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 $
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/module.h>
102 #include <linux/sched.h>
103 #include <linux/timer.h>
104 #include <linux/string.h>
105 #include <linux/sockios.h>
106 #include <linux/net.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>
123 #include <linux/filter.h>
129 /* Take into consideration the size of the struct sk_buff overhead in the
130 * determination of these values, since that is non-constant across
131 * platforms. This makes socket queueing behavior and performance
132 * not depend upon such differences.
134 #define _SK_MEM_PACKETS 256
135 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
136 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
137 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
139 /* Run time adjustable parameters. */
140 __u32 sysctl_wmem_max = SK_WMEM_MAX;
141 __u32 sysctl_rmem_max = SK_RMEM_MAX;
142 __u32 sysctl_wmem_default = SK_WMEM_MAX;
143 __u32 sysctl_rmem_default = SK_RMEM_MAX;
145 /* Maximal space eaten by iovec or ancilliary data plus some space */
146 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
148 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
152 if (optlen < sizeof(tv))
154 if (copy_from_user(&tv, optval, sizeof(tv)))
157 *timeo_p = MAX_SCHEDULE_TIMEOUT;
158 if (tv.tv_sec == 0 && tv.tv_usec == 0)
160 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
161 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
165 static void sock_warn_obsolete_bsdism(const char *name)
168 static char warncomm[16];
169 if (strcmp(warncomm, current->comm) && warned < 5) {
170 strcpy(warncomm, current->comm);
171 printk(KERN_WARNING "process `%s' is using obsolete "
172 "%s SO_BSDCOMPAT\n", warncomm, name);
178 * This is meant for all protocols to use and covers goings on
179 * at the socket level. Everything here is generic.
182 int sock_setsockopt(struct socket *sock, int level, int optname,
183 char __user *optval, int optlen)
185 struct sock *sk=sock->sk;
186 struct sk_filter *filter;
193 * Options without arguments
196 #ifdef SO_DONTLINGER /* Compatibility item... */
199 sock_reset_flag(sk, SOCK_LINGER);
204 if(optlen<sizeof(int))
207 if (get_user(val, (int __user *)optval))
217 if(val && !capable(CAP_NET_ADMIN))
222 sk->sk_debug = valbool;
225 sk->sk_reuse = valbool;
232 sk->sk_localroute = valbool;
235 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
238 /* Don't error on this BSD doesn't and if you think
239 about it this is right. Otherwise apps have to
240 play 'guess the biggest size' games. RCVBUF/SNDBUF
241 are treated in BSD as hints */
243 if (val > sysctl_wmem_max)
244 val = sysctl_wmem_max;
246 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
247 if ((val * 2) < SOCK_MIN_SNDBUF)
248 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
250 sk->sk_sndbuf = val * 2;
253 * Wake up sending tasks if we
256 sk->sk_write_space(sk);
260 /* Don't error on this BSD doesn't and if you think
261 about it this is right. Otherwise apps have to
262 play 'guess the biggest size' games. RCVBUF/SNDBUF
263 are treated in BSD as hints */
265 if (val > sysctl_rmem_max)
266 val = sysctl_rmem_max;
268 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
269 /* FIXME: is this lower bound the right one? */
270 if ((val * 2) < SOCK_MIN_RCVBUF)
271 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
273 sk->sk_rcvbuf = val * 2;
278 if (sk->sk_protocol == IPPROTO_TCP)
279 tcp_set_keepalive(sk, valbool);
281 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
285 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
289 sk->sk_no_check = valbool;
293 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
294 sk->sk_priority = val;
300 if(optlen<sizeof(ling)) {
301 ret = -EINVAL; /* 1003.1g */
304 if (copy_from_user(&ling,optval,sizeof(ling))) {
309 sock_reset_flag(sk, SOCK_LINGER);
311 #if (BITS_PER_LONG == 32)
312 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
313 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
316 sk->sk_lingertime = ling.l_linger * HZ;
317 sock_set_flag(sk, SOCK_LINGER);
322 sock_warn_obsolete_bsdism("setsockopt");
327 set_bit(SOCK_PASS_CRED, &sock->flags);
329 clear_bit(SOCK_PASS_CRED, &sock->flags);
333 sk->sk_rcvtstamp = valbool;
335 sock_enable_timestamp(sk);
341 sk->sk_rcvlowat = val ? : 1;
345 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
349 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
352 #ifdef CONFIG_NETDEVICES
353 case SO_BINDTODEVICE:
355 char devname[IFNAMSIZ];
358 if (!capable(CAP_NET_RAW)) {
363 /* Bind this socket to a particular device like "eth0",
364 * as specified in the passed interface name. If the
365 * name is "" or the option length is zero the socket
370 sk->sk_bound_dev_if = 0;
372 if (optlen > IFNAMSIZ)
374 if (copy_from_user(devname, optval, optlen)) {
379 /* Remove any cached route for this socket. */
382 if (devname[0] == '\0') {
383 sk->sk_bound_dev_if = 0;
385 struct net_device *dev = dev_get_by_name(devname);
390 sk->sk_bound_dev_if = dev->ifindex;
399 case SO_ATTACH_FILTER:
401 if (optlen == sizeof(struct sock_fprog)) {
402 struct sock_fprog fprog;
405 if (copy_from_user(&fprog, optval, sizeof(fprog)))
408 ret = sk_attach_filter(&fprog, sk);
412 case SO_DETACH_FILTER:
413 spin_lock_bh(&sk->sk_lock.slock);
414 filter = sk->sk_filter;
416 sk->sk_filter = NULL;
417 spin_unlock_bh(&sk->sk_lock.slock);
418 sk_filter_release(sk, filter);
421 spin_unlock_bh(&sk->sk_lock.slock);
425 /* We implement the SO_SNDLOWAT etc to
426 not be settable (1003.1g 5.3) */
436 int sock_getsockopt(struct socket *sock, int level, int optname,
437 char __user *optval, int __user *optlen)
439 struct sock *sk = sock->sk;
448 unsigned int lv=sizeof(int),len;
450 if(get_user(len,optlen))
458 v.val = sk->sk_debug;
462 v.val = sk->sk_localroute;
466 v.val = !!sock_flag(sk, SOCK_BROADCAST);
470 v.val = sk->sk_sndbuf;
474 v.val = sk->sk_rcvbuf;
478 v.val = sk->sk_reuse;
482 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
490 v.val = -sock_error(sk);
492 v.val = xchg(&sk->sk_err_soft, 0);
496 v.val = !!sock_flag(sk, SOCK_URGINLINE);
500 v.val = sk->sk_no_check;
504 v.val = sk->sk_priority;
509 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
510 v.ling.l_linger = sk->sk_lingertime / HZ;
514 sock_warn_obsolete_bsdism("getsockopt");
518 v.val = sk->sk_rcvtstamp;
522 lv=sizeof(struct timeval);
523 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
527 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
528 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000) / HZ;
533 lv=sizeof(struct timeval);
534 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
538 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
539 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000) / HZ;
544 v.val = sk->sk_rcvlowat;
552 v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
556 if (len > sizeof(sk->sk_peercred))
557 len = sizeof(sk->sk_peercred);
558 if (copy_to_user(optval, &sk->sk_peercred, len))
566 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
570 if (copy_to_user(optval, address, len))
575 /* Dubious BSD thing... Probably nobody even uses it, but
576 * the UNIX standard wants it for whatever reason... -DaveM
579 v.val = sk->sk_state == TCP_LISTEN;
583 return security_socket_getpeersec(sock, optval, optlen, len);
586 return(-ENOPROTOOPT);
590 if (copy_to_user(optval, &v, len))
593 if (put_user(len, optlen))
598 static kmem_cache_t *sk_cachep;
601 * sk_alloc - All socket objects are allocated here
602 * @family - protocol family
603 * @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
604 * @zero_it - zeroes the allocated sock
605 * @slab - alternate slab
607 * All socket objects are allocated here. If @zero_it is non-zero
608 * it should have the size of the are to be zeroed, because the
609 * private slabcaches have different sizes of the generic struct sock.
610 * 1 has been kept as a way to say sizeof(struct sock).
612 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
614 struct sock *sk = NULL;
618 sk = kmem_cache_alloc(slab, priority);
622 zero_it == 1 ? sizeof(struct sock) : zero_it);
623 sk->sk_family = family;
630 if (security_sk_alloc(sk, family, priority)) {
631 kmem_cache_free(slab, sk);
638 void sk_free(struct sock *sk)
640 struct sk_filter *filter;
641 struct module *owner = sk->sk_owner;
646 filter = sk->sk_filter;
648 sk_filter_release(sk, filter);
649 sk->sk_filter = NULL;
652 sock_disable_timestamp(sk);
654 if (atomic_read(&sk->sk_omem_alloc))
655 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
656 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
658 security_sk_free(sk);
659 BUG_ON(sk->sk_vx_info);
660 BUG_ON(sk->sk_nx_info);
661 /* clr_vx_info(&sk->sk_vx_info);
662 clr_nx_info(&sk->sk_nx_info); */
663 kmem_cache_free(sk->sk_slab, sk);
667 void __init sk_init(void)
669 sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
670 SLAB_HWCACHE_ALIGN, 0, 0);
672 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
674 if (num_physpages <= 4096) {
675 sysctl_wmem_max = 32767;
676 sysctl_rmem_max = 32767;
677 sysctl_wmem_default = 32767;
678 sysctl_rmem_default = 32767;
679 } else if (num_physpages >= 131072) {
680 sysctl_wmem_max = 131071;
681 sysctl_rmem_max = 131071;
686 * Simple resource managers for sockets.
691 * Write buffer destructor automatically called from kfree_skb.
693 void sock_wfree(struct sk_buff *skb)
695 struct sock *sk = skb->sk;
697 /* In case it might be waiting for more memory. */
698 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
699 if (!sk->sk_use_write_queue)
700 sk->sk_write_space(sk);
705 * Read buffer destructor automatically called from kfree_skb.
707 void sock_rfree(struct sk_buff *skb)
709 struct sock *sk = skb->sk;
711 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
715 * Allocate a skb from the socket's send buffer.
717 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
719 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
720 struct sk_buff * skb = alloc_skb(size, priority);
722 skb_set_owner_w(skb, sk);
730 * Allocate a skb from the socket's receive buffer.
732 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
734 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
735 struct sk_buff *skb = alloc_skb(size, priority);
737 skb_set_owner_r(skb, sk);
745 * Allocate a memory block from the socket's option memory buffer.
747 void *sock_kmalloc(struct sock *sk, int size, int priority)
749 if ((unsigned)size <= sysctl_optmem_max &&
750 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
752 /* First do the add, to avoid the race if kmalloc
755 atomic_add(size, &sk->sk_omem_alloc);
756 mem = kmalloc(size, priority);
759 atomic_sub(size, &sk->sk_omem_alloc);
765 * Free an option memory block.
767 void sock_kfree_s(struct sock *sk, void *mem, int size)
770 atomic_sub(size, &sk->sk_omem_alloc);
773 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
774 I think, these locks should be removed for datagram sockets.
776 static long sock_wait_for_wmem(struct sock * sk, long timeo)
780 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
784 if (signal_pending(current))
786 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
787 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
788 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
790 if (sk->sk_shutdown & SEND_SHUTDOWN)
794 timeo = schedule_timeout(timeo);
796 finish_wait(sk->sk_sleep, &wait);
802 * Generic send/receive buffer handlers
805 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
806 unsigned long data_len, int noblock, int *errcode)
809 unsigned int gfp_mask;
813 gfp_mask = sk->sk_allocation;
814 if (gfp_mask & __GFP_WAIT)
815 gfp_mask |= __GFP_REPEAT;
817 timeo = sock_sndtimeo(sk, noblock);
819 err = sock_error(sk);
824 if (sk->sk_shutdown & SEND_SHUTDOWN)
827 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
828 skb = alloc_skb(header_len, sk->sk_allocation);
833 /* No pages, we're done... */
837 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
838 skb->truesize += data_len;
839 skb_shinfo(skb)->nr_frags = npages;
840 for (i = 0; i < npages; i++) {
844 page = alloc_pages(sk->sk_allocation, 0);
847 skb_shinfo(skb)->nr_frags = i;
852 frag = &skb_shinfo(skb)->frags[i];
854 frag->page_offset = 0;
855 frag->size = (data_len >= PAGE_SIZE ?
858 data_len -= PAGE_SIZE;
861 /* Full success... */
867 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
868 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
872 if (signal_pending(current))
874 timeo = sock_wait_for_wmem(sk, timeo);
877 skb_set_owner_w(skb, sk);
881 err = sock_intr_errno(timeo);
887 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
888 int noblock, int *errcode)
890 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
893 void __lock_sock(struct sock *sk)
898 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
899 TASK_UNINTERRUPTIBLE);
900 spin_unlock_bh(&sk->sk_lock.slock);
902 spin_lock_bh(&sk->sk_lock.slock);
903 if(!sock_owned_by_user(sk))
906 finish_wait(&sk->sk_lock.wq, &wait);
909 void __release_sock(struct sock *sk)
911 struct sk_buff *skb = sk->sk_backlog.head;
914 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
918 struct sk_buff *next = skb->next;
921 sk->sk_backlog_rcv(sk, skb);
923 } while (skb != NULL);
926 } while((skb = sk->sk_backlog.head) != NULL);
930 * Set of default routines for initialising struct proto_ops when
931 * the protocol does not support a particular function. In certain
932 * cases where it makes no sense for a protocol to have a "do nothing"
933 * function, some default processing is provided.
936 int sock_no_release(struct socket *sock)
941 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
946 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
952 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
957 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
962 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
968 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
973 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
978 int sock_no_listen(struct socket *sock, int backlog)
983 int sock_no_shutdown(struct socket *sock, int how)
988 int sock_no_setsockopt(struct socket *sock, int level, int optname,
989 char __user *optval, int optlen)
994 int sock_no_getsockopt(struct socket *sock, int level, int optname,
995 char __user *optval, int __user *optlen)
1000 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1006 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1007 size_t len, int flags)
1012 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1014 /* Mirror missing mmap method error code */
1018 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1023 mm_segment_t old_fs;
1028 msg.msg_name = NULL;
1029 msg.msg_namelen = 0;
1032 msg.msg_control = NULL;
1033 msg.msg_controllen = 0;
1034 msg.msg_flags = flags;
1036 /* This cast is ok because of the "set_fs(KERNEL_DS)" */
1037 iov.iov_base = (void __user *) (kaddr + offset);
1042 res = sock_sendmsg(sock, &msg, size);
1050 * Default Socket Callbacks
1053 void sock_def_wakeup(struct sock *sk)
1055 read_lock(&sk->sk_callback_lock);
1056 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1057 wake_up_interruptible_all(sk->sk_sleep);
1058 read_unlock(&sk->sk_callback_lock);
1061 void sock_def_error_report(struct sock *sk)
1063 read_lock(&sk->sk_callback_lock);
1064 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1065 wake_up_interruptible(sk->sk_sleep);
1066 sk_wake_async(sk,0,POLL_ERR);
1067 read_unlock(&sk->sk_callback_lock);
1070 void sock_def_readable(struct sock *sk, int len)
1072 read_lock(&sk->sk_callback_lock);
1073 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1074 wake_up_interruptible(sk->sk_sleep);
1075 sk_wake_async(sk,1,POLL_IN);
1076 read_unlock(&sk->sk_callback_lock);
1079 void sock_def_write_space(struct sock *sk)
1081 read_lock(&sk->sk_callback_lock);
1083 /* Do not wake up a writer until he can make "significant"
1086 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1087 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1088 wake_up_interruptible(sk->sk_sleep);
1090 /* Should agree with poll, otherwise some programs break */
1091 if (sock_writeable(sk))
1092 sk_wake_async(sk, 2, POLL_OUT);
1095 read_unlock(&sk->sk_callback_lock);
1098 void sock_def_destruct(struct sock *sk)
1100 if (sk->sk_protinfo)
1101 kfree(sk->sk_protinfo);
1104 void sk_send_sigurg(struct sock *sk)
1106 if (sk->sk_socket && sk->sk_socket->file)
1107 if (send_sigurg(&sk->sk_socket->file->f_owner))
1108 sk_wake_async(sk, 3, POLL_PRI);
1111 void sock_init_data(struct socket *sock, struct sock *sk)
1113 skb_queue_head_init(&sk->sk_receive_queue);
1114 skb_queue_head_init(&sk->sk_write_queue);
1115 skb_queue_head_init(&sk->sk_error_queue);
1117 init_timer(&sk->sk_timer);
1119 sk->sk_allocation = GFP_KERNEL;
1120 sk->sk_rcvbuf = sysctl_rmem_default;
1121 sk->sk_sndbuf = sysctl_wmem_default;
1122 sk->sk_state = TCP_CLOSE;
1124 sk->sk_socket = sock;
1128 sk->sk_type = sock->type;
1129 sk->sk_sleep = &sock->wait;
1132 sk->sk_sleep = NULL;
1134 sk->sk_dst_lock = RW_LOCK_UNLOCKED;
1135 sk->sk_callback_lock = RW_LOCK_UNLOCKED;
1137 sk->sk_state_change = sock_def_wakeup;
1138 sk->sk_data_ready = sock_def_readable;
1139 sk->sk_write_space = sock_def_write_space;
1140 sk->sk_error_report = sock_def_error_report;
1141 sk->sk_destruct = sock_def_destruct;
1143 sk->sk_peercred.pid = 0;
1144 sk->sk_peercred.uid = -1;
1145 sk->sk_peercred.gid = -1;
1146 sk->sk_rcvlowat = 1;
1147 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1148 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1149 sk->sk_owner = NULL;
1151 sk->sk_stamp.tv_sec = -1L;
1152 sk->sk_stamp.tv_usec = -1L;
1154 sk->sk_vx_info = NULL;
1156 sk->sk_nx_info = NULL;
1159 atomic_set(&sk->sk_refcnt, 1);
1162 void fastcall lock_sock(struct sock *sk)
1165 spin_lock_bh(&(sk->sk_lock.slock));
1166 if (sk->sk_lock.owner)
1168 sk->sk_lock.owner = (void *)1;
1169 spin_unlock_bh(&(sk->sk_lock.slock));
1172 EXPORT_SYMBOL(lock_sock);
1174 void fastcall release_sock(struct sock *sk)
1176 spin_lock_bh(&(sk->sk_lock.slock));
1177 if (sk->sk_backlog.tail)
1179 sk->sk_lock.owner = NULL;
1180 if (waitqueue_active(&(sk->sk_lock.wq)))
1181 wake_up(&(sk->sk_lock.wq));
1182 spin_unlock_bh(&(sk->sk_lock.slock));
1184 EXPORT_SYMBOL(release_sock);
1186 /* When > 0 there are consumers of rx skb time stamps */
1187 atomic_t netstamp_needed = ATOMIC_INIT(0);
1189 int sock_get_timestamp(struct sock *sk, struct timeval *userstamp)
1191 if (!sock_flag(sk, SOCK_TIMESTAMP))
1192 sock_enable_timestamp(sk);
1193 if (sk->sk_stamp.tv_sec == -1)
1195 if (sk->sk_stamp.tv_sec == 0)
1196 do_gettimeofday(&sk->sk_stamp);
1197 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1200 EXPORT_SYMBOL(sock_get_timestamp);
1202 void sock_enable_timestamp(struct sock *sk)
1204 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1205 sock_set_flag(sk, SOCK_TIMESTAMP);
1206 atomic_inc(&netstamp_needed);
1209 EXPORT_SYMBOL(sock_enable_timestamp);
1211 void sock_disable_timestamp(struct sock *sk)
1213 if (sock_flag(sk, SOCK_TIMESTAMP)) {
1214 sock_reset_flag(sk, SOCK_TIMESTAMP);
1215 atomic_dec(&netstamp_needed);
1218 EXPORT_SYMBOL(sock_disable_timestamp);
1220 EXPORT_SYMBOL(__lock_sock);
1221 EXPORT_SYMBOL(__release_sock);
1222 EXPORT_SYMBOL(sk_alloc);
1223 EXPORT_SYMBOL(sk_free);
1224 EXPORT_SYMBOL(sk_send_sigurg);
1225 EXPORT_SYMBOL(sock_alloc_send_pskb);
1226 EXPORT_SYMBOL(sock_alloc_send_skb);
1227 EXPORT_SYMBOL(sock_getsockopt);
1228 EXPORT_SYMBOL(sock_init_data);
1229 EXPORT_SYMBOL(sock_kfree_s);
1230 EXPORT_SYMBOL(sock_kmalloc);
1231 EXPORT_SYMBOL(sock_no_accept);
1232 EXPORT_SYMBOL(sock_no_bind);
1233 EXPORT_SYMBOL(sock_no_connect);
1234 EXPORT_SYMBOL(sock_no_getname);
1235 EXPORT_SYMBOL(sock_no_getsockopt);
1236 EXPORT_SYMBOL(sock_no_ioctl);
1237 EXPORT_SYMBOL(sock_no_listen);
1238 EXPORT_SYMBOL(sock_no_mmap);
1239 EXPORT_SYMBOL(sock_no_poll);
1240 EXPORT_SYMBOL(sock_no_recvmsg);
1241 EXPORT_SYMBOL(sock_no_release);
1242 EXPORT_SYMBOL(sock_no_sendmsg);
1243 EXPORT_SYMBOL(sock_no_sendpage);
1244 EXPORT_SYMBOL(sock_no_setsockopt);
1245 EXPORT_SYMBOL(sock_no_shutdown);
1246 EXPORT_SYMBOL(sock_no_socketpair);
1247 EXPORT_SYMBOL(sock_rfree);
1248 EXPORT_SYMBOL(sock_rmalloc);
1249 EXPORT_SYMBOL(sock_setsockopt);
1250 EXPORT_SYMBOL(sock_wfree);
1251 EXPORT_SYMBOL(sock_wmalloc);
1252 #ifdef CONFIG_SYSCTL
1253 EXPORT_SYMBOL(sysctl_optmem_max);
1254 EXPORT_SYMBOL(sysctl_rmem_max);
1255 EXPORT_SYMBOL(sysctl_wmem_max);