1 // SPDX-License-Identifier: GPL-2.0-only
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.
7 * The Internet Protocol (IP) output module.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Donald Becker, <becker@super.org>
12 * Alan Cox, <Alan.Cox@linux.org>
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>
19 * See ip_input.c for original log
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
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
43 * Hirokazu Takahashi: sendfile() on UDP works now.
46 #include <linux/uaccess.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54 #include <linux/slab.h>
56 #include <linux/socket.h>
57 #include <linux/sockios.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>
68 #include <net/protocol.h>
69 #include <net/route.h>
71 #include <linux/skbuff.h>
75 #include <net/checksum.h>
76 #include <net/inetpeer.h>
77 #include <net/lwtunnel.h>
78 #include <linux/bpf-cgroup.h>
79 #include <linux/igmp.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/netfilter_bridge.h>
82 #include <linux/netlink.h>
83 #include <linux/tcp.h>
86 ip_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
88 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*));
90 /* Generate a checksum for an outgoing IP datagram. */
91 void ip_send_check(struct iphdr
*iph
)
94 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
96 EXPORT_SYMBOL(ip_send_check
);
98 int __ip_local_out(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
100 struct iphdr
*iph
= ip_hdr(skb
);
102 iph
->tot_len
= htons(skb
->len
);
105 /* if egress device is enslaved to an L3 master device pass the
106 * skb to its handler for processing
108 skb
= l3mdev_ip_out(sk
, skb
);
112 skb
->protocol
= htons(ETH_P_IP
);
114 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
,
115 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
119 int ip_local_out(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
123 err
= __ip_local_out(net
, sk
, skb
);
124 if (likely(err
== 1))
125 err
= dst_output(net
, sk
, skb
);
129 EXPORT_SYMBOL_GPL(ip_local_out
);
131 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
133 int ttl
= inet
->uc_ttl
;
136 ttl
= ip4_dst_hoplimit(dst
);
141 * Add an ip header to a skbuff and send it out.
144 int ip_build_and_send_pkt(struct sk_buff
*skb
, const struct sock
*sk
,
145 __be32 saddr
, __be32 daddr
, struct ip_options_rcu
*opt
)
147 struct inet_sock
*inet
= inet_sk(sk
);
148 struct rtable
*rt
= skb_rtable(skb
);
149 struct net
*net
= sock_net(sk
);
152 /* Build the IP header. */
153 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->opt
.optlen
: 0));
154 skb_reset_network_header(skb
);
158 iph
->tos
= inet
->tos
;
159 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
160 iph
->daddr
= (opt
&& opt
->opt
.srr
? opt
->opt
.faddr
: daddr
);
162 iph
->protocol
= sk
->sk_protocol
;
163 if (ip_dont_fragment(sk
, &rt
->dst
)) {
164 iph
->frag_off
= htons(IP_DF
);
168 __ip_select_ident(net
, iph
, 1);
171 if (opt
&& opt
->opt
.optlen
) {
172 iph
->ihl
+= opt
->opt
.optlen
>>2;
173 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
176 skb
->priority
= sk
->sk_priority
;
178 skb
->mark
= sk
->sk_mark
;
181 return ip_local_out(net
, skb
->sk
, skb
);
183 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
185 static int ip_finish_output2(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
187 struct dst_entry
*dst
= skb_dst(skb
);
188 struct rtable
*rt
= (struct rtable
*)dst
;
189 struct net_device
*dev
= dst
->dev
;
190 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
191 struct neighbour
*neigh
;
192 bool is_v6gw
= false;
194 if (rt
->rt_type
== RTN_MULTICAST
) {
195 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTMCAST
, skb
->len
);
196 } else if (rt
->rt_type
== RTN_BROADCAST
)
197 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTBCAST
, skb
->len
);
199 /* Be paranoid, rather than too clever. */
200 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
201 struct sk_buff
*skb2
;
203 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
209 skb_set_owner_w(skb2
, skb
->sk
);
214 if (lwtunnel_xmit_redirect(dst
->lwtstate
)) {
215 int res
= lwtunnel_xmit(skb
);
217 if (res
< 0 || res
== LWTUNNEL_XMIT_DONE
)
222 neigh
= ip_neigh_for_gw(rt
, skb
, &is_v6gw
);
223 if (!IS_ERR(neigh
)) {
226 sock_confirm_neigh(skb
, neigh
);
227 /* if crossing protocols, can not use the cached header */
228 res
= neigh_output(neigh
, skb
, is_v6gw
);
229 rcu_read_unlock_bh();
232 rcu_read_unlock_bh();
234 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
240 static int ip_finish_output_gso(struct net
*net
, struct sock
*sk
,
241 struct sk_buff
*skb
, unsigned int mtu
)
243 netdev_features_t features
;
244 struct sk_buff
*segs
;
247 /* common case: seglen is <= mtu
249 if (skb_gso_validate_network_len(skb
, mtu
))
250 return ip_finish_output2(net
, sk
, skb
);
252 /* Slowpath - GSO segment length exceeds the egress MTU.
254 * This can happen in several cases:
255 * - Forwarding of a TCP GRO skb, when DF flag is not set.
256 * - Forwarding of an skb that arrived on a virtualization interface
257 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
259 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
260 * interface with a smaller MTU.
261 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
262 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
265 features
= netif_skb_features(skb
);
266 BUILD_BUG_ON(sizeof(*IPCB(skb
)) > SKB_SGO_CB_OFFSET
);
267 segs
= skb_gso_segment(skb
, features
& ~NETIF_F_GSO_MASK
);
268 if (IS_ERR_OR_NULL(segs
)) {
276 struct sk_buff
*nskb
= segs
->next
;
279 skb_mark_not_on_list(segs
);
280 err
= ip_fragment(net
, sk
, segs
, mtu
, ip_finish_output2
);
290 static int ip_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
295 ret
= BPF_CGROUP_RUN_PROG_INET_EGRESS(sk
, skb
);
301 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
302 /* Policy lookup after SNAT yielded a new policy */
303 if (skb_dst(skb
)->xfrm
) {
304 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
305 return dst_output(net
, sk
, skb
);
308 mtu
= ip_skb_dst_mtu(sk
, skb
);
310 return ip_finish_output_gso(net
, sk
, skb
, mtu
);
312 if (skb
->len
> mtu
|| (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
))
313 return ip_fragment(net
, sk
, skb
, mtu
, ip_finish_output2
);
315 return ip_finish_output2(net
, sk
, skb
);
318 static int ip_mc_finish_output(struct net
*net
, struct sock
*sk
,
323 ret
= BPF_CGROUP_RUN_PROG_INET_EGRESS(sk
, skb
);
329 return dev_loopback_xmit(net
, sk
, skb
);
332 int ip_mc_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
334 struct rtable
*rt
= skb_rtable(skb
);
335 struct net_device
*dev
= rt
->dst
.dev
;
338 * If the indicated interface is up and running, send the packet.
340 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
343 skb
->protocol
= htons(ETH_P_IP
);
346 * Multicasts are looped back for other local users
349 if (rt
->rt_flags
&RTCF_MULTICAST
) {
351 #ifdef CONFIG_IP_MROUTE
352 /* Small optimization: do not loopback not local frames,
353 which returned after forwarding; they will be dropped
354 by ip_mr_input in any case.
355 Note, that local frames are looped back to be delivered
358 This check is duplicated in ip_mr_input at the moment.
361 ((rt
->rt_flags
& RTCF_LOCAL
) ||
362 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
365 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
367 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
368 net
, sk
, newskb
, NULL
, newskb
->dev
,
369 ip_mc_finish_output
);
372 /* Multicasts with ttl 0 must not go beyond the host */
374 if (ip_hdr(skb
)->ttl
== 0) {
380 if (rt
->rt_flags
&RTCF_BROADCAST
) {
381 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
383 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
384 net
, sk
, newskb
, NULL
, newskb
->dev
,
385 ip_mc_finish_output
);
388 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
389 net
, sk
, skb
, NULL
, skb
->dev
,
391 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
394 int ip_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
396 struct net_device
*dev
= skb_dst(skb
)->dev
;
398 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
401 skb
->protocol
= htons(ETH_P_IP
);
403 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
404 net
, sk
, skb
, NULL
, dev
,
406 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
410 * copy saddr and daddr, possibly using 64bit load/stores
412 * iph->saddr = fl4->saddr;
413 * iph->daddr = fl4->daddr;
415 static void ip_copy_addrs(struct iphdr
*iph
, const struct flowi4
*fl4
)
417 BUILD_BUG_ON(offsetof(typeof(*fl4
), daddr
) !=
418 offsetof(typeof(*fl4
), saddr
) + sizeof(fl4
->saddr
));
419 memcpy(&iph
->saddr
, &fl4
->saddr
,
420 sizeof(fl4
->saddr
) + sizeof(fl4
->daddr
));
423 /* Note: skb->sk can be different from sk, in case of tunnels */
424 int __ip_queue_xmit(struct sock
*sk
, struct sk_buff
*skb
, struct flowi
*fl
,
427 struct inet_sock
*inet
= inet_sk(sk
);
428 struct net
*net
= sock_net(sk
);
429 struct ip_options_rcu
*inet_opt
;
435 /* Skip all of this if the packet is already routed,
436 * f.e. by something like SCTP.
439 inet_opt
= rcu_dereference(inet
->inet_opt
);
441 rt
= skb_rtable(skb
);
445 /* Make sure we can route this packet. */
446 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
450 /* Use correct destination address if we have options. */
451 daddr
= inet
->inet_daddr
;
452 if (inet_opt
&& inet_opt
->opt
.srr
)
453 daddr
= inet_opt
->opt
.faddr
;
455 /* If this fails, retransmit mechanism of transport layer will
456 * keep trying until route appears or the connection times
459 rt
= ip_route_output_ports(net
, fl4
, sk
,
460 daddr
, inet
->inet_saddr
,
464 RT_CONN_FLAGS_TOS(sk
, tos
),
465 sk
->sk_bound_dev_if
);
468 sk_setup_caps(sk
, &rt
->dst
);
470 skb_dst_set_noref(skb
, &rt
->dst
);
473 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_gw_family
)
476 /* OK, we know where to send it, allocate and build IP header. */
477 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
478 skb_reset_network_header(skb
);
480 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (tos
& 0xff));
481 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->ignore_df
)
482 iph
->frag_off
= htons(IP_DF
);
485 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
486 iph
->protocol
= sk
->sk_protocol
;
487 ip_copy_addrs(iph
, fl4
);
489 /* Transport layer set skb->h.foo itself. */
491 if (inet_opt
&& inet_opt
->opt
.optlen
) {
492 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
493 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
496 ip_select_ident_segs(net
, skb
, sk
,
497 skb_shinfo(skb
)->gso_segs
?: 1);
499 /* TODO : should we use skb->sk here instead of sk ? */
500 skb
->priority
= sk
->sk_priority
;
501 skb
->mark
= sk
->sk_mark
;
503 res
= ip_local_out(net
, sk
, skb
);
509 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
511 return -EHOSTUNREACH
;
513 EXPORT_SYMBOL(__ip_queue_xmit
);
515 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
517 to
->pkt_type
= from
->pkt_type
;
518 to
->priority
= from
->priority
;
519 to
->protocol
= from
->protocol
;
520 to
->skb_iif
= from
->skb_iif
;
522 skb_dst_copy(to
, from
);
524 to
->mark
= from
->mark
;
526 skb_copy_hash(to
, from
);
528 /* Copy the flags to each fragment. */
529 IPCB(to
)->flags
= IPCB(from
)->flags
;
531 #ifdef CONFIG_NET_SCHED
532 to
->tc_index
= from
->tc_index
;
535 skb_ext_copy(to
, from
);
536 #if IS_ENABLED(CONFIG_IP_VS)
537 to
->ipvs_property
= from
->ipvs_property
;
539 skb_copy_secmark(to
, from
);
542 static int ip_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
544 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*))
546 struct iphdr
*iph
= ip_hdr(skb
);
548 if ((iph
->frag_off
& htons(IP_DF
)) == 0)
549 return ip_do_fragment(net
, sk
, skb
, output
);
551 if (unlikely(!skb
->ignore_df
||
552 (IPCB(skb
)->frag_max_size
&&
553 IPCB(skb
)->frag_max_size
> mtu
))) {
554 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
555 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
561 return ip_do_fragment(net
, sk
, skb
, output
);
565 * This IP datagram is too large to be sent in one piece. Break it up into
566 * smaller pieces (each of size equal to IP header plus
567 * a block of the data of the original IP data part) that will yet fit in a
568 * single device frame, and queue such a frame for sending.
571 int ip_do_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
572 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*))
576 struct sk_buff
*skb2
;
577 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
579 __be16 not_last_frag
;
580 struct rtable
*rt
= skb_rtable(skb
);
583 /* for offloaded checksums cleanup checksum before fragmentation */
584 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
585 (err
= skb_checksum_help(skb
)))
589 * Point into the IP datagram header.
594 mtu
= ip_skb_dst_mtu(sk
, skb
);
595 if (IPCB(skb
)->frag_max_size
&& IPCB(skb
)->frag_max_size
< mtu
)
596 mtu
= IPCB(skb
)->frag_max_size
;
599 * Setup starting values.
603 mtu
= mtu
- hlen
; /* Size of data space */
604 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
605 ll_rs
= LL_RESERVED_SPACE(rt
->dst
.dev
);
607 /* When frag_list is given, use it. First, check its validity:
608 * some transformers could create wrong frag_list or break existing
609 * one, it is not prohibited. In this case fall back to copying.
611 * LATER: this step can be merged to real generation of fragments,
612 * we can switch to copy when see the first bad fragment.
614 if (skb_has_frag_list(skb
)) {
615 struct sk_buff
*frag
, *frag2
;
616 unsigned int first_len
= skb_pagelen(skb
);
618 if (first_len
- hlen
> mtu
||
619 ((first_len
- hlen
) & 7) ||
620 ip_is_fragment(iph
) ||
622 skb_headroom(skb
) < ll_rs
)
625 skb_walk_frags(skb
, frag
) {
626 /* Correct geometry. */
627 if (frag
->len
> mtu
||
628 ((frag
->len
& 7) && frag
->next
) ||
629 skb_headroom(frag
) < hlen
+ ll_rs
)
630 goto slow_path_clean
;
632 /* Partially cloned skb? */
633 if (skb_shared(frag
))
634 goto slow_path_clean
;
639 frag
->destructor
= sock_wfree
;
641 skb
->truesize
-= frag
->truesize
;
644 /* Everything is OK. Generate! */
648 frag
= skb_shinfo(skb
)->frag_list
;
649 skb_frag_list_init(skb
);
650 skb
->data_len
= first_len
- skb_headlen(skb
);
651 skb
->len
= first_len
;
652 iph
->tot_len
= htons(first_len
);
653 iph
->frag_off
= htons(IP_MF
);
657 /* Prepare header of the next frame,
658 * before previous one went down. */
660 frag
->ip_summed
= CHECKSUM_NONE
;
661 skb_reset_transport_header(frag
);
662 __skb_push(frag
, hlen
);
663 skb_reset_network_header(frag
);
664 memcpy(skb_network_header(frag
), iph
, hlen
);
666 iph
->tot_len
= htons(frag
->len
);
667 ip_copy_metadata(frag
, skb
);
669 ip_options_fragment(frag
);
670 offset
+= skb
->len
- hlen
;
671 iph
->frag_off
= htons(offset
>>3);
673 iph
->frag_off
|= htons(IP_MF
);
674 /* Ready, complete checksum */
678 err
= output(net
, sk
, skb
);
681 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
687 skb_mark_not_on_list(skb
);
691 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
695 kfree_skb_list(frag
);
697 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
701 skb_walk_frags(skb
, frag2
) {
705 frag2
->destructor
= NULL
;
706 skb
->truesize
+= frag2
->truesize
;
713 left
= skb
->len
- hlen
; /* Space per frame */
714 ptr
= hlen
; /* Where to start from */
717 * Fragment the datagram.
720 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
721 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
724 * Keep copying data until we run out.
729 /* IF: it doesn't fit, use 'mtu' - the data space left */
732 /* IF: we are not sending up to and including the packet end
733 then align the next start on an eight byte boundary */
738 /* Allocate buffer */
739 skb2
= alloc_skb(len
+ hlen
+ ll_rs
, GFP_ATOMIC
);
746 * Set up data on packet
749 ip_copy_metadata(skb2
, skb
);
750 skb_reserve(skb2
, ll_rs
);
751 skb_put(skb2
, len
+ hlen
);
752 skb_reset_network_header(skb2
);
753 skb2
->transport_header
= skb2
->network_header
+ hlen
;
756 * Charge the memory for the fragment to any owner
761 skb_set_owner_w(skb2
, skb
->sk
);
764 * Copy the packet header into the new buffer.
767 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
770 * Copy a block of the IP datagram.
772 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
777 * Fill in the new header fields.
780 iph
->frag_off
= htons((offset
>> 3));
782 if (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
)
783 iph
->frag_off
|= htons(IP_DF
);
785 /* ANK: dirty, but effective trick. Upgrade options only if
786 * the segment to be fragmented was THE FIRST (otherwise,
787 * options are already fixed) and make it ONCE
788 * on the initial skb, so that all the following fragments
789 * will inherit fixed options.
792 ip_options_fragment(skb
);
795 * Added AC : If we are fragmenting a fragment that's not the
796 * last fragment then keep MF on each bit
798 if (left
> 0 || not_last_frag
)
799 iph
->frag_off
|= htons(IP_MF
);
804 * Put this fragment into the sending queue.
806 iph
->tot_len
= htons(len
+ hlen
);
810 err
= output(net
, sk
, skb2
);
814 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
817 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
822 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
825 EXPORT_SYMBOL(ip_do_fragment
);
828 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
830 struct msghdr
*msg
= from
;
832 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
833 if (!copy_from_iter_full(to
, len
, &msg
->msg_iter
))
837 if (!csum_and_copy_from_iter_full(to
, len
, &csum
, &msg
->msg_iter
))
839 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
843 EXPORT_SYMBOL(ip_generic_getfrag
);
846 csum_page(struct page
*page
, int offset
, int copy
)
851 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
856 static int __ip_append_data(struct sock
*sk
,
858 struct sk_buff_head
*queue
,
859 struct inet_cork
*cork
,
860 struct page_frag
*pfrag
,
861 int getfrag(void *from
, char *to
, int offset
,
862 int len
, int odd
, struct sk_buff
*skb
),
863 void *from
, int length
, int transhdrlen
,
866 struct inet_sock
*inet
= inet_sk(sk
);
867 struct ubuf_info
*uarg
= NULL
;
870 struct ip_options
*opt
= cork
->opt
;
877 unsigned int maxfraglen
, fragheaderlen
, maxnonfragsize
;
878 int csummode
= CHECKSUM_NONE
;
879 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
880 unsigned int wmem_alloc_delta
= 0;
881 bool paged
, extra_uref
;
884 skb
= skb_peek_tail(queue
);
886 exthdrlen
= !skb
? rt
->dst
.header_len
: 0;
887 mtu
= cork
->gso_size
? IP_MAX_MTU
: cork
->fragsize
;
888 paged
= !!cork
->gso_size
;
890 if (cork
->tx_flags
& SKBTX_ANY_SW_TSTAMP
&&
891 sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)
892 tskey
= sk
->sk_tskey
++;
894 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
896 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
897 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
898 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
900 if (cork
->length
+ length
> maxnonfragsize
- fragheaderlen
) {
901 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
902 mtu
- (opt
? opt
->optlen
: 0));
907 * transhdrlen > 0 means that this is the first fragment and we wish
908 * it won't be fragmented in the future.
911 length
+ fragheaderlen
<= mtu
&&
912 rt
->dst
.dev
->features
& (NETIF_F_HW_CSUM
| NETIF_F_IP_CSUM
) &&
913 (!(flags
& MSG_MORE
) || cork
->gso_size
) &&
914 (!exthdrlen
|| (rt
->dst
.dev
->features
& NETIF_F_HW_ESP_TX_CSUM
)))
915 csummode
= CHECKSUM_PARTIAL
;
917 if (flags
& MSG_ZEROCOPY
&& length
&& sock_flag(sk
, SOCK_ZEROCOPY
)) {
918 uarg
= sock_zerocopy_realloc(sk
, length
, skb_zcopy(skb
));
922 if (rt
->dst
.dev
->features
& NETIF_F_SG
&&
923 csummode
== CHECKSUM_PARTIAL
) {
927 skb_zcopy_set(skb
, uarg
, &extra_uref
);
931 cork
->length
+= length
;
933 /* So, what's going on in the loop below?
935 * We use calculated fragment length to generate chained skb,
936 * each of segments is IP fragment ready for sending to network after
937 * adding appropriate IP header.
944 /* Check if the remaining data fits into current packet. */
945 copy
= mtu
- skb
->len
;
947 copy
= maxfraglen
- skb
->len
;
950 unsigned int datalen
;
951 unsigned int fraglen
;
952 unsigned int fraggap
;
953 unsigned int alloclen
;
954 unsigned int pagedlen
;
955 struct sk_buff
*skb_prev
;
959 fraggap
= skb_prev
->len
- maxfraglen
;
964 * If remaining data exceeds the mtu,
965 * we know we need more fragment(s).
967 datalen
= length
+ fraggap
;
968 if (datalen
> mtu
- fragheaderlen
)
969 datalen
= maxfraglen
- fragheaderlen
;
970 fraglen
= datalen
+ fragheaderlen
;
973 if ((flags
& MSG_MORE
) &&
974 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
979 alloclen
= min_t(int, fraglen
, MAX_HEADER
);
980 pagedlen
= fraglen
- alloclen
;
983 alloclen
+= exthdrlen
;
985 /* The last fragment gets additional space at tail.
986 * Note, with MSG_MORE we overallocate on fragments,
987 * because we have no idea what fragment will be
990 if (datalen
== length
+ fraggap
)
991 alloclen
+= rt
->dst
.trailer_len
;
994 skb
= sock_alloc_send_skb(sk
,
995 alloclen
+ hh_len
+ 15,
996 (flags
& MSG_DONTWAIT
), &err
);
999 if (refcount_read(&sk
->sk_wmem_alloc
) + wmem_alloc_delta
<=
1001 skb
= alloc_skb(alloclen
+ hh_len
+ 15,
1010 * Fill in the control structures
1012 skb
->ip_summed
= csummode
;
1014 skb_reserve(skb
, hh_len
);
1017 * Find where to start putting bytes.
1019 data
= skb_put(skb
, fraglen
+ exthdrlen
- pagedlen
);
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
- pagedlen
;
1036 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1043 length
-= copy
+ transhdrlen
;
1046 csummode
= CHECKSUM_NONE
;
1048 /* only the initial fragment is time stamped */
1049 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
1051 skb_shinfo(skb
)->tskey
= tskey
;
1053 skb_zcopy_set(skb
, uarg
, &extra_uref
);
1055 if ((flags
& MSG_CONFIRM
) && !skb_prev
)
1056 skb_set_dst_pending_confirm(skb
, 1);
1059 * Put the packet on the pending queue.
1061 if (!skb
->destructor
) {
1062 skb
->destructor
= sock_wfree
;
1064 wmem_alloc_delta
+= skb
->truesize
;
1066 __skb_queue_tail(queue
, skb
);
1073 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
) &&
1074 skb_tailroom(skb
) >= copy
) {
1078 if (getfrag(from
, skb_put(skb
, copy
),
1079 offset
, copy
, off
, skb
) < 0) {
1080 __skb_trim(skb
, off
);
1084 } else if (!uarg
|| !uarg
->zerocopy
) {
1085 int i
= skb_shinfo(skb
)->nr_frags
;
1088 if (!sk_page_frag_refill(sk
, pfrag
))
1091 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
1094 if (i
== MAX_SKB_FRAGS
)
1097 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1099 skb_shinfo(skb
)->nr_frags
= ++i
;
1100 get_page(pfrag
->page
);
1102 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1104 page_address(pfrag
->page
) + pfrag
->offset
,
1105 offset
, copy
, skb
->len
, skb
) < 0)
1108 pfrag
->offset
+= copy
;
1109 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1111 skb
->data_len
+= copy
;
1112 skb
->truesize
+= copy
;
1113 wmem_alloc_delta
+= copy
;
1115 err
= skb_zerocopy_iter_dgram(skb
, from
, copy
);
1123 if (wmem_alloc_delta
)
1124 refcount_add(wmem_alloc_delta
, &sk
->sk_wmem_alloc
);
1131 sock_zerocopy_put_abort(uarg
, extra_uref
);
1132 cork
->length
-= length
;
1133 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1134 refcount_add(wmem_alloc_delta
, &sk
->sk_wmem_alloc
);
1138 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1139 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1141 struct ip_options_rcu
*opt
;
1149 * setup for corking.
1154 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1156 if (unlikely(!cork
->opt
))
1159 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1160 cork
->flags
|= IPCORK_OPT
;
1161 cork
->addr
= ipc
->addr
;
1165 * We steal reference to this route, caller should not release it
1168 cork
->fragsize
= ip_sk_use_pmtu(sk
) ?
1169 dst_mtu(&rt
->dst
) : rt
->dst
.dev
->mtu
;
1171 cork
->gso_size
= ipc
->gso_size
;
1172 cork
->dst
= &rt
->dst
;
1174 cork
->ttl
= ipc
->ttl
;
1175 cork
->tos
= ipc
->tos
;
1176 cork
->priority
= ipc
->priority
;
1177 cork
->transmit_time
= ipc
->sockc
.transmit_time
;
1179 sock_tx_timestamp(sk
, ipc
->sockc
.tsflags
, &cork
->tx_flags
);
1185 * ip_append_data() and ip_append_page() can make one large IP datagram
1186 * from many pieces of data. Each pieces will be holded on the socket
1187 * until ip_push_pending_frames() is called. Each piece can be a page
1190 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1191 * this interface potentially.
1193 * LATER: length must be adjusted by pad at tail, when it is required.
1195 int ip_append_data(struct sock
*sk
, struct flowi4
*fl4
,
1196 int getfrag(void *from
, char *to
, int offset
, int len
,
1197 int odd
, struct sk_buff
*skb
),
1198 void *from
, int length
, int transhdrlen
,
1199 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1202 struct inet_sock
*inet
= inet_sk(sk
);
1205 if (flags
&MSG_PROBE
)
1208 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1209 err
= ip_setup_cork(sk
, &inet
->cork
.base
, ipc
, rtp
);
1216 return __ip_append_data(sk
, fl4
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1217 sk_page_frag(sk
), getfrag
,
1218 from
, length
, transhdrlen
, flags
);
1221 ssize_t
ip_append_page(struct sock
*sk
, struct flowi4
*fl4
, struct page
*page
,
1222 int offset
, size_t size
, int flags
)
1224 struct inet_sock
*inet
= inet_sk(sk
);
1225 struct sk_buff
*skb
;
1227 struct ip_options
*opt
= NULL
;
1228 struct inet_cork
*cork
;
1233 unsigned int maxfraglen
, fragheaderlen
, fraggap
, maxnonfragsize
;
1238 if (flags
&MSG_PROBE
)
1241 if (skb_queue_empty(&sk
->sk_write_queue
))
1244 cork
= &inet
->cork
.base
;
1245 rt
= (struct rtable
*)cork
->dst
;
1246 if (cork
->flags
& IPCORK_OPT
)
1249 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1252 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1253 mtu
= cork
->gso_size
? IP_MAX_MTU
: cork
->fragsize
;
1255 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1256 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1257 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
1259 if (cork
->length
+ size
> maxnonfragsize
- fragheaderlen
) {
1260 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
1261 mtu
- (opt
? opt
->optlen
: 0));
1265 skb
= skb_peek_tail(&sk
->sk_write_queue
);
1269 cork
->length
+= size
;
1272 /* Check if the remaining data fits into current packet. */
1273 len
= mtu
- skb
->len
;
1275 len
= maxfraglen
- skb
->len
;
1278 struct sk_buff
*skb_prev
;
1282 fraggap
= skb_prev
->len
- maxfraglen
;
1284 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1285 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1286 if (unlikely(!skb
)) {
1292 * Fill in the control structures
1294 skb
->ip_summed
= CHECKSUM_NONE
;
1296 skb_reserve(skb
, hh_len
);
1299 * Find where to start putting bytes.
1301 skb_put(skb
, fragheaderlen
+ fraggap
);
1302 skb_reset_network_header(skb
);
1303 skb
->transport_header
= (skb
->network_header
+
1306 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1308 skb_transport_header(skb
),
1310 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1312 pskb_trim_unique(skb_prev
, maxfraglen
);
1316 * Put the packet on the pending queue.
1318 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1325 if (skb_append_pagefrags(skb
, page
, offset
, len
)) {
1330 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1332 csum
= csum_page(page
, offset
, len
);
1333 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1337 skb
->data_len
+= len
;
1338 skb
->truesize
+= len
;
1339 refcount_add(len
, &sk
->sk_wmem_alloc
);
1346 cork
->length
-= size
;
1347 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1351 static void ip_cork_release(struct inet_cork
*cork
)
1353 cork
->flags
&= ~IPCORK_OPT
;
1356 dst_release(cork
->dst
);
1361 * Combined all pending IP fragments on the socket as one IP datagram
1362 * and push them out.
1364 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1366 struct sk_buff_head
*queue
,
1367 struct inet_cork
*cork
)
1369 struct sk_buff
*skb
, *tmp_skb
;
1370 struct sk_buff
**tail_skb
;
1371 struct inet_sock
*inet
= inet_sk(sk
);
1372 struct net
*net
= sock_net(sk
);
1373 struct ip_options
*opt
= NULL
;
1374 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1379 skb
= __skb_dequeue(queue
);
1382 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1384 /* move skb->data to ip header from ext header */
1385 if (skb
->data
< skb_network_header(skb
))
1386 __skb_pull(skb
, skb_network_offset(skb
));
1387 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1388 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1389 *tail_skb
= tmp_skb
;
1390 tail_skb
= &(tmp_skb
->next
);
1391 skb
->len
+= tmp_skb
->len
;
1392 skb
->data_len
+= tmp_skb
->len
;
1393 skb
->truesize
+= tmp_skb
->truesize
;
1394 tmp_skb
->destructor
= NULL
;
1398 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1399 * to fragment the frame generated here. No matter, what transforms
1400 * how transforms change size of the packet, it will come out.
1402 skb
->ignore_df
= ip_sk_ignore_df(sk
);
1404 /* DF bit is set when we want to see DF on outgoing frames.
1405 * If ignore_df is set too, we still allow to fragment this frame
1407 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1408 inet
->pmtudisc
== IP_PMTUDISC_PROBE
||
1409 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1410 ip_dont_fragment(sk
, &rt
->dst
)))
1413 if (cork
->flags
& IPCORK_OPT
)
1418 else if (rt
->rt_type
== RTN_MULTICAST
)
1421 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1426 iph
->tos
= (cork
->tos
!= -1) ? cork
->tos
: inet
->tos
;
1429 iph
->protocol
= sk
->sk_protocol
;
1430 ip_copy_addrs(iph
, fl4
);
1431 ip_select_ident(net
, skb
, sk
);
1434 iph
->ihl
+= opt
->optlen
>>2;
1435 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1438 skb
->priority
= (cork
->tos
!= -1) ? cork
->priority
: sk
->sk_priority
;
1439 skb
->mark
= sk
->sk_mark
;
1440 skb
->tstamp
= cork
->transmit_time
;
1442 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1446 skb_dst_set(skb
, &rt
->dst
);
1448 if (iph
->protocol
== IPPROTO_ICMP
)
1449 icmp_out_count(net
, ((struct icmphdr
*)
1450 skb_transport_header(skb
))->type
);
1452 ip_cork_release(cork
);
1457 int ip_send_skb(struct net
*net
, struct sk_buff
*skb
)
1461 err
= ip_local_out(net
, skb
->sk
, skb
);
1464 err
= net_xmit_errno(err
);
1466 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1472 int ip_push_pending_frames(struct sock
*sk
, struct flowi4
*fl4
)
1474 struct sk_buff
*skb
;
1476 skb
= ip_finish_skb(sk
, fl4
);
1480 /* Netfilter gets whole the not fragmented skb. */
1481 return ip_send_skb(sock_net(sk
), skb
);
1485 * Throw away all pending data on the socket.
1487 static void __ip_flush_pending_frames(struct sock
*sk
,
1488 struct sk_buff_head
*queue
,
1489 struct inet_cork
*cork
)
1491 struct sk_buff
*skb
;
1493 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1496 ip_cork_release(cork
);
1499 void ip_flush_pending_frames(struct sock
*sk
)
1501 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
.base
);
1504 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1506 int getfrag(void *from
, char *to
, int offset
,
1507 int len
, int odd
, struct sk_buff
*skb
),
1508 void *from
, int length
, int transhdrlen
,
1509 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1510 struct inet_cork
*cork
, unsigned int flags
)
1512 struct sk_buff_head queue
;
1515 if (flags
& MSG_PROBE
)
1518 __skb_queue_head_init(&queue
);
1523 err
= ip_setup_cork(sk
, cork
, ipc
, rtp
);
1525 return ERR_PTR(err
);
1527 err
= __ip_append_data(sk
, fl4
, &queue
, cork
,
1528 ¤t
->task_frag
, getfrag
,
1529 from
, length
, transhdrlen
, flags
);
1531 __ip_flush_pending_frames(sk
, &queue
, cork
);
1532 return ERR_PTR(err
);
1535 return __ip_make_skb(sk
, fl4
, &queue
, cork
);
1539 * Fetch data from kernel space and fill in checksum if needed.
1541 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1542 int len
, int odd
, struct sk_buff
*skb
)
1546 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1547 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1552 * Generic function to send a packet as reply to another packet.
1553 * Used to send some TCP resets/acks so far.
1555 void ip_send_unicast_reply(struct sock
*sk
, struct sk_buff
*skb
,
1556 const struct ip_options
*sopt
,
1557 __be32 daddr
, __be32 saddr
,
1558 const struct ip_reply_arg
*arg
,
1561 struct ip_options_data replyopts
;
1562 struct ipcm_cookie ipc
;
1564 struct rtable
*rt
= skb_rtable(skb
);
1565 struct net
*net
= sock_net(sk
);
1566 struct sk_buff
*nskb
;
1570 if (__ip_options_echo(net
, &replyopts
.opt
.opt
, skb
, sopt
))
1576 if (replyopts
.opt
.opt
.optlen
) {
1577 ipc
.opt
= &replyopts
.opt
;
1579 if (replyopts
.opt
.opt
.srr
)
1580 daddr
= replyopts
.opt
.opt
.faddr
;
1583 oif
= arg
->bound_dev_if
;
1584 if (!oif
&& netif_index_is_l3_master(net
, skb
->skb_iif
))
1587 flowi4_init_output(&fl4
, oif
,
1588 IP4_REPLY_MARK(net
, skb
->mark
) ?: sk
->sk_mark
,
1590 RT_SCOPE_UNIVERSE
, ip_hdr(skb
)->protocol
,
1591 ip_reply_arg_flowi_flags(arg
),
1593 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
,
1595 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1596 rt
= ip_route_output_key(net
, &fl4
);
1600 inet_sk(sk
)->tos
= arg
->tos
;
1602 sk
->sk_priority
= skb
->priority
;
1603 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1604 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1605 sk
->sk_sndbuf
= sysctl_wmem_default
;
1606 sk
->sk_mark
= fl4
.flowi4_mark
;
1607 err
= ip_append_data(sk
, &fl4
, ip_reply_glue_bits
, arg
->iov
->iov_base
,
1608 len
, 0, &ipc
, &rt
, MSG_DONTWAIT
);
1609 if (unlikely(err
)) {
1610 ip_flush_pending_frames(sk
);
1614 nskb
= skb_peek(&sk
->sk_write_queue
);
1616 if (arg
->csumoffset
>= 0)
1617 *((__sum16
*)skb_transport_header(nskb
) +
1618 arg
->csumoffset
) = csum_fold(csum_add(nskb
->csum
,
1620 nskb
->ip_summed
= CHECKSUM_NONE
;
1621 ip_push_pending_frames(sk
, &fl4
);
1627 void __init
ip_init(void)
1632 #if defined(CONFIG_IP_MULTICAST)