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>
124 #include <linux/vs_socket.h>
130 /* Take into consideration the size of the struct sk_buff overhead in the
131 * determination of these values, since that is non-constant across
132 * platforms. This makes socket queueing behavior and performance
133 * not depend upon such differences.
135 #define _SK_MEM_PACKETS 256
136 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
137 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
138 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
140 /* Run time adjustable parameters. */
141 __u32 sysctl_wmem_max = SK_WMEM_MAX;
142 __u32 sysctl_rmem_max = SK_RMEM_MAX;
143 __u32 sysctl_wmem_default = SK_WMEM_MAX;
144 __u32 sysctl_rmem_default = SK_RMEM_MAX;
146 /* Maximal space eaten by iovec or ancilliary data plus some space */
147 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
149 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
153 if (optlen < sizeof(tv))
155 if (copy_from_user(&tv, optval, sizeof(tv)))
158 *timeo_p = MAX_SCHEDULE_TIMEOUT;
159 if (tv.tv_sec == 0 && tv.tv_usec == 0)
161 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
162 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
166 static void sock_warn_obsolete_bsdism(const char *name)
169 static char warncomm[16];
170 if (strcmp(warncomm, current->comm) && warned < 5) {
171 strcpy(warncomm, current->comm);
172 printk(KERN_WARNING "process `%s' is using obsolete "
173 "%s SO_BSDCOMPAT\n", warncomm, name);
179 * This is meant for all protocols to use and covers goings on
180 * at the socket level. Everything here is generic.
183 int sock_setsockopt(struct socket *sock, int level, int optname,
184 char __user *optval, int optlen)
186 struct sock *sk=sock->sk;
187 struct sk_filter *filter;
194 * Options without arguments
197 #ifdef SO_DONTLINGER /* Compatibility item... */
200 sock_reset_flag(sk, SOCK_LINGER);
205 if(optlen<sizeof(int))
208 if (get_user(val, (int __user *)optval))
218 if(val && !capable(CAP_NET_ADMIN))
223 sk->sk_debug = valbool;
226 sk->sk_reuse = valbool;
233 sk->sk_localroute = valbool;
236 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
239 /* Don't error on this BSD doesn't and if you think
240 about it this is right. Otherwise apps have to
241 play 'guess the biggest size' games. RCVBUF/SNDBUF
242 are treated in BSD as hints */
244 if (val > sysctl_wmem_max)
245 val = sysctl_wmem_max;
247 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
248 if ((val * 2) < SOCK_MIN_SNDBUF)
249 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
251 sk->sk_sndbuf = val * 2;
254 * Wake up sending tasks if we
257 sk->sk_write_space(sk);
261 /* Don't error on this BSD doesn't and if you think
262 about it this is right. Otherwise apps have to
263 play 'guess the biggest size' games. RCVBUF/SNDBUF
264 are treated in BSD as hints */
266 if (val > sysctl_rmem_max)
267 val = sysctl_rmem_max;
269 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
270 /* FIXME: is this lower bound the right one? */
271 if ((val * 2) < SOCK_MIN_RCVBUF)
272 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
274 sk->sk_rcvbuf = val * 2;
279 if (sk->sk_protocol == IPPROTO_TCP)
280 tcp_set_keepalive(sk, valbool);
282 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
286 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
290 sk->sk_no_check = valbool;
294 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
295 sk->sk_priority = val;
301 if(optlen<sizeof(ling)) {
302 ret = -EINVAL; /* 1003.1g */
305 if (copy_from_user(&ling,optval,sizeof(ling))) {
310 sock_reset_flag(sk, SOCK_LINGER);
312 #if (BITS_PER_LONG == 32)
313 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
314 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
317 sk->sk_lingertime = ling.l_linger * HZ;
318 sock_set_flag(sk, SOCK_LINGER);
323 sock_warn_obsolete_bsdism("setsockopt");
328 set_bit(SOCK_PASS_CRED, &sock->flags);
330 clear_bit(SOCK_PASS_CRED, &sock->flags);
334 sk->sk_rcvtstamp = valbool;
336 sock_enable_timestamp(sk);
342 sk->sk_rcvlowat = val ? : 1;
346 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
350 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
353 #ifdef CONFIG_NETDEVICES
354 case SO_BINDTODEVICE:
356 char devname[IFNAMSIZ];
359 if (!capable(CAP_NET_RAW)) {
364 /* Bind this socket to a particular device like "eth0",
365 * as specified in the passed interface name. If the
366 * name is "" or the option length is zero the socket
371 sk->sk_bound_dev_if = 0;
373 if (optlen > IFNAMSIZ)
375 if (copy_from_user(devname, optval, optlen)) {
380 /* Remove any cached route for this socket. */
383 if (devname[0] == '\0') {
384 sk->sk_bound_dev_if = 0;
386 struct net_device *dev = dev_get_by_name(devname);
391 sk->sk_bound_dev_if = dev->ifindex;
400 case SO_ATTACH_FILTER:
402 if (optlen == sizeof(struct sock_fprog)) {
403 struct sock_fprog fprog;
406 if (copy_from_user(&fprog, optval, sizeof(fprog)))
409 ret = sk_attach_filter(&fprog, sk);
413 case SO_DETACH_FILTER:
414 spin_lock_bh(&sk->sk_lock.slock);
415 filter = sk->sk_filter;
417 sk->sk_filter = NULL;
418 spin_unlock_bh(&sk->sk_lock.slock);
419 sk_filter_release(sk, filter);
422 spin_unlock_bh(&sk->sk_lock.slock);
426 /* We implement the SO_SNDLOWAT etc to
427 not be settable (1003.1g 5.3) */
437 int sock_getsockopt(struct socket *sock, int level, int optname,
438 char __user *optval, int __user *optlen)
440 struct sock *sk = sock->sk;
449 unsigned int lv=sizeof(int),len;
451 if(get_user(len,optlen))
459 v.val = sk->sk_debug;
463 v.val = sk->sk_localroute;
467 v.val = !!sock_flag(sk, SOCK_BROADCAST);
471 v.val = sk->sk_sndbuf;
475 v.val = sk->sk_rcvbuf;
479 v.val = sk->sk_reuse;
483 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
491 v.val = -sock_error(sk);
493 v.val = xchg(&sk->sk_err_soft, 0);
497 v.val = !!sock_flag(sk, SOCK_URGINLINE);
501 v.val = sk->sk_no_check;
505 v.val = sk->sk_priority;
510 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
511 v.ling.l_linger = sk->sk_lingertime / HZ;
515 sock_warn_obsolete_bsdism("getsockopt");
519 v.val = sk->sk_rcvtstamp;
523 lv=sizeof(struct timeval);
524 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
528 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
529 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000) / HZ;
534 lv=sizeof(struct timeval);
535 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
539 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
540 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000) / HZ;
545 v.val = sk->sk_rcvlowat;
553 v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
557 if (len > sizeof(sk->sk_peercred))
558 len = sizeof(sk->sk_peercred);
559 if (copy_to_user(optval, &sk->sk_peercred, len))
567 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
571 if (copy_to_user(optval, address, len))
576 /* Dubious BSD thing... Probably nobody even uses it, but
577 * the UNIX standard wants it for whatever reason... -DaveM
580 v.val = sk->sk_state == TCP_LISTEN;
584 return security_socket_getpeersec(sock, optval, optlen, len);
587 return(-ENOPROTOOPT);
591 if (copy_to_user(optval, &v, len))
594 if (put_user(len, optlen))
599 static kmem_cache_t *sk_cachep;
602 * sk_alloc - All socket objects are allocated here
603 * @family - protocol family
604 * @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
605 * @zero_it - zeroes the allocated sock
606 * @slab - alternate slab
608 * All socket objects are allocated here. If @zero_it is non-zero
609 * it should have the size of the are to be zeroed, because the
610 * private slabcaches have different sizes of the generic struct sock.
611 * 1 has been kept as a way to say sizeof(struct sock).
613 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
615 struct sock *sk = NULL;
619 sk = kmem_cache_alloc(slab, priority);
623 zero_it == 1 ? sizeof(struct sock) : zero_it);
624 sk->sk_family = family;
631 if (security_sk_alloc(sk, family, priority)) {
632 kmem_cache_free(slab, sk);
639 void sk_free(struct sock *sk)
641 struct sk_filter *filter;
642 struct module *owner = sk->sk_owner;
647 filter = sk->sk_filter;
649 sk_filter_release(sk, filter);
650 sk->sk_filter = NULL;
653 sock_disable_timestamp(sk);
655 if (atomic_read(&sk->sk_omem_alloc))
656 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
657 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
659 security_sk_free(sk);
660 BUG_ON(sk->sk_vx_info);
661 BUG_ON(sk->sk_nx_info);
662 /* clr_vx_info(&sk->sk_vx_info);
663 clr_nx_info(&sk->sk_nx_info); */
664 kmem_cache_free(sk->sk_slab, sk);
668 void __init sk_init(void)
670 sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
671 SLAB_HWCACHE_ALIGN, 0, 0);
673 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
675 if (num_physpages <= 4096) {
676 sysctl_wmem_max = 32767;
677 sysctl_rmem_max = 32767;
678 sysctl_wmem_default = 32767;
679 sysctl_rmem_default = 32767;
680 } else if (num_physpages >= 131072) {
681 sysctl_wmem_max = 131071;
682 sysctl_rmem_max = 131071;
687 * Simple resource managers for sockets.
692 * Write buffer destructor automatically called from kfree_skb.
694 void sock_wfree(struct sk_buff *skb)
696 struct sock *sk = skb->sk;
698 /* In case it might be waiting for more memory. */
699 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
700 if (!sk->sk_use_write_queue)
701 sk->sk_write_space(sk);
706 * Read buffer destructor automatically called from kfree_skb.
708 void sock_rfree(struct sk_buff *skb)
710 struct sock *sk = skb->sk;
712 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
716 * Allocate a skb from the socket's send buffer.
718 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
720 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
721 struct sk_buff * skb = alloc_skb(size, priority);
723 skb_set_owner_w(skb, sk);
731 * Allocate a skb from the socket's receive buffer.
733 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
735 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
736 struct sk_buff *skb = alloc_skb(size, priority);
738 skb_set_owner_r(skb, sk);
746 * Allocate a memory block from the socket's option memory buffer.
748 void *sock_kmalloc(struct sock *sk, int size, int priority)
750 if ((unsigned)size <= sysctl_optmem_max &&
751 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
753 /* First do the add, to avoid the race if kmalloc
756 atomic_add(size, &sk->sk_omem_alloc);
757 mem = kmalloc(size, priority);
760 atomic_sub(size, &sk->sk_omem_alloc);
766 * Free an option memory block.
768 void sock_kfree_s(struct sock *sk, void *mem, int size)
771 atomic_sub(size, &sk->sk_omem_alloc);
774 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
775 I think, these locks should be removed for datagram sockets.
777 static long sock_wait_for_wmem(struct sock * sk, long timeo)
781 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
785 if (signal_pending(current))
787 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
788 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
789 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
791 if (sk->sk_shutdown & SEND_SHUTDOWN)
795 timeo = schedule_timeout(timeo);
797 finish_wait(sk->sk_sleep, &wait);
803 * Generic send/receive buffer handlers
806 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
807 unsigned long data_len, int noblock, int *errcode)
810 unsigned int gfp_mask;
814 gfp_mask = sk->sk_allocation;
815 if (gfp_mask & __GFP_WAIT)
816 gfp_mask |= __GFP_REPEAT;
818 timeo = sock_sndtimeo(sk, noblock);
820 err = sock_error(sk);
825 if (sk->sk_shutdown & SEND_SHUTDOWN)
828 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
829 skb = alloc_skb(header_len, sk->sk_allocation);
834 /* No pages, we're done... */
838 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
839 skb->truesize += data_len;
840 skb_shinfo(skb)->nr_frags = npages;
841 for (i = 0; i < npages; i++) {
845 page = alloc_pages(sk->sk_allocation, 0);
848 skb_shinfo(skb)->nr_frags = i;
853 frag = &skb_shinfo(skb)->frags[i];
855 frag->page_offset = 0;
856 frag->size = (data_len >= PAGE_SIZE ?
859 data_len -= PAGE_SIZE;
862 /* Full success... */
868 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
869 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
873 if (signal_pending(current))
875 timeo = sock_wait_for_wmem(sk, timeo);
878 skb_set_owner_w(skb, sk);
882 err = sock_intr_errno(timeo);
888 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
889 int noblock, int *errcode)
891 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
894 void __lock_sock(struct sock *sk)
899 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
900 TASK_UNINTERRUPTIBLE);
901 spin_unlock_bh(&sk->sk_lock.slock);
903 spin_lock_bh(&sk->sk_lock.slock);
904 if(!sock_owned_by_user(sk))
907 finish_wait(&sk->sk_lock.wq, &wait);
910 void __release_sock(struct sock *sk)
912 struct sk_buff *skb = sk->sk_backlog.head;
915 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
919 struct sk_buff *next = skb->next;
922 sk->sk_backlog_rcv(sk, skb);
924 } while (skb != NULL);
927 } while((skb = sk->sk_backlog.head) != NULL);
931 * sk_wait_data - wait for data to arrive at sk_receive_queue
932 * sk - sock to wait on
933 * timeo - for how long
935 * Now socket state including sk->sk_err is changed only under lock,
936 * hence we may omit checks after joining wait queue.
937 * We check receive queue before schedule() only as optimization;
938 * it is very likely that release_sock() added new data.
940 int sk_wait_data(struct sock *sk, long *timeo)
945 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
946 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
947 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
948 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
949 finish_wait(sk->sk_sleep, &wait);
953 EXPORT_SYMBOL(sk_wait_data);
956 * Set of default routines for initialising struct proto_ops when
957 * the protocol does not support a particular function. In certain
958 * cases where it makes no sense for a protocol to have a "do nothing"
959 * function, some default processing is provided.
962 int sock_no_release(struct socket *sock)
967 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
972 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
978 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
983 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
988 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
994 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
999 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1004 int sock_no_listen(struct socket *sock, int backlog)
1009 int sock_no_shutdown(struct socket *sock, int how)
1014 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1015 char __user *optval, int optlen)
1020 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1021 char __user *optval, int __user *optlen)
1026 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1032 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1033 size_t len, int flags)
1038 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1040 /* Mirror missing mmap method error code */
1044 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1049 mm_segment_t old_fs;
1054 msg.msg_name = NULL;
1055 msg.msg_namelen = 0;
1058 msg.msg_control = NULL;
1059 msg.msg_controllen = 0;
1060 msg.msg_flags = flags;
1062 /* This cast is ok because of the "set_fs(KERNEL_DS)" */
1063 iov.iov_base = (void __user *) (kaddr + offset);
1068 res = sock_sendmsg(sock, &msg, size);
1076 * Default Socket Callbacks
1079 void sock_def_wakeup(struct sock *sk)
1081 read_lock(&sk->sk_callback_lock);
1082 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1083 wake_up_interruptible_all(sk->sk_sleep);
1084 read_unlock(&sk->sk_callback_lock);
1087 void sock_def_error_report(struct sock *sk)
1089 read_lock(&sk->sk_callback_lock);
1090 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1091 wake_up_interruptible(sk->sk_sleep);
1092 sk_wake_async(sk,0,POLL_ERR);
1093 read_unlock(&sk->sk_callback_lock);
1096 void sock_def_readable(struct sock *sk, int len)
1098 read_lock(&sk->sk_callback_lock);
1099 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1100 wake_up_interruptible(sk->sk_sleep);
1101 sk_wake_async(sk,1,POLL_IN);
1102 read_unlock(&sk->sk_callback_lock);
1105 void sock_def_write_space(struct sock *sk)
1107 read_lock(&sk->sk_callback_lock);
1109 /* Do not wake up a writer until he can make "significant"
1112 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1113 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1114 wake_up_interruptible(sk->sk_sleep);
1116 /* Should agree with poll, otherwise some programs break */
1117 if (sock_writeable(sk))
1118 sk_wake_async(sk, 2, POLL_OUT);
1121 read_unlock(&sk->sk_callback_lock);
1124 void sock_def_destruct(struct sock *sk)
1126 if (sk->sk_protinfo)
1127 kfree(sk->sk_protinfo);
1130 void sk_send_sigurg(struct sock *sk)
1132 if (sk->sk_socket && sk->sk_socket->file)
1133 if (send_sigurg(&sk->sk_socket->file->f_owner))
1134 sk_wake_async(sk, 3, POLL_PRI);
1137 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1138 unsigned long expires)
1140 if (!mod_timer(timer, expires))
1144 EXPORT_SYMBOL(sk_reset_timer);
1146 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1148 if (timer_pending(timer) && del_timer(timer))
1152 EXPORT_SYMBOL(sk_stop_timer);
1154 void sock_init_data(struct socket *sock, struct sock *sk)
1156 skb_queue_head_init(&sk->sk_receive_queue);
1157 skb_queue_head_init(&sk->sk_write_queue);
1158 skb_queue_head_init(&sk->sk_error_queue);
1160 init_timer(&sk->sk_timer);
1162 sk->sk_allocation = GFP_KERNEL;
1163 sk->sk_rcvbuf = sysctl_rmem_default;
1164 sk->sk_sndbuf = sysctl_wmem_default;
1165 sk->sk_state = TCP_CLOSE;
1167 sk->sk_socket = sock;
1171 sk->sk_type = sock->type;
1172 sk->sk_sleep = &sock->wait;
1175 sk->sk_sleep = NULL;
1177 sk->sk_dst_lock = RW_LOCK_UNLOCKED;
1178 sk->sk_callback_lock = RW_LOCK_UNLOCKED;
1180 sk->sk_state_change = sock_def_wakeup;
1181 sk->sk_data_ready = sock_def_readable;
1182 sk->sk_write_space = sock_def_write_space;
1183 sk->sk_error_report = sock_def_error_report;
1184 sk->sk_destruct = sock_def_destruct;
1186 sk->sk_peercred.pid = 0;
1187 sk->sk_peercred.uid = -1;
1188 sk->sk_peercred.gid = -1;
1189 sk->sk_rcvlowat = 1;
1190 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1191 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1192 sk->sk_owner = NULL;
1194 sk->sk_stamp.tv_sec = -1L;
1195 sk->sk_stamp.tv_usec = -1L;
1197 sk->sk_vx_info = NULL;
1199 sk->sk_nx_info = NULL;
1202 atomic_set(&sk->sk_refcnt, 1);
1205 void fastcall lock_sock(struct sock *sk)
1208 spin_lock_bh(&(sk->sk_lock.slock));
1209 if (sk->sk_lock.owner)
1211 sk->sk_lock.owner = (void *)1;
1212 spin_unlock_bh(&(sk->sk_lock.slock));
1215 EXPORT_SYMBOL(lock_sock);
1217 void fastcall release_sock(struct sock *sk)
1219 spin_lock_bh(&(sk->sk_lock.slock));
1220 if (sk->sk_backlog.tail)
1222 sk->sk_lock.owner = NULL;
1223 if (waitqueue_active(&(sk->sk_lock.wq)))
1224 wake_up(&(sk->sk_lock.wq));
1225 spin_unlock_bh(&(sk->sk_lock.slock));
1227 EXPORT_SYMBOL(release_sock);
1229 /* When > 0 there are consumers of rx skb time stamps */
1230 atomic_t netstamp_needed = ATOMIC_INIT(0);
1232 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1234 if (!sock_flag(sk, SOCK_TIMESTAMP))
1235 sock_enable_timestamp(sk);
1236 if (sk->sk_stamp.tv_sec == -1)
1238 if (sk->sk_stamp.tv_sec == 0)
1239 do_gettimeofday(&sk->sk_stamp);
1240 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1243 EXPORT_SYMBOL(sock_get_timestamp);
1245 void sock_enable_timestamp(struct sock *sk)
1247 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1248 sock_set_flag(sk, SOCK_TIMESTAMP);
1249 atomic_inc(&netstamp_needed);
1252 EXPORT_SYMBOL(sock_enable_timestamp);
1254 void sock_disable_timestamp(struct sock *sk)
1256 if (sock_flag(sk, SOCK_TIMESTAMP)) {
1257 sock_reset_flag(sk, SOCK_TIMESTAMP);
1258 atomic_dec(&netstamp_needed);
1261 EXPORT_SYMBOL(sock_disable_timestamp);
1263 EXPORT_SYMBOL(__lock_sock);
1264 EXPORT_SYMBOL(__release_sock);
1265 EXPORT_SYMBOL(sk_alloc);
1266 EXPORT_SYMBOL(sk_free);
1267 EXPORT_SYMBOL(sk_send_sigurg);
1268 EXPORT_SYMBOL(sock_alloc_send_pskb);
1269 EXPORT_SYMBOL(sock_alloc_send_skb);
1270 EXPORT_SYMBOL(sock_getsockopt);
1271 EXPORT_SYMBOL(sock_init_data);
1272 EXPORT_SYMBOL(sock_kfree_s);
1273 EXPORT_SYMBOL(sock_kmalloc);
1274 EXPORT_SYMBOL(sock_no_accept);
1275 EXPORT_SYMBOL(sock_no_bind);
1276 EXPORT_SYMBOL(sock_no_connect);
1277 EXPORT_SYMBOL(sock_no_getname);
1278 EXPORT_SYMBOL(sock_no_getsockopt);
1279 EXPORT_SYMBOL(sock_no_ioctl);
1280 EXPORT_SYMBOL(sock_no_listen);
1281 EXPORT_SYMBOL(sock_no_mmap);
1282 EXPORT_SYMBOL(sock_no_poll);
1283 EXPORT_SYMBOL(sock_no_recvmsg);
1284 EXPORT_SYMBOL(sock_no_release);
1285 EXPORT_SYMBOL(sock_no_sendmsg);
1286 EXPORT_SYMBOL(sock_no_sendpage);
1287 EXPORT_SYMBOL(sock_no_setsockopt);
1288 EXPORT_SYMBOL(sock_no_shutdown);
1289 EXPORT_SYMBOL(sock_no_socketpair);
1290 EXPORT_SYMBOL(sock_rfree);
1291 EXPORT_SYMBOL(sock_rmalloc);
1292 EXPORT_SYMBOL(sock_setsockopt);
1293 EXPORT_SYMBOL(sock_wfree);
1294 EXPORT_SYMBOL(sock_wmalloc);
1295 #ifdef CONFIG_SYSCTL
1296 EXPORT_SYMBOL(sysctl_optmem_max);
1297 EXPORT_SYMBOL(sysctl_rmem_max);
1298 EXPORT_SYMBOL(sysctl_wmem_max);