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 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
128 #include <linux/filter.h>
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
138 static struct lock_class_key af_family_keys[AF_MAX];
139 static struct lock_class_key af_family_slock_keys[AF_MAX];
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
147 static const char *af_family_key_strings[AF_MAX+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
171 "slock-27" , "slock-28" , "slock-AF_CAN" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-29" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_MAX"
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
195 static struct lock_class_key af_callback_keys[AF_MAX];
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
220 if (optlen < sizeof(tv))
222 if (copy_from_user(&tv, optval, sizeof(tv)))
224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
228 static int warned __read_mostly;
231 if (warned < 10 && net_ratelimit())
233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
235 current->comm, task_pid_nr(current));
238 *timeo_p = MAX_SCHEDULE_TIMEOUT;
239 if (tv.tv_sec == 0 && tv.tv_usec == 0)
241 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
242 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
246 static void sock_warn_obsolete_bsdism(const char *name)
249 static char warncomm[TASK_COMM_LEN];
250 if (strcmp(warncomm, current->comm) && warned < 5) {
251 strcpy(warncomm, current->comm);
252 printk(KERN_WARNING "process `%s' is using obsolete "
253 "%s SO_BSDCOMPAT\n", warncomm, name);
258 static void sock_disable_timestamp(struct sock *sk)
260 if (sock_flag(sk, SOCK_TIMESTAMP)) {
261 sock_reset_flag(sk, SOCK_TIMESTAMP);
262 net_disable_timestamp();
267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
272 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
273 number of warnings when compiling with -W --ANK
275 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
276 (unsigned)sk->sk_rcvbuf) {
281 err = sk_filter(sk, skb);
285 if (!sk_rmem_schedule(sk, skb->truesize)) {
291 skb_set_owner_r(skb, sk);
293 /* Cache the SKB length before we tack it onto the receive
294 * queue. Once it is added it no longer belongs to us and
295 * may be freed by other threads of control pulling packets
300 skb_queue_tail(&sk->sk_receive_queue, skb);
302 if (!sock_flag(sk, SOCK_DEAD))
303 sk->sk_data_ready(sk, skb_len);
307 EXPORT_SYMBOL(sock_queue_rcv_skb);
309 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
311 int rc = NET_RX_SUCCESS;
313 if (sk_filter(sk, skb))
314 goto discard_and_relse;
319 bh_lock_sock_nested(sk);
322 if (!sock_owned_by_user(sk)) {
324 * trylock + unlock semantics:
326 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
328 rc = sk->sk_backlog_rcv(sk, skb);
330 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
332 sk_add_backlog(sk, skb);
341 EXPORT_SYMBOL(sk_receive_skb);
343 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
345 struct dst_entry *dst = sk->sk_dst_cache;
347 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
348 sk->sk_dst_cache = NULL;
355 EXPORT_SYMBOL(__sk_dst_check);
357 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
359 struct dst_entry *dst = sk_dst_get(sk);
361 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
369 EXPORT_SYMBOL(sk_dst_check);
371 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
373 int ret = -ENOPROTOOPT;
374 #ifdef CONFIG_NETDEVICES
375 struct net *net = sock_net(sk);
376 char devname[IFNAMSIZ];
381 if (!capable(CAP_NET_RAW))
388 /* Bind this socket to a particular device like "eth0",
389 * as specified in the passed interface name. If the
390 * name is "" or the option length is zero the socket
393 if (optlen > IFNAMSIZ - 1)
394 optlen = IFNAMSIZ - 1;
395 memset(devname, 0, sizeof(devname));
398 if (copy_from_user(devname, optval, optlen))
401 if (devname[0] == '\0') {
404 struct net_device *dev = dev_get_by_name(net, devname);
410 index = dev->ifindex;
415 sk->sk_bound_dev_if = index;
427 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
430 sock_set_flag(sk, bit);
432 sock_reset_flag(sk, bit);
436 * This is meant for all protocols to use and covers goings on
437 * at the socket level. Everything here is generic.
440 int sock_setsockopt(struct socket *sock, int level, int optname,
441 char __user *optval, int optlen)
443 struct sock *sk=sock->sk;
450 * Options without arguments
453 #ifdef SO_DONTLINGER /* Compatibility item... */
454 if (optname == SO_DONTLINGER) {
456 sock_reset_flag(sk, SOCK_LINGER);
462 if (optname == SO_BINDTODEVICE)
463 return sock_bindtodevice(sk, optval, optlen);
465 if (optlen < sizeof(int))
468 if (get_user(val, (int __user *)optval))
477 if (val && !capable(CAP_NET_ADMIN)) {
480 sock_valbool_flag(sk, SOCK_DBG, valbool);
483 sk->sk_reuse = valbool;
490 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
493 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
496 /* Don't error on this BSD doesn't and if you think
497 about it this is right. Otherwise apps have to
498 play 'guess the biggest size' games. RCVBUF/SNDBUF
499 are treated in BSD as hints */
501 if (val > sysctl_wmem_max)
502 val = sysctl_wmem_max;
504 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
505 if ((val * 2) < SOCK_MIN_SNDBUF)
506 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
508 sk->sk_sndbuf = val * 2;
511 * Wake up sending tasks if we
514 sk->sk_write_space(sk);
518 if (!capable(CAP_NET_ADMIN)) {
525 /* Don't error on this BSD doesn't and if you think
526 about it this is right. Otherwise apps have to
527 play 'guess the biggest size' games. RCVBUF/SNDBUF
528 are treated in BSD as hints */
530 if (val > sysctl_rmem_max)
531 val = sysctl_rmem_max;
533 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
535 * We double it on the way in to account for
536 * "struct sk_buff" etc. overhead. Applications
537 * assume that the SO_RCVBUF setting they make will
538 * allow that much actual data to be received on that
541 * Applications are unaware that "struct sk_buff" and
542 * other overheads allocate from the receive buffer
543 * during socket buffer allocation.
545 * And after considering the possible alternatives,
546 * returning the value we actually used in getsockopt
547 * is the most desirable behavior.
549 if ((val * 2) < SOCK_MIN_RCVBUF)
550 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
552 sk->sk_rcvbuf = val * 2;
556 if (!capable(CAP_NET_ADMIN)) {
564 if (sk->sk_protocol == IPPROTO_TCP)
565 tcp_set_keepalive(sk, valbool);
567 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
571 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
575 sk->sk_no_check = valbool;
579 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
580 sk->sk_priority = val;
586 if (optlen < sizeof(ling)) {
587 ret = -EINVAL; /* 1003.1g */
590 if (copy_from_user(&ling,optval,sizeof(ling))) {
595 sock_reset_flag(sk, SOCK_LINGER);
597 #if (BITS_PER_LONG == 32)
598 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
599 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
602 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
603 sock_set_flag(sk, SOCK_LINGER);
608 sock_warn_obsolete_bsdism("setsockopt");
613 set_bit(SOCK_PASSCRED, &sock->flags);
615 clear_bit(SOCK_PASSCRED, &sock->flags);
621 if (optname == SO_TIMESTAMP)
622 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
624 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
625 sock_set_flag(sk, SOCK_RCVTSTAMP);
626 sock_enable_timestamp(sk);
628 sock_reset_flag(sk, SOCK_RCVTSTAMP);
629 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
636 sk->sk_rcvlowat = val ? : 1;
640 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
644 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
647 case SO_ATTACH_FILTER:
649 if (optlen == sizeof(struct sock_fprog)) {
650 struct sock_fprog fprog;
653 if (copy_from_user(&fprog, optval, sizeof(fprog)))
656 ret = sk_attach_filter(&fprog, sk);
660 case SO_DETACH_FILTER:
661 ret = sk_detach_filter(sk);
666 set_bit(SOCK_PASSSEC, &sock->flags);
668 clear_bit(SOCK_PASSSEC, &sock->flags);
671 if (!capable(CAP_NET_ADMIN))
678 /* We implement the SO_SNDLOWAT etc to
679 not be settable (1003.1g 5.3) */
689 int sock_getsockopt(struct socket *sock, int level, int optname,
690 char __user *optval, int __user *optlen)
692 struct sock *sk = sock->sk;
700 unsigned int lv = sizeof(int);
703 if (get_user(len, optlen))
710 v.val = sock_flag(sk, SOCK_DBG);
714 v.val = sock_flag(sk, SOCK_LOCALROUTE);
718 v.val = !!sock_flag(sk, SOCK_BROADCAST);
722 v.val = sk->sk_sndbuf;
726 v.val = sk->sk_rcvbuf;
730 v.val = sk->sk_reuse;
734 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
742 v.val = -sock_error(sk);
744 v.val = xchg(&sk->sk_err_soft, 0);
748 v.val = !!sock_flag(sk, SOCK_URGINLINE);
752 v.val = sk->sk_no_check;
756 v.val = sk->sk_priority;
761 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
762 v.ling.l_linger = sk->sk_lingertime / HZ;
766 sock_warn_obsolete_bsdism("getsockopt");
770 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
771 !sock_flag(sk, SOCK_RCVTSTAMPNS);
775 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
779 lv=sizeof(struct timeval);
780 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
784 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
785 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
790 lv=sizeof(struct timeval);
791 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
795 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
796 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
801 v.val = sk->sk_rcvlowat;
809 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
813 if (len > sizeof(sk->sk_peercred))
814 len = sizeof(sk->sk_peercred);
815 if (copy_to_user(optval, &sk->sk_peercred, len))
823 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
827 if (copy_to_user(optval, address, len))
832 /* Dubious BSD thing... Probably nobody even uses it, but
833 * the UNIX standard wants it for whatever reason... -DaveM
836 v.val = sk->sk_state == TCP_LISTEN;
840 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
844 return security_socket_getpeersec_stream(sock, optval, optlen, len);
856 if (copy_to_user(optval, &v, len))
859 if (put_user(len, optlen))
865 * Initialize an sk_lock.
867 * (We also register the sk_lock with the lock validator.)
869 static inline void sock_lock_init(struct sock *sk)
871 sock_lock_init_class_and_name(sk,
872 af_family_slock_key_strings[sk->sk_family],
873 af_family_slock_keys + sk->sk_family,
874 af_family_key_strings[sk->sk_family],
875 af_family_keys + sk->sk_family);
878 static void sock_copy(struct sock *nsk, const struct sock *osk)
880 #ifdef CONFIG_SECURITY_NETWORK
881 void *sptr = nsk->sk_security;
884 memcpy(nsk, osk, osk->sk_prot->obj_size);
885 #ifdef CONFIG_SECURITY_NETWORK
886 nsk->sk_security = sptr;
887 security_sk_clone(osk, nsk);
891 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
895 struct kmem_cache *slab;
899 sk = kmem_cache_alloc(slab, priority);
901 sk = kmalloc(prot->obj_size, priority);
904 if (security_sk_alloc(sk, family, priority))
907 if (!try_module_get(prot->owner))
914 security_sk_free(sk);
917 kmem_cache_free(slab, sk);
923 static void sk_prot_free(struct proto *prot, struct sock *sk)
925 struct kmem_cache *slab;
926 struct module *owner;
931 security_sk_free(sk);
933 kmem_cache_free(slab, sk);
940 * sk_alloc - All socket objects are allocated here
941 * @net: the applicable net namespace
942 * @family: protocol family
943 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
944 * @prot: struct proto associated with this new sock instance
946 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
951 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
953 sk->sk_family = family;
955 * See comment in struct sock definition to understand
956 * why we need sk_prot_creator -acme
958 sk->sk_prot = sk->sk_prot_creator = prot;
960 sock_net_set(sk, get_net(net));
966 void sk_free(struct sock *sk)
968 struct sk_filter *filter;
973 filter = rcu_dereference(sk->sk_filter);
975 sk_filter_uncharge(sk, filter);
976 rcu_assign_pointer(sk->sk_filter, NULL);
979 sock_disable_timestamp(sk);
981 if (atomic_read(&sk->sk_omem_alloc))
982 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
983 __func__, atomic_read(&sk->sk_omem_alloc));
985 put_net(sock_net(sk));
986 sk_prot_free(sk->sk_prot_creator, sk);
990 * Last sock_put should drop referrence to sk->sk_net. It has already
991 * been dropped in sk_change_net. Taking referrence to stopping namespace
993 * Take referrence to a socket to remove it from hash _alive_ and after that
994 * destroy it in the context of init_net.
996 void sk_release_kernel(struct sock *sk)
998 if (sk == NULL || sk->sk_socket == NULL)
1002 sock_release(sk->sk_socket);
1003 release_net(sock_net(sk));
1004 sock_net_set(sk, get_net(&init_net));
1007 EXPORT_SYMBOL(sk_release_kernel);
1009 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1013 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1014 if (newsk != NULL) {
1015 struct sk_filter *filter;
1017 sock_copy(newsk, sk);
1020 get_net(sock_net(newsk));
1021 sk_node_init(&newsk->sk_node);
1022 sock_lock_init(newsk);
1023 bh_lock_sock(newsk);
1024 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1026 atomic_set(&newsk->sk_rmem_alloc, 0);
1027 atomic_set(&newsk->sk_wmem_alloc, 0);
1028 atomic_set(&newsk->sk_omem_alloc, 0);
1029 skb_queue_head_init(&newsk->sk_receive_queue);
1030 skb_queue_head_init(&newsk->sk_write_queue);
1031 #ifdef CONFIG_NET_DMA
1032 skb_queue_head_init(&newsk->sk_async_wait_queue);
1035 rwlock_init(&newsk->sk_dst_lock);
1036 rwlock_init(&newsk->sk_callback_lock);
1037 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1038 af_callback_keys + newsk->sk_family,
1039 af_family_clock_key_strings[newsk->sk_family]);
1041 newsk->sk_dst_cache = NULL;
1042 newsk->sk_wmem_queued = 0;
1043 newsk->sk_forward_alloc = 0;
1044 newsk->sk_send_head = NULL;
1045 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1047 sock_reset_flag(newsk, SOCK_DONE);
1048 skb_queue_head_init(&newsk->sk_error_queue);
1050 filter = newsk->sk_filter;
1052 sk_filter_charge(newsk, filter);
1054 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1055 /* It is still raw copy of parent, so invalidate
1056 * destructor and make plain sk_free() */
1057 newsk->sk_destruct = NULL;
1064 newsk->sk_priority = 0;
1065 atomic_set(&newsk->sk_refcnt, 2);
1068 * Increment the counter in the same struct proto as the master
1069 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1070 * is the same as sk->sk_prot->socks, as this field was copied
1073 * This _changes_ the previous behaviour, where
1074 * tcp_create_openreq_child always was incrementing the
1075 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1076 * to be taken into account in all callers. -acme
1078 sk_refcnt_debug_inc(newsk);
1079 newsk->sk_socket = NULL;
1080 newsk->sk_sleep = NULL;
1082 if (newsk->sk_prot->sockets_allocated)
1083 atomic_inc(newsk->sk_prot->sockets_allocated);
1089 EXPORT_SYMBOL_GPL(sk_clone);
1091 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1093 __sk_dst_set(sk, dst);
1094 sk->sk_route_caps = dst->dev->features;
1095 if (sk->sk_route_caps & NETIF_F_GSO)
1096 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1097 if (sk_can_gso(sk)) {
1098 if (dst->header_len) {
1099 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1101 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1102 sk->sk_gso_max_size = dst->dev->gso_max_size;
1106 EXPORT_SYMBOL_GPL(sk_setup_caps);
1108 void __init sk_init(void)
1110 if (num_physpages <= 4096) {
1111 sysctl_wmem_max = 32767;
1112 sysctl_rmem_max = 32767;
1113 sysctl_wmem_default = 32767;
1114 sysctl_rmem_default = 32767;
1115 } else if (num_physpages >= 131072) {
1116 sysctl_wmem_max = 131071;
1117 sysctl_rmem_max = 131071;
1122 * Simple resource managers for sockets.
1127 * Write buffer destructor automatically called from kfree_skb.
1129 void sock_wfree(struct sk_buff *skb)
1131 struct sock *sk = skb->sk;
1133 /* In case it might be waiting for more memory. */
1134 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1135 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1136 sk->sk_write_space(sk);
1141 * Read buffer destructor automatically called from kfree_skb.
1143 void sock_rfree(struct sk_buff *skb)
1145 struct sock *sk = skb->sk;
1147 skb_truesize_check(skb);
1148 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1149 sk_mem_uncharge(skb->sk, skb->truesize);
1153 int sock_i_uid(struct sock *sk)
1157 read_lock(&sk->sk_callback_lock);
1158 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1159 read_unlock(&sk->sk_callback_lock);
1163 unsigned long sock_i_ino(struct sock *sk)
1167 read_lock(&sk->sk_callback_lock);
1168 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1169 read_unlock(&sk->sk_callback_lock);
1174 * Allocate a skb from the socket's send buffer.
1176 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1179 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1180 struct sk_buff * skb = alloc_skb(size, priority);
1182 skb_set_owner_w(skb, sk);
1190 * Allocate a skb from the socket's receive buffer.
1192 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1195 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1196 struct sk_buff *skb = alloc_skb(size, priority);
1198 skb_set_owner_r(skb, sk);
1206 * Allocate a memory block from the socket's option memory buffer.
1208 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1210 if ((unsigned)size <= sysctl_optmem_max &&
1211 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1213 /* First do the add, to avoid the race if kmalloc
1216 atomic_add(size, &sk->sk_omem_alloc);
1217 mem = kmalloc(size, priority);
1220 atomic_sub(size, &sk->sk_omem_alloc);
1226 * Free an option memory block.
1228 void sock_kfree_s(struct sock *sk, void *mem, int size)
1231 atomic_sub(size, &sk->sk_omem_alloc);
1234 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1235 I think, these locks should be removed for datagram sockets.
1237 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1241 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1245 if (signal_pending(current))
1247 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1248 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1249 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1251 if (sk->sk_shutdown & SEND_SHUTDOWN)
1255 timeo = schedule_timeout(timeo);
1257 finish_wait(sk->sk_sleep, &wait);
1263 * Generic send/receive buffer handlers
1266 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1267 unsigned long header_len,
1268 unsigned long data_len,
1269 int noblock, int *errcode)
1271 struct sk_buff *skb;
1276 gfp_mask = sk->sk_allocation;
1277 if (gfp_mask & __GFP_WAIT)
1278 gfp_mask |= __GFP_REPEAT;
1280 timeo = sock_sndtimeo(sk, noblock);
1282 err = sock_error(sk);
1287 if (sk->sk_shutdown & SEND_SHUTDOWN)
1290 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1291 skb = alloc_skb(header_len, gfp_mask);
1296 /* No pages, we're done... */
1300 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1301 skb->truesize += data_len;
1302 skb_shinfo(skb)->nr_frags = npages;
1303 for (i = 0; i < npages; i++) {
1307 page = alloc_pages(sk->sk_allocation, 0);
1310 skb_shinfo(skb)->nr_frags = i;
1315 frag = &skb_shinfo(skb)->frags[i];
1317 frag->page_offset = 0;
1318 frag->size = (data_len >= PAGE_SIZE ?
1321 data_len -= PAGE_SIZE;
1324 /* Full success... */
1330 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1331 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1335 if (signal_pending(current))
1337 timeo = sock_wait_for_wmem(sk, timeo);
1340 skb_set_owner_w(skb, sk);
1344 err = sock_intr_errno(timeo);
1350 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1351 int noblock, int *errcode)
1353 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1356 static void __lock_sock(struct sock *sk)
1361 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1362 TASK_UNINTERRUPTIBLE);
1363 spin_unlock_bh(&sk->sk_lock.slock);
1365 spin_lock_bh(&sk->sk_lock.slock);
1366 if (!sock_owned_by_user(sk))
1369 finish_wait(&sk->sk_lock.wq, &wait);
1372 static void __release_sock(struct sock *sk)
1374 struct sk_buff *skb = sk->sk_backlog.head;
1377 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1381 struct sk_buff *next = skb->next;
1384 sk->sk_backlog_rcv(sk, skb);
1387 * We are in process context here with softirqs
1388 * disabled, use cond_resched_softirq() to preempt.
1389 * This is safe to do because we've taken the backlog
1392 cond_resched_softirq();
1395 } while (skb != NULL);
1398 } while ((skb = sk->sk_backlog.head) != NULL);
1402 * sk_wait_data - wait for data to arrive at sk_receive_queue
1403 * @sk: sock to wait on
1404 * @timeo: for how long
1406 * Now socket state including sk->sk_err is changed only under lock,
1407 * hence we may omit checks after joining wait queue.
1408 * We check receive queue before schedule() only as optimization;
1409 * it is very likely that release_sock() added new data.
1411 int sk_wait_data(struct sock *sk, long *timeo)
1416 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1417 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1418 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1419 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1420 finish_wait(sk->sk_sleep, &wait);
1424 EXPORT_SYMBOL(sk_wait_data);
1427 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1429 * @size: memory size to allocate
1430 * @kind: allocation type
1432 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1433 * rmem allocation. This function assumes that protocols which have
1434 * memory_pressure use sk_wmem_queued as write buffer accounting.
1436 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1438 struct proto *prot = sk->sk_prot;
1439 int amt = sk_mem_pages(size);
1442 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1443 allocated = atomic_add_return(amt, prot->memory_allocated);
1446 if (allocated <= prot->sysctl_mem[0]) {
1447 if (prot->memory_pressure && *prot->memory_pressure)
1448 *prot->memory_pressure = 0;
1452 /* Under pressure. */
1453 if (allocated > prot->sysctl_mem[1])
1454 if (prot->enter_memory_pressure)
1455 prot->enter_memory_pressure();
1457 /* Over hard limit. */
1458 if (allocated > prot->sysctl_mem[2])
1459 goto suppress_allocation;
1461 /* guarantee minimum buffer size under pressure */
1462 if (kind == SK_MEM_RECV) {
1463 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1465 } else { /* SK_MEM_SEND */
1466 if (sk->sk_type == SOCK_STREAM) {
1467 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1469 } else if (atomic_read(&sk->sk_wmem_alloc) <
1470 prot->sysctl_wmem[0])
1474 if (prot->memory_pressure) {
1475 if (!*prot->memory_pressure ||
1476 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1477 sk_mem_pages(sk->sk_wmem_queued +
1478 atomic_read(&sk->sk_rmem_alloc) +
1479 sk->sk_forward_alloc))
1483 suppress_allocation:
1485 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1486 sk_stream_moderate_sndbuf(sk);
1488 /* Fail only if socket is _under_ its sndbuf.
1489 * In this case we cannot block, so that we have to fail.
1491 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1495 /* Alas. Undo changes. */
1496 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1497 atomic_sub(amt, prot->memory_allocated);
1501 EXPORT_SYMBOL(__sk_mem_schedule);
1504 * __sk_reclaim - reclaim memory_allocated
1507 void __sk_mem_reclaim(struct sock *sk)
1509 struct proto *prot = sk->sk_prot;
1511 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1512 prot->memory_allocated);
1513 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1515 if (prot->memory_pressure && *prot->memory_pressure &&
1516 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1517 *prot->memory_pressure = 0;
1520 EXPORT_SYMBOL(__sk_mem_reclaim);
1524 * Set of default routines for initialising struct proto_ops when
1525 * the protocol does not support a particular function. In certain
1526 * cases where it makes no sense for a protocol to have a "do nothing"
1527 * function, some default processing is provided.
1530 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1535 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1541 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1546 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1551 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1557 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1562 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1567 int sock_no_listen(struct socket *sock, int backlog)
1572 int sock_no_shutdown(struct socket *sock, int how)
1577 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1578 char __user *optval, int optlen)
1583 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1584 char __user *optval, int __user *optlen)
1589 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1595 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1596 size_t len, int flags)
1601 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1603 /* Mirror missing mmap method error code */
1607 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1610 struct msghdr msg = {.msg_flags = flags};
1612 char *kaddr = kmap(page);
1613 iov.iov_base = kaddr + offset;
1615 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1621 * Default Socket Callbacks
1624 static void sock_def_wakeup(struct sock *sk)
1626 read_lock(&sk->sk_callback_lock);
1627 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1628 wake_up_interruptible_all(sk->sk_sleep);
1629 read_unlock(&sk->sk_callback_lock);
1632 static void sock_def_error_report(struct sock *sk)
1634 read_lock(&sk->sk_callback_lock);
1635 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1636 wake_up_interruptible(sk->sk_sleep);
1637 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1638 read_unlock(&sk->sk_callback_lock);
1641 static void sock_def_readable(struct sock *sk, int len)
1643 read_lock(&sk->sk_callback_lock);
1644 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1645 wake_up_interruptible_sync(sk->sk_sleep);
1646 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1647 read_unlock(&sk->sk_callback_lock);
1650 static void sock_def_write_space(struct sock *sk)
1652 read_lock(&sk->sk_callback_lock);
1654 /* Do not wake up a writer until he can make "significant"
1657 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1658 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1659 wake_up_interruptible_sync(sk->sk_sleep);
1661 /* Should agree with poll, otherwise some programs break */
1662 if (sock_writeable(sk))
1663 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1666 read_unlock(&sk->sk_callback_lock);
1669 static void sock_def_destruct(struct sock *sk)
1671 kfree(sk->sk_protinfo);
1674 void sk_send_sigurg(struct sock *sk)
1676 if (sk->sk_socket && sk->sk_socket->file)
1677 if (send_sigurg(&sk->sk_socket->file->f_owner))
1678 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1681 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1682 unsigned long expires)
1684 if (!mod_timer(timer, expires))
1688 EXPORT_SYMBOL(sk_reset_timer);
1690 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1692 if (timer_pending(timer) && del_timer(timer))
1696 EXPORT_SYMBOL(sk_stop_timer);
1698 void sock_init_data(struct socket *sock, struct sock *sk)
1700 skb_queue_head_init(&sk->sk_receive_queue);
1701 skb_queue_head_init(&sk->sk_write_queue);
1702 skb_queue_head_init(&sk->sk_error_queue);
1703 #ifdef CONFIG_NET_DMA
1704 skb_queue_head_init(&sk->sk_async_wait_queue);
1707 sk->sk_send_head = NULL;
1709 init_timer(&sk->sk_timer);
1711 sk->sk_allocation = GFP_KERNEL;
1712 sk->sk_rcvbuf = sysctl_rmem_default;
1713 sk->sk_sndbuf = sysctl_wmem_default;
1714 sk->sk_state = TCP_CLOSE;
1715 sk->sk_socket = sock;
1717 sock_set_flag(sk, SOCK_ZAPPED);
1720 sk->sk_type = sock->type;
1721 sk->sk_sleep = &sock->wait;
1724 sk->sk_sleep = NULL;
1726 rwlock_init(&sk->sk_dst_lock);
1727 rwlock_init(&sk->sk_callback_lock);
1728 lockdep_set_class_and_name(&sk->sk_callback_lock,
1729 af_callback_keys + sk->sk_family,
1730 af_family_clock_key_strings[sk->sk_family]);
1732 sk->sk_state_change = sock_def_wakeup;
1733 sk->sk_data_ready = sock_def_readable;
1734 sk->sk_write_space = sock_def_write_space;
1735 sk->sk_error_report = sock_def_error_report;
1736 sk->sk_destruct = sock_def_destruct;
1738 sk->sk_sndmsg_page = NULL;
1739 sk->sk_sndmsg_off = 0;
1741 sk->sk_peercred.pid = 0;
1742 sk->sk_peercred.uid = -1;
1743 sk->sk_peercred.gid = -1;
1744 sk->sk_write_pending = 0;
1745 sk->sk_rcvlowat = 1;
1746 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1747 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1749 sk->sk_stamp = ktime_set(-1L, 0);
1751 atomic_set(&sk->sk_refcnt, 1);
1752 atomic_set(&sk->sk_drops, 0);
1755 void lock_sock_nested(struct sock *sk, int subclass)
1758 spin_lock_bh(&sk->sk_lock.slock);
1759 if (sk->sk_lock.owned)
1761 sk->sk_lock.owned = 1;
1762 spin_unlock(&sk->sk_lock.slock);
1764 * The sk_lock has mutex_lock() semantics here:
1766 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1770 EXPORT_SYMBOL(lock_sock_nested);
1772 void release_sock(struct sock *sk)
1775 * The sk_lock has mutex_unlock() semantics:
1777 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1779 spin_lock_bh(&sk->sk_lock.slock);
1780 if (sk->sk_backlog.tail)
1782 sk->sk_lock.owned = 0;
1783 if (waitqueue_active(&sk->sk_lock.wq))
1784 wake_up(&sk->sk_lock.wq);
1785 spin_unlock_bh(&sk->sk_lock.slock);
1787 EXPORT_SYMBOL(release_sock);
1789 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1792 if (!sock_flag(sk, SOCK_TIMESTAMP))
1793 sock_enable_timestamp(sk);
1794 tv = ktime_to_timeval(sk->sk_stamp);
1795 if (tv.tv_sec == -1)
1797 if (tv.tv_sec == 0) {
1798 sk->sk_stamp = ktime_get_real();
1799 tv = ktime_to_timeval(sk->sk_stamp);
1801 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1803 EXPORT_SYMBOL(sock_get_timestamp);
1805 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1808 if (!sock_flag(sk, SOCK_TIMESTAMP))
1809 sock_enable_timestamp(sk);
1810 ts = ktime_to_timespec(sk->sk_stamp);
1811 if (ts.tv_sec == -1)
1813 if (ts.tv_sec == 0) {
1814 sk->sk_stamp = ktime_get_real();
1815 ts = ktime_to_timespec(sk->sk_stamp);
1817 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1819 EXPORT_SYMBOL(sock_get_timestampns);
1821 void sock_enable_timestamp(struct sock *sk)
1823 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1824 sock_set_flag(sk, SOCK_TIMESTAMP);
1825 net_enable_timestamp();
1830 * Get a socket option on an socket.
1832 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1833 * asynchronous errors should be reported by getsockopt. We assume
1834 * this means if you specify SO_ERROR (otherwise whats the point of it).
1836 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1837 char __user *optval, int __user *optlen)
1839 struct sock *sk = sock->sk;
1841 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1844 EXPORT_SYMBOL(sock_common_getsockopt);
1846 #ifdef CONFIG_COMPAT
1847 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1848 char __user *optval, int __user *optlen)
1850 struct sock *sk = sock->sk;
1852 if (sk->sk_prot->compat_getsockopt != NULL)
1853 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1855 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1857 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1860 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1861 struct msghdr *msg, size_t size, int flags)
1863 struct sock *sk = sock->sk;
1867 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1868 flags & ~MSG_DONTWAIT, &addr_len);
1870 msg->msg_namelen = addr_len;
1874 EXPORT_SYMBOL(sock_common_recvmsg);
1877 * Set socket options on an inet socket.
1879 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1880 char __user *optval, int optlen)
1882 struct sock *sk = sock->sk;
1884 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1887 EXPORT_SYMBOL(sock_common_setsockopt);
1889 #ifdef CONFIG_COMPAT
1890 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1891 char __user *optval, int optlen)
1893 struct sock *sk = sock->sk;
1895 if (sk->sk_prot->compat_setsockopt != NULL)
1896 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1898 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1900 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1903 void sk_common_release(struct sock *sk)
1905 if (sk->sk_prot->destroy)
1906 sk->sk_prot->destroy(sk);
1909 * Observation: when sock_common_release is called, processes have
1910 * no access to socket. But net still has.
1911 * Step one, detach it from networking:
1913 * A. Remove from hash tables.
1916 sk->sk_prot->unhash(sk);
1919 * In this point socket cannot receive new packets, but it is possible
1920 * that some packets are in flight because some CPU runs receiver and
1921 * did hash table lookup before we unhashed socket. They will achieve
1922 * receive queue and will be purged by socket destructor.
1924 * Also we still have packets pending on receive queue and probably,
1925 * our own packets waiting in device queues. sock_destroy will drain
1926 * receive queue, but transmitted packets will delay socket destruction
1927 * until the last reference will be released.
1932 xfrm_sk_free_policy(sk);
1934 sk_refcnt_debug_release(sk);
1938 EXPORT_SYMBOL(sk_common_release);
1940 static DEFINE_RWLOCK(proto_list_lock);
1941 static LIST_HEAD(proto_list);
1943 #ifdef CONFIG_PROC_FS
1944 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1946 int val[PROTO_INUSE_NR];
1949 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1951 #ifdef CONFIG_NET_NS
1952 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1954 int cpu = smp_processor_id();
1955 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1957 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1959 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1961 int cpu, idx = prot->inuse_idx;
1964 for_each_possible_cpu(cpu)
1965 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1967 return res >= 0 ? res : 0;
1969 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1971 static int sock_inuse_init_net(struct net *net)
1973 net->core.inuse = alloc_percpu(struct prot_inuse);
1974 return net->core.inuse ? 0 : -ENOMEM;
1977 static void sock_inuse_exit_net(struct net *net)
1979 free_percpu(net->core.inuse);
1982 static struct pernet_operations net_inuse_ops = {
1983 .init = sock_inuse_init_net,
1984 .exit = sock_inuse_exit_net,
1987 static __init int net_inuse_init(void)
1989 if (register_pernet_subsys(&net_inuse_ops))
1990 panic("Cannot initialize net inuse counters");
1995 core_initcall(net_inuse_init);
1997 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1999 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2001 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2003 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2005 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2007 int cpu, idx = prot->inuse_idx;
2010 for_each_possible_cpu(cpu)
2011 res += per_cpu(prot_inuse, cpu).val[idx];
2013 return res >= 0 ? res : 0;
2015 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2018 static void assign_proto_idx(struct proto *prot)
2020 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2022 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2023 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2027 set_bit(prot->inuse_idx, proto_inuse_idx);
2030 static void release_proto_idx(struct proto *prot)
2032 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2033 clear_bit(prot->inuse_idx, proto_inuse_idx);
2036 static inline void assign_proto_idx(struct proto *prot)
2040 static inline void release_proto_idx(struct proto *prot)
2045 int proto_register(struct proto *prot, int alloc_slab)
2047 char *request_sock_slab_name = NULL;
2048 char *timewait_sock_slab_name;
2051 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2052 SLAB_HWCACHE_ALIGN, NULL);
2054 if (prot->slab == NULL) {
2055 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2060 if (prot->rsk_prot != NULL) {
2061 static const char mask[] = "request_sock_%s";
2063 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2064 if (request_sock_slab_name == NULL)
2065 goto out_free_sock_slab;
2067 sprintf(request_sock_slab_name, mask, prot->name);
2068 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
2069 prot->rsk_prot->obj_size, 0,
2070 SLAB_HWCACHE_ALIGN, NULL);
2072 if (prot->rsk_prot->slab == NULL) {
2073 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2075 goto out_free_request_sock_slab_name;
2079 if (prot->twsk_prot != NULL) {
2080 static const char mask[] = "tw_sock_%s";
2082 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2084 if (timewait_sock_slab_name == NULL)
2085 goto out_free_request_sock_slab;
2087 sprintf(timewait_sock_slab_name, mask, prot->name);
2088 prot->twsk_prot->twsk_slab =
2089 kmem_cache_create(timewait_sock_slab_name,
2090 prot->twsk_prot->twsk_obj_size,
2091 0, SLAB_HWCACHE_ALIGN,
2093 if (prot->twsk_prot->twsk_slab == NULL)
2094 goto out_free_timewait_sock_slab_name;
2098 write_lock(&proto_list_lock);
2099 list_add(&prot->node, &proto_list);
2100 assign_proto_idx(prot);
2101 write_unlock(&proto_list_lock);
2104 out_free_timewait_sock_slab_name:
2105 kfree(timewait_sock_slab_name);
2106 out_free_request_sock_slab:
2107 if (prot->rsk_prot && prot->rsk_prot->slab) {
2108 kmem_cache_destroy(prot->rsk_prot->slab);
2109 prot->rsk_prot->slab = NULL;
2111 out_free_request_sock_slab_name:
2112 kfree(request_sock_slab_name);
2114 kmem_cache_destroy(prot->slab);
2120 EXPORT_SYMBOL(proto_register);
2122 void proto_unregister(struct proto *prot)
2124 write_lock(&proto_list_lock);
2125 release_proto_idx(prot);
2126 list_del(&prot->node);
2127 write_unlock(&proto_list_lock);
2129 if (prot->slab != NULL) {
2130 kmem_cache_destroy(prot->slab);
2134 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2135 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2137 kmem_cache_destroy(prot->rsk_prot->slab);
2139 prot->rsk_prot->slab = NULL;
2142 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2143 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
2145 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2147 prot->twsk_prot->twsk_slab = NULL;
2151 EXPORT_SYMBOL(proto_unregister);
2153 #ifdef CONFIG_PROC_FS
2154 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2155 __acquires(proto_list_lock)
2157 read_lock(&proto_list_lock);
2158 return seq_list_start_head(&proto_list, *pos);
2161 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2163 return seq_list_next(v, &proto_list, pos);
2166 static void proto_seq_stop(struct seq_file *seq, void *v)
2167 __releases(proto_list_lock)
2169 read_unlock(&proto_list_lock);
2172 static char proto_method_implemented(const void *method)
2174 return method == NULL ? 'n' : 'y';
2177 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2179 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2180 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2183 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2184 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2185 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2187 proto->slab == NULL ? "no" : "yes",
2188 module_name(proto->owner),
2189 proto_method_implemented(proto->close),
2190 proto_method_implemented(proto->connect),
2191 proto_method_implemented(proto->disconnect),
2192 proto_method_implemented(proto->accept),
2193 proto_method_implemented(proto->ioctl),
2194 proto_method_implemented(proto->init),
2195 proto_method_implemented(proto->destroy),
2196 proto_method_implemented(proto->shutdown),
2197 proto_method_implemented(proto->setsockopt),
2198 proto_method_implemented(proto->getsockopt),
2199 proto_method_implemented(proto->sendmsg),
2200 proto_method_implemented(proto->recvmsg),
2201 proto_method_implemented(proto->sendpage),
2202 proto_method_implemented(proto->bind),
2203 proto_method_implemented(proto->backlog_rcv),
2204 proto_method_implemented(proto->hash),
2205 proto_method_implemented(proto->unhash),
2206 proto_method_implemented(proto->get_port),
2207 proto_method_implemented(proto->enter_memory_pressure));
2210 static int proto_seq_show(struct seq_file *seq, void *v)
2212 if (v == &proto_list)
2213 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2222 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2224 proto_seq_printf(seq, list_entry(v, struct proto, node));
2228 static const struct seq_operations proto_seq_ops = {
2229 .start = proto_seq_start,
2230 .next = proto_seq_next,
2231 .stop = proto_seq_stop,
2232 .show = proto_seq_show,
2235 static int proto_seq_open(struct inode *inode, struct file *file)
2237 return seq_open(file, &proto_seq_ops);
2240 static const struct file_operations proto_seq_fops = {
2241 .owner = THIS_MODULE,
2242 .open = proto_seq_open,
2244 .llseek = seq_lseek,
2245 .release = seq_release,
2248 static int __init proto_init(void)
2250 /* register /proc/net/protocols */
2251 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2254 subsys_initcall(proto_init);
2256 #endif /* PROC_FS */
2258 EXPORT_SYMBOL(sk_alloc);
2259 EXPORT_SYMBOL(sk_free);
2260 EXPORT_SYMBOL(sk_send_sigurg);
2261 EXPORT_SYMBOL(sock_alloc_send_skb);
2262 EXPORT_SYMBOL(sock_init_data);
2263 EXPORT_SYMBOL(sock_kfree_s);
2264 EXPORT_SYMBOL(sock_kmalloc);
2265 EXPORT_SYMBOL(sock_no_accept);
2266 EXPORT_SYMBOL(sock_no_bind);
2267 EXPORT_SYMBOL(sock_no_connect);
2268 EXPORT_SYMBOL(sock_no_getname);
2269 EXPORT_SYMBOL(sock_no_getsockopt);
2270 EXPORT_SYMBOL(sock_no_ioctl);
2271 EXPORT_SYMBOL(sock_no_listen);
2272 EXPORT_SYMBOL(sock_no_mmap);
2273 EXPORT_SYMBOL(sock_no_poll);
2274 EXPORT_SYMBOL(sock_no_recvmsg);
2275 EXPORT_SYMBOL(sock_no_sendmsg);
2276 EXPORT_SYMBOL(sock_no_sendpage);
2277 EXPORT_SYMBOL(sock_no_setsockopt);
2278 EXPORT_SYMBOL(sock_no_shutdown);
2279 EXPORT_SYMBOL(sock_no_socketpair);
2280 EXPORT_SYMBOL(sock_rfree);
2281 EXPORT_SYMBOL(sock_setsockopt);
2282 EXPORT_SYMBOL(sock_wfree);
2283 EXPORT_SYMBOL(sock_wmalloc);
2284 EXPORT_SYMBOL(sock_i_uid);
2285 EXPORT_SYMBOL(sock_i_ino);
2286 EXPORT_SYMBOL(sysctl_optmem_max);