ip_output.c 38 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545
  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. #include <asm/uaccess.h>
  45. #include <linux/module.h>
  46. #include <linux/types.h>
  47. #include <linux/kernel.h>
  48. #include <linux/mm.h>
  49. #include <linux/string.h>
  50. #include <linux/errno.h>
  51. #include <linux/highmem.h>
  52. #include <linux/slab.h>
  53. #include <linux/socket.h>
  54. #include <linux/sockios.h>
  55. #include <linux/in.h>
  56. #include <linux/inet.h>
  57. #include <linux/netdevice.h>
  58. #include <linux/etherdevice.h>
  59. #include <linux/proc_fs.h>
  60. #include <linux/stat.h>
  61. #include <linux/init.h>
  62. #include <net/snmp.h>
  63. #include <net/ip.h>
  64. #include <net/protocol.h>
  65. #include <net/route.h>
  66. #include <net/xfrm.h>
  67. #include <linux/skbuff.h>
  68. #include <net/sock.h>
  69. #include <net/arp.h>
  70. #include <net/icmp.h>
  71. #include <net/checksum.h>
  72. #include <net/inetpeer.h>
  73. #include <linux/igmp.h>
  74. #include <linux/netfilter_ipv4.h>
  75. #include <linux/netfilter_bridge.h>
  76. #include <linux/mroute.h>
  77. #include <linux/netlink.h>
  78. #include <linux/tcp.h>
  79. int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
  80. EXPORT_SYMBOL(sysctl_ip_default_ttl);
  81. /* Generate a checksum for an outgoing IP datagram. */
  82. void ip_send_check(struct iphdr *iph)
  83. {
  84. iph->check = 0;
  85. iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
  86. }
  87. EXPORT_SYMBOL(ip_send_check);
  88. int __ip_local_out(struct sk_buff *skb)
  89. {
  90. struct iphdr *iph = ip_hdr(skb);
  91. iph->tot_len = htons(skb->len);
  92. ip_send_check(iph);
  93. return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
  94. skb_dst(skb)->dev, dst_output);
  95. }
  96. int ip_local_out(struct sk_buff *skb)
  97. {
  98. int err;
  99. err = __ip_local_out(skb);
  100. if (likely(err == 1))
  101. err = dst_output(skb);
  102. return err;
  103. }
  104. EXPORT_SYMBOL_GPL(ip_local_out);
  105. static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
  106. {
  107. int ttl = inet->uc_ttl;
  108. if (ttl < 0)
  109. ttl = ip4_dst_hoplimit(dst);
  110. return ttl;
  111. }
  112. /*
  113. * Add an ip header to a skbuff and send it out.
  114. *
  115. */
  116. int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
  117. __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
  118. {
  119. struct inet_sock *inet = inet_sk(sk);
  120. struct rtable *rt = skb_rtable(skb);
  121. struct iphdr *iph;
  122. /* Build the IP header. */
  123. skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
  124. skb_reset_network_header(skb);
  125. iph = ip_hdr(skb);
  126. iph->version = 4;
  127. iph->ihl = 5;
  128. iph->tos = inet->tos;
  129. if (ip_dont_fragment(sk, &rt->dst))
  130. iph->frag_off = htons(IP_DF);
  131. else
  132. iph->frag_off = 0;
  133. iph->ttl = ip_select_ttl(inet, &rt->dst);
  134. iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
  135. iph->saddr = saddr;
  136. iph->protocol = sk->sk_protocol;
  137. ip_select_ident(skb, &rt->dst, sk);
  138. if (opt && opt->opt.optlen) {
  139. iph->ihl += opt->opt.optlen>>2;
  140. ip_options_build(skb, &opt->opt, daddr, rt, 0);
  141. }
  142. skb->priority = sk->sk_priority;
  143. skb->mark = sk->sk_mark;
  144. /* Send it out. */
  145. return ip_local_out(skb);
  146. }
  147. EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
  148. static inline int ip_finish_output2(struct sk_buff *skb)
  149. {
  150. struct dst_entry *dst = skb_dst(skb);
  151. struct rtable *rt = (struct rtable *)dst;
  152. struct net_device *dev = dst->dev;
  153. unsigned int hh_len = LL_RESERVED_SPACE(dev);
  154. struct neighbour *neigh;
  155. u32 nexthop;
  156. if (rt->rt_type == RTN_MULTICAST) {
  157. IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
  158. } else if (rt->rt_type == RTN_BROADCAST)
  159. IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
  160. /* Be paranoid, rather than too clever. */
  161. if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
  162. struct sk_buff *skb2;
  163. skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
  164. if (skb2 == NULL) {
  165. kfree_skb(skb);
  166. return -ENOMEM;
  167. }
  168. if (skb->sk)
  169. skb_set_owner_w(skb2, skb->sk);
  170. consume_skb(skb);
  171. skb = skb2;
  172. }
  173. rcu_read_lock_bh();
  174. nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
  175. neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
  176. if (unlikely(!neigh))
  177. neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
  178. if (!IS_ERR(neigh)) {
  179. int res = dst_neigh_output(dst, neigh, skb);
  180. rcu_read_unlock_bh();
  181. return res;
  182. }
  183. rcu_read_unlock_bh();
  184. net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
  185. __func__);
  186. kfree_skb(skb);
  187. return -EINVAL;
  188. }
  189. static int ip_finish_output(struct sk_buff *skb)
  190. {
  191. #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
  192. /* Policy lookup after SNAT yielded a new policy */
  193. if (skb_dst(skb)->xfrm != NULL) {
  194. IPCB(skb)->flags |= IPSKB_REROUTED;
  195. return dst_output(skb);
  196. }
  197. #endif
  198. if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
  199. return ip_fragment(skb, ip_finish_output2);
  200. else
  201. return ip_finish_output2(skb);
  202. }
  203. int ip_mc_output(struct sk_buff *skb)
  204. {
  205. struct sock *sk = skb->sk;
  206. struct rtable *rt = skb_rtable(skb);
  207. struct net_device *dev = rt->dst.dev;
  208. /*
  209. * If the indicated interface is up and running, send the packet.
  210. */
  211. IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
  212. skb->dev = dev;
  213. skb->protocol = htons(ETH_P_IP);
  214. /*
  215. * Multicasts are looped back for other local users
  216. */
  217. if (rt->rt_flags&RTCF_MULTICAST) {
  218. if (sk_mc_loop(sk)
  219. #ifdef CONFIG_IP_MROUTE
  220. /* Small optimization: do not loopback not local frames,
  221. which returned after forwarding; they will be dropped
  222. by ip_mr_input in any case.
  223. Note, that local frames are looped back to be delivered
  224. to local recipients.
  225. This check is duplicated in ip_mr_input at the moment.
  226. */
  227. &&
  228. ((rt->rt_flags & RTCF_LOCAL) ||
  229. !(IPCB(skb)->flags & IPSKB_FORWARDED))
  230. #endif
  231. ) {
  232. struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
  233. if (newskb)
  234. NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
  235. newskb, NULL, newskb->dev,
  236. dev_loopback_xmit);
  237. }
  238. /* Multicasts with ttl 0 must not go beyond the host */
  239. if (ip_hdr(skb)->ttl == 0) {
  240. kfree_skb(skb);
  241. return 0;
  242. }
  243. }
  244. if (rt->rt_flags&RTCF_BROADCAST) {
  245. struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
  246. if (newskb)
  247. NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
  248. NULL, newskb->dev, dev_loopback_xmit);
  249. }
  250. return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
  251. skb->dev, ip_finish_output,
  252. !(IPCB(skb)->flags & IPSKB_REROUTED));
  253. }
  254. int ip_output(struct sk_buff *skb)
  255. {
  256. struct net_device *dev = skb_dst(skb)->dev;
  257. IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
  258. skb->dev = dev;
  259. skb->protocol = htons(ETH_P_IP);
  260. return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
  261. ip_finish_output,
  262. !(IPCB(skb)->flags & IPSKB_REROUTED));
  263. }
  264. /*
  265. * copy saddr and daddr, possibly using 64bit load/stores
  266. * Equivalent to :
  267. * iph->saddr = fl4->saddr;
  268. * iph->daddr = fl4->daddr;
  269. */
  270. static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
  271. {
  272. BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
  273. offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
  274. memcpy(&iph->saddr, &fl4->saddr,
  275. sizeof(fl4->saddr) + sizeof(fl4->daddr));
  276. }
  277. int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl)
  278. {
  279. struct sock *sk = skb->sk;
  280. struct inet_sock *inet = inet_sk(sk);
  281. struct ip_options_rcu *inet_opt;
  282. struct flowi4 *fl4;
  283. struct rtable *rt;
  284. struct iphdr *iph;
  285. int res;
  286. /* Skip all of this if the packet is already routed,
  287. * f.e. by something like SCTP.
  288. */
  289. rcu_read_lock();
  290. inet_opt = rcu_dereference(inet->inet_opt);
  291. fl4 = &fl->u.ip4;
  292. rt = skb_rtable(skb);
  293. if (rt != NULL)
  294. goto packet_routed;
  295. /* Make sure we can route this packet. */
  296. rt = (struct rtable *)__sk_dst_check(sk, 0);
  297. if (rt == NULL) {
  298. __be32 daddr;
  299. /* Use correct destination address if we have options. */
  300. daddr = inet->inet_daddr;
  301. if (inet_opt && inet_opt->opt.srr)
  302. daddr = inet_opt->opt.faddr;
  303. /* If this fails, retransmit mechanism of transport layer will
  304. * keep trying until route appears or the connection times
  305. * itself out.
  306. */
  307. rt = ip_route_output_ports(sock_net(sk), fl4, sk,
  308. daddr, inet->inet_saddr,
  309. inet->inet_dport,
  310. inet->inet_sport,
  311. sk->sk_protocol,
  312. RT_CONN_FLAGS(sk),
  313. sk->sk_bound_dev_if);
  314. if (IS_ERR(rt))
  315. goto no_route;
  316. sk_setup_caps(sk, &rt->dst);
  317. }
  318. skb_dst_set_noref(skb, &rt->dst);
  319. packet_routed:
  320. if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
  321. goto no_route;
  322. /* OK, we know where to send it, allocate and build IP header. */
  323. skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
  324. skb_reset_network_header(skb);
  325. iph = ip_hdr(skb);
  326. *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
  327. if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
  328. iph->frag_off = htons(IP_DF);
  329. else
  330. iph->frag_off = 0;
  331. iph->ttl = ip_select_ttl(inet, &rt->dst);
  332. iph->protocol = sk->sk_protocol;
  333. ip_copy_addrs(iph, fl4);
  334. /* Transport layer set skb->h.foo itself. */
  335. if (inet_opt && inet_opt->opt.optlen) {
  336. iph->ihl += inet_opt->opt.optlen >> 2;
  337. ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
  338. }
  339. ip_select_ident_more(skb, &rt->dst, sk,
  340. (skb_shinfo(skb)->gso_segs ?: 1) - 1);
  341. skb->priority = sk->sk_priority;
  342. skb->mark = sk->sk_mark;
  343. res = ip_local_out(skb);
  344. rcu_read_unlock();
  345. return res;
  346. no_route:
  347. rcu_read_unlock();
  348. IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
  349. kfree_skb(skb);
  350. return -EHOSTUNREACH;
  351. }
  352. EXPORT_SYMBOL(ip_queue_xmit);
  353. static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
  354. {
  355. to->pkt_type = from->pkt_type;
  356. to->priority = from->priority;
  357. to->protocol = from->protocol;
  358. skb_dst_drop(to);
  359. skb_dst_copy(to, from);
  360. to->dev = from->dev;
  361. to->mark = from->mark;
  362. /* Copy the flags to each fragment. */
  363. IPCB(to)->flags = IPCB(from)->flags;
  364. #ifdef CONFIG_NET_SCHED
  365. to->tc_index = from->tc_index;
  366. #endif
  367. nf_copy(to, from);
  368. #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
  369. to->nf_trace = from->nf_trace;
  370. #endif
  371. #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
  372. to->ipvs_property = from->ipvs_property;
  373. #endif
  374. skb_copy_secmark(to, from);
  375. }
  376. /*
  377. * This IP datagram is too large to be sent in one piece. Break it up into
  378. * smaller pieces (each of size equal to IP header plus
  379. * a block of the data of the original IP data part) that will yet fit in a
  380. * single device frame, and queue such a frame for sending.
  381. */
  382. int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
  383. {
  384. struct iphdr *iph;
  385. int ptr;
  386. struct net_device *dev;
  387. struct sk_buff *skb2;
  388. unsigned int mtu, hlen, left, len, ll_rs;
  389. int offset;
  390. __be16 not_last_frag;
  391. struct rtable *rt = skb_rtable(skb);
  392. int err = 0;
  393. dev = rt->dst.dev;
  394. /*
  395. * Point into the IP datagram header.
  396. */
  397. iph = ip_hdr(skb);
  398. if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) ||
  399. (IPCB(skb)->frag_max_size &&
  400. IPCB(skb)->frag_max_size > dst_mtu(&rt->dst)))) {
  401. IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
  402. icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
  403. htonl(ip_skb_dst_mtu(skb)));
  404. kfree_skb(skb);
  405. return -EMSGSIZE;
  406. }
  407. /*
  408. * Setup starting values.
  409. */
  410. hlen = iph->ihl * 4;
  411. mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */
  412. #ifdef CONFIG_BRIDGE_NETFILTER
  413. if (skb->nf_bridge)
  414. mtu -= nf_bridge_mtu_reduction(skb);
  415. #endif
  416. IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
  417. /* When frag_list is given, use it. First, check its validity:
  418. * some transformers could create wrong frag_list or break existing
  419. * one, it is not prohibited. In this case fall back to copying.
  420. *
  421. * LATER: this step can be merged to real generation of fragments,
  422. * we can switch to copy when see the first bad fragment.
  423. */
  424. if (skb_has_frag_list(skb)) {
  425. struct sk_buff *frag, *frag2;
  426. int first_len = skb_pagelen(skb);
  427. if (first_len - hlen > mtu ||
  428. ((first_len - hlen) & 7) ||
  429. ip_is_fragment(iph) ||
  430. skb_cloned(skb))
  431. goto slow_path;
  432. skb_walk_frags(skb, frag) {
  433. /* Correct geometry. */
  434. if (frag->len > mtu ||
  435. ((frag->len & 7) && frag->next) ||
  436. skb_headroom(frag) < hlen)
  437. goto slow_path_clean;
  438. /* Partially cloned skb? */
  439. if (skb_shared(frag))
  440. goto slow_path_clean;
  441. BUG_ON(frag->sk);
  442. if (skb->sk) {
  443. frag->sk = skb->sk;
  444. frag->destructor = sock_wfree;
  445. }
  446. skb->truesize -= frag->truesize;
  447. }
  448. /* Everything is OK. Generate! */
  449. err = 0;
  450. offset = 0;
  451. frag = skb_shinfo(skb)->frag_list;
  452. skb_frag_list_init(skb);
  453. skb->data_len = first_len - skb_headlen(skb);
  454. skb->len = first_len;
  455. iph->tot_len = htons(first_len);
  456. iph->frag_off = htons(IP_MF);
  457. ip_send_check(iph);
  458. for (;;) {
  459. /* Prepare header of the next frame,
  460. * before previous one went down. */
  461. if (frag) {
  462. frag->ip_summed = CHECKSUM_NONE;
  463. skb_reset_transport_header(frag);
  464. __skb_push(frag, hlen);
  465. skb_reset_network_header(frag);
  466. memcpy(skb_network_header(frag), iph, hlen);
  467. iph = ip_hdr(frag);
  468. iph->tot_len = htons(frag->len);
  469. ip_copy_metadata(frag, skb);
  470. if (offset == 0)
  471. ip_options_fragment(frag);
  472. offset += skb->len - hlen;
  473. iph->frag_off = htons(offset>>3);
  474. if (frag->next != NULL)
  475. iph->frag_off |= htons(IP_MF);
  476. /* Ready, complete checksum */
  477. ip_send_check(iph);
  478. }
  479. err = output(skb);
  480. if (!err)
  481. IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
  482. if (err || !frag)
  483. break;
  484. skb = frag;
  485. frag = skb->next;
  486. skb->next = NULL;
  487. }
  488. if (err == 0) {
  489. IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
  490. return 0;
  491. }
  492. while (frag) {
  493. skb = frag->next;
  494. kfree_skb(frag);
  495. frag = skb;
  496. }
  497. IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
  498. return err;
  499. slow_path_clean:
  500. skb_walk_frags(skb, frag2) {
  501. if (frag2 == frag)
  502. break;
  503. frag2->sk = NULL;
  504. frag2->destructor = NULL;
  505. skb->truesize += frag2->truesize;
  506. }
  507. }
  508. slow_path:
  509. /* for offloaded checksums cleanup checksum before fragmentation */
  510. if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
  511. goto fail;
  512. iph = ip_hdr(skb);
  513. left = skb->len - hlen; /* Space per frame */
  514. ptr = hlen; /* Where to start from */
  515. /* for bridged IP traffic encapsulated inside f.e. a vlan header,
  516. * we need to make room for the encapsulating header
  517. */
  518. ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
  519. /*
  520. * Fragment the datagram.
  521. */
  522. offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
  523. not_last_frag = iph->frag_off & htons(IP_MF);
  524. /*
  525. * Keep copying data until we run out.
  526. */
  527. while (left > 0) {
  528. len = left;
  529. /* IF: it doesn't fit, use 'mtu' - the data space left */
  530. if (len > mtu)
  531. len = mtu;
  532. /* IF: we are not sending up to and including the packet end
  533. then align the next start on an eight byte boundary */
  534. if (len < left) {
  535. len &= ~7;
  536. }
  537. /*
  538. * Allocate buffer.
  539. */
  540. if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
  541. NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
  542. err = -ENOMEM;
  543. goto fail;
  544. }
  545. /*
  546. * Set up data on packet
  547. */
  548. ip_copy_metadata(skb2, skb);
  549. skb_reserve(skb2, ll_rs);
  550. skb_put(skb2, len + hlen);
  551. skb_reset_network_header(skb2);
  552. skb2->transport_header = skb2->network_header + hlen;
  553. /*
  554. * Charge the memory for the fragment to any owner
  555. * it might possess
  556. */
  557. if (skb->sk)
  558. skb_set_owner_w(skb2, skb->sk);
  559. /*
  560. * Copy the packet header into the new buffer.
  561. */
  562. skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
  563. /*
  564. * Copy a block of the IP datagram.
  565. */
  566. if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
  567. BUG();
  568. left -= len;
  569. /*
  570. * Fill in the new header fields.
  571. */
  572. iph = ip_hdr(skb2);
  573. iph->frag_off = htons((offset >> 3));
  574. /* ANK: dirty, but effective trick. Upgrade options only if
  575. * the segment to be fragmented was THE FIRST (otherwise,
  576. * options are already fixed) and make it ONCE
  577. * on the initial skb, so that all the following fragments
  578. * will inherit fixed options.
  579. */
  580. if (offset == 0)
  581. ip_options_fragment(skb);
  582. /*
  583. * Added AC : If we are fragmenting a fragment that's not the
  584. * last fragment then keep MF on each bit
  585. */
  586. if (left > 0 || not_last_frag)
  587. iph->frag_off |= htons(IP_MF);
  588. ptr += len;
  589. offset += len;
  590. /*
  591. * Put this fragment into the sending queue.
  592. */
  593. iph->tot_len = htons(len + hlen);
  594. ip_send_check(iph);
  595. err = output(skb2);
  596. if (err)
  597. goto fail;
  598. IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
  599. }
  600. consume_skb(skb);
  601. IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
  602. return err;
  603. fail:
  604. kfree_skb(skb);
  605. IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
  606. return err;
  607. }
  608. EXPORT_SYMBOL(ip_fragment);
  609. int
  610. ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
  611. {
  612. struct iovec *iov = from;
  613. if (skb->ip_summed == CHECKSUM_PARTIAL) {
  614. if (memcpy_fromiovecend(to, iov, offset, len) < 0)
  615. return -EFAULT;
  616. } else {
  617. __wsum csum = 0;
  618. if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
  619. return -EFAULT;
  620. skb->csum = csum_block_add(skb->csum, csum, odd);
  621. }
  622. return 0;
  623. }
  624. EXPORT_SYMBOL(ip_generic_getfrag);
  625. static inline __wsum
  626. csum_page(struct page *page, int offset, int copy)
  627. {
  628. char *kaddr;
  629. __wsum csum;
  630. kaddr = kmap(page);
  631. csum = csum_partial(kaddr + offset, copy, 0);
  632. kunmap(page);
  633. return csum;
  634. }
  635. static inline int ip_ufo_append_data(struct sock *sk,
  636. struct sk_buff_head *queue,
  637. int getfrag(void *from, char *to, int offset, int len,
  638. int odd, struct sk_buff *skb),
  639. void *from, int length, int hh_len, int fragheaderlen,
  640. int transhdrlen, int maxfraglen, unsigned int flags)
  641. {
  642. struct sk_buff *skb;
  643. int err;
  644. /* There is support for UDP fragmentation offload by network
  645. * device, so create one single skb packet containing complete
  646. * udp datagram
  647. */
  648. if ((skb = skb_peek_tail(queue)) == NULL) {
  649. skb = sock_alloc_send_skb(sk,
  650. hh_len + fragheaderlen + transhdrlen + 20,
  651. (flags & MSG_DONTWAIT), &err);
  652. if (skb == NULL)
  653. return err;
  654. /* reserve space for Hardware header */
  655. skb_reserve(skb, hh_len);
  656. /* create space for UDP/IP header */
  657. skb_put(skb, fragheaderlen + transhdrlen);
  658. /* initialize network header pointer */
  659. skb_reset_network_header(skb);
  660. /* initialize protocol header pointer */
  661. skb->transport_header = skb->network_header + fragheaderlen;
  662. skb->csum = 0;
  663. __skb_queue_tail(queue, skb);
  664. } else if (skb_is_gso(skb)) {
  665. goto append;
  666. }
  667. skb->ip_summed = CHECKSUM_PARTIAL;
  668. /* specify the length of each IP datagram fragment */
  669. skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
  670. skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
  671. append:
  672. return skb_append_datato_frags(sk, skb, getfrag, from,
  673. (length - transhdrlen));
  674. }
  675. static int __ip_append_data(struct sock *sk,
  676. struct flowi4 *fl4,
  677. struct sk_buff_head *queue,
  678. struct inet_cork *cork,
  679. struct page_frag *pfrag,
  680. int getfrag(void *from, char *to, int offset,
  681. int len, int odd, struct sk_buff *skb),
  682. void *from, int length, int transhdrlen,
  683. unsigned int flags)
  684. {
  685. struct inet_sock *inet = inet_sk(sk);
  686. struct sk_buff *skb;
  687. struct ip_options *opt = cork->opt;
  688. int hh_len;
  689. int exthdrlen;
  690. int mtu;
  691. int copy;
  692. int err;
  693. int offset = 0;
  694. unsigned int maxfraglen, fragheaderlen;
  695. int csummode = CHECKSUM_NONE;
  696. struct rtable *rt = (struct rtable *)cork->dst;
  697. skb = skb_peek_tail(queue);
  698. exthdrlen = !skb ? rt->dst.header_len : 0;
  699. mtu = cork->fragsize;
  700. hh_len = LL_RESERVED_SPACE(rt->dst.dev);
  701. fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
  702. maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
  703. if (cork->length + length > 0xFFFF - fragheaderlen) {
  704. ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
  705. mtu-exthdrlen);
  706. return -EMSGSIZE;
  707. }
  708. /*
  709. * transhdrlen > 0 means that this is the first fragment and we wish
  710. * it won't be fragmented in the future.
  711. */
  712. if (transhdrlen &&
  713. length + fragheaderlen <= mtu &&
  714. rt->dst.dev->features & NETIF_F_V4_CSUM &&
  715. !exthdrlen)
  716. csummode = CHECKSUM_PARTIAL;
  717. cork->length += length;
  718. if (((length > mtu) || (skb && skb_is_gso(skb))) &&
  719. (sk->sk_protocol == IPPROTO_UDP) &&
  720. (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
  721. err = ip_ufo_append_data(sk, queue, getfrag, from, length,
  722. hh_len, fragheaderlen, transhdrlen,
  723. maxfraglen, flags);
  724. if (err)
  725. goto error;
  726. return 0;
  727. }
  728. /* So, what's going on in the loop below?
  729. *
  730. * We use calculated fragment length to generate chained skb,
  731. * each of segments is IP fragment ready for sending to network after
  732. * adding appropriate IP header.
  733. */
  734. if (!skb)
  735. goto alloc_new_skb;
  736. while (length > 0) {
  737. /* Check if the remaining data fits into current packet. */
  738. copy = mtu - skb->len;
  739. if (copy < length)
  740. copy = maxfraglen - skb->len;
  741. if (copy <= 0) {
  742. char *data;
  743. unsigned int datalen;
  744. unsigned int fraglen;
  745. unsigned int fraggap;
  746. unsigned int alloclen;
  747. struct sk_buff *skb_prev;
  748. alloc_new_skb:
  749. skb_prev = skb;
  750. if (skb_prev)
  751. fraggap = skb_prev->len - maxfraglen;
  752. else
  753. fraggap = 0;
  754. /*
  755. * If remaining data exceeds the mtu,
  756. * we know we need more fragment(s).
  757. */
  758. datalen = length + fraggap;
  759. if (datalen > mtu - fragheaderlen)
  760. datalen = maxfraglen - fragheaderlen;
  761. fraglen = datalen + fragheaderlen;
  762. if ((flags & MSG_MORE) &&
  763. !(rt->dst.dev->features&NETIF_F_SG))
  764. alloclen = mtu;
  765. else
  766. alloclen = fraglen;
  767. alloclen += exthdrlen;
  768. /* The last fragment gets additional space at tail.
  769. * Note, with MSG_MORE we overallocate on fragments,
  770. * because we have no idea what fragment will be
  771. * the last.
  772. */
  773. if (datalen == length + fraggap)
  774. alloclen += rt->dst.trailer_len;
  775. if (transhdrlen) {
  776. skb = sock_alloc_send_skb(sk,
  777. alloclen + hh_len + 15,
  778. (flags & MSG_DONTWAIT), &err);
  779. } else {
  780. skb = NULL;
  781. if (atomic_read(&sk->sk_wmem_alloc) <=
  782. 2 * sk->sk_sndbuf)
  783. skb = sock_wmalloc(sk,
  784. alloclen + hh_len + 15, 1,
  785. sk->sk_allocation);
  786. if (unlikely(skb == NULL))
  787. err = -ENOBUFS;
  788. else
  789. /* only the initial fragment is
  790. time stamped */
  791. cork->tx_flags = 0;
  792. }
  793. if (skb == NULL)
  794. goto error;
  795. /*
  796. * Fill in the control structures
  797. */
  798. skb->ip_summed = csummode;
  799. skb->csum = 0;
  800. skb_reserve(skb, hh_len);
  801. skb_shinfo(skb)->tx_flags = cork->tx_flags;
  802. /*
  803. * Find where to start putting bytes.
  804. */
  805. data = skb_put(skb, fraglen + exthdrlen);
  806. skb_set_network_header(skb, exthdrlen);
  807. skb->transport_header = (skb->network_header +
  808. fragheaderlen);
  809. data += fragheaderlen + exthdrlen;
  810. if (fraggap) {
  811. skb->csum = skb_copy_and_csum_bits(
  812. skb_prev, maxfraglen,
  813. data + transhdrlen, fraggap, 0);
  814. skb_prev->csum = csum_sub(skb_prev->csum,
  815. skb->csum);
  816. data += fraggap;
  817. pskb_trim_unique(skb_prev, maxfraglen);
  818. }
  819. copy = datalen - transhdrlen - fraggap;
  820. if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
  821. err = -EFAULT;
  822. kfree_skb(skb);
  823. goto error;
  824. }
  825. offset += copy;
  826. length -= datalen - fraggap;
  827. transhdrlen = 0;
  828. exthdrlen = 0;
  829. csummode = CHECKSUM_NONE;
  830. /*
  831. * Put the packet on the pending queue.
  832. */
  833. __skb_queue_tail(queue, skb);
  834. continue;
  835. }
  836. if (copy > length)
  837. copy = length;
  838. if (!(rt->dst.dev->features&NETIF_F_SG)) {
  839. unsigned int off;
  840. off = skb->len;
  841. if (getfrag(from, skb_put(skb, copy),
  842. offset, copy, off, skb) < 0) {
  843. __skb_trim(skb, off);
  844. err = -EFAULT;
  845. goto error;
  846. }
  847. } else {
  848. int i = skb_shinfo(skb)->nr_frags;
  849. err = -ENOMEM;
  850. if (!sk_page_frag_refill(sk, pfrag))
  851. goto error;
  852. if (!skb_can_coalesce(skb, i, pfrag->page,
  853. pfrag->offset)) {
  854. err = -EMSGSIZE;
  855. if (i == MAX_SKB_FRAGS)
  856. goto error;
  857. __skb_fill_page_desc(skb, i, pfrag->page,
  858. pfrag->offset, 0);
  859. skb_shinfo(skb)->nr_frags = ++i;
  860. get_page(pfrag->page);
  861. }
  862. copy = min_t(int, copy, pfrag->size - pfrag->offset);
  863. if (getfrag(from,
  864. page_address(pfrag->page) + pfrag->offset,
  865. offset, copy, skb->len, skb) < 0)
  866. goto error_efault;
  867. pfrag->offset += copy;
  868. skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
  869. skb->len += copy;
  870. skb->data_len += copy;
  871. skb->truesize += copy;
  872. atomic_add(copy, &sk->sk_wmem_alloc);
  873. }
  874. offset += copy;
  875. length -= copy;
  876. }
  877. return 0;
  878. error_efault:
  879. err = -EFAULT;
  880. error:
  881. cork->length -= length;
  882. IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
  883. return err;
  884. }
  885. static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
  886. struct ipcm_cookie *ipc, struct rtable **rtp)
  887. {
  888. struct inet_sock *inet = inet_sk(sk);
  889. struct ip_options_rcu *opt;
  890. struct rtable *rt;
  891. /*
  892. * setup for corking.
  893. */
  894. opt = ipc->opt;
  895. if (opt) {
  896. if (cork->opt == NULL) {
  897. cork->opt = kmalloc(sizeof(struct ip_options) + 40,
  898. sk->sk_allocation);
  899. if (unlikely(cork->opt == NULL))
  900. return -ENOBUFS;
  901. }
  902. memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
  903. cork->flags |= IPCORK_OPT;
  904. cork->addr = ipc->addr;
  905. }
  906. rt = *rtp;
  907. if (unlikely(!rt))
  908. return -EFAULT;
  909. /*
  910. * We steal reference to this route, caller should not release it
  911. */
  912. *rtp = NULL;
  913. cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ?
  914. rt->dst.dev->mtu : dst_mtu(&rt->dst);
  915. cork->dst = &rt->dst;
  916. cork->length = 0;
  917. cork->tx_flags = ipc->tx_flags;
  918. return 0;
  919. }
  920. /*
  921. * ip_append_data() and ip_append_page() can make one large IP datagram
  922. * from many pieces of data. Each pieces will be holded on the socket
  923. * until ip_push_pending_frames() is called. Each piece can be a page
  924. * or non-page data.
  925. *
  926. * Not only UDP, other transport protocols - e.g. raw sockets - can use
  927. * this interface potentially.
  928. *
  929. * LATER: length must be adjusted by pad at tail, when it is required.
  930. */
  931. int ip_append_data(struct sock *sk, struct flowi4 *fl4,
  932. int getfrag(void *from, char *to, int offset, int len,
  933. int odd, struct sk_buff *skb),
  934. void *from, int length, int transhdrlen,
  935. struct ipcm_cookie *ipc, struct rtable **rtp,
  936. unsigned int flags)
  937. {
  938. struct inet_sock *inet = inet_sk(sk);
  939. int err;
  940. if (flags&MSG_PROBE)
  941. return 0;
  942. if (skb_queue_empty(&sk->sk_write_queue)) {
  943. err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
  944. if (err)
  945. return err;
  946. } else {
  947. transhdrlen = 0;
  948. }
  949. return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
  950. sk_page_frag(sk), getfrag,
  951. from, length, transhdrlen, flags);
  952. }
  953. ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
  954. int offset, size_t size, int flags)
  955. {
  956. struct inet_sock *inet = inet_sk(sk);
  957. struct sk_buff *skb;
  958. struct rtable *rt;
  959. struct ip_options *opt = NULL;
  960. struct inet_cork *cork;
  961. int hh_len;
  962. int mtu;
  963. int len;
  964. int err;
  965. unsigned int maxfraglen, fragheaderlen, fraggap;
  966. if (inet->hdrincl)
  967. return -EPERM;
  968. if (flags&MSG_PROBE)
  969. return 0;
  970. if (skb_queue_empty(&sk->sk_write_queue))
  971. return -EINVAL;
  972. cork = &inet->cork.base;
  973. rt = (struct rtable *)cork->dst;
  974. if (cork->flags & IPCORK_OPT)
  975. opt = cork->opt;
  976. if (!(rt->dst.dev->features&NETIF_F_SG))
  977. return -EOPNOTSUPP;
  978. hh_len = LL_RESERVED_SPACE(rt->dst.dev);
  979. mtu = cork->fragsize;
  980. fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
  981. maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
  982. if (cork->length + size > 0xFFFF - fragheaderlen) {
  983. ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, mtu);
  984. return -EMSGSIZE;
  985. }
  986. if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
  987. return -EINVAL;
  988. cork->length += size;
  989. if ((size + skb->len > mtu) &&
  990. (sk->sk_protocol == IPPROTO_UDP) &&
  991. (rt->dst.dev->features & NETIF_F_UFO)) {
  992. skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
  993. skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
  994. }
  995. while (size > 0) {
  996. int i;
  997. if (skb_is_gso(skb))
  998. len = size;
  999. else {
  1000. /* Check if the remaining data fits into current packet. */
  1001. len = mtu - skb->len;
  1002. if (len < size)
  1003. len = maxfraglen - skb->len;
  1004. }
  1005. if (len <= 0) {
  1006. struct sk_buff *skb_prev;
  1007. int alloclen;
  1008. skb_prev = skb;
  1009. fraggap = skb_prev->len - maxfraglen;
  1010. alloclen = fragheaderlen + hh_len + fraggap + 15;
  1011. skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
  1012. if (unlikely(!skb)) {
  1013. err = -ENOBUFS;
  1014. goto error;
  1015. }
  1016. /*
  1017. * Fill in the control structures
  1018. */
  1019. skb->ip_summed = CHECKSUM_NONE;
  1020. skb->csum = 0;
  1021. skb_reserve(skb, hh_len);
  1022. /*
  1023. * Find where to start putting bytes.
  1024. */
  1025. skb_put(skb, fragheaderlen + fraggap);
  1026. skb_reset_network_header(skb);
  1027. skb->transport_header = (skb->network_header +
  1028. fragheaderlen);
  1029. if (fraggap) {
  1030. skb->csum = skb_copy_and_csum_bits(skb_prev,
  1031. maxfraglen,
  1032. skb_transport_header(skb),
  1033. fraggap, 0);
  1034. skb_prev->csum = csum_sub(skb_prev->csum,
  1035. skb->csum);
  1036. pskb_trim_unique(skb_prev, maxfraglen);
  1037. }
  1038. /*
  1039. * Put the packet on the pending queue.
  1040. */
  1041. __skb_queue_tail(&sk->sk_write_queue, skb);
  1042. continue;
  1043. }
  1044. i = skb_shinfo(skb)->nr_frags;
  1045. if (len > size)
  1046. len = size;
  1047. if (skb_can_coalesce(skb, i, page, offset)) {
  1048. skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
  1049. } else if (i < MAX_SKB_FRAGS) {
  1050. get_page(page);
  1051. skb_fill_page_desc(skb, i, page, offset, len);
  1052. } else {
  1053. err = -EMSGSIZE;
  1054. goto error;
  1055. }
  1056. if (skb->ip_summed == CHECKSUM_NONE) {
  1057. __wsum csum;
  1058. csum = csum_page(page, offset, len);
  1059. skb->csum = csum_block_add(skb->csum, csum, skb->len);
  1060. }
  1061. skb->len += len;
  1062. skb->data_len += len;
  1063. skb->truesize += len;
  1064. atomic_add(len, &sk->sk_wmem_alloc);
  1065. offset += len;
  1066. size -= len;
  1067. }
  1068. return 0;
  1069. error:
  1070. cork->length -= size;
  1071. IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
  1072. return err;
  1073. }
  1074. static void ip_cork_release(struct inet_cork *cork)
  1075. {
  1076. cork->flags &= ~IPCORK_OPT;
  1077. kfree(cork->opt);
  1078. cork->opt = NULL;
  1079. dst_release(cork->dst);
  1080. cork->dst = NULL;
  1081. }
  1082. /*
  1083. * Combined all pending IP fragments on the socket as one IP datagram
  1084. * and push them out.
  1085. */
  1086. struct sk_buff *__ip_make_skb(struct sock *sk,
  1087. struct flowi4 *fl4,
  1088. struct sk_buff_head *queue,
  1089. struct inet_cork *cork)
  1090. {
  1091. struct sk_buff *skb, *tmp_skb;
  1092. struct sk_buff **tail_skb;
  1093. struct inet_sock *inet = inet_sk(sk);
  1094. struct net *net = sock_net(sk);
  1095. struct ip_options *opt = NULL;
  1096. struct rtable *rt = (struct rtable *)cork->dst;
  1097. struct iphdr *iph;
  1098. __be16 df = 0;
  1099. __u8 ttl;
  1100. if ((skb = __skb_dequeue(queue)) == NULL)
  1101. goto out;
  1102. tail_skb = &(skb_shinfo(skb)->frag_list);
  1103. /* move skb->data to ip header from ext header */
  1104. if (skb->data < skb_network_header(skb))
  1105. __skb_pull(skb, skb_network_offset(skb));
  1106. while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
  1107. __skb_pull(tmp_skb, skb_network_header_len(skb));
  1108. *tail_skb = tmp_skb;
  1109. tail_skb = &(tmp_skb->next);
  1110. skb->len += tmp_skb->len;
  1111. skb->data_len += tmp_skb->len;
  1112. skb->truesize += tmp_skb->truesize;
  1113. tmp_skb->destructor = NULL;
  1114. tmp_skb->sk = NULL;
  1115. }
  1116. /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
  1117. * to fragment the frame generated here. No matter, what transforms
  1118. * how transforms change size of the packet, it will come out.
  1119. */
  1120. if (inet->pmtudisc < IP_PMTUDISC_DO)
  1121. skb->local_df = 1;
  1122. /* DF bit is set when we want to see DF on outgoing frames.
  1123. * If local_df is set too, we still allow to fragment this frame
  1124. * locally. */
  1125. if (inet->pmtudisc >= IP_PMTUDISC_DO ||
  1126. (skb->len <= dst_mtu(&rt->dst) &&
  1127. ip_dont_fragment(sk, &rt->dst)))
  1128. df = htons(IP_DF);
  1129. if (cork->flags & IPCORK_OPT)
  1130. opt = cork->opt;
  1131. if (rt->rt_type == RTN_MULTICAST)
  1132. ttl = inet->mc_ttl;
  1133. else
  1134. ttl = ip_select_ttl(inet, &rt->dst);
  1135. iph = ip_hdr(skb);
  1136. iph->version = 4;
  1137. iph->ihl = 5;
  1138. iph->tos = inet->tos;
  1139. iph->frag_off = df;
  1140. iph->ttl = ttl;
  1141. iph->protocol = sk->sk_protocol;
  1142. ip_copy_addrs(iph, fl4);
  1143. ip_select_ident(skb, &rt->dst, sk);
  1144. if (opt) {
  1145. iph->ihl += opt->optlen>>2;
  1146. ip_options_build(skb, opt, cork->addr, rt, 0);
  1147. }
  1148. skb->priority = sk->sk_priority;
  1149. skb->mark = sk->sk_mark;
  1150. /*
  1151. * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
  1152. * on dst refcount
  1153. */
  1154. cork->dst = NULL;
  1155. skb_dst_set(skb, &rt->dst);
  1156. if (iph->protocol == IPPROTO_ICMP)
  1157. icmp_out_count(net, ((struct icmphdr *)
  1158. skb_transport_header(skb))->type);
  1159. ip_cork_release(cork);
  1160. out:
  1161. return skb;
  1162. }
  1163. int ip_send_skb(struct net *net, struct sk_buff *skb)
  1164. {
  1165. int err;
  1166. err = ip_local_out(skb);
  1167. if (err) {
  1168. if (err > 0)
  1169. err = net_xmit_errno(err);
  1170. if (err)
  1171. IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
  1172. }
  1173. return err;
  1174. }
  1175. int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
  1176. {
  1177. struct sk_buff *skb;
  1178. skb = ip_finish_skb(sk, fl4);
  1179. if (!skb)
  1180. return 0;
  1181. /* Netfilter gets whole the not fragmented skb. */
  1182. return ip_send_skb(sock_net(sk), skb);
  1183. }
  1184. /*
  1185. * Throw away all pending data on the socket.
  1186. */
  1187. static void __ip_flush_pending_frames(struct sock *sk,
  1188. struct sk_buff_head *queue,
  1189. struct inet_cork *cork)
  1190. {
  1191. struct sk_buff *skb;
  1192. while ((skb = __skb_dequeue_tail(queue)) != NULL)
  1193. kfree_skb(skb);
  1194. ip_cork_release(cork);
  1195. }
  1196. void ip_flush_pending_frames(struct sock *sk)
  1197. {
  1198. __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
  1199. }
  1200. struct sk_buff *ip_make_skb(struct sock *sk,
  1201. struct flowi4 *fl4,
  1202. int getfrag(void *from, char *to, int offset,
  1203. int len, int odd, struct sk_buff *skb),
  1204. void *from, int length, int transhdrlen,
  1205. struct ipcm_cookie *ipc, struct rtable **rtp,
  1206. unsigned int flags)
  1207. {
  1208. struct inet_cork cork;
  1209. struct sk_buff_head queue;
  1210. int err;
  1211. if (flags & MSG_PROBE)
  1212. return NULL;
  1213. __skb_queue_head_init(&queue);
  1214. cork.flags = 0;
  1215. cork.addr = 0;
  1216. cork.opt = NULL;
  1217. err = ip_setup_cork(sk, &cork, ipc, rtp);
  1218. if (err)
  1219. return ERR_PTR(err);
  1220. err = __ip_append_data(sk, fl4, &queue, &cork,
  1221. &current->task_frag, getfrag,
  1222. from, length, transhdrlen, flags);
  1223. if (err) {
  1224. __ip_flush_pending_frames(sk, &queue, &cork);
  1225. return ERR_PTR(err);
  1226. }
  1227. return __ip_make_skb(sk, fl4, &queue, &cork);
  1228. }
  1229. /*
  1230. * Fetch data from kernel space and fill in checksum if needed.
  1231. */
  1232. static int ip_reply_glue_bits(void *dptr, char *to, int offset,
  1233. int len, int odd, struct sk_buff *skb)
  1234. {
  1235. __wsum csum;
  1236. csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
  1237. skb->csum = csum_block_add(skb->csum, csum, odd);
  1238. return 0;
  1239. }
  1240. /*
  1241. * Generic function to send a packet as reply to another packet.
  1242. * Used to send some TCP resets/acks so far.
  1243. *
  1244. * Use a fake percpu inet socket to avoid false sharing and contention.
  1245. */
  1246. static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
  1247. .sk = {
  1248. .__sk_common = {
  1249. .skc_refcnt = ATOMIC_INIT(1),
  1250. },
  1251. .sk_wmem_alloc = ATOMIC_INIT(1),
  1252. .sk_allocation = GFP_ATOMIC,
  1253. .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE),
  1254. },
  1255. .pmtudisc = IP_PMTUDISC_WANT,
  1256. .uc_ttl = -1,
  1257. };
  1258. void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
  1259. __be32 saddr, const struct ip_reply_arg *arg,
  1260. unsigned int len)
  1261. {
  1262. struct ip_options_data replyopts;
  1263. struct ipcm_cookie ipc;
  1264. struct flowi4 fl4;
  1265. struct rtable *rt = skb_rtable(skb);
  1266. struct sk_buff *nskb;
  1267. struct sock *sk;
  1268. struct inet_sock *inet;
  1269. if (ip_options_echo(&replyopts.opt.opt, skb))
  1270. return;
  1271. ipc.addr = daddr;
  1272. ipc.opt = NULL;
  1273. ipc.tx_flags = 0;
  1274. if (replyopts.opt.opt.optlen) {
  1275. ipc.opt = &replyopts.opt;
  1276. if (replyopts.opt.opt.srr)
  1277. daddr = replyopts.opt.opt.faddr;
  1278. }
  1279. flowi4_init_output(&fl4, arg->bound_dev_if, 0,
  1280. RT_TOS(arg->tos),
  1281. RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
  1282. ip_reply_arg_flowi_flags(arg),
  1283. daddr, saddr,
  1284. tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
  1285. security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
  1286. rt = ip_route_output_key(net, &fl4);
  1287. if (IS_ERR(rt))
  1288. return;
  1289. inet = &get_cpu_var(unicast_sock);
  1290. inet->tos = arg->tos;
  1291. sk = &inet->sk;
  1292. sk->sk_priority = skb->priority;
  1293. sk->sk_protocol = ip_hdr(skb)->protocol;
  1294. sk->sk_bound_dev_if = arg->bound_dev_if;
  1295. sock_net_set(sk, net);
  1296. __skb_queue_head_init(&sk->sk_write_queue);
  1297. sk->sk_sndbuf = sysctl_wmem_default;
  1298. ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
  1299. &ipc, &rt, MSG_DONTWAIT);
  1300. nskb = skb_peek(&sk->sk_write_queue);
  1301. if (nskb) {
  1302. if (arg->csumoffset >= 0)
  1303. *((__sum16 *)skb_transport_header(nskb) +
  1304. arg->csumoffset) = csum_fold(csum_add(nskb->csum,
  1305. arg->csum));
  1306. nskb->ip_summed = CHECKSUM_NONE;
  1307. skb_orphan(nskb);
  1308. skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
  1309. ip_push_pending_frames(sk, &fl4);
  1310. }
  1311. put_cpu_var(unicast_sock);
  1312. ip_rt_put(rt);
  1313. }
  1314. void __init ip_init(void)
  1315. {
  1316. ip_rt_init();
  1317. inet_initpeers();
  1318. #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
  1319. igmp_mc_proc_init();
  1320. #endif
  1321. }