2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The Internet Protocol (IP) output module.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 * Hirokazu Takahashi, <taka@valinux.co.jp>
18 * See ip_input.c for original log
21 * Alan Cox : Missing nonblock feature in ip_build_xmit.
22 * Mike Kilburn : htons() missing in ip_build_xmit.
23 * Bradford Johnson: Fix faulty handling of some frames when
25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
26 * (in case if packet not accepted by
27 * output firewall rules)
28 * Mike McLagan : Routing by source
29 * Alexey Kuznetsov: use new route cache
30 * Andi Kleen: Fix broken PMTU recovery and remove
31 * some redundant tests.
32 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
33 * Andi Kleen : Replace ip_reply with ip_send_reply.
34 * Andi Kleen : Split fast and slow ip_build_xmit path
35 * for decreased register pressure on x86
36 * and more readibility.
37 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
38 * silently drop skb instead of failing with -EPERM.
39 * Detlev Wengorz : Copy protocol for fragments.
40 * Hirokazu Takahashi: HW checksumming for outgoing UDP
42 * Hirokazu Takahashi: sendfile() on UDP works now.
45 #include <asm/uaccess.h>
46 #include <linux/module.h>
47 #include <linux/types.h>
48 #include <linux/kernel.h>
50 #include <linux/string.h>
51 #include <linux/errno.h>
52 #include <linux/highmem.h>
53 #include <linux/slab.h>
55 #include <linux/socket.h>
56 #include <linux/sockios.h>
58 #include <linux/inet.h>
59 #include <linux/netdevice.h>
60 #include <linux/etherdevice.h>
61 #include <linux/proc_fs.h>
62 #include <linux/stat.h>
63 #include <linux/init.h>
67 #include <net/protocol.h>
68 #include <net/route.h>
70 #include <linux/skbuff.h>
74 #include <net/checksum.h>
75 #include <net/inetpeer.h>
76 #include <linux/igmp.h>
77 #include <linux/netfilter_ipv4.h>
78 #include <linux/netfilter_bridge.h>
79 #include <linux/mroute.h>
80 #include <linux/netlink.h>
81 #include <linux/tcp.h>
83 int sysctl_ip_default_ttl __read_mostly
= IPDEFTTL
;
84 EXPORT_SYMBOL(sysctl_ip_default_ttl
);
87 ip_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
89 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*));
91 /* Generate a checksum for an outgoing IP datagram. */
92 void ip_send_check(struct iphdr
*iph
)
95 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
97 EXPORT_SYMBOL(ip_send_check
);
99 int __ip_local_out(struct sock
*sk
, struct sk_buff
*skb
)
101 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
102 struct iphdr
*iph
= ip_hdr(skb
);
104 iph
->tot_len
= htons(skb
->len
);
106 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
,
107 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
111 int ip_local_out(struct sock
*sk
, struct sk_buff
*skb
)
113 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
116 err
= __ip_local_out(sk
, skb
);
117 if (likely(err
== 1))
118 err
= dst_output(net
, sk
, skb
);
122 EXPORT_SYMBOL_GPL(ip_local_out
);
124 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
126 int ttl
= inet
->uc_ttl
;
129 ttl
= ip4_dst_hoplimit(dst
);
134 * Add an ip header to a skbuff and send it out.
137 int ip_build_and_send_pkt(struct sk_buff
*skb
, const struct sock
*sk
,
138 __be32 saddr
, __be32 daddr
, struct ip_options_rcu
*opt
)
140 struct inet_sock
*inet
= inet_sk(sk
);
141 struct rtable
*rt
= skb_rtable(skb
);
144 /* Build the IP header. */
145 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->opt
.optlen
: 0));
146 skb_reset_network_header(skb
);
150 iph
->tos
= inet
->tos
;
151 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
152 iph
->daddr
= (opt
&& opt
->opt
.srr
? opt
->opt
.faddr
: daddr
);
154 iph
->protocol
= sk
->sk_protocol
;
155 if (ip_dont_fragment(sk
, &rt
->dst
)) {
156 iph
->frag_off
= htons(IP_DF
);
160 __ip_select_ident(sock_net(sk
), iph
, 1);
163 if (opt
&& opt
->opt
.optlen
) {
164 iph
->ihl
+= opt
->opt
.optlen
>>2;
165 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
168 skb
->priority
= sk
->sk_priority
;
169 skb
->mark
= sk
->sk_mark
;
172 return ip_local_out(skb
->sk
, skb
);
174 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
176 static int ip_finish_output2(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
178 struct dst_entry
*dst
= skb_dst(skb
);
179 struct rtable
*rt
= (struct rtable
*)dst
;
180 struct net_device
*dev
= dst
->dev
;
181 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
182 struct neighbour
*neigh
;
185 if (rt
->rt_type
== RTN_MULTICAST
) {
186 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTMCAST
, skb
->len
);
187 } else if (rt
->rt_type
== RTN_BROADCAST
)
188 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTBCAST
, skb
->len
);
190 /* Be paranoid, rather than too clever. */
191 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
192 struct sk_buff
*skb2
;
194 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
200 skb_set_owner_w(skb2
, skb
->sk
);
206 nexthop
= (__force u32
) rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
207 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
208 if (unlikely(!neigh
))
209 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
210 if (!IS_ERR(neigh
)) {
211 int res
= dst_neigh_output(dst
, neigh
, skb
);
213 rcu_read_unlock_bh();
216 rcu_read_unlock_bh();
218 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
224 static int ip_finish_output_gso(struct net
*net
, struct sock
*sk
,
225 struct sk_buff
*skb
, unsigned int mtu
)
227 netdev_features_t features
;
228 struct sk_buff
*segs
;
231 /* common case: locally created skb or seglen is <= mtu */
232 if (((IPCB(skb
)->flags
& IPSKB_FORWARDED
) == 0) ||
233 skb_gso_network_seglen(skb
) <= mtu
)
234 return ip_finish_output2(net
, sk
, skb
);
236 /* Slowpath - GSO segment length is exceeding the dst MTU.
238 * This can happen in two cases:
239 * 1) TCP GRO packet, DF bit not set
240 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
241 * from host network stack.
243 features
= netif_skb_features(skb
);
244 segs
= skb_gso_segment(skb
, features
& ~NETIF_F_GSO_MASK
);
245 if (IS_ERR_OR_NULL(segs
)) {
253 struct sk_buff
*nskb
= segs
->next
;
257 err
= ip_fragment(net
, sk
, segs
, mtu
, ip_finish_output2
);
267 static int ip_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
271 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
272 /* Policy lookup after SNAT yielded a new policy */
273 if (skb_dst(skb
)->xfrm
) {
274 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
275 return dst_output(net
, sk
, skb
);
278 mtu
= ip_skb_dst_mtu(skb
);
280 return ip_finish_output_gso(net
, sk
, skb
, mtu
);
282 if (skb
->len
> mtu
|| (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
))
283 return ip_fragment(net
, sk
, skb
, mtu
, ip_finish_output2
);
285 return ip_finish_output2(net
, sk
, skb
);
288 int ip_mc_output(struct sock
*sk
, struct sk_buff
*skb
)
290 struct rtable
*rt
= skb_rtable(skb
);
291 struct net_device
*dev
= rt
->dst
.dev
;
292 struct net
*net
= dev_net(dev
);
295 * If the indicated interface is up and running, send the packet.
297 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
300 skb
->protocol
= htons(ETH_P_IP
);
303 * Multicasts are looped back for other local users
306 if (rt
->rt_flags
&RTCF_MULTICAST
) {
308 #ifdef CONFIG_IP_MROUTE
309 /* Small optimization: do not loopback not local frames,
310 which returned after forwarding; they will be dropped
311 by ip_mr_input in any case.
312 Note, that local frames are looped back to be delivered
315 This check is duplicated in ip_mr_input at the moment.
318 ((rt
->rt_flags
& RTCF_LOCAL
) ||
319 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
322 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
324 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
325 net
, sk
, newskb
, NULL
, newskb
->dev
,
329 /* Multicasts with ttl 0 must not go beyond the host */
331 if (ip_hdr(skb
)->ttl
== 0) {
337 if (rt
->rt_flags
&RTCF_BROADCAST
) {
338 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
340 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
341 net
, sk
, newskb
, NULL
, newskb
->dev
,
345 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
346 net
, sk
, skb
, NULL
, skb
->dev
,
348 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
351 int ip_output(struct sock
*sk
, struct sk_buff
*skb
)
353 struct net_device
*dev
= skb_dst(skb
)->dev
;
354 struct net
*net
= dev_net(dev
);
356 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
359 skb
->protocol
= htons(ETH_P_IP
);
361 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
362 net
, sk
, skb
, NULL
, dev
,
364 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
368 * copy saddr and daddr, possibly using 64bit load/stores
370 * iph->saddr = fl4->saddr;
371 * iph->daddr = fl4->daddr;
373 static void ip_copy_addrs(struct iphdr
*iph
, const struct flowi4
*fl4
)
375 BUILD_BUG_ON(offsetof(typeof(*fl4
), daddr
) !=
376 offsetof(typeof(*fl4
), saddr
) + sizeof(fl4
->saddr
));
377 memcpy(&iph
->saddr
, &fl4
->saddr
,
378 sizeof(fl4
->saddr
) + sizeof(fl4
->daddr
));
381 /* Note: skb->sk can be different from sk, in case of tunnels */
382 int ip_queue_xmit(struct sock
*sk
, struct sk_buff
*skb
, struct flowi
*fl
)
384 struct inet_sock
*inet
= inet_sk(sk
);
385 struct ip_options_rcu
*inet_opt
;
391 /* Skip all of this if the packet is already routed,
392 * f.e. by something like SCTP.
395 inet_opt
= rcu_dereference(inet
->inet_opt
);
397 rt
= skb_rtable(skb
);
401 /* Make sure we can route this packet. */
402 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
406 /* Use correct destination address if we have options. */
407 daddr
= inet
->inet_daddr
;
408 if (inet_opt
&& inet_opt
->opt
.srr
)
409 daddr
= inet_opt
->opt
.faddr
;
411 /* If this fails, retransmit mechanism of transport layer will
412 * keep trying until route appears or the connection times
415 rt
= ip_route_output_ports(sock_net(sk
), fl4
, sk
,
416 daddr
, inet
->inet_saddr
,
421 sk
->sk_bound_dev_if
);
424 sk_setup_caps(sk
, &rt
->dst
);
426 skb_dst_set_noref(skb
, &rt
->dst
);
429 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_uses_gateway
)
432 /* OK, we know where to send it, allocate and build IP header. */
433 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
434 skb_reset_network_header(skb
);
436 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
437 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->ignore_df
)
438 iph
->frag_off
= htons(IP_DF
);
441 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
442 iph
->protocol
= sk
->sk_protocol
;
443 ip_copy_addrs(iph
, fl4
);
445 /* Transport layer set skb->h.foo itself. */
447 if (inet_opt
&& inet_opt
->opt
.optlen
) {
448 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
449 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
452 ip_select_ident_segs(sock_net(sk
), skb
, sk
,
453 skb_shinfo(skb
)->gso_segs
?: 1);
455 /* TODO : should we use skb->sk here instead of sk ? */
456 skb
->priority
= sk
->sk_priority
;
457 skb
->mark
= sk
->sk_mark
;
459 res
= ip_local_out(sk
, skb
);
465 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
467 return -EHOSTUNREACH
;
469 EXPORT_SYMBOL(ip_queue_xmit
);
471 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
473 to
->pkt_type
= from
->pkt_type
;
474 to
->priority
= from
->priority
;
475 to
->protocol
= from
->protocol
;
477 skb_dst_copy(to
, from
);
479 to
->mark
= from
->mark
;
481 /* Copy the flags to each fragment. */
482 IPCB(to
)->flags
= IPCB(from
)->flags
;
484 #ifdef CONFIG_NET_SCHED
485 to
->tc_index
= from
->tc_index
;
488 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
489 to
->ipvs_property
= from
->ipvs_property
;
491 skb_copy_secmark(to
, from
);
494 static int ip_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
496 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*))
498 struct iphdr
*iph
= ip_hdr(skb
);
500 if ((iph
->frag_off
& htons(IP_DF
)) == 0)
501 return ip_do_fragment(net
, sk
, skb
, output
);
503 if (unlikely(!skb
->ignore_df
||
504 (IPCB(skb
)->frag_max_size
&&
505 IPCB(skb
)->frag_max_size
> mtu
))) {
506 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
507 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
513 return ip_do_fragment(net
, sk
, skb
, output
);
517 * This IP datagram is too large to be sent in one piece. Break it up into
518 * smaller pieces (each of size equal to IP header plus
519 * a block of the data of the original IP data part) that will yet fit in a
520 * single device frame, and queue such a frame for sending.
523 int ip_do_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
524 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*))
528 struct net_device
*dev
;
529 struct sk_buff
*skb2
;
530 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
532 __be16 not_last_frag
;
533 struct rtable
*rt
= skb_rtable(skb
);
539 * Point into the IP datagram header.
544 mtu
= ip_skb_dst_mtu(skb
);
545 if (IPCB(skb
)->frag_max_size
&& IPCB(skb
)->frag_max_size
< mtu
)
546 mtu
= IPCB(skb
)->frag_max_size
;
549 * Setup starting values.
553 mtu
= mtu
- hlen
; /* Size of data space */
554 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
556 /* When frag_list is given, use it. First, check its validity:
557 * some transformers could create wrong frag_list or break existing
558 * one, it is not prohibited. In this case fall back to copying.
560 * LATER: this step can be merged to real generation of fragments,
561 * we can switch to copy when see the first bad fragment.
563 if (skb_has_frag_list(skb
)) {
564 struct sk_buff
*frag
, *frag2
;
565 int first_len
= skb_pagelen(skb
);
567 if (first_len
- hlen
> mtu
||
568 ((first_len
- hlen
) & 7) ||
569 ip_is_fragment(iph
) ||
573 skb_walk_frags(skb
, frag
) {
574 /* Correct geometry. */
575 if (frag
->len
> mtu
||
576 ((frag
->len
& 7) && frag
->next
) ||
577 skb_headroom(frag
) < hlen
)
578 goto slow_path_clean
;
580 /* Partially cloned skb? */
581 if (skb_shared(frag
))
582 goto slow_path_clean
;
587 frag
->destructor
= sock_wfree
;
589 skb
->truesize
-= frag
->truesize
;
592 /* Everything is OK. Generate! */
596 frag
= skb_shinfo(skb
)->frag_list
;
597 skb_frag_list_init(skb
);
598 skb
->data_len
= first_len
- skb_headlen(skb
);
599 skb
->len
= first_len
;
600 iph
->tot_len
= htons(first_len
);
601 iph
->frag_off
= htons(IP_MF
);
605 /* Prepare header of the next frame,
606 * before previous one went down. */
608 frag
->ip_summed
= CHECKSUM_NONE
;
609 skb_reset_transport_header(frag
);
610 __skb_push(frag
, hlen
);
611 skb_reset_network_header(frag
);
612 memcpy(skb_network_header(frag
), iph
, hlen
);
614 iph
->tot_len
= htons(frag
->len
);
615 ip_copy_metadata(frag
, skb
);
617 ip_options_fragment(frag
);
618 offset
+= skb
->len
- hlen
;
619 iph
->frag_off
= htons(offset
>>3);
621 iph
->frag_off
|= htons(IP_MF
);
622 /* Ready, complete checksum */
626 err
= output(net
, sk
, skb
);
629 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
639 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
648 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
652 skb_walk_frags(skb
, frag2
) {
656 frag2
->destructor
= NULL
;
657 skb
->truesize
+= frag2
->truesize
;
662 /* for offloaded checksums cleanup checksum before fragmentation */
663 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) && skb_checksum_help(skb
))
667 left
= skb
->len
- hlen
; /* Space per frame */
668 ptr
= hlen
; /* Where to start from */
670 ll_rs
= LL_RESERVED_SPACE(rt
->dst
.dev
);
673 * Fragment the datagram.
676 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
677 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
680 * Keep copying data until we run out.
685 /* IF: it doesn't fit, use 'mtu' - the data space left */
688 /* IF: we are not sending up to and including the packet end
689 then align the next start on an eight byte boundary */
694 /* Allocate buffer */
695 skb2
= alloc_skb(len
+ hlen
+ ll_rs
, GFP_ATOMIC
);
702 * Set up data on packet
705 ip_copy_metadata(skb2
, skb
);
706 skb_reserve(skb2
, ll_rs
);
707 skb_put(skb2
, len
+ hlen
);
708 skb_reset_network_header(skb2
);
709 skb2
->transport_header
= skb2
->network_header
+ hlen
;
712 * Charge the memory for the fragment to any owner
717 skb_set_owner_w(skb2
, skb
->sk
);
720 * Copy the packet header into the new buffer.
723 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
726 * Copy a block of the IP datagram.
728 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
733 * Fill in the new header fields.
736 iph
->frag_off
= htons((offset
>> 3));
738 if (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
)
739 iph
->frag_off
|= htons(IP_DF
);
741 /* ANK: dirty, but effective trick. Upgrade options only if
742 * the segment to be fragmented was THE FIRST (otherwise,
743 * options are already fixed) and make it ONCE
744 * on the initial skb, so that all the following fragments
745 * will inherit fixed options.
748 ip_options_fragment(skb
);
751 * Added AC : If we are fragmenting a fragment that's not the
752 * last fragment then keep MF on each bit
754 if (left
> 0 || not_last_frag
)
755 iph
->frag_off
|= htons(IP_MF
);
760 * Put this fragment into the sending queue.
762 iph
->tot_len
= htons(len
+ hlen
);
766 err
= output(net
, sk
, skb2
);
770 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
773 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
778 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
781 EXPORT_SYMBOL(ip_do_fragment
);
784 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
786 struct msghdr
*msg
= from
;
788 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
789 if (copy_from_iter(to
, len
, &msg
->msg_iter
) != len
)
793 if (csum_and_copy_from_iter(to
, len
, &csum
, &msg
->msg_iter
) != len
)
795 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
799 EXPORT_SYMBOL(ip_generic_getfrag
);
802 csum_page(struct page
*page
, int offset
, int copy
)
807 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
812 static inline int ip_ufo_append_data(struct sock
*sk
,
813 struct sk_buff_head
*queue
,
814 int getfrag(void *from
, char *to
, int offset
, int len
,
815 int odd
, struct sk_buff
*skb
),
816 void *from
, int length
, int hh_len
, int fragheaderlen
,
817 int transhdrlen
, int maxfraglen
, unsigned int flags
)
822 /* There is support for UDP fragmentation offload by network
823 * device, so create one single skb packet containing complete
826 skb
= skb_peek_tail(queue
);
828 skb
= sock_alloc_send_skb(sk
,
829 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
830 (flags
& MSG_DONTWAIT
), &err
);
835 /* reserve space for Hardware header */
836 skb_reserve(skb
, hh_len
);
838 /* create space for UDP/IP header */
839 skb_put(skb
, fragheaderlen
+ transhdrlen
);
841 /* initialize network header pointer */
842 skb_reset_network_header(skb
);
844 /* initialize protocol header pointer */
845 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
849 __skb_queue_tail(queue
, skb
);
850 } else if (skb_is_gso(skb
)) {
854 skb
->ip_summed
= CHECKSUM_PARTIAL
;
855 /* specify the length of each IP datagram fragment */
856 skb_shinfo(skb
)->gso_size
= maxfraglen
- fragheaderlen
;
857 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
860 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
861 (length
- transhdrlen
));
864 static int __ip_append_data(struct sock
*sk
,
866 struct sk_buff_head
*queue
,
867 struct inet_cork
*cork
,
868 struct page_frag
*pfrag
,
869 int getfrag(void *from
, char *to
, int offset
,
870 int len
, int odd
, struct sk_buff
*skb
),
871 void *from
, int length
, int transhdrlen
,
874 struct inet_sock
*inet
= inet_sk(sk
);
877 struct ip_options
*opt
= cork
->opt
;
884 unsigned int maxfraglen
, fragheaderlen
, maxnonfragsize
;
885 int csummode
= CHECKSUM_NONE
;
886 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
889 skb
= skb_peek_tail(queue
);
891 exthdrlen
= !skb
? rt
->dst
.header_len
: 0;
892 mtu
= cork
->fragsize
;
893 if (cork
->tx_flags
& SKBTX_ANY_SW_TSTAMP
&&
894 sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)
895 tskey
= sk
->sk_tskey
++;
897 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
899 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
900 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
901 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
903 if (cork
->length
+ length
> maxnonfragsize
- fragheaderlen
) {
904 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
905 mtu
- (opt
? opt
->optlen
: 0));
910 * transhdrlen > 0 means that this is the first fragment and we wish
911 * it won't be fragmented in the future.
914 length
+ fragheaderlen
<= mtu
&&
915 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
917 csummode
= CHECKSUM_PARTIAL
;
919 cork
->length
+= length
;
920 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
921 (sk
->sk_protocol
== IPPROTO_UDP
) &&
922 (rt
->dst
.dev
->features
& NETIF_F_UFO
) && !rt
->dst
.header_len
&&
923 (sk
->sk_type
== SOCK_DGRAM
)) {
924 err
= ip_ufo_append_data(sk
, queue
, getfrag
, from
, length
,
925 hh_len
, fragheaderlen
, transhdrlen
,
932 /* So, what's going on in the loop below?
934 * We use calculated fragment length to generate chained skb,
935 * each of segments is IP fragment ready for sending to network after
936 * adding appropriate IP header.
943 /* Check if the remaining data fits into current packet. */
944 copy
= mtu
- skb
->len
;
946 copy
= maxfraglen
- skb
->len
;
949 unsigned int datalen
;
950 unsigned int fraglen
;
951 unsigned int fraggap
;
952 unsigned int alloclen
;
953 struct sk_buff
*skb_prev
;
957 fraggap
= skb_prev
->len
- maxfraglen
;
962 * If remaining data exceeds the mtu,
963 * we know we need more fragment(s).
965 datalen
= length
+ fraggap
;
966 if (datalen
> mtu
- fragheaderlen
)
967 datalen
= maxfraglen
- fragheaderlen
;
968 fraglen
= datalen
+ fragheaderlen
;
970 if ((flags
& MSG_MORE
) &&
971 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
976 alloclen
+= exthdrlen
;
978 /* The last fragment gets additional space at tail.
979 * Note, with MSG_MORE we overallocate on fragments,
980 * because we have no idea what fragment will be
983 if (datalen
== length
+ fraggap
)
984 alloclen
+= rt
->dst
.trailer_len
;
987 skb
= sock_alloc_send_skb(sk
,
988 alloclen
+ hh_len
+ 15,
989 (flags
& MSG_DONTWAIT
), &err
);
992 if (atomic_read(&sk
->sk_wmem_alloc
) <=
994 skb
= sock_wmalloc(sk
,
995 alloclen
+ hh_len
+ 15, 1,
1004 * Fill in the control structures
1006 skb
->ip_summed
= csummode
;
1008 skb_reserve(skb
, hh_len
);
1010 /* only the initial fragment is time stamped */
1011 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
1013 skb_shinfo(skb
)->tskey
= tskey
;
1017 * Find where to start putting bytes.
1019 data
= skb_put(skb
, fraglen
+ exthdrlen
);
1020 skb_set_network_header(skb
, exthdrlen
);
1021 skb
->transport_header
= (skb
->network_header
+
1023 data
+= fragheaderlen
+ exthdrlen
;
1026 skb
->csum
= skb_copy_and_csum_bits(
1027 skb_prev
, maxfraglen
,
1028 data
+ transhdrlen
, fraggap
, 0);
1029 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1032 pskb_trim_unique(skb_prev
, maxfraglen
);
1035 copy
= datalen
- transhdrlen
- fraggap
;
1036 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1043 length
-= datalen
- fraggap
;
1046 csummode
= CHECKSUM_NONE
;
1049 * Put the packet on the pending queue.
1051 __skb_queue_tail(queue
, skb
);
1058 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
1062 if (getfrag(from
, skb_put(skb
, copy
),
1063 offset
, copy
, off
, skb
) < 0) {
1064 __skb_trim(skb
, off
);
1069 int i
= skb_shinfo(skb
)->nr_frags
;
1072 if (!sk_page_frag_refill(sk
, pfrag
))
1075 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
1078 if (i
== MAX_SKB_FRAGS
)
1081 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1083 skb_shinfo(skb
)->nr_frags
= ++i
;
1084 get_page(pfrag
->page
);
1086 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1088 page_address(pfrag
->page
) + pfrag
->offset
,
1089 offset
, copy
, skb
->len
, skb
) < 0)
1092 pfrag
->offset
+= copy
;
1093 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1095 skb
->data_len
+= copy
;
1096 skb
->truesize
+= copy
;
1097 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1108 cork
->length
-= length
;
1109 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1113 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1114 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1116 struct ip_options_rcu
*opt
;
1120 * setup for corking.
1125 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1127 if (unlikely(!cork
->opt
))
1130 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1131 cork
->flags
|= IPCORK_OPT
;
1132 cork
->addr
= ipc
->addr
;
1138 * We steal reference to this route, caller should not release it
1141 cork
->fragsize
= ip_sk_use_pmtu(sk
) ?
1142 dst_mtu(&rt
->dst
) : rt
->dst
.dev
->mtu
;
1143 cork
->dst
= &rt
->dst
;
1145 cork
->ttl
= ipc
->ttl
;
1146 cork
->tos
= ipc
->tos
;
1147 cork
->priority
= ipc
->priority
;
1148 cork
->tx_flags
= ipc
->tx_flags
;
1154 * ip_append_data() and ip_append_page() can make one large IP datagram
1155 * from many pieces of data. Each pieces will be holded on the socket
1156 * until ip_push_pending_frames() is called. Each piece can be a page
1159 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1160 * this interface potentially.
1162 * LATER: length must be adjusted by pad at tail, when it is required.
1164 int ip_append_data(struct sock
*sk
, struct flowi4
*fl4
,
1165 int getfrag(void *from
, char *to
, int offset
, int len
,
1166 int odd
, struct sk_buff
*skb
),
1167 void *from
, int length
, int transhdrlen
,
1168 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1171 struct inet_sock
*inet
= inet_sk(sk
);
1174 if (flags
&MSG_PROBE
)
1177 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1178 err
= ip_setup_cork(sk
, &inet
->cork
.base
, ipc
, rtp
);
1185 return __ip_append_data(sk
, fl4
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1186 sk_page_frag(sk
), getfrag
,
1187 from
, length
, transhdrlen
, flags
);
1190 ssize_t
ip_append_page(struct sock
*sk
, struct flowi4
*fl4
, struct page
*page
,
1191 int offset
, size_t size
, int flags
)
1193 struct inet_sock
*inet
= inet_sk(sk
);
1194 struct sk_buff
*skb
;
1196 struct ip_options
*opt
= NULL
;
1197 struct inet_cork
*cork
;
1202 unsigned int maxfraglen
, fragheaderlen
, fraggap
, maxnonfragsize
;
1207 if (flags
&MSG_PROBE
)
1210 if (skb_queue_empty(&sk
->sk_write_queue
))
1213 cork
= &inet
->cork
.base
;
1214 rt
= (struct rtable
*)cork
->dst
;
1215 if (cork
->flags
& IPCORK_OPT
)
1218 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1221 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1222 mtu
= cork
->fragsize
;
1224 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1225 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1226 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
1228 if (cork
->length
+ size
> maxnonfragsize
- fragheaderlen
) {
1229 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
1230 mtu
- (opt
? opt
->optlen
: 0));
1234 skb
= skb_peek_tail(&sk
->sk_write_queue
);
1238 cork
->length
+= size
;
1239 if ((size
+ skb
->len
> mtu
) &&
1240 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1241 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1242 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1243 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1247 if (skb_is_gso(skb
)) {
1251 /* Check if the remaining data fits into current packet. */
1252 len
= mtu
- skb
->len
;
1254 len
= maxfraglen
- skb
->len
;
1257 struct sk_buff
*skb_prev
;
1261 fraggap
= skb_prev
->len
- maxfraglen
;
1263 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1264 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1265 if (unlikely(!skb
)) {
1271 * Fill in the control structures
1273 skb
->ip_summed
= CHECKSUM_NONE
;
1275 skb_reserve(skb
, hh_len
);
1278 * Find where to start putting bytes.
1280 skb_put(skb
, fragheaderlen
+ fraggap
);
1281 skb_reset_network_header(skb
);
1282 skb
->transport_header
= (skb
->network_header
+
1285 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1287 skb_transport_header(skb
),
1289 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1291 pskb_trim_unique(skb_prev
, maxfraglen
);
1295 * Put the packet on the pending queue.
1297 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1304 if (skb_append_pagefrags(skb
, page
, offset
, len
)) {
1309 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1311 csum
= csum_page(page
, offset
, len
);
1312 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1316 skb
->data_len
+= len
;
1317 skb
->truesize
+= len
;
1318 atomic_add(len
, &sk
->sk_wmem_alloc
);
1325 cork
->length
-= size
;
1326 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1330 static void ip_cork_release(struct inet_cork
*cork
)
1332 cork
->flags
&= ~IPCORK_OPT
;
1335 dst_release(cork
->dst
);
1340 * Combined all pending IP fragments on the socket as one IP datagram
1341 * and push them out.
1343 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1345 struct sk_buff_head
*queue
,
1346 struct inet_cork
*cork
)
1348 struct sk_buff
*skb
, *tmp_skb
;
1349 struct sk_buff
**tail_skb
;
1350 struct inet_sock
*inet
= inet_sk(sk
);
1351 struct net
*net
= sock_net(sk
);
1352 struct ip_options
*opt
= NULL
;
1353 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1358 skb
= __skb_dequeue(queue
);
1361 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1363 /* move skb->data to ip header from ext header */
1364 if (skb
->data
< skb_network_header(skb
))
1365 __skb_pull(skb
, skb_network_offset(skb
));
1366 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1367 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1368 *tail_skb
= tmp_skb
;
1369 tail_skb
= &(tmp_skb
->next
);
1370 skb
->len
+= tmp_skb
->len
;
1371 skb
->data_len
+= tmp_skb
->len
;
1372 skb
->truesize
+= tmp_skb
->truesize
;
1373 tmp_skb
->destructor
= NULL
;
1377 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1378 * to fragment the frame generated here. No matter, what transforms
1379 * how transforms change size of the packet, it will come out.
1381 skb
->ignore_df
= ip_sk_ignore_df(sk
);
1383 /* DF bit is set when we want to see DF on outgoing frames.
1384 * If ignore_df is set too, we still allow to fragment this frame
1386 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1387 inet
->pmtudisc
== IP_PMTUDISC_PROBE
||
1388 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1389 ip_dont_fragment(sk
, &rt
->dst
)))
1392 if (cork
->flags
& IPCORK_OPT
)
1397 else if (rt
->rt_type
== RTN_MULTICAST
)
1400 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1405 iph
->tos
= (cork
->tos
!= -1) ? cork
->tos
: inet
->tos
;
1408 iph
->protocol
= sk
->sk_protocol
;
1409 ip_copy_addrs(iph
, fl4
);
1410 ip_select_ident(net
, skb
, sk
);
1413 iph
->ihl
+= opt
->optlen
>>2;
1414 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1417 skb
->priority
= (cork
->tos
!= -1) ? cork
->priority
: sk
->sk_priority
;
1418 skb
->mark
= sk
->sk_mark
;
1420 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1424 skb_dst_set(skb
, &rt
->dst
);
1426 if (iph
->protocol
== IPPROTO_ICMP
)
1427 icmp_out_count(net
, ((struct icmphdr
*)
1428 skb_transport_header(skb
))->type
);
1430 ip_cork_release(cork
);
1435 int ip_send_skb(struct net
*net
, struct sk_buff
*skb
)
1439 err
= ip_local_out(skb
->sk
, skb
);
1442 err
= net_xmit_errno(err
);
1444 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1450 int ip_push_pending_frames(struct sock
*sk
, struct flowi4
*fl4
)
1452 struct sk_buff
*skb
;
1454 skb
= ip_finish_skb(sk
, fl4
);
1458 /* Netfilter gets whole the not fragmented skb. */
1459 return ip_send_skb(sock_net(sk
), skb
);
1463 * Throw away all pending data on the socket.
1465 static void __ip_flush_pending_frames(struct sock
*sk
,
1466 struct sk_buff_head
*queue
,
1467 struct inet_cork
*cork
)
1469 struct sk_buff
*skb
;
1471 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1474 ip_cork_release(cork
);
1477 void ip_flush_pending_frames(struct sock
*sk
)
1479 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
.base
);
1482 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1484 int getfrag(void *from
, char *to
, int offset
,
1485 int len
, int odd
, struct sk_buff
*skb
),
1486 void *from
, int length
, int transhdrlen
,
1487 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1490 struct inet_cork cork
;
1491 struct sk_buff_head queue
;
1494 if (flags
& MSG_PROBE
)
1497 __skb_queue_head_init(&queue
);
1502 err
= ip_setup_cork(sk
, &cork
, ipc
, rtp
);
1504 return ERR_PTR(err
);
1506 err
= __ip_append_data(sk
, fl4
, &queue
, &cork
,
1507 ¤t
->task_frag
, getfrag
,
1508 from
, length
, transhdrlen
, flags
);
1510 __ip_flush_pending_frames(sk
, &queue
, &cork
);
1511 return ERR_PTR(err
);
1514 return __ip_make_skb(sk
, fl4
, &queue
, &cork
);
1518 * Fetch data from kernel space and fill in checksum if needed.
1520 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1521 int len
, int odd
, struct sk_buff
*skb
)
1525 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1526 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1531 * Generic function to send a packet as reply to another packet.
1532 * Used to send some TCP resets/acks so far.
1534 void ip_send_unicast_reply(struct sock
*sk
, struct sk_buff
*skb
,
1535 const struct ip_options
*sopt
,
1536 __be32 daddr
, __be32 saddr
,
1537 const struct ip_reply_arg
*arg
,
1540 struct ip_options_data replyopts
;
1541 struct ipcm_cookie ipc
;
1543 struct rtable
*rt
= skb_rtable(skb
);
1544 struct net
*net
= sock_net(sk
);
1545 struct sk_buff
*nskb
;
1549 if (__ip_options_echo(&replyopts
.opt
.opt
, skb
, sopt
))
1558 if (replyopts
.opt
.opt
.optlen
) {
1559 ipc
.opt
= &replyopts
.opt
;
1561 if (replyopts
.opt
.opt
.srr
)
1562 daddr
= replyopts
.opt
.opt
.faddr
;
1565 oif
= arg
->bound_dev_if
;
1566 if (!oif
&& netif_index_is_l3_master(net
, skb
->skb_iif
))
1569 flowi4_init_output(&fl4
, oif
,
1570 IP4_REPLY_MARK(net
, skb
->mark
),
1572 RT_SCOPE_UNIVERSE
, ip_hdr(skb
)->protocol
,
1573 ip_reply_arg_flowi_flags(arg
),
1575 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
);
1576 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1577 rt
= ip_route_output_key(net
, &fl4
);
1581 inet_sk(sk
)->tos
= arg
->tos
;
1583 sk
->sk_priority
= skb
->priority
;
1584 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1585 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1586 sk
->sk_sndbuf
= sysctl_wmem_default
;
1587 err
= ip_append_data(sk
, &fl4
, ip_reply_glue_bits
, arg
->iov
->iov_base
,
1588 len
, 0, &ipc
, &rt
, MSG_DONTWAIT
);
1589 if (unlikely(err
)) {
1590 ip_flush_pending_frames(sk
);
1594 nskb
= skb_peek(&sk
->sk_write_queue
);
1596 if (arg
->csumoffset
>= 0)
1597 *((__sum16
*)skb_transport_header(nskb
) +
1598 arg
->csumoffset
) = csum_fold(csum_add(nskb
->csum
,
1600 nskb
->ip_summed
= CHECKSUM_NONE
;
1601 skb_set_queue_mapping(nskb
, skb_get_queue_mapping(skb
));
1602 ip_push_pending_frames(sk
, &fl4
);
1608 void __init
ip_init(void)
1613 #if defined(CONFIG_IP_MULTICAST)