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
);
86 /* Generate a checksum for an outgoing IP datagram. */
87 void ip_send_check(struct iphdr
*iph
)
90 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
92 EXPORT_SYMBOL(ip_send_check
);
94 int __ip_local_out(struct sk_buff
*skb
)
96 struct iphdr
*iph
= ip_hdr(skb
);
98 iph
->tot_len
= htons(skb
->len
);
100 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, skb
, NULL
,
101 skb_dst(skb
)->dev
, dst_output
);
104 int ip_local_out_sk(struct sock
*sk
, struct sk_buff
*skb
)
108 err
= __ip_local_out(skb
);
109 if (likely(err
== 1))
110 err
= dst_output_sk(sk
, skb
);
114 EXPORT_SYMBOL_GPL(ip_local_out_sk
);
116 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
118 int ttl
= inet
->uc_ttl
;
121 ttl
= ip4_dst_hoplimit(dst
);
126 * Add an ip header to a skbuff and send it out.
129 int ip_build_and_send_pkt(struct sk_buff
*skb
, struct sock
*sk
,
130 __be32 saddr
, __be32 daddr
, struct ip_options_rcu
*opt
)
132 struct inet_sock
*inet
= inet_sk(sk
);
133 struct rtable
*rt
= skb_rtable(skb
);
136 /* Build the IP header. */
137 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->opt
.optlen
: 0));
138 skb_reset_network_header(skb
);
142 iph
->tos
= inet
->tos
;
143 if (ip_dont_fragment(sk
, &rt
->dst
))
144 iph
->frag_off
= htons(IP_DF
);
147 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
148 iph
->daddr
= (opt
&& opt
->opt
.srr
? opt
->opt
.faddr
: daddr
);
150 iph
->protocol
= sk
->sk_protocol
;
151 ip_select_ident(sock_net(sk
), skb
, sk
);
153 if (opt
&& opt
->opt
.optlen
) {
154 iph
->ihl
+= opt
->opt
.optlen
>>2;
155 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
158 skb
->priority
= sk
->sk_priority
;
159 skb
->mark
= sk
->sk_mark
;
162 return ip_local_out(skb
);
164 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
166 static inline int ip_finish_output2(struct sk_buff
*skb
)
168 struct dst_entry
*dst
= skb_dst(skb
);
169 struct rtable
*rt
= (struct rtable
*)dst
;
170 struct net_device
*dev
= dst
->dev
;
171 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
172 struct neighbour
*neigh
;
175 if (rt
->rt_type
== RTN_MULTICAST
) {
176 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTMCAST
, skb
->len
);
177 } else if (rt
->rt_type
== RTN_BROADCAST
)
178 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTBCAST
, skb
->len
);
180 /* Be paranoid, rather than too clever. */
181 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
182 struct sk_buff
*skb2
;
184 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
190 skb_set_owner_w(skb2
, skb
->sk
);
196 nexthop
= (__force u32
) rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
197 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
198 if (unlikely(!neigh
))
199 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
200 if (!IS_ERR(neigh
)) {
201 int res
= dst_neigh_output(dst
, neigh
, skb
);
203 rcu_read_unlock_bh();
206 rcu_read_unlock_bh();
208 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
214 static int ip_finish_output_gso(struct sk_buff
*skb
)
216 netdev_features_t features
;
217 struct sk_buff
*segs
;
220 /* common case: locally created skb or seglen is <= mtu */
221 if (((IPCB(skb
)->flags
& IPSKB_FORWARDED
) == 0) ||
222 skb_gso_network_seglen(skb
) <= ip_skb_dst_mtu(skb
))
223 return ip_finish_output2(skb
);
225 /* Slowpath - GSO segment length is exceeding the dst MTU.
227 * This can happen in two cases:
228 * 1) TCP GRO packet, DF bit not set
229 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
230 * from host network stack.
232 features
= netif_skb_features(skb
);
233 segs
= skb_gso_segment(skb
, features
& ~NETIF_F_GSO_MASK
);
234 if (IS_ERR_OR_NULL(segs
)) {
242 struct sk_buff
*nskb
= segs
->next
;
246 err
= ip_fragment(segs
, ip_finish_output2
);
256 static int ip_finish_output(struct sk_buff
*skb
)
258 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
259 /* Policy lookup after SNAT yielded a new policy */
260 if (skb_dst(skb
)->xfrm
!= NULL
) {
261 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
262 return dst_output(skb
);
266 return ip_finish_output_gso(skb
);
268 if (skb
->len
> ip_skb_dst_mtu(skb
))
269 return ip_fragment(skb
, ip_finish_output2
);
271 return ip_finish_output2(skb
);
274 int ip_mc_output(struct sock
*sk
, struct sk_buff
*skb
)
276 struct rtable
*rt
= skb_rtable(skb
);
277 struct net_device
*dev
= rt
->dst
.dev
;
280 * If the indicated interface is up and running, send the packet.
282 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
285 skb
->protocol
= htons(ETH_P_IP
);
288 * Multicasts are looped back for other local users
291 if (rt
->rt_flags
&RTCF_MULTICAST
) {
293 #ifdef CONFIG_IP_MROUTE
294 /* Small optimization: do not loopback not local frames,
295 which returned after forwarding; they will be dropped
296 by ip_mr_input in any case.
297 Note, that local frames are looped back to be delivered
300 This check is duplicated in ip_mr_input at the moment.
303 ((rt
->rt_flags
& RTCF_LOCAL
) ||
304 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
307 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
309 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
310 newskb
, NULL
, newskb
->dev
,
314 /* Multicasts with ttl 0 must not go beyond the host */
316 if (ip_hdr(skb
)->ttl
== 0) {
322 if (rt
->rt_flags
&RTCF_BROADCAST
) {
323 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
325 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, newskb
,
326 NULL
, newskb
->dev
, dev_loopback_xmit
);
329 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
,
330 skb
->dev
, ip_finish_output
,
331 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
334 int ip_output(struct sock
*sk
, struct sk_buff
*skb
)
336 struct net_device
*dev
= skb_dst(skb
)->dev
;
338 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
341 skb
->protocol
= htons(ETH_P_IP
);
343 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
, dev
,
345 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
349 * copy saddr and daddr, possibly using 64bit load/stores
351 * iph->saddr = fl4->saddr;
352 * iph->daddr = fl4->daddr;
354 static void ip_copy_addrs(struct iphdr
*iph
, const struct flowi4
*fl4
)
356 BUILD_BUG_ON(offsetof(typeof(*fl4
), daddr
) !=
357 offsetof(typeof(*fl4
), saddr
) + sizeof(fl4
->saddr
));
358 memcpy(&iph
->saddr
, &fl4
->saddr
,
359 sizeof(fl4
->saddr
) + sizeof(fl4
->daddr
));
362 /* Note: skb->sk can be different from sk, in case of tunnels */
363 int ip_queue_xmit(struct sock
*sk
, struct sk_buff
*skb
, struct flowi
*fl
)
365 struct inet_sock
*inet
= inet_sk(sk
);
366 struct ip_options_rcu
*inet_opt
;
372 /* Skip all of this if the packet is already routed,
373 * f.e. by something like SCTP.
376 inet_opt
= rcu_dereference(inet
->inet_opt
);
378 rt
= skb_rtable(skb
);
382 /* Make sure we can route this packet. */
383 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
387 /* Use correct destination address if we have options. */
388 daddr
= inet
->inet_daddr
;
389 if (inet_opt
&& inet_opt
->opt
.srr
)
390 daddr
= inet_opt
->opt
.faddr
;
392 /* If this fails, retransmit mechanism of transport layer will
393 * keep trying until route appears or the connection times
396 rt
= ip_route_output_ports(sock_net(sk
), fl4
, sk
,
397 daddr
, inet
->inet_saddr
,
402 sk
->sk_bound_dev_if
);
405 sk_setup_caps(sk
, &rt
->dst
);
407 skb_dst_set_noref(skb
, &rt
->dst
);
410 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_uses_gateway
)
413 /* OK, we know where to send it, allocate and build IP header. */
414 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
415 skb_reset_network_header(skb
);
417 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
418 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->ignore_df
)
419 iph
->frag_off
= htons(IP_DF
);
422 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
423 iph
->protocol
= sk
->sk_protocol
;
424 ip_copy_addrs(iph
, fl4
);
426 /* Transport layer set skb->h.foo itself. */
428 if (inet_opt
&& inet_opt
->opt
.optlen
) {
429 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
430 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
433 ip_select_ident_segs(sock_net(sk
), skb
, sk
,
434 skb_shinfo(skb
)->gso_segs
?: 1);
436 /* TODO : should we use skb->sk here instead of sk ? */
437 skb
->priority
= sk
->sk_priority
;
438 skb
->mark
= sk
->sk_mark
;
440 res
= ip_local_out(skb
);
446 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
448 return -EHOSTUNREACH
;
450 EXPORT_SYMBOL(ip_queue_xmit
);
453 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
455 to
->pkt_type
= from
->pkt_type
;
456 to
->priority
= from
->priority
;
457 to
->protocol
= from
->protocol
;
459 skb_dst_copy(to
, from
);
461 to
->mark
= from
->mark
;
463 /* Copy the flags to each fragment. */
464 IPCB(to
)->flags
= IPCB(from
)->flags
;
466 #ifdef CONFIG_NET_SCHED
467 to
->tc_index
= from
->tc_index
;
470 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
471 to
->ipvs_property
= from
->ipvs_property
;
473 skb_copy_secmark(to
, from
);
477 * This IP datagram is too large to be sent in one piece. Break it up into
478 * smaller pieces (each of size equal to IP header plus
479 * a block of the data of the original IP data part) that will yet fit in a
480 * single device frame, and queue such a frame for sending.
483 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
487 struct net_device
*dev
;
488 struct sk_buff
*skb2
;
489 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
491 __be16 not_last_frag
;
492 struct rtable
*rt
= skb_rtable(skb
);
498 * Point into the IP datagram header.
503 mtu
= ip_skb_dst_mtu(skb
);
504 if (unlikely(((iph
->frag_off
& htons(IP_DF
)) && !skb
->ignore_df
) ||
505 (IPCB(skb
)->frag_max_size
&&
506 IPCB(skb
)->frag_max_size
> mtu
))) {
507 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
508 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
515 * Setup starting values.
519 mtu
= mtu
- hlen
; /* Size of data space */
520 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
522 mtu
-= nf_bridge_mtu_reduction(skb
);
524 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
526 /* When frag_list is given, use it. First, check its validity:
527 * some transformers could create wrong frag_list or break existing
528 * one, it is not prohibited. In this case fall back to copying.
530 * LATER: this step can be merged to real generation of fragments,
531 * we can switch to copy when see the first bad fragment.
533 if (skb_has_frag_list(skb
)) {
534 struct sk_buff
*frag
, *frag2
;
535 int first_len
= skb_pagelen(skb
);
537 if (first_len
- hlen
> mtu
||
538 ((first_len
- hlen
) & 7) ||
539 ip_is_fragment(iph
) ||
543 skb_walk_frags(skb
, frag
) {
544 /* Correct geometry. */
545 if (frag
->len
> mtu
||
546 ((frag
->len
& 7) && frag
->next
) ||
547 skb_headroom(frag
) < hlen
)
548 goto slow_path_clean
;
550 /* Partially cloned skb? */
551 if (skb_shared(frag
))
552 goto slow_path_clean
;
557 frag
->destructor
= sock_wfree
;
559 skb
->truesize
-= frag
->truesize
;
562 /* Everything is OK. Generate! */
566 frag
= skb_shinfo(skb
)->frag_list
;
567 skb_frag_list_init(skb
);
568 skb
->data_len
= first_len
- skb_headlen(skb
);
569 skb
->len
= first_len
;
570 iph
->tot_len
= htons(first_len
);
571 iph
->frag_off
= htons(IP_MF
);
575 /* Prepare header of the next frame,
576 * before previous one went down. */
578 frag
->ip_summed
= CHECKSUM_NONE
;
579 skb_reset_transport_header(frag
);
580 __skb_push(frag
, hlen
);
581 skb_reset_network_header(frag
);
582 memcpy(skb_network_header(frag
), iph
, hlen
);
584 iph
->tot_len
= htons(frag
->len
);
585 ip_copy_metadata(frag
, skb
);
587 ip_options_fragment(frag
);
588 offset
+= skb
->len
- hlen
;
589 iph
->frag_off
= htons(offset
>>3);
590 if (frag
->next
!= NULL
)
591 iph
->frag_off
|= htons(IP_MF
);
592 /* Ready, complete checksum */
599 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
609 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
618 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
622 skb_walk_frags(skb
, frag2
) {
626 frag2
->destructor
= NULL
;
627 skb
->truesize
+= frag2
->truesize
;
632 /* for offloaded checksums cleanup checksum before fragmentation */
633 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) && skb_checksum_help(skb
))
637 left
= skb
->len
- hlen
; /* Space per frame */
638 ptr
= hlen
; /* Where to start from */
640 ll_rs
= LL_RESERVED_SPACE(rt
->dst
.dev
);
643 * Fragment the datagram.
646 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
647 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
650 * Keep copying data until we run out.
655 /* IF: it doesn't fit, use 'mtu' - the data space left */
658 /* IF: we are not sending up to and including the packet end
659 then align the next start on an eight byte boundary */
664 /* Allocate buffer */
665 skb2
= alloc_skb(len
+ hlen
+ ll_rs
, GFP_ATOMIC
);
672 * Set up data on packet
675 ip_copy_metadata(skb2
, skb
);
676 skb_reserve(skb2
, ll_rs
);
677 skb_put(skb2
, len
+ hlen
);
678 skb_reset_network_header(skb2
);
679 skb2
->transport_header
= skb2
->network_header
+ hlen
;
682 * Charge the memory for the fragment to any owner
687 skb_set_owner_w(skb2
, skb
->sk
);
690 * Copy the packet header into the new buffer.
693 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
696 * Copy a block of the IP datagram.
698 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
703 * Fill in the new header fields.
706 iph
->frag_off
= htons((offset
>> 3));
708 /* ANK: dirty, but effective trick. Upgrade options only if
709 * the segment to be fragmented was THE FIRST (otherwise,
710 * options are already fixed) and make it ONCE
711 * on the initial skb, so that all the following fragments
712 * will inherit fixed options.
715 ip_options_fragment(skb
);
718 * Added AC : If we are fragmenting a fragment that's not the
719 * last fragment then keep MF on each bit
721 if (left
> 0 || not_last_frag
)
722 iph
->frag_off
|= htons(IP_MF
);
727 * Put this fragment into the sending queue.
729 iph
->tot_len
= htons(len
+ hlen
);
737 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
740 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
745 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
748 EXPORT_SYMBOL(ip_fragment
);
751 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
753 struct msghdr
*msg
= from
;
755 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
756 if (copy_from_iter(to
, len
, &msg
->msg_iter
) != len
)
760 if (csum_and_copy_from_iter(to
, len
, &csum
, &msg
->msg_iter
) != len
)
762 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
766 EXPORT_SYMBOL(ip_generic_getfrag
);
769 csum_page(struct page
*page
, int offset
, int copy
)
774 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
779 static inline int ip_ufo_append_data(struct sock
*sk
,
780 struct sk_buff_head
*queue
,
781 int getfrag(void *from
, char *to
, int offset
, int len
,
782 int odd
, struct sk_buff
*skb
),
783 void *from
, int length
, int hh_len
, int fragheaderlen
,
784 int transhdrlen
, int maxfraglen
, unsigned int flags
)
789 /* There is support for UDP fragmentation offload by network
790 * device, so create one single skb packet containing complete
793 if ((skb
= skb_peek_tail(queue
)) == NULL
) {
794 skb
= sock_alloc_send_skb(sk
,
795 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
796 (flags
& MSG_DONTWAIT
), &err
);
801 /* reserve space for Hardware header */
802 skb_reserve(skb
, hh_len
);
804 /* create space for UDP/IP header */
805 skb_put(skb
, fragheaderlen
+ transhdrlen
);
807 /* initialize network header pointer */
808 skb_reset_network_header(skb
);
810 /* initialize protocol header pointer */
811 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
816 __skb_queue_tail(queue
, skb
);
817 } else if (skb_is_gso(skb
)) {
821 skb
->ip_summed
= CHECKSUM_PARTIAL
;
822 /* specify the length of each IP datagram fragment */
823 skb_shinfo(skb
)->gso_size
= maxfraglen
- fragheaderlen
;
824 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
827 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
828 (length
- transhdrlen
));
831 static int __ip_append_data(struct sock
*sk
,
833 struct sk_buff_head
*queue
,
834 struct inet_cork
*cork
,
835 struct page_frag
*pfrag
,
836 int getfrag(void *from
, char *to
, int offset
,
837 int len
, int odd
, struct sk_buff
*skb
),
838 void *from
, int length
, int transhdrlen
,
841 struct inet_sock
*inet
= inet_sk(sk
);
844 struct ip_options
*opt
= cork
->opt
;
851 unsigned int maxfraglen
, fragheaderlen
, maxnonfragsize
;
852 int csummode
= CHECKSUM_NONE
;
853 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
856 skb
= skb_peek_tail(queue
);
858 exthdrlen
= !skb
? rt
->dst
.header_len
: 0;
859 mtu
= cork
->fragsize
;
860 if (cork
->tx_flags
& SKBTX_ANY_SW_TSTAMP
&&
861 sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)
862 tskey
= sk
->sk_tskey
++;
864 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
866 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
867 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
868 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
870 if (cork
->length
+ length
> maxnonfragsize
- fragheaderlen
) {
871 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
872 mtu
- (opt
? opt
->optlen
: 0));
877 * transhdrlen > 0 means that this is the first fragment and we wish
878 * it won't be fragmented in the future.
881 length
+ fragheaderlen
<= mtu
&&
882 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
884 csummode
= CHECKSUM_PARTIAL
;
886 cork
->length
+= length
;
887 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
888 (sk
->sk_protocol
== IPPROTO_UDP
) &&
889 (rt
->dst
.dev
->features
& NETIF_F_UFO
) && !rt
->dst
.header_len
&&
890 (sk
->sk_type
== SOCK_DGRAM
)) {
891 err
= ip_ufo_append_data(sk
, queue
, getfrag
, from
, length
,
892 hh_len
, fragheaderlen
, transhdrlen
,
899 /* So, what's going on in the loop below?
901 * We use calculated fragment length to generate chained skb,
902 * each of segments is IP fragment ready for sending to network after
903 * adding appropriate IP header.
910 /* Check if the remaining data fits into current packet. */
911 copy
= mtu
- skb
->len
;
913 copy
= maxfraglen
- skb
->len
;
916 unsigned int datalen
;
917 unsigned int fraglen
;
918 unsigned int fraggap
;
919 unsigned int alloclen
;
920 struct sk_buff
*skb_prev
;
924 fraggap
= skb_prev
->len
- maxfraglen
;
929 * If remaining data exceeds the mtu,
930 * we know we need more fragment(s).
932 datalen
= length
+ fraggap
;
933 if (datalen
> mtu
- fragheaderlen
)
934 datalen
= maxfraglen
- fragheaderlen
;
935 fraglen
= datalen
+ fragheaderlen
;
937 if ((flags
& MSG_MORE
) &&
938 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
943 alloclen
+= exthdrlen
;
945 /* The last fragment gets additional space at tail.
946 * Note, with MSG_MORE we overallocate on fragments,
947 * because we have no idea what fragment will be
950 if (datalen
== length
+ fraggap
)
951 alloclen
+= rt
->dst
.trailer_len
;
954 skb
= sock_alloc_send_skb(sk
,
955 alloclen
+ hh_len
+ 15,
956 (flags
& MSG_DONTWAIT
), &err
);
959 if (atomic_read(&sk
->sk_wmem_alloc
) <=
961 skb
= sock_wmalloc(sk
,
962 alloclen
+ hh_len
+ 15, 1,
964 if (unlikely(skb
== NULL
))
971 * Fill in the control structures
973 skb
->ip_summed
= csummode
;
975 skb_reserve(skb
, hh_len
);
977 /* only the initial fragment is time stamped */
978 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
980 skb_shinfo(skb
)->tskey
= tskey
;
984 * Find where to start putting bytes.
986 data
= skb_put(skb
, fraglen
+ exthdrlen
);
987 skb_set_network_header(skb
, exthdrlen
);
988 skb
->transport_header
= (skb
->network_header
+
990 data
+= fragheaderlen
+ exthdrlen
;
993 skb
->csum
= skb_copy_and_csum_bits(
994 skb_prev
, maxfraglen
,
995 data
+ transhdrlen
, fraggap
, 0);
996 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
999 pskb_trim_unique(skb_prev
, maxfraglen
);
1002 copy
= datalen
- transhdrlen
- fraggap
;
1003 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1010 length
-= datalen
- fraggap
;
1013 csummode
= CHECKSUM_NONE
;
1016 * Put the packet on the pending queue.
1018 __skb_queue_tail(queue
, skb
);
1025 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
1029 if (getfrag(from
, skb_put(skb
, copy
),
1030 offset
, copy
, off
, skb
) < 0) {
1031 __skb_trim(skb
, off
);
1036 int i
= skb_shinfo(skb
)->nr_frags
;
1039 if (!sk_page_frag_refill(sk
, pfrag
))
1042 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
1045 if (i
== MAX_SKB_FRAGS
)
1048 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1050 skb_shinfo(skb
)->nr_frags
= ++i
;
1051 get_page(pfrag
->page
);
1053 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1055 page_address(pfrag
->page
) + pfrag
->offset
,
1056 offset
, copy
, skb
->len
, skb
) < 0)
1059 pfrag
->offset
+= copy
;
1060 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1062 skb
->data_len
+= copy
;
1063 skb
->truesize
+= copy
;
1064 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1075 cork
->length
-= length
;
1076 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1080 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1081 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1083 struct ip_options_rcu
*opt
;
1087 * setup for corking.
1091 if (cork
->opt
== NULL
) {
1092 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1094 if (unlikely(cork
->opt
== NULL
))
1097 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1098 cork
->flags
|= IPCORK_OPT
;
1099 cork
->addr
= ipc
->addr
;
1105 * We steal reference to this route, caller should not release it
1108 cork
->fragsize
= ip_sk_use_pmtu(sk
) ?
1109 dst_mtu(&rt
->dst
) : rt
->dst
.dev
->mtu
;
1110 cork
->dst
= &rt
->dst
;
1112 cork
->ttl
= ipc
->ttl
;
1113 cork
->tos
= ipc
->tos
;
1114 cork
->priority
= ipc
->priority
;
1115 cork
->tx_flags
= ipc
->tx_flags
;
1121 * ip_append_data() and ip_append_page() can make one large IP datagram
1122 * from many pieces of data. Each pieces will be holded on the socket
1123 * until ip_push_pending_frames() is called. Each piece can be a page
1126 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1127 * this interface potentially.
1129 * LATER: length must be adjusted by pad at tail, when it is required.
1131 int ip_append_data(struct sock
*sk
, struct flowi4
*fl4
,
1132 int getfrag(void *from
, char *to
, int offset
, int len
,
1133 int odd
, struct sk_buff
*skb
),
1134 void *from
, int length
, int transhdrlen
,
1135 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1138 struct inet_sock
*inet
= inet_sk(sk
);
1141 if (flags
&MSG_PROBE
)
1144 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1145 err
= ip_setup_cork(sk
, &inet
->cork
.base
, ipc
, rtp
);
1152 return __ip_append_data(sk
, fl4
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1153 sk_page_frag(sk
), getfrag
,
1154 from
, length
, transhdrlen
, flags
);
1157 ssize_t
ip_append_page(struct sock
*sk
, struct flowi4
*fl4
, struct page
*page
,
1158 int offset
, size_t size
, int flags
)
1160 struct inet_sock
*inet
= inet_sk(sk
);
1161 struct sk_buff
*skb
;
1163 struct ip_options
*opt
= NULL
;
1164 struct inet_cork
*cork
;
1169 unsigned int maxfraglen
, fragheaderlen
, fraggap
, maxnonfragsize
;
1174 if (flags
&MSG_PROBE
)
1177 if (skb_queue_empty(&sk
->sk_write_queue
))
1180 cork
= &inet
->cork
.base
;
1181 rt
= (struct rtable
*)cork
->dst
;
1182 if (cork
->flags
& IPCORK_OPT
)
1185 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1188 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1189 mtu
= cork
->fragsize
;
1191 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1192 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1193 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
1195 if (cork
->length
+ size
> maxnonfragsize
- fragheaderlen
) {
1196 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
1197 mtu
- (opt
? opt
->optlen
: 0));
1201 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1204 cork
->length
+= size
;
1205 if ((size
+ skb
->len
> mtu
) &&
1206 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1207 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1208 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1209 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1216 if (skb_is_gso(skb
))
1220 /* Check if the remaining data fits into current packet. */
1221 len
= mtu
- skb
->len
;
1223 len
= maxfraglen
- skb
->len
;
1226 struct sk_buff
*skb_prev
;
1230 fraggap
= skb_prev
->len
- maxfraglen
;
1232 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1233 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1234 if (unlikely(!skb
)) {
1240 * Fill in the control structures
1242 skb
->ip_summed
= CHECKSUM_NONE
;
1244 skb_reserve(skb
, hh_len
);
1247 * Find where to start putting bytes.
1249 skb_put(skb
, fragheaderlen
+ fraggap
);
1250 skb_reset_network_header(skb
);
1251 skb
->transport_header
= (skb
->network_header
+
1254 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1256 skb_transport_header(skb
),
1258 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1260 pskb_trim_unique(skb_prev
, maxfraglen
);
1264 * Put the packet on the pending queue.
1266 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1270 i
= skb_shinfo(skb
)->nr_frags
;
1273 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1274 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
-1], len
);
1275 } else if (i
< MAX_SKB_FRAGS
) {
1277 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1283 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1285 csum
= csum_page(page
, offset
, len
);
1286 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1290 skb
->data_len
+= len
;
1291 skb
->truesize
+= len
;
1292 atomic_add(len
, &sk
->sk_wmem_alloc
);
1299 cork
->length
-= size
;
1300 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1304 static void ip_cork_release(struct inet_cork
*cork
)
1306 cork
->flags
&= ~IPCORK_OPT
;
1309 dst_release(cork
->dst
);
1314 * Combined all pending IP fragments on the socket as one IP datagram
1315 * and push them out.
1317 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1319 struct sk_buff_head
*queue
,
1320 struct inet_cork
*cork
)
1322 struct sk_buff
*skb
, *tmp_skb
;
1323 struct sk_buff
**tail_skb
;
1324 struct inet_sock
*inet
= inet_sk(sk
);
1325 struct net
*net
= sock_net(sk
);
1326 struct ip_options
*opt
= NULL
;
1327 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1332 if ((skb
= __skb_dequeue(queue
)) == NULL
)
1334 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1336 /* move skb->data to ip header from ext header */
1337 if (skb
->data
< skb_network_header(skb
))
1338 __skb_pull(skb
, skb_network_offset(skb
));
1339 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1340 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1341 *tail_skb
= tmp_skb
;
1342 tail_skb
= &(tmp_skb
->next
);
1343 skb
->len
+= tmp_skb
->len
;
1344 skb
->data_len
+= tmp_skb
->len
;
1345 skb
->truesize
+= tmp_skb
->truesize
;
1346 tmp_skb
->destructor
= NULL
;
1350 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1351 * to fragment the frame generated here. No matter, what transforms
1352 * how transforms change size of the packet, it will come out.
1354 skb
->ignore_df
= ip_sk_ignore_df(sk
);
1356 /* DF bit is set when we want to see DF on outgoing frames.
1357 * If ignore_df is set too, we still allow to fragment this frame
1359 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1360 inet
->pmtudisc
== IP_PMTUDISC_PROBE
||
1361 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1362 ip_dont_fragment(sk
, &rt
->dst
)))
1365 if (cork
->flags
& IPCORK_OPT
)
1370 else if (rt
->rt_type
== RTN_MULTICAST
)
1373 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1378 iph
->tos
= (cork
->tos
!= -1) ? cork
->tos
: inet
->tos
;
1381 iph
->protocol
= sk
->sk_protocol
;
1382 ip_copy_addrs(iph
, fl4
);
1383 ip_select_ident(net
, skb
, sk
);
1386 iph
->ihl
+= opt
->optlen
>>2;
1387 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1390 skb
->priority
= (cork
->tos
!= -1) ? cork
->priority
: sk
->sk_priority
;
1391 skb
->mark
= sk
->sk_mark
;
1393 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1397 skb_dst_set(skb
, &rt
->dst
);
1399 if (iph
->protocol
== IPPROTO_ICMP
)
1400 icmp_out_count(net
, ((struct icmphdr
*)
1401 skb_transport_header(skb
))->type
);
1403 ip_cork_release(cork
);
1408 int ip_send_skb(struct net
*net
, struct sk_buff
*skb
)
1412 err
= ip_local_out(skb
);
1415 err
= net_xmit_errno(err
);
1417 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1423 int ip_push_pending_frames(struct sock
*sk
, struct flowi4
*fl4
)
1425 struct sk_buff
*skb
;
1427 skb
= ip_finish_skb(sk
, fl4
);
1431 /* Netfilter gets whole the not fragmented skb. */
1432 return ip_send_skb(sock_net(sk
), skb
);
1436 * Throw away all pending data on the socket.
1438 static void __ip_flush_pending_frames(struct sock
*sk
,
1439 struct sk_buff_head
*queue
,
1440 struct inet_cork
*cork
)
1442 struct sk_buff
*skb
;
1444 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1447 ip_cork_release(cork
);
1450 void ip_flush_pending_frames(struct sock
*sk
)
1452 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
.base
);
1455 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1457 int getfrag(void *from
, char *to
, int offset
,
1458 int len
, int odd
, struct sk_buff
*skb
),
1459 void *from
, int length
, int transhdrlen
,
1460 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1463 struct inet_cork cork
;
1464 struct sk_buff_head queue
;
1467 if (flags
& MSG_PROBE
)
1470 __skb_queue_head_init(&queue
);
1475 err
= ip_setup_cork(sk
, &cork
, ipc
, rtp
);
1477 return ERR_PTR(err
);
1479 err
= __ip_append_data(sk
, fl4
, &queue
, &cork
,
1480 ¤t
->task_frag
, getfrag
,
1481 from
, length
, transhdrlen
, flags
);
1483 __ip_flush_pending_frames(sk
, &queue
, &cork
);
1484 return ERR_PTR(err
);
1487 return __ip_make_skb(sk
, fl4
, &queue
, &cork
);
1491 * Fetch data from kernel space and fill in checksum if needed.
1493 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1494 int len
, int odd
, struct sk_buff
*skb
)
1498 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1499 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1504 * Generic function to send a packet as reply to another packet.
1505 * Used to send some TCP resets/acks so far.
1507 void ip_send_unicast_reply(struct sock
*sk
, struct sk_buff
*skb
,
1508 const struct ip_options
*sopt
,
1509 __be32 daddr
, __be32 saddr
,
1510 const struct ip_reply_arg
*arg
,
1513 struct ip_options_data replyopts
;
1514 struct ipcm_cookie ipc
;
1516 struct rtable
*rt
= skb_rtable(skb
);
1517 struct net
*net
= sock_net(sk
);
1518 struct sk_buff
*nskb
;
1521 if (__ip_options_echo(&replyopts
.opt
.opt
, skb
, sopt
))
1530 if (replyopts
.opt
.opt
.optlen
) {
1531 ipc
.opt
= &replyopts
.opt
;
1533 if (replyopts
.opt
.opt
.srr
)
1534 daddr
= replyopts
.opt
.opt
.faddr
;
1537 flowi4_init_output(&fl4
, arg
->bound_dev_if
,
1538 IP4_REPLY_MARK(net
, skb
->mark
),
1540 RT_SCOPE_UNIVERSE
, ip_hdr(skb
)->protocol
,
1541 ip_reply_arg_flowi_flags(arg
),
1543 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
);
1544 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1545 rt
= ip_route_output_key(net
, &fl4
);
1549 inet_sk(sk
)->tos
= arg
->tos
;
1551 sk
->sk_priority
= skb
->priority
;
1552 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1553 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1554 sk
->sk_sndbuf
= sysctl_wmem_default
;
1555 err
= ip_append_data(sk
, &fl4
, ip_reply_glue_bits
, arg
->iov
->iov_base
,
1556 len
, 0, &ipc
, &rt
, MSG_DONTWAIT
);
1557 if (unlikely(err
)) {
1558 ip_flush_pending_frames(sk
);
1562 nskb
= skb_peek(&sk
->sk_write_queue
);
1564 if (arg
->csumoffset
>= 0)
1565 *((__sum16
*)skb_transport_header(nskb
) +
1566 arg
->csumoffset
) = csum_fold(csum_add(nskb
->csum
,
1568 nskb
->ip_summed
= CHECKSUM_NONE
;
1569 skb_set_queue_mapping(nskb
, skb_get_queue_mapping(skb
));
1570 ip_push_pending_frames(sk
, &fl4
);
1576 void __init
ip_init(void)
1581 #if defined(CONFIG_IP_MULTICAST)