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