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-29" ,
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-29" ,
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);
286 skb_set_owner_r(skb, sk);
288 /* Cache the SKB length before we tack it onto the receive
289 * queue. Once it is added it no longer belongs to us and
290 * may be freed by other threads of control pulling packets
295 skb_queue_tail(&sk->sk_receive_queue, skb);
297 if (!sock_flag(sk, SOCK_DEAD))
298 sk->sk_data_ready(sk, skb_len);
302 EXPORT_SYMBOL(sock_queue_rcv_skb);
304 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
306 int rc = NET_RX_SUCCESS;
308 if (sk_filter(sk, skb))
309 goto discard_and_relse;
314 bh_lock_sock_nested(sk);
317 if (!sock_owned_by_user(sk)) {
319 * trylock + unlock semantics:
321 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
323 rc = sk->sk_backlog_rcv(sk, skb);
325 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
327 sk_add_backlog(sk, skb);
336 EXPORT_SYMBOL(sk_receive_skb);
338 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
340 struct dst_entry *dst = sk->sk_dst_cache;
342 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
343 sk->sk_dst_cache = NULL;
350 EXPORT_SYMBOL(__sk_dst_check);
352 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
354 struct dst_entry *dst = sk_dst_get(sk);
356 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
364 EXPORT_SYMBOL(sk_dst_check);
366 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
368 int ret = -ENOPROTOOPT;
369 #ifdef CONFIG_NETDEVICES
370 struct net *net = sk->sk_net;
371 char devname[IFNAMSIZ];
376 if (!capable(CAP_NET_RAW))
383 /* Bind this socket to a particular device like "eth0",
384 * as specified in the passed interface name. If the
385 * name is "" or the option length is zero the socket
388 if (optlen > IFNAMSIZ - 1)
389 optlen = IFNAMSIZ - 1;
390 memset(devname, 0, sizeof(devname));
393 if (copy_from_user(devname, optval, optlen))
396 if (devname[0] == '\0') {
399 struct net_device *dev = dev_get_by_name(net, devname);
405 index = dev->ifindex;
410 sk->sk_bound_dev_if = index;
422 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
425 sock_set_flag(sk, bit);
427 sock_reset_flag(sk, bit);
431 * This is meant for all protocols to use and covers goings on
432 * at the socket level. Everything here is generic.
435 int sock_setsockopt(struct socket *sock, int level, int optname,
436 char __user *optval, int optlen)
438 struct sock *sk=sock->sk;
445 * Options without arguments
448 #ifdef SO_DONTLINGER /* Compatibility item... */
449 if (optname == SO_DONTLINGER) {
451 sock_reset_flag(sk, SOCK_LINGER);
457 if (optname == SO_BINDTODEVICE)
458 return sock_bindtodevice(sk, optval, optlen);
460 if (optlen < sizeof(int))
463 if (get_user(val, (int __user *)optval))
472 if (val && !capable(CAP_NET_ADMIN)) {
475 sock_valbool_flag(sk, SOCK_DBG, valbool);
478 sk->sk_reuse = valbool;
485 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
488 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
491 /* Don't error on this BSD doesn't and if you think
492 about it this is right. Otherwise apps have to
493 play 'guess the biggest size' games. RCVBUF/SNDBUF
494 are treated in BSD as hints */
496 if (val > sysctl_wmem_max)
497 val = sysctl_wmem_max;
499 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
500 if ((val * 2) < SOCK_MIN_SNDBUF)
501 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
503 sk->sk_sndbuf = val * 2;
506 * Wake up sending tasks if we
509 sk->sk_write_space(sk);
513 if (!capable(CAP_NET_ADMIN)) {
520 /* Don't error on this BSD doesn't and if you think
521 about it this is right. Otherwise apps have to
522 play 'guess the biggest size' games. RCVBUF/SNDBUF
523 are treated in BSD as hints */
525 if (val > sysctl_rmem_max)
526 val = sysctl_rmem_max;
528 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
530 * We double it on the way in to account for
531 * "struct sk_buff" etc. overhead. Applications
532 * assume that the SO_RCVBUF setting they make will
533 * allow that much actual data to be received on that
536 * Applications are unaware that "struct sk_buff" and
537 * other overheads allocate from the receive buffer
538 * during socket buffer allocation.
540 * And after considering the possible alternatives,
541 * returning the value we actually used in getsockopt
542 * is the most desirable behavior.
544 if ((val * 2) < SOCK_MIN_RCVBUF)
545 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
547 sk->sk_rcvbuf = val * 2;
551 if (!capable(CAP_NET_ADMIN)) {
559 if (sk->sk_protocol == IPPROTO_TCP)
560 tcp_set_keepalive(sk, valbool);
562 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
566 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
570 sk->sk_no_check = valbool;
574 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
575 sk->sk_priority = val;
581 if (optlen < sizeof(ling)) {
582 ret = -EINVAL; /* 1003.1g */
585 if (copy_from_user(&ling,optval,sizeof(ling))) {
590 sock_reset_flag(sk, SOCK_LINGER);
592 #if (BITS_PER_LONG == 32)
593 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
594 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
597 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
598 sock_set_flag(sk, SOCK_LINGER);
603 sock_warn_obsolete_bsdism("setsockopt");
608 set_bit(SOCK_PASSCRED, &sock->flags);
610 clear_bit(SOCK_PASSCRED, &sock->flags);
616 if (optname == SO_TIMESTAMP)
617 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
619 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
620 sock_set_flag(sk, SOCK_RCVTSTAMP);
621 sock_enable_timestamp(sk);
623 sock_reset_flag(sk, SOCK_RCVTSTAMP);
624 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
631 sk->sk_rcvlowat = val ? : 1;
635 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
639 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
642 case SO_ATTACH_FILTER:
644 if (optlen == sizeof(struct sock_fprog)) {
645 struct sock_fprog fprog;
648 if (copy_from_user(&fprog, optval, sizeof(fprog)))
651 ret = sk_attach_filter(&fprog, sk);
655 case SO_DETACH_FILTER:
656 ret = sk_detach_filter(sk);
661 set_bit(SOCK_PASSSEC, &sock->flags);
663 clear_bit(SOCK_PASSSEC, &sock->flags);
666 /* We implement the SO_SNDLOWAT etc to
667 not be settable (1003.1g 5.3) */
677 int sock_getsockopt(struct socket *sock, int level, int optname,
678 char __user *optval, int __user *optlen)
680 struct sock *sk = sock->sk;
688 unsigned int lv = sizeof(int);
691 if (get_user(len, optlen))
698 v.val = sock_flag(sk, SOCK_DBG);
702 v.val = sock_flag(sk, SOCK_LOCALROUTE);
706 v.val = !!sock_flag(sk, SOCK_BROADCAST);
710 v.val = sk->sk_sndbuf;
714 v.val = sk->sk_rcvbuf;
718 v.val = sk->sk_reuse;
722 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
730 v.val = -sock_error(sk);
732 v.val = xchg(&sk->sk_err_soft, 0);
736 v.val = !!sock_flag(sk, SOCK_URGINLINE);
740 v.val = sk->sk_no_check;
744 v.val = sk->sk_priority;
749 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
750 v.ling.l_linger = sk->sk_lingertime / HZ;
754 sock_warn_obsolete_bsdism("getsockopt");
758 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
759 !sock_flag(sk, SOCK_RCVTSTAMPNS);
763 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
767 lv=sizeof(struct timeval);
768 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
772 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
773 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
778 lv=sizeof(struct timeval);
779 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
783 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
784 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
789 v.val = sk->sk_rcvlowat;
797 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
801 if (len > sizeof(sk->sk_peercred))
802 len = sizeof(sk->sk_peercred);
803 if (copy_to_user(optval, &sk->sk_peercred, len))
811 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
815 if (copy_to_user(optval, address, len))
820 /* Dubious BSD thing... Probably nobody even uses it, but
821 * the UNIX standard wants it for whatever reason... -DaveM
824 v.val = sk->sk_state == TCP_LISTEN;
828 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
832 return security_socket_getpeersec_stream(sock, optval, optlen, len);
840 if (copy_to_user(optval, &v, len))
843 if (put_user(len, optlen))
849 * Initialize an sk_lock.
851 * (We also register the sk_lock with the lock validator.)
853 static inline void sock_lock_init(struct sock *sk)
855 sock_lock_init_class_and_name(sk,
856 af_family_slock_key_strings[sk->sk_family],
857 af_family_slock_keys + sk->sk_family,
858 af_family_key_strings[sk->sk_family],
859 af_family_keys + sk->sk_family);
862 static void sock_copy(struct sock *nsk, const struct sock *osk)
864 #ifdef CONFIG_SECURITY_NETWORK
865 void *sptr = nsk->sk_security;
868 memcpy(nsk, osk, osk->sk_prot->obj_size);
869 #ifdef CONFIG_SECURITY_NETWORK
870 nsk->sk_security = sptr;
871 security_sk_clone(osk, nsk);
875 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
879 struct kmem_cache *slab;
883 sk = kmem_cache_alloc(slab, priority);
885 sk = kmalloc(prot->obj_size, priority);
888 if (security_sk_alloc(sk, family, priority))
891 if (!try_module_get(prot->owner))
898 security_sk_free(sk);
901 kmem_cache_free(slab, sk);
907 static void sk_prot_free(struct proto *prot, struct sock *sk)
909 struct kmem_cache *slab;
910 struct module *owner;
915 security_sk_free(sk);
917 kmem_cache_free(slab, sk);
924 * sk_alloc - All socket objects are allocated here
925 * @net: the applicable net namespace
926 * @family: protocol family
927 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
928 * @prot: struct proto associated with this new sock instance
929 * @zero_it: if we should zero the newly allocated sock
931 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
936 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
938 sk->sk_family = family;
940 * See comment in struct sock definition to understand
941 * why we need sk_prot_creator -acme
943 sk->sk_prot = sk->sk_prot_creator = prot;
945 sk->sk_net = get_net(net);
951 void sk_free(struct sock *sk)
953 struct sk_filter *filter;
958 filter = rcu_dereference(sk->sk_filter);
960 sk_filter_uncharge(sk, filter);
961 rcu_assign_pointer(sk->sk_filter, NULL);
964 sock_disable_timestamp(sk);
966 if (atomic_read(&sk->sk_omem_alloc))
967 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
968 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
971 sk_prot_free(sk->sk_prot_creator, sk);
974 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
978 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
980 struct sk_filter *filter;
982 sock_copy(newsk, sk);
985 get_net(newsk->sk_net);
986 sk_node_init(&newsk->sk_node);
987 sock_lock_init(newsk);
989 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
991 atomic_set(&newsk->sk_rmem_alloc, 0);
992 atomic_set(&newsk->sk_wmem_alloc, 0);
993 atomic_set(&newsk->sk_omem_alloc, 0);
994 skb_queue_head_init(&newsk->sk_receive_queue);
995 skb_queue_head_init(&newsk->sk_write_queue);
996 #ifdef CONFIG_NET_DMA
997 skb_queue_head_init(&newsk->sk_async_wait_queue);
1000 rwlock_init(&newsk->sk_dst_lock);
1001 rwlock_init(&newsk->sk_callback_lock);
1002 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1003 af_callback_keys + newsk->sk_family,
1004 af_family_clock_key_strings[newsk->sk_family]);
1006 newsk->sk_dst_cache = NULL;
1007 newsk->sk_wmem_queued = 0;
1008 newsk->sk_forward_alloc = 0;
1009 newsk->sk_send_head = NULL;
1010 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1012 sock_reset_flag(newsk, SOCK_DONE);
1013 skb_queue_head_init(&newsk->sk_error_queue);
1015 filter = newsk->sk_filter;
1017 sk_filter_charge(newsk, filter);
1019 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1020 /* It is still raw copy of parent, so invalidate
1021 * destructor and make plain sk_free() */
1022 newsk->sk_destruct = NULL;
1029 newsk->sk_priority = 0;
1030 atomic_set(&newsk->sk_refcnt, 2);
1033 * Increment the counter in the same struct proto as the master
1034 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1035 * is the same as sk->sk_prot->socks, as this field was copied
1038 * This _changes_ the previous behaviour, where
1039 * tcp_create_openreq_child always was incrementing the
1040 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1041 * to be taken into account in all callers. -acme
1043 sk_refcnt_debug_inc(newsk);
1044 newsk->sk_socket = NULL;
1045 newsk->sk_sleep = NULL;
1047 if (newsk->sk_prot->sockets_allocated)
1048 atomic_inc(newsk->sk_prot->sockets_allocated);
1054 EXPORT_SYMBOL_GPL(sk_clone);
1056 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1058 __sk_dst_set(sk, dst);
1059 sk->sk_route_caps = dst->dev->features;
1060 if (sk->sk_route_caps & NETIF_F_GSO)
1061 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1062 if (sk_can_gso(sk)) {
1063 if (dst->header_len)
1064 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1066 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1069 EXPORT_SYMBOL_GPL(sk_setup_caps);
1071 void __init sk_init(void)
1073 if (num_physpages <= 4096) {
1074 sysctl_wmem_max = 32767;
1075 sysctl_rmem_max = 32767;
1076 sysctl_wmem_default = 32767;
1077 sysctl_rmem_default = 32767;
1078 } else if (num_physpages >= 131072) {
1079 sysctl_wmem_max = 131071;
1080 sysctl_rmem_max = 131071;
1085 * Simple resource managers for sockets.
1090 * Write buffer destructor automatically called from kfree_skb.
1092 void sock_wfree(struct sk_buff *skb)
1094 struct sock *sk = skb->sk;
1096 /* In case it might be waiting for more memory. */
1097 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1098 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1099 sk->sk_write_space(sk);
1104 * Read buffer destructor automatically called from kfree_skb.
1106 void sock_rfree(struct sk_buff *skb)
1108 struct sock *sk = skb->sk;
1110 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1114 int sock_i_uid(struct sock *sk)
1118 read_lock(&sk->sk_callback_lock);
1119 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1120 read_unlock(&sk->sk_callback_lock);
1124 unsigned long sock_i_ino(struct sock *sk)
1128 read_lock(&sk->sk_callback_lock);
1129 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1130 read_unlock(&sk->sk_callback_lock);
1135 * Allocate a skb from the socket's send buffer.
1137 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1140 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1141 struct sk_buff * skb = alloc_skb(size, priority);
1143 skb_set_owner_w(skb, sk);
1151 * Allocate a skb from the socket's receive buffer.
1153 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1156 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1157 struct sk_buff *skb = alloc_skb(size, priority);
1159 skb_set_owner_r(skb, sk);
1167 * Allocate a memory block from the socket's option memory buffer.
1169 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1171 if ((unsigned)size <= sysctl_optmem_max &&
1172 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1174 /* First do the add, to avoid the race if kmalloc
1177 atomic_add(size, &sk->sk_omem_alloc);
1178 mem = kmalloc(size, priority);
1181 atomic_sub(size, &sk->sk_omem_alloc);
1187 * Free an option memory block.
1189 void sock_kfree_s(struct sock *sk, void *mem, int size)
1192 atomic_sub(size, &sk->sk_omem_alloc);
1195 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1196 I think, these locks should be removed for datagram sockets.
1198 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1202 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1206 if (signal_pending(current))
1208 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1209 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1210 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1212 if (sk->sk_shutdown & SEND_SHUTDOWN)
1216 timeo = schedule_timeout(timeo);
1218 finish_wait(sk->sk_sleep, &wait);
1224 * Generic send/receive buffer handlers
1227 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1228 unsigned long header_len,
1229 unsigned long data_len,
1230 int noblock, int *errcode)
1232 struct sk_buff *skb;
1237 gfp_mask = sk->sk_allocation;
1238 if (gfp_mask & __GFP_WAIT)
1239 gfp_mask |= __GFP_REPEAT;
1241 timeo = sock_sndtimeo(sk, noblock);
1243 err = sock_error(sk);
1248 if (sk->sk_shutdown & SEND_SHUTDOWN)
1251 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1252 skb = alloc_skb(header_len, gfp_mask);
1257 /* No pages, we're done... */
1261 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1262 skb->truesize += data_len;
1263 skb_shinfo(skb)->nr_frags = npages;
1264 for (i = 0; i < npages; i++) {
1268 page = alloc_pages(sk->sk_allocation, 0);
1271 skb_shinfo(skb)->nr_frags = i;
1276 frag = &skb_shinfo(skb)->frags[i];
1278 frag->page_offset = 0;
1279 frag->size = (data_len >= PAGE_SIZE ?
1282 data_len -= PAGE_SIZE;
1285 /* Full success... */
1291 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1292 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1296 if (signal_pending(current))
1298 timeo = sock_wait_for_wmem(sk, timeo);
1301 skb_set_owner_w(skb, sk);
1305 err = sock_intr_errno(timeo);
1311 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1312 int noblock, int *errcode)
1314 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1317 static void __lock_sock(struct sock *sk)
1322 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1323 TASK_UNINTERRUPTIBLE);
1324 spin_unlock_bh(&sk->sk_lock.slock);
1326 spin_lock_bh(&sk->sk_lock.slock);
1327 if (!sock_owned_by_user(sk))
1330 finish_wait(&sk->sk_lock.wq, &wait);
1333 static void __release_sock(struct sock *sk)
1335 struct sk_buff *skb = sk->sk_backlog.head;
1338 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1342 struct sk_buff *next = skb->next;
1345 sk->sk_backlog_rcv(sk, skb);
1348 * We are in process context here with softirqs
1349 * disabled, use cond_resched_softirq() to preempt.
1350 * This is safe to do because we've taken the backlog
1353 cond_resched_softirq();
1356 } while (skb != NULL);
1359 } while ((skb = sk->sk_backlog.head) != NULL);
1363 * sk_wait_data - wait for data to arrive at sk_receive_queue
1364 * @sk: sock to wait on
1365 * @timeo: for how long
1367 * Now socket state including sk->sk_err is changed only under lock,
1368 * hence we may omit checks after joining wait queue.
1369 * We check receive queue before schedule() only as optimization;
1370 * it is very likely that release_sock() added new data.
1372 int sk_wait_data(struct sock *sk, long *timeo)
1377 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1378 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1379 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1380 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1381 finish_wait(sk->sk_sleep, &wait);
1385 EXPORT_SYMBOL(sk_wait_data);
1388 * Set of default routines for initialising struct proto_ops when
1389 * the protocol does not support a particular function. In certain
1390 * cases where it makes no sense for a protocol to have a "do nothing"
1391 * function, some default processing is provided.
1394 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1399 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1405 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1410 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1415 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1421 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1426 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1431 int sock_no_listen(struct socket *sock, int backlog)
1436 int sock_no_shutdown(struct socket *sock, int how)
1441 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1442 char __user *optval, int optlen)
1447 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1448 char __user *optval, int __user *optlen)
1453 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1459 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1460 size_t len, int flags)
1465 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1467 /* Mirror missing mmap method error code */
1471 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1474 struct msghdr msg = {.msg_flags = flags};
1476 char *kaddr = kmap(page);
1477 iov.iov_base = kaddr + offset;
1479 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1485 * Default Socket Callbacks
1488 static void sock_def_wakeup(struct sock *sk)
1490 read_lock(&sk->sk_callback_lock);
1491 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1492 wake_up_interruptible_all(sk->sk_sleep);
1493 read_unlock(&sk->sk_callback_lock);
1496 static void sock_def_error_report(struct sock *sk)
1498 read_lock(&sk->sk_callback_lock);
1499 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1500 wake_up_interruptible(sk->sk_sleep);
1501 sk_wake_async(sk,0,POLL_ERR);
1502 read_unlock(&sk->sk_callback_lock);
1505 static void sock_def_readable(struct sock *sk, int len)
1507 read_lock(&sk->sk_callback_lock);
1508 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1509 wake_up_interruptible(sk->sk_sleep);
1510 sk_wake_async(sk,1,POLL_IN);
1511 read_unlock(&sk->sk_callback_lock);
1514 static void sock_def_write_space(struct sock *sk)
1516 read_lock(&sk->sk_callback_lock);
1518 /* Do not wake up a writer until he can make "significant"
1521 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1522 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1523 wake_up_interruptible(sk->sk_sleep);
1525 /* Should agree with poll, otherwise some programs break */
1526 if (sock_writeable(sk))
1527 sk_wake_async(sk, 2, POLL_OUT);
1530 read_unlock(&sk->sk_callback_lock);
1533 static void sock_def_destruct(struct sock *sk)
1535 kfree(sk->sk_protinfo);
1538 void sk_send_sigurg(struct sock *sk)
1540 if (sk->sk_socket && sk->sk_socket->file)
1541 if (send_sigurg(&sk->sk_socket->file->f_owner))
1542 sk_wake_async(sk, 3, POLL_PRI);
1545 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1546 unsigned long expires)
1548 if (!mod_timer(timer, expires))
1552 EXPORT_SYMBOL(sk_reset_timer);
1554 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1556 if (timer_pending(timer) && del_timer(timer))
1560 EXPORT_SYMBOL(sk_stop_timer);
1562 void sock_init_data(struct socket *sock, struct sock *sk)
1564 skb_queue_head_init(&sk->sk_receive_queue);
1565 skb_queue_head_init(&sk->sk_write_queue);
1566 skb_queue_head_init(&sk->sk_error_queue);
1567 #ifdef CONFIG_NET_DMA
1568 skb_queue_head_init(&sk->sk_async_wait_queue);
1571 sk->sk_send_head = NULL;
1573 init_timer(&sk->sk_timer);
1575 sk->sk_allocation = GFP_KERNEL;
1576 sk->sk_rcvbuf = sysctl_rmem_default;
1577 sk->sk_sndbuf = sysctl_wmem_default;
1578 sk->sk_state = TCP_CLOSE;
1579 sk->sk_socket = sock;
1581 sock_set_flag(sk, SOCK_ZAPPED);
1584 sk->sk_type = sock->type;
1585 sk->sk_sleep = &sock->wait;
1588 sk->sk_sleep = NULL;
1590 rwlock_init(&sk->sk_dst_lock);
1591 rwlock_init(&sk->sk_callback_lock);
1592 lockdep_set_class_and_name(&sk->sk_callback_lock,
1593 af_callback_keys + sk->sk_family,
1594 af_family_clock_key_strings[sk->sk_family]);
1596 sk->sk_state_change = sock_def_wakeup;
1597 sk->sk_data_ready = sock_def_readable;
1598 sk->sk_write_space = sock_def_write_space;
1599 sk->sk_error_report = sock_def_error_report;
1600 sk->sk_destruct = sock_def_destruct;
1602 sk->sk_sndmsg_page = NULL;
1603 sk->sk_sndmsg_off = 0;
1605 sk->sk_peercred.pid = 0;
1606 sk->sk_peercred.uid = -1;
1607 sk->sk_peercred.gid = -1;
1608 sk->sk_write_pending = 0;
1609 sk->sk_rcvlowat = 1;
1610 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1611 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1613 sk->sk_stamp = ktime_set(-1L, -1L);
1615 atomic_set(&sk->sk_refcnt, 1);
1616 atomic_set(&sk->sk_drops, 0);
1619 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1622 spin_lock_bh(&sk->sk_lock.slock);
1623 if (sk->sk_lock.owned)
1625 sk->sk_lock.owned = 1;
1626 spin_unlock(&sk->sk_lock.slock);
1628 * The sk_lock has mutex_lock() semantics here:
1630 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1634 EXPORT_SYMBOL(lock_sock_nested);
1636 void fastcall release_sock(struct sock *sk)
1639 * The sk_lock has mutex_unlock() semantics:
1641 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1643 spin_lock_bh(&sk->sk_lock.slock);
1644 if (sk->sk_backlog.tail)
1646 sk->sk_lock.owned = 0;
1647 if (waitqueue_active(&sk->sk_lock.wq))
1648 wake_up(&sk->sk_lock.wq);
1649 spin_unlock_bh(&sk->sk_lock.slock);
1651 EXPORT_SYMBOL(release_sock);
1653 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1656 if (!sock_flag(sk, SOCK_TIMESTAMP))
1657 sock_enable_timestamp(sk);
1658 tv = ktime_to_timeval(sk->sk_stamp);
1659 if (tv.tv_sec == -1)
1661 if (tv.tv_sec == 0) {
1662 sk->sk_stamp = ktime_get_real();
1663 tv = ktime_to_timeval(sk->sk_stamp);
1665 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1667 EXPORT_SYMBOL(sock_get_timestamp);
1669 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1672 if (!sock_flag(sk, SOCK_TIMESTAMP))
1673 sock_enable_timestamp(sk);
1674 ts = ktime_to_timespec(sk->sk_stamp);
1675 if (ts.tv_sec == -1)
1677 if (ts.tv_sec == 0) {
1678 sk->sk_stamp = ktime_get_real();
1679 ts = ktime_to_timespec(sk->sk_stamp);
1681 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1683 EXPORT_SYMBOL(sock_get_timestampns);
1685 void sock_enable_timestamp(struct sock *sk)
1687 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1688 sock_set_flag(sk, SOCK_TIMESTAMP);
1689 net_enable_timestamp();
1694 * Get a socket option on an socket.
1696 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1697 * asynchronous errors should be reported by getsockopt. We assume
1698 * this means if you specify SO_ERROR (otherwise whats the point of it).
1700 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1701 char __user *optval, int __user *optlen)
1703 struct sock *sk = sock->sk;
1705 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1708 EXPORT_SYMBOL(sock_common_getsockopt);
1710 #ifdef CONFIG_COMPAT
1711 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1712 char __user *optval, int __user *optlen)
1714 struct sock *sk = sock->sk;
1716 if (sk->sk_prot->compat_getsockopt != NULL)
1717 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1719 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1721 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1724 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1725 struct msghdr *msg, size_t size, int flags)
1727 struct sock *sk = sock->sk;
1731 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1732 flags & ~MSG_DONTWAIT, &addr_len);
1734 msg->msg_namelen = addr_len;
1738 EXPORT_SYMBOL(sock_common_recvmsg);
1741 * Set socket options on an inet socket.
1743 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1744 char __user *optval, int optlen)
1746 struct sock *sk = sock->sk;
1748 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1751 EXPORT_SYMBOL(sock_common_setsockopt);
1753 #ifdef CONFIG_COMPAT
1754 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1755 char __user *optval, int optlen)
1757 struct sock *sk = sock->sk;
1759 if (sk->sk_prot->compat_setsockopt != NULL)
1760 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1762 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1764 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1767 void sk_common_release(struct sock *sk)
1769 if (sk->sk_prot->destroy)
1770 sk->sk_prot->destroy(sk);
1773 * Observation: when sock_common_release is called, processes have
1774 * no access to socket. But net still has.
1775 * Step one, detach it from networking:
1777 * A. Remove from hash tables.
1780 sk->sk_prot->unhash(sk);
1783 * In this point socket cannot receive new packets, but it is possible
1784 * that some packets are in flight because some CPU runs receiver and
1785 * did hash table lookup before we unhashed socket. They will achieve
1786 * receive queue and will be purged by socket destructor.
1788 * Also we still have packets pending on receive queue and probably,
1789 * our own packets waiting in device queues. sock_destroy will drain
1790 * receive queue, but transmitted packets will delay socket destruction
1791 * until the last reference will be released.
1796 xfrm_sk_free_policy(sk);
1798 sk_refcnt_debug_release(sk);
1802 EXPORT_SYMBOL(sk_common_release);
1804 static DEFINE_RWLOCK(proto_list_lock);
1805 static LIST_HEAD(proto_list);
1809 * Define default functions to keep track of inuse sockets per protocol
1810 * Note that often used protocols use dedicated functions to get a speed increase.
1811 * (see DEFINE_PROTO_INUSE/REF_PROTO_INUSE)
1813 static void inuse_add(struct proto *prot, int inc)
1815 per_cpu_ptr(prot->inuse_ptr, smp_processor_id())[0] += inc;
1818 static int inuse_get(const struct proto *prot)
1821 for_each_possible_cpu(cpu)
1822 res += per_cpu_ptr(prot->inuse_ptr, cpu)[0];
1826 static int inuse_init(struct proto *prot)
1828 if (!prot->inuse_getval || !prot->inuse_add) {
1829 prot->inuse_ptr = alloc_percpu(int);
1830 if (prot->inuse_ptr == NULL)
1833 prot->inuse_getval = inuse_get;
1834 prot->inuse_add = inuse_add;
1839 static void inuse_fini(struct proto *prot)
1841 if (prot->inuse_ptr != NULL) {
1842 free_percpu(prot->inuse_ptr);
1843 prot->inuse_ptr = NULL;
1844 prot->inuse_getval = NULL;
1845 prot->inuse_add = NULL;
1849 static inline int inuse_init(struct proto *prot)
1854 static inline void inuse_fini(struct proto *prot)
1859 int proto_register(struct proto *prot, int alloc_slab)
1861 char *request_sock_slab_name = NULL;
1862 char *timewait_sock_slab_name;
1864 if (inuse_init(prot))
1868 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1869 SLAB_HWCACHE_ALIGN, NULL);
1871 if (prot->slab == NULL) {
1872 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1874 goto out_free_inuse;
1877 if (prot->rsk_prot != NULL) {
1878 static const char mask[] = "request_sock_%s";
1880 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1881 if (request_sock_slab_name == NULL)
1882 goto out_free_sock_slab;
1884 sprintf(request_sock_slab_name, mask, prot->name);
1885 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1886 prot->rsk_prot->obj_size, 0,
1887 SLAB_HWCACHE_ALIGN, NULL);
1889 if (prot->rsk_prot->slab == NULL) {
1890 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1892 goto out_free_request_sock_slab_name;
1896 if (prot->twsk_prot != NULL) {
1897 static const char mask[] = "tw_sock_%s";
1899 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1901 if (timewait_sock_slab_name == NULL)
1902 goto out_free_request_sock_slab;
1904 sprintf(timewait_sock_slab_name, mask, prot->name);
1905 prot->twsk_prot->twsk_slab =
1906 kmem_cache_create(timewait_sock_slab_name,
1907 prot->twsk_prot->twsk_obj_size,
1908 0, SLAB_HWCACHE_ALIGN,
1910 if (prot->twsk_prot->twsk_slab == NULL)
1911 goto out_free_timewait_sock_slab_name;
1915 write_lock(&proto_list_lock);
1916 list_add(&prot->node, &proto_list);
1917 write_unlock(&proto_list_lock);
1920 out_free_timewait_sock_slab_name:
1921 kfree(timewait_sock_slab_name);
1922 out_free_request_sock_slab:
1923 if (prot->rsk_prot && prot->rsk_prot->slab) {
1924 kmem_cache_destroy(prot->rsk_prot->slab);
1925 prot->rsk_prot->slab = NULL;
1927 out_free_request_sock_slab_name:
1928 kfree(request_sock_slab_name);
1930 kmem_cache_destroy(prot->slab);
1938 EXPORT_SYMBOL(proto_register);
1940 void proto_unregister(struct proto *prot)
1942 write_lock(&proto_list_lock);
1943 list_del(&prot->node);
1944 write_unlock(&proto_list_lock);
1947 if (prot->slab != NULL) {
1948 kmem_cache_destroy(prot->slab);
1952 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1953 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1955 kmem_cache_destroy(prot->rsk_prot->slab);
1957 prot->rsk_prot->slab = NULL;
1960 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1961 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1963 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1965 prot->twsk_prot->twsk_slab = NULL;
1969 EXPORT_SYMBOL(proto_unregister);
1971 #ifdef CONFIG_PROC_FS
1972 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1974 read_lock(&proto_list_lock);
1975 return seq_list_start_head(&proto_list, *pos);
1978 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1980 return seq_list_next(v, &proto_list, pos);
1983 static void proto_seq_stop(struct seq_file *seq, void *v)
1985 read_unlock(&proto_list_lock);
1988 static char proto_method_implemented(const void *method)
1990 return method == NULL ? 'n' : 'y';
1993 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1995 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1996 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1999 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2000 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2001 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2003 proto->slab == NULL ? "no" : "yes",
2004 module_name(proto->owner),
2005 proto_method_implemented(proto->close),
2006 proto_method_implemented(proto->connect),
2007 proto_method_implemented(proto->disconnect),
2008 proto_method_implemented(proto->accept),
2009 proto_method_implemented(proto->ioctl),
2010 proto_method_implemented(proto->init),
2011 proto_method_implemented(proto->destroy),
2012 proto_method_implemented(proto->shutdown),
2013 proto_method_implemented(proto->setsockopt),
2014 proto_method_implemented(proto->getsockopt),
2015 proto_method_implemented(proto->sendmsg),
2016 proto_method_implemented(proto->recvmsg),
2017 proto_method_implemented(proto->sendpage),
2018 proto_method_implemented(proto->bind),
2019 proto_method_implemented(proto->backlog_rcv),
2020 proto_method_implemented(proto->hash),
2021 proto_method_implemented(proto->unhash),
2022 proto_method_implemented(proto->get_port),
2023 proto_method_implemented(proto->enter_memory_pressure));
2026 static int proto_seq_show(struct seq_file *seq, void *v)
2028 if (v == &proto_list)
2029 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2038 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2040 proto_seq_printf(seq, list_entry(v, struct proto, node));
2044 static const struct seq_operations proto_seq_ops = {
2045 .start = proto_seq_start,
2046 .next = proto_seq_next,
2047 .stop = proto_seq_stop,
2048 .show = proto_seq_show,
2051 static int proto_seq_open(struct inode *inode, struct file *file)
2053 return seq_open(file, &proto_seq_ops);
2056 static const struct file_operations proto_seq_fops = {
2057 .owner = THIS_MODULE,
2058 .open = proto_seq_open,
2060 .llseek = seq_lseek,
2061 .release = seq_release,
2064 static int __init proto_init(void)
2066 /* register /proc/net/protocols */
2067 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2070 subsys_initcall(proto_init);
2072 #endif /* PROC_FS */
2074 EXPORT_SYMBOL(sk_alloc);
2075 EXPORT_SYMBOL(sk_free);
2076 EXPORT_SYMBOL(sk_send_sigurg);
2077 EXPORT_SYMBOL(sock_alloc_send_skb);
2078 EXPORT_SYMBOL(sock_init_data);
2079 EXPORT_SYMBOL(sock_kfree_s);
2080 EXPORT_SYMBOL(sock_kmalloc);
2081 EXPORT_SYMBOL(sock_no_accept);
2082 EXPORT_SYMBOL(sock_no_bind);
2083 EXPORT_SYMBOL(sock_no_connect);
2084 EXPORT_SYMBOL(sock_no_getname);
2085 EXPORT_SYMBOL(sock_no_getsockopt);
2086 EXPORT_SYMBOL(sock_no_ioctl);
2087 EXPORT_SYMBOL(sock_no_listen);
2088 EXPORT_SYMBOL(sock_no_mmap);
2089 EXPORT_SYMBOL(sock_no_poll);
2090 EXPORT_SYMBOL(sock_no_recvmsg);
2091 EXPORT_SYMBOL(sock_no_sendmsg);
2092 EXPORT_SYMBOL(sock_no_sendpage);
2093 EXPORT_SYMBOL(sock_no_setsockopt);
2094 EXPORT_SYMBOL(sock_no_shutdown);
2095 EXPORT_SYMBOL(sock_no_socketpair);
2096 EXPORT_SYMBOL(sock_rfree);
2097 EXPORT_SYMBOL(sock_setsockopt);
2098 EXPORT_SYMBOL(sock_wfree);
2099 EXPORT_SYMBOL(sock_wmalloc);
2100 EXPORT_SYMBOL(sock_i_uid);
2101 EXPORT_SYMBOL(sock_i_ino);
2102 EXPORT_SYMBOL(sysctl_optmem_max);