]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - net/ipv4/ip_output.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ipv4: tcp: send zero IPID in SYNACK messages
[mirror_ubuntu-jammy-kernel.git] / net / ipv4 / ip_output.c
1 // SPDX-License-Identifier: GPL-2.0-only
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 *
9 * Authors: Ross Biro
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.
24 * Bradford Johnson: Fix faulty handling of some frames when
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.
35 * Andi Kleen : Split fast and slow ip_build_xmit path
36 * for decreased register pressure on x86
37 * and more readability.
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
46 #include <linux/uaccess.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
50 #include <linux/mm.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54 #include <linux/slab.h>
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>
70 #include <net/xfrm.h>
71 #include <linux/skbuff.h>
72 #include <net/sock.h>
73 #include <net/arp.h>
74 #include <net/icmp.h>
75 #include <net/checksum.h>
76 #include <net/inetpeer.h>
77 #include <net/inet_ecn.h>
78 #include <net/lwtunnel.h>
79 #include <linux/bpf-cgroup.h>
80 #include <linux/igmp.h>
81 #include <linux/netfilter_ipv4.h>
82 #include <linux/netfilter_bridge.h>
83 #include <linux/netlink.h>
84 #include <linux/tcp.h>
85
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 *));
90
91 /* Generate a checksum for an outgoing IP datagram. */
92 void ip_send_check(struct iphdr *iph)
93 {
94 iph->check = 0;
95 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
96 }
97 EXPORT_SYMBOL(ip_send_check);
98
99 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
100 {
101 struct iphdr *iph = ip_hdr(skb);
102
103 iph->tot_len = htons(skb->len);
104 ip_send_check(iph);
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
113 skb->protocol = htons(ETH_P_IP);
114
115 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
116 net, sk, skb, NULL, skb_dst(skb)->dev,
117 dst_output);
118 }
119
120 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
121 {
122 int err;
123
124 err = __ip_local_out(net, sk, skb);
125 if (likely(err == 1))
126 err = dst_output(net, sk, skb);
127
128 return err;
129 }
130 EXPORT_SYMBOL_GPL(ip_local_out);
131
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)
137 ttl = ip4_dst_hoplimit(dst);
138 return ttl;
139 }
140
141 /*
142 * Add an ip header to a skbuff and send it out.
143 *
144 */
145 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
146 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
147 u8 tos)
148 {
149 struct inet_sock *inet = inet_sk(sk);
150 struct rtable *rt = skb_rtable(skb);
151 struct net *net = sock_net(sk);
152 struct iphdr *iph;
153
154 /* Build the IP header. */
155 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
156 skb_reset_network_header(skb);
157 iph = ip_hdr(skb);
158 iph->version = 4;
159 iph->ihl = 5;
160 iph->tos = tos;
161 iph->ttl = ip_select_ttl(inet, &rt->dst);
162 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
163 iph->saddr = saddr;
164 iph->protocol = sk->sk_protocol;
165 /* Do not bother generating IPID for small packets (eg SYNACK) */
166 if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, &rt->dst)) {
167 iph->frag_off = htons(IP_DF);
168 iph->id = 0;
169 } else {
170 iph->frag_off = 0;
171 /* TCP packets here are SYNACK with fat IPv4/TCP options.
172 * Avoid using the hashed IP ident generator.
173 */
174 if (sk->sk_protocol == IPPROTO_TCP)
175 iph->id = (__force __be16)prandom_u32();
176 else
177 __ip_select_ident(net, iph, 1);
178 }
179
180 if (opt && opt->opt.optlen) {
181 iph->ihl += opt->opt.optlen>>2;
182 ip_options_build(skb, &opt->opt, daddr, rt, 0);
183 }
184
185 skb->priority = sk->sk_priority;
186 if (!skb->mark)
187 skb->mark = sk->sk_mark;
188
189 /* Send it out. */
190 return ip_local_out(net, skb->sk, skb);
191 }
192 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
193
194 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
195 {
196 struct dst_entry *dst = skb_dst(skb);
197 struct rtable *rt = (struct rtable *)dst;
198 struct net_device *dev = dst->dev;
199 unsigned int hh_len = LL_RESERVED_SPACE(dev);
200 struct neighbour *neigh;
201 bool is_v6gw = false;
202
203 if (rt->rt_type == RTN_MULTICAST) {
204 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
205 } else if (rt->rt_type == RTN_BROADCAST)
206 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
207
208 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
209 skb = skb_expand_head(skb, hh_len);
210 if (!skb)
211 return -ENOMEM;
212 }
213
214 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
215 int res = lwtunnel_xmit(skb);
216
217 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
218 return res;
219 }
220
221 rcu_read_lock_bh();
222 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
223 if (!IS_ERR(neigh)) {
224 int res;
225
226 sock_confirm_neigh(skb, neigh);
227 /* if crossing protocols, can not use the cached header */
228 res = neigh_output(neigh, skb, is_v6gw);
229 rcu_read_unlock_bh();
230 return res;
231 }
232 rcu_read_unlock_bh();
233
234 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
235 __func__);
236 kfree_skb(skb);
237 return -EINVAL;
238 }
239
240 static int ip_finish_output_gso(struct net *net, struct sock *sk,
241 struct sk_buff *skb, unsigned int mtu)
242 {
243 struct sk_buff *segs, *nskb;
244 netdev_features_t features;
245 int ret = 0;
246
247 /* common case: seglen is <= mtu
248 */
249 if (skb_gso_validate_network_len(skb, mtu))
250 return ip_finish_output2(net, sk, skb);
251
252 /* Slowpath - GSO segment length exceeds the egress MTU.
253 *
254 * This can happen in several cases:
255 * - Forwarding of a TCP GRO skb, when DF flag is not set.
256 * - Forwarding of an skb that arrived on a virtualization interface
257 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
258 * stack.
259 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
260 * interface with a smaller MTU.
261 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
262 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
263 * insufficient MTU.
264 */
265 features = netif_skb_features(skb);
266 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
267 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
268 if (IS_ERR_OR_NULL(segs)) {
269 kfree_skb(skb);
270 return -ENOMEM;
271 }
272
273 consume_skb(skb);
274
275 skb_list_walk_safe(segs, segs, nskb) {
276 int err;
277
278 skb_mark_not_on_list(segs);
279 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
280
281 if (err && ret == 0)
282 ret = err;
283 }
284
285 return ret;
286 }
287
288 static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
289 {
290 unsigned int mtu;
291
292 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
293 /* Policy lookup after SNAT yielded a new policy */
294 if (skb_dst(skb)->xfrm) {
295 IPCB(skb)->flags |= IPSKB_REROUTED;
296 return dst_output(net, sk, skb);
297 }
298 #endif
299 mtu = ip_skb_dst_mtu(sk, skb);
300 if (skb_is_gso(skb))
301 return ip_finish_output_gso(net, sk, skb, mtu);
302
303 if (skb->len > mtu || IPCB(skb)->frag_max_size)
304 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
305
306 return ip_finish_output2(net, sk, skb);
307 }
308
309 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
310 {
311 int ret;
312
313 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
314 switch (ret) {
315 case NET_XMIT_SUCCESS:
316 return __ip_finish_output(net, sk, skb);
317 case NET_XMIT_CN:
318 return __ip_finish_output(net, sk, skb) ? : ret;
319 default:
320 kfree_skb(skb);
321 return ret;
322 }
323 }
324
325 static int ip_mc_finish_output(struct net *net, struct sock *sk,
326 struct sk_buff *skb)
327 {
328 struct rtable *new_rt;
329 bool do_cn = false;
330 int ret, err;
331
332 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
333 switch (ret) {
334 case NET_XMIT_CN:
335 do_cn = true;
336 fallthrough;
337 case NET_XMIT_SUCCESS:
338 break;
339 default:
340 kfree_skb(skb);
341 return ret;
342 }
343
344 /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
345 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
346 * see ipv4_pktinfo_prepare().
347 */
348 new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
349 if (new_rt) {
350 new_rt->rt_iif = 0;
351 skb_dst_drop(skb);
352 skb_dst_set(skb, &new_rt->dst);
353 }
354
355 err = dev_loopback_xmit(net, sk, skb);
356 return (do_cn && err) ? ret : err;
357 }
358
359 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
360 {
361 struct rtable *rt = skb_rtable(skb);
362 struct net_device *dev = rt->dst.dev;
363
364 /*
365 * If the indicated interface is up and running, send the packet.
366 */
367 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
368
369 skb->dev = dev;
370 skb->protocol = htons(ETH_P_IP);
371
372 /*
373 * Multicasts are looped back for other local users
374 */
375
376 if (rt->rt_flags&RTCF_MULTICAST) {
377 if (sk_mc_loop(sk)
378 #ifdef CONFIG_IP_MROUTE
379 /* Small optimization: do not loopback not local frames,
380 which returned after forwarding; they will be dropped
381 by ip_mr_input in any case.
382 Note, that local frames are looped back to be delivered
383 to local recipients.
384
385 This check is duplicated in ip_mr_input at the moment.
386 */
387 &&
388 ((rt->rt_flags & RTCF_LOCAL) ||
389 !(IPCB(skb)->flags & IPSKB_FORWARDED))
390 #endif
391 ) {
392 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
393 if (newskb)
394 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
395 net, sk, newskb, NULL, newskb->dev,
396 ip_mc_finish_output);
397 }
398
399 /* Multicasts with ttl 0 must not go beyond the host */
400
401 if (ip_hdr(skb)->ttl == 0) {
402 kfree_skb(skb);
403 return 0;
404 }
405 }
406
407 if (rt->rt_flags&RTCF_BROADCAST) {
408 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
409 if (newskb)
410 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
411 net, sk, newskb, NULL, newskb->dev,
412 ip_mc_finish_output);
413 }
414
415 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
416 net, sk, skb, NULL, skb->dev,
417 ip_finish_output,
418 !(IPCB(skb)->flags & IPSKB_REROUTED));
419 }
420
421 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
422 {
423 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
424
425 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
426
427 skb->dev = dev;
428 skb->protocol = htons(ETH_P_IP);
429
430 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
431 net, sk, skb, indev, dev,
432 ip_finish_output,
433 !(IPCB(skb)->flags & IPSKB_REROUTED));
434 }
435 EXPORT_SYMBOL(ip_output);
436
437 /*
438 * copy saddr and daddr, possibly using 64bit load/stores
439 * Equivalent to :
440 * iph->saddr = fl4->saddr;
441 * iph->daddr = fl4->daddr;
442 */
443 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
444 {
445 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
446 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
447
448 iph->saddr = fl4->saddr;
449 iph->daddr = fl4->daddr;
450 }
451
452 /* Note: skb->sk can be different from sk, in case of tunnels */
453 int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
454 __u8 tos)
455 {
456 struct inet_sock *inet = inet_sk(sk);
457 struct net *net = sock_net(sk);
458 struct ip_options_rcu *inet_opt;
459 struct flowi4 *fl4;
460 struct rtable *rt;
461 struct iphdr *iph;
462 int res;
463
464 /* Skip all of this if the packet is already routed,
465 * f.e. by something like SCTP.
466 */
467 rcu_read_lock();
468 inet_opt = rcu_dereference(inet->inet_opt);
469 fl4 = &fl->u.ip4;
470 rt = skb_rtable(skb);
471 if (rt)
472 goto packet_routed;
473
474 /* Make sure we can route this packet. */
475 rt = (struct rtable *)__sk_dst_check(sk, 0);
476 if (!rt) {
477 __be32 daddr;
478
479 /* Use correct destination address if we have options. */
480 daddr = inet->inet_daddr;
481 if (inet_opt && inet_opt->opt.srr)
482 daddr = inet_opt->opt.faddr;
483
484 /* If this fails, retransmit mechanism of transport layer will
485 * keep trying until route appears or the connection times
486 * itself out.
487 */
488 rt = ip_route_output_ports(net, fl4, sk,
489 daddr, inet->inet_saddr,
490 inet->inet_dport,
491 inet->inet_sport,
492 sk->sk_protocol,
493 RT_CONN_FLAGS_TOS(sk, tos),
494 sk->sk_bound_dev_if);
495 if (IS_ERR(rt))
496 goto no_route;
497 sk_setup_caps(sk, &rt->dst);
498 }
499 skb_dst_set_noref(skb, &rt->dst);
500
501 packet_routed:
502 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
503 goto no_route;
504
505 /* OK, we know where to send it, allocate and build IP header. */
506 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
507 skb_reset_network_header(skb);
508 iph = ip_hdr(skb);
509 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
510 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
511 iph->frag_off = htons(IP_DF);
512 else
513 iph->frag_off = 0;
514 iph->ttl = ip_select_ttl(inet, &rt->dst);
515 iph->protocol = sk->sk_protocol;
516 ip_copy_addrs(iph, fl4);
517
518 /* Transport layer set skb->h.foo itself. */
519
520 if (inet_opt && inet_opt->opt.optlen) {
521 iph->ihl += inet_opt->opt.optlen >> 2;
522 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
523 }
524
525 ip_select_ident_segs(net, skb, sk,
526 skb_shinfo(skb)->gso_segs ?: 1);
527
528 /* TODO : should we use skb->sk here instead of sk ? */
529 skb->priority = sk->sk_priority;
530 skb->mark = sk->sk_mark;
531
532 res = ip_local_out(net, sk, skb);
533 rcu_read_unlock();
534 return res;
535
536 no_route:
537 rcu_read_unlock();
538 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
539 kfree_skb(skb);
540 return -EHOSTUNREACH;
541 }
542 EXPORT_SYMBOL(__ip_queue_xmit);
543
544 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
545 {
546 return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
547 }
548 EXPORT_SYMBOL(ip_queue_xmit);
549
550 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
551 {
552 to->pkt_type = from->pkt_type;
553 to->priority = from->priority;
554 to->protocol = from->protocol;
555 to->skb_iif = from->skb_iif;
556 skb_dst_drop(to);
557 skb_dst_copy(to, from);
558 to->dev = from->dev;
559 to->mark = from->mark;
560
561 skb_copy_hash(to, from);
562
563 #ifdef CONFIG_NET_SCHED
564 to->tc_index = from->tc_index;
565 #endif
566 nf_copy(to, from);
567 skb_ext_copy(to, from);
568 #if IS_ENABLED(CONFIG_IP_VS)
569 to->ipvs_property = from->ipvs_property;
570 #endif
571 skb_copy_secmark(to, from);
572 }
573
574 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
575 unsigned int mtu,
576 int (*output)(struct net *, struct sock *, struct sk_buff *))
577 {
578 struct iphdr *iph = ip_hdr(skb);
579
580 if ((iph->frag_off & htons(IP_DF)) == 0)
581 return ip_do_fragment(net, sk, skb, output);
582
583 if (unlikely(!skb->ignore_df ||
584 (IPCB(skb)->frag_max_size &&
585 IPCB(skb)->frag_max_size > mtu))) {
586 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
587 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
588 htonl(mtu));
589 kfree_skb(skb);
590 return -EMSGSIZE;
591 }
592
593 return ip_do_fragment(net, sk, skb, output);
594 }
595
596 void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
597 unsigned int hlen, struct ip_fraglist_iter *iter)
598 {
599 unsigned int first_len = skb_pagelen(skb);
600
601 iter->frag = skb_shinfo(skb)->frag_list;
602 skb_frag_list_init(skb);
603
604 iter->offset = 0;
605 iter->iph = iph;
606 iter->hlen = hlen;
607
608 skb->data_len = first_len - skb_headlen(skb);
609 skb->len = first_len;
610 iph->tot_len = htons(first_len);
611 iph->frag_off = htons(IP_MF);
612 ip_send_check(iph);
613 }
614 EXPORT_SYMBOL(ip_fraglist_init);
615
616 void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
617 {
618 unsigned int hlen = iter->hlen;
619 struct iphdr *iph = iter->iph;
620 struct sk_buff *frag;
621
622 frag = iter->frag;
623 frag->ip_summed = CHECKSUM_NONE;
624 skb_reset_transport_header(frag);
625 __skb_push(frag, hlen);
626 skb_reset_network_header(frag);
627 memcpy(skb_network_header(frag), iph, hlen);
628 iter->iph = ip_hdr(frag);
629 iph = iter->iph;
630 iph->tot_len = htons(frag->len);
631 ip_copy_metadata(frag, skb);
632 iter->offset += skb->len - hlen;
633 iph->frag_off = htons(iter->offset >> 3);
634 if (frag->next)
635 iph->frag_off |= htons(IP_MF);
636 /* Ready, complete checksum */
637 ip_send_check(iph);
638 }
639 EXPORT_SYMBOL(ip_fraglist_prepare);
640
641 void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
642 unsigned int ll_rs, unsigned int mtu, bool DF,
643 struct ip_frag_state *state)
644 {
645 struct iphdr *iph = ip_hdr(skb);
646
647 state->DF = DF;
648 state->hlen = hlen;
649 state->ll_rs = ll_rs;
650 state->mtu = mtu;
651
652 state->left = skb->len - hlen; /* Space per frame */
653 state->ptr = hlen; /* Where to start from */
654
655 state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
656 state->not_last_frag = iph->frag_off & htons(IP_MF);
657 }
658 EXPORT_SYMBOL(ip_frag_init);
659
660 static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
661 bool first_frag)
662 {
663 /* Copy the flags to each fragment. */
664 IPCB(to)->flags = IPCB(from)->flags;
665
666 /* ANK: dirty, but effective trick. Upgrade options only if
667 * the segment to be fragmented was THE FIRST (otherwise,
668 * options are already fixed) and make it ONCE
669 * on the initial skb, so that all the following fragments
670 * will inherit fixed options.
671 */
672 if (first_frag)
673 ip_options_fragment(from);
674 }
675
676 struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
677 {
678 unsigned int len = state->left;
679 struct sk_buff *skb2;
680 struct iphdr *iph;
681
682 len = state->left;
683 /* IF: it doesn't fit, use 'mtu' - the data space left */
684 if (len > state->mtu)
685 len = state->mtu;
686 /* IF: we are not sending up to and including the packet end
687 then align the next start on an eight byte boundary */
688 if (len < state->left) {
689 len &= ~7;
690 }
691
692 /* Allocate buffer */
693 skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
694 if (!skb2)
695 return ERR_PTR(-ENOMEM);
696
697 /*
698 * Set up data on packet
699 */
700
701 ip_copy_metadata(skb2, skb);
702 skb_reserve(skb2, state->ll_rs);
703 skb_put(skb2, len + state->hlen);
704 skb_reset_network_header(skb2);
705 skb2->transport_header = skb2->network_header + state->hlen;
706
707 /*
708 * Charge the memory for the fragment to any owner
709 * it might possess
710 */
711
712 if (skb->sk)
713 skb_set_owner_w(skb2, skb->sk);
714
715 /*
716 * Copy the packet header into the new buffer.
717 */
718
719 skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
720
721 /*
722 * Copy a block of the IP datagram.
723 */
724 if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
725 BUG();
726 state->left -= len;
727
728 /*
729 * Fill in the new header fields.
730 */
731 iph = ip_hdr(skb2);
732 iph->frag_off = htons((state->offset >> 3));
733 if (state->DF)
734 iph->frag_off |= htons(IP_DF);
735
736 /*
737 * Added AC : If we are fragmenting a fragment that's not the
738 * last fragment then keep MF on each bit
739 */
740 if (state->left > 0 || state->not_last_frag)
741 iph->frag_off |= htons(IP_MF);
742 state->ptr += len;
743 state->offset += len;
744
745 iph->tot_len = htons(len + state->hlen);
746
747 ip_send_check(iph);
748
749 return skb2;
750 }
751 EXPORT_SYMBOL(ip_frag_next);
752
753 /*
754 * This IP datagram is too large to be sent in one piece. Break it up into
755 * smaller pieces (each of size equal to IP header plus
756 * a block of the data of the original IP data part) that will yet fit in a
757 * single device frame, and queue such a frame for sending.
758 */
759
760 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
761 int (*output)(struct net *, struct sock *, struct sk_buff *))
762 {
763 struct iphdr *iph;
764 struct sk_buff *skb2;
765 struct rtable *rt = skb_rtable(skb);
766 unsigned int mtu, hlen, ll_rs;
767 struct ip_fraglist_iter iter;
768 ktime_t tstamp = skb->tstamp;
769 struct ip_frag_state state;
770 int err = 0;
771
772 /* for offloaded checksums cleanup checksum before fragmentation */
773 if (skb->ip_summed == CHECKSUM_PARTIAL &&
774 (err = skb_checksum_help(skb)))
775 goto fail;
776
777 /*
778 * Point into the IP datagram header.
779 */
780
781 iph = ip_hdr(skb);
782
783 mtu = ip_skb_dst_mtu(sk, skb);
784 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
785 mtu = IPCB(skb)->frag_max_size;
786
787 /*
788 * Setup starting values.
789 */
790
791 hlen = iph->ihl * 4;
792 mtu = mtu - hlen; /* Size of data space */
793 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
794 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
795
796 /* When frag_list is given, use it. First, check its validity:
797 * some transformers could create wrong frag_list or break existing
798 * one, it is not prohibited. In this case fall back to copying.
799 *
800 * LATER: this step can be merged to real generation of fragments,
801 * we can switch to copy when see the first bad fragment.
802 */
803 if (skb_has_frag_list(skb)) {
804 struct sk_buff *frag, *frag2;
805 unsigned int first_len = skb_pagelen(skb);
806
807 if (first_len - hlen > mtu ||
808 ((first_len - hlen) & 7) ||
809 ip_is_fragment(iph) ||
810 skb_cloned(skb) ||
811 skb_headroom(skb) < ll_rs)
812 goto slow_path;
813
814 skb_walk_frags(skb, frag) {
815 /* Correct geometry. */
816 if (frag->len > mtu ||
817 ((frag->len & 7) && frag->next) ||
818 skb_headroom(frag) < hlen + ll_rs)
819 goto slow_path_clean;
820
821 /* Partially cloned skb? */
822 if (skb_shared(frag))
823 goto slow_path_clean;
824
825 BUG_ON(frag->sk);
826 if (skb->sk) {
827 frag->sk = skb->sk;
828 frag->destructor = sock_wfree;
829 }
830 skb->truesize -= frag->truesize;
831 }
832
833 /* Everything is OK. Generate! */
834 ip_fraglist_init(skb, iph, hlen, &iter);
835
836 for (;;) {
837 /* Prepare header of the next frame,
838 * before previous one went down. */
839 if (iter.frag) {
840 bool first_frag = (iter.offset == 0);
841
842 IPCB(iter.frag)->flags = IPCB(skb)->flags;
843 ip_fraglist_prepare(skb, &iter);
844 if (first_frag && IPCB(skb)->opt.optlen) {
845 /* ipcb->opt is not populated for frags
846 * coming from __ip_make_skb(),
847 * ip_options_fragment() needs optlen
848 */
849 IPCB(iter.frag)->opt.optlen =
850 IPCB(skb)->opt.optlen;
851 ip_options_fragment(iter.frag);
852 ip_send_check(iter.iph);
853 }
854 }
855
856 skb->tstamp = tstamp;
857 err = output(net, sk, skb);
858
859 if (!err)
860 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
861 if (err || !iter.frag)
862 break;
863
864 skb = ip_fraglist_next(&iter);
865 }
866
867 if (err == 0) {
868 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
869 return 0;
870 }
871
872 kfree_skb_list(iter.frag);
873
874 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
875 return err;
876
877 slow_path_clean:
878 skb_walk_frags(skb, frag2) {
879 if (frag2 == frag)
880 break;
881 frag2->sk = NULL;
882 frag2->destructor = NULL;
883 skb->truesize += frag2->truesize;
884 }
885 }
886
887 slow_path:
888 /*
889 * Fragment the datagram.
890 */
891
892 ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
893 &state);
894
895 /*
896 * Keep copying data until we run out.
897 */
898
899 while (state.left > 0) {
900 bool first_frag = (state.offset == 0);
901
902 skb2 = ip_frag_next(skb, &state);
903 if (IS_ERR(skb2)) {
904 err = PTR_ERR(skb2);
905 goto fail;
906 }
907 ip_frag_ipcb(skb, skb2, first_frag);
908
909 /*
910 * Put this fragment into the sending queue.
911 */
912 skb2->tstamp = tstamp;
913 err = output(net, sk, skb2);
914 if (err)
915 goto fail;
916
917 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
918 }
919 consume_skb(skb);
920 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
921 return err;
922
923 fail:
924 kfree_skb(skb);
925 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
926 return err;
927 }
928 EXPORT_SYMBOL(ip_do_fragment);
929
930 int
931 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
932 {
933 struct msghdr *msg = from;
934
935 if (skb->ip_summed == CHECKSUM_PARTIAL) {
936 if (!copy_from_iter_full(to, len, &msg->msg_iter))
937 return -EFAULT;
938 } else {
939 __wsum csum = 0;
940 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
941 return -EFAULT;
942 skb->csum = csum_block_add(skb->csum, csum, odd);
943 }
944 return 0;
945 }
946 EXPORT_SYMBOL(ip_generic_getfrag);
947
948 static inline __wsum
949 csum_page(struct page *page, int offset, int copy)
950 {
951 char *kaddr;
952 __wsum csum;
953 kaddr = kmap(page);
954 csum = csum_partial(kaddr + offset, copy, 0);
955 kunmap(page);
956 return csum;
957 }
958
959 static int __ip_append_data(struct sock *sk,
960 struct flowi4 *fl4,
961 struct sk_buff_head *queue,
962 struct inet_cork *cork,
963 struct page_frag *pfrag,
964 int getfrag(void *from, char *to, int offset,
965 int len, int odd, struct sk_buff *skb),
966 void *from, int length, int transhdrlen,
967 unsigned int flags)
968 {
969 struct inet_sock *inet = inet_sk(sk);
970 struct ubuf_info *uarg = NULL;
971 struct sk_buff *skb;
972
973 struct ip_options *opt = cork->opt;
974 int hh_len;
975 int exthdrlen;
976 int mtu;
977 int copy;
978 int err;
979 int offset = 0;
980 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
981 int csummode = CHECKSUM_NONE;
982 struct rtable *rt = (struct rtable *)cork->dst;
983 unsigned int wmem_alloc_delta = 0;
984 bool paged, extra_uref = false;
985 u32 tskey = 0;
986
987 skb = skb_peek_tail(queue);
988
989 exthdrlen = !skb ? rt->dst.header_len : 0;
990 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
991 paged = !!cork->gso_size;
992
993 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
994 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
995 tskey = sk->sk_tskey++;
996
997 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
998
999 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1000 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1001 maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
1002
1003 if (cork->length + length > maxnonfragsize - fragheaderlen) {
1004 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1005 mtu - (opt ? opt->optlen : 0));
1006 return -EMSGSIZE;
1007 }
1008
1009 /*
1010 * transhdrlen > 0 means that this is the first fragment and we wish
1011 * it won't be fragmented in the future.
1012 */
1013 if (transhdrlen &&
1014 length + fragheaderlen <= mtu &&
1015 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1016 (!(flags & MSG_MORE) || cork->gso_size) &&
1017 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1018 csummode = CHECKSUM_PARTIAL;
1019
1020 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1021 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1022 if (!uarg)
1023 return -ENOBUFS;
1024 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1025 if (rt->dst.dev->features & NETIF_F_SG &&
1026 csummode == CHECKSUM_PARTIAL) {
1027 paged = true;
1028 } else {
1029 uarg->zerocopy = 0;
1030 skb_zcopy_set(skb, uarg, &extra_uref);
1031 }
1032 }
1033
1034 cork->length += length;
1035
1036 /* So, what's going on in the loop below?
1037 *
1038 * We use calculated fragment length to generate chained skb,
1039 * each of segments is IP fragment ready for sending to network after
1040 * adding appropriate IP header.
1041 */
1042
1043 if (!skb)
1044 goto alloc_new_skb;
1045
1046 while (length > 0) {
1047 /* Check if the remaining data fits into current packet. */
1048 copy = mtu - skb->len;
1049 if (copy < length)
1050 copy = maxfraglen - skb->len;
1051 if (copy <= 0) {
1052 char *data;
1053 unsigned int datalen;
1054 unsigned int fraglen;
1055 unsigned int fraggap;
1056 unsigned int alloclen, alloc_extra;
1057 unsigned int pagedlen;
1058 struct sk_buff *skb_prev;
1059 alloc_new_skb:
1060 skb_prev = skb;
1061 if (skb_prev)
1062 fraggap = skb_prev->len - maxfraglen;
1063 else
1064 fraggap = 0;
1065
1066 /*
1067 * If remaining data exceeds the mtu,
1068 * we know we need more fragment(s).
1069 */
1070 datalen = length + fraggap;
1071 if (datalen > mtu - fragheaderlen)
1072 datalen = maxfraglen - fragheaderlen;
1073 fraglen = datalen + fragheaderlen;
1074 pagedlen = 0;
1075
1076 alloc_extra = hh_len + 15;
1077 alloc_extra += exthdrlen;
1078
1079 /* The last fragment gets additional space at tail.
1080 * Note, with MSG_MORE we overallocate on fragments,
1081 * because we have no idea what fragment will be
1082 * the last.
1083 */
1084 if (datalen == length + fraggap)
1085 alloc_extra += rt->dst.trailer_len;
1086
1087 if ((flags & MSG_MORE) &&
1088 !(rt->dst.dev->features&NETIF_F_SG))
1089 alloclen = mtu;
1090 else if (!paged &&
1091 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1092 !(rt->dst.dev->features & NETIF_F_SG)))
1093 alloclen = fraglen;
1094 else {
1095 alloclen = min_t(int, fraglen, MAX_HEADER);
1096 pagedlen = fraglen - alloclen;
1097 }
1098
1099 alloclen += alloc_extra;
1100
1101 if (transhdrlen) {
1102 skb = sock_alloc_send_skb(sk, alloclen,
1103 (flags & MSG_DONTWAIT), &err);
1104 } else {
1105 skb = NULL;
1106 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1107 2 * sk->sk_sndbuf)
1108 skb = alloc_skb(alloclen,
1109 sk->sk_allocation);
1110 if (unlikely(!skb))
1111 err = -ENOBUFS;
1112 }
1113 if (!skb)
1114 goto error;
1115
1116 /*
1117 * Fill in the control structures
1118 */
1119 skb->ip_summed = csummode;
1120 skb->csum = 0;
1121 skb_reserve(skb, hh_len);
1122
1123 /*
1124 * Find where to start putting bytes.
1125 */
1126 data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1127 skb_set_network_header(skb, exthdrlen);
1128 skb->transport_header = (skb->network_header +
1129 fragheaderlen);
1130 data += fragheaderlen + exthdrlen;
1131
1132 if (fraggap) {
1133 skb->csum = skb_copy_and_csum_bits(
1134 skb_prev, maxfraglen,
1135 data + transhdrlen, fraggap);
1136 skb_prev->csum = csum_sub(skb_prev->csum,
1137 skb->csum);
1138 data += fraggap;
1139 pskb_trim_unique(skb_prev, maxfraglen);
1140 }
1141
1142 copy = datalen - transhdrlen - fraggap - pagedlen;
1143 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1144 err = -EFAULT;
1145 kfree_skb(skb);
1146 goto error;
1147 }
1148
1149 offset += copy;
1150 length -= copy + transhdrlen;
1151 transhdrlen = 0;
1152 exthdrlen = 0;
1153 csummode = CHECKSUM_NONE;
1154
1155 /* only the initial fragment is time stamped */
1156 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1157 cork->tx_flags = 0;
1158 skb_shinfo(skb)->tskey = tskey;
1159 tskey = 0;
1160 skb_zcopy_set(skb, uarg, &extra_uref);
1161
1162 if ((flags & MSG_CONFIRM) && !skb_prev)
1163 skb_set_dst_pending_confirm(skb, 1);
1164
1165 /*
1166 * Put the packet on the pending queue.
1167 */
1168 if (!skb->destructor) {
1169 skb->destructor = sock_wfree;
1170 skb->sk = sk;
1171 wmem_alloc_delta += skb->truesize;
1172 }
1173 __skb_queue_tail(queue, skb);
1174 continue;
1175 }
1176
1177 if (copy > length)
1178 copy = length;
1179
1180 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1181 skb_tailroom(skb) >= copy) {
1182 unsigned int off;
1183
1184 off = skb->len;
1185 if (getfrag(from, skb_put(skb, copy),
1186 offset, copy, off, skb) < 0) {
1187 __skb_trim(skb, off);
1188 err = -EFAULT;
1189 goto error;
1190 }
1191 } else if (!uarg || !uarg->zerocopy) {
1192 int i = skb_shinfo(skb)->nr_frags;
1193
1194 err = -ENOMEM;
1195 if (!sk_page_frag_refill(sk, pfrag))
1196 goto error;
1197
1198 if (!skb_can_coalesce(skb, i, pfrag->page,
1199 pfrag->offset)) {
1200 err = -EMSGSIZE;
1201 if (i == MAX_SKB_FRAGS)
1202 goto error;
1203
1204 __skb_fill_page_desc(skb, i, pfrag->page,
1205 pfrag->offset, 0);
1206 skb_shinfo(skb)->nr_frags = ++i;
1207 get_page(pfrag->page);
1208 }
1209 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1210 if (getfrag(from,
1211 page_address(pfrag->page) + pfrag->offset,
1212 offset, copy, skb->len, skb) < 0)
1213 goto error_efault;
1214
1215 pfrag->offset += copy;
1216 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1217 skb->len += copy;
1218 skb->data_len += copy;
1219 skb->truesize += copy;
1220 wmem_alloc_delta += copy;
1221 } else {
1222 err = skb_zerocopy_iter_dgram(skb, from, copy);
1223 if (err < 0)
1224 goto error;
1225 }
1226 offset += copy;
1227 length -= copy;
1228 }
1229
1230 if (wmem_alloc_delta)
1231 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1232 return 0;
1233
1234 error_efault:
1235 err = -EFAULT;
1236 error:
1237 net_zcopy_put_abort(uarg, extra_uref);
1238 cork->length -= length;
1239 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1240 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1241 return err;
1242 }
1243
1244 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1245 struct ipcm_cookie *ipc, struct rtable **rtp)
1246 {
1247 struct ip_options_rcu *opt;
1248 struct rtable *rt;
1249
1250 rt = *rtp;
1251 if (unlikely(!rt))
1252 return -EFAULT;
1253
1254 /*
1255 * setup for corking.
1256 */
1257 opt = ipc->opt;
1258 if (opt) {
1259 if (!cork->opt) {
1260 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1261 sk->sk_allocation);
1262 if (unlikely(!cork->opt))
1263 return -ENOBUFS;
1264 }
1265 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1266 cork->flags |= IPCORK_OPT;
1267 cork->addr = ipc->addr;
1268 }
1269
1270 cork->fragsize = ip_sk_use_pmtu(sk) ?
1271 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1272
1273 if (!inetdev_valid_mtu(cork->fragsize))
1274 return -ENETUNREACH;
1275
1276 cork->gso_size = ipc->gso_size;
1277
1278 cork->dst = &rt->dst;
1279 /* We stole this route, caller should not release it. */
1280 *rtp = NULL;
1281
1282 cork->length = 0;
1283 cork->ttl = ipc->ttl;
1284 cork->tos = ipc->tos;
1285 cork->mark = ipc->sockc.mark;
1286 cork->priority = ipc->priority;
1287 cork->transmit_time = ipc->sockc.transmit_time;
1288 cork->tx_flags = 0;
1289 sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags);
1290
1291 return 0;
1292 }
1293
1294 /*
1295 * ip_append_data() and ip_append_page() can make one large IP datagram
1296 * from many pieces of data. Each pieces will be holded on the socket
1297 * until ip_push_pending_frames() is called. Each piece can be a page
1298 * or non-page data.
1299 *
1300 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1301 * this interface potentially.
1302 *
1303 * LATER: length must be adjusted by pad at tail, when it is required.
1304 */
1305 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1306 int getfrag(void *from, char *to, int offset, int len,
1307 int odd, struct sk_buff *skb),
1308 void *from, int length, int transhdrlen,
1309 struct ipcm_cookie *ipc, struct rtable **rtp,
1310 unsigned int flags)
1311 {
1312 struct inet_sock *inet = inet_sk(sk);
1313 int err;
1314
1315 if (flags&MSG_PROBE)
1316 return 0;
1317
1318 if (skb_queue_empty(&sk->sk_write_queue)) {
1319 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1320 if (err)
1321 return err;
1322 } else {
1323 transhdrlen = 0;
1324 }
1325
1326 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1327 sk_page_frag(sk), getfrag,
1328 from, length, transhdrlen, flags);
1329 }
1330
1331 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1332 int offset, size_t size, int flags)
1333 {
1334 struct inet_sock *inet = inet_sk(sk);
1335 struct sk_buff *skb;
1336 struct rtable *rt;
1337 struct ip_options *opt = NULL;
1338 struct inet_cork *cork;
1339 int hh_len;
1340 int mtu;
1341 int len;
1342 int err;
1343 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1344
1345 if (inet->hdrincl)
1346 return -EPERM;
1347
1348 if (flags&MSG_PROBE)
1349 return 0;
1350
1351 if (skb_queue_empty(&sk->sk_write_queue))
1352 return -EINVAL;
1353
1354 cork = &inet->cork.base;
1355 rt = (struct rtable *)cork->dst;
1356 if (cork->flags & IPCORK_OPT)
1357 opt = cork->opt;
1358
1359 if (!(rt->dst.dev->features & NETIF_F_SG))
1360 return -EOPNOTSUPP;
1361
1362 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1363 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
1364
1365 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1366 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1367 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1368
1369 if (cork->length + size > maxnonfragsize - fragheaderlen) {
1370 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1371 mtu - (opt ? opt->optlen : 0));
1372 return -EMSGSIZE;
1373 }
1374
1375 skb = skb_peek_tail(&sk->sk_write_queue);
1376 if (!skb)
1377 return -EINVAL;
1378
1379 cork->length += size;
1380
1381 while (size > 0) {
1382 /* Check if the remaining data fits into current packet. */
1383 len = mtu - skb->len;
1384 if (len < size)
1385 len = maxfraglen - skb->len;
1386
1387 if (len <= 0) {
1388 struct sk_buff *skb_prev;
1389 int alloclen;
1390
1391 skb_prev = skb;
1392 fraggap = skb_prev->len - maxfraglen;
1393
1394 alloclen = fragheaderlen + hh_len + fraggap + 15;
1395 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1396 if (unlikely(!skb)) {
1397 err = -ENOBUFS;
1398 goto error;
1399 }
1400
1401 /*
1402 * Fill in the control structures
1403 */
1404 skb->ip_summed = CHECKSUM_NONE;
1405 skb->csum = 0;
1406 skb_reserve(skb, hh_len);
1407
1408 /*
1409 * Find where to start putting bytes.
1410 */
1411 skb_put(skb, fragheaderlen + fraggap);
1412 skb_reset_network_header(skb);
1413 skb->transport_header = (skb->network_header +
1414 fragheaderlen);
1415 if (fraggap) {
1416 skb->csum = skb_copy_and_csum_bits(skb_prev,
1417 maxfraglen,
1418 skb_transport_header(skb),
1419 fraggap);
1420 skb_prev->csum = csum_sub(skb_prev->csum,
1421 skb->csum);
1422 pskb_trim_unique(skb_prev, maxfraglen);
1423 }
1424
1425 /*
1426 * Put the packet on the pending queue.
1427 */
1428 __skb_queue_tail(&sk->sk_write_queue, skb);
1429 continue;
1430 }
1431
1432 if (len > size)
1433 len = size;
1434
1435 if (skb_append_pagefrags(skb, page, offset, len)) {
1436 err = -EMSGSIZE;
1437 goto error;
1438 }
1439
1440 if (skb->ip_summed == CHECKSUM_NONE) {
1441 __wsum csum;
1442 csum = csum_page(page, offset, len);
1443 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1444 }
1445
1446 skb->len += len;
1447 skb->data_len += len;
1448 skb->truesize += len;
1449 refcount_add(len, &sk->sk_wmem_alloc);
1450 offset += len;
1451 size -= len;
1452 }
1453 return 0;
1454
1455 error:
1456 cork->length -= size;
1457 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1458 return err;
1459 }
1460
1461 static void ip_cork_release(struct inet_cork *cork)
1462 {
1463 cork->flags &= ~IPCORK_OPT;
1464 kfree(cork->opt);
1465 cork->opt = NULL;
1466 dst_release(cork->dst);
1467 cork->dst = NULL;
1468 }
1469
1470 /*
1471 * Combined all pending IP fragments on the socket as one IP datagram
1472 * and push them out.
1473 */
1474 struct sk_buff *__ip_make_skb(struct sock *sk,
1475 struct flowi4 *fl4,
1476 struct sk_buff_head *queue,
1477 struct inet_cork *cork)
1478 {
1479 struct sk_buff *skb, *tmp_skb;
1480 struct sk_buff **tail_skb;
1481 struct inet_sock *inet = inet_sk(sk);
1482 struct net *net = sock_net(sk);
1483 struct ip_options *opt = NULL;
1484 struct rtable *rt = (struct rtable *)cork->dst;
1485 struct iphdr *iph;
1486 __be16 df = 0;
1487 __u8 ttl;
1488
1489 skb = __skb_dequeue(queue);
1490 if (!skb)
1491 goto out;
1492 tail_skb = &(skb_shinfo(skb)->frag_list);
1493
1494 /* move skb->data to ip header from ext header */
1495 if (skb->data < skb_network_header(skb))
1496 __skb_pull(skb, skb_network_offset(skb));
1497 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1498 __skb_pull(tmp_skb, skb_network_header_len(skb));
1499 *tail_skb = tmp_skb;
1500 tail_skb = &(tmp_skb->next);
1501 skb->len += tmp_skb->len;
1502 skb->data_len += tmp_skb->len;
1503 skb->truesize += tmp_skb->truesize;
1504 tmp_skb->destructor = NULL;
1505 tmp_skb->sk = NULL;
1506 }
1507
1508 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1509 * to fragment the frame generated here. No matter, what transforms
1510 * how transforms change size of the packet, it will come out.
1511 */
1512 skb->ignore_df = ip_sk_ignore_df(sk);
1513
1514 /* DF bit is set when we want to see DF on outgoing frames.
1515 * If ignore_df is set too, we still allow to fragment this frame
1516 * locally. */
1517 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1518 inet->pmtudisc == IP_PMTUDISC_PROBE ||
1519 (skb->len <= dst_mtu(&rt->dst) &&
1520 ip_dont_fragment(sk, &rt->dst)))
1521 df = htons(IP_DF);
1522
1523 if (cork->flags & IPCORK_OPT)
1524 opt = cork->opt;
1525
1526 if (cork->ttl != 0)
1527 ttl = cork->ttl;
1528 else if (rt->rt_type == RTN_MULTICAST)
1529 ttl = inet->mc_ttl;
1530 else
1531 ttl = ip_select_ttl(inet, &rt->dst);
1532
1533 iph = ip_hdr(skb);
1534 iph->version = 4;
1535 iph->ihl = 5;
1536 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1537 iph->frag_off = df;
1538 iph->ttl = ttl;
1539 iph->protocol = sk->sk_protocol;
1540 ip_copy_addrs(iph, fl4);
1541 ip_select_ident(net, skb, sk);
1542
1543 if (opt) {
1544 iph->ihl += opt->optlen >> 2;
1545 ip_options_build(skb, opt, cork->addr, rt, 0);
1546 }
1547
1548 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1549 skb->mark = cork->mark;
1550 skb->tstamp = cork->transmit_time;
1551 /*
1552 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1553 * on dst refcount
1554 */
1555 cork->dst = NULL;
1556 skb_dst_set(skb, &rt->dst);
1557
1558 if (iph->protocol == IPPROTO_ICMP)
1559 icmp_out_count(net, ((struct icmphdr *)
1560 skb_transport_header(skb))->type);
1561
1562 ip_cork_release(cork);
1563 out:
1564 return skb;
1565 }
1566
1567 int ip_send_skb(struct net *net, struct sk_buff *skb)
1568 {
1569 int err;
1570
1571 err = ip_local_out(net, skb->sk, skb);
1572 if (err) {
1573 if (err > 0)
1574 err = net_xmit_errno(err);
1575 if (err)
1576 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1577 }
1578
1579 return err;
1580 }
1581
1582 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1583 {
1584 struct sk_buff *skb;
1585
1586 skb = ip_finish_skb(sk, fl4);
1587 if (!skb)
1588 return 0;
1589
1590 /* Netfilter gets whole the not fragmented skb. */
1591 return ip_send_skb(sock_net(sk), skb);
1592 }
1593
1594 /*
1595 * Throw away all pending data on the socket.
1596 */
1597 static void __ip_flush_pending_frames(struct sock *sk,
1598 struct sk_buff_head *queue,
1599 struct inet_cork *cork)
1600 {
1601 struct sk_buff *skb;
1602
1603 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1604 kfree_skb(skb);
1605
1606 ip_cork_release(cork);
1607 }
1608
1609 void ip_flush_pending_frames(struct sock *sk)
1610 {
1611 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1612 }
1613
1614 struct sk_buff *ip_make_skb(struct sock *sk,
1615 struct flowi4 *fl4,
1616 int getfrag(void *from, char *to, int offset,
1617 int len, int odd, struct sk_buff *skb),
1618 void *from, int length, int transhdrlen,
1619 struct ipcm_cookie *ipc, struct rtable **rtp,
1620 struct inet_cork *cork, unsigned int flags)
1621 {
1622 struct sk_buff_head queue;
1623 int err;
1624
1625 if (flags & MSG_PROBE)
1626 return NULL;
1627
1628 __skb_queue_head_init(&queue);
1629
1630 cork->flags = 0;
1631 cork->addr = 0;
1632 cork->opt = NULL;
1633 err = ip_setup_cork(sk, cork, ipc, rtp);
1634 if (err)
1635 return ERR_PTR(err);
1636
1637 err = __ip_append_data(sk, fl4, &queue, cork,
1638 &current->task_frag, getfrag,
1639 from, length, transhdrlen, flags);
1640 if (err) {
1641 __ip_flush_pending_frames(sk, &queue, cork);
1642 return ERR_PTR(err);
1643 }
1644
1645 return __ip_make_skb(sk, fl4, &queue, cork);
1646 }
1647
1648 /*
1649 * Fetch data from kernel space and fill in checksum if needed.
1650 */
1651 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1652 int len, int odd, struct sk_buff *skb)
1653 {
1654 __wsum csum;
1655
1656 csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1657 skb->csum = csum_block_add(skb->csum, csum, odd);
1658 return 0;
1659 }
1660
1661 /*
1662 * Generic function to send a packet as reply to another packet.
1663 * Used to send some TCP resets/acks so far.
1664 */
1665 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1666 const struct ip_options *sopt,
1667 __be32 daddr, __be32 saddr,
1668 const struct ip_reply_arg *arg,
1669 unsigned int len, u64 transmit_time)
1670 {
1671 struct ip_options_data replyopts;
1672 struct ipcm_cookie ipc;
1673 struct flowi4 fl4;
1674 struct rtable *rt = skb_rtable(skb);
1675 struct net *net = sock_net(sk);
1676 struct sk_buff *nskb;
1677 int err;
1678 int oif;
1679
1680 if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
1681 return;
1682
1683 ipcm_init(&ipc);
1684 ipc.addr = daddr;
1685 ipc.sockc.transmit_time = transmit_time;
1686
1687 if (replyopts.opt.opt.optlen) {
1688 ipc.opt = &replyopts.opt;
1689
1690 if (replyopts.opt.opt.srr)
1691 daddr = replyopts.opt.opt.faddr;
1692 }
1693
1694 oif = arg->bound_dev_if;
1695 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1696 oif = skb->skb_iif;
1697
1698 flowi4_init_output(&fl4, oif,
1699 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1700 RT_TOS(arg->tos),
1701 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1702 ip_reply_arg_flowi_flags(arg),
1703 daddr, saddr,
1704 tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1705 arg->uid);
1706 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1707 rt = ip_route_output_key(net, &fl4);
1708 if (IS_ERR(rt))
1709 return;
1710
1711 inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1712
1713 sk->sk_protocol = ip_hdr(skb)->protocol;
1714 sk->sk_bound_dev_if = arg->bound_dev_if;
1715 sk->sk_sndbuf = sysctl_wmem_default;
1716 ipc.sockc.mark = fl4.flowi4_mark;
1717 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1718 len, 0, &ipc, &rt, MSG_DONTWAIT);
1719 if (unlikely(err)) {
1720 ip_flush_pending_frames(sk);
1721 goto out;
1722 }
1723
1724 nskb = skb_peek(&sk->sk_write_queue);
1725 if (nskb) {
1726 if (arg->csumoffset >= 0)
1727 *((__sum16 *)skb_transport_header(nskb) +
1728 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1729 arg->csum));
1730 nskb->ip_summed = CHECKSUM_NONE;
1731 ip_push_pending_frames(sk, &fl4);
1732 }
1733 out:
1734 ip_rt_put(rt);
1735 }
1736
1737 void __init ip_init(void)
1738 {
1739 ip_rt_init();
1740 inet_initpeers();
1741
1742 #if defined(CONFIG_IP_MULTICAST)
1743 igmp_mc_init();
1744 #endif
1745 }
Ubuntu Jammy Linux kernel mirror