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 int sysctl_udp_rmem_min __read_mostly
;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
128 int sysctl_udp_wmem_min __read_mostly
;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
131 atomic_long_t udp_memory_allocated
;
132 EXPORT_SYMBOL(udp_memory_allocated
);
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
137 /* IPCB reference means this can not be used from early demux */
138 static bool udp_lib_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
140 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
141 if (!net
->ipv4
.sysctl_udp_l3mdev_accept
&&
142 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
148 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
149 const struct udp_hslot
*hslot
,
150 unsigned long *bitmap
,
151 struct sock
*sk
, unsigned int log
)
154 kuid_t uid
= sock_i_uid(sk
);
156 sk_for_each(sk2
, &hslot
->head
) {
157 if (net_eq(sock_net(sk2
), net
) &&
159 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
160 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
161 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
162 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
163 inet_rcv_saddr_equal(sk
, sk2
, true)) {
164 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
165 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
166 uid_eq(uid
, sock_i_uid(sk2
))) {
172 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
181 * Note: we still hold spinlock of primary hash chain, so no other writer
182 * can insert/delete a socket with local_port == num
184 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
185 struct udp_hslot
*hslot2
,
189 kuid_t uid
= sock_i_uid(sk
);
192 spin_lock(&hslot2
->lock
);
193 udp_portaddr_for_each_entry(sk2
, &hslot2
->head
) {
194 if (net_eq(sock_net(sk2
), net
) &&
196 (udp_sk(sk2
)->udp_port_hash
== num
) &&
197 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
198 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
199 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
200 inet_rcv_saddr_equal(sk
, sk2
, true)) {
201 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
202 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
203 uid_eq(uid
, sock_i_uid(sk2
))) {
211 spin_unlock(&hslot2
->lock
);
215 static int udp_reuseport_add_sock(struct sock
*sk
, struct udp_hslot
*hslot
)
217 struct net
*net
= sock_net(sk
);
218 kuid_t uid
= sock_i_uid(sk
);
221 sk_for_each(sk2
, &hslot
->head
) {
222 if (net_eq(sock_net(sk2
), net
) &&
224 sk2
->sk_family
== sk
->sk_family
&&
225 ipv6_only_sock(sk2
) == ipv6_only_sock(sk
) &&
226 (udp_sk(sk2
)->udp_port_hash
== udp_sk(sk
)->udp_port_hash
) &&
227 (sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
228 sk2
->sk_reuseport
&& uid_eq(uid
, sock_i_uid(sk2
)) &&
229 inet_rcv_saddr_equal(sk
, sk2
, false)) {
230 return reuseport_add_sock(sk
, sk2
);
234 return reuseport_alloc(sk
);
238 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
240 * @sk: socket struct in question
241 * @snum: port number to look up
242 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
245 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
246 unsigned int hash2_nulladdr
)
248 struct udp_hslot
*hslot
, *hslot2
;
249 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
251 struct net
*net
= sock_net(sk
);
254 int low
, high
, remaining
;
256 unsigned short first
, last
;
257 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
259 inet_get_local_port_range(net
, &low
, &high
);
260 remaining
= (high
- low
) + 1;
262 rand
= prandom_u32();
263 first
= reciprocal_scale(rand
, remaining
) + low
;
265 * force rand to be an odd multiple of UDP_HTABLE_SIZE
267 rand
= (rand
| 1) * (udptable
->mask
+ 1);
268 last
= first
+ udptable
->mask
+ 1;
270 hslot
= udp_hashslot(udptable
, net
, first
);
271 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
272 spin_lock_bh(&hslot
->lock
);
273 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
278 * Iterate on all possible values of snum for this hash.
279 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 * give us randomization and full range coverage.
283 if (low
<= snum
&& snum
<= high
&&
284 !test_bit(snum
>> udptable
->log
, bitmap
) &&
285 !inet_is_local_reserved_port(net
, snum
))
288 } while (snum
!= first
);
289 spin_unlock_bh(&hslot
->lock
);
291 } while (++first
!= last
);
294 hslot
= udp_hashslot(udptable
, net
, snum
);
295 spin_lock_bh(&hslot
->lock
);
296 if (hslot
->count
> 10) {
298 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
300 slot2
&= udptable
->mask
;
301 hash2_nulladdr
&= udptable
->mask
;
303 hslot2
= udp_hashslot2(udptable
, slot2
);
304 if (hslot
->count
< hslot2
->count
)
305 goto scan_primary_hash
;
307 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
, sk
);
308 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
309 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
310 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
319 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
, 0))
323 inet_sk(sk
)->inet_num
= snum
;
324 udp_sk(sk
)->udp_port_hash
= snum
;
325 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
326 if (sk_unhashed(sk
)) {
327 if (sk
->sk_reuseport
&&
328 udp_reuseport_add_sock(sk
, hslot
)) {
329 inet_sk(sk
)->inet_num
= 0;
330 udp_sk(sk
)->udp_port_hash
= 0;
331 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
335 sk_add_node_rcu(sk
, &hslot
->head
);
337 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
339 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
340 spin_lock(&hslot2
->lock
);
341 if (IS_ENABLED(CONFIG_IPV6
) && sk
->sk_reuseport
&&
342 sk
->sk_family
== AF_INET6
)
343 hlist_add_tail_rcu(&udp_sk(sk
)->udp_portaddr_node
,
346 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
349 spin_unlock(&hslot2
->lock
);
351 sock_set_flag(sk
, SOCK_RCU_FREE
);
354 spin_unlock_bh(&hslot
->lock
);
358 EXPORT_SYMBOL(udp_lib_get_port
);
360 static u32
udp4_portaddr_hash(const struct net
*net
, __be32 saddr
,
363 return jhash_1word((__force u32
)saddr
, net_hash_mix(net
)) ^ port
;
366 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
368 unsigned int hash2_nulladdr
=
369 udp4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
370 unsigned int hash2_partial
=
371 udp4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
373 /* precompute partial secondary hash */
374 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
375 return udp_lib_get_port(sk
, snum
, hash2_nulladdr
);
378 static int compute_score(struct sock
*sk
, struct net
*net
,
379 __be32 saddr
, __be16 sport
,
380 __be32 daddr
, unsigned short hnum
, int dif
,
384 struct inet_sock
*inet
;
386 if (!net_eq(sock_net(sk
), net
) ||
387 udp_sk(sk
)->udp_port_hash
!= hnum
||
391 score
= (sk
->sk_family
== PF_INET
) ? 2 : 1;
394 if (inet
->inet_rcv_saddr
) {
395 if (inet
->inet_rcv_saddr
!= daddr
)
400 if (inet
->inet_daddr
) {
401 if (inet
->inet_daddr
!= saddr
)
406 if (inet
->inet_dport
) {
407 if (inet
->inet_dport
!= sport
)
412 if (sk
->sk_bound_dev_if
|| exact_dif
) {
413 if (sk
->sk_bound_dev_if
!= dif
)
417 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
422 static u32
udp_ehashfn(const struct net
*net
, const __be32 laddr
,
423 const __u16 lport
, const __be32 faddr
,
426 static u32 udp_ehash_secret __read_mostly
;
428 net_get_random_once(&udp_ehash_secret
, sizeof(udp_ehash_secret
));
430 return __inet_ehashfn(laddr
, lport
, faddr
, fport
,
431 udp_ehash_secret
+ net_hash_mix(net
));
434 /* called with rcu_read_lock() */
435 static struct sock
*udp4_lib_lookup2(struct net
*net
,
436 __be32 saddr
, __be16 sport
,
437 __be32 daddr
, unsigned int hnum
, int dif
, bool exact_dif
,
438 struct udp_hslot
*hslot2
,
441 struct sock
*sk
, *result
;
442 int score
, badness
, matches
= 0, reuseport
= 0;
447 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
448 score
= compute_score(sk
, net
, saddr
, sport
,
449 daddr
, hnum
, dif
, exact_dif
);
450 if (score
> badness
) {
451 reuseport
= sk
->sk_reuseport
;
453 hash
= udp_ehashfn(net
, daddr
, hnum
,
455 result
= reuseport_select_sock(sk
, hash
, skb
,
456 sizeof(struct udphdr
));
463 } else if (score
== badness
&& reuseport
) {
465 if (reciprocal_scale(hash
, matches
) == 0)
467 hash
= next_pseudo_random32(hash
);
473 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
474 * harder than this. -DaveM
476 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
477 __be16 sport
, __be32 daddr
, __be16 dport
,
478 int dif
, struct udp_table
*udptable
, struct sk_buff
*skb
)
480 struct sock
*sk
, *result
;
481 unsigned short hnum
= ntohs(dport
);
482 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
483 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
484 bool exact_dif
= udp_lib_exact_dif_match(net
, skb
);
485 int score
, badness
, matches
= 0, reuseport
= 0;
488 if (hslot
->count
> 10) {
489 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
);
490 slot2
= hash2
& udptable
->mask
;
491 hslot2
= &udptable
->hash2
[slot2
];
492 if (hslot
->count
< hslot2
->count
)
495 result
= udp4_lib_lookup2(net
, saddr
, sport
,
497 exact_dif
, hslot2
, skb
);
499 unsigned int old_slot2
= slot2
;
500 hash2
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
501 slot2
= hash2
& udptable
->mask
;
502 /* avoid searching the same slot again. */
503 if (unlikely(slot2
== old_slot2
))
506 hslot2
= &udptable
->hash2
[slot2
];
507 if (hslot
->count
< hslot2
->count
)
510 result
= udp4_lib_lookup2(net
, saddr
, sport
,
512 exact_dif
, hslot2
, skb
);
519 sk_for_each_rcu(sk
, &hslot
->head
) {
520 score
= compute_score(sk
, net
, saddr
, sport
,
521 daddr
, hnum
, dif
, exact_dif
);
522 if (score
> badness
) {
523 reuseport
= sk
->sk_reuseport
;
525 hash
= udp_ehashfn(net
, daddr
, hnum
,
527 result
= reuseport_select_sock(sk
, hash
, skb
,
528 sizeof(struct udphdr
));
535 } else if (score
== badness
&& reuseport
) {
537 if (reciprocal_scale(hash
, matches
) == 0)
539 hash
= next_pseudo_random32(hash
);
544 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
546 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
547 __be16 sport
, __be16 dport
,
548 struct udp_table
*udptable
)
550 const struct iphdr
*iph
= ip_hdr(skb
);
552 return __udp4_lib_lookup(dev_net(skb
->dev
), iph
->saddr
, sport
,
553 iph
->daddr
, dport
, inet_iif(skb
),
557 struct sock
*udp4_lib_lookup_skb(struct sk_buff
*skb
,
558 __be16 sport
, __be16 dport
)
560 return __udp4_lib_lookup_skb(skb
, sport
, dport
, &udp_table
);
562 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb
);
564 /* Must be called under rcu_read_lock().
565 * Does increment socket refcount.
567 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
568 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
569 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
570 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
571 __be32 daddr
, __be16 dport
, int dif
)
575 sk
= __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
576 dif
, &udp_table
, NULL
);
577 if (sk
&& !refcount_inc_not_zero(&sk
->sk_refcnt
))
581 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
584 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
585 __be16 loc_port
, __be32 loc_addr
,
586 __be16 rmt_port
, __be32 rmt_addr
,
587 int dif
, unsigned short hnum
)
589 struct inet_sock
*inet
= inet_sk(sk
);
591 if (!net_eq(sock_net(sk
), net
) ||
592 udp_sk(sk
)->udp_port_hash
!= hnum
||
593 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
594 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
595 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
596 ipv6_only_sock(sk
) ||
597 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
))
599 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
))
605 * This routine is called by the ICMP module when it gets some
606 * sort of error condition. If err < 0 then the socket should
607 * be closed and the error returned to the user. If err > 0
608 * it's just the icmp type << 8 | icmp code.
609 * Header points to the ip header of the error packet. We move
610 * on past this. Then (as it used to claim before adjustment)
611 * header points to the first 8 bytes of the udp header. We need
612 * to find the appropriate port.
615 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
617 struct inet_sock
*inet
;
618 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
619 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
620 const int type
= icmp_hdr(skb
)->type
;
621 const int code
= icmp_hdr(skb
)->code
;
625 struct net
*net
= dev_net(skb
->dev
);
627 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
628 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
,
631 __ICMP_INC_STATS(net
, ICMP_MIB_INERRORS
);
632 return; /* No socket for error */
641 case ICMP_TIME_EXCEEDED
:
644 case ICMP_SOURCE_QUENCH
:
646 case ICMP_PARAMETERPROB
:
650 case ICMP_DEST_UNREACH
:
651 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
652 ipv4_sk_update_pmtu(skb
, sk
, info
);
653 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
661 if (code
<= NR_ICMP_UNREACH
) {
662 harderr
= icmp_err_convert
[code
].fatal
;
663 err
= icmp_err_convert
[code
].errno
;
667 ipv4_sk_redirect(skb
, sk
);
672 * RFC1122: OK. Passes ICMP errors back to application, as per
675 if (!inet
->recverr
) {
676 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
679 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
682 sk
->sk_error_report(sk
);
687 void udp_err(struct sk_buff
*skb
, u32 info
)
689 __udp4_lib_err(skb
, info
, &udp_table
);
693 * Throw away all pending data and cancel the corking. Socket is locked.
695 void udp_flush_pending_frames(struct sock
*sk
)
697 struct udp_sock
*up
= udp_sk(sk
);
702 ip_flush_pending_frames(sk
);
705 EXPORT_SYMBOL(udp_flush_pending_frames
);
708 * udp4_hwcsum - handle outgoing HW checksumming
709 * @skb: sk_buff containing the filled-in UDP header
710 * (checksum field must be zeroed out)
711 * @src: source IP address
712 * @dst: destination IP address
714 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
716 struct udphdr
*uh
= udp_hdr(skb
);
717 int offset
= skb_transport_offset(skb
);
718 int len
= skb
->len
- offset
;
722 if (!skb_has_frag_list(skb
)) {
724 * Only one fragment on the socket.
726 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
727 skb
->csum_offset
= offsetof(struct udphdr
, check
);
728 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
731 struct sk_buff
*frags
;
734 * HW-checksum won't work as there are two or more
735 * fragments on the socket so that all csums of sk_buffs
738 skb_walk_frags(skb
, frags
) {
739 csum
= csum_add(csum
, frags
->csum
);
743 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
744 skb
->ip_summed
= CHECKSUM_NONE
;
746 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
748 uh
->check
= CSUM_MANGLED_0
;
751 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
753 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
754 * for the simple case like when setting the checksum for a UDP tunnel.
756 void udp_set_csum(bool nocheck
, struct sk_buff
*skb
,
757 __be32 saddr
, __be32 daddr
, int len
)
759 struct udphdr
*uh
= udp_hdr(skb
);
763 } else if (skb_is_gso(skb
)) {
764 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
765 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
767 uh
->check
= udp_v4_check(len
, saddr
, daddr
, lco_csum(skb
));
769 uh
->check
= CSUM_MANGLED_0
;
771 skb
->ip_summed
= CHECKSUM_PARTIAL
;
772 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
773 skb
->csum_offset
= offsetof(struct udphdr
, check
);
774 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
777 EXPORT_SYMBOL(udp_set_csum
);
779 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
)
781 struct sock
*sk
= skb
->sk
;
782 struct inet_sock
*inet
= inet_sk(sk
);
785 int is_udplite
= IS_UDPLITE(sk
);
786 int offset
= skb_transport_offset(skb
);
787 int len
= skb
->len
- offset
;
791 * Create a UDP header
794 uh
->source
= inet
->inet_sport
;
795 uh
->dest
= fl4
->fl4_dport
;
796 uh
->len
= htons(len
);
799 if (is_udplite
) /* UDP-Lite */
800 csum
= udplite_csum(skb
);
802 else if (sk
->sk_no_check_tx
&& !skb_is_gso(skb
)) { /* UDP csum off */
804 skb
->ip_summed
= CHECKSUM_NONE
;
807 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
809 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
813 csum
= udp_csum(skb
);
815 /* add protocol-dependent pseudo-header */
816 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
817 sk
->sk_protocol
, csum
);
819 uh
->check
= CSUM_MANGLED_0
;
822 err
= ip_send_skb(sock_net(sk
), skb
);
824 if (err
== -ENOBUFS
&& !inet
->recverr
) {
825 UDP_INC_STATS(sock_net(sk
),
826 UDP_MIB_SNDBUFERRORS
, is_udplite
);
830 UDP_INC_STATS(sock_net(sk
),
831 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
836 * Push out all pending data as one UDP datagram. Socket is locked.
838 int udp_push_pending_frames(struct sock
*sk
)
840 struct udp_sock
*up
= udp_sk(sk
);
841 struct inet_sock
*inet
= inet_sk(sk
);
842 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
846 skb
= ip_finish_skb(sk
, fl4
);
850 err
= udp_send_skb(skb
, fl4
);
857 EXPORT_SYMBOL(udp_push_pending_frames
);
859 int udp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
861 struct inet_sock
*inet
= inet_sk(sk
);
862 struct udp_sock
*up
= udp_sk(sk
);
863 struct flowi4 fl4_stack
;
866 struct ipcm_cookie ipc
;
867 struct rtable
*rt
= NULL
;
870 __be32 daddr
, faddr
, saddr
;
873 int err
, is_udplite
= IS_UDPLITE(sk
);
874 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
875 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
877 struct ip_options_data opt_copy
;
886 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
894 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
896 fl4
= &inet
->cork
.fl
.u
.ip4
;
899 * There are pending frames.
900 * The socket lock must be held while it's corked.
903 if (likely(up
->pending
)) {
904 if (unlikely(up
->pending
!= AF_INET
)) {
912 ulen
+= sizeof(struct udphdr
);
915 * Get and verify the address.
918 DECLARE_SOCKADDR(struct sockaddr_in
*, usin
, msg
->msg_name
);
919 if (msg
->msg_namelen
< sizeof(*usin
))
921 if (usin
->sin_family
!= AF_INET
) {
922 if (usin
->sin_family
!= AF_UNSPEC
)
923 return -EAFNOSUPPORT
;
926 daddr
= usin
->sin_addr
.s_addr
;
927 dport
= usin
->sin_port
;
931 if (sk
->sk_state
!= TCP_ESTABLISHED
)
932 return -EDESTADDRREQ
;
933 daddr
= inet
->inet_daddr
;
934 dport
= inet
->inet_dport
;
935 /* Open fast path for connected socket.
936 Route will not be used, if at least one option is set.
941 ipc
.sockc
.tsflags
= sk
->sk_tsflags
;
942 ipc
.addr
= inet
->inet_saddr
;
943 ipc
.oif
= sk
->sk_bound_dev_if
;
945 if (msg
->msg_controllen
) {
946 err
= ip_cmsg_send(sk
, msg
, &ipc
, sk
->sk_family
== AF_INET6
);
956 struct ip_options_rcu
*inet_opt
;
959 inet_opt
= rcu_dereference(inet
->inet_opt
);
961 memcpy(&opt_copy
, inet_opt
,
962 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
963 ipc
.opt
= &opt_copy
.opt
;
969 ipc
.addr
= faddr
= daddr
;
971 sock_tx_timestamp(sk
, ipc
.sockc
.tsflags
, &ipc
.tx_flags
);
973 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
976 faddr
= ipc
.opt
->opt
.faddr
;
979 tos
= get_rttos(&ipc
, inet
);
980 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
981 (msg
->msg_flags
& MSG_DONTROUTE
) ||
982 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
987 if (ipv4_is_multicast(daddr
)) {
989 ipc
.oif
= inet
->mc_index
;
991 saddr
= inet
->mc_addr
;
994 ipc
.oif
= inet
->uc_index
;
997 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
1000 struct net
*net
= sock_net(sk
);
1001 __u8 flow_flags
= inet_sk_flowi_flags(sk
);
1005 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
1006 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1008 faddr
, saddr
, dport
, inet
->inet_sport
,
1011 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
1012 rt
= ip_route_output_flow(net
, fl4
, sk
);
1016 if (err
== -ENETUNREACH
)
1017 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
1022 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1023 !sock_flag(sk
, SOCK_BROADCAST
))
1026 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1029 if (msg
->msg_flags
&MSG_CONFIRM
)
1035 daddr
= ipc
.addr
= fl4
->daddr
;
1037 /* Lockless fast path for the non-corking case. */
1039 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
, ulen
,
1040 sizeof(struct udphdr
), &ipc
, &rt
,
1043 if (!IS_ERR_OR_NULL(skb
))
1044 err
= udp_send_skb(skb
, fl4
);
1049 if (unlikely(up
->pending
)) {
1050 /* The socket is already corked while preparing it. */
1051 /* ... which is an evident application bug. --ANK */
1054 net_dbg_ratelimited("cork app bug 2\n");
1059 * Now cork the socket to pend data.
1061 fl4
= &inet
->cork
.fl
.u
.ip4
;
1064 fl4
->fl4_dport
= dport
;
1065 fl4
->fl4_sport
= inet
->inet_sport
;
1066 up
->pending
= AF_INET
;
1070 err
= ip_append_data(sk
, fl4
, getfrag
, msg
, ulen
,
1071 sizeof(struct udphdr
), &ipc
, &rt
,
1072 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1074 udp_flush_pending_frames(sk
);
1076 err
= udp_push_pending_frames(sk
);
1077 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1088 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1089 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1090 * we don't have a good statistic (IpOutDiscards but it can be too many
1091 * things). We could add another new stat but at least for now that
1092 * seems like overkill.
1094 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1095 UDP_INC_STATS(sock_net(sk
),
1096 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1101 if (msg
->msg_flags
& MSG_PROBE
)
1102 dst_confirm_neigh(&rt
->dst
, &fl4
->daddr
);
1103 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1104 goto back_from_confirm
;
1108 EXPORT_SYMBOL(udp_sendmsg
);
1110 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1111 size_t size
, int flags
)
1113 struct inet_sock
*inet
= inet_sk(sk
);
1114 struct udp_sock
*up
= udp_sk(sk
);
1117 if (flags
& MSG_SENDPAGE_NOTLAST
)
1121 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1123 /* Call udp_sendmsg to specify destination address which
1124 * sendpage interface can't pass.
1125 * This will succeed only when the socket is connected.
1127 ret
= udp_sendmsg(sk
, &msg
, 0);
1134 if (unlikely(!up
->pending
)) {
1137 net_dbg_ratelimited("udp cork app bug 3\n");
1141 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1142 page
, offset
, size
, flags
);
1143 if (ret
== -EOPNOTSUPP
) {
1145 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1149 udp_flush_pending_frames(sk
);
1154 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1155 ret
= udp_push_pending_frames(sk
);
1163 #define UDP_SKB_IS_STATELESS 0x80000000
1165 static void udp_set_dev_scratch(struct sk_buff
*skb
)
1167 struct udp_dev_scratch
*scratch
= udp_skb_scratch(skb
);
1169 BUILD_BUG_ON(sizeof(struct udp_dev_scratch
) > sizeof(long));
1170 scratch
->_tsize_state
= skb
->truesize
;
1171 #if BITS_PER_LONG == 64
1172 scratch
->len
= skb
->len
;
1173 scratch
->csum_unnecessary
= !!skb_csum_unnecessary(skb
);
1174 scratch
->is_linear
= !skb_is_nonlinear(skb
);
1176 if (likely(!skb
->_skb_refdst
&& !skb_sec_path(skb
)))
1177 scratch
->_tsize_state
|= UDP_SKB_IS_STATELESS
;
1180 static int udp_skb_truesize(struct sk_buff
*skb
)
1182 return udp_skb_scratch(skb
)->_tsize_state
& ~UDP_SKB_IS_STATELESS
;
1185 static bool udp_skb_has_head_state(struct sk_buff
*skb
)
1187 return !(udp_skb_scratch(skb
)->_tsize_state
& UDP_SKB_IS_STATELESS
);
1190 /* fully reclaim rmem/fwd memory allocated for skb */
1191 static void udp_rmem_release(struct sock
*sk
, int size
, int partial
,
1192 bool rx_queue_lock_held
)
1194 struct udp_sock
*up
= udp_sk(sk
);
1195 struct sk_buff_head
*sk_queue
;
1198 if (likely(partial
)) {
1199 up
->forward_deficit
+= size
;
1200 size
= up
->forward_deficit
;
1201 if (size
< (sk
->sk_rcvbuf
>> 2) &&
1202 !skb_queue_empty(&up
->reader_queue
))
1205 size
+= up
->forward_deficit
;
1207 up
->forward_deficit
= 0;
1209 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1210 * if the called don't held it already
1212 sk_queue
= &sk
->sk_receive_queue
;
1213 if (!rx_queue_lock_held
)
1214 spin_lock(&sk_queue
->lock
);
1217 sk
->sk_forward_alloc
+= size
;
1218 amt
= (sk
->sk_forward_alloc
- partial
) & ~(SK_MEM_QUANTUM
- 1);
1219 sk
->sk_forward_alloc
-= amt
;
1222 __sk_mem_reduce_allocated(sk
, amt
>> SK_MEM_QUANTUM_SHIFT
);
1224 atomic_sub(size
, &sk
->sk_rmem_alloc
);
1226 /* this can save us from acquiring the rx queue lock on next receive */
1227 skb_queue_splice_tail_init(sk_queue
, &up
->reader_queue
);
1229 if (!rx_queue_lock_held
)
1230 spin_unlock(&sk_queue
->lock
);
1233 /* Note: called with reader_queue.lock held.
1234 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1235 * This avoids a cache line miss while receive_queue lock is held.
1236 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1238 void udp_skb_destructor(struct sock
*sk
, struct sk_buff
*skb
)
1240 prefetch(&skb
->data
);
1241 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, false);
1243 EXPORT_SYMBOL(udp_skb_destructor
);
1245 /* as above, but the caller held the rx queue lock, too */
1246 static void udp_skb_dtor_locked(struct sock
*sk
, struct sk_buff
*skb
)
1248 prefetch(&skb
->data
);
1249 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, true);
1252 /* Idea of busylocks is to let producers grab an extra spinlock
1253 * to relieve pressure on the receive_queue spinlock shared by consumer.
1254 * Under flood, this means that only one producer can be in line
1255 * trying to acquire the receive_queue spinlock.
1256 * These busylock can be allocated on a per cpu manner, instead of a
1257 * per socket one (that would consume a cache line per socket)
1259 static int udp_busylocks_log __read_mostly
;
1260 static spinlock_t
*udp_busylocks __read_mostly
;
1262 static spinlock_t
*busylock_acquire(void *ptr
)
1266 busy
= udp_busylocks
+ hash_ptr(ptr
, udp_busylocks_log
);
1271 static void busylock_release(spinlock_t
*busy
)
1277 int __udp_enqueue_schedule_skb(struct sock
*sk
, struct sk_buff
*skb
)
1279 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
1280 int rmem
, delta
, amt
, err
= -ENOMEM
;
1281 spinlock_t
*busy
= NULL
;
1284 /* try to avoid the costly atomic add/sub pair when the receive
1285 * queue is full; always allow at least a packet
1287 rmem
= atomic_read(&sk
->sk_rmem_alloc
);
1288 if (rmem
> sk
->sk_rcvbuf
)
1291 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1292 * having linear skbs :
1293 * - Reduce memory overhead and thus increase receive queue capacity
1294 * - Less cache line misses at copyout() time
1295 * - Less work at consume_skb() (less alien page frag freeing)
1297 if (rmem
> (sk
->sk_rcvbuf
>> 1)) {
1300 busy
= busylock_acquire(sk
);
1302 size
= skb
->truesize
;
1303 udp_set_dev_scratch(skb
);
1305 /* we drop only if the receive buf is full and the receive
1306 * queue contains some other skb
1308 rmem
= atomic_add_return(size
, &sk
->sk_rmem_alloc
);
1309 if (rmem
> (size
+ sk
->sk_rcvbuf
))
1312 spin_lock(&list
->lock
);
1313 if (size
>= sk
->sk_forward_alloc
) {
1314 amt
= sk_mem_pages(size
);
1315 delta
= amt
<< SK_MEM_QUANTUM_SHIFT
;
1316 if (!__sk_mem_raise_allocated(sk
, delta
, amt
, SK_MEM_RECV
)) {
1318 spin_unlock(&list
->lock
);
1322 sk
->sk_forward_alloc
+= delta
;
1325 sk
->sk_forward_alloc
-= size
;
1327 /* no need to setup a destructor, we will explicitly release the
1328 * forward allocated memory on dequeue
1330 sock_skb_set_dropcount(sk
, skb
);
1332 __skb_queue_tail(list
, skb
);
1333 spin_unlock(&list
->lock
);
1335 if (!sock_flag(sk
, SOCK_DEAD
))
1336 sk
->sk_data_ready(sk
);
1338 busylock_release(busy
);
1342 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1345 atomic_inc(&sk
->sk_drops
);
1346 busylock_release(busy
);
1349 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb
);
1351 void udp_destruct_sock(struct sock
*sk
)
1353 /* reclaim completely the forward allocated memory */
1354 struct udp_sock
*up
= udp_sk(sk
);
1355 unsigned int total
= 0;
1356 struct sk_buff
*skb
;
1358 skb_queue_splice_tail_init(&sk
->sk_receive_queue
, &up
->reader_queue
);
1359 while ((skb
= __skb_dequeue(&up
->reader_queue
)) != NULL
) {
1360 total
+= skb
->truesize
;
1363 udp_rmem_release(sk
, total
, 0, true);
1365 inet_sock_destruct(sk
);
1367 EXPORT_SYMBOL_GPL(udp_destruct_sock
);
1369 int udp_init_sock(struct sock
*sk
)
1371 skb_queue_head_init(&udp_sk(sk
)->reader_queue
);
1372 sk
->sk_destruct
= udp_destruct_sock
;
1375 EXPORT_SYMBOL_GPL(udp_init_sock
);
1377 void skb_consume_udp(struct sock
*sk
, struct sk_buff
*skb
, int len
)
1379 if (unlikely(READ_ONCE(sk
->sk_peek_off
) >= 0)) {
1380 bool slow
= lock_sock_fast(sk
);
1382 sk_peek_offset_bwd(sk
, len
);
1383 unlock_sock_fast(sk
, slow
);
1386 if (!skb_unref(skb
))
1389 /* In the more common cases we cleared the head states previously,
1390 * see __udp_queue_rcv_skb().
1392 if (unlikely(udp_skb_has_head_state(skb
)))
1393 skb_release_head_state(skb
);
1394 __consume_stateless_skb(skb
);
1396 EXPORT_SYMBOL_GPL(skb_consume_udp
);
1398 static struct sk_buff
*__first_packet_length(struct sock
*sk
,
1399 struct sk_buff_head
*rcvq
,
1402 struct sk_buff
*skb
;
1404 while ((skb
= skb_peek(rcvq
)) != NULL
) {
1405 if (udp_lib_checksum_complete(skb
)) {
1406 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1408 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
1410 atomic_inc(&sk
->sk_drops
);
1411 __skb_unlink(skb
, rcvq
);
1412 *total
+= skb
->truesize
;
1415 /* the csum related bits could be changed, refresh
1418 udp_set_dev_scratch(skb
);
1426 * first_packet_length - return length of first packet in receive queue
1429 * Drops all bad checksum frames, until a valid one is found.
1430 * Returns the length of found skb, or -1 if none is found.
1432 static int first_packet_length(struct sock
*sk
)
1434 struct sk_buff_head
*rcvq
= &udp_sk(sk
)->reader_queue
;
1435 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1436 struct sk_buff
*skb
;
1440 spin_lock_bh(&rcvq
->lock
);
1441 skb
= __first_packet_length(sk
, rcvq
, &total
);
1442 if (!skb
&& !skb_queue_empty(sk_queue
)) {
1443 spin_lock(&sk_queue
->lock
);
1444 skb_queue_splice_tail_init(sk_queue
, rcvq
);
1445 spin_unlock(&sk_queue
->lock
);
1447 skb
= __first_packet_length(sk
, rcvq
, &total
);
1449 res
= skb
? skb
->len
: -1;
1451 udp_rmem_release(sk
, total
, 1, false);
1452 spin_unlock_bh(&rcvq
->lock
);
1457 * IOCTL requests applicable to the UDP protocol
1460 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1465 int amount
= sk_wmem_alloc_get(sk
);
1467 return put_user(amount
, (int __user
*)arg
);
1472 int amount
= max_t(int, 0, first_packet_length(sk
));
1474 return put_user(amount
, (int __user
*)arg
);
1478 return -ENOIOCTLCMD
;
1483 EXPORT_SYMBOL(udp_ioctl
);
1485 struct sk_buff
*__skb_recv_udp(struct sock
*sk
, unsigned int flags
,
1486 int noblock
, int *peeked
, int *off
, int *err
)
1488 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1489 struct sk_buff_head
*queue
;
1490 struct sk_buff
*last
;
1494 queue
= &udp_sk(sk
)->reader_queue
;
1495 flags
|= noblock
? MSG_DONTWAIT
: 0;
1496 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
1498 struct sk_buff
*skb
;
1500 error
= sock_error(sk
);
1507 spin_lock_bh(&queue
->lock
);
1508 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1513 spin_unlock_bh(&queue
->lock
);
1517 if (skb_queue_empty(sk_queue
)) {
1518 spin_unlock_bh(&queue
->lock
);
1522 /* refill the reader queue and walk it again
1523 * keep both queues locked to avoid re-acquiring
1524 * the sk_receive_queue lock if fwd memory scheduling
1527 spin_lock(&sk_queue
->lock
);
1528 skb_queue_splice_tail_init(sk_queue
, queue
);
1530 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1531 udp_skb_dtor_locked
,
1534 spin_unlock(&sk_queue
->lock
);
1535 spin_unlock_bh(&queue
->lock
);
1540 if (!sk_can_busy_loop(sk
))
1543 sk_busy_loop(sk
, flags
& MSG_DONTWAIT
);
1544 } while (!skb_queue_empty(sk_queue
));
1546 /* sk_queue is empty, reader_queue may contain peeked packets */
1548 !__skb_wait_for_more_packets(sk
, &error
, &timeo
,
1549 (struct sk_buff
*)sk_queue
));
1554 EXPORT_SYMBOL_GPL(__skb_recv_udp
);
1557 * This should be easy, if there is something there we
1558 * return it, otherwise we block.
1561 int udp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int noblock
,
1562 int flags
, int *addr_len
)
1564 struct inet_sock
*inet
= inet_sk(sk
);
1565 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, msg
->msg_name
);
1566 struct sk_buff
*skb
;
1567 unsigned int ulen
, copied
;
1568 int peeked
, peeking
, off
;
1570 int is_udplite
= IS_UDPLITE(sk
);
1571 bool checksum_valid
= false;
1573 if (flags
& MSG_ERRQUEUE
)
1574 return ip_recv_error(sk
, msg
, len
, addr_len
);
1577 peeking
= flags
& MSG_PEEK
;
1578 off
= sk_peek_offset(sk
, flags
);
1579 skb
= __skb_recv_udp(sk
, flags
, noblock
, &peeked
, &off
, &err
);
1583 ulen
= udp_skb_len(skb
);
1585 if (copied
> ulen
- off
)
1586 copied
= ulen
- off
;
1587 else if (copied
< ulen
)
1588 msg
->msg_flags
|= MSG_TRUNC
;
1591 * If checksum is needed at all, try to do it while copying the
1592 * data. If the data is truncated, or if we only want a partial
1593 * coverage checksum (UDP-Lite), do it before the copy.
1596 if (copied
< ulen
|| peeking
||
1597 (is_udplite
&& UDP_SKB_CB(skb
)->partial_cov
)) {
1598 checksum_valid
= udp_skb_csum_unnecessary(skb
) ||
1599 !__udp_lib_checksum_complete(skb
);
1600 if (!checksum_valid
)
1604 if (checksum_valid
|| udp_skb_csum_unnecessary(skb
)) {
1605 if (udp_skb_is_linear(skb
))
1606 err
= copy_linear_skb(skb
, copied
, off
, &msg
->msg_iter
);
1608 err
= skb_copy_datagram_msg(skb
, off
, msg
, copied
);
1610 err
= skb_copy_and_csum_datagram_msg(skb
, off
, msg
);
1616 if (unlikely(err
)) {
1618 atomic_inc(&sk
->sk_drops
);
1619 UDP_INC_STATS(sock_net(sk
),
1620 UDP_MIB_INERRORS
, is_udplite
);
1627 UDP_INC_STATS(sock_net(sk
),
1628 UDP_MIB_INDATAGRAMS
, is_udplite
);
1630 sock_recv_ts_and_drops(msg
, sk
, skb
);
1632 /* Copy the address. */
1634 sin
->sin_family
= AF_INET
;
1635 sin
->sin_port
= udp_hdr(skb
)->source
;
1636 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1637 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1638 *addr_len
= sizeof(*sin
);
1640 if (inet
->cmsg_flags
)
1641 ip_cmsg_recv_offset(msg
, sk
, skb
, sizeof(struct udphdr
), off
);
1644 if (flags
& MSG_TRUNC
)
1647 skb_consume_udp(sk
, skb
, peeking
? -err
: err
);
1651 if (!__sk_queue_drop_skb(sk
, &udp_sk(sk
)->reader_queue
, skb
, flags
,
1652 udp_skb_destructor
)) {
1653 UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1654 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1658 /* starting over for a new packet, but check if we need to yield */
1660 msg
->msg_flags
&= ~MSG_TRUNC
;
1664 int __udp_disconnect(struct sock
*sk
, int flags
)
1666 struct inet_sock
*inet
= inet_sk(sk
);
1668 * 1003.1g - break association.
1671 sk
->sk_state
= TCP_CLOSE
;
1672 inet
->inet_daddr
= 0;
1673 inet
->inet_dport
= 0;
1674 sock_rps_reset_rxhash(sk
);
1675 sk
->sk_bound_dev_if
= 0;
1676 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1677 inet_reset_saddr(sk
);
1679 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1680 sk
->sk_prot
->unhash(sk
);
1681 inet
->inet_sport
= 0;
1686 EXPORT_SYMBOL(__udp_disconnect
);
1688 int udp_disconnect(struct sock
*sk
, int flags
)
1691 __udp_disconnect(sk
, flags
);
1695 EXPORT_SYMBOL(udp_disconnect
);
1697 void udp_lib_unhash(struct sock
*sk
)
1699 if (sk_hashed(sk
)) {
1700 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1701 struct udp_hslot
*hslot
, *hslot2
;
1703 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1704 udp_sk(sk
)->udp_port_hash
);
1705 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1707 spin_lock_bh(&hslot
->lock
);
1708 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1709 reuseport_detach_sock(sk
);
1710 if (sk_del_node_init_rcu(sk
)) {
1712 inet_sk(sk
)->inet_num
= 0;
1713 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1715 spin_lock(&hslot2
->lock
);
1716 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1718 spin_unlock(&hslot2
->lock
);
1720 spin_unlock_bh(&hslot
->lock
);
1723 EXPORT_SYMBOL(udp_lib_unhash
);
1726 * inet_rcv_saddr was changed, we must rehash secondary hash
1728 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1730 if (sk_hashed(sk
)) {
1731 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1732 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1734 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1735 nhslot2
= udp_hashslot2(udptable
, newhash
);
1736 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1738 if (hslot2
!= nhslot2
||
1739 rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1740 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1741 udp_sk(sk
)->udp_port_hash
);
1742 /* we must lock primary chain too */
1743 spin_lock_bh(&hslot
->lock
);
1744 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1745 reuseport_detach_sock(sk
);
1747 if (hslot2
!= nhslot2
) {
1748 spin_lock(&hslot2
->lock
);
1749 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1751 spin_unlock(&hslot2
->lock
);
1753 spin_lock(&nhslot2
->lock
);
1754 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1757 spin_unlock(&nhslot2
->lock
);
1760 spin_unlock_bh(&hslot
->lock
);
1764 EXPORT_SYMBOL(udp_lib_rehash
);
1766 static void udp_v4_rehash(struct sock
*sk
)
1768 u16 new_hash
= udp4_portaddr_hash(sock_net(sk
),
1769 inet_sk(sk
)->inet_rcv_saddr
,
1770 inet_sk(sk
)->inet_num
);
1771 udp_lib_rehash(sk
, new_hash
);
1774 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1778 if (inet_sk(sk
)->inet_daddr
) {
1779 sock_rps_save_rxhash(sk
, skb
);
1780 sk_mark_napi_id(sk
, skb
);
1781 sk_incoming_cpu_update(sk
);
1783 sk_mark_napi_id_once(sk
, skb
);
1786 /* At recvmsg() time we may access skb->dst or skb->sp depending on
1787 * the IP options and the cmsg flags, elsewhere can we clear all
1788 * pending head states while they are hot in the cache
1790 if (likely(IPCB(skb
)->opt
.optlen
== 0 && !skb_sec_path(skb
)))
1791 skb_release_head_state(skb
);
1793 rc
= __udp_enqueue_schedule_skb(sk
, skb
);
1795 int is_udplite
= IS_UDPLITE(sk
);
1797 /* Note that an ENOMEM error is charged twice */
1799 UDP_INC_STATS(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1801 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1803 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1810 static struct static_key udp_encap_needed __read_mostly
;
1811 void udp_encap_enable(void)
1813 if (!static_key_enabled(&udp_encap_needed
))
1814 static_key_slow_inc(&udp_encap_needed
);
1816 EXPORT_SYMBOL(udp_encap_enable
);
1821 * >0: "udp encap" protocol resubmission
1823 * Note that in the success and error cases, the skb is assumed to
1824 * have either been requeued or freed.
1826 static int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1828 struct udp_sock
*up
= udp_sk(sk
);
1829 int is_udplite
= IS_UDPLITE(sk
);
1832 * Charge it to the socket, dropping if the queue is full.
1834 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1838 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1839 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1842 * This is an encapsulation socket so pass the skb to
1843 * the socket's udp_encap_rcv() hook. Otherwise, just
1844 * fall through and pass this up the UDP socket.
1845 * up->encap_rcv() returns the following value:
1846 * =0 if skb was successfully passed to the encap
1847 * handler or was discarded by it.
1848 * >0 if skb should be passed on to UDP.
1849 * <0 if skb should be resubmitted as proto -N
1852 /* if we're overly short, let UDP handle it */
1853 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
1857 /* Verify checksum before giving to encap */
1858 if (udp_lib_checksum_complete(skb
))
1861 ret
= encap_rcv(sk
, skb
);
1863 __UDP_INC_STATS(sock_net(sk
),
1864 UDP_MIB_INDATAGRAMS
,
1870 /* FALLTHROUGH -- it's a UDP Packet */
1874 * UDP-Lite specific tests, ignored on UDP sockets
1876 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1879 * MIB statistics other than incrementing the error count are
1880 * disabled for the following two types of errors: these depend
1881 * on the application settings, not on the functioning of the
1882 * protocol stack as such.
1884 * RFC 3828 here recommends (sec 3.3): "There should also be a
1885 * way ... to ... at least let the receiving application block
1886 * delivery of packets with coverage values less than a value
1887 * provided by the application."
1889 if (up
->pcrlen
== 0) { /* full coverage was set */
1890 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1891 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1894 /* The next case involves violating the min. coverage requested
1895 * by the receiver. This is subtle: if receiver wants x and x is
1896 * greater than the buffersize/MTU then receiver will complain
1897 * that it wants x while sender emits packets of smaller size y.
1898 * Therefore the above ...()->partial_cov statement is essential.
1900 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1901 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1902 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1907 prefetch(&sk
->sk_rmem_alloc
);
1908 if (rcu_access_pointer(sk
->sk_filter
) &&
1909 udp_lib_checksum_complete(skb
))
1912 if (sk_filter_trim_cap(sk
, skb
, sizeof(struct udphdr
)))
1915 udp_csum_pull_header(skb
);
1917 ipv4_pktinfo_prepare(sk
, skb
);
1918 return __udp_queue_rcv_skb(sk
, skb
);
1921 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1923 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1924 atomic_inc(&sk
->sk_drops
);
1929 /* For TCP sockets, sk_rx_dst is protected by socket lock
1930 * For UDP, we use xchg() to guard against concurrent changes.
1932 bool udp_sk_rx_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
1934 struct dst_entry
*old
;
1936 if (dst_hold_safe(dst
)) {
1937 old
= xchg(&sk
->sk_rx_dst
, dst
);
1943 EXPORT_SYMBOL(udp_sk_rx_dst_set
);
1946 * Multicasts and broadcasts go to each listener.
1948 * Note: called only from the BH handler context.
1950 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1952 __be32 saddr
, __be32 daddr
,
1953 struct udp_table
*udptable
,
1956 struct sock
*sk
, *first
= NULL
;
1957 unsigned short hnum
= ntohs(uh
->dest
);
1958 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
1959 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
1960 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
1961 int dif
= skb
->dev
->ifindex
;
1962 struct hlist_node
*node
;
1963 struct sk_buff
*nskb
;
1966 hash2_any
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
) &
1968 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
) & udptable
->mask
;
1970 hslot
= &udptable
->hash2
[hash2
];
1971 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
1974 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
1975 if (!__udp_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
1976 uh
->source
, saddr
, dif
, hnum
))
1983 nskb
= skb_clone(skb
, GFP_ATOMIC
);
1985 if (unlikely(!nskb
)) {
1986 atomic_inc(&sk
->sk_drops
);
1987 __UDP_INC_STATS(net
, UDP_MIB_RCVBUFERRORS
,
1989 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
,
1993 if (udp_queue_rcv_skb(sk
, nskb
) > 0)
1997 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1998 if (use_hash2
&& hash2
!= hash2_any
) {
2004 if (udp_queue_rcv_skb(first
, skb
) > 0)
2008 __UDP_INC_STATS(net
, UDP_MIB_IGNOREDMULTI
,
2009 proto
== IPPROTO_UDPLITE
);
2014 /* Initialize UDP checksum. If exited with zero value (success),
2015 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2016 * Otherwise, csum completion requires chacksumming packet body,
2017 * including udp header and folding it to skb->csum.
2019 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
2024 UDP_SKB_CB(skb
)->partial_cov
= 0;
2025 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
2027 if (proto
== IPPROTO_UDPLITE
) {
2028 err
= udplite_checksum_init(skb
, uh
);
2033 /* Note, we are only interested in != 0 or == 0, thus the
2036 return (__force
int)skb_checksum_init_zero_check(skb
, proto
, uh
->check
,
2037 inet_compute_pseudo
);
2041 * All we need to do is get the socket, and then do a checksum.
2044 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
2049 unsigned short ulen
;
2050 struct rtable
*rt
= skb_rtable(skb
);
2051 __be32 saddr
, daddr
;
2052 struct net
*net
= dev_net(skb
->dev
);
2055 * Validate the packet.
2057 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
2058 goto drop
; /* No space for header. */
2061 ulen
= ntohs(uh
->len
);
2062 saddr
= ip_hdr(skb
)->saddr
;
2063 daddr
= ip_hdr(skb
)->daddr
;
2065 if (ulen
> skb
->len
)
2068 if (proto
== IPPROTO_UDP
) {
2069 /* UDP validates ulen. */
2070 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
2075 if (udp4_csum_init(skb
, uh
, proto
))
2078 sk
= skb_steal_sock(skb
);
2080 struct dst_entry
*dst
= skb_dst(skb
);
2083 if (unlikely(sk
->sk_rx_dst
!= dst
))
2084 udp_sk_rx_dst_set(sk
, dst
);
2086 ret
= udp_queue_rcv_skb(sk
, skb
);
2088 /* a return value > 0 means to resubmit the input, but
2089 * it wants the return to be -protocol, or 0
2096 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
2097 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
2098 saddr
, daddr
, udptable
, proto
);
2100 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
2104 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
2105 skb_checksum_try_convert(skb
, IPPROTO_UDP
, uh
->check
,
2106 inet_compute_pseudo
);
2108 ret
= udp_queue_rcv_skb(sk
, skb
);
2110 /* a return value > 0 means to resubmit the input, but
2111 * it wants the return to be -protocol, or 0
2118 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2122 /* No socket. Drop packet silently, if checksum is wrong */
2123 if (udp_lib_checksum_complete(skb
))
2126 __UDP_INC_STATS(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
2127 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
2130 * Hmm. We got an UDP packet to a port to which we
2131 * don't wanna listen. Ignore it.
2137 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2138 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2139 &saddr
, ntohs(uh
->source
),
2141 &daddr
, ntohs(uh
->dest
));
2146 * RFC1122: OK. Discards the bad packet silently (as far as
2147 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2149 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2150 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2151 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
2153 __UDP_INC_STATS(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
2155 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
2160 /* We can only early demux multicast if there is a single matching socket.
2161 * If more than one socket found returns NULL
2163 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
2164 __be16 loc_port
, __be32 loc_addr
,
2165 __be16 rmt_port
, __be32 rmt_addr
,
2168 struct sock
*sk
, *result
;
2169 unsigned short hnum
= ntohs(loc_port
);
2170 unsigned int slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
2171 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
2173 /* Do not bother scanning a too big list */
2174 if (hslot
->count
> 10)
2178 sk_for_each_rcu(sk
, &hslot
->head
) {
2179 if (__udp_is_mcast_sock(net
, sk
, loc_port
, loc_addr
,
2180 rmt_port
, rmt_addr
, dif
, hnum
)) {
2190 /* For unicast we should only early demux connected sockets or we can
2191 * break forwarding setups. The chains here can be long so only check
2192 * if the first socket is an exact match and if not move on.
2194 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
2195 __be16 loc_port
, __be32 loc_addr
,
2196 __be16 rmt_port
, __be32 rmt_addr
,
2199 unsigned short hnum
= ntohs(loc_port
);
2200 unsigned int hash2
= udp4_portaddr_hash(net
, loc_addr
, hnum
);
2201 unsigned int slot2
= hash2
& udp_table
.mask
;
2202 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
2203 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
);
2204 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
2207 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
2208 if (INET_MATCH(sk
, net
, acookie
, rmt_addr
,
2209 loc_addr
, ports
, dif
))
2211 /* Only check first socket in chain */
2217 int udp_v4_early_demux(struct sk_buff
*skb
)
2219 struct net
*net
= dev_net(skb
->dev
);
2220 struct in_device
*in_dev
= NULL
;
2221 const struct iphdr
*iph
;
2222 const struct udphdr
*uh
;
2223 struct sock
*sk
= NULL
;
2224 struct dst_entry
*dst
;
2225 int dif
= skb
->dev
->ifindex
;
2228 /* validate the packet */
2229 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
2235 if (skb
->pkt_type
== PACKET_MULTICAST
) {
2236 in_dev
= __in_dev_get_rcu(skb
->dev
);
2241 ours
= ip_check_mc_rcu(in_dev
, iph
->daddr
, iph
->saddr
,
2246 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2247 uh
->source
, iph
->saddr
, dif
);
2248 } else if (skb
->pkt_type
== PACKET_HOST
) {
2249 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2250 uh
->source
, iph
->saddr
, dif
);
2253 if (!sk
|| !refcount_inc_not_zero(&sk
->sk_refcnt
))
2257 skb
->destructor
= sock_efree
;
2258 dst
= READ_ONCE(sk
->sk_rx_dst
);
2261 dst
= dst_check(dst
, 0);
2265 /* set noref for now.
2266 * any place which wants to hold dst has to call
2269 skb_dst_set_noref(skb
, dst
);
2271 /* for unconnected multicast sockets we need to validate
2272 * the source on each packet
2274 if (!inet_sk(sk
)->inet_daddr
&& in_dev
)
2275 return ip_mc_validate_source(skb
, iph
->daddr
,
2276 iph
->saddr
, iph
->tos
,
2277 skb
->dev
, in_dev
, &itag
);
2282 int udp_rcv(struct sk_buff
*skb
)
2284 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
2287 void udp_destroy_sock(struct sock
*sk
)
2289 struct udp_sock
*up
= udp_sk(sk
);
2290 bool slow
= lock_sock_fast(sk
);
2291 udp_flush_pending_frames(sk
);
2292 unlock_sock_fast(sk
, slow
);
2293 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
2294 void (*encap_destroy
)(struct sock
*sk
);
2295 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
2302 * Socket option code for UDP
2304 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
2305 char __user
*optval
, unsigned int optlen
,
2306 int (*push_pending_frames
)(struct sock
*))
2308 struct udp_sock
*up
= udp_sk(sk
);
2311 int is_udplite
= IS_UDPLITE(sk
);
2313 if (optlen
< sizeof(int))
2316 if (get_user(val
, (int __user
*)optval
))
2319 valbool
= val
? 1 : 0;
2328 push_pending_frames(sk
);
2336 case UDP_ENCAP_ESPINUDP
:
2337 case UDP_ENCAP_ESPINUDP_NON_IKE
:
2338 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
2340 case UDP_ENCAP_L2TPINUDP
:
2341 up
->encap_type
= val
;
2350 case UDP_NO_CHECK6_TX
:
2351 up
->no_check6_tx
= valbool
;
2354 case UDP_NO_CHECK6_RX
:
2355 up
->no_check6_rx
= valbool
;
2359 * UDP-Lite's partial checksum coverage (RFC 3828).
2361 /* The sender sets actual checksum coverage length via this option.
2362 * The case coverage > packet length is handled by send module. */
2363 case UDPLITE_SEND_CSCOV
:
2364 if (!is_udplite
) /* Disable the option on UDP sockets */
2365 return -ENOPROTOOPT
;
2366 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2368 else if (val
> USHRT_MAX
)
2371 up
->pcflag
|= UDPLITE_SEND_CC
;
2374 /* The receiver specifies a minimum checksum coverage value. To make
2375 * sense, this should be set to at least 8 (as done below). If zero is
2376 * used, this again means full checksum coverage. */
2377 case UDPLITE_RECV_CSCOV
:
2378 if (!is_udplite
) /* Disable the option on UDP sockets */
2379 return -ENOPROTOOPT
;
2380 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2382 else if (val
> USHRT_MAX
)
2385 up
->pcflag
|= UDPLITE_RECV_CC
;
2395 EXPORT_SYMBOL(udp_lib_setsockopt
);
2397 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2398 char __user
*optval
, unsigned int optlen
)
2400 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2401 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2402 udp_push_pending_frames
);
2403 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2406 #ifdef CONFIG_COMPAT
2407 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2408 char __user
*optval
, unsigned int optlen
)
2410 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2411 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2412 udp_push_pending_frames
);
2413 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2417 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2418 char __user
*optval
, int __user
*optlen
)
2420 struct udp_sock
*up
= udp_sk(sk
);
2423 if (get_user(len
, optlen
))
2426 len
= min_t(unsigned int, len
, sizeof(int));
2437 val
= up
->encap_type
;
2440 case UDP_NO_CHECK6_TX
:
2441 val
= up
->no_check6_tx
;
2444 case UDP_NO_CHECK6_RX
:
2445 val
= up
->no_check6_rx
;
2448 /* The following two cannot be changed on UDP sockets, the return is
2449 * always 0 (which corresponds to the full checksum coverage of UDP). */
2450 case UDPLITE_SEND_CSCOV
:
2454 case UDPLITE_RECV_CSCOV
:
2459 return -ENOPROTOOPT
;
2462 if (put_user(len
, optlen
))
2464 if (copy_to_user(optval
, &val
, len
))
2468 EXPORT_SYMBOL(udp_lib_getsockopt
);
2470 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2471 char __user
*optval
, int __user
*optlen
)
2473 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2474 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2475 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2478 #ifdef CONFIG_COMPAT
2479 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2480 char __user
*optval
, int __user
*optlen
)
2482 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2483 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2484 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2488 * udp_poll - wait for a UDP event.
2489 * @file - file struct
2491 * @wait - poll table
2493 * This is same as datagram poll, except for the special case of
2494 * blocking sockets. If application is using a blocking fd
2495 * and a packet with checksum error is in the queue;
2496 * then it could get return from select indicating data available
2497 * but then block when reading it. Add special case code
2498 * to work around these arguably broken applications.
2500 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2502 unsigned int mask
= datagram_poll(file
, sock
, wait
);
2503 struct sock
*sk
= sock
->sk
;
2505 if (!skb_queue_empty(&udp_sk(sk
)->reader_queue
))
2506 mask
|= POLLIN
| POLLRDNORM
;
2508 sock_rps_record_flow(sk
);
2510 /* Check for false positives due to checksum errors */
2511 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2512 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && first_packet_length(sk
) == -1)
2513 mask
&= ~(POLLIN
| POLLRDNORM
);
2518 EXPORT_SYMBOL(udp_poll
);
2520 int udp_abort(struct sock
*sk
, int err
)
2525 sk
->sk_error_report(sk
);
2526 __udp_disconnect(sk
, 0);
2532 EXPORT_SYMBOL_GPL(udp_abort
);
2534 struct proto udp_prot
= {
2536 .owner
= THIS_MODULE
,
2537 .close
= udp_lib_close
,
2538 .connect
= ip4_datagram_connect
,
2539 .disconnect
= udp_disconnect
,
2541 .init
= udp_init_sock
,
2542 .destroy
= udp_destroy_sock
,
2543 .setsockopt
= udp_setsockopt
,
2544 .getsockopt
= udp_getsockopt
,
2545 .sendmsg
= udp_sendmsg
,
2546 .recvmsg
= udp_recvmsg
,
2547 .sendpage
= udp_sendpage
,
2548 .release_cb
= ip4_datagram_release_cb
,
2549 .hash
= udp_lib_hash
,
2550 .unhash
= udp_lib_unhash
,
2551 .rehash
= udp_v4_rehash
,
2552 .get_port
= udp_v4_get_port
,
2553 .memory_allocated
= &udp_memory_allocated
,
2554 .sysctl_mem
= sysctl_udp_mem
,
2555 .sysctl_wmem
= &sysctl_udp_wmem_min
,
2556 .sysctl_rmem
= &sysctl_udp_rmem_min
,
2557 .obj_size
= sizeof(struct udp_sock
),
2558 .h
.udp_table
= &udp_table
,
2559 #ifdef CONFIG_COMPAT
2560 .compat_setsockopt
= compat_udp_setsockopt
,
2561 .compat_getsockopt
= compat_udp_getsockopt
,
2563 .diag_destroy
= udp_abort
,
2565 EXPORT_SYMBOL(udp_prot
);
2567 /* ------------------------------------------------------------------------ */
2568 #ifdef CONFIG_PROC_FS
2570 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2573 struct udp_iter_state
*state
= seq
->private;
2574 struct net
*net
= seq_file_net(seq
);
2576 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
2578 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
2580 if (hlist_empty(&hslot
->head
))
2583 spin_lock_bh(&hslot
->lock
);
2584 sk_for_each(sk
, &hslot
->head
) {
2585 if (!net_eq(sock_net(sk
), net
))
2587 if (sk
->sk_family
== state
->family
)
2590 spin_unlock_bh(&hslot
->lock
);
2597 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2599 struct udp_iter_state
*state
= seq
->private;
2600 struct net
*net
= seq_file_net(seq
);
2604 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
2607 if (state
->bucket
<= state
->udp_table
->mask
)
2608 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2609 return udp_get_first(seq
, state
->bucket
+ 1);
2614 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2616 struct sock
*sk
= udp_get_first(seq
, 0);
2619 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2621 return pos
? NULL
: sk
;
2624 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2626 struct udp_iter_state
*state
= seq
->private;
2627 state
->bucket
= MAX_UDP_PORTS
;
2629 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2632 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2636 if (v
== SEQ_START_TOKEN
)
2637 sk
= udp_get_idx(seq
, 0);
2639 sk
= udp_get_next(seq
, v
);
2645 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
2647 struct udp_iter_state
*state
= seq
->private;
2649 if (state
->bucket
<= state
->udp_table
->mask
)
2650 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2653 int udp_seq_open(struct inode
*inode
, struct file
*file
)
2655 struct udp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2656 struct udp_iter_state
*s
;
2659 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2660 sizeof(struct udp_iter_state
));
2664 s
= ((struct seq_file
*)file
->private_data
)->private;
2665 s
->family
= afinfo
->family
;
2666 s
->udp_table
= afinfo
->udp_table
;
2669 EXPORT_SYMBOL(udp_seq_open
);
2671 /* ------------------------------------------------------------------------ */
2672 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2674 struct proc_dir_entry
*p
;
2677 afinfo
->seq_ops
.start
= udp_seq_start
;
2678 afinfo
->seq_ops
.next
= udp_seq_next
;
2679 afinfo
->seq_ops
.stop
= udp_seq_stop
;
2681 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2682 afinfo
->seq_fops
, afinfo
);
2687 EXPORT_SYMBOL(udp_proc_register
);
2689 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2691 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2693 EXPORT_SYMBOL(udp_proc_unregister
);
2695 /* ------------------------------------------------------------------------ */
2696 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2699 struct inet_sock
*inet
= inet_sk(sp
);
2700 __be32 dest
= inet
->inet_daddr
;
2701 __be32 src
= inet
->inet_rcv_saddr
;
2702 __u16 destp
= ntohs(inet
->inet_dport
);
2703 __u16 srcp
= ntohs(inet
->inet_sport
);
2705 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2706 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2707 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2708 sk_wmem_alloc_get(sp
),
2709 sk_rmem_alloc_get(sp
),
2711 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2713 refcount_read(&sp
->sk_refcnt
), sp
,
2714 atomic_read(&sp
->sk_drops
));
2717 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2719 seq_setwidth(seq
, 127);
2720 if (v
== SEQ_START_TOKEN
)
2721 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2722 "rx_queue tr tm->when retrnsmt uid timeout "
2723 "inode ref pointer drops");
2725 struct udp_iter_state
*state
= seq
->private;
2727 udp4_format_sock(v
, seq
, state
->bucket
);
2733 static const struct file_operations udp_afinfo_seq_fops
= {
2734 .owner
= THIS_MODULE
,
2735 .open
= udp_seq_open
,
2737 .llseek
= seq_lseek
,
2738 .release
= seq_release_net
2741 /* ------------------------------------------------------------------------ */
2742 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2745 .udp_table
= &udp_table
,
2746 .seq_fops
= &udp_afinfo_seq_fops
,
2748 .show
= udp4_seq_show
,
2752 static int __net_init
udp4_proc_init_net(struct net
*net
)
2754 return udp_proc_register(net
, &udp4_seq_afinfo
);
2757 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2759 udp_proc_unregister(net
, &udp4_seq_afinfo
);
2762 static struct pernet_operations udp4_net_ops
= {
2763 .init
= udp4_proc_init_net
,
2764 .exit
= udp4_proc_exit_net
,
2767 int __init
udp4_proc_init(void)
2769 return register_pernet_subsys(&udp4_net_ops
);
2772 void udp4_proc_exit(void)
2774 unregister_pernet_subsys(&udp4_net_ops
);
2776 #endif /* CONFIG_PROC_FS */
2778 static __initdata
unsigned long uhash_entries
;
2779 static int __init
set_uhash_entries(char *str
)
2786 ret
= kstrtoul(str
, 0, &uhash_entries
);
2790 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2791 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2794 __setup("uhash_entries=", set_uhash_entries
);
2796 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2800 table
->hash
= alloc_large_system_hash(name
,
2801 2 * sizeof(struct udp_hslot
),
2803 21, /* one slot per 2 MB */
2807 UDP_HTABLE_SIZE_MIN
,
2810 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2811 for (i
= 0; i
<= table
->mask
; i
++) {
2812 INIT_HLIST_HEAD(&table
->hash
[i
].head
);
2813 table
->hash
[i
].count
= 0;
2814 spin_lock_init(&table
->hash
[i
].lock
);
2816 for (i
= 0; i
<= table
->mask
; i
++) {
2817 INIT_HLIST_HEAD(&table
->hash2
[i
].head
);
2818 table
->hash2
[i
].count
= 0;
2819 spin_lock_init(&table
->hash2
[i
].lock
);
2823 u32
udp_flow_hashrnd(void)
2825 static u32 hashrnd __read_mostly
;
2827 net_get_random_once(&hashrnd
, sizeof(hashrnd
));
2831 EXPORT_SYMBOL(udp_flow_hashrnd
);
2833 void __init
udp_init(void)
2835 unsigned long limit
;
2838 udp_table_init(&udp_table
, "UDP");
2839 limit
= nr_free_buffer_pages() / 8;
2840 limit
= max(limit
, 128UL);
2841 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2842 sysctl_udp_mem
[1] = limit
;
2843 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2845 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2846 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2848 /* 16 spinlocks per cpu */
2849 udp_busylocks_log
= ilog2(nr_cpu_ids
) + 4;
2850 udp_busylocks
= kmalloc(sizeof(spinlock_t
) << udp_busylocks_log
,
2853 panic("UDP: failed to alloc udp_busylocks\n");
2854 for (i
= 0; i
< (1U << udp_busylocks_log
); i
++)
2855 spin_lock_init(udp_busylocks
+ i
);