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 User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #define pr_fmt(fmt) "UDP: " fmt
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
94 #include <linux/errno.h>
95 #include <linux/timer.h>
97 #include <linux/inet.h>
98 #include <linux/netdevice.h>
99 #include <linux/slab.h>
100 #include <net/tcp_states.h>
101 #include <linux/skbuff.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <net/net_namespace.h>
105 #include <net/icmp.h>
106 #include <net/inet_hashtables.h>
107 #include <net/route.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <trace/events/udp.h>
111 #include <linux/static_key.h>
112 #include <trace/events/skb.h>
113 #include <net/busy_poll.h>
114 #include "udp_impl.h"
116 struct udp_table udp_table __read_mostly
;
117 EXPORT_SYMBOL(udp_table
);
119 long sysctl_udp_mem
[3] __read_mostly
;
120 EXPORT_SYMBOL(sysctl_udp_mem
);
122 int sysctl_udp_rmem_min __read_mostly
;
123 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
125 int sysctl_udp_wmem_min __read_mostly
;
126 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
128 atomic_long_t udp_memory_allocated
;
129 EXPORT_SYMBOL(udp_memory_allocated
);
131 #define MAX_UDP_PORTS 65536
132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
134 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
135 const struct udp_hslot
*hslot
,
136 unsigned long *bitmap
,
138 int (*saddr_comp
)(const struct sock
*sk1
,
139 const struct sock
*sk2
),
143 struct hlist_nulls_node
*node
;
144 kuid_t uid
= sock_i_uid(sk
);
146 sk_nulls_for_each(sk2
, node
, &hslot
->head
)
147 if (net_eq(sock_net(sk2
), net
) &&
149 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
150 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
151 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
152 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
153 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
154 !uid_eq(uid
, sock_i_uid(sk2
))) &&
155 (*saddr_comp
)(sk
, sk2
)) {
157 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
169 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
170 struct udp_hslot
*hslot2
,
172 int (*saddr_comp
)(const struct sock
*sk1
,
173 const struct sock
*sk2
))
176 struct hlist_nulls_node
*node
;
177 kuid_t uid
= sock_i_uid(sk
);
180 spin_lock(&hslot2
->lock
);
181 udp_portaddr_for_each_entry(sk2
, node
, &hslot2
->head
)
182 if (net_eq(sock_net(sk2
), net
) &&
184 (udp_sk(sk2
)->udp_port_hash
== num
) &&
185 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
186 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
187 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
188 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
189 !uid_eq(uid
, sock_i_uid(sk2
))) &&
190 (*saddr_comp
)(sk
, sk2
)) {
194 spin_unlock(&hslot2
->lock
);
199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
201 * @sk: socket struct in question
202 * @snum: port number to look up
203 * @saddr_comp: AF-dependent comparison of bound local IP addresses
204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
207 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
208 int (*saddr_comp
)(const struct sock
*sk1
,
209 const struct sock
*sk2
),
210 unsigned int hash2_nulladdr
)
212 struct udp_hslot
*hslot
, *hslot2
;
213 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
215 struct net
*net
= sock_net(sk
);
218 int low
, high
, remaining
;
220 unsigned short first
, last
;
221 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
223 inet_get_local_port_range(net
, &low
, &high
);
224 remaining
= (high
- low
) + 1;
227 first
= (((u64
)rand
* remaining
) >> 32) + low
;
229 * force rand to be an odd multiple of UDP_HTABLE_SIZE
231 rand
= (rand
| 1) * (udptable
->mask
+ 1);
232 last
= first
+ udptable
->mask
+ 1;
234 hslot
= udp_hashslot(udptable
, net
, first
);
235 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
236 spin_lock_bh(&hslot
->lock
);
237 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
238 saddr_comp
, udptable
->log
);
242 * Iterate on all possible values of snum for this hash.
243 * Using steps of an odd multiple of UDP_HTABLE_SIZE
244 * give us randomization and full range coverage.
247 if (low
<= snum
&& snum
<= high
&&
248 !test_bit(snum
>> udptable
->log
, bitmap
) &&
249 !inet_is_reserved_local_port(snum
))
252 } while (snum
!= first
);
253 spin_unlock_bh(&hslot
->lock
);
254 } while (++first
!= last
);
257 hslot
= udp_hashslot(udptable
, net
, snum
);
258 spin_lock_bh(&hslot
->lock
);
259 if (hslot
->count
> 10) {
261 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
263 slot2
&= udptable
->mask
;
264 hash2_nulladdr
&= udptable
->mask
;
266 hslot2
= udp_hashslot2(udptable
, slot2
);
267 if (hslot
->count
< hslot2
->count
)
268 goto scan_primary_hash
;
270 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
272 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
273 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
274 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
283 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
,
288 inet_sk(sk
)->inet_num
= snum
;
289 udp_sk(sk
)->udp_port_hash
= snum
;
290 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
291 if (sk_unhashed(sk
)) {
292 sk_nulls_add_node_rcu(sk
, &hslot
->head
);
294 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
296 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
297 spin_lock(&hslot2
->lock
);
298 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
301 spin_unlock(&hslot2
->lock
);
305 spin_unlock_bh(&hslot
->lock
);
309 EXPORT_SYMBOL(udp_lib_get_port
);
311 static int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
313 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
315 return (!ipv6_only_sock(sk2
) &&
316 (!inet1
->inet_rcv_saddr
|| !inet2
->inet_rcv_saddr
||
317 inet1
->inet_rcv_saddr
== inet2
->inet_rcv_saddr
));
320 static unsigned int udp4_portaddr_hash(struct net
*net
, __be32 saddr
,
323 return jhash_1word((__force u32
)saddr
, net_hash_mix(net
)) ^ port
;
326 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
328 unsigned int hash2_nulladdr
=
329 udp4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
330 unsigned int hash2_partial
=
331 udp4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
333 /* precompute partial secondary hash */
334 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
335 return udp_lib_get_port(sk
, snum
, ipv4_rcv_saddr_equal
, hash2_nulladdr
);
338 static inline int compute_score(struct sock
*sk
, struct net
*net
, __be32 saddr
,
340 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
)
344 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
345 !ipv6_only_sock(sk
)) {
346 struct inet_sock
*inet
= inet_sk(sk
);
348 score
= (sk
->sk_family
== PF_INET
? 2 : 1);
349 if (inet
->inet_rcv_saddr
) {
350 if (inet
->inet_rcv_saddr
!= daddr
)
354 if (inet
->inet_daddr
) {
355 if (inet
->inet_daddr
!= saddr
)
359 if (inet
->inet_dport
) {
360 if (inet
->inet_dport
!= sport
)
364 if (sk
->sk_bound_dev_if
) {
365 if (sk
->sk_bound_dev_if
!= dif
)
374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
376 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
377 __be32 saddr
, __be16 sport
,
378 __be32 daddr
, unsigned int hnum
, int dif
)
382 if (net_eq(sock_net(sk
), net
) && !ipv6_only_sock(sk
)) {
383 struct inet_sock
*inet
= inet_sk(sk
);
385 if (inet
->inet_rcv_saddr
!= daddr
)
387 if (inet
->inet_num
!= hnum
)
390 score
= (sk
->sk_family
== PF_INET
? 2 : 1);
391 if (inet
->inet_daddr
) {
392 if (inet
->inet_daddr
!= saddr
)
396 if (inet
->inet_dport
) {
397 if (inet
->inet_dport
!= sport
)
401 if (sk
->sk_bound_dev_if
) {
402 if (sk
->sk_bound_dev_if
!= dif
)
411 /* called with read_rcu_lock() */
412 static struct sock
*udp4_lib_lookup2(struct net
*net
,
413 __be32 saddr
, __be16 sport
,
414 __be32 daddr
, unsigned int hnum
, int dif
,
415 struct udp_hslot
*hslot2
, unsigned int slot2
)
417 struct sock
*sk
, *result
;
418 struct hlist_nulls_node
*node
;
419 int score
, badness
, matches
= 0, reuseport
= 0;
425 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
426 score
= compute_score2(sk
, net
, saddr
, sport
,
428 if (score
> badness
) {
431 reuseport
= sk
->sk_reuseport
;
433 hash
= inet_ehashfn(net
, daddr
, hnum
,
437 } else if (score
== badness
&& reuseport
) {
439 if (((u64
)hash
* matches
) >> 32 == 0)
441 hash
= next_pseudo_random32(hash
);
445 * if the nulls value we got at the end of this lookup is
446 * not the expected one, we must restart lookup.
447 * We probably met an item that was moved to another chain.
449 if (get_nulls_value(node
) != slot2
)
452 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
454 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
455 daddr
, hnum
, dif
) < badness
)) {
463 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
464 * harder than this. -DaveM
466 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
467 __be16 sport
, __be32 daddr
, __be16 dport
,
468 int dif
, struct udp_table
*udptable
)
470 struct sock
*sk
, *result
;
471 struct hlist_nulls_node
*node
;
472 unsigned short hnum
= ntohs(dport
);
473 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
474 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
475 int score
, badness
, matches
= 0, reuseport
= 0;
479 if (hslot
->count
> 10) {
480 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
);
481 slot2
= hash2
& udptable
->mask
;
482 hslot2
= &udptable
->hash2
[slot2
];
483 if (hslot
->count
< hslot2
->count
)
486 result
= udp4_lib_lookup2(net
, saddr
, sport
,
490 hash2
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
491 slot2
= hash2
& udptable
->mask
;
492 hslot2
= &udptable
->hash2
[slot2
];
493 if (hslot
->count
< hslot2
->count
)
496 result
= udp4_lib_lookup2(net
, saddr
, sport
,
497 htonl(INADDR_ANY
), hnum
, dif
,
506 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
507 score
= compute_score(sk
, net
, saddr
, hnum
, sport
,
509 if (score
> badness
) {
512 reuseport
= sk
->sk_reuseport
;
514 hash
= inet_ehashfn(net
, daddr
, hnum
,
518 } else if (score
== badness
&& reuseport
) {
520 if (((u64
)hash
* matches
) >> 32 == 0)
522 hash
= next_pseudo_random32(hash
);
526 * if the nulls value we got at the end of this lookup is
527 * not the expected one, we must restart lookup.
528 * We probably met an item that was moved to another chain.
530 if (get_nulls_value(node
) != slot
)
534 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
536 else if (unlikely(compute_score(result
, net
, saddr
, hnum
, sport
,
537 daddr
, dport
, dif
) < badness
)) {
545 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
547 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
548 __be16 sport
, __be16 dport
,
549 struct udp_table
*udptable
)
552 const struct iphdr
*iph
= ip_hdr(skb
);
554 if (unlikely(sk
= skb_steal_sock(skb
)))
557 return __udp4_lib_lookup(dev_net(skb_dst(skb
)->dev
), iph
->saddr
, sport
,
558 iph
->daddr
, dport
, inet_iif(skb
),
562 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
563 __be32 daddr
, __be16 dport
, int dif
)
565 return __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
567 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
569 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
570 __be16 loc_port
, __be32 loc_addr
,
571 __be16 rmt_port
, __be32 rmt_addr
,
572 int dif
, unsigned short hnum
)
574 struct inet_sock
*inet
= inet_sk(sk
);
576 if (!net_eq(sock_net(sk
), net
) ||
577 udp_sk(sk
)->udp_port_hash
!= hnum
||
578 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
579 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
580 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
581 ipv6_only_sock(sk
) ||
582 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
))
584 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
))
589 static inline struct sock
*udp_v4_mcast_next(struct net
*net
, struct sock
*sk
,
590 __be16 loc_port
, __be32 loc_addr
,
591 __be16 rmt_port
, __be32 rmt_addr
,
594 struct hlist_nulls_node
*node
;
596 unsigned short hnum
= ntohs(loc_port
);
598 sk_nulls_for_each_from(s
, node
) {
599 if (__udp_is_mcast_sock(net
, s
,
611 * This routine is called by the ICMP module when it gets some
612 * sort of error condition. If err < 0 then the socket should
613 * be closed and the error returned to the user. If err > 0
614 * it's just the icmp type << 8 | icmp code.
615 * Header points to the ip header of the error packet. We move
616 * on past this. Then (as it used to claim before adjustment)
617 * header points to the first 8 bytes of the udp header. We need
618 * to find the appropriate port.
621 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
623 struct inet_sock
*inet
;
624 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
625 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
626 const int type
= icmp_hdr(skb
)->type
;
627 const int code
= icmp_hdr(skb
)->code
;
631 struct net
*net
= dev_net(skb
->dev
);
633 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
634 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
);
636 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
637 return; /* No socket for error */
646 case ICMP_TIME_EXCEEDED
:
649 case ICMP_SOURCE_QUENCH
:
651 case ICMP_PARAMETERPROB
:
655 case ICMP_DEST_UNREACH
:
656 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
657 ipv4_sk_update_pmtu(skb
, sk
, info
);
658 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
666 if (code
<= NR_ICMP_UNREACH
) {
667 harderr
= icmp_err_convert
[code
].fatal
;
668 err
= icmp_err_convert
[code
].errno
;
672 ipv4_sk_redirect(skb
, sk
);
677 * RFC1122: OK. Passes ICMP errors back to application, as per
680 if (!inet
->recverr
) {
681 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
684 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
687 sk
->sk_error_report(sk
);
692 void udp_err(struct sk_buff
*skb
, u32 info
)
694 __udp4_lib_err(skb
, info
, &udp_table
);
698 * Throw away all pending data and cancel the corking. Socket is locked.
700 void udp_flush_pending_frames(struct sock
*sk
)
702 struct udp_sock
*up
= udp_sk(sk
);
707 ip_flush_pending_frames(sk
);
710 EXPORT_SYMBOL(udp_flush_pending_frames
);
713 * udp4_hwcsum - handle outgoing HW checksumming
714 * @skb: sk_buff containing the filled-in UDP header
715 * (checksum field must be zeroed out)
716 * @src: source IP address
717 * @dst: destination IP address
719 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
721 struct udphdr
*uh
= udp_hdr(skb
);
722 struct sk_buff
*frags
= skb_shinfo(skb
)->frag_list
;
723 int offset
= skb_transport_offset(skb
);
724 int len
= skb
->len
- offset
;
730 * Only one fragment on the socket.
732 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
733 skb
->csum_offset
= offsetof(struct udphdr
, check
);
734 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
738 * HW-checksum won't work as there are two or more
739 * fragments on the socket so that all csums of sk_buffs
743 csum
= csum_add(csum
, frags
->csum
);
745 } while ((frags
= frags
->next
));
747 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
748 skb
->ip_summed
= CHECKSUM_NONE
;
750 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
752 uh
->check
= CSUM_MANGLED_0
;
755 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
757 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
)
759 struct sock
*sk
= skb
->sk
;
760 struct inet_sock
*inet
= inet_sk(sk
);
763 int is_udplite
= IS_UDPLITE(sk
);
764 int offset
= skb_transport_offset(skb
);
765 int len
= skb
->len
- offset
;
769 * Create a UDP header
772 uh
->source
= inet
->inet_sport
;
773 uh
->dest
= fl4
->fl4_dport
;
774 uh
->len
= htons(len
);
777 if (is_udplite
) /* UDP-Lite */
778 csum
= udplite_csum(skb
);
780 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
782 skb
->ip_summed
= CHECKSUM_NONE
;
785 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
787 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
791 csum
= udp_csum(skb
);
793 /* add protocol-dependent pseudo-header */
794 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
795 sk
->sk_protocol
, csum
);
797 uh
->check
= CSUM_MANGLED_0
;
800 err
= ip_send_skb(sock_net(sk
), skb
);
802 if (err
== -ENOBUFS
&& !inet
->recverr
) {
803 UDP_INC_STATS_USER(sock_net(sk
),
804 UDP_MIB_SNDBUFERRORS
, is_udplite
);
808 UDP_INC_STATS_USER(sock_net(sk
),
809 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
814 * Push out all pending data as one UDP datagram. Socket is locked.
816 int udp_push_pending_frames(struct sock
*sk
)
818 struct udp_sock
*up
= udp_sk(sk
);
819 struct inet_sock
*inet
= inet_sk(sk
);
820 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
824 skb
= ip_finish_skb(sk
, fl4
);
828 err
= udp_send_skb(skb
, fl4
);
835 EXPORT_SYMBOL(udp_push_pending_frames
);
837 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
840 struct inet_sock
*inet
= inet_sk(sk
);
841 struct udp_sock
*up
= udp_sk(sk
);
842 struct flowi4 fl4_stack
;
845 struct ipcm_cookie ipc
;
846 struct rtable
*rt
= NULL
;
849 __be32 daddr
, faddr
, saddr
;
852 int err
, is_udplite
= IS_UDPLITE(sk
);
853 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
854 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
856 struct ip_options_data opt_copy
;
865 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
873 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
875 fl4
= &inet
->cork
.fl
.u
.ip4
;
878 * There are pending frames.
879 * The socket lock must be held while it's corked.
882 if (likely(up
->pending
)) {
883 if (unlikely(up
->pending
!= AF_INET
)) {
891 ulen
+= sizeof(struct udphdr
);
894 * Get and verify the address.
897 struct sockaddr_in
*usin
= (struct sockaddr_in
*)msg
->msg_name
;
898 if (msg
->msg_namelen
< sizeof(*usin
))
900 if (usin
->sin_family
!= AF_INET
) {
901 if (usin
->sin_family
!= AF_UNSPEC
)
902 return -EAFNOSUPPORT
;
905 daddr
= usin
->sin_addr
.s_addr
;
906 dport
= usin
->sin_port
;
910 if (sk
->sk_state
!= TCP_ESTABLISHED
)
911 return -EDESTADDRREQ
;
912 daddr
= inet
->inet_daddr
;
913 dport
= inet
->inet_dport
;
914 /* Open fast path for connected socket.
915 Route will not be used, if at least one option is set.
919 ipc
.addr
= inet
->inet_saddr
;
921 ipc
.oif
= sk
->sk_bound_dev_if
;
923 sock_tx_timestamp(sk
, &ipc
.tx_flags
);
925 if (msg
->msg_controllen
) {
926 err
= ip_cmsg_send(sock_net(sk
), msg
, &ipc
);
934 struct ip_options_rcu
*inet_opt
;
937 inet_opt
= rcu_dereference(inet
->inet_opt
);
939 memcpy(&opt_copy
, inet_opt
,
940 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
941 ipc
.opt
= &opt_copy
.opt
;
947 ipc
.addr
= faddr
= daddr
;
949 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
952 faddr
= ipc
.opt
->opt
.faddr
;
955 tos
= get_rttos(&ipc
, inet
);
956 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
957 (msg
->msg_flags
& MSG_DONTROUTE
) ||
958 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
963 if (ipv4_is_multicast(daddr
)) {
965 ipc
.oif
= inet
->mc_index
;
967 saddr
= inet
->mc_addr
;
970 ipc
.oif
= inet
->uc_index
;
973 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
976 struct net
*net
= sock_net(sk
);
979 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
980 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
981 inet_sk_flowi_flags(sk
)|FLOWI_FLAG_CAN_SLEEP
,
982 faddr
, saddr
, dport
, inet
->inet_sport
);
984 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
985 rt
= ip_route_output_flow(net
, fl4
, sk
);
989 if (err
== -ENETUNREACH
)
990 IP_INC_STATS_BH(net
, IPSTATS_MIB_OUTNOROUTES
);
995 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
996 !sock_flag(sk
, SOCK_BROADCAST
))
999 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1002 if (msg
->msg_flags
&MSG_CONFIRM
)
1008 daddr
= ipc
.addr
= fl4
->daddr
;
1010 /* Lockless fast path for the non-corking case. */
1012 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
->msg_iov
, ulen
,
1013 sizeof(struct udphdr
), &ipc
, &rt
,
1016 if (!IS_ERR_OR_NULL(skb
))
1017 err
= udp_send_skb(skb
, fl4
);
1022 if (unlikely(up
->pending
)) {
1023 /* The socket is already corked while preparing it. */
1024 /* ... which is an evident application bug. --ANK */
1027 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("cork app bug 2\n"));
1032 * Now cork the socket to pend data.
1034 fl4
= &inet
->cork
.fl
.u
.ip4
;
1037 fl4
->fl4_dport
= dport
;
1038 fl4
->fl4_sport
= inet
->inet_sport
;
1039 up
->pending
= AF_INET
;
1043 err
= ip_append_data(sk
, fl4
, getfrag
, msg
->msg_iov
, ulen
,
1044 sizeof(struct udphdr
), &ipc
, &rt
,
1045 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1047 udp_flush_pending_frames(sk
);
1049 err
= udp_push_pending_frames(sk
);
1050 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1061 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1062 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1063 * we don't have a good statistic (IpOutDiscards but it can be too many
1064 * things). We could add another new stat but at least for now that
1065 * seems like overkill.
1067 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1068 UDP_INC_STATS_USER(sock_net(sk
),
1069 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1074 dst_confirm(&rt
->dst
);
1075 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1076 goto back_from_confirm
;
1080 EXPORT_SYMBOL(udp_sendmsg
);
1082 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1083 size_t size
, int flags
)
1085 struct inet_sock
*inet
= inet_sk(sk
);
1086 struct udp_sock
*up
= udp_sk(sk
);
1090 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1092 /* Call udp_sendmsg to specify destination address which
1093 * sendpage interface can't pass.
1094 * This will succeed only when the socket is connected.
1096 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
1103 if (unlikely(!up
->pending
)) {
1106 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("udp cork app bug 3\n"));
1110 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1111 page
, offset
, size
, flags
);
1112 if (ret
== -EOPNOTSUPP
) {
1114 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1118 udp_flush_pending_frames(sk
);
1123 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1124 ret
= udp_push_pending_frames(sk
);
1134 * first_packet_length - return length of first packet in receive queue
1137 * Drops all bad checksum frames, until a valid one is found.
1138 * Returns the length of found skb, or 0 if none is found.
1140 static unsigned int first_packet_length(struct sock
*sk
)
1142 struct sk_buff_head list_kill
, *rcvq
= &sk
->sk_receive_queue
;
1143 struct sk_buff
*skb
;
1146 __skb_queue_head_init(&list_kill
);
1148 spin_lock_bh(&rcvq
->lock
);
1149 while ((skb
= skb_peek(rcvq
)) != NULL
&&
1150 udp_lib_checksum_complete(skb
)) {
1151 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1153 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1155 atomic_inc(&sk
->sk_drops
);
1156 __skb_unlink(skb
, rcvq
);
1157 __skb_queue_tail(&list_kill
, skb
);
1159 res
= skb
? skb
->len
: 0;
1160 spin_unlock_bh(&rcvq
->lock
);
1162 if (!skb_queue_empty(&list_kill
)) {
1163 bool slow
= lock_sock_fast(sk
);
1165 __skb_queue_purge(&list_kill
);
1166 sk_mem_reclaim_partial(sk
);
1167 unlock_sock_fast(sk
, slow
);
1173 * IOCTL requests applicable to the UDP protocol
1176 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1181 int amount
= sk_wmem_alloc_get(sk
);
1183 return put_user(amount
, (int __user
*)arg
);
1188 unsigned int amount
= first_packet_length(sk
);
1192 * We will only return the amount
1193 * of this packet since that is all
1194 * that will be read.
1196 amount
-= sizeof(struct udphdr
);
1198 return put_user(amount
, (int __user
*)arg
);
1202 return -ENOIOCTLCMD
;
1207 EXPORT_SYMBOL(udp_ioctl
);
1210 * This should be easy, if there is something there we
1211 * return it, otherwise we block.
1214 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
1215 size_t len
, int noblock
, int flags
, int *addr_len
)
1217 struct inet_sock
*inet
= inet_sk(sk
);
1218 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
1219 struct sk_buff
*skb
;
1220 unsigned int ulen
, copied
;
1221 int peeked
, off
= 0;
1223 int is_udplite
= IS_UDPLITE(sk
);
1227 * Check any passed addresses
1230 *addr_len
= sizeof(*sin
);
1232 if (flags
& MSG_ERRQUEUE
)
1233 return ip_recv_error(sk
, msg
, len
);
1236 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
1237 &peeked
, &off
, &err
);
1241 ulen
= skb
->len
- sizeof(struct udphdr
);
1245 else if (copied
< ulen
)
1246 msg
->msg_flags
|= MSG_TRUNC
;
1249 * If checksum is needed at all, try to do it while copying the
1250 * data. If the data is truncated, or if we only want a partial
1251 * coverage checksum (UDP-Lite), do it before the copy.
1254 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
1255 if (udp_lib_checksum_complete(skb
))
1259 if (skb_csum_unnecessary(skb
))
1260 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
1261 msg
->msg_iov
, copied
);
1263 err
= skb_copy_and_csum_datagram_iovec(skb
,
1264 sizeof(struct udphdr
),
1271 if (unlikely(err
)) {
1272 trace_kfree_skb(skb
, udp_recvmsg
);
1274 atomic_inc(&sk
->sk_drops
);
1275 UDP_INC_STATS_USER(sock_net(sk
),
1276 UDP_MIB_INERRORS
, is_udplite
);
1282 UDP_INC_STATS_USER(sock_net(sk
),
1283 UDP_MIB_INDATAGRAMS
, is_udplite
);
1285 sock_recv_ts_and_drops(msg
, sk
, skb
);
1287 /* Copy the address. */
1289 sin
->sin_family
= AF_INET
;
1290 sin
->sin_port
= udp_hdr(skb
)->source
;
1291 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1292 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1294 if (inet
->cmsg_flags
)
1295 ip_cmsg_recv(msg
, skb
);
1298 if (flags
& MSG_TRUNC
)
1302 skb_free_datagram_locked(sk
, skb
);
1307 slow
= lock_sock_fast(sk
);
1308 if (!skb_kill_datagram(sk
, skb
, flags
)) {
1309 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1310 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1312 unlock_sock_fast(sk
, slow
);
1317 /* starting over for a new packet */
1318 msg
->msg_flags
&= ~MSG_TRUNC
;
1323 int udp_disconnect(struct sock
*sk
, int flags
)
1325 struct inet_sock
*inet
= inet_sk(sk
);
1327 * 1003.1g - break association.
1330 sk
->sk_state
= TCP_CLOSE
;
1331 inet
->inet_daddr
= 0;
1332 inet
->inet_dport
= 0;
1333 sock_rps_reset_rxhash(sk
);
1334 sk
->sk_bound_dev_if
= 0;
1335 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1336 inet_reset_saddr(sk
);
1338 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1339 sk
->sk_prot
->unhash(sk
);
1340 inet
->inet_sport
= 0;
1345 EXPORT_SYMBOL(udp_disconnect
);
1347 void udp_lib_unhash(struct sock
*sk
)
1349 if (sk_hashed(sk
)) {
1350 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1351 struct udp_hslot
*hslot
, *hslot2
;
1353 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1354 udp_sk(sk
)->udp_port_hash
);
1355 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1357 spin_lock_bh(&hslot
->lock
);
1358 if (sk_nulls_del_node_init_rcu(sk
)) {
1360 inet_sk(sk
)->inet_num
= 0;
1361 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1363 spin_lock(&hslot2
->lock
);
1364 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1366 spin_unlock(&hslot2
->lock
);
1368 spin_unlock_bh(&hslot
->lock
);
1371 EXPORT_SYMBOL(udp_lib_unhash
);
1374 * inet_rcv_saddr was changed, we must rehash secondary hash
1376 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1378 if (sk_hashed(sk
)) {
1379 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1380 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1382 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1383 nhslot2
= udp_hashslot2(udptable
, newhash
);
1384 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1385 if (hslot2
!= nhslot2
) {
1386 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1387 udp_sk(sk
)->udp_port_hash
);
1388 /* we must lock primary chain too */
1389 spin_lock_bh(&hslot
->lock
);
1391 spin_lock(&hslot2
->lock
);
1392 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1394 spin_unlock(&hslot2
->lock
);
1396 spin_lock(&nhslot2
->lock
);
1397 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1400 spin_unlock(&nhslot2
->lock
);
1402 spin_unlock_bh(&hslot
->lock
);
1406 EXPORT_SYMBOL(udp_lib_rehash
);
1408 static void udp_v4_rehash(struct sock
*sk
)
1410 u16 new_hash
= udp4_portaddr_hash(sock_net(sk
),
1411 inet_sk(sk
)->inet_rcv_saddr
,
1412 inet_sk(sk
)->inet_num
);
1413 udp_lib_rehash(sk
, new_hash
);
1416 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1420 if (inet_sk(sk
)->inet_daddr
) {
1421 sock_rps_save_rxhash(sk
, skb
);
1422 sk_mark_napi_id(sk
, skb
);
1425 rc
= sock_queue_rcv_skb(sk
, skb
);
1427 int is_udplite
= IS_UDPLITE(sk
);
1429 /* Note that an ENOMEM error is charged twice */
1431 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1433 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1435 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1443 static struct static_key udp_encap_needed __read_mostly
;
1444 void udp_encap_enable(void)
1446 if (!static_key_enabled(&udp_encap_needed
))
1447 static_key_slow_inc(&udp_encap_needed
);
1449 EXPORT_SYMBOL(udp_encap_enable
);
1454 * >0: "udp encap" protocol resubmission
1456 * Note that in the success and error cases, the skb is assumed to
1457 * have either been requeued or freed.
1459 int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1461 struct udp_sock
*up
= udp_sk(sk
);
1463 int is_udplite
= IS_UDPLITE(sk
);
1466 * Charge it to the socket, dropping if the queue is full.
1468 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1472 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1473 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1476 * This is an encapsulation socket so pass the skb to
1477 * the socket's udp_encap_rcv() hook. Otherwise, just
1478 * fall through and pass this up the UDP socket.
1479 * up->encap_rcv() returns the following value:
1480 * =0 if skb was successfully passed to the encap
1481 * handler or was discarded by it.
1482 * >0 if skb should be passed on to UDP.
1483 * <0 if skb should be resubmitted as proto -N
1486 /* if we're overly short, let UDP handle it */
1487 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
1488 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
!= NULL
) {
1491 ret
= encap_rcv(sk
, skb
);
1493 UDP_INC_STATS_BH(sock_net(sk
),
1494 UDP_MIB_INDATAGRAMS
,
1500 /* FALLTHROUGH -- it's a UDP Packet */
1504 * UDP-Lite specific tests, ignored on UDP sockets
1506 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1509 * MIB statistics other than incrementing the error count are
1510 * disabled for the following two types of errors: these depend
1511 * on the application settings, not on the functioning of the
1512 * protocol stack as such.
1514 * RFC 3828 here recommends (sec 3.3): "There should also be a
1515 * way ... to ... at least let the receiving application block
1516 * delivery of packets with coverage values less than a value
1517 * provided by the application."
1519 if (up
->pcrlen
== 0) { /* full coverage was set */
1520 LIMIT_NETDEBUG(KERN_WARNING
"UDPLite: partial coverage %d while full coverage %d requested\n",
1521 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1524 /* The next case involves violating the min. coverage requested
1525 * by the receiver. This is subtle: if receiver wants x and x is
1526 * greater than the buffersize/MTU then receiver will complain
1527 * that it wants x while sender emits packets of smaller size y.
1528 * Therefore the above ...()->partial_cov statement is essential.
1530 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1531 LIMIT_NETDEBUG(KERN_WARNING
"UDPLite: coverage %d too small, need min %d\n",
1532 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1537 if (rcu_access_pointer(sk
->sk_filter
) &&
1538 udp_lib_checksum_complete(skb
))
1542 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
))
1547 ipv4_pktinfo_prepare(sk
, skb
);
1549 if (!sock_owned_by_user(sk
))
1550 rc
= __udp_queue_rcv_skb(sk
, skb
);
1551 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
1560 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1562 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1563 atomic_inc(&sk
->sk_drops
);
1569 static void flush_stack(struct sock
**stack
, unsigned int count
,
1570 struct sk_buff
*skb
, unsigned int final
)
1573 struct sk_buff
*skb1
= NULL
;
1576 for (i
= 0; i
< count
; i
++) {
1578 if (likely(skb1
== NULL
))
1579 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
1582 atomic_inc(&sk
->sk_drops
);
1583 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1585 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1589 if (skb1
&& udp_queue_rcv_skb(sk
, skb1
) <= 0)
1596 static void udp_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
1598 struct dst_entry
*dst
= skb_dst(skb
);
1601 sk
->sk_rx_dst
= dst
;
1605 * Multicasts and broadcasts go to each listener.
1607 * Note: called only from the BH handler context.
1609 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1611 __be32 saddr
, __be32 daddr
,
1612 struct udp_table
*udptable
)
1614 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
1615 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
1617 unsigned int i
, count
= 0;
1619 spin_lock(&hslot
->lock
);
1620 sk
= sk_nulls_head(&hslot
->head
);
1621 dif
= skb
->dev
->ifindex
;
1622 sk
= udp_v4_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1624 stack
[count
++] = sk
;
1625 sk
= udp_v4_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
,
1626 daddr
, uh
->source
, saddr
, dif
);
1627 if (unlikely(count
== ARRAY_SIZE(stack
))) {
1630 flush_stack(stack
, count
, skb
, ~0);
1635 * before releasing chain lock, we must take a reference on sockets
1637 for (i
= 0; i
< count
; i
++)
1638 sock_hold(stack
[i
]);
1640 spin_unlock(&hslot
->lock
);
1643 * do the slow work with no lock held
1646 flush_stack(stack
, count
, skb
, count
- 1);
1648 for (i
= 0; i
< count
; i
++)
1656 /* Initialize UDP checksum. If exited with zero value (success),
1657 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1658 * Otherwise, csum completion requires chacksumming packet body,
1659 * including udp header and folding it to skb->csum.
1661 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1664 const struct iphdr
*iph
;
1667 UDP_SKB_CB(skb
)->partial_cov
= 0;
1668 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1670 if (proto
== IPPROTO_UDPLITE
) {
1671 err
= udplite_checksum_init(skb
, uh
);
1677 if (uh
->check
== 0) {
1678 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1679 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1680 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1682 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1684 if (!skb_csum_unnecessary(skb
))
1685 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1686 skb
->len
, proto
, 0);
1687 /* Probably, we should checksum udp header (it should be in cache
1688 * in any case) and data in tiny packets (< rx copybreak).
1695 * All we need to do is get the socket, and then do a checksum.
1698 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
1703 unsigned short ulen
;
1704 struct rtable
*rt
= skb_rtable(skb
);
1705 __be32 saddr
, daddr
;
1706 struct net
*net
= dev_net(skb
->dev
);
1709 * Validate the packet.
1711 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1712 goto drop
; /* No space for header. */
1715 ulen
= ntohs(uh
->len
);
1716 saddr
= ip_hdr(skb
)->saddr
;
1717 daddr
= ip_hdr(skb
)->daddr
;
1719 if (ulen
> skb
->len
)
1722 if (proto
== IPPROTO_UDP
) {
1723 /* UDP validates ulen. */
1724 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1729 if (udp4_csum_init(skb
, uh
, proto
))
1736 if (unlikely(sk
->sk_rx_dst
== NULL
))
1737 udp_sk_rx_dst_set(sk
, skb
);
1739 ret
= udp_queue_rcv_skb(sk
, skb
);
1741 /* a return value > 0 means to resubmit the input, but
1742 * it wants the return to be -protocol, or 0
1748 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1749 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
1750 saddr
, daddr
, udptable
);
1752 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
1758 ret
= udp_queue_rcv_skb(sk
, skb
);
1761 /* a return value > 0 means to resubmit the input, but
1762 * it wants the return to be -protocol, or 0
1769 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1773 /* No socket. Drop packet silently, if checksum is wrong */
1774 if (udp_lib_checksum_complete(skb
))
1777 UDP_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1778 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1781 * Hmm. We got an UDP packet to a port to which we
1782 * don't wanna listen. Ignore it.
1788 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1789 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1790 &saddr
, ntohs(uh
->source
),
1792 &daddr
, ntohs(uh
->dest
));
1797 * RFC1122: OK. Discards the bad packet silently (as far as
1798 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1800 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1801 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1802 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
1804 UDP_INC_STATS_BH(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
1806 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1811 /* We can only early demux multicast if there is a single matching socket.
1812 * If more than one socket found returns NULL
1814 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
1815 __be16 loc_port
, __be32 loc_addr
,
1816 __be16 rmt_port
, __be32 rmt_addr
,
1819 struct sock
*sk
, *result
;
1820 struct hlist_nulls_node
*node
;
1821 unsigned short hnum
= ntohs(loc_port
);
1822 unsigned int count
, slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
1823 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
1829 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
1830 if (__udp_is_mcast_sock(net
, sk
,
1839 * if the nulls value we got at the end of this lookup is
1840 * not the expected one, we must restart lookup.
1841 * We probably met an item that was moved to another chain.
1843 if (get_nulls_value(node
) != slot
)
1848 unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
1850 else if (unlikely(!__udp_is_mcast_sock(net
, result
,
1862 /* For unicast we should only early demux connected sockets or we can
1863 * break forwarding setups. The chains here can be long so only check
1864 * if the first socket is an exact match and if not move on.
1866 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
1867 __be16 loc_port
, __be32 loc_addr
,
1868 __be16 rmt_port
, __be32 rmt_addr
,
1871 struct sock
*sk
, *result
;
1872 struct hlist_nulls_node
*node
;
1873 unsigned short hnum
= ntohs(loc_port
);
1874 unsigned int hash2
= udp4_portaddr_hash(net
, loc_addr
, hnum
);
1875 unsigned int slot2
= hash2
& udp_table
.mask
;
1876 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
1877 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
)
1878 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
1882 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
1883 if (INET_MATCH(sk
, net
, acookie
,
1884 rmt_addr
, loc_addr
, ports
, dif
))
1886 /* Only check first socket in chain */
1891 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
1893 else if (unlikely(!INET_MATCH(sk
, net
, acookie
,
1904 void udp_v4_early_demux(struct sk_buff
*skb
)
1906 const struct iphdr
*iph
= ip_hdr(skb
);
1907 const struct udphdr
*uh
= udp_hdr(skb
);
1909 struct dst_entry
*dst
;
1910 struct net
*net
= dev_net(skb
->dev
);
1911 int dif
= skb
->dev
->ifindex
;
1913 /* validate the packet */
1914 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
1917 if (skb
->pkt_type
== PACKET_BROADCAST
||
1918 skb
->pkt_type
== PACKET_MULTICAST
)
1919 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
1920 uh
->source
, iph
->saddr
, dif
);
1921 else if (skb
->pkt_type
== PACKET_HOST
)
1922 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
1923 uh
->source
, iph
->saddr
, dif
);
1931 skb
->destructor
= sock_edemux
;
1932 dst
= sk
->sk_rx_dst
;
1935 dst
= dst_check(dst
, 0);
1937 skb_dst_set_noref(skb
, dst
);
1940 int udp_rcv(struct sk_buff
*skb
)
1942 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
1945 void udp_destroy_sock(struct sock
*sk
)
1947 struct udp_sock
*up
= udp_sk(sk
);
1948 bool slow
= lock_sock_fast(sk
);
1949 udp_flush_pending_frames(sk
);
1950 unlock_sock_fast(sk
, slow
);
1951 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1952 void (*encap_destroy
)(struct sock
*sk
);
1953 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1960 * Socket option code for UDP
1962 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1963 char __user
*optval
, unsigned int optlen
,
1964 int (*push_pending_frames
)(struct sock
*))
1966 struct udp_sock
*up
= udp_sk(sk
);
1969 int is_udplite
= IS_UDPLITE(sk
);
1971 if (optlen
< sizeof(int))
1974 if (get_user(val
, (int __user
*)optval
))
1984 (*push_pending_frames
)(sk
);
1992 case UDP_ENCAP_ESPINUDP
:
1993 case UDP_ENCAP_ESPINUDP_NON_IKE
:
1994 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
1996 case UDP_ENCAP_L2TPINUDP
:
1997 up
->encap_type
= val
;
2007 * UDP-Lite's partial checksum coverage (RFC 3828).
2009 /* The sender sets actual checksum coverage length via this option.
2010 * The case coverage > packet length is handled by send module. */
2011 case UDPLITE_SEND_CSCOV
:
2012 if (!is_udplite
) /* Disable the option on UDP sockets */
2013 return -ENOPROTOOPT
;
2014 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2016 else if (val
> USHRT_MAX
)
2019 up
->pcflag
|= UDPLITE_SEND_CC
;
2022 /* The receiver specifies a minimum checksum coverage value. To make
2023 * sense, this should be set to at least 8 (as done below). If zero is
2024 * used, this again means full checksum coverage. */
2025 case UDPLITE_RECV_CSCOV
:
2026 if (!is_udplite
) /* Disable the option on UDP sockets */
2027 return -ENOPROTOOPT
;
2028 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2030 else if (val
> USHRT_MAX
)
2033 up
->pcflag
|= UDPLITE_RECV_CC
;
2043 EXPORT_SYMBOL(udp_lib_setsockopt
);
2045 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2046 char __user
*optval
, unsigned int optlen
)
2048 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2049 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2050 udp_push_pending_frames
);
2051 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2054 #ifdef CONFIG_COMPAT
2055 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2056 char __user
*optval
, unsigned int optlen
)
2058 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2059 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2060 udp_push_pending_frames
);
2061 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2065 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2066 char __user
*optval
, int __user
*optlen
)
2068 struct udp_sock
*up
= udp_sk(sk
);
2071 if (get_user(len
, optlen
))
2074 len
= min_t(unsigned int, len
, sizeof(int));
2085 val
= up
->encap_type
;
2088 /* The following two cannot be changed on UDP sockets, the return is
2089 * always 0 (which corresponds to the full checksum coverage of UDP). */
2090 case UDPLITE_SEND_CSCOV
:
2094 case UDPLITE_RECV_CSCOV
:
2099 return -ENOPROTOOPT
;
2102 if (put_user(len
, optlen
))
2104 if (copy_to_user(optval
, &val
, len
))
2108 EXPORT_SYMBOL(udp_lib_getsockopt
);
2110 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2111 char __user
*optval
, int __user
*optlen
)
2113 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2114 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2115 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2118 #ifdef CONFIG_COMPAT
2119 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2120 char __user
*optval
, int __user
*optlen
)
2122 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2123 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2124 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2128 * udp_poll - wait for a UDP event.
2129 * @file - file struct
2131 * @wait - poll table
2133 * This is same as datagram poll, except for the special case of
2134 * blocking sockets. If application is using a blocking fd
2135 * and a packet with checksum error is in the queue;
2136 * then it could get return from select indicating data available
2137 * but then block when reading it. Add special case code
2138 * to work around these arguably broken applications.
2140 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2142 unsigned int mask
= datagram_poll(file
, sock
, wait
);
2143 struct sock
*sk
= sock
->sk
;
2145 sock_rps_record_flow(sk
);
2147 /* Check for false positives due to checksum errors */
2148 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2149 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && !first_packet_length(sk
))
2150 mask
&= ~(POLLIN
| POLLRDNORM
);
2155 EXPORT_SYMBOL(udp_poll
);
2157 struct proto udp_prot
= {
2159 .owner
= THIS_MODULE
,
2160 .close
= udp_lib_close
,
2161 .connect
= ip4_datagram_connect
,
2162 .disconnect
= udp_disconnect
,
2164 .destroy
= udp_destroy_sock
,
2165 .setsockopt
= udp_setsockopt
,
2166 .getsockopt
= udp_getsockopt
,
2167 .sendmsg
= udp_sendmsg
,
2168 .recvmsg
= udp_recvmsg
,
2169 .sendpage
= udp_sendpage
,
2170 .backlog_rcv
= __udp_queue_rcv_skb
,
2171 .release_cb
= ip4_datagram_release_cb
,
2172 .hash
= udp_lib_hash
,
2173 .unhash
= udp_lib_unhash
,
2174 .rehash
= udp_v4_rehash
,
2175 .get_port
= udp_v4_get_port
,
2176 .memory_allocated
= &udp_memory_allocated
,
2177 .sysctl_mem
= sysctl_udp_mem
,
2178 .sysctl_wmem
= &sysctl_udp_wmem_min
,
2179 .sysctl_rmem
= &sysctl_udp_rmem_min
,
2180 .obj_size
= sizeof(struct udp_sock
),
2181 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2182 .h
.udp_table
= &udp_table
,
2183 #ifdef CONFIG_COMPAT
2184 .compat_setsockopt
= compat_udp_setsockopt
,
2185 .compat_getsockopt
= compat_udp_getsockopt
,
2187 .clear_sk
= sk_prot_clear_portaddr_nulls
,
2189 EXPORT_SYMBOL(udp_prot
);
2191 /* ------------------------------------------------------------------------ */
2192 #ifdef CONFIG_PROC_FS
2194 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2197 struct udp_iter_state
*state
= seq
->private;
2198 struct net
*net
= seq_file_net(seq
);
2200 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
2202 struct hlist_nulls_node
*node
;
2203 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
2205 if (hlist_nulls_empty(&hslot
->head
))
2208 spin_lock_bh(&hslot
->lock
);
2209 sk_nulls_for_each(sk
, node
, &hslot
->head
) {
2210 if (!net_eq(sock_net(sk
), net
))
2212 if (sk
->sk_family
== state
->family
)
2215 spin_unlock_bh(&hslot
->lock
);
2222 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2224 struct udp_iter_state
*state
= seq
->private;
2225 struct net
*net
= seq_file_net(seq
);
2228 sk
= sk_nulls_next(sk
);
2229 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
2232 if (state
->bucket
<= state
->udp_table
->mask
)
2233 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2234 return udp_get_first(seq
, state
->bucket
+ 1);
2239 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2241 struct sock
*sk
= udp_get_first(seq
, 0);
2244 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2246 return pos
? NULL
: sk
;
2249 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2251 struct udp_iter_state
*state
= seq
->private;
2252 state
->bucket
= MAX_UDP_PORTS
;
2254 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2257 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2261 if (v
== SEQ_START_TOKEN
)
2262 sk
= udp_get_idx(seq
, 0);
2264 sk
= udp_get_next(seq
, v
);
2270 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
2272 struct udp_iter_state
*state
= seq
->private;
2274 if (state
->bucket
<= state
->udp_table
->mask
)
2275 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2278 int udp_seq_open(struct inode
*inode
, struct file
*file
)
2280 struct udp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2281 struct udp_iter_state
*s
;
2284 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2285 sizeof(struct udp_iter_state
));
2289 s
= ((struct seq_file
*)file
->private_data
)->private;
2290 s
->family
= afinfo
->family
;
2291 s
->udp_table
= afinfo
->udp_table
;
2294 EXPORT_SYMBOL(udp_seq_open
);
2296 /* ------------------------------------------------------------------------ */
2297 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2299 struct proc_dir_entry
*p
;
2302 afinfo
->seq_ops
.start
= udp_seq_start
;
2303 afinfo
->seq_ops
.next
= udp_seq_next
;
2304 afinfo
->seq_ops
.stop
= udp_seq_stop
;
2306 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2307 afinfo
->seq_fops
, afinfo
);
2312 EXPORT_SYMBOL(udp_proc_register
);
2314 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2316 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2318 EXPORT_SYMBOL(udp_proc_unregister
);
2320 /* ------------------------------------------------------------------------ */
2321 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2322 int bucket
, int *len
)
2324 struct inet_sock
*inet
= inet_sk(sp
);
2325 __be32 dest
= inet
->inet_daddr
;
2326 __be32 src
= inet
->inet_rcv_saddr
;
2327 __u16 destp
= ntohs(inet
->inet_dport
);
2328 __u16 srcp
= ntohs(inet
->inet_sport
);
2330 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2331 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d%n",
2332 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2333 sk_wmem_alloc_get(sp
),
2334 sk_rmem_alloc_get(sp
),
2336 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2338 atomic_read(&sp
->sk_refcnt
), sp
,
2339 atomic_read(&sp
->sk_drops
), len
);
2342 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2344 if (v
== SEQ_START_TOKEN
)
2345 seq_printf(seq
, "%-127s\n",
2346 " sl local_address rem_address st tx_queue "
2347 "rx_queue tr tm->when retrnsmt uid timeout "
2348 "inode ref pointer drops");
2350 struct udp_iter_state
*state
= seq
->private;
2353 udp4_format_sock(v
, seq
, state
->bucket
, &len
);
2354 seq_printf(seq
, "%*s\n", 127 - len
, "");
2359 static const struct file_operations udp_afinfo_seq_fops
= {
2360 .owner
= THIS_MODULE
,
2361 .open
= udp_seq_open
,
2363 .llseek
= seq_lseek
,
2364 .release
= seq_release_net
2367 /* ------------------------------------------------------------------------ */
2368 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2371 .udp_table
= &udp_table
,
2372 .seq_fops
= &udp_afinfo_seq_fops
,
2374 .show
= udp4_seq_show
,
2378 static int __net_init
udp4_proc_init_net(struct net
*net
)
2380 return udp_proc_register(net
, &udp4_seq_afinfo
);
2383 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2385 udp_proc_unregister(net
, &udp4_seq_afinfo
);
2388 static struct pernet_operations udp4_net_ops
= {
2389 .init
= udp4_proc_init_net
,
2390 .exit
= udp4_proc_exit_net
,
2393 int __init
udp4_proc_init(void)
2395 return register_pernet_subsys(&udp4_net_ops
);
2398 void udp4_proc_exit(void)
2400 unregister_pernet_subsys(&udp4_net_ops
);
2402 #endif /* CONFIG_PROC_FS */
2404 static __initdata
unsigned long uhash_entries
;
2405 static int __init
set_uhash_entries(char *str
)
2412 ret
= kstrtoul(str
, 0, &uhash_entries
);
2416 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2417 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2420 __setup("uhash_entries=", set_uhash_entries
);
2422 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2426 table
->hash
= alloc_large_system_hash(name
,
2427 2 * sizeof(struct udp_hslot
),
2429 21, /* one slot per 2 MB */
2433 UDP_HTABLE_SIZE_MIN
,
2436 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2437 for (i
= 0; i
<= table
->mask
; i
++) {
2438 INIT_HLIST_NULLS_HEAD(&table
->hash
[i
].head
, i
);
2439 table
->hash
[i
].count
= 0;
2440 spin_lock_init(&table
->hash
[i
].lock
);
2442 for (i
= 0; i
<= table
->mask
; i
++) {
2443 INIT_HLIST_NULLS_HEAD(&table
->hash2
[i
].head
, i
);
2444 table
->hash2
[i
].count
= 0;
2445 spin_lock_init(&table
->hash2
[i
].lock
);
2449 void __init
udp_init(void)
2451 unsigned long limit
;
2453 udp_table_init(&udp_table
, "UDP");
2454 limit
= nr_free_buffer_pages() / 8;
2455 limit
= max(limit
, 128UL);
2456 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2457 sysctl_udp_mem
[1] = limit
;
2458 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2460 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2461 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2464 struct sk_buff
*skb_udp_tunnel_segment(struct sk_buff
*skb
,
2465 netdev_features_t features
)
2467 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
2468 int mac_len
= skb
->mac_len
;
2469 int tnl_hlen
= skb_inner_mac_header(skb
) - skb_transport_header(skb
);
2470 __be16 protocol
= skb
->protocol
;
2471 netdev_features_t enc_features
;
2474 if (unlikely(!pskb_may_pull(skb
, tnl_hlen
)))
2477 skb
->encapsulation
= 0;
2478 __skb_pull(skb
, tnl_hlen
);
2479 skb_reset_mac_header(skb
);
2480 skb_set_network_header(skb
, skb_inner_network_offset(skb
));
2481 skb
->mac_len
= skb_inner_network_offset(skb
);
2482 skb
->protocol
= htons(ETH_P_TEB
);
2484 /* segment inner packet. */
2485 enc_features
= skb
->dev
->hw_enc_features
& netif_skb_features(skb
);
2486 segs
= skb_mac_gso_segment(skb
, enc_features
);
2487 if (!segs
|| IS_ERR(segs
))
2490 outer_hlen
= skb_tnl_header_len(skb
);
2494 int udp_offset
= outer_hlen
- tnl_hlen
;
2496 skb_reset_inner_headers(skb
);
2497 skb
->encapsulation
= 1;
2499 skb
->mac_len
= mac_len
;
2501 skb_push(skb
, outer_hlen
);
2502 skb_reset_mac_header(skb
);
2503 skb_set_network_header(skb
, mac_len
);
2504 skb_set_transport_header(skb
, udp_offset
);
2506 uh
->len
= htons(skb
->len
- udp_offset
);
2508 /* csum segment if tunnel sets skb with csum. */
2509 if (protocol
== htons(ETH_P_IP
) && unlikely(uh
->check
)) {
2510 struct iphdr
*iph
= ip_hdr(skb
);
2512 uh
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
2513 skb
->len
- udp_offset
,
2515 uh
->check
= csum_fold(skb_checksum(skb
, udp_offset
,
2516 skb
->len
- udp_offset
, 0));
2518 uh
->check
= CSUM_MANGLED_0
;
2520 } else if (protocol
== htons(ETH_P_IPV6
)) {
2521 struct ipv6hdr
*ipv6h
= ipv6_hdr(skb
);
2522 u32 len
= skb
->len
- udp_offset
;
2524 uh
->check
= ~csum_ipv6_magic(&ipv6h
->saddr
, &ipv6h
->daddr
,
2525 len
, IPPROTO_UDP
, 0);
2526 uh
->check
= csum_fold(skb_checksum(skb
, udp_offset
, len
, 0));
2528 uh
->check
= CSUM_MANGLED_0
;
2529 skb
->ip_summed
= CHECKSUM_NONE
;
2532 skb
->protocol
= protocol
;
2533 } while ((skb
= skb
->next
));