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 <linux/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>
117 #include <net/addrconf.h>
119 struct udp_table udp_table __read_mostly
;
120 EXPORT_SYMBOL(udp_table
);
122 long sysctl_udp_mem
[3] __read_mostly
;
123 EXPORT_SYMBOL(sysctl_udp_mem
);
125 atomic_long_t udp_memory_allocated
;
126 EXPORT_SYMBOL(udp_memory_allocated
);
128 #define MAX_UDP_PORTS 65536
129 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
131 /* IPCB reference means this can not be used from early demux */
132 static bool udp_lib_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
134 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
135 if (!net
->ipv4
.sysctl_udp_l3mdev_accept
&&
136 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
142 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
143 const struct udp_hslot
*hslot
,
144 unsigned long *bitmap
,
145 struct sock
*sk
, unsigned int log
)
148 kuid_t uid
= sock_i_uid(sk
);
150 sk_for_each(sk2
, &hslot
->head
) {
151 if (net_eq(sock_net(sk2
), net
) &&
153 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
154 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
155 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
156 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
157 inet_rcv_saddr_equal(sk
, sk2
, true)) {
158 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
159 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
160 uid_eq(uid
, sock_i_uid(sk2
))) {
166 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
175 * Note: we still hold spinlock of primary hash chain, so no other writer
176 * can insert/delete a socket with local_port == num
178 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
179 struct udp_hslot
*hslot2
,
183 kuid_t uid
= sock_i_uid(sk
);
186 spin_lock(&hslot2
->lock
);
187 udp_portaddr_for_each_entry(sk2
, &hslot2
->head
) {
188 if (net_eq(sock_net(sk2
), net
) &&
190 (udp_sk(sk2
)->udp_port_hash
== num
) &&
191 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
192 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
193 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
194 inet_rcv_saddr_equal(sk
, sk2
, true)) {
195 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
196 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
197 uid_eq(uid
, sock_i_uid(sk2
))) {
205 spin_unlock(&hslot2
->lock
);
209 static int udp_reuseport_add_sock(struct sock
*sk
, struct udp_hslot
*hslot
)
211 struct net
*net
= sock_net(sk
);
212 kuid_t uid
= sock_i_uid(sk
);
215 sk_for_each(sk2
, &hslot
->head
) {
216 if (net_eq(sock_net(sk2
), net
) &&
218 sk2
->sk_family
== sk
->sk_family
&&
219 ipv6_only_sock(sk2
) == ipv6_only_sock(sk
) &&
220 (udp_sk(sk2
)->udp_port_hash
== udp_sk(sk
)->udp_port_hash
) &&
221 (sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
222 sk2
->sk_reuseport
&& uid_eq(uid
, sock_i_uid(sk2
)) &&
223 inet_rcv_saddr_equal(sk
, sk2
, false)) {
224 return reuseport_add_sock(sk
, sk2
,
225 inet_rcv_saddr_any(sk
));
229 return reuseport_alloc(sk
, inet_rcv_saddr_any(sk
));
233 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
235 * @sk: socket struct in question
236 * @snum: port number to look up
237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
240 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
241 unsigned int hash2_nulladdr
)
243 struct udp_hslot
*hslot
, *hslot2
;
244 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
246 struct net
*net
= sock_net(sk
);
249 int low
, high
, remaining
;
251 unsigned short first
, last
;
252 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
254 inet_get_local_port_range(net
, &low
, &high
);
255 remaining
= (high
- low
) + 1;
257 rand
= prandom_u32();
258 first
= reciprocal_scale(rand
, remaining
) + low
;
260 * force rand to be an odd multiple of UDP_HTABLE_SIZE
262 rand
= (rand
| 1) * (udptable
->mask
+ 1);
263 last
= first
+ udptable
->mask
+ 1;
265 hslot
= udp_hashslot(udptable
, net
, first
);
266 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
267 spin_lock_bh(&hslot
->lock
);
268 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
273 * Iterate on all possible values of snum for this hash.
274 * Using steps of an odd multiple of UDP_HTABLE_SIZE
275 * give us randomization and full range coverage.
278 if (low
<= snum
&& snum
<= high
&&
279 !test_bit(snum
>> udptable
->log
, bitmap
) &&
280 !inet_is_local_reserved_port(net
, snum
))
283 } while (snum
!= first
);
284 spin_unlock_bh(&hslot
->lock
);
286 } while (++first
!= last
);
289 hslot
= udp_hashslot(udptable
, net
, snum
);
290 spin_lock_bh(&hslot
->lock
);
291 if (hslot
->count
> 10) {
293 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
295 slot2
&= udptable
->mask
;
296 hash2_nulladdr
&= udptable
->mask
;
298 hslot2
= udp_hashslot2(udptable
, slot2
);
299 if (hslot
->count
< hslot2
->count
)
300 goto scan_primary_hash
;
302 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
, sk
);
303 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
304 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
305 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
314 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
, 0))
318 inet_sk(sk
)->inet_num
= snum
;
319 udp_sk(sk
)->udp_port_hash
= snum
;
320 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
321 if (sk_unhashed(sk
)) {
322 if (sk
->sk_reuseport
&&
323 udp_reuseport_add_sock(sk
, hslot
)) {
324 inet_sk(sk
)->inet_num
= 0;
325 udp_sk(sk
)->udp_port_hash
= 0;
326 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
330 sk_add_node_rcu(sk
, &hslot
->head
);
332 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
334 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
335 spin_lock(&hslot2
->lock
);
336 if (IS_ENABLED(CONFIG_IPV6
) && sk
->sk_reuseport
&&
337 sk
->sk_family
== AF_INET6
)
338 hlist_add_tail_rcu(&udp_sk(sk
)->udp_portaddr_node
,
341 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
344 spin_unlock(&hslot2
->lock
);
346 sock_set_flag(sk
, SOCK_RCU_FREE
);
349 spin_unlock_bh(&hslot
->lock
);
353 EXPORT_SYMBOL(udp_lib_get_port
);
355 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
357 unsigned int hash2_nulladdr
=
358 ipv4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
359 unsigned int hash2_partial
=
360 ipv4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
362 /* precompute partial secondary hash */
363 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
364 return udp_lib_get_port(sk
, snum
, hash2_nulladdr
);
367 static int compute_score(struct sock
*sk
, struct net
*net
,
368 __be32 saddr
, __be16 sport
,
369 __be32 daddr
, unsigned short hnum
,
370 int dif
, int sdif
, bool exact_dif
)
373 struct inet_sock
*inet
;
375 if (!net_eq(sock_net(sk
), net
) ||
376 udp_sk(sk
)->udp_port_hash
!= hnum
||
380 score
= (sk
->sk_family
== PF_INET
) ? 2 : 1;
383 if (inet
->inet_rcv_saddr
) {
384 if (inet
->inet_rcv_saddr
!= daddr
)
389 if (inet
->inet_daddr
) {
390 if (inet
->inet_daddr
!= saddr
)
395 if (inet
->inet_dport
) {
396 if (inet
->inet_dport
!= sport
)
401 if (sk
->sk_bound_dev_if
|| exact_dif
) {
402 bool dev_match
= (sk
->sk_bound_dev_if
== dif
||
403 sk
->sk_bound_dev_if
== sdif
);
407 if (sk
->sk_bound_dev_if
)
411 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
416 static u32
udp_ehashfn(const struct net
*net
, const __be32 laddr
,
417 const __u16 lport
, const __be32 faddr
,
420 static u32 udp_ehash_secret __read_mostly
;
422 net_get_random_once(&udp_ehash_secret
, sizeof(udp_ehash_secret
));
424 return __inet_ehashfn(laddr
, lport
, faddr
, fport
,
425 udp_ehash_secret
+ net_hash_mix(net
));
428 /* called with rcu_read_lock() */
429 static struct sock
*udp4_lib_lookup2(struct net
*net
,
430 __be32 saddr
, __be16 sport
,
431 __be32 daddr
, unsigned int hnum
,
432 int dif
, int sdif
, bool exact_dif
,
433 struct udp_hslot
*hslot2
,
436 struct sock
*sk
, *result
;
442 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
443 score
= compute_score(sk
, net
, saddr
, sport
,
444 daddr
, hnum
, dif
, sdif
, exact_dif
);
445 if (score
> badness
) {
446 if (sk
->sk_reuseport
) {
447 hash
= udp_ehashfn(net
, daddr
, hnum
,
449 result
= reuseport_select_sock(sk
, hash
, skb
,
450 sizeof(struct udphdr
));
461 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
462 * harder than this. -DaveM
464 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
465 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
,
466 int sdif
, struct udp_table
*udptable
, struct sk_buff
*skb
)
468 struct sock
*sk
, *result
;
469 unsigned short hnum
= ntohs(dport
);
470 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
471 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
472 bool exact_dif
= udp_lib_exact_dif_match(net
, skb
);
476 if (hslot
->count
> 10) {
477 hash2
= ipv4_portaddr_hash(net
, daddr
, hnum
);
478 slot2
= hash2
& udptable
->mask
;
479 hslot2
= &udptable
->hash2
[slot2
];
480 if (hslot
->count
< hslot2
->count
)
483 result
= udp4_lib_lookup2(net
, saddr
, sport
,
484 daddr
, hnum
, dif
, sdif
,
485 exact_dif
, hslot2
, skb
);
487 unsigned int old_slot2
= slot2
;
488 hash2
= ipv4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
489 slot2
= hash2
& udptable
->mask
;
490 /* avoid searching the same slot again. */
491 if (unlikely(slot2
== old_slot2
))
494 hslot2
= &udptable
->hash2
[slot2
];
495 if (hslot
->count
< hslot2
->count
)
498 result
= udp4_lib_lookup2(net
, saddr
, sport
,
499 daddr
, hnum
, dif
, sdif
,
500 exact_dif
, hslot2
, skb
);
502 if (unlikely(IS_ERR(result
)))
509 sk_for_each_rcu(sk
, &hslot
->head
) {
510 score
= compute_score(sk
, net
, saddr
, sport
,
511 daddr
, hnum
, dif
, sdif
, exact_dif
);
512 if (score
> badness
) {
513 if (sk
->sk_reuseport
) {
514 hash
= udp_ehashfn(net
, daddr
, hnum
,
516 result
= reuseport_select_sock(sk
, hash
, skb
,
517 sizeof(struct udphdr
));
518 if (unlikely(IS_ERR(result
)))
529 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
531 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
532 __be16 sport
, __be16 dport
,
533 struct udp_table
*udptable
)
535 const struct iphdr
*iph
= ip_hdr(skb
);
537 return __udp4_lib_lookup(dev_net(skb
->dev
), iph
->saddr
, sport
,
538 iph
->daddr
, dport
, inet_iif(skb
),
539 inet_sdif(skb
), udptable
, skb
);
542 struct sock
*udp4_lib_lookup_skb(struct sk_buff
*skb
,
543 __be16 sport
, __be16 dport
)
545 return __udp4_lib_lookup_skb(skb
, sport
, dport
, &udp_table
);
547 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb
);
549 /* Must be called under rcu_read_lock().
550 * Does increment socket refcount.
552 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
553 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
554 __be32 daddr
, __be16 dport
, int dif
)
558 sk
= __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
559 dif
, 0, &udp_table
, NULL
);
560 if (sk
&& !refcount_inc_not_zero(&sk
->sk_refcnt
))
564 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
567 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
568 __be16 loc_port
, __be32 loc_addr
,
569 __be16 rmt_port
, __be32 rmt_addr
,
570 int dif
, int sdif
, unsigned short hnum
)
572 struct inet_sock
*inet
= inet_sk(sk
);
574 if (!net_eq(sock_net(sk
), net
) ||
575 udp_sk(sk
)->udp_port_hash
!= hnum
||
576 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
577 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
578 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
579 ipv6_only_sock(sk
) ||
580 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
&&
581 sk
->sk_bound_dev_if
!= sdif
))
583 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
, sdif
))
589 * This routine is called by the ICMP module when it gets some
590 * sort of error condition. If err < 0 then the socket should
591 * be closed and the error returned to the user. If err > 0
592 * it's just the icmp type << 8 | icmp code.
593 * Header points to the ip header of the error packet. We move
594 * on past this. Then (as it used to claim before adjustment)
595 * header points to the first 8 bytes of the udp header. We need
596 * to find the appropriate port.
599 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
601 struct inet_sock
*inet
;
602 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
603 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
604 const int type
= icmp_hdr(skb
)->type
;
605 const int code
= icmp_hdr(skb
)->code
;
609 struct net
*net
= dev_net(skb
->dev
);
611 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
612 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, 0,
615 __ICMP_INC_STATS(net
, ICMP_MIB_INERRORS
);
616 return; /* No socket for error */
625 case ICMP_TIME_EXCEEDED
:
628 case ICMP_SOURCE_QUENCH
:
630 case ICMP_PARAMETERPROB
:
634 case ICMP_DEST_UNREACH
:
635 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
636 ipv4_sk_update_pmtu(skb
, sk
, info
);
637 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
645 if (code
<= NR_ICMP_UNREACH
) {
646 harderr
= icmp_err_convert
[code
].fatal
;
647 err
= icmp_err_convert
[code
].errno
;
651 ipv4_sk_redirect(skb
, sk
);
656 * RFC1122: OK. Passes ICMP errors back to application, as per
659 if (!inet
->recverr
) {
660 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
663 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
666 sk
->sk_error_report(sk
);
671 void udp_err(struct sk_buff
*skb
, u32 info
)
673 __udp4_lib_err(skb
, info
, &udp_table
);
677 * Throw away all pending data and cancel the corking. Socket is locked.
679 void udp_flush_pending_frames(struct sock
*sk
)
681 struct udp_sock
*up
= udp_sk(sk
);
686 ip_flush_pending_frames(sk
);
689 EXPORT_SYMBOL(udp_flush_pending_frames
);
692 * udp4_hwcsum - handle outgoing HW checksumming
693 * @skb: sk_buff containing the filled-in UDP header
694 * (checksum field must be zeroed out)
695 * @src: source IP address
696 * @dst: destination IP address
698 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
700 struct udphdr
*uh
= udp_hdr(skb
);
701 int offset
= skb_transport_offset(skb
);
702 int len
= skb
->len
- offset
;
706 if (!skb_has_frag_list(skb
)) {
708 * Only one fragment on the socket.
710 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
711 skb
->csum_offset
= offsetof(struct udphdr
, check
);
712 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
715 struct sk_buff
*frags
;
718 * HW-checksum won't work as there are two or more
719 * fragments on the socket so that all csums of sk_buffs
722 skb_walk_frags(skb
, frags
) {
723 csum
= csum_add(csum
, frags
->csum
);
727 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
728 skb
->ip_summed
= CHECKSUM_NONE
;
730 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
732 uh
->check
= CSUM_MANGLED_0
;
735 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
737 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
738 * for the simple case like when setting the checksum for a UDP tunnel.
740 void udp_set_csum(bool nocheck
, struct sk_buff
*skb
,
741 __be32 saddr
, __be32 daddr
, int len
)
743 struct udphdr
*uh
= udp_hdr(skb
);
747 } else if (skb_is_gso(skb
)) {
748 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
749 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
751 uh
->check
= udp_v4_check(len
, saddr
, daddr
, lco_csum(skb
));
753 uh
->check
= CSUM_MANGLED_0
;
755 skb
->ip_summed
= CHECKSUM_PARTIAL
;
756 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
757 skb
->csum_offset
= offsetof(struct udphdr
, check
);
758 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
761 EXPORT_SYMBOL(udp_set_csum
);
763 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
,
764 struct inet_cork
*cork
)
766 struct sock
*sk
= skb
->sk
;
767 struct inet_sock
*inet
= inet_sk(sk
);
770 int is_udplite
= IS_UDPLITE(sk
);
771 int offset
= skb_transport_offset(skb
);
772 int len
= skb
->len
- offset
;
776 * Create a UDP header
779 uh
->source
= inet
->inet_sport
;
780 uh
->dest
= fl4
->fl4_dport
;
781 uh
->len
= htons(len
);
784 if (cork
->gso_size
) {
785 const int hlen
= skb_network_header_len(skb
) +
786 sizeof(struct udphdr
);
788 if (hlen
+ cork
->gso_size
> cork
->fragsize
)
790 if (skb
->len
> cork
->gso_size
* UDP_MAX_SEGMENTS
)
792 if (sk
->sk_no_check_tx
)
794 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
|| is_udplite
||
795 dst_xfrm(skb_dst(skb
)))
798 skb_shinfo(skb
)->gso_size
= cork
->gso_size
;
799 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP_L4
;
800 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(len
- sizeof(uh
),
805 if (is_udplite
) /* UDP-Lite */
806 csum
= udplite_csum(skb
);
808 else if (sk
->sk_no_check_tx
) { /* UDP csum off */
810 skb
->ip_summed
= CHECKSUM_NONE
;
813 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
816 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
820 csum
= udp_csum(skb
);
822 /* add protocol-dependent pseudo-header */
823 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
824 sk
->sk_protocol
, csum
);
826 uh
->check
= CSUM_MANGLED_0
;
829 err
= ip_send_skb(sock_net(sk
), skb
);
831 if (err
== -ENOBUFS
&& !inet
->recverr
) {
832 UDP_INC_STATS(sock_net(sk
),
833 UDP_MIB_SNDBUFERRORS
, is_udplite
);
837 UDP_INC_STATS(sock_net(sk
),
838 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
843 * Push out all pending data as one UDP datagram. Socket is locked.
845 int udp_push_pending_frames(struct sock
*sk
)
847 struct udp_sock
*up
= udp_sk(sk
);
848 struct inet_sock
*inet
= inet_sk(sk
);
849 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
853 skb
= ip_finish_skb(sk
, fl4
);
857 err
= udp_send_skb(skb
, fl4
, &inet
->cork
.base
);
864 EXPORT_SYMBOL(udp_push_pending_frames
);
866 static int __udp_cmsg_send(struct cmsghdr
*cmsg
, u16
*gso_size
)
868 switch (cmsg
->cmsg_type
) {
870 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(__u16
)))
872 *gso_size
= *(__u16
*)CMSG_DATA(cmsg
);
879 int udp_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, u16
*gso_size
)
881 struct cmsghdr
*cmsg
;
882 bool need_ip
= false;
885 for_each_cmsghdr(cmsg
, msg
) {
886 if (!CMSG_OK(msg
, cmsg
))
889 if (cmsg
->cmsg_level
!= SOL_UDP
) {
894 err
= __udp_cmsg_send(cmsg
, gso_size
);
901 EXPORT_SYMBOL_GPL(udp_cmsg_send
);
903 int udp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
905 struct inet_sock
*inet
= inet_sk(sk
);
906 struct udp_sock
*up
= udp_sk(sk
);
907 DECLARE_SOCKADDR(struct sockaddr_in
*, usin
, msg
->msg_name
);
908 struct flowi4 fl4_stack
;
911 struct ipcm_cookie ipc
;
912 struct rtable
*rt
= NULL
;
915 __be32 daddr
, faddr
, saddr
;
918 int err
, is_udplite
= IS_UDPLITE(sk
);
919 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
920 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
922 struct ip_options_data opt_copy
;
931 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
934 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
936 fl4
= &inet
->cork
.fl
.u
.ip4
;
939 * There are pending frames.
940 * The socket lock must be held while it's corked.
943 if (likely(up
->pending
)) {
944 if (unlikely(up
->pending
!= AF_INET
)) {
952 ulen
+= sizeof(struct udphdr
);
955 * Get and verify the address.
958 if (msg
->msg_namelen
< sizeof(*usin
))
960 if (usin
->sin_family
!= AF_INET
) {
961 if (usin
->sin_family
!= AF_UNSPEC
)
962 return -EAFNOSUPPORT
;
965 daddr
= usin
->sin_addr
.s_addr
;
966 dport
= usin
->sin_port
;
970 if (sk
->sk_state
!= TCP_ESTABLISHED
)
971 return -EDESTADDRREQ
;
972 daddr
= inet
->inet_daddr
;
973 dport
= inet
->inet_dport
;
974 /* Open fast path for connected socket.
975 Route will not be used, if at least one option is set.
980 ipcm_init_sk(&ipc
, inet
);
981 ipc
.gso_size
= up
->gso_size
;
983 if (msg
->msg_controllen
) {
984 err
= udp_cmsg_send(sk
, msg
, &ipc
.gso_size
);
986 err
= ip_cmsg_send(sk
, msg
, &ipc
,
987 sk
->sk_family
== AF_INET6
);
988 if (unlikely(err
< 0)) {
997 struct ip_options_rcu
*inet_opt
;
1000 inet_opt
= rcu_dereference(inet
->inet_opt
);
1002 memcpy(&opt_copy
, inet_opt
,
1003 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
1004 ipc
.opt
= &opt_copy
.opt
;
1009 if (cgroup_bpf_enabled
&& !connected
) {
1010 err
= BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk
,
1011 (struct sockaddr
*)usin
, &ipc
.addr
);
1015 if (usin
->sin_port
== 0) {
1016 /* BPF program set invalid port. Reject it. */
1020 daddr
= usin
->sin_addr
.s_addr
;
1021 dport
= usin
->sin_port
;
1026 ipc
.addr
= faddr
= daddr
;
1028 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
1033 faddr
= ipc
.opt
->opt
.faddr
;
1036 tos
= get_rttos(&ipc
, inet
);
1037 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
1038 (msg
->msg_flags
& MSG_DONTROUTE
) ||
1039 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
1044 if (ipv4_is_multicast(daddr
)) {
1046 ipc
.oif
= inet
->mc_index
;
1048 saddr
= inet
->mc_addr
;
1050 } else if (!ipc
.oif
) {
1051 ipc
.oif
= inet
->uc_index
;
1052 } else if (ipv4_is_lbcast(daddr
) && inet
->uc_index
) {
1053 /* oif is set, packet is to local broadcast and
1054 * and uc_index is set. oif is most likely set
1055 * by sk_bound_dev_if. If uc_index != oif check if the
1056 * oif is an L3 master and uc_index is an L3 slave.
1057 * If so, we want to allow the send using the uc_index.
1059 if (ipc
.oif
!= inet
->uc_index
&&
1060 ipc
.oif
== l3mdev_master_ifindex_by_index(sock_net(sk
),
1062 ipc
.oif
= inet
->uc_index
;
1067 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
1070 struct net
*net
= sock_net(sk
);
1071 __u8 flow_flags
= inet_sk_flowi_flags(sk
);
1075 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
1076 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1078 faddr
, saddr
, dport
, inet
->inet_sport
,
1081 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
1082 rt
= ip_route_output_flow(net
, fl4
, sk
);
1086 if (err
== -ENETUNREACH
)
1087 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
1092 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1093 !sock_flag(sk
, SOCK_BROADCAST
))
1096 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1099 if (msg
->msg_flags
&MSG_CONFIRM
)
1105 daddr
= ipc
.addr
= fl4
->daddr
;
1107 /* Lockless fast path for the non-corking case. */
1109 struct inet_cork cork
;
1111 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
, ulen
,
1112 sizeof(struct udphdr
), &ipc
, &rt
,
1113 &cork
, msg
->msg_flags
);
1115 if (!IS_ERR_OR_NULL(skb
))
1116 err
= udp_send_skb(skb
, fl4
, &cork
);
1121 if (unlikely(up
->pending
)) {
1122 /* The socket is already corked while preparing it. */
1123 /* ... which is an evident application bug. --ANK */
1126 net_dbg_ratelimited("socket already corked\n");
1131 * Now cork the socket to pend data.
1133 fl4
= &inet
->cork
.fl
.u
.ip4
;
1136 fl4
->fl4_dport
= dport
;
1137 fl4
->fl4_sport
= inet
->inet_sport
;
1138 up
->pending
= AF_INET
;
1142 err
= ip_append_data(sk
, fl4
, getfrag
, msg
, ulen
,
1143 sizeof(struct udphdr
), &ipc
, &rt
,
1144 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1146 udp_flush_pending_frames(sk
);
1148 err
= udp_push_pending_frames(sk
);
1149 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1161 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1162 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1163 * we don't have a good statistic (IpOutDiscards but it can be too many
1164 * things). We could add another new stat but at least for now that
1165 * seems like overkill.
1167 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1168 UDP_INC_STATS(sock_net(sk
),
1169 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1174 if (msg
->msg_flags
& MSG_PROBE
)
1175 dst_confirm_neigh(&rt
->dst
, &fl4
->daddr
);
1176 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1177 goto back_from_confirm
;
1181 EXPORT_SYMBOL(udp_sendmsg
);
1183 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1184 size_t size
, int flags
)
1186 struct inet_sock
*inet
= inet_sk(sk
);
1187 struct udp_sock
*up
= udp_sk(sk
);
1190 if (flags
& MSG_SENDPAGE_NOTLAST
)
1194 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1196 /* Call udp_sendmsg to specify destination address which
1197 * sendpage interface can't pass.
1198 * This will succeed only when the socket is connected.
1200 ret
= udp_sendmsg(sk
, &msg
, 0);
1207 if (unlikely(!up
->pending
)) {
1210 net_dbg_ratelimited("cork failed\n");
1214 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1215 page
, offset
, size
, flags
);
1216 if (ret
== -EOPNOTSUPP
) {
1218 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1222 udp_flush_pending_frames(sk
);
1227 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1228 ret
= udp_push_pending_frames(sk
);
1236 #define UDP_SKB_IS_STATELESS 0x80000000
1238 static void udp_set_dev_scratch(struct sk_buff
*skb
)
1240 struct udp_dev_scratch
*scratch
= udp_skb_scratch(skb
);
1242 BUILD_BUG_ON(sizeof(struct udp_dev_scratch
) > sizeof(long));
1243 scratch
->_tsize_state
= skb
->truesize
;
1244 #if BITS_PER_LONG == 64
1245 scratch
->len
= skb
->len
;
1246 scratch
->csum_unnecessary
= !!skb_csum_unnecessary(skb
);
1247 scratch
->is_linear
= !skb_is_nonlinear(skb
);
1249 /* all head states execept sp (dst, sk, nf) are always cleared by
1250 * udp_rcv() and we need to preserve secpath, if present, to eventually
1251 * process IP_CMSG_PASSSEC at recvmsg() time
1253 if (likely(!skb_sec_path(skb
)))
1254 scratch
->_tsize_state
|= UDP_SKB_IS_STATELESS
;
1257 static int udp_skb_truesize(struct sk_buff
*skb
)
1259 return udp_skb_scratch(skb
)->_tsize_state
& ~UDP_SKB_IS_STATELESS
;
1262 static bool udp_skb_has_head_state(struct sk_buff
*skb
)
1264 return !(udp_skb_scratch(skb
)->_tsize_state
& UDP_SKB_IS_STATELESS
);
1267 /* fully reclaim rmem/fwd memory allocated for skb */
1268 static void udp_rmem_release(struct sock
*sk
, int size
, int partial
,
1269 bool rx_queue_lock_held
)
1271 struct udp_sock
*up
= udp_sk(sk
);
1272 struct sk_buff_head
*sk_queue
;
1275 if (likely(partial
)) {
1276 up
->forward_deficit
+= size
;
1277 size
= up
->forward_deficit
;
1278 if (size
< (sk
->sk_rcvbuf
>> 2))
1281 size
+= up
->forward_deficit
;
1283 up
->forward_deficit
= 0;
1285 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1286 * if the called don't held it already
1288 sk_queue
= &sk
->sk_receive_queue
;
1289 if (!rx_queue_lock_held
)
1290 spin_lock(&sk_queue
->lock
);
1293 sk
->sk_forward_alloc
+= size
;
1294 amt
= (sk
->sk_forward_alloc
- partial
) & ~(SK_MEM_QUANTUM
- 1);
1295 sk
->sk_forward_alloc
-= amt
;
1298 __sk_mem_reduce_allocated(sk
, amt
>> SK_MEM_QUANTUM_SHIFT
);
1300 atomic_sub(size
, &sk
->sk_rmem_alloc
);
1302 /* this can save us from acquiring the rx queue lock on next receive */
1303 skb_queue_splice_tail_init(sk_queue
, &up
->reader_queue
);
1305 if (!rx_queue_lock_held
)
1306 spin_unlock(&sk_queue
->lock
);
1309 /* Note: called with reader_queue.lock held.
1310 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1311 * This avoids a cache line miss while receive_queue lock is held.
1312 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1314 void udp_skb_destructor(struct sock
*sk
, struct sk_buff
*skb
)
1316 prefetch(&skb
->data
);
1317 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, false);
1319 EXPORT_SYMBOL(udp_skb_destructor
);
1321 /* as above, but the caller held the rx queue lock, too */
1322 static void udp_skb_dtor_locked(struct sock
*sk
, struct sk_buff
*skb
)
1324 prefetch(&skb
->data
);
1325 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, true);
1328 /* Idea of busylocks is to let producers grab an extra spinlock
1329 * to relieve pressure on the receive_queue spinlock shared by consumer.
1330 * Under flood, this means that only one producer can be in line
1331 * trying to acquire the receive_queue spinlock.
1332 * These busylock can be allocated on a per cpu manner, instead of a
1333 * per socket one (that would consume a cache line per socket)
1335 static int udp_busylocks_log __read_mostly
;
1336 static spinlock_t
*udp_busylocks __read_mostly
;
1338 static spinlock_t
*busylock_acquire(void *ptr
)
1342 busy
= udp_busylocks
+ hash_ptr(ptr
, udp_busylocks_log
);
1347 static void busylock_release(spinlock_t
*busy
)
1353 int __udp_enqueue_schedule_skb(struct sock
*sk
, struct sk_buff
*skb
)
1355 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
1356 int rmem
, delta
, amt
, err
= -ENOMEM
;
1357 spinlock_t
*busy
= NULL
;
1360 /* try to avoid the costly atomic add/sub pair when the receive
1361 * queue is full; always allow at least a packet
1363 rmem
= atomic_read(&sk
->sk_rmem_alloc
);
1364 if (rmem
> sk
->sk_rcvbuf
)
1367 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1368 * having linear skbs :
1369 * - Reduce memory overhead and thus increase receive queue capacity
1370 * - Less cache line misses at copyout() time
1371 * - Less work at consume_skb() (less alien page frag freeing)
1373 if (rmem
> (sk
->sk_rcvbuf
>> 1)) {
1376 busy
= busylock_acquire(sk
);
1378 size
= skb
->truesize
;
1379 udp_set_dev_scratch(skb
);
1381 /* we drop only if the receive buf is full and the receive
1382 * queue contains some other skb
1384 rmem
= atomic_add_return(size
, &sk
->sk_rmem_alloc
);
1385 if (rmem
> (size
+ sk
->sk_rcvbuf
))
1388 spin_lock(&list
->lock
);
1389 if (size
>= sk
->sk_forward_alloc
) {
1390 amt
= sk_mem_pages(size
);
1391 delta
= amt
<< SK_MEM_QUANTUM_SHIFT
;
1392 if (!__sk_mem_raise_allocated(sk
, delta
, amt
, SK_MEM_RECV
)) {
1394 spin_unlock(&list
->lock
);
1398 sk
->sk_forward_alloc
+= delta
;
1401 sk
->sk_forward_alloc
-= size
;
1403 /* no need to setup a destructor, we will explicitly release the
1404 * forward allocated memory on dequeue
1406 sock_skb_set_dropcount(sk
, skb
);
1408 __skb_queue_tail(list
, skb
);
1409 spin_unlock(&list
->lock
);
1411 if (!sock_flag(sk
, SOCK_DEAD
))
1412 sk
->sk_data_ready(sk
);
1414 busylock_release(busy
);
1418 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1421 atomic_inc(&sk
->sk_drops
);
1422 busylock_release(busy
);
1425 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb
);
1427 void udp_destruct_sock(struct sock
*sk
)
1429 /* reclaim completely the forward allocated memory */
1430 struct udp_sock
*up
= udp_sk(sk
);
1431 unsigned int total
= 0;
1432 struct sk_buff
*skb
;
1434 skb_queue_splice_tail_init(&sk
->sk_receive_queue
, &up
->reader_queue
);
1435 while ((skb
= __skb_dequeue(&up
->reader_queue
)) != NULL
) {
1436 total
+= skb
->truesize
;
1439 udp_rmem_release(sk
, total
, 0, true);
1441 inet_sock_destruct(sk
);
1443 EXPORT_SYMBOL_GPL(udp_destruct_sock
);
1445 int udp_init_sock(struct sock
*sk
)
1447 skb_queue_head_init(&udp_sk(sk
)->reader_queue
);
1448 sk
->sk_destruct
= udp_destruct_sock
;
1451 EXPORT_SYMBOL_GPL(udp_init_sock
);
1453 void skb_consume_udp(struct sock
*sk
, struct sk_buff
*skb
, int len
)
1455 if (unlikely(READ_ONCE(sk
->sk_peek_off
) >= 0)) {
1456 bool slow
= lock_sock_fast(sk
);
1458 sk_peek_offset_bwd(sk
, len
);
1459 unlock_sock_fast(sk
, slow
);
1462 if (!skb_unref(skb
))
1465 /* In the more common cases we cleared the head states previously,
1466 * see __udp_queue_rcv_skb().
1468 if (unlikely(udp_skb_has_head_state(skb
)))
1469 skb_release_head_state(skb
);
1470 __consume_stateless_skb(skb
);
1472 EXPORT_SYMBOL_GPL(skb_consume_udp
);
1474 static struct sk_buff
*__first_packet_length(struct sock
*sk
,
1475 struct sk_buff_head
*rcvq
,
1478 struct sk_buff
*skb
;
1480 while ((skb
= skb_peek(rcvq
)) != NULL
) {
1481 if (udp_lib_checksum_complete(skb
)) {
1482 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1484 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
1486 atomic_inc(&sk
->sk_drops
);
1487 __skb_unlink(skb
, rcvq
);
1488 *total
+= skb
->truesize
;
1491 /* the csum related bits could be changed, refresh
1494 udp_set_dev_scratch(skb
);
1502 * first_packet_length - return length of first packet in receive queue
1505 * Drops all bad checksum frames, until a valid one is found.
1506 * Returns the length of found skb, or -1 if none is found.
1508 static int first_packet_length(struct sock
*sk
)
1510 struct sk_buff_head
*rcvq
= &udp_sk(sk
)->reader_queue
;
1511 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1512 struct sk_buff
*skb
;
1516 spin_lock_bh(&rcvq
->lock
);
1517 skb
= __first_packet_length(sk
, rcvq
, &total
);
1518 if (!skb
&& !skb_queue_empty(sk_queue
)) {
1519 spin_lock(&sk_queue
->lock
);
1520 skb_queue_splice_tail_init(sk_queue
, rcvq
);
1521 spin_unlock(&sk_queue
->lock
);
1523 skb
= __first_packet_length(sk
, rcvq
, &total
);
1525 res
= skb
? skb
->len
: -1;
1527 udp_rmem_release(sk
, total
, 1, false);
1528 spin_unlock_bh(&rcvq
->lock
);
1533 * IOCTL requests applicable to the UDP protocol
1536 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1541 int amount
= sk_wmem_alloc_get(sk
);
1543 return put_user(amount
, (int __user
*)arg
);
1548 int amount
= max_t(int, 0, first_packet_length(sk
));
1550 return put_user(amount
, (int __user
*)arg
);
1554 return -ENOIOCTLCMD
;
1559 EXPORT_SYMBOL(udp_ioctl
);
1561 struct sk_buff
*__skb_recv_udp(struct sock
*sk
, unsigned int flags
,
1562 int noblock
, int *peeked
, int *off
, int *err
)
1564 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1565 struct sk_buff_head
*queue
;
1566 struct sk_buff
*last
;
1570 queue
= &udp_sk(sk
)->reader_queue
;
1571 flags
|= noblock
? MSG_DONTWAIT
: 0;
1572 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
1574 struct sk_buff
*skb
;
1576 error
= sock_error(sk
);
1583 spin_lock_bh(&queue
->lock
);
1584 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1589 spin_unlock_bh(&queue
->lock
);
1593 if (skb_queue_empty(sk_queue
)) {
1594 spin_unlock_bh(&queue
->lock
);
1598 /* refill the reader queue and walk it again
1599 * keep both queues locked to avoid re-acquiring
1600 * the sk_receive_queue lock if fwd memory scheduling
1603 spin_lock(&sk_queue
->lock
);
1604 skb_queue_splice_tail_init(sk_queue
, queue
);
1606 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1607 udp_skb_dtor_locked
,
1610 spin_unlock(&sk_queue
->lock
);
1611 spin_unlock_bh(&queue
->lock
);
1616 if (!sk_can_busy_loop(sk
))
1619 sk_busy_loop(sk
, flags
& MSG_DONTWAIT
);
1620 } while (!skb_queue_empty(sk_queue
));
1622 /* sk_queue is empty, reader_queue may contain peeked packets */
1624 !__skb_wait_for_more_packets(sk
, &error
, &timeo
,
1625 (struct sk_buff
*)sk_queue
));
1630 EXPORT_SYMBOL_GPL(__skb_recv_udp
);
1633 * This should be easy, if there is something there we
1634 * return it, otherwise we block.
1637 int udp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int noblock
,
1638 int flags
, int *addr_len
)
1640 struct inet_sock
*inet
= inet_sk(sk
);
1641 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, msg
->msg_name
);
1642 struct sk_buff
*skb
;
1643 unsigned int ulen
, copied
;
1644 int peeked
, peeking
, off
;
1646 int is_udplite
= IS_UDPLITE(sk
);
1647 bool checksum_valid
= false;
1649 if (flags
& MSG_ERRQUEUE
)
1650 return ip_recv_error(sk
, msg
, len
, addr_len
);
1653 peeking
= flags
& MSG_PEEK
;
1654 off
= sk_peek_offset(sk
, flags
);
1655 skb
= __skb_recv_udp(sk
, flags
, noblock
, &peeked
, &off
, &err
);
1659 ulen
= udp_skb_len(skb
);
1661 if (copied
> ulen
- off
)
1662 copied
= ulen
- off
;
1663 else if (copied
< ulen
)
1664 msg
->msg_flags
|= MSG_TRUNC
;
1667 * If checksum is needed at all, try to do it while copying the
1668 * data. If the data is truncated, or if we only want a partial
1669 * coverage checksum (UDP-Lite), do it before the copy.
1672 if (copied
< ulen
|| peeking
||
1673 (is_udplite
&& UDP_SKB_CB(skb
)->partial_cov
)) {
1674 checksum_valid
= udp_skb_csum_unnecessary(skb
) ||
1675 !__udp_lib_checksum_complete(skb
);
1676 if (!checksum_valid
)
1680 if (checksum_valid
|| udp_skb_csum_unnecessary(skb
)) {
1681 if (udp_skb_is_linear(skb
))
1682 err
= copy_linear_skb(skb
, copied
, off
, &msg
->msg_iter
);
1684 err
= skb_copy_datagram_msg(skb
, off
, msg
, copied
);
1686 err
= skb_copy_and_csum_datagram_msg(skb
, off
, msg
);
1692 if (unlikely(err
)) {
1694 atomic_inc(&sk
->sk_drops
);
1695 UDP_INC_STATS(sock_net(sk
),
1696 UDP_MIB_INERRORS
, is_udplite
);
1703 UDP_INC_STATS(sock_net(sk
),
1704 UDP_MIB_INDATAGRAMS
, is_udplite
);
1706 sock_recv_ts_and_drops(msg
, sk
, skb
);
1708 /* Copy the address. */
1710 sin
->sin_family
= AF_INET
;
1711 sin
->sin_port
= udp_hdr(skb
)->source
;
1712 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1713 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1714 *addr_len
= sizeof(*sin
);
1716 if (inet
->cmsg_flags
)
1717 ip_cmsg_recv_offset(msg
, sk
, skb
, sizeof(struct udphdr
), off
);
1720 if (flags
& MSG_TRUNC
)
1723 skb_consume_udp(sk
, skb
, peeking
? -err
: err
);
1727 if (!__sk_queue_drop_skb(sk
, &udp_sk(sk
)->reader_queue
, skb
, flags
,
1728 udp_skb_destructor
)) {
1729 UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1730 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1734 /* starting over for a new packet, but check if we need to yield */
1736 msg
->msg_flags
&= ~MSG_TRUNC
;
1740 int udp_pre_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
1742 /* This check is replicated from __ip4_datagram_connect() and
1743 * intended to prevent BPF program called below from accessing bytes
1744 * that are out of the bound specified by user in addr_len.
1746 if (addr_len
< sizeof(struct sockaddr_in
))
1749 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk
, uaddr
);
1751 EXPORT_SYMBOL(udp_pre_connect
);
1753 int __udp_disconnect(struct sock
*sk
, int flags
)
1755 struct inet_sock
*inet
= inet_sk(sk
);
1757 * 1003.1g - break association.
1760 sk
->sk_state
= TCP_CLOSE
;
1761 inet
->inet_daddr
= 0;
1762 inet
->inet_dport
= 0;
1763 sock_rps_reset_rxhash(sk
);
1764 sk
->sk_bound_dev_if
= 0;
1765 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1766 inet_reset_saddr(sk
);
1768 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1769 sk
->sk_prot
->unhash(sk
);
1770 inet
->inet_sport
= 0;
1775 EXPORT_SYMBOL(__udp_disconnect
);
1777 int udp_disconnect(struct sock
*sk
, int flags
)
1780 __udp_disconnect(sk
, flags
);
1784 EXPORT_SYMBOL(udp_disconnect
);
1786 void udp_lib_unhash(struct sock
*sk
)
1788 if (sk_hashed(sk
)) {
1789 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1790 struct udp_hslot
*hslot
, *hslot2
;
1792 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1793 udp_sk(sk
)->udp_port_hash
);
1794 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1796 spin_lock_bh(&hslot
->lock
);
1797 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1798 reuseport_detach_sock(sk
);
1799 if (sk_del_node_init_rcu(sk
)) {
1801 inet_sk(sk
)->inet_num
= 0;
1802 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1804 spin_lock(&hslot2
->lock
);
1805 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1807 spin_unlock(&hslot2
->lock
);
1809 spin_unlock_bh(&hslot
->lock
);
1812 EXPORT_SYMBOL(udp_lib_unhash
);
1815 * inet_rcv_saddr was changed, we must rehash secondary hash
1817 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1819 if (sk_hashed(sk
)) {
1820 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1821 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1823 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1824 nhslot2
= udp_hashslot2(udptable
, newhash
);
1825 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1827 if (hslot2
!= nhslot2
||
1828 rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1829 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1830 udp_sk(sk
)->udp_port_hash
);
1831 /* we must lock primary chain too */
1832 spin_lock_bh(&hslot
->lock
);
1833 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1834 reuseport_detach_sock(sk
);
1836 if (hslot2
!= nhslot2
) {
1837 spin_lock(&hslot2
->lock
);
1838 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1840 spin_unlock(&hslot2
->lock
);
1842 spin_lock(&nhslot2
->lock
);
1843 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1846 spin_unlock(&nhslot2
->lock
);
1849 spin_unlock_bh(&hslot
->lock
);
1853 EXPORT_SYMBOL(udp_lib_rehash
);
1855 static void udp_v4_rehash(struct sock
*sk
)
1857 u16 new_hash
= ipv4_portaddr_hash(sock_net(sk
),
1858 inet_sk(sk
)->inet_rcv_saddr
,
1859 inet_sk(sk
)->inet_num
);
1860 udp_lib_rehash(sk
, new_hash
);
1863 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1867 if (inet_sk(sk
)->inet_daddr
) {
1868 sock_rps_save_rxhash(sk
, skb
);
1869 sk_mark_napi_id(sk
, skb
);
1870 sk_incoming_cpu_update(sk
);
1872 sk_mark_napi_id_once(sk
, skb
);
1875 rc
= __udp_enqueue_schedule_skb(sk
, skb
);
1877 int is_udplite
= IS_UDPLITE(sk
);
1879 /* Note that an ENOMEM error is charged twice */
1881 UDP_INC_STATS(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1883 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1885 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1892 static DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key
);
1893 void udp_encap_enable(void)
1895 static_branch_enable(&udp_encap_needed_key
);
1897 EXPORT_SYMBOL(udp_encap_enable
);
1902 * >0: "udp encap" protocol resubmission
1904 * Note that in the success and error cases, the skb is assumed to
1905 * have either been requeued or freed.
1907 static int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1909 struct udp_sock
*up
= udp_sk(sk
);
1910 int is_udplite
= IS_UDPLITE(sk
);
1913 * Charge it to the socket, dropping if the queue is full.
1915 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1919 if (static_branch_unlikely(&udp_encap_needed_key
) && up
->encap_type
) {
1920 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1923 * This is an encapsulation socket so pass the skb to
1924 * the socket's udp_encap_rcv() hook. Otherwise, just
1925 * fall through and pass this up the UDP socket.
1926 * up->encap_rcv() returns the following value:
1927 * =0 if skb was successfully passed to the encap
1928 * handler or was discarded by it.
1929 * >0 if skb should be passed on to UDP.
1930 * <0 if skb should be resubmitted as proto -N
1933 /* if we're overly short, let UDP handle it */
1934 encap_rcv
= READ_ONCE(up
->encap_rcv
);
1938 /* Verify checksum before giving to encap */
1939 if (udp_lib_checksum_complete(skb
))
1942 ret
= encap_rcv(sk
, skb
);
1944 __UDP_INC_STATS(sock_net(sk
),
1945 UDP_MIB_INDATAGRAMS
,
1951 /* FALLTHROUGH -- it's a UDP Packet */
1955 * UDP-Lite specific tests, ignored on UDP sockets
1957 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1960 * MIB statistics other than incrementing the error count are
1961 * disabled for the following two types of errors: these depend
1962 * on the application settings, not on the functioning of the
1963 * protocol stack as such.
1965 * RFC 3828 here recommends (sec 3.3): "There should also be a
1966 * way ... to ... at least let the receiving application block
1967 * delivery of packets with coverage values less than a value
1968 * provided by the application."
1970 if (up
->pcrlen
== 0) { /* full coverage was set */
1971 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1972 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1975 /* The next case involves violating the min. coverage requested
1976 * by the receiver. This is subtle: if receiver wants x and x is
1977 * greater than the buffersize/MTU then receiver will complain
1978 * that it wants x while sender emits packets of smaller size y.
1979 * Therefore the above ...()->partial_cov statement is essential.
1981 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1982 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1983 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1988 prefetch(&sk
->sk_rmem_alloc
);
1989 if (rcu_access_pointer(sk
->sk_filter
) &&
1990 udp_lib_checksum_complete(skb
))
1993 if (sk_filter_trim_cap(sk
, skb
, sizeof(struct udphdr
)))
1996 udp_csum_pull_header(skb
);
1998 ipv4_pktinfo_prepare(sk
, skb
);
1999 return __udp_queue_rcv_skb(sk
, skb
);
2002 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
2004 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
2005 atomic_inc(&sk
->sk_drops
);
2010 /* For TCP sockets, sk_rx_dst is protected by socket lock
2011 * For UDP, we use xchg() to guard against concurrent changes.
2013 bool udp_sk_rx_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
2015 struct dst_entry
*old
;
2017 if (dst_hold_safe(dst
)) {
2018 old
= xchg(&sk
->sk_rx_dst
, dst
);
2024 EXPORT_SYMBOL(udp_sk_rx_dst_set
);
2027 * Multicasts and broadcasts go to each listener.
2029 * Note: called only from the BH handler context.
2031 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
2033 __be32 saddr
, __be32 daddr
,
2034 struct udp_table
*udptable
,
2037 struct sock
*sk
, *first
= NULL
;
2038 unsigned short hnum
= ntohs(uh
->dest
);
2039 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
2040 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
2041 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
2042 int dif
= skb
->dev
->ifindex
;
2043 int sdif
= inet_sdif(skb
);
2044 struct hlist_node
*node
;
2045 struct sk_buff
*nskb
;
2048 hash2_any
= ipv4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
) &
2050 hash2
= ipv4_portaddr_hash(net
, daddr
, hnum
) & udptable
->mask
;
2052 hslot
= &udptable
->hash2
[hash2
];
2053 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
2056 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
2057 if (!__udp_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
2058 uh
->source
, saddr
, dif
, sdif
, hnum
))
2065 nskb
= skb_clone(skb
, GFP_ATOMIC
);
2067 if (unlikely(!nskb
)) {
2068 atomic_inc(&sk
->sk_drops
);
2069 __UDP_INC_STATS(net
, UDP_MIB_RCVBUFERRORS
,
2071 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
,
2075 if (udp_queue_rcv_skb(sk
, nskb
) > 0)
2079 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2080 if (use_hash2
&& hash2
!= hash2_any
) {
2086 if (udp_queue_rcv_skb(first
, skb
) > 0)
2090 __UDP_INC_STATS(net
, UDP_MIB_IGNOREDMULTI
,
2091 proto
== IPPROTO_UDPLITE
);
2096 /* Initialize UDP checksum. If exited with zero value (success),
2097 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2098 * Otherwise, csum completion requires chacksumming packet body,
2099 * including udp header and folding it to skb->csum.
2101 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
2106 UDP_SKB_CB(skb
)->partial_cov
= 0;
2107 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
2109 if (proto
== IPPROTO_UDPLITE
) {
2110 err
= udplite_checksum_init(skb
, uh
);
2114 if (UDP_SKB_CB(skb
)->partial_cov
) {
2115 skb
->csum
= inet_compute_pseudo(skb
, proto
);
2120 /* Note, we are only interested in != 0 or == 0, thus the
2123 return (__force
int)skb_checksum_init_zero_check(skb
, proto
, uh
->check
,
2124 inet_compute_pseudo
);
2128 * All we need to do is get the socket, and then do a checksum.
2131 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
2136 unsigned short ulen
;
2137 struct rtable
*rt
= skb_rtable(skb
);
2138 __be32 saddr
, daddr
;
2139 struct net
*net
= dev_net(skb
->dev
);
2142 * Validate the packet.
2144 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
2145 goto drop
; /* No space for header. */
2148 ulen
= ntohs(uh
->len
);
2149 saddr
= ip_hdr(skb
)->saddr
;
2150 daddr
= ip_hdr(skb
)->daddr
;
2152 if (ulen
> skb
->len
)
2155 if (proto
== IPPROTO_UDP
) {
2156 /* UDP validates ulen. */
2157 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
2162 if (udp4_csum_init(skb
, uh
, proto
))
2165 sk
= skb_steal_sock(skb
);
2167 struct dst_entry
*dst
= skb_dst(skb
);
2170 if (unlikely(sk
->sk_rx_dst
!= dst
))
2171 udp_sk_rx_dst_set(sk
, dst
);
2173 ret
= udp_queue_rcv_skb(sk
, skb
);
2175 /* a return value > 0 means to resubmit the input, but
2176 * it wants the return to be -protocol, or 0
2183 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
2184 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
2185 saddr
, daddr
, udptable
, proto
);
2187 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
2191 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
2192 skb_checksum_try_convert(skb
, IPPROTO_UDP
, uh
->check
,
2193 inet_compute_pseudo
);
2195 ret
= udp_queue_rcv_skb(sk
, skb
);
2197 /* a return value > 0 means to resubmit the input, but
2198 * it wants the return to be -protocol, or 0
2205 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2209 /* No socket. Drop packet silently, if checksum is wrong */
2210 if (udp_lib_checksum_complete(skb
))
2213 __UDP_INC_STATS(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
2214 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
2217 * Hmm. We got an UDP packet to a port to which we
2218 * don't wanna listen. Ignore it.
2224 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2225 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2226 &saddr
, ntohs(uh
->source
),
2228 &daddr
, ntohs(uh
->dest
));
2233 * RFC1122: OK. Discards the bad packet silently (as far as
2234 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2236 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2237 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2238 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
2240 __UDP_INC_STATS(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
2242 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
2247 /* We can only early demux multicast if there is a single matching socket.
2248 * If more than one socket found returns NULL
2250 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
2251 __be16 loc_port
, __be32 loc_addr
,
2252 __be16 rmt_port
, __be32 rmt_addr
,
2255 struct sock
*sk
, *result
;
2256 unsigned short hnum
= ntohs(loc_port
);
2257 unsigned int slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
2258 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
2260 /* Do not bother scanning a too big list */
2261 if (hslot
->count
> 10)
2265 sk_for_each_rcu(sk
, &hslot
->head
) {
2266 if (__udp_is_mcast_sock(net
, sk
, loc_port
, loc_addr
,
2267 rmt_port
, rmt_addr
, dif
, sdif
, hnum
)) {
2277 /* For unicast we should only early demux connected sockets or we can
2278 * break forwarding setups. The chains here can be long so only check
2279 * if the first socket is an exact match and if not move on.
2281 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
2282 __be16 loc_port
, __be32 loc_addr
,
2283 __be16 rmt_port
, __be32 rmt_addr
,
2286 unsigned short hnum
= ntohs(loc_port
);
2287 unsigned int hash2
= ipv4_portaddr_hash(net
, loc_addr
, hnum
);
2288 unsigned int slot2
= hash2
& udp_table
.mask
;
2289 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
2290 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
);
2291 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
2294 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
2295 if (INET_MATCH(sk
, net
, acookie
, rmt_addr
,
2296 loc_addr
, ports
, dif
, sdif
))
2298 /* Only check first socket in chain */
2304 int udp_v4_early_demux(struct sk_buff
*skb
)
2306 struct net
*net
= dev_net(skb
->dev
);
2307 struct in_device
*in_dev
= NULL
;
2308 const struct iphdr
*iph
;
2309 const struct udphdr
*uh
;
2310 struct sock
*sk
= NULL
;
2311 struct dst_entry
*dst
;
2312 int dif
= skb
->dev
->ifindex
;
2313 int sdif
= inet_sdif(skb
);
2316 /* validate the packet */
2317 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
2323 if (skb
->pkt_type
== PACKET_MULTICAST
) {
2324 in_dev
= __in_dev_get_rcu(skb
->dev
);
2329 ours
= ip_check_mc_rcu(in_dev
, iph
->daddr
, iph
->saddr
,
2334 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2335 uh
->source
, iph
->saddr
,
2337 } else if (skb
->pkt_type
== PACKET_HOST
) {
2338 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2339 uh
->source
, iph
->saddr
, dif
, sdif
);
2342 if (!sk
|| !refcount_inc_not_zero(&sk
->sk_refcnt
))
2346 skb
->destructor
= sock_efree
;
2347 dst
= READ_ONCE(sk
->sk_rx_dst
);
2350 dst
= dst_check(dst
, 0);
2354 /* set noref for now.
2355 * any place which wants to hold dst has to call
2358 skb_dst_set_noref(skb
, dst
);
2360 /* for unconnected multicast sockets we need to validate
2361 * the source on each packet
2363 if (!inet_sk(sk
)->inet_daddr
&& in_dev
)
2364 return ip_mc_validate_source(skb
, iph
->daddr
,
2365 iph
->saddr
, iph
->tos
,
2366 skb
->dev
, in_dev
, &itag
);
2371 int udp_rcv(struct sk_buff
*skb
)
2373 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
2376 void udp_destroy_sock(struct sock
*sk
)
2378 struct udp_sock
*up
= udp_sk(sk
);
2379 bool slow
= lock_sock_fast(sk
);
2380 udp_flush_pending_frames(sk
);
2381 unlock_sock_fast(sk
, slow
);
2382 if (static_branch_unlikely(&udp_encap_needed_key
) && up
->encap_type
) {
2383 void (*encap_destroy
)(struct sock
*sk
);
2384 encap_destroy
= READ_ONCE(up
->encap_destroy
);
2391 * Socket option code for UDP
2393 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
2394 char __user
*optval
, unsigned int optlen
,
2395 int (*push_pending_frames
)(struct sock
*))
2397 struct udp_sock
*up
= udp_sk(sk
);
2400 int is_udplite
= IS_UDPLITE(sk
);
2402 if (optlen
< sizeof(int))
2405 if (get_user(val
, (int __user
*)optval
))
2408 valbool
= val
? 1 : 0;
2417 push_pending_frames(sk
);
2425 case UDP_ENCAP_ESPINUDP
:
2426 case UDP_ENCAP_ESPINUDP_NON_IKE
:
2427 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
2429 case UDP_ENCAP_L2TPINUDP
:
2430 up
->encap_type
= val
;
2439 case UDP_NO_CHECK6_TX
:
2440 up
->no_check6_tx
= valbool
;
2443 case UDP_NO_CHECK6_RX
:
2444 up
->no_check6_rx
= valbool
;
2448 if (val
< 0 || val
> USHRT_MAX
)
2454 * UDP-Lite's partial checksum coverage (RFC 3828).
2456 /* The sender sets actual checksum coverage length via this option.
2457 * The case coverage > packet length is handled by send module. */
2458 case UDPLITE_SEND_CSCOV
:
2459 if (!is_udplite
) /* Disable the option on UDP sockets */
2460 return -ENOPROTOOPT
;
2461 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2463 else if (val
> USHRT_MAX
)
2466 up
->pcflag
|= UDPLITE_SEND_CC
;
2469 /* The receiver specifies a minimum checksum coverage value. To make
2470 * sense, this should be set to at least 8 (as done below). If zero is
2471 * used, this again means full checksum coverage. */
2472 case UDPLITE_RECV_CSCOV
:
2473 if (!is_udplite
) /* Disable the option on UDP sockets */
2474 return -ENOPROTOOPT
;
2475 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2477 else if (val
> USHRT_MAX
)
2480 up
->pcflag
|= UDPLITE_RECV_CC
;
2490 EXPORT_SYMBOL(udp_lib_setsockopt
);
2492 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2493 char __user
*optval
, unsigned int optlen
)
2495 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2496 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2497 udp_push_pending_frames
);
2498 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2501 #ifdef CONFIG_COMPAT
2502 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2503 char __user
*optval
, unsigned int optlen
)
2505 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2506 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2507 udp_push_pending_frames
);
2508 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2512 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2513 char __user
*optval
, int __user
*optlen
)
2515 struct udp_sock
*up
= udp_sk(sk
);
2518 if (get_user(len
, optlen
))
2521 len
= min_t(unsigned int, len
, sizeof(int));
2532 val
= up
->encap_type
;
2535 case UDP_NO_CHECK6_TX
:
2536 val
= up
->no_check6_tx
;
2539 case UDP_NO_CHECK6_RX
:
2540 val
= up
->no_check6_rx
;
2547 /* The following two cannot be changed on UDP sockets, the return is
2548 * always 0 (which corresponds to the full checksum coverage of UDP). */
2549 case UDPLITE_SEND_CSCOV
:
2553 case UDPLITE_RECV_CSCOV
:
2558 return -ENOPROTOOPT
;
2561 if (put_user(len
, optlen
))
2563 if (copy_to_user(optval
, &val
, len
))
2567 EXPORT_SYMBOL(udp_lib_getsockopt
);
2569 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2570 char __user
*optval
, int __user
*optlen
)
2572 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2573 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2574 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2577 #ifdef CONFIG_COMPAT
2578 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2579 char __user
*optval
, int __user
*optlen
)
2581 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2582 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2583 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2587 * udp_poll - wait for a UDP event.
2588 * @file - file struct
2590 * @wait - poll table
2592 * This is same as datagram poll, except for the special case of
2593 * blocking sockets. If application is using a blocking fd
2594 * and a packet with checksum error is in the queue;
2595 * then it could get return from select indicating data available
2596 * but then block when reading it. Add special case code
2597 * to work around these arguably broken applications.
2599 __poll_t
udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2601 __poll_t mask
= datagram_poll(file
, sock
, wait
);
2602 struct sock
*sk
= sock
->sk
;
2604 if (!skb_queue_empty(&udp_sk(sk
)->reader_queue
))
2605 mask
|= EPOLLIN
| EPOLLRDNORM
;
2607 /* Check for false positives due to checksum errors */
2608 if ((mask
& EPOLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2609 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && first_packet_length(sk
) == -1)
2610 mask
&= ~(EPOLLIN
| EPOLLRDNORM
);
2615 EXPORT_SYMBOL(udp_poll
);
2617 int udp_abort(struct sock
*sk
, int err
)
2622 sk
->sk_error_report(sk
);
2623 __udp_disconnect(sk
, 0);
2629 EXPORT_SYMBOL_GPL(udp_abort
);
2631 struct proto udp_prot
= {
2633 .owner
= THIS_MODULE
,
2634 .close
= udp_lib_close
,
2635 .pre_connect
= udp_pre_connect
,
2636 .connect
= ip4_datagram_connect
,
2637 .disconnect
= udp_disconnect
,
2639 .init
= udp_init_sock
,
2640 .destroy
= udp_destroy_sock
,
2641 .setsockopt
= udp_setsockopt
,
2642 .getsockopt
= udp_getsockopt
,
2643 .sendmsg
= udp_sendmsg
,
2644 .recvmsg
= udp_recvmsg
,
2645 .sendpage
= udp_sendpage
,
2646 .release_cb
= ip4_datagram_release_cb
,
2647 .hash
= udp_lib_hash
,
2648 .unhash
= udp_lib_unhash
,
2649 .rehash
= udp_v4_rehash
,
2650 .get_port
= udp_v4_get_port
,
2651 .memory_allocated
= &udp_memory_allocated
,
2652 .sysctl_mem
= sysctl_udp_mem
,
2653 .sysctl_wmem_offset
= offsetof(struct net
, ipv4
.sysctl_udp_wmem_min
),
2654 .sysctl_rmem_offset
= offsetof(struct net
, ipv4
.sysctl_udp_rmem_min
),
2655 .obj_size
= sizeof(struct udp_sock
),
2656 .h
.udp_table
= &udp_table
,
2657 #ifdef CONFIG_COMPAT
2658 .compat_setsockopt
= compat_udp_setsockopt
,
2659 .compat_getsockopt
= compat_udp_getsockopt
,
2661 .diag_destroy
= udp_abort
,
2663 EXPORT_SYMBOL(udp_prot
);
2665 /* ------------------------------------------------------------------------ */
2666 #ifdef CONFIG_PROC_FS
2668 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2671 struct udp_seq_afinfo
*afinfo
= PDE_DATA(file_inode(seq
->file
));
2672 struct udp_iter_state
*state
= seq
->private;
2673 struct net
*net
= seq_file_net(seq
);
2675 for (state
->bucket
= start
; state
->bucket
<= afinfo
->udp_table
->mask
;
2677 struct udp_hslot
*hslot
= &afinfo
->udp_table
->hash
[state
->bucket
];
2679 if (hlist_empty(&hslot
->head
))
2682 spin_lock_bh(&hslot
->lock
);
2683 sk_for_each(sk
, &hslot
->head
) {
2684 if (!net_eq(sock_net(sk
), net
))
2686 if (sk
->sk_family
== afinfo
->family
)
2689 spin_unlock_bh(&hslot
->lock
);
2696 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2698 struct udp_seq_afinfo
*afinfo
= PDE_DATA(file_inode(seq
->file
));
2699 struct udp_iter_state
*state
= seq
->private;
2700 struct net
*net
= seq_file_net(seq
);
2704 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= afinfo
->family
));
2707 if (state
->bucket
<= afinfo
->udp_table
->mask
)
2708 spin_unlock_bh(&afinfo
->udp_table
->hash
[state
->bucket
].lock
);
2709 return udp_get_first(seq
, state
->bucket
+ 1);
2714 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2716 struct sock
*sk
= udp_get_first(seq
, 0);
2719 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2721 return pos
? NULL
: sk
;
2724 void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2726 struct udp_iter_state
*state
= seq
->private;
2727 state
->bucket
= MAX_UDP_PORTS
;
2729 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2731 EXPORT_SYMBOL(udp_seq_start
);
2733 void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2737 if (v
== SEQ_START_TOKEN
)
2738 sk
= udp_get_idx(seq
, 0);
2740 sk
= udp_get_next(seq
, v
);
2745 EXPORT_SYMBOL(udp_seq_next
);
2747 void udp_seq_stop(struct seq_file
*seq
, void *v
)
2749 struct udp_seq_afinfo
*afinfo
= PDE_DATA(file_inode(seq
->file
));
2750 struct udp_iter_state
*state
= seq
->private;
2752 if (state
->bucket
<= afinfo
->udp_table
->mask
)
2753 spin_unlock_bh(&afinfo
->udp_table
->hash
[state
->bucket
].lock
);
2755 EXPORT_SYMBOL(udp_seq_stop
);
2757 /* ------------------------------------------------------------------------ */
2758 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2761 struct inet_sock
*inet
= inet_sk(sp
);
2762 __be32 dest
= inet
->inet_daddr
;
2763 __be32 src
= inet
->inet_rcv_saddr
;
2764 __u16 destp
= ntohs(inet
->inet_dport
);
2765 __u16 srcp
= ntohs(inet
->inet_sport
);
2767 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2768 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2769 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2770 sk_wmem_alloc_get(sp
),
2773 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2775 refcount_read(&sp
->sk_refcnt
), sp
,
2776 atomic_read(&sp
->sk_drops
));
2779 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2781 seq_setwidth(seq
, 127);
2782 if (v
== SEQ_START_TOKEN
)
2783 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2784 "rx_queue tr tm->when retrnsmt uid timeout "
2785 "inode ref pointer drops");
2787 struct udp_iter_state
*state
= seq
->private;
2789 udp4_format_sock(v
, seq
, state
->bucket
);
2795 const struct seq_operations udp_seq_ops
= {
2796 .start
= udp_seq_start
,
2797 .next
= udp_seq_next
,
2798 .stop
= udp_seq_stop
,
2799 .show
= udp4_seq_show
,
2801 EXPORT_SYMBOL(udp_seq_ops
);
2803 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2805 .udp_table
= &udp_table
,
2808 static int __net_init
udp4_proc_init_net(struct net
*net
)
2810 if (!proc_create_net_data("udp", 0444, net
->proc_net
, &udp_seq_ops
,
2811 sizeof(struct udp_iter_state
), &udp4_seq_afinfo
))
2816 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2818 remove_proc_entry("udp", net
->proc_net
);
2821 static struct pernet_operations udp4_net_ops
= {
2822 .init
= udp4_proc_init_net
,
2823 .exit
= udp4_proc_exit_net
,
2826 int __init
udp4_proc_init(void)
2828 return register_pernet_subsys(&udp4_net_ops
);
2831 void udp4_proc_exit(void)
2833 unregister_pernet_subsys(&udp4_net_ops
);
2835 #endif /* CONFIG_PROC_FS */
2837 static __initdata
unsigned long uhash_entries
;
2838 static int __init
set_uhash_entries(char *str
)
2845 ret
= kstrtoul(str
, 0, &uhash_entries
);
2849 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2850 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2853 __setup("uhash_entries=", set_uhash_entries
);
2855 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2859 table
->hash
= alloc_large_system_hash(name
,
2860 2 * sizeof(struct udp_hslot
),
2862 21, /* one slot per 2 MB */
2866 UDP_HTABLE_SIZE_MIN
,
2869 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2870 for (i
= 0; i
<= table
->mask
; i
++) {
2871 INIT_HLIST_HEAD(&table
->hash
[i
].head
);
2872 table
->hash
[i
].count
= 0;
2873 spin_lock_init(&table
->hash
[i
].lock
);
2875 for (i
= 0; i
<= table
->mask
; i
++) {
2876 INIT_HLIST_HEAD(&table
->hash2
[i
].head
);
2877 table
->hash2
[i
].count
= 0;
2878 spin_lock_init(&table
->hash2
[i
].lock
);
2882 u32
udp_flow_hashrnd(void)
2884 static u32 hashrnd __read_mostly
;
2886 net_get_random_once(&hashrnd
, sizeof(hashrnd
));
2890 EXPORT_SYMBOL(udp_flow_hashrnd
);
2892 static void __udp_sysctl_init(struct net
*net
)
2894 net
->ipv4
.sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2895 net
->ipv4
.sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2897 #ifdef CONFIG_NET_L3_MASTER_DEV
2898 net
->ipv4
.sysctl_udp_l3mdev_accept
= 0;
2902 static int __net_init
udp_sysctl_init(struct net
*net
)
2904 __udp_sysctl_init(net
);
2908 static struct pernet_operations __net_initdata udp_sysctl_ops
= {
2909 .init
= udp_sysctl_init
,
2912 void __init
udp_init(void)
2914 unsigned long limit
;
2917 udp_table_init(&udp_table
, "UDP");
2918 limit
= nr_free_buffer_pages() / 8;
2919 limit
= max(limit
, 128UL);
2920 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2921 sysctl_udp_mem
[1] = limit
;
2922 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2924 __udp_sysctl_init(&init_net
);
2926 /* 16 spinlocks per cpu */
2927 udp_busylocks_log
= ilog2(nr_cpu_ids
) + 4;
2928 udp_busylocks
= kmalloc(sizeof(spinlock_t
) << udp_busylocks_log
,
2931 panic("UDP: failed to alloc udp_busylocks\n");
2932 for (i
= 0; i
< (1U << udp_busylocks_log
); i
++)
2933 spin_lock_init(udp_busylocks
+ i
);
2935 if (register_pernet_subsys(&udp_sysctl_ops
))
2936 panic("UDP: failed to init sysctl parameters.\n");