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>
93 #include <linux/inetdevice.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
118 struct udp_table udp_table __read_mostly
;
119 EXPORT_SYMBOL(udp_table
);
121 long sysctl_udp_mem
[3] __read_mostly
;
122 EXPORT_SYMBOL(sysctl_udp_mem
);
124 int sysctl_udp_rmem_min __read_mostly
;
125 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
127 int sysctl_udp_wmem_min __read_mostly
;
128 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
130 atomic_long_t udp_memory_allocated
;
131 EXPORT_SYMBOL(udp_memory_allocated
);
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
136 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
137 const struct udp_hslot
*hslot
,
138 unsigned long *bitmap
,
140 int (*saddr_comp
)(const struct sock
*sk1
,
141 const struct sock
*sk2
,
142 bool match_wildcard
),
146 kuid_t uid
= sock_i_uid(sk
);
148 sk_for_each(sk2
, &hslot
->head
) {
149 if (net_eq(sock_net(sk2
), net
) &&
151 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
152 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
153 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
154 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
155 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
156 rcu_access_pointer(sk
->sk_reuseport_cb
) ||
157 !uid_eq(uid
, sock_i_uid(sk2
))) &&
158 saddr_comp(sk
, sk2
, true)) {
161 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
, bitmap
);
168 * Note: we still hold spinlock of primary hash chain, so no other writer
169 * can insert/delete a socket with local_port == num
171 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
172 struct udp_hslot
*hslot2
,
174 int (*saddr_comp
)(const struct sock
*sk1
,
175 const struct sock
*sk2
,
176 bool match_wildcard
))
179 kuid_t uid
= sock_i_uid(sk
);
182 spin_lock(&hslot2
->lock
);
183 udp_portaddr_for_each_entry(sk2
, &hslot2
->head
) {
184 if (net_eq(sock_net(sk2
), net
) &&
186 (udp_sk(sk2
)->udp_port_hash
== num
) &&
187 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
188 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
189 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
190 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
191 rcu_access_pointer(sk
->sk_reuseport_cb
) ||
192 !uid_eq(uid
, sock_i_uid(sk2
))) &&
193 saddr_comp(sk
, sk2
, true)) {
198 spin_unlock(&hslot2
->lock
);
202 static int udp_reuseport_add_sock(struct sock
*sk
, struct udp_hslot
*hslot
,
203 int (*saddr_same
)(const struct sock
*sk1
,
204 const struct sock
*sk2
,
205 bool match_wildcard
))
207 struct net
*net
= sock_net(sk
);
208 kuid_t uid
= sock_i_uid(sk
);
211 sk_for_each(sk2
, &hslot
->head
) {
212 if (net_eq(sock_net(sk2
), net
) &&
214 sk2
->sk_family
== sk
->sk_family
&&
215 ipv6_only_sock(sk2
) == ipv6_only_sock(sk
) &&
216 (udp_sk(sk2
)->udp_port_hash
== udp_sk(sk
)->udp_port_hash
) &&
217 (sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
218 sk2
->sk_reuseport
&& uid_eq(uid
, sock_i_uid(sk2
)) &&
219 (*saddr_same
)(sk
, sk2
, false)) {
220 return reuseport_add_sock(sk
, sk2
);
224 /* Initial allocation may have already happened via setsockopt */
225 if (!rcu_access_pointer(sk
->sk_reuseport_cb
))
226 return reuseport_alloc(sk
);
231 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 * @sk: socket struct in question
234 * @snum: port number to look up
235 * @saddr_comp: AF-dependent comparison of bound local IP addresses
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
239 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
240 int (*saddr_comp
)(const struct sock
*sk1
,
241 const struct sock
*sk2
,
242 bool match_wildcard
),
243 unsigned int hash2_nulladdr
)
245 struct udp_hslot
*hslot
, *hslot2
;
246 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
248 struct net
*net
= sock_net(sk
);
251 int low
, high
, remaining
;
253 unsigned short first
, last
;
254 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
256 inet_get_local_port_range(net
, &low
, &high
);
257 remaining
= (high
- low
) + 1;
259 rand
= prandom_u32();
260 first
= reciprocal_scale(rand
, remaining
) + low
;
262 * force rand to be an odd multiple of UDP_HTABLE_SIZE
264 rand
= (rand
| 1) * (udptable
->mask
+ 1);
265 last
= first
+ udptable
->mask
+ 1;
267 hslot
= udp_hashslot(udptable
, net
, first
);
268 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
269 spin_lock_bh(&hslot
->lock
);
270 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
271 saddr_comp
, udptable
->log
);
275 * Iterate on all possible values of snum for this hash.
276 * Using steps of an odd multiple of UDP_HTABLE_SIZE
277 * give us randomization and full range coverage.
280 if (low
<= snum
&& snum
<= high
&&
281 !test_bit(snum
>> udptable
->log
, bitmap
) &&
282 !inet_is_local_reserved_port(net
, snum
))
285 } while (snum
!= first
);
286 spin_unlock_bh(&hslot
->lock
);
287 } while (++first
!= last
);
290 hslot
= udp_hashslot(udptable
, net
, snum
);
291 spin_lock_bh(&hslot
->lock
);
292 if (hslot
->count
> 10) {
294 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
296 slot2
&= udptable
->mask
;
297 hash2_nulladdr
&= udptable
->mask
;
299 hslot2
= udp_hashslot2(udptable
, slot2
);
300 if (hslot
->count
< hslot2
->count
)
301 goto scan_primary_hash
;
303 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
305 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
306 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
307 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
316 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
,
321 inet_sk(sk
)->inet_num
= snum
;
322 udp_sk(sk
)->udp_port_hash
= snum
;
323 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
324 if (sk_unhashed(sk
)) {
325 if (sk
->sk_reuseport
&&
326 udp_reuseport_add_sock(sk
, hslot
, saddr_comp
)) {
327 inet_sk(sk
)->inet_num
= 0;
328 udp_sk(sk
)->udp_port_hash
= 0;
329 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
333 sk_add_node_rcu(sk
, &hslot
->head
);
335 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
337 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
338 spin_lock(&hslot2
->lock
);
339 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
342 spin_unlock(&hslot2
->lock
);
344 sock_set_flag(sk
, SOCK_RCU_FREE
);
347 spin_unlock_bh(&hslot
->lock
);
351 EXPORT_SYMBOL(udp_lib_get_port
);
353 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
354 * match_wildcard == false: addresses must be exactly the same, i.e.
355 * 0.0.0.0 only equals to 0.0.0.0
357 int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
,
360 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
362 if (!ipv6_only_sock(sk2
)) {
363 if (inet1
->inet_rcv_saddr
== inet2
->inet_rcv_saddr
)
365 if (!inet1
->inet_rcv_saddr
|| !inet2
->inet_rcv_saddr
)
366 return match_wildcard
;
371 static u32
udp4_portaddr_hash(const struct net
*net
, __be32 saddr
,
374 return jhash_1word((__force u32
)saddr
, net_hash_mix(net
)) ^ port
;
377 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
379 unsigned int hash2_nulladdr
=
380 udp4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
381 unsigned int hash2_partial
=
382 udp4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
384 /* precompute partial secondary hash */
385 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
386 return udp_lib_get_port(sk
, snum
, ipv4_rcv_saddr_equal
, hash2_nulladdr
);
389 static inline int compute_score(struct sock
*sk
, struct net
*net
,
390 __be32 saddr
, unsigned short hnum
, __be16 sport
,
391 __be32 daddr
, __be16 dport
, int dif
)
394 struct inet_sock
*inet
;
396 if (!net_eq(sock_net(sk
), net
) ||
397 udp_sk(sk
)->udp_port_hash
!= hnum
||
401 score
= (sk
->sk_family
== PF_INET
) ? 2 : 1;
404 if (inet
->inet_rcv_saddr
) {
405 if (inet
->inet_rcv_saddr
!= daddr
)
410 if (inet
->inet_daddr
) {
411 if (inet
->inet_daddr
!= saddr
)
416 if (inet
->inet_dport
) {
417 if (inet
->inet_dport
!= sport
)
422 if (sk
->sk_bound_dev_if
) {
423 if (sk
->sk_bound_dev_if
!= dif
)
427 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
433 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
435 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
436 __be32 saddr
, __be16 sport
,
437 __be32 daddr
, unsigned int hnum
, int dif
)
440 struct inet_sock
*inet
;
442 if (!net_eq(sock_net(sk
), net
) ||
448 if (inet
->inet_rcv_saddr
!= daddr
||
449 inet
->inet_num
!= hnum
)
452 score
= (sk
->sk_family
== PF_INET
) ? 2 : 1;
454 if (inet
->inet_daddr
) {
455 if (inet
->inet_daddr
!= saddr
)
460 if (inet
->inet_dport
) {
461 if (inet
->inet_dport
!= sport
)
466 if (sk
->sk_bound_dev_if
) {
467 if (sk
->sk_bound_dev_if
!= dif
)
472 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
478 static u32
udp_ehashfn(const struct net
*net
, const __be32 laddr
,
479 const __u16 lport
, const __be32 faddr
,
482 static u32 udp_ehash_secret __read_mostly
;
484 net_get_random_once(&udp_ehash_secret
, sizeof(udp_ehash_secret
));
486 return __inet_ehashfn(laddr
, lport
, faddr
, fport
,
487 udp_ehash_secret
+ net_hash_mix(net
));
490 /* called with read_rcu_lock() */
491 static struct sock
*udp4_lib_lookup2(struct net
*net
,
492 __be32 saddr
, __be16 sport
,
493 __be32 daddr
, unsigned int hnum
, int dif
,
494 struct udp_hslot
*hslot2
, unsigned int slot2
,
497 struct sock
*sk
, *result
;
498 int score
, badness
, matches
= 0, reuseport
= 0;
503 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
504 score
= compute_score2(sk
, net
, saddr
, sport
,
506 if (score
> badness
) {
507 reuseport
= sk
->sk_reuseport
;
509 hash
= udp_ehashfn(net
, daddr
, hnum
,
511 result
= reuseport_select_sock(sk
, hash
, skb
,
512 sizeof(struct udphdr
));
519 } else if (score
== badness
&& reuseport
) {
521 if (reciprocal_scale(hash
, matches
) == 0)
523 hash
= next_pseudo_random32(hash
);
529 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
530 * harder than this. -DaveM
532 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
533 __be16 sport
, __be32 daddr
, __be16 dport
,
534 int dif
, struct udp_table
*udptable
, struct sk_buff
*skb
)
536 struct sock
*sk
, *result
;
537 unsigned short hnum
= ntohs(dport
);
538 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
539 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
540 int score
, badness
, matches
= 0, reuseport
= 0;
543 if (hslot
->count
> 10) {
544 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
);
545 slot2
= hash2
& udptable
->mask
;
546 hslot2
= &udptable
->hash2
[slot2
];
547 if (hslot
->count
< hslot2
->count
)
550 result
= udp4_lib_lookup2(net
, saddr
, sport
,
554 hash2
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
555 slot2
= hash2
& udptable
->mask
;
556 hslot2
= &udptable
->hash2
[slot2
];
557 if (hslot
->count
< hslot2
->count
)
560 result
= udp4_lib_lookup2(net
, saddr
, sport
,
561 htonl(INADDR_ANY
), hnum
, dif
,
569 sk_for_each_rcu(sk
, &hslot
->head
) {
570 score
= compute_score(sk
, net
, saddr
, hnum
, sport
,
572 if (score
> badness
) {
573 reuseport
= sk
->sk_reuseport
;
575 hash
= udp_ehashfn(net
, daddr
, hnum
,
577 result
= reuseport_select_sock(sk
, hash
, skb
,
578 sizeof(struct udphdr
));
585 } else if (score
== badness
&& reuseport
) {
587 if (reciprocal_scale(hash
, matches
) == 0)
589 hash
= next_pseudo_random32(hash
);
594 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
596 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
597 __be16 sport
, __be16 dport
,
598 struct udp_table
*udptable
)
600 const struct iphdr
*iph
= ip_hdr(skb
);
602 return __udp4_lib_lookup(dev_net(skb_dst(skb
)->dev
), iph
->saddr
, sport
,
603 iph
->daddr
, dport
, inet_iif(skb
),
607 /* Must be called under rcu_read_lock().
608 * Does increment socket refcount.
610 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
611 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
612 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
613 __be32 daddr
, __be16 dport
, int dif
)
617 sk
= __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
618 dif
, &udp_table
, NULL
);
619 if (sk
&& !atomic_inc_not_zero(&sk
->sk_refcnt
))
623 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
626 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
627 __be16 loc_port
, __be32 loc_addr
,
628 __be16 rmt_port
, __be32 rmt_addr
,
629 int dif
, unsigned short hnum
)
631 struct inet_sock
*inet
= inet_sk(sk
);
633 if (!net_eq(sock_net(sk
), net
) ||
634 udp_sk(sk
)->udp_port_hash
!= hnum
||
635 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
636 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
637 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
638 ipv6_only_sock(sk
) ||
639 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
))
641 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
))
647 * This routine is called by the ICMP module when it gets some
648 * sort of error condition. If err < 0 then the socket should
649 * be closed and the error returned to the user. If err > 0
650 * it's just the icmp type << 8 | icmp code.
651 * Header points to the ip header of the error packet. We move
652 * on past this. Then (as it used to claim before adjustment)
653 * header points to the first 8 bytes of the udp header. We need
654 * to find the appropriate port.
657 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
659 struct inet_sock
*inet
;
660 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
661 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
662 const int type
= icmp_hdr(skb
)->type
;
663 const int code
= icmp_hdr(skb
)->code
;
667 struct net
*net
= dev_net(skb
->dev
);
669 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
670 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
,
673 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
674 return; /* No socket for error */
683 case ICMP_TIME_EXCEEDED
:
686 case ICMP_SOURCE_QUENCH
:
688 case ICMP_PARAMETERPROB
:
692 case ICMP_DEST_UNREACH
:
693 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
694 ipv4_sk_update_pmtu(skb
, sk
, info
);
695 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
703 if (code
<= NR_ICMP_UNREACH
) {
704 harderr
= icmp_err_convert
[code
].fatal
;
705 err
= icmp_err_convert
[code
].errno
;
709 ipv4_sk_redirect(skb
, sk
);
714 * RFC1122: OK. Passes ICMP errors back to application, as per
717 if (!inet
->recverr
) {
718 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
721 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
724 sk
->sk_error_report(sk
);
729 void udp_err(struct sk_buff
*skb
, u32 info
)
731 __udp4_lib_err(skb
, info
, &udp_table
);
735 * Throw away all pending data and cancel the corking. Socket is locked.
737 void udp_flush_pending_frames(struct sock
*sk
)
739 struct udp_sock
*up
= udp_sk(sk
);
744 ip_flush_pending_frames(sk
);
747 EXPORT_SYMBOL(udp_flush_pending_frames
);
750 * udp4_hwcsum - handle outgoing HW checksumming
751 * @skb: sk_buff containing the filled-in UDP header
752 * (checksum field must be zeroed out)
753 * @src: source IP address
754 * @dst: destination IP address
756 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
758 struct udphdr
*uh
= udp_hdr(skb
);
759 int offset
= skb_transport_offset(skb
);
760 int len
= skb
->len
- offset
;
764 if (!skb_has_frag_list(skb
)) {
766 * Only one fragment on the socket.
768 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
769 skb
->csum_offset
= offsetof(struct udphdr
, check
);
770 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
773 struct sk_buff
*frags
;
776 * HW-checksum won't work as there are two or more
777 * fragments on the socket so that all csums of sk_buffs
780 skb_walk_frags(skb
, frags
) {
781 csum
= csum_add(csum
, frags
->csum
);
785 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
786 skb
->ip_summed
= CHECKSUM_NONE
;
788 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
790 uh
->check
= CSUM_MANGLED_0
;
793 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
795 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
796 * for the simple case like when setting the checksum for a UDP tunnel.
798 void udp_set_csum(bool nocheck
, struct sk_buff
*skb
,
799 __be32 saddr
, __be32 daddr
, int len
)
801 struct udphdr
*uh
= udp_hdr(skb
);
805 } else if (skb_is_gso(skb
)) {
806 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
807 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
809 uh
->check
= udp_v4_check(len
, saddr
, daddr
, lco_csum(skb
));
811 uh
->check
= CSUM_MANGLED_0
;
813 skb
->ip_summed
= CHECKSUM_PARTIAL
;
814 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
815 skb
->csum_offset
= offsetof(struct udphdr
, check
);
816 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
819 EXPORT_SYMBOL(udp_set_csum
);
821 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
)
823 struct sock
*sk
= skb
->sk
;
824 struct inet_sock
*inet
= inet_sk(sk
);
827 int is_udplite
= IS_UDPLITE(sk
);
828 int offset
= skb_transport_offset(skb
);
829 int len
= skb
->len
- offset
;
833 * Create a UDP header
836 uh
->source
= inet
->inet_sport
;
837 uh
->dest
= fl4
->fl4_dport
;
838 uh
->len
= htons(len
);
841 if (is_udplite
) /* UDP-Lite */
842 csum
= udplite_csum(skb
);
844 else if (sk
->sk_no_check_tx
) { /* UDP csum disabled */
846 skb
->ip_summed
= CHECKSUM_NONE
;
849 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
851 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
855 csum
= udp_csum(skb
);
857 /* add protocol-dependent pseudo-header */
858 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
859 sk
->sk_protocol
, csum
);
861 uh
->check
= CSUM_MANGLED_0
;
864 err
= ip_send_skb(sock_net(sk
), skb
);
866 if (err
== -ENOBUFS
&& !inet
->recverr
) {
867 UDP_INC_STATS_USER(sock_net(sk
),
868 UDP_MIB_SNDBUFERRORS
, is_udplite
);
872 UDP_INC_STATS_USER(sock_net(sk
),
873 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
878 * Push out all pending data as one UDP datagram. Socket is locked.
880 int udp_push_pending_frames(struct sock
*sk
)
882 struct udp_sock
*up
= udp_sk(sk
);
883 struct inet_sock
*inet
= inet_sk(sk
);
884 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
888 skb
= ip_finish_skb(sk
, fl4
);
892 err
= udp_send_skb(skb
, fl4
);
899 EXPORT_SYMBOL(udp_push_pending_frames
);
901 int udp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
903 struct inet_sock
*inet
= inet_sk(sk
);
904 struct udp_sock
*up
= udp_sk(sk
);
905 struct flowi4 fl4_stack
;
908 struct ipcm_cookie ipc
;
909 struct rtable
*rt
= NULL
;
912 __be32 daddr
, faddr
, saddr
;
915 int err
, is_udplite
= IS_UDPLITE(sk
);
916 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
917 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
919 struct ip_options_data opt_copy
;
928 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
936 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
938 fl4
= &inet
->cork
.fl
.u
.ip4
;
941 * There are pending frames.
942 * The socket lock must be held while it's corked.
945 if (likely(up
->pending
)) {
946 if (unlikely(up
->pending
!= AF_INET
)) {
954 ulen
+= sizeof(struct udphdr
);
957 * Get and verify the address.
960 DECLARE_SOCKADDR(struct sockaddr_in
*, usin
, msg
->msg_name
);
961 if (msg
->msg_namelen
< sizeof(*usin
))
963 if (usin
->sin_family
!= AF_INET
) {
964 if (usin
->sin_family
!= AF_UNSPEC
)
965 return -EAFNOSUPPORT
;
968 daddr
= usin
->sin_addr
.s_addr
;
969 dport
= usin
->sin_port
;
973 if (sk
->sk_state
!= TCP_ESTABLISHED
)
974 return -EDESTADDRREQ
;
975 daddr
= inet
->inet_daddr
;
976 dport
= inet
->inet_dport
;
977 /* Open fast path for connected socket.
978 Route will not be used, if at least one option is set.
983 ipc
.sockc
.tsflags
= sk
->sk_tsflags
;
984 ipc
.addr
= inet
->inet_saddr
;
985 ipc
.oif
= sk
->sk_bound_dev_if
;
987 if (msg
->msg_controllen
) {
988 err
= ip_cmsg_send(sk
, msg
, &ipc
, sk
->sk_family
== AF_INET6
);
998 struct ip_options_rcu
*inet_opt
;
1001 inet_opt
= rcu_dereference(inet
->inet_opt
);
1003 memcpy(&opt_copy
, inet_opt
,
1004 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
1005 ipc
.opt
= &opt_copy
.opt
;
1011 ipc
.addr
= faddr
= daddr
;
1013 sock_tx_timestamp(sk
, ipc
.sockc
.tsflags
, &ipc
.tx_flags
);
1015 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
1018 faddr
= ipc
.opt
->opt
.faddr
;
1021 tos
= get_rttos(&ipc
, inet
);
1022 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
1023 (msg
->msg_flags
& MSG_DONTROUTE
) ||
1024 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
1029 if (ipv4_is_multicast(daddr
)) {
1031 ipc
.oif
= inet
->mc_index
;
1033 saddr
= inet
->mc_addr
;
1035 } else if (!ipc
.oif
)
1036 ipc
.oif
= inet
->uc_index
;
1039 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
1042 struct net
*net
= sock_net(sk
);
1043 __u8 flow_flags
= inet_sk_flowi_flags(sk
);
1047 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
1048 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1050 faddr
, saddr
, dport
, inet
->inet_sport
);
1052 if (!saddr
&& ipc
.oif
) {
1053 err
= l3mdev_get_saddr(net
, ipc
.oif
, fl4
);
1058 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
1059 rt
= ip_route_output_flow(net
, fl4
, sk
);
1063 if (err
== -ENETUNREACH
)
1064 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
1069 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1070 !sock_flag(sk
, SOCK_BROADCAST
))
1073 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1076 if (msg
->msg_flags
&MSG_CONFIRM
)
1082 daddr
= ipc
.addr
= fl4
->daddr
;
1084 /* Lockless fast path for the non-corking case. */
1086 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
, ulen
,
1087 sizeof(struct udphdr
), &ipc
, &rt
,
1090 if (!IS_ERR_OR_NULL(skb
))
1091 err
= udp_send_skb(skb
, fl4
);
1096 if (unlikely(up
->pending
)) {
1097 /* The socket is already corked while preparing it. */
1098 /* ... which is an evident application bug. --ANK */
1101 net_dbg_ratelimited("cork app bug 2\n");
1106 * Now cork the socket to pend data.
1108 fl4
= &inet
->cork
.fl
.u
.ip4
;
1111 fl4
->fl4_dport
= dport
;
1112 fl4
->fl4_sport
= inet
->inet_sport
;
1113 up
->pending
= AF_INET
;
1117 err
= ip_append_data(sk
, fl4
, getfrag
, msg
, ulen
,
1118 sizeof(struct udphdr
), &ipc
, &rt
,
1119 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1121 udp_flush_pending_frames(sk
);
1123 err
= udp_push_pending_frames(sk
);
1124 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1135 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1136 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1137 * we don't have a good statistic (IpOutDiscards but it can be too many
1138 * things). We could add another new stat but at least for now that
1139 * seems like overkill.
1141 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1142 UDP_INC_STATS_USER(sock_net(sk
),
1143 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1148 dst_confirm(&rt
->dst
);
1149 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1150 goto back_from_confirm
;
1154 EXPORT_SYMBOL(udp_sendmsg
);
1156 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1157 size_t size
, int flags
)
1159 struct inet_sock
*inet
= inet_sk(sk
);
1160 struct udp_sock
*up
= udp_sk(sk
);
1163 if (flags
& MSG_SENDPAGE_NOTLAST
)
1167 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1169 /* Call udp_sendmsg to specify destination address which
1170 * sendpage interface can't pass.
1171 * This will succeed only when the socket is connected.
1173 ret
= udp_sendmsg(sk
, &msg
, 0);
1180 if (unlikely(!up
->pending
)) {
1183 net_dbg_ratelimited("udp cork app bug 3\n");
1187 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1188 page
, offset
, size
, flags
);
1189 if (ret
== -EOPNOTSUPP
) {
1191 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1195 udp_flush_pending_frames(sk
);
1200 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1201 ret
= udp_push_pending_frames(sk
);
1210 * first_packet_length - return length of first packet in receive queue
1213 * Drops all bad checksum frames, until a valid one is found.
1214 * Returns the length of found skb, or 0 if none is found.
1216 static unsigned int first_packet_length(struct sock
*sk
)
1218 struct sk_buff_head list_kill
, *rcvq
= &sk
->sk_receive_queue
;
1219 struct sk_buff
*skb
;
1222 __skb_queue_head_init(&list_kill
);
1224 spin_lock_bh(&rcvq
->lock
);
1225 while ((skb
= skb_peek(rcvq
)) != NULL
&&
1226 udp_lib_checksum_complete(skb
)) {
1227 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1229 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1231 atomic_inc(&sk
->sk_drops
);
1232 __skb_unlink(skb
, rcvq
);
1233 __skb_queue_tail(&list_kill
, skb
);
1235 res
= skb
? skb
->len
: 0;
1236 spin_unlock_bh(&rcvq
->lock
);
1238 if (!skb_queue_empty(&list_kill
)) {
1239 bool slow
= lock_sock_fast(sk
);
1241 __skb_queue_purge(&list_kill
);
1242 sk_mem_reclaim_partial(sk
);
1243 unlock_sock_fast(sk
, slow
);
1249 * IOCTL requests applicable to the UDP protocol
1252 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1257 int amount
= sk_wmem_alloc_get(sk
);
1259 return put_user(amount
, (int __user
*)arg
);
1264 unsigned int amount
= first_packet_length(sk
);
1268 * We will only return the amount
1269 * of this packet since that is all
1270 * that will be read.
1272 amount
-= sizeof(struct udphdr
);
1274 return put_user(amount
, (int __user
*)arg
);
1278 return -ENOIOCTLCMD
;
1283 EXPORT_SYMBOL(udp_ioctl
);
1286 * This should be easy, if there is something there we
1287 * return it, otherwise we block.
1290 int udp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int noblock
,
1291 int flags
, int *addr_len
)
1293 struct inet_sock
*inet
= inet_sk(sk
);
1294 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, msg
->msg_name
);
1295 struct sk_buff
*skb
;
1296 unsigned int ulen
, copied
;
1297 int peeked
, off
= 0;
1299 int is_udplite
= IS_UDPLITE(sk
);
1300 bool checksum_valid
= false;
1303 if (flags
& MSG_ERRQUEUE
)
1304 return ip_recv_error(sk
, msg
, len
, addr_len
);
1307 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
1308 &peeked
, &off
, &err
);
1316 else if (copied
< ulen
)
1317 msg
->msg_flags
|= MSG_TRUNC
;
1320 * If checksum is needed at all, try to do it while copying the
1321 * data. If the data is truncated, or if we only want a partial
1322 * coverage checksum (UDP-Lite), do it before the copy.
1325 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
1326 checksum_valid
= !udp_lib_checksum_complete(skb
);
1327 if (!checksum_valid
)
1331 if (checksum_valid
|| skb_csum_unnecessary(skb
))
1332 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
1334 err
= skb_copy_and_csum_datagram_msg(skb
, 0, msg
);
1340 if (unlikely(err
)) {
1341 trace_kfree_skb(skb
, udp_recvmsg
);
1343 atomic_inc(&sk
->sk_drops
);
1344 UDP_INC_STATS_USER(sock_net(sk
),
1345 UDP_MIB_INERRORS
, is_udplite
);
1351 UDP_INC_STATS_USER(sock_net(sk
),
1352 UDP_MIB_INDATAGRAMS
, is_udplite
);
1354 sock_recv_ts_and_drops(msg
, sk
, skb
);
1356 /* Copy the address. */
1358 sin
->sin_family
= AF_INET
;
1359 sin
->sin_port
= udp_hdr(skb
)->source
;
1360 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1361 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1362 *addr_len
= sizeof(*sin
);
1364 if (inet
->cmsg_flags
)
1365 ip_cmsg_recv_offset(msg
, skb
, sizeof(struct udphdr
));
1368 if (flags
& MSG_TRUNC
)
1372 skb_free_datagram_locked(sk
, skb
);
1377 slow
= lock_sock_fast(sk
);
1378 if (!skb_kill_datagram(sk
, skb
, flags
)) {
1379 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1380 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1382 unlock_sock_fast(sk
, slow
);
1384 /* starting over for a new packet, but check if we need to yield */
1386 msg
->msg_flags
&= ~MSG_TRUNC
;
1390 int udp_disconnect(struct sock
*sk
, int flags
)
1392 struct inet_sock
*inet
= inet_sk(sk
);
1394 * 1003.1g - break association.
1397 sk
->sk_state
= TCP_CLOSE
;
1398 inet
->inet_daddr
= 0;
1399 inet
->inet_dport
= 0;
1400 sock_rps_reset_rxhash(sk
);
1401 sk
->sk_bound_dev_if
= 0;
1402 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1403 inet_reset_saddr(sk
);
1405 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1406 sk
->sk_prot
->unhash(sk
);
1407 inet
->inet_sport
= 0;
1412 EXPORT_SYMBOL(udp_disconnect
);
1414 void udp_lib_unhash(struct sock
*sk
)
1416 if (sk_hashed(sk
)) {
1417 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1418 struct udp_hslot
*hslot
, *hslot2
;
1420 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1421 udp_sk(sk
)->udp_port_hash
);
1422 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1424 spin_lock_bh(&hslot
->lock
);
1425 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1426 reuseport_detach_sock(sk
);
1427 if (sk_del_node_init_rcu(sk
)) {
1429 inet_sk(sk
)->inet_num
= 0;
1430 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1432 spin_lock(&hslot2
->lock
);
1433 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1435 spin_unlock(&hslot2
->lock
);
1437 spin_unlock_bh(&hslot
->lock
);
1440 EXPORT_SYMBOL(udp_lib_unhash
);
1443 * inet_rcv_saddr was changed, we must rehash secondary hash
1445 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1447 if (sk_hashed(sk
)) {
1448 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1449 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1451 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1452 nhslot2
= udp_hashslot2(udptable
, newhash
);
1453 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1455 if (hslot2
!= nhslot2
||
1456 rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1457 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1458 udp_sk(sk
)->udp_port_hash
);
1459 /* we must lock primary chain too */
1460 spin_lock_bh(&hslot
->lock
);
1461 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1462 reuseport_detach_sock(sk
);
1464 if (hslot2
!= nhslot2
) {
1465 spin_lock(&hslot2
->lock
);
1466 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1468 spin_unlock(&hslot2
->lock
);
1470 spin_lock(&nhslot2
->lock
);
1471 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1474 spin_unlock(&nhslot2
->lock
);
1477 spin_unlock_bh(&hslot
->lock
);
1481 EXPORT_SYMBOL(udp_lib_rehash
);
1483 static void udp_v4_rehash(struct sock
*sk
)
1485 u16 new_hash
= udp4_portaddr_hash(sock_net(sk
),
1486 inet_sk(sk
)->inet_rcv_saddr
,
1487 inet_sk(sk
)->inet_num
);
1488 udp_lib_rehash(sk
, new_hash
);
1491 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1495 if (inet_sk(sk
)->inet_daddr
) {
1496 sock_rps_save_rxhash(sk
, skb
);
1497 sk_mark_napi_id(sk
, skb
);
1498 sk_incoming_cpu_update(sk
);
1501 rc
= __sock_queue_rcv_skb(sk
, skb
);
1503 int is_udplite
= IS_UDPLITE(sk
);
1505 /* Note that an ENOMEM error is charged twice */
1507 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1509 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1511 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1519 static struct static_key udp_encap_needed __read_mostly
;
1520 void udp_encap_enable(void)
1522 if (!static_key_enabled(&udp_encap_needed
))
1523 static_key_slow_inc(&udp_encap_needed
);
1525 EXPORT_SYMBOL(udp_encap_enable
);
1530 * >0: "udp encap" protocol resubmission
1532 * Note that in the success and error cases, the skb is assumed to
1533 * have either been requeued or freed.
1535 int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1537 struct udp_sock
*up
= udp_sk(sk
);
1539 int is_udplite
= IS_UDPLITE(sk
);
1542 * Charge it to the socket, dropping if the queue is full.
1544 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1548 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1549 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1552 * This is an encapsulation socket so pass the skb to
1553 * the socket's udp_encap_rcv() hook. Otherwise, just
1554 * fall through and pass this up the UDP socket.
1555 * up->encap_rcv() returns the following value:
1556 * =0 if skb was successfully passed to the encap
1557 * handler or was discarded by it.
1558 * >0 if skb should be passed on to UDP.
1559 * <0 if skb should be resubmitted as proto -N
1562 /* if we're overly short, let UDP handle it */
1563 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
1564 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
) {
1567 /* Verify checksum before giving to encap */
1568 if (udp_lib_checksum_complete(skb
))
1571 ret
= encap_rcv(sk
, skb
);
1573 UDP_INC_STATS_BH(sock_net(sk
),
1574 UDP_MIB_INDATAGRAMS
,
1580 /* FALLTHROUGH -- it's a UDP Packet */
1584 * UDP-Lite specific tests, ignored on UDP sockets
1586 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1589 * MIB statistics other than incrementing the error count are
1590 * disabled for the following two types of errors: these depend
1591 * on the application settings, not on the functioning of the
1592 * protocol stack as such.
1594 * RFC 3828 here recommends (sec 3.3): "There should also be a
1595 * way ... to ... at least let the receiving application block
1596 * delivery of packets with coverage values less than a value
1597 * provided by the application."
1599 if (up
->pcrlen
== 0) { /* full coverage was set */
1600 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1601 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1604 /* The next case involves violating the min. coverage requested
1605 * by the receiver. This is subtle: if receiver wants x and x is
1606 * greater than the buffersize/MTU then receiver will complain
1607 * that it wants x while sender emits packets of smaller size y.
1608 * Therefore the above ...()->partial_cov statement is essential.
1610 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1611 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1612 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1617 if (rcu_access_pointer(sk
->sk_filter
)) {
1618 if (udp_lib_checksum_complete(skb
))
1620 if (sk_filter(sk
, skb
))
1624 udp_csum_pull_header(skb
);
1625 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
1626 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1633 ipv4_pktinfo_prepare(sk
, skb
);
1635 if (!sock_owned_by_user(sk
))
1636 rc
= __udp_queue_rcv_skb(sk
, skb
);
1637 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
1646 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1648 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1649 atomic_inc(&sk
->sk_drops
);
1654 /* For TCP sockets, sk_rx_dst is protected by socket lock
1655 * For UDP, we use xchg() to guard against concurrent changes.
1657 static void udp_sk_rx_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
1659 struct dst_entry
*old
;
1662 old
= xchg(&sk
->sk_rx_dst
, dst
);
1667 * Multicasts and broadcasts go to each listener.
1669 * Note: called only from the BH handler context.
1671 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1673 __be32 saddr
, __be32 daddr
,
1674 struct udp_table
*udptable
,
1677 struct sock
*sk
, *first
= NULL
;
1678 unsigned short hnum
= ntohs(uh
->dest
);
1679 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
1680 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
1681 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
1682 int dif
= skb
->dev
->ifindex
;
1683 struct hlist_node
*node
;
1684 struct sk_buff
*nskb
;
1687 hash2_any
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
) &
1689 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
) & udp_table
.mask
;
1691 hslot
= &udp_table
.hash2
[hash2
];
1692 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
1695 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
1696 if (!__udp_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
1697 uh
->source
, saddr
, dif
, hnum
))
1704 nskb
= skb_clone(skb
, GFP_ATOMIC
);
1706 if (unlikely(!nskb
)) {
1707 atomic_inc(&sk
->sk_drops
);
1708 UDP_INC_STATS_BH(net
, UDP_MIB_RCVBUFERRORS
,
1710 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
,
1714 if (udp_queue_rcv_skb(sk
, nskb
) > 0)
1718 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1719 if (use_hash2
&& hash2
!= hash2_any
) {
1725 if (udp_queue_rcv_skb(first
, skb
) > 0)
1729 UDP_INC_STATS_BH(net
, UDP_MIB_IGNOREDMULTI
,
1730 proto
== IPPROTO_UDPLITE
);
1735 /* Initialize UDP checksum. If exited with zero value (success),
1736 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1737 * Otherwise, csum completion requires chacksumming packet body,
1738 * including udp header and folding it to skb->csum.
1740 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1745 UDP_SKB_CB(skb
)->partial_cov
= 0;
1746 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1748 if (proto
== IPPROTO_UDPLITE
) {
1749 err
= udplite_checksum_init(skb
, uh
);
1754 return skb_checksum_init_zero_check(skb
, proto
, uh
->check
,
1755 inet_compute_pseudo
);
1759 * All we need to do is get the socket, and then do a checksum.
1762 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
1767 unsigned short ulen
;
1768 struct rtable
*rt
= skb_rtable(skb
);
1769 __be32 saddr
, daddr
;
1770 struct net
*net
= dev_net(skb
->dev
);
1773 * Validate the packet.
1775 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1776 goto drop
; /* No space for header. */
1779 ulen
= ntohs(uh
->len
);
1780 saddr
= ip_hdr(skb
)->saddr
;
1781 daddr
= ip_hdr(skb
)->daddr
;
1783 if (ulen
> skb
->len
)
1786 if (proto
== IPPROTO_UDP
) {
1787 /* UDP validates ulen. */
1788 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1793 if (udp4_csum_init(skb
, uh
, proto
))
1796 sk
= skb_steal_sock(skb
);
1798 struct dst_entry
*dst
= skb_dst(skb
);
1801 if (unlikely(sk
->sk_rx_dst
!= dst
))
1802 udp_sk_rx_dst_set(sk
, dst
);
1804 ret
= udp_queue_rcv_skb(sk
, skb
);
1806 /* a return value > 0 means to resubmit the input, but
1807 * it wants the return to be -protocol, or 0
1814 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1815 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
1816 saddr
, daddr
, udptable
, proto
);
1818 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
1822 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
1823 skb_checksum_try_convert(skb
, IPPROTO_UDP
, uh
->check
,
1824 inet_compute_pseudo
);
1826 ret
= udp_queue_rcv_skb(sk
, skb
);
1828 /* a return value > 0 means to resubmit the input, but
1829 * it wants the return to be -protocol, or 0
1836 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1840 /* No socket. Drop packet silently, if checksum is wrong */
1841 if (udp_lib_checksum_complete(skb
))
1844 UDP_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1845 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1848 * Hmm. We got an UDP packet to a port to which we
1849 * don't wanna listen. Ignore it.
1855 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1856 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1857 &saddr
, ntohs(uh
->source
),
1859 &daddr
, ntohs(uh
->dest
));
1864 * RFC1122: OK. Discards the bad packet silently (as far as
1865 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1867 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1868 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1869 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
1871 UDP_INC_STATS_BH(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
1873 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1878 /* We can only early demux multicast if there is a single matching socket.
1879 * If more than one socket found returns NULL
1881 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
1882 __be16 loc_port
, __be32 loc_addr
,
1883 __be16 rmt_port
, __be32 rmt_addr
,
1886 struct sock
*sk
, *result
;
1887 unsigned short hnum
= ntohs(loc_port
);
1888 unsigned int slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
1889 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
1891 /* Do not bother scanning a too big list */
1892 if (hslot
->count
> 10)
1896 sk_for_each_rcu(sk
, &hslot
->head
) {
1897 if (__udp_is_mcast_sock(net
, sk
, loc_port
, loc_addr
,
1898 rmt_port
, rmt_addr
, dif
, hnum
)) {
1908 /* For unicast we should only early demux connected sockets or we can
1909 * break forwarding setups. The chains here can be long so only check
1910 * if the first socket is an exact match and if not move on.
1912 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
1913 __be16 loc_port
, __be32 loc_addr
,
1914 __be16 rmt_port
, __be32 rmt_addr
,
1917 unsigned short hnum
= ntohs(loc_port
);
1918 unsigned int hash2
= udp4_portaddr_hash(net
, loc_addr
, hnum
);
1919 unsigned int slot2
= hash2
& udp_table
.mask
;
1920 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
1921 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
);
1922 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
1925 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
1926 if (INET_MATCH(sk
, net
, acookie
, rmt_addr
,
1927 loc_addr
, ports
, dif
))
1929 /* Only check first socket in chain */
1935 void udp_v4_early_demux(struct sk_buff
*skb
)
1937 struct net
*net
= dev_net(skb
->dev
);
1938 const struct iphdr
*iph
;
1939 const struct udphdr
*uh
;
1940 struct sock
*sk
= NULL
;
1941 struct dst_entry
*dst
;
1942 int dif
= skb
->dev
->ifindex
;
1945 /* validate the packet */
1946 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
1952 if (skb
->pkt_type
== PACKET_BROADCAST
||
1953 skb
->pkt_type
== PACKET_MULTICAST
) {
1954 struct in_device
*in_dev
= __in_dev_get_rcu(skb
->dev
);
1959 /* we are supposed to accept bcast packets */
1960 if (skb
->pkt_type
== PACKET_MULTICAST
) {
1961 ours
= ip_check_mc_rcu(in_dev
, iph
->daddr
, iph
->saddr
,
1967 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
1968 uh
->source
, iph
->saddr
, dif
);
1969 } else if (skb
->pkt_type
== PACKET_HOST
) {
1970 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
1971 uh
->source
, iph
->saddr
, dif
);
1974 if (!sk
|| !atomic_inc_not_zero_hint(&sk
->sk_refcnt
, 2))
1978 skb
->destructor
= sock_efree
;
1979 dst
= READ_ONCE(sk
->sk_rx_dst
);
1982 dst
= dst_check(dst
, 0);
1984 /* DST_NOCACHE can not be used without taking a reference */
1985 if (dst
->flags
& DST_NOCACHE
) {
1986 if (likely(atomic_inc_not_zero(&dst
->__refcnt
)))
1987 skb_dst_set(skb
, dst
);
1989 skb_dst_set_noref(skb
, dst
);
1994 int udp_rcv(struct sk_buff
*skb
)
1996 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
1999 void udp_destroy_sock(struct sock
*sk
)
2001 struct udp_sock
*up
= udp_sk(sk
);
2002 bool slow
= lock_sock_fast(sk
);
2003 udp_flush_pending_frames(sk
);
2004 unlock_sock_fast(sk
, slow
);
2005 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
2006 void (*encap_destroy
)(struct sock
*sk
);
2007 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
2014 * Socket option code for UDP
2016 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
2017 char __user
*optval
, unsigned int optlen
,
2018 int (*push_pending_frames
)(struct sock
*))
2020 struct udp_sock
*up
= udp_sk(sk
);
2023 int is_udplite
= IS_UDPLITE(sk
);
2025 if (optlen
< sizeof(int))
2028 if (get_user(val
, (int __user
*)optval
))
2031 valbool
= val
? 1 : 0;
2040 push_pending_frames(sk
);
2048 case UDP_ENCAP_ESPINUDP
:
2049 case UDP_ENCAP_ESPINUDP_NON_IKE
:
2050 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
2052 case UDP_ENCAP_L2TPINUDP
:
2053 up
->encap_type
= val
;
2062 case UDP_NO_CHECK6_TX
:
2063 up
->no_check6_tx
= valbool
;
2066 case UDP_NO_CHECK6_RX
:
2067 up
->no_check6_rx
= valbool
;
2071 * UDP-Lite's partial checksum coverage (RFC 3828).
2073 /* The sender sets actual checksum coverage length via this option.
2074 * The case coverage > packet length is handled by send module. */
2075 case UDPLITE_SEND_CSCOV
:
2076 if (!is_udplite
) /* Disable the option on UDP sockets */
2077 return -ENOPROTOOPT
;
2078 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2080 else if (val
> USHRT_MAX
)
2083 up
->pcflag
|= UDPLITE_SEND_CC
;
2086 /* The receiver specifies a minimum checksum coverage value. To make
2087 * sense, this should be set to at least 8 (as done below). If zero is
2088 * used, this again means full checksum coverage. */
2089 case UDPLITE_RECV_CSCOV
:
2090 if (!is_udplite
) /* Disable the option on UDP sockets */
2091 return -ENOPROTOOPT
;
2092 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2094 else if (val
> USHRT_MAX
)
2097 up
->pcflag
|= UDPLITE_RECV_CC
;
2107 EXPORT_SYMBOL(udp_lib_setsockopt
);
2109 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2110 char __user
*optval
, unsigned int optlen
)
2112 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2113 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2114 udp_push_pending_frames
);
2115 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2118 #ifdef CONFIG_COMPAT
2119 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2120 char __user
*optval
, unsigned int optlen
)
2122 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2123 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2124 udp_push_pending_frames
);
2125 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2129 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2130 char __user
*optval
, int __user
*optlen
)
2132 struct udp_sock
*up
= udp_sk(sk
);
2135 if (get_user(len
, optlen
))
2138 len
= min_t(unsigned int, len
, sizeof(int));
2149 val
= up
->encap_type
;
2152 case UDP_NO_CHECK6_TX
:
2153 val
= up
->no_check6_tx
;
2156 case UDP_NO_CHECK6_RX
:
2157 val
= up
->no_check6_rx
;
2160 /* The following two cannot be changed on UDP sockets, the return is
2161 * always 0 (which corresponds to the full checksum coverage of UDP). */
2162 case UDPLITE_SEND_CSCOV
:
2166 case UDPLITE_RECV_CSCOV
:
2171 return -ENOPROTOOPT
;
2174 if (put_user(len
, optlen
))
2176 if (copy_to_user(optval
, &val
, len
))
2180 EXPORT_SYMBOL(udp_lib_getsockopt
);
2182 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2183 char __user
*optval
, int __user
*optlen
)
2185 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2186 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2187 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2190 #ifdef CONFIG_COMPAT
2191 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2192 char __user
*optval
, int __user
*optlen
)
2194 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2195 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2196 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2200 * udp_poll - wait for a UDP event.
2201 * @file - file struct
2203 * @wait - poll table
2205 * This is same as datagram poll, except for the special case of
2206 * blocking sockets. If application is using a blocking fd
2207 * and a packet with checksum error is in the queue;
2208 * then it could get return from select indicating data available
2209 * but then block when reading it. Add special case code
2210 * to work around these arguably broken applications.
2212 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2214 unsigned int mask
= datagram_poll(file
, sock
, wait
);
2215 struct sock
*sk
= sock
->sk
;
2217 sock_rps_record_flow(sk
);
2219 /* Check for false positives due to checksum errors */
2220 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2221 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && !first_packet_length(sk
))
2222 mask
&= ~(POLLIN
| POLLRDNORM
);
2227 EXPORT_SYMBOL(udp_poll
);
2229 struct proto udp_prot
= {
2231 .owner
= THIS_MODULE
,
2232 .close
= udp_lib_close
,
2233 .connect
= ip4_datagram_connect
,
2234 .disconnect
= udp_disconnect
,
2236 .destroy
= udp_destroy_sock
,
2237 .setsockopt
= udp_setsockopt
,
2238 .getsockopt
= udp_getsockopt
,
2239 .sendmsg
= udp_sendmsg
,
2240 .recvmsg
= udp_recvmsg
,
2241 .sendpage
= udp_sendpage
,
2242 .backlog_rcv
= __udp_queue_rcv_skb
,
2243 .release_cb
= ip4_datagram_release_cb
,
2244 .hash
= udp_lib_hash
,
2245 .unhash
= udp_lib_unhash
,
2246 .rehash
= udp_v4_rehash
,
2247 .get_port
= udp_v4_get_port
,
2248 .memory_allocated
= &udp_memory_allocated
,
2249 .sysctl_mem
= sysctl_udp_mem
,
2250 .sysctl_wmem
= &sysctl_udp_wmem_min
,
2251 .sysctl_rmem
= &sysctl_udp_rmem_min
,
2252 .obj_size
= sizeof(struct udp_sock
),
2253 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2254 .h
.udp_table
= &udp_table
,
2255 #ifdef CONFIG_COMPAT
2256 .compat_setsockopt
= compat_udp_setsockopt
,
2257 .compat_getsockopt
= compat_udp_getsockopt
,
2259 .clear_sk
= sk_prot_clear_portaddr_nulls
,
2261 EXPORT_SYMBOL(udp_prot
);
2263 /* ------------------------------------------------------------------------ */
2264 #ifdef CONFIG_PROC_FS
2266 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2269 struct udp_iter_state
*state
= seq
->private;
2270 struct net
*net
= seq_file_net(seq
);
2272 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
2274 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
2276 if (hlist_empty(&hslot
->head
))
2279 spin_lock_bh(&hslot
->lock
);
2280 sk_for_each(sk
, &hslot
->head
) {
2281 if (!net_eq(sock_net(sk
), net
))
2283 if (sk
->sk_family
== state
->family
)
2286 spin_unlock_bh(&hslot
->lock
);
2293 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2295 struct udp_iter_state
*state
= seq
->private;
2296 struct net
*net
= seq_file_net(seq
);
2300 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
2303 if (state
->bucket
<= state
->udp_table
->mask
)
2304 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2305 return udp_get_first(seq
, state
->bucket
+ 1);
2310 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2312 struct sock
*sk
= udp_get_first(seq
, 0);
2315 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2317 return pos
? NULL
: sk
;
2320 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2322 struct udp_iter_state
*state
= seq
->private;
2323 state
->bucket
= MAX_UDP_PORTS
;
2325 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2328 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2332 if (v
== SEQ_START_TOKEN
)
2333 sk
= udp_get_idx(seq
, 0);
2335 sk
= udp_get_next(seq
, v
);
2341 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
2343 struct udp_iter_state
*state
= seq
->private;
2345 if (state
->bucket
<= state
->udp_table
->mask
)
2346 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2349 int udp_seq_open(struct inode
*inode
, struct file
*file
)
2351 struct udp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2352 struct udp_iter_state
*s
;
2355 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2356 sizeof(struct udp_iter_state
));
2360 s
= ((struct seq_file
*)file
->private_data
)->private;
2361 s
->family
= afinfo
->family
;
2362 s
->udp_table
= afinfo
->udp_table
;
2365 EXPORT_SYMBOL(udp_seq_open
);
2367 /* ------------------------------------------------------------------------ */
2368 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2370 struct proc_dir_entry
*p
;
2373 afinfo
->seq_ops
.start
= udp_seq_start
;
2374 afinfo
->seq_ops
.next
= udp_seq_next
;
2375 afinfo
->seq_ops
.stop
= udp_seq_stop
;
2377 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2378 afinfo
->seq_fops
, afinfo
);
2383 EXPORT_SYMBOL(udp_proc_register
);
2385 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2387 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2389 EXPORT_SYMBOL(udp_proc_unregister
);
2391 /* ------------------------------------------------------------------------ */
2392 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2395 struct inet_sock
*inet
= inet_sk(sp
);
2396 __be32 dest
= inet
->inet_daddr
;
2397 __be32 src
= inet
->inet_rcv_saddr
;
2398 __u16 destp
= ntohs(inet
->inet_dport
);
2399 __u16 srcp
= ntohs(inet
->inet_sport
);
2401 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2402 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2403 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2404 sk_wmem_alloc_get(sp
),
2405 sk_rmem_alloc_get(sp
),
2407 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2409 atomic_read(&sp
->sk_refcnt
), sp
,
2410 atomic_read(&sp
->sk_drops
));
2413 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2415 seq_setwidth(seq
, 127);
2416 if (v
== SEQ_START_TOKEN
)
2417 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2418 "rx_queue tr tm->when retrnsmt uid timeout "
2419 "inode ref pointer drops");
2421 struct udp_iter_state
*state
= seq
->private;
2423 udp4_format_sock(v
, seq
, state
->bucket
);
2429 static const struct file_operations udp_afinfo_seq_fops
= {
2430 .owner
= THIS_MODULE
,
2431 .open
= udp_seq_open
,
2433 .llseek
= seq_lseek
,
2434 .release
= seq_release_net
2437 /* ------------------------------------------------------------------------ */
2438 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2441 .udp_table
= &udp_table
,
2442 .seq_fops
= &udp_afinfo_seq_fops
,
2444 .show
= udp4_seq_show
,
2448 static int __net_init
udp4_proc_init_net(struct net
*net
)
2450 return udp_proc_register(net
, &udp4_seq_afinfo
);
2453 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2455 udp_proc_unregister(net
, &udp4_seq_afinfo
);
2458 static struct pernet_operations udp4_net_ops
= {
2459 .init
= udp4_proc_init_net
,
2460 .exit
= udp4_proc_exit_net
,
2463 int __init
udp4_proc_init(void)
2465 return register_pernet_subsys(&udp4_net_ops
);
2468 void udp4_proc_exit(void)
2470 unregister_pernet_subsys(&udp4_net_ops
);
2472 #endif /* CONFIG_PROC_FS */
2474 static __initdata
unsigned long uhash_entries
;
2475 static int __init
set_uhash_entries(char *str
)
2482 ret
= kstrtoul(str
, 0, &uhash_entries
);
2486 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2487 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2490 __setup("uhash_entries=", set_uhash_entries
);
2492 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2496 table
->hash
= alloc_large_system_hash(name
,
2497 2 * sizeof(struct udp_hslot
),
2499 21, /* one slot per 2 MB */
2503 UDP_HTABLE_SIZE_MIN
,
2506 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2507 for (i
= 0; i
<= table
->mask
; i
++) {
2508 INIT_HLIST_HEAD(&table
->hash
[i
].head
);
2509 table
->hash
[i
].count
= 0;
2510 spin_lock_init(&table
->hash
[i
].lock
);
2512 for (i
= 0; i
<= table
->mask
; i
++) {
2513 INIT_HLIST_HEAD(&table
->hash2
[i
].head
);
2514 table
->hash2
[i
].count
= 0;
2515 spin_lock_init(&table
->hash2
[i
].lock
);
2519 u32
udp_flow_hashrnd(void)
2521 static u32 hashrnd __read_mostly
;
2523 net_get_random_once(&hashrnd
, sizeof(hashrnd
));
2527 EXPORT_SYMBOL(udp_flow_hashrnd
);
2529 void __init
udp_init(void)
2531 unsigned long limit
;
2533 udp_table_init(&udp_table
, "UDP");
2534 limit
= nr_free_buffer_pages() / 8;
2535 limit
= max(limit
, 128UL);
2536 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2537 sysctl_udp_mem
[1] = limit
;
2538 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2540 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2541 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;