /* When frag_list is given, use it. First, check its validity: | |
810 | * some transformers could create wrong frag_list or break existing | |
811 | * one, it is not prohibited. In this case fall back to copying. | |
812 | * | |
813 | * LATER: this step can be merged to real generation of fragments, | |
814 | * we can switch to copy when see the first bad fragment. | |
815 | */ | |
21dc3301 | 816 | if (skb_has_frag_list(skb)) { |
3d13008e | 817 | struct sk_buff *frag, *frag2; |
c72d8cda | 818 | unsigned int first_len = skb_pagelen(skb); |
1da177e4 LT |
819 | |
820 | if (first_len - hlen > mtu || | |
821 | ((first_len - hlen) & 7) || | |
56f8a75c | 822 | ip_is_fragment(iph) || |
254d900b VA |
823 | skb_cloned(skb) || |
824 | skb_headroom(skb) < ll_rs) | |
1da177e4 LT |
825 | goto slow_path; |
826 | ||
d7fcf1a5 | 827 | skb_walk_frags(skb, frag) { |
1da177e4 LT |
828 | /* Correct geometry. */ |
829 | if (frag->len > mtu || | |
830 | ((frag->len & 7) && frag->next) || | |
254d900b | 831 | skb_headroom(frag) < hlen + ll_rs) |
3d13008e | 832 | goto slow_path_clean; |
1da177e4 LT |
833 | |
834 | /* Partially cloned skb? */ | |
835 | if (skb_shared(frag)) | |
3d13008e | 836 | goto slow_path_clean; |
2fdba6b0 HX |
837 | |
838 | BUG_ON(frag->sk); | |
839 | if (skb->sk) { | |
2fdba6b0 HX |
840 | frag->sk = skb->sk; |
841 | frag->destructor = sock_wfree; | |
2fdba6b0 | 842 | } |
3d13008e | 843 | skb->truesize -= frag->truesize; |
1da177e4 LT |
844 | } |
845 | ||
846 | /* Everything is OK. Generate! */ | |
c8b17be0 | 847 | ip_fraglist_init(skb, iph, hlen, &iter); |
1da177e4 LT |
848 | |
849 | for (;;) { | |
850 | /* Prepare header of the next frame, | |
851 | * before previous one went down. */ | |
19c3401a PNA |
852 | if (iter.frag) { |
853 | ip_fraglist_ipcb_prepare(skb, &iter); | |
c8b17be0 | 854 | ip_fraglist_prepare(skb, &iter); |
19c3401a | 855 | } |
1da177e4 | 856 | |
9669fffc | 857 | skb->tstamp = tstamp; |
694869b3 | 858 | err = output(net, sk, skb); |
1da177e4 | 859 | |
dafee490 | 860 | if (!err) |
26a949db | 861 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); |
c8b17be0 | 862 | if (err || !iter.frag) |
1da177e4 LT |
863 | break; |
864 | ||
c8b17be0 | 865 | skb = ip_fraglist_next(&iter); |
1da177e4 LT |
866 | } |
867 | ||
868 | if (err == 0) { | |
26a949db | 869 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
870 | return 0; |
871 | } | |
872 | ||
b7034146 | 873 | kfree_skb_list(iter.frag); |
942f146a | 874 | |
26a949db | 875 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 876 | return err; |
3d13008e ED |
877 | |
878 | slow_path_clean: | |
879 | skb_walk_frags(skb, frag2) { | |
880 | if (frag2 == frag) | |
881 | break; | |
882 | frag2->sk = NULL; | |
883 | frag2->destructor = NULL; | |
884 | skb->truesize += frag2->truesize; | |
885 | } | |
1da177e4 LT |
886 | } |
887 | ||
888 | slow_path: | |
1da177e4 LT |
889 | /* |
890 | * Fragment the datagram. | |
891 | */ | |
892 | ||
e7a409c3 ED |
893 | ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU, |
894 | &state); | |
1da177e4 LT |
895 | |
896 | /* | |
897 | * Keep copying data until we run out. | |
898 | */ | |
899 | ||
065ff79f | 900 | while (state.left > 0) { |
19c3401a PNA |
901 | bool first_frag = (state.offset == 0); |
902 | ||
065ff79f PNA |
903 | skb2 = ip_frag_next(skb, &state); |
904 | if (IS_ERR(skb2)) { | |
905 | err = PTR_ERR(skb2); | |
1da177e4 LT |
906 | goto fail; |
907 | } | |
19c3401a | 908 | ip_frag_ipcb(skb, skb2, first_frag, &state); |
1da177e4 | 909 | |
1da177e4 LT |
910 | /* |
911 | * Put this fragment into the sending queue. | |
912 | */ | |
9669fffc | 913 | skb2->tstamp = tstamp; |
694869b3 | 914 | err = output(net, sk, skb2); |
1da177e4 LT |
915 | if (err) |
916 | goto fail; | |
dafee490 | 917 | |
26a949db | 918 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); |
1da177e4 | 919 | } |
5d0ba55b | 920 | consume_skb(skb); |
26a949db | 921 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
922 | return err; |
923 | ||
924 | fail: | |
e905a9ed | 925 | kfree_skb(skb); |
26a949db | 926 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
927 | return err; |
928 | } | |
49d16b23 | 929 | EXPORT_SYMBOL(ip_do_fragment); |
2e2f7aef | 930 | |
1da177e4 LT |
931 | int |
932 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
933 | { | |
f69e6d13 | 934 | struct msghdr *msg = from; |
1da177e4 | 935 | |
84fa7933 | 936 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
0b62fca2 | 937 | if (!copy_from_iter_full(to, len, &msg->msg_iter)) |
1da177e4 LT |
938 | return -EFAULT; |
939 | } else { | |
44bb9363 | 940 | __wsum csum = 0; |
0b62fca2 | 941 | if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter)) |
1da177e4 LT |
942 | return -EFAULT; |
943 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
944 | } | |
945 | return 0; | |
946 | } | |
4bc2f18b | 947 | EXPORT_SYMBOL(ip_generic_getfrag); |
1da177e4 | 948 | |
44bb9363 | 949 | static inline __wsum |
1da177e4 LT |
950 | csum_page(struct page *page, int offset, int copy) |
951 | { | |
952 | char *kaddr; | |
44bb9363 | 953 | __wsum csum; |
1da177e4 LT |
954 | kaddr = kmap(page); |
955 | csum = csum_partial(kaddr + offset, copy, 0); | |
956 | kunmap(page); | |
957 | return csum; | |
958 | } | |
959 | ||
f5fca608 DM |
960 | static int __ip_append_data(struct sock *sk, |
961 | struct flowi4 *fl4, | |
962 | struct sk_buff_head *queue, | |
1470ddf7 | 963 | struct inet_cork *cork, |
5640f768 | 964 | struct page_frag *pfrag, |
1470ddf7 HX |
965 | int getfrag(void *from, char *to, int offset, |
966 | int len, int odd, struct sk_buff *skb), | |
967 | void *from, int length, int transhdrlen, | |
968 | unsigned int flags) | |
1da177e4 LT |
969 | { |
970 | struct inet_sock *inet = inet_sk(sk); | |
b5947e5d | 971 | struct ubuf_info *uarg = NULL; |
1da177e4 LT |
972 | struct sk_buff *skb; |
973 | ||
07df5294 | 974 | struct ip_options *opt = cork->opt; |
1da177e4 LT |
975 | int hh_len; |
976 | int exthdrlen; | |
977 | int mtu; | |
978 | int copy; | |
979 | int err; | |
980 | int offset = 0; | |
daba287b | 981 | unsigned int maxfraglen, fragheaderlen, maxnonfragsize; |
1da177e4 | 982 | int csummode = CHECKSUM_NONE; |
1470ddf7 | 983 | struct rtable *rt = (struct rtable *)cork->dst; |
694aba69 | 984 | unsigned int wmem_alloc_delta = 0; |
100f6d8e | 985 | bool paged, extra_uref = false; |
09c2d251 | 986 | u32 tskey = 0; |
1da177e4 | 987 | |
96d7303e SK |
988 | skb = skb_peek_tail(queue); |
989 | ||
990 | exthdrlen = !skb ? rt->dst.header_len : 0; | |
bec1f6f6 | 991 | mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; |
15e36f5b | 992 | paged = !!cork->gso_size; |
bec1f6f6 | 993 | |
09c2d251 WB |
994 | if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && |
995 | sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) | |
996 | tskey = sk->sk_tskey++; | |
1da177e4 | 997 | |
d8d1f30b | 998 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
1da177e4 LT |
999 | |
1000 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1001 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
cbc08a33 | 1002 | maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu; |
1da177e4 | 1003 | |
daba287b | 1004 | if (cork->length + length > maxnonfragsize - fragheaderlen) { |
f5fca608 | 1005 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
61e7f09d | 1006 | mtu - (opt ? opt->optlen : 0)); |
1da177e4 LT |
1007 | return -EMSGSIZE; |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * transhdrlen > 0 means that this is the first fragment and we wish | |
1012 | * it won't be fragmented in the future. | |
1013 | */ | |
1014 | if (transhdrlen && | |
1015 | length + fragheaderlen <= mtu && | |
c8cd0989 | 1016 | rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && |
bec1f6f6 | 1017 | (!(flags & MSG_MORE) || cork->gso_size) && |
cd027a54 | 1018 | (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM))) |
84fa7933 | 1019 | csummode = CHECKSUM_PARTIAL; |
1da177e4 | 1020 | |
b5947e5d | 1021 | if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) { |
8c793822 | 1022 | uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb)); |
b5947e5d WB |
1023 | if (!uarg) |
1024 | return -ENOBUFS; | |
522924b5 | 1025 | extra_uref = !skb_zcopy(skb); /* only ref on new uarg */ |
b5947e5d WB |
1026 | if (rt->dst.dev->features & NETIF_F_SG && |
1027 | csummode == CHECKSUM_PARTIAL) { | |
1028 | paged = true; | |
1029 | } else { | |
1030 | uarg->zerocopy = 0; | |
52900d22 | 1031 | skb_zcopy_set(skb, uarg, &extra_uref); |
b5947e5d WB |
1032 | } |
1033 | } | |
1034 | ||
1470ddf7 | 1035 | cork->length += length; |
1da177e4 LT |
1036 | |
1037 | /* So, what's going on in the loop below? | |
1038 | * | |
1039 | * We use calculated fragment length to generate chained skb, | |
1040 | * each of segments is IP fragment ready for sending to network after | |
1041 | * adding appropriate IP header. | |
1042 | */ | |
1043 | ||
26cde9f7 | 1044 | if (!skb) |
1da177e4 LT |
1045 | goto alloc_new_skb; |
1046 | ||
1047 | while (length > 0) { | |
1048 | /* Check if the remaining data fits into current packet. */ | |
1049 | copy = mtu - skb->len; | |
1050 | if (copy < length) | |
1051 | copy = maxfraglen - skb->len; | |
1052 | if (copy <= 0) { | |
1053 | char *data; | |
1054 | unsigned int datalen; | |
1055 | unsigned int fraglen; | |
1056 | unsigned int fraggap; | |
6d123b81 | 1057 | unsigned int alloclen, alloc_extra; |
aba36930 | 1058 | unsigned int pagedlen; |
1da177e4 LT |
1059 | struct sk_buff *skb_prev; |
1060 | alloc_new_skb: | |
1061 | skb_prev = skb; | |
1062 | if (skb_prev) | |
1063 | fraggap = skb_prev->len - maxfraglen; | |
1064 | else | |
1065 | fraggap = 0; | |
1066 | ||
1067 | /* | |
1068 | * If remaining data exceeds the mtu, | |
1069 | * we know we need more fragment(s). | |
1070 | */ | |
1071 | datalen = length + fraggap; | |
1072 | if (datalen > mtu - fragheaderlen) | |
1073 | datalen = maxfraglen - fragheaderlen; | |
1074 | fraglen = datalen + fragheaderlen; | |
aba36930 | 1075 | pagedlen = 0; |
1da177e4 | 1076 | |
6d123b81 JK |
1077 | alloc_extra = hh_len + 15; |
1078 | alloc_extra += exthdrlen; | |
1079 | ||
1080 | /* The last fragment gets additional space at tail. | |
1081 | * Note, with MSG_MORE we overallocate on fragments, | |
1082 | * because we have no idea what fragment will be | |
1083 | * the last. | |
1084 | */ | |
1085 | if (datalen == length + fraggap) | |
1086 | alloc_extra += rt->dst.trailer_len; | |
1087 | ||
e905a9ed | 1088 | if ((flags & MSG_MORE) && |
d8d1f30b | 1089 | !(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 | 1090 | alloclen = mtu; |
6d123b81 JK |
1091 | else if (!paged && |
1092 | (fraglen + alloc_extra < SKB_MAX_ALLOC || | |
1093 | !(rt->dst.dev->features & NETIF_F_SG))) | |
59104f06 | 1094 | alloclen = fraglen; |
15e36f5b WB |
1095 | else { |
1096 | alloclen = min_t(int, fraglen, MAX_HEADER); | |
1097 | pagedlen = fraglen - alloclen; | |
1098 | } | |
1da177e4 | 1099 | |
6d123b81 | 1100 | alloclen += alloc_extra; |
33f99dc7 | 1101 | |
1da177e4 | 1102 | if (transhdrlen) { |
6d123b81 | 1103 | skb = sock_alloc_send_skb(sk, alloclen, |
1da177e4 LT |
1104 | (flags & MSG_DONTWAIT), &err); |
1105 | } else { | |
1106 | skb = NULL; | |
694aba69 | 1107 | if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <= |
1da177e4 | 1108 | 2 * sk->sk_sndbuf) |
6d123b81 | 1109 | skb = alloc_skb(alloclen, |
694aba69 | 1110 | sk->sk_allocation); |
51456b29 | 1111 | if (unlikely(!skb)) |
1da177e4 LT |
1112 | err = -ENOBUFS; |
1113 | } | |
51456b29 | 1114 | if (!skb) |
1da177e4 LT |
1115 | goto error; |
1116 | ||
1117 | /* | |
1118 | * Fill in the control structures | |
1119 | */ | |
1120 | skb->ip_summed = csummode; | |
1121 | skb->csum = 0; | |
1122 | skb_reserve(skb, hh_len); | |
11878b40 | 1123 | |
1da177e4 LT |
1124 | /* |
1125 | * Find where to start putting bytes. | |
1126 | */ | |
15e36f5b | 1127 | data = skb_put(skb, fraglen + exthdrlen - pagedlen); |
c14d2450 | 1128 | skb_set_network_header(skb, exthdrlen); |
b0e380b1 ACM |
1129 | skb->transport_header = (skb->network_header + |
1130 | fragheaderlen); | |
353e5c9a | 1131 | data += fragheaderlen + exthdrlen; |
1da177e4 LT |
1132 | |
1133 | if (fraggap) { | |
1134 | skb->csum = skb_copy_and_csum_bits( | |
1135 | skb_prev, maxfraglen, | |
8d5930df | 1136 | data + transhdrlen, fraggap); |
1da177e4 LT |
1137 | skb_prev->csum = csum_sub(skb_prev->csum, |
1138 | skb->csum); | |
1139 | data += fraggap; | |
e9fa4f7b | 1140 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1141 | } |
1142 | ||
15e36f5b | 1143 | copy = datalen - transhdrlen - fraggap - pagedlen; |
1da177e4 LT |
1144 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { |
1145 | err = -EFAULT; | |
1146 | kfree_skb(skb); | |
1147 | goto error; | |
1148 | } | |
1149 | ||
1150 | offset += copy; | |
15e36f5b | 1151 | length -= copy + transhdrlen; |
1da177e4 LT |
1152 | transhdrlen = 0; |
1153 | exthdrlen = 0; | |
1154 | csummode = CHECKSUM_NONE; | |
1155 | ||
52900d22 WB |
1156 | /* only the initial fragment is time stamped */ |
1157 | skb_shinfo(skb)->tx_flags = cork->tx_flags; | |
1158 | cork->tx_flags = 0; | |
1159 | skb_shinfo(skb)->tskey = tskey; | |
1160 | tskey = 0; | |
1161 | skb_zcopy_set(skb, uarg, &extra_uref); | |
1162 | ||
0dec879f JA |
1163 | if ((flags & MSG_CONFIRM) && !skb_prev) |
1164 | skb_set_dst_pending_confirm(skb, 1); | |
1165 | ||
1da177e4 LT |
1166 | /* |
1167 | * Put the packet on the pending queue. | |
1168 | */ | |
694aba69 ED |
1169 | if (!skb->destructor) { |
1170 | skb->destructor = sock_wfree; | |
1171 | skb->sk = sk; | |
1172 | wmem_alloc_delta += skb->truesize; | |
1173 | } | |
1470ddf7 | 1174 | __skb_queue_tail(queue, skb); |
1da177e4 LT |
1175 | continue; |
1176 | } | |
1177 | ||
1178 | if (copy > length) | |
1179 | copy = length; | |
1180 | ||
113f99c3 WB |
1181 | if (!(rt->dst.dev->features&NETIF_F_SG) && |
1182 | skb_tailroom(skb) >= copy) { | |
1da177e4 LT |
1183 | unsigned int off; |
1184 | ||
1185 | off = skb->len; | |
e905a9ed | 1186 | if (getfrag(from, skb_put(skb, copy), |
1da177e4 LT |
1187 | offset, copy, off, skb) < 0) { |
1188 | __skb_trim(skb, off); | |
1189 | err = -EFAULT; | |
1190 | goto error; | |
1191 | } | |
b5947e5d | 1192 | } else if (!uarg || !uarg->zerocopy) { |
1da177e4 | 1193 | int i = skb_shinfo(skb)->nr_frags; |
1da177e4 | 1194 | |
5640f768 ED |
1195 | err = -ENOMEM; |
1196 | if (!sk_page_frag_refill(sk, pfrag)) | |
1da177e4 | 1197 | goto error; |
5640f768 ED |
1198 | |
1199 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
1200 | pfrag->offset)) { | |
1201 | err = -EMSGSIZE; | |
1202 | if (i == MAX_SKB_FRAGS) | |
1203 | goto error; | |
1204 | ||
1205 | __skb_fill_page_desc(skb, i, pfrag->page, | |
1206 | pfrag->offset, 0); | |
1207 | skb_shinfo(skb)->nr_frags = ++i; | |
1208 | get_page(pfrag->page); | |
1da177e4 | 1209 | } |
5640f768 ED |
1210 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1211 | if (getfrag(from, | |
1212 | page_address(pfrag->page) + pfrag->offset, | |
1213 | offset, copy, skb->len, skb) < 0) | |
1214 | goto error_efault; | |
1215 | ||
1216 | pfrag->offset += copy; | |
1217 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1da177e4 LT |
1218 | skb->len += copy; |
1219 | skb->data_len += copy; | |
f945fa7a | 1220 | skb->truesize += copy; |
694aba69 | 1221 | wmem_alloc_delta += copy; |
b5947e5d WB |
1222 | } else { |
1223 | err = skb_zerocopy_iter_dgram(skb, from, copy); | |
1224 | if (err < 0) | |
1225 | goto error; | |
1da177e4 LT |
1226 | } |
1227 | offset += copy; | |
1228 | length -= copy; | |
1229 | } | |
1230 | ||
9e8445a5 PA |
1231 | if (wmem_alloc_delta) |
1232 | refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1233 | return 0; |
1234 | ||
5640f768 ED |
1235 | error_efault: |
1236 | err = -EFAULT; | |
1da177e4 | 1237 | error: |
8e044917 | 1238 | net_zcopy_put_abort(uarg, extra_uref); |
1470ddf7 | 1239 | cork->length -= length; |
5e38e270 | 1240 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
694aba69 | 1241 | refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); |
e905a9ed | 1242 | return err; |
1da177e4 LT |
1243 | } |
1244 | ||
1470ddf7 HX |
1245 | static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, |
1246 | struct ipcm_cookie *ipc, struct rtable **rtp) | |
1247 | { | |
f6d8bd05 | 1248 | struct ip_options_rcu *opt; |
1470ddf7 HX |
1249 | struct rtable *rt; |
1250 | ||
9783ccd0 GF |
1251 | rt = *rtp; |
1252 | if (unlikely(!rt)) | |
1253 | return -EFAULT; | |
1254 | ||
1470ddf7 HX |
1255 | /* |
1256 | * setup for corking. | |
1257 | */ | |
1258 | opt = ipc->opt; | |
1259 | if (opt) { | |
51456b29 | 1260 | if (!cork->opt) { |
1470ddf7 HX |
1261 | cork->opt = kmalloc(sizeof(struct ip_options) + 40, |
1262 | sk->sk_allocation); | |
51456b29 | 1263 | if (unlikely(!cork->opt)) |
1470ddf7 HX |
1264 | return -ENOBUFS; |
1265 | } | |
f6d8bd05 | 1266 | memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); |
1470ddf7 HX |
1267 | cork->flags |= IPCORK_OPT; |
1268 | cork->addr = ipc->addr; | |
1269 | } | |
9783ccd0 | 1270 | |
482fc609 | 1271 | cork->fragsize = ip_sk_use_pmtu(sk) ? |
501a90c9 ED |
1272 | dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu); |
1273 | ||
1274 | if (!inetdev_valid_mtu(cork->fragsize)) | |
1275 | return -ENETUNREACH; | |
bec1f6f6 | 1276 | |
fbf47813 | 1277 | cork->gso_size = ipc->gso_size; |
501a90c9 | 1278 | |
1470ddf7 | 1279 | cork->dst = &rt->dst; |
501a90c9 ED |
1280 | /* We stole this route, caller should not release it. */ |
1281 | *rtp = NULL; | |
1282 | ||
1470ddf7 | 1283 | cork->length = 0; |
aa661581 FF |
1284 | cork->ttl = ipc->ttl; |
1285 | cork->tos = ipc->tos; | |
c6af0c22 | 1286 | cork->mark = ipc->sockc.mark; |
aa661581 | 1287 | cork->priority = ipc->priority; |
bc969a97 | 1288 | cork->transmit_time = ipc->sockc.transmit_time; |
678ca42d WB |
1289 | cork->tx_flags = 0; |
1290 | sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags); | |
1470ddf7 HX |
1291 | |
1292 | return 0; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
1297 | * from many pieces of data. Each pieces will be holded on the socket | |
1298 | * until ip_push_pending_frames() is called. Each piece can be a page | |
1299 | * or non-page data. | |
1300 | * | |
1301 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
1302 | * this interface potentially. | |
1303 | * | |
1304 | * LATER: length must be adjusted by pad at tail, when it is required. | |
1305 | */ | |
f5fca608 | 1306 | int ip_append_data(struct sock *sk, struct flowi4 *fl4, |
1470ddf7 HX |
1307 | int getfrag(void *from, char *to, int offset, int len, |
1308 | int odd, struct sk_buff *skb), | |
1309 | void *from, int length, int transhdrlen, | |
1310 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1311 | unsigned int flags) | |
1312 | { | |
1313 | struct inet_sock *inet = inet_sk(sk); | |
1314 | int err; | |
1315 | ||
1316 | if (flags&MSG_PROBE) | |
1317 | return 0; | |
1318 | ||
1319 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
bdc712b4 | 1320 | err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); |
1470ddf7 HX |
1321 | if (err) |
1322 | return err; | |
1323 | } else { | |
1324 | transhdrlen = 0; | |
1325 | } | |
1326 | ||
5640f768 ED |
1327 | return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, |
1328 | sk_page_frag(sk), getfrag, | |
1470ddf7 HX |
1329 | from, length, transhdrlen, flags); |
1330 | } | |
1331 | ||
f5fca608 | 1332 | ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, |
1da177e4 LT |
1333 | int offset, size_t size, int flags) |
1334 | { | |
1335 | struct inet_sock *inet = inet_sk(sk); | |
1336 | struct sk_buff *skb; | |
1337 | struct rtable *rt; | |
1338 | struct ip_options *opt = NULL; | |
bdc712b4 | 1339 | struct inet_cork *cork; |
1da177e4 LT |
1340 | int hh_len; |
1341 | int mtu; | |
1342 | int len; | |
1343 | int err; | |
daba287b | 1344 | unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; |
1da177e4 LT |
1345 | |
1346 | if (inet->hdrincl) | |
1347 | return -EPERM; | |
1348 | ||
1349 | if (flags&MSG_PROBE) | |
1350 | return 0; | |
1351 | ||
1352 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1353 | return -EINVAL; | |
1354 | ||
bdc712b4 DM |
1355 | cork = &inet->cork.base; |
1356 | rt = (struct rtable *)cork->dst; | |
1357 | if (cork->flags & IPCORK_OPT) | |
1358 | opt = cork->opt; | |
1da177e4 | 1359 | |
343d8c60 | 1360 | if (!(rt->dst.dev->features & NETIF_F_SG)) |
1da177e4 LT |
1361 | return -EOPNOTSUPP; |
1362 | ||
d8d1f30b | 1363 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
bec1f6f6 | 1364 | mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; |
1da177e4 LT |
1365 | |
1366 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1367 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
60ff7467 | 1368 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; |
1da177e4 | 1369 | |
daba287b | 1370 | if (cork->length + size > maxnonfragsize - fragheaderlen) { |
61e7f09d HFS |
1371 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
1372 | mtu - (opt ? opt->optlen : 0)); | |
1da177e4 LT |
1373 | return -EMSGSIZE; |
1374 | } | |
1375 | ||
51456b29 IM |
1376 | skb = skb_peek_tail(&sk->sk_write_queue); |
1377 | if (!skb) | |
1da177e4 LT |
1378 | return -EINVAL; |
1379 | ||
a8c4a252 | 1380 | cork->length += size; |
e89e9cf5 | 1381 | |
1da177e4 | 1382 | while (size > 0) { |
ab2fb7e3 WB |
1383 | /* Check if the remaining data fits into current packet. */ |
1384 | len = mtu - skb->len; | |
1385 | if (len < size) | |
1386 | len = maxfraglen - skb->len; | |
e89e9cf5 | 1387 | |
1da177e4 LT |
1388 | if (len <= 0) { |
1389 | struct sk_buff *skb_prev; | |
1da177e4 LT |
1390 | int alloclen; |
1391 | ||
1392 | skb_prev = skb; | |
0d0d2bba | 1393 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1394 | |
1395 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1396 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1397 | if (unlikely(!skb)) { | |
1398 | err = -ENOBUFS; | |
1399 | goto error; | |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * Fill in the control structures | |
1404 | */ | |
1405 | skb->ip_summed = CHECKSUM_NONE; | |
1406 | skb->csum = 0; | |
1407 | skb_reserve(skb, hh_len); | |
1408 | ||
1409 | /* | |
1410 | * Find where to start putting bytes. | |
1411 | */ | |
967b05f6 | 1412 | skb_put(skb, fragheaderlen + fraggap); |
2ca9e6f2 | 1413 | skb_reset_network_header(skb); |
b0e380b1 ACM |
1414 | skb->transport_header = (skb->network_header + |
1415 | fragheaderlen); | |
1da177e4 | 1416 | if (fraggap) { |
967b05f6 ACM |
1417 | skb->csum = skb_copy_and_csum_bits(skb_prev, |
1418 | maxfraglen, | |
9c70220b | 1419 | skb_transport_header(skb), |
8d5930df | 1420 | fraggap); |
1da177e4 LT |
1421 | skb_prev->csum = csum_sub(skb_prev->csum, |
1422 | skb->csum); | |
e9fa4f7b | 1423 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1424 | } |
1425 | ||
1426 | /* | |
1427 | * Put the packet on the pending queue. | |
1428 | */ | |
1429 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1430 | continue; | |
1431 | } | |
1432 | ||
1da177e4 LT |
1433 | if (len > size) |
1434 | len = size; | |
be12a1fe HFS |
1435 | |
1436 | if (skb_append_pagefrags(skb, page, offset, len)) { | |
1da177e4 LT |
1437 | err = -EMSGSIZE; |
1438 | goto error; | |
1439 | } | |
1440 | ||
1441 | if (skb->ip_summed == CHECKSUM_NONE) { | |
44bb9363 | 1442 | __wsum csum; |
1da177e4 LT |
1443 | csum = csum_page(page, offset, len); |
1444 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1445 | } | |
1446 | ||
1447 | skb->len += len; | |
1448 | skb->data_len += len; | |
1e34a11d | 1449 | skb->truesize += len; |
14afee4b | 1450 | refcount_add(len, &sk->sk_wmem_alloc); |
1da177e4 LT |
1451 | offset += len; |
1452 | size -= len; | |
1453 | } | |
1454 | return 0; | |
1455 | ||
1456 | error: | |
bdc712b4 | 1457 | cork->length -= size; |
5e38e270 | 1458 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1459 | return err; |
1460 | } | |
1461 | ||
1470ddf7 | 1462 | static void ip_cork_release(struct inet_cork *cork) |
429f08e9 | 1463 | { |
1470ddf7 HX |
1464 | cork->flags &= ~IPCORK_OPT; |
1465 | kfree(cork->opt); | |
1466 | cork->opt = NULL; | |
1467 | dst_release(cork->dst); | |
1468 | cork->dst = NULL; | |
429f08e9 PE |
1469 | } |
1470 | ||
1da177e4 LT |
1471 | /* |
1472 | * Combined all pending IP fragments on the socket as one IP datagram | |
1473 | * and push them out. | |
1474 | */ | |
1c32c5ad | 1475 | struct sk_buff *__ip_make_skb(struct sock *sk, |
77968b78 | 1476 | struct flowi4 *fl4, |
1c32c5ad HX |
1477 | struct sk_buff_head *queue, |
1478 | struct inet_cork *cork) | |
1da177e4 LT |
1479 | { |
1480 | struct sk_buff *skb, *tmp_skb; | |
1481 | struct sk_buff **tail_skb; | |
1482 | struct inet_sock *inet = inet_sk(sk); | |
0388b004 | 1483 | struct net *net = sock_net(sk); |
1da177e4 | 1484 | struct ip_options *opt = NULL; |
1470ddf7 | 1485 | struct rtable *rt = (struct rtable *)cork->dst; |
1da177e4 | 1486 | struct iphdr *iph; |
76ab608d | 1487 | __be16 df = 0; |
1da177e4 | 1488 | __u8 ttl; |
1da177e4 | 1489 | |
51456b29 IM |
1490 | skb = __skb_dequeue(queue); |
1491 | if (!skb) | |
1da177e4 LT |
1492 | goto out; |
1493 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1494 | ||
1495 | /* move skb->data to ip header from ext header */ | |
d56f90a7 | 1496 | if (skb->data < skb_network_header(skb)) |
bbe735e4 | 1497 | __skb_pull(skb, skb_network_offset(skb)); |
1470ddf7 | 1498 | while ((tmp_skb = __skb_dequeue(queue)) != NULL) { |
cfe1fc77 | 1499 | __skb_pull(tmp_skb, skb_network_header_len(skb)); |
1da177e4 LT |
1500 | *tail_skb = tmp_skb; |
1501 | tail_skb = &(tmp_skb->next); | |
1502 | skb->len += tmp_skb->len; | |
1503 | skb->data_len += tmp_skb->len; | |
1504 | skb->truesize += tmp_skb->truesize; | |
1da177e4 LT |
1505 | tmp_skb->destructor = NULL; |
1506 | tmp_skb->sk = NULL; | |
1507 | } | |
1508 | ||
1509 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1510 | * to fragment the frame generated here. No matter, what transforms | |
1511 | * how transforms change size of the packet, it will come out. | |
1512 | */ | |
60ff7467 | 1513 | skb->ignore_df = ip_sk_ignore_df(sk); |
1da177e4 LT |
1514 | |
1515 | /* DF bit is set when we want to see DF on outgoing frames. | |
60ff7467 | 1516 | * If ignore_df is set too, we still allow to fragment this frame |
1da177e4 | 1517 | * locally. */ |
482fc609 HFS |
1518 | if (inet->pmtudisc == IP_PMTUDISC_DO || |
1519 | inet->pmtudisc == IP_PMTUDISC_PROBE || | |
d8d1f30b CG |
1520 | (skb->len <= dst_mtu(&rt->dst) && |
1521 | ip_dont_fragment(sk, &rt->dst))) | |
1da177e4 LT |
1522 | df = htons(IP_DF); |
1523 | ||
1470ddf7 HX |
1524 | if (cork->flags & IPCORK_OPT) |
1525 | opt = cork->opt; | |
1da177e4 | 1526 | |
aa661581 FF |
1527 | if (cork->ttl != 0) |
1528 | ttl = cork->ttl; | |
1529 | else if (rt->rt_type == RTN_MULTICAST) | |
1da177e4 LT |
1530 | ttl = inet->mc_ttl; |
1531 | else | |
d8d1f30b | 1532 | ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 1533 | |
749154aa | 1534 | iph = ip_hdr(skb); |
1da177e4 LT |
1535 | iph->version = 4; |
1536 | iph->ihl = 5; | |
aa661581 | 1537 | iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; |
1da177e4 | 1538 | iph->frag_off = df; |
1da177e4 LT |
1539 | iph->ttl = ttl; |
1540 | iph->protocol = sk->sk_protocol; | |
84f9307c | 1541 | ip_copy_addrs(iph, fl4); |
b6a7719a | 1542 | ip_select_ident(net, skb, sk); |
1da177e4 | 1543 | |
22f728f8 | 1544 | if (opt) { |
5af68891 | 1545 | iph->ihl += opt->optlen >> 2; |
22f728f8 DM |
1546 | ip_options_build(skb, opt, cork->addr, rt, 0); |
1547 | } | |
1548 | ||
aa661581 | 1549 | skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; |
c6af0c22 | 1550 | skb->mark = cork->mark; |
bc969a97 | 1551 | skb->tstamp = cork->transmit_time; |
a21bba94 ED |
1552 | /* |
1553 | * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec | |
1554 | * on dst refcount | |
1555 | */ | |
1470ddf7 | 1556 | cork->dst = NULL; |
d8d1f30b | 1557 | skb_dst_set(skb, &rt->dst); |
1da177e4 | 1558 | |
96793b48 | 1559 | if (iph->protocol == IPPROTO_ICMP) |
0388b004 | 1560 | icmp_out_count(net, ((struct icmphdr *) |
96793b48 DS |
1561 | skb_transport_header(skb))->type); |
1562 | ||
1c32c5ad HX |
1563 | ip_cork_release(cork); |
1564 | out: | |
1565 | return skb; | |
1566 | } | |
1567 | ||
b5ec8eea | 1568 | int ip_send_skb(struct net *net, struct sk_buff *skb) |
1c32c5ad | 1569 | { |
1c32c5ad HX |
1570 | int err; |
1571 | ||
33224b16 | 1572 | err = ip_local_out(net, skb->sk, skb); |
1da177e4 LT |
1573 | if (err) { |
1574 | if (err > 0) | |
6ce9e7b5 | 1575 | err = net_xmit_errno(err); |
1da177e4 | 1576 | if (err) |
1c32c5ad | 1577 | IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1578 | } |
1579 | ||
1da177e4 | 1580 | return err; |
1da177e4 LT |
1581 | } |
1582 | ||
77968b78 | 1583 | int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) |
1470ddf7 | 1584 | { |
1c32c5ad HX |
1585 | struct sk_buff *skb; |
1586 | ||
77968b78 | 1587 | skb = ip_finish_skb(sk, fl4); |
1c32c5ad HX |
1588 | if (!skb) |
1589 | return 0; | |
1590 | ||
1591 | /* Netfilter gets whole the not fragmented skb. */ | |
b5ec8eea | 1592 | return ip_send_skb(sock_net(sk), skb); |
1470ddf7 HX |
1593 | } |
1594 | ||
1da177e4 LT |
1595 | /* |
1596 | * Throw away all pending data on the socket. | |
1597 | */ | |
1470ddf7 HX |
1598 | static void __ip_flush_pending_frames(struct sock *sk, |
1599 | struct sk_buff_head *queue, | |
1600 | struct inet_cork *cork) | |
1da177e4 | 1601 | { |
1da177e4 LT |
1602 | struct sk_buff *skb; |
1603 | ||
1470ddf7 | 1604 | while ((skb = __skb_dequeue_tail(queue)) != NULL) |
1da177e4 LT |
1605 | kfree_skb(skb); |
1606 | ||
1470ddf7 HX |
1607 | ip_cork_release(cork); |
1608 | } | |
1609 | ||
1610 | void ip_flush_pending_frames(struct sock *sk) | |
1611 | { | |
bdc712b4 | 1612 | __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); |
1da177e4 LT |
1613 | } |
1614 | ||
1c32c5ad | 1615 | struct sk_buff *ip_make_skb(struct sock *sk, |
77968b78 | 1616 | struct flowi4 *fl4, |
1c32c5ad HX |
1617 | int getfrag(void *from, char *to, int offset, |
1618 | int len, int odd, struct sk_buff *skb), | |
1619 | void *from, int length, int transhdrlen, | |
1620 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1cd7884d | 1621 | struct inet_cork *cork, unsigned int flags) |
1c32c5ad | 1622 | { |
1c32c5ad HX |
1623 | struct sk_buff_head queue; |
1624 | int err; | |
1625 | ||
1626 | if (flags & MSG_PROBE) | |
1627 | return NULL; | |
1628 | ||
1629 | __skb_queue_head_init(&queue); | |
1630 | ||
1cd7884d WB |
1631 | cork->flags = 0; |
1632 | cork->addr = 0; | |
1633 | cork->opt = NULL; | |
1634 | err = ip_setup_cork(sk, cork, ipc, rtp); | |
1c32c5ad HX |
1635 | if (err) |
1636 | return ERR_PTR(err); | |
1637 | ||
1cd7884d | 1638 | err = __ip_append_data(sk, fl4, &queue, cork, |
5640f768 | 1639 | ¤t->task_frag, getfrag, |
1c32c5ad HX |
1640 | from, length, transhdrlen, flags); |
1641 | if (err) { | |
1cd7884d | 1642 | __ip_flush_pending_frames(sk, &queue, cork); |
1c32c5ad HX |
1643 | return ERR_PTR(err); |
1644 | } | |
1645 | ||
1cd7884d | 1646 | return __ip_make_skb(sk, fl4, &queue, cork); |
1c32c5ad | 1647 | } |
1da177e4 LT |
1648 | |
1649 | /* | |
1650 | * Fetch data from kernel space and fill in checksum if needed. | |
1651 | */ | |
e905a9ed | 1652 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, |
1da177e4 LT |
1653 | int len, int odd, struct sk_buff *skb) |
1654 | { | |
5084205f | 1655 | __wsum csum; |
1da177e4 | 1656 | |
cc44c17b | 1657 | csum = csum_partial_copy_nocheck(dptr+offset, to, len); |
1da177e4 | 1658 | skb->csum = csum_block_add(skb->csum, csum, odd); |
e905a9ed | 1659 | return 0; |
1da177e4 LT |
1660 | } |
1661 | ||
e905a9ed | 1662 | /* |
1da177e4 | 1663 | * Generic function to send a packet as reply to another packet. |
be9f4a44 | 1664 | * Used to send some TCP resets/acks so far. |
1da177e4 | 1665 | */ |
bdbbb852 | 1666 | void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, |
24a2d43d ED |
1667 | const struct ip_options *sopt, |
1668 | __be32 daddr, __be32 saddr, | |
1669 | const struct ip_reply_arg *arg, | |
d6fb396c | 1670 | unsigned int len, u64 transmit_time) |
1da177e4 | 1671 | { |
f6d8bd05 | 1672 | struct ip_options_data replyopts; |
1da177e4 | 1673 | struct ipcm_cookie ipc; |
77968b78 | 1674 | struct flowi4 fl4; |
511c3f92 | 1675 | struct rtable *rt = skb_rtable(skb); |
bdbbb852 | 1676 | struct net *net = sock_net(sk); |
be9f4a44 | 1677 | struct sk_buff *nskb; |
4062090e | 1678 | int err; |
f7ba868b | 1679 | int oif; |
1da177e4 | 1680 | |
91ed1e66 | 1681 | if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt)) |
1da177e4 LT |
1682 | return; |
1683 | ||
35178206 | 1684 | ipcm_init(&ipc); |
0a5ebb80 | 1685 | ipc.addr = daddr; |
d6fb396c | 1686 | ipc.sockc.transmit_time = transmit_time; |
1da177e4 | 1687 | |
f6d8bd05 | 1688 | if (replyopts.opt.opt.optlen) { |
1da177e4 LT |
1689 | ipc.opt = &replyopts.opt; |
1690 | ||
f6d8bd05 ED |
1691 | if (replyopts.opt.opt.srr) |
1692 | daddr = replyopts.opt.opt.faddr; | |
1da177e4 LT |
1693 | } |
1694 | ||
f7ba868b | 1695 | oif = arg->bound_dev_if; |
9b6c14d5 DA |
1696 | if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) |
1697 | oif = skb->skb_iif; | |
f7ba868b DA |
1698 | |
1699 | flowi4_init_output(&fl4, oif, | |
00483690 | 1700 | IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark, |
66b13d99 | 1701 | RT_TOS(arg->tos), |
be9f4a44 | 1702 | RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, |
77968b78 | 1703 | ip_reply_arg_flowi_flags(arg), |
70e73416 | 1704 | daddr, saddr, |
e2d118a1 LC |
1705 | tcp_hdr(skb)->source, tcp_hdr(skb)->dest, |
1706 | arg->uid); | |
3df98d79 | 1707 | security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); |
be9f4a44 | 1708 | rt = ip_route_output_key(net, &fl4); |
77968b78 DM |
1709 | if (IS_ERR(rt)) |
1710 | return; | |
1da177e4 | 1711 | |
ba9e04a7 | 1712 | inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK; |
1da177e4 | 1713 | |
eddc9ec5 | 1714 | sk->sk_protocol = ip_hdr(skb)->protocol; |
f0e48dbf | 1715 | sk->sk_bound_dev_if = arg->bound_dev_if; |
be9f4a44 | 1716 | sk->sk_sndbuf = sysctl_wmem_default; |
0da7536f | 1717 | ipc.sockc.mark = fl4.flowi4_mark; |
4062090e VA |
1718 | err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, |
1719 | len, 0, &ipc, &rt, MSG_DONTWAIT); | |
1720 | if (unlikely(err)) { | |
1721 | ip_flush_pending_frames(sk); | |
1722 | goto out; | |
1723 | } | |
1724 | ||
be9f4a44 ED |
1725 | nskb = skb_peek(&sk->sk_write_queue); |
1726 | if (nskb) { | |
1da177e4 | 1727 | if (arg->csumoffset >= 0) |
be9f4a44 ED |
1728 | *((__sum16 *)skb_transport_header(nskb) + |
1729 | arg->csumoffset) = csum_fold(csum_add(nskb->csum, | |
9c70220b | 1730 | arg->csum)); |
be9f4a44 | 1731 | nskb->ip_summed = CHECKSUM_NONE; |
77968b78 | 1732 | ip_push_pending_frames(sk, &fl4); |
1da177e4 | 1733 | } |
4062090e | 1734 | out: |
1da177e4 LT |
1735 | ip_rt_put(rt); |
1736 | } | |
1737 | ||
1da177e4 LT |
1738 | void __init ip_init(void) |
1739 | { | |
1da177e4 LT |
1740 | ip_rt_init(); |
1741 | inet_initpeers(); | |
1742 | ||
72c1d3bd WC |
1743 | #if defined(CONFIG_IP_MULTICAST) |
1744 | igmp_mc_init(); | |
1da177e4 LT |
1745 | #endif |
1746 | } |