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 1.116 2001/11/08 04:20:06 davem Exp $
12 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
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/config.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/major.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/fcntl.h>
107 #include <linux/mm.h>
108 #include <linux/slab.h>
109 #include <linux/interrupt.h>
110 #include <linux/poll.h>
111 #include <linux/tcp.h>
112 #include <linux/init.h>
114 #include <asm/uaccess.h>
115 #include <asm/system.h>
117 #include <linux/netdevice.h>
118 #include <net/protocol.h>
119 #include <linux/skbuff.h>
120 #include <net/sock.h>
121 #include <linux/ipsec.h>
124 #include <linux/filter.h>
131 /* Take into consideration the size of the struct sk_buff overhead in the
132 * determination of these values, since that is non-constant across
133 * platforms. This makes socket queueing behavior and performance
134 * not depend upon such differences.
136 #define _SK_MEM_PACKETS 256
137 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
138 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
139 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
141 /* Run time adjustable parameters. */
142 __u32 sysctl_wmem_max = SK_WMEM_MAX;
143 __u32 sysctl_rmem_max = SK_RMEM_MAX;
144 __u32 sysctl_wmem_default = SK_WMEM_MAX;
145 __u32 sysctl_rmem_default = SK_RMEM_MAX;
147 /* Maximal space eaten by iovec or ancilliary data plus some space */
148 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
150 static int sock_set_timeout(long *timeo_p, char *optval, int optlen)
154 if (optlen < sizeof(tv))
156 if (copy_from_user(&tv, optval, sizeof(tv)))
159 *timeo_p = MAX_SCHEDULE_TIMEOUT;
160 if (tv.tv_sec == 0 && tv.tv_usec == 0)
162 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
163 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
168 * This is meant for all protocols to use and covers goings on
169 * at the socket level. Everything here is generic.
172 int sock_setsockopt(struct socket *sock, int level, int optname,
173 char *optval, int optlen)
175 struct sock *sk=sock->sk;
177 struct sk_filter *filter;
185 * Options without arguments
188 #ifdef SO_DONTLINGER /* Compatibility item... */
197 if(optlen<sizeof(int))
200 if (get_user(val, (int *)optval))
210 if(val && !capable(CAP_NET_ADMIN))
225 sk->localroute=valbool;
228 sk->broadcast=valbool;
231 /* Don't error on this BSD doesn't and if you think
232 about it this is right. Otherwise apps have to
233 play 'guess the biggest size' games. RCVBUF/SNDBUF
234 are treated in BSD as hints */
236 if (val > sysctl_wmem_max)
237 val = sysctl_wmem_max;
239 sk->userlocks |= SOCK_SNDBUF_LOCK;
240 if ((val * 2) < SOCK_MIN_SNDBUF)
241 sk->sndbuf = SOCK_MIN_SNDBUF;
243 sk->sndbuf = (val * 2);
246 * Wake up sending tasks if we
253 /* Don't error on this BSD doesn't and if you think
254 about it this is right. Otherwise apps have to
255 play 'guess the biggest size' games. RCVBUF/SNDBUF
256 are treated in BSD as hints */
258 if (val > sysctl_rmem_max)
259 val = sysctl_rmem_max;
261 sk->userlocks |= SOCK_RCVBUF_LOCK;
262 /* FIXME: is this lower bound the right one? */
263 if ((val * 2) < SOCK_MIN_RCVBUF)
264 sk->rcvbuf = SOCK_MIN_RCVBUF;
266 sk->rcvbuf = (val * 2);
271 if (sk->protocol == IPPROTO_TCP)
273 tcp_set_keepalive(sk, valbool);
276 sk->keepopen = valbool;
280 sk->urginline = valbool;
284 sk->no_check = valbool;
288 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
295 if(optlen<sizeof(ling)) {
296 ret = -EINVAL; /* 1003.1g */
299 if (copy_from_user(&ling,optval,sizeof(ling))) {
303 if(ling.l_onoff==0) {
306 #if (BITS_PER_LONG == 32)
307 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
308 sk->lingertime=MAX_SCHEDULE_TIMEOUT;
311 sk->lingertime=ling.l_linger*HZ;
317 sk->bsdism = valbool;
321 sock->passcred = valbool;
325 sk->rcvtstamp = valbool;
331 sk->rcvlowat = val ? : 1;
335 ret = sock_set_timeout(&sk->rcvtimeo, optval, optlen);
339 ret = sock_set_timeout(&sk->sndtimeo, optval, optlen);
342 #ifdef CONFIG_NETDEVICES
343 case SO_BINDTODEVICE:
345 char devname[IFNAMSIZ];
348 if (!capable(CAP_NET_RAW)) {
353 /* Bind this socket to a particular device like "eth0",
354 * as specified in the passed interface name. If the
355 * name is "" or the option length is zero the socket
360 sk->bound_dev_if = 0;
362 if (optlen > IFNAMSIZ)
364 if (copy_from_user(devname, optval, optlen)) {
369 /* Remove any cached route for this socket. */
372 if (devname[0] == '\0') {
373 sk->bound_dev_if = 0;
375 struct net_device *dev = dev_get_by_name(devname);
380 sk->bound_dev_if = dev->ifindex;
390 case SO_ATTACH_FILTER:
392 if (optlen == sizeof(struct sock_fprog)) {
393 struct sock_fprog fprog;
396 if (copy_from_user(&fprog, optval, sizeof(fprog)))
399 ret = sk_attach_filter(&fprog, sk);
403 case SO_DETACH_FILTER:
404 spin_lock_bh(&sk->lock.slock);
408 spin_unlock_bh(&sk->lock.slock);
409 sk_filter_release(sk, filter);
412 spin_unlock_bh(&sk->lock.slock);
416 /* We implement the SO_SNDLOWAT etc to
417 not be settable (1003.1g 5.3) */
427 int sock_getsockopt(struct socket *sock, int level, int optname,
428 char *optval, int *optlen)
430 struct sock *sk = sock->sk;
439 unsigned int lv=sizeof(int),len;
441 if(get_user(len,optlen))
453 v.val = sk->localroute;
457 v.val= sk->broadcast;
473 v.val = sk->keepopen;
481 v.val = -sock_error(sk);
483 v.val=xchg(&sk->err_soft,0);
487 v.val = sk->urginline;
491 v.val = sk->no_check;
495 v.val = sk->priority;
500 v.ling.l_onoff=sk->linger;
501 v.ling.l_linger=sk->lingertime/HZ;
509 v.val = sk->rcvtstamp;
513 lv=sizeof(struct timeval);
514 if (sk->rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
518 v.tm.tv_sec = sk->rcvtimeo/HZ;
519 v.tm.tv_usec = ((sk->rcvtimeo%HZ)*1000000)/HZ;
524 lv=sizeof(struct timeval);
525 if (sk->sndtimeo == MAX_SCHEDULE_TIMEOUT) {
529 v.tm.tv_sec = sk->sndtimeo/HZ;
530 v.tm.tv_usec = ((sk->sndtimeo%HZ)*1000000)/HZ;
535 v.val = sk->rcvlowat;
543 v.val = sock->passcred;
547 if (len > sizeof(sk->peercred))
548 len = sizeof(sk->peercred);
549 if (copy_to_user(optval, &sk->peercred, len))
557 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
561 if(copy_to_user((void*)optval, address, len))
566 /* Dubious BSD thing... Probably nobody even uses it, but
567 * the UNIX standard wants it for whatever reason... -DaveM
570 v.val = (sk->state == TCP_LISTEN);
574 return(-ENOPROTOOPT);
578 if (copy_to_user(optval, &v, len))
581 if (put_user(len, optlen))
586 static kmem_cache_t *sk_cachep;
589 * All socket objects are allocated here. This is for future
593 struct sock *sk_alloc(int family, int priority, int zero_it)
595 struct sock *sk = kmem_cache_alloc(sk_cachep, priority);
598 memset(sk, 0, sizeof(struct sock));
606 void sk_free(struct sock *sk)
609 struct sk_filter *filter;
618 sk_filter_release(sk, filter);
623 if (atomic_read(&sk->omem_alloc))
624 printk(KERN_DEBUG "sk_free: optmem leakage (%d bytes) detected.\n", atomic_read(&sk->omem_alloc));
626 kmem_cache_free(sk_cachep, sk);
629 void __init sk_init(void)
631 sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
632 SLAB_HWCACHE_ALIGN, 0, 0);
634 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
636 if (num_physpages <= 4096) {
637 sysctl_wmem_max = 32767;
638 sysctl_rmem_max = 32767;
639 sysctl_wmem_default = 32767;
640 sysctl_rmem_default = 32767;
641 } else if (num_physpages >= 131072) {
642 sysctl_wmem_max = 131071;
643 sysctl_rmem_max = 131071;
648 * Simple resource managers for sockets.
653 * Write buffer destructor automatically called from kfree_skb.
655 void sock_wfree(struct sk_buff *skb)
657 struct sock *sk = skb->sk;
659 /* In case it might be waiting for more memory. */
660 atomic_sub(skb->truesize, &sk->wmem_alloc);
661 if (!sk->use_write_queue)
667 * Read buffer destructor automatically called from kfree_skb.
669 void sock_rfree(struct sk_buff *skb)
671 struct sock *sk = skb->sk;
673 atomic_sub(skb->truesize, &sk->rmem_alloc);
677 * Allocate a skb from the socket's send buffer.
679 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
681 if (force || atomic_read(&sk->wmem_alloc) < sk->sndbuf) {
682 struct sk_buff * skb = alloc_skb(size, priority);
684 skb_set_owner_w(skb, sk);
692 * Allocate a skb from the socket's receive buffer.
694 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
696 if (force || atomic_read(&sk->rmem_alloc) < sk->rcvbuf) {
697 struct sk_buff *skb = alloc_skb(size, priority);
699 skb_set_owner_r(skb, sk);
707 * Allocate a memory block from the socket's option memory buffer.
709 void *sock_kmalloc(struct sock *sk, int size, int priority)
711 if ((unsigned)size <= sysctl_optmem_max &&
712 atomic_read(&sk->omem_alloc)+size < sysctl_optmem_max) {
714 /* First do the add, to avoid the race if kmalloc
717 atomic_add(size, &sk->omem_alloc);
718 mem = kmalloc(size, priority);
721 atomic_sub(size, &sk->omem_alloc);
727 * Free an option memory block.
729 void sock_kfree_s(struct sock *sk, void *mem, int size)
732 atomic_sub(size, &sk->omem_alloc);
735 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
736 I think, these locks should be removed for datagram sockets.
738 static long sock_wait_for_wmem(struct sock * sk, long timeo)
740 DECLARE_WAITQUEUE(wait, current);
742 clear_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags);
743 add_wait_queue(sk->sleep, &wait);
747 if (signal_pending(current))
749 set_bit(SOCK_NOSPACE, &sk->socket->flags);
750 set_current_state(TASK_INTERRUPTIBLE);
751 if (atomic_read(&sk->wmem_alloc) < sk->sndbuf)
753 if (sk->shutdown & SEND_SHUTDOWN)
757 timeo = schedule_timeout(timeo);
759 __set_current_state(TASK_RUNNING);
760 remove_wait_queue(sk->sleep, &wait);
766 * Generic send/receive buffer handlers
769 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
770 unsigned long data_len, int noblock, int *errcode)
776 timeo = sock_sndtimeo(sk, noblock);
778 err = sock_error(sk);
783 if (sk->shutdown & SEND_SHUTDOWN)
786 if (atomic_read(&sk->wmem_alloc) < sk->sndbuf) {
787 skb = alloc_skb(header_len, sk->allocation);
792 /* No pages, we're done... */
796 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
797 skb->truesize += data_len;
798 skb_shinfo(skb)->nr_frags = npages;
799 for (i = 0; i < npages; i++) {
803 page = alloc_pages(sk->allocation, 0);
806 skb_shinfo(skb)->nr_frags = i;
811 frag = &skb_shinfo(skb)->frags[i];
813 frag->page_offset = 0;
814 frag->size = (data_len >= PAGE_SIZE ?
817 data_len -= PAGE_SIZE;
820 /* Full success... */
826 set_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags);
827 set_bit(SOCK_NOSPACE, &sk->socket->flags);
831 if (signal_pending(current))
833 timeo = sock_wait_for_wmem(sk, timeo);
836 skb_set_owner_w(skb, sk);
840 err = sock_intr_errno(timeo);
846 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
847 int noblock, int *errcode)
849 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
852 void __lock_sock(struct sock *sk)
854 DECLARE_WAITQUEUE(wait, current);
856 add_wait_queue_exclusive(&sk->lock.wq, &wait);
858 current->state = TASK_UNINTERRUPTIBLE;
859 spin_unlock_bh(&sk->lock.slock);
861 spin_lock_bh(&sk->lock.slock);
865 current->state = TASK_RUNNING;
866 remove_wait_queue(&sk->lock.wq, &wait);
869 void __release_sock(struct sock *sk)
871 struct sk_buff *skb = sk->backlog.head;
874 sk->backlog.head = sk->backlog.tail = NULL;
878 struct sk_buff *next = skb->next;
881 sk->backlog_rcv(sk, skb);
883 } while (skb != NULL);
886 } while((skb = sk->backlog.head) != NULL);
890 * Generic socket manager library. Most simpler socket families
891 * use this to manage their socket lists. At some point we should
892 * hash these. By making this generic we get the lot hashed for free.
894 * It is broken by design. All the protocols using it must be fixed. --ANK
897 rwlock_t net_big_sklist_lock = RW_LOCK_UNLOCKED;
899 void sklist_remove_socket(struct sock **list, struct sock *sk)
903 write_lock_bh(&net_big_sklist_lock);
905 while ((s = *list) != NULL) {
913 write_unlock_bh(&net_big_sklist_lock);
918 void sklist_insert_socket(struct sock **list, struct sock *sk)
920 write_lock_bh(&net_big_sklist_lock);
924 write_unlock_bh(&net_big_sklist_lock);
928 * This is only called from user mode. Thus it protects itself against
929 * interrupt users but doesn't worry about being called during work.
930 * Once it is removed from the queue no interrupt or bottom half will
931 * touch it and we are (fairly 8-) ) safe.
934 void sklist_destroy_socket(struct sock **list, struct sock *sk);
937 * Handler for deferred kills.
940 static void sklist_destroy_timer(unsigned long data)
942 struct sock *sk=(struct sock *)data;
943 sklist_destroy_socket(NULL,sk);
947 * Destroy a socket. We pass NULL for a list if we know the
948 * socket is not on a list.
951 void sklist_destroy_socket(struct sock **list,struct sock *sk)
954 sklist_remove_socket(list, sk);
956 skb_queue_purge(&sk->receive_queue);
958 if(atomic_read(&sk->wmem_alloc) == 0 &&
959 atomic_read(&sk->rmem_alloc) == 0 &&
967 * Someone is using our buffers still.. defer
969 init_timer(&sk->timer);
970 sk->timer.expires=jiffies+SOCK_DESTROY_TIME;
971 sk->timer.function=sklist_destroy_timer;
972 sk->timer.data = (unsigned long)sk;
973 add_timer(&sk->timer);
978 * Set of default routines for initialising struct proto_ops when
979 * the protocol does not support a particular function. In certain
980 * cases where it makes no sense for a protocol to have a "do nothing"
981 * function, some default processing is provided.
984 int sock_no_release(struct socket *sock)
989 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
994 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1000 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1005 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1010 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1016 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1021 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1026 int sock_no_listen(struct socket *sock, int backlog)
1031 int sock_no_shutdown(struct socket *sock, int how)
1036 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1037 char *optval, int optlen)
1042 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1043 char *optval, int *optlen)
1049 * Note: if you add something that sleeps here then change sock_fcntl()
1050 * to do proper fd locking.
1052 int sock_no_fcntl(struct socket *sock, unsigned int cmd, unsigned long arg)
1054 struct sock *sk = sock->sk;
1060 * This is a little restrictive, but it's the only
1061 * way to make sure that you can't send a sigurg to
1064 if (current->pgrp != -arg &&
1065 current->pid != arg &&
1066 !capable(CAP_KILL)) return(-EPERM);
1076 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, int flags,
1077 struct scm_cookie *scm)
1082 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, int len, int flags,
1083 struct scm_cookie *scm)
1088 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1090 /* Mirror missing mmap method error code */
1094 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1099 mm_segment_t old_fs;
1104 msg.msg_name = NULL;
1105 msg.msg_namelen = 0;
1108 msg.msg_control = NULL;
1109 msg.msg_controllen = 0;
1110 msg.msg_flags = flags;
1112 iov.iov_base = kaddr + offset;
1117 res = sock_sendmsg(sock, &msg, size);
1125 * Default Socket Callbacks
1128 void sock_def_wakeup(struct sock *sk)
1130 read_lock(&sk->callback_lock);
1131 if (sk->sleep && waitqueue_active(sk->sleep))
1132 wake_up_interruptible_all(sk->sleep);
1133 read_unlock(&sk->callback_lock);
1136 void sock_def_error_report(struct sock *sk)
1138 read_lock(&sk->callback_lock);
1139 if (sk->sleep && waitqueue_active(sk->sleep))
1140 wake_up_interruptible(sk->sleep);
1141 sk_wake_async(sk,0,POLL_ERR);
1142 read_unlock(&sk->callback_lock);
1145 void sock_def_readable(struct sock *sk, int len)
1147 read_lock(&sk->callback_lock);
1148 if (sk->sleep && waitqueue_active(sk->sleep))
1149 wake_up_interruptible(sk->sleep);
1150 sk_wake_async(sk,1,POLL_IN);
1151 read_unlock(&sk->callback_lock);
1154 void sock_def_write_space(struct sock *sk)
1156 read_lock(&sk->callback_lock);
1158 /* Do not wake up a writer until he can make "significant"
1161 if((atomic_read(&sk->wmem_alloc) << 1) <= sk->sndbuf) {
1162 if (sk->sleep && waitqueue_active(sk->sleep))
1163 wake_up_interruptible(sk->sleep);
1165 /* Should agree with poll, otherwise some programs break */
1166 if (sock_writeable(sk))
1167 sk_wake_async(sk, 2, POLL_OUT);
1170 read_unlock(&sk->callback_lock);
1173 void sock_def_destruct(struct sock *sk)
1175 if (sk->protinfo.destruct_hook)
1176 kfree(sk->protinfo.destruct_hook);
1179 void sock_init_data(struct socket *sock, struct sock *sk)
1181 skb_queue_head_init(&sk->receive_queue);
1182 skb_queue_head_init(&sk->write_queue);
1183 skb_queue_head_init(&sk->error_queue);
1185 init_timer(&sk->timer);
1187 sk->allocation = GFP_KERNEL;
1188 sk->rcvbuf = sysctl_rmem_default;
1189 sk->sndbuf = sysctl_wmem_default;
1190 sk->state = TCP_CLOSE;
1196 sk->type = sock->type;
1197 sk->sleep = &sock->wait;
1202 sk->dst_lock = RW_LOCK_UNLOCKED;
1203 sk->callback_lock = RW_LOCK_UNLOCKED;
1205 sk->state_change = sock_def_wakeup;
1206 sk->data_ready = sock_def_readable;
1207 sk->write_space = sock_def_write_space;
1208 sk->error_report = sock_def_error_report;
1209 sk->destruct = sock_def_destruct;
1211 sk->peercred.pid = 0;
1212 sk->peercred.uid = -1;
1213 sk->peercred.gid = -1;
1215 sk->rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1216 sk->sndtimeo = MAX_SCHEDULE_TIMEOUT;
1218 atomic_set(&sk->refcnt, 1);