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 kmem_cache_free(sk->sk_slab, sk);
666 void __init sk_init(void)
668 sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
669 SLAB_HWCACHE_ALIGN, 0, 0);
671 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
673 if (num_physpages <= 4096) {
674 sysctl_wmem_max = 32767;
675 sysctl_rmem_max = 32767;
676 sysctl_wmem_default = 32767;
677 sysctl_rmem_default = 32767;
678 } else if (num_physpages >= 131072) {
679 sysctl_wmem_max = 131071;
680 sysctl_rmem_max = 131071;
685 * Simple resource managers for sockets.
690 * Write buffer destructor automatically called from kfree_skb.
692 void sock_wfree(struct sk_buff *skb)
694 struct sock *sk = skb->sk;
696 /* In case it might be waiting for more memory. */
697 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
698 if (!sk->sk_use_write_queue)
699 sk->sk_write_space(sk);
704 * Read buffer destructor automatically called from kfree_skb.
706 void sock_rfree(struct sk_buff *skb)
708 struct sock *sk = skb->sk;
710 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
714 * Allocate a skb from the socket's send buffer.
716 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
718 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
719 struct sk_buff * skb = alloc_skb(size, priority);
721 skb_set_owner_w(skb, sk);
729 * Allocate a skb from the socket's receive buffer.
731 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
733 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
734 struct sk_buff *skb = alloc_skb(size, priority);
736 skb_set_owner_r(skb, sk);
744 * Allocate a memory block from the socket's option memory buffer.
746 void *sock_kmalloc(struct sock *sk, int size, int priority)
748 if ((unsigned)size <= sysctl_optmem_max &&
749 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
751 /* First do the add, to avoid the race if kmalloc
754 atomic_add(size, &sk->sk_omem_alloc);
755 mem = kmalloc(size, priority);
758 atomic_sub(size, &sk->sk_omem_alloc);
764 * Free an option memory block.
766 void sock_kfree_s(struct sock *sk, void *mem, int size)
769 atomic_sub(size, &sk->sk_omem_alloc);
772 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
773 I think, these locks should be removed for datagram sockets.
775 static long sock_wait_for_wmem(struct sock * sk, long timeo)
779 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
783 if (signal_pending(current))
785 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
786 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
787 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
789 if (sk->sk_shutdown & SEND_SHUTDOWN)
793 timeo = schedule_timeout(timeo);
795 finish_wait(sk->sk_sleep, &wait);
801 * Generic send/receive buffer handlers
804 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
805 unsigned long data_len, int noblock, int *errcode)
808 unsigned int gfp_mask;
812 gfp_mask = sk->sk_allocation;
813 if (gfp_mask & __GFP_WAIT)
814 gfp_mask |= __GFP_REPEAT;
816 timeo = sock_sndtimeo(sk, noblock);
818 err = sock_error(sk);
823 if (sk->sk_shutdown & SEND_SHUTDOWN)
826 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
827 skb = alloc_skb(header_len, sk->sk_allocation);
832 /* No pages, we're done... */
836 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
837 skb->truesize += data_len;
838 skb_shinfo(skb)->nr_frags = npages;
839 for (i = 0; i < npages; i++) {
843 page = alloc_pages(sk->sk_allocation, 0);
846 skb_shinfo(skb)->nr_frags = i;
851 frag = &skb_shinfo(skb)->frags[i];
853 frag->page_offset = 0;
854 frag->size = (data_len >= PAGE_SIZE ?
857 data_len -= PAGE_SIZE;
860 /* Full success... */
866 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
867 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
871 if (signal_pending(current))
873 timeo = sock_wait_for_wmem(sk, timeo);
876 skb_set_owner_w(skb, sk);
880 err = sock_intr_errno(timeo);
886 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
887 int noblock, int *errcode)
889 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
892 void __lock_sock(struct sock *sk)
897 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
898 TASK_UNINTERRUPTIBLE);
899 spin_unlock_bh(&sk->sk_lock.slock);
901 spin_lock_bh(&sk->sk_lock.slock);
902 if(!sock_owned_by_user(sk))
905 finish_wait(&sk->sk_lock.wq, &wait);
908 void __release_sock(struct sock *sk)
910 struct sk_buff *skb = sk->sk_backlog.head;
913 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
917 struct sk_buff *next = skb->next;
920 sk->sk_backlog_rcv(sk, skb);
922 } while (skb != NULL);
925 } while((skb = sk->sk_backlog.head) != NULL);
929 * sk_wait_data - wait for data to arrive at sk_receive_queue
930 * sk - sock to wait on
931 * timeo - for how long
933 * Now socket state including sk->sk_err is changed only under lock,
934 * hence we may omit checks after joining wait queue.
935 * We check receive queue before schedule() only as optimization;
936 * it is very likely that release_sock() added new data.
938 int sk_wait_data(struct sock *sk, long *timeo)
943 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
944 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
945 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
946 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
947 finish_wait(sk->sk_sleep, &wait);
951 EXPORT_SYMBOL(sk_wait_data);
954 * Set of default routines for initialising struct proto_ops when
955 * the protocol does not support a particular function. In certain
956 * cases where it makes no sense for a protocol to have a "do nothing"
957 * function, some default processing is provided.
960 int sock_no_release(struct socket *sock)
965 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
970 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
976 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
981 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
986 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
992 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
997 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1002 int sock_no_listen(struct socket *sock, int backlog)
1007 int sock_no_shutdown(struct socket *sock, int how)
1012 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1013 char __user *optval, int optlen)
1018 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1019 char __user *optval, int __user *optlen)
1024 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1030 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1031 size_t len, int flags)
1036 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1038 /* Mirror missing mmap method error code */
1042 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1047 mm_segment_t old_fs;
1052 msg.msg_name = NULL;
1053 msg.msg_namelen = 0;
1056 msg.msg_control = NULL;
1057 msg.msg_controllen = 0;
1058 msg.msg_flags = flags;
1060 /* This cast is ok because of the "set_fs(KERNEL_DS)" */
1061 iov.iov_base = (void __user *) (kaddr + offset);
1066 res = sock_sendmsg(sock, &msg, size);
1074 * Default Socket Callbacks
1077 void sock_def_wakeup(struct sock *sk)
1079 read_lock(&sk->sk_callback_lock);
1080 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1081 wake_up_interruptible_all(sk->sk_sleep);
1082 read_unlock(&sk->sk_callback_lock);
1085 void sock_def_error_report(struct sock *sk)
1087 read_lock(&sk->sk_callback_lock);
1088 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1089 wake_up_interruptible(sk->sk_sleep);
1090 sk_wake_async(sk,0,POLL_ERR);
1091 read_unlock(&sk->sk_callback_lock);
1094 void sock_def_readable(struct sock *sk, int len)
1096 read_lock(&sk->sk_callback_lock);
1097 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1098 wake_up_interruptible(sk->sk_sleep);
1099 sk_wake_async(sk,1,POLL_IN);
1100 read_unlock(&sk->sk_callback_lock);
1103 void sock_def_write_space(struct sock *sk)
1105 read_lock(&sk->sk_callback_lock);
1107 /* Do not wake up a writer until he can make "significant"
1110 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1111 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1112 wake_up_interruptible(sk->sk_sleep);
1114 /* Should agree with poll, otherwise some programs break */
1115 if (sock_writeable(sk))
1116 sk_wake_async(sk, 2, POLL_OUT);
1119 read_unlock(&sk->sk_callback_lock);
1122 void sock_def_destruct(struct sock *sk)
1124 if (sk->sk_protinfo)
1125 kfree(sk->sk_protinfo);
1128 void sk_send_sigurg(struct sock *sk)
1130 if (sk->sk_socket && sk->sk_socket->file)
1131 if (send_sigurg(&sk->sk_socket->file->f_owner))
1132 sk_wake_async(sk, 3, POLL_PRI);
1135 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1136 unsigned long expires)
1138 if (!mod_timer(timer, expires))
1142 EXPORT_SYMBOL(sk_reset_timer);
1144 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1146 if (timer_pending(timer) && del_timer(timer))
1150 EXPORT_SYMBOL(sk_stop_timer);
1152 void sock_init_data(struct socket *sock, struct sock *sk)
1154 skb_queue_head_init(&sk->sk_receive_queue);
1155 skb_queue_head_init(&sk->sk_write_queue);
1156 skb_queue_head_init(&sk->sk_error_queue);
1158 init_timer(&sk->sk_timer);
1160 sk->sk_allocation = GFP_KERNEL;
1161 sk->sk_rcvbuf = sysctl_rmem_default;
1162 sk->sk_sndbuf = sysctl_wmem_default;
1163 sk->sk_state = TCP_CLOSE;
1165 sk->sk_socket = sock;
1169 sk->sk_type = sock->type;
1170 sk->sk_sleep = &sock->wait;
1173 sk->sk_sleep = NULL;
1175 sk->sk_dst_lock = RW_LOCK_UNLOCKED;
1176 sk->sk_callback_lock = RW_LOCK_UNLOCKED;
1178 sk->sk_state_change = sock_def_wakeup;
1179 sk->sk_data_ready = sock_def_readable;
1180 sk->sk_write_space = sock_def_write_space;
1181 sk->sk_error_report = sock_def_error_report;
1182 sk->sk_destruct = sock_def_destruct;
1184 sk->sk_peercred.pid = 0;
1185 sk->sk_peercred.uid = -1;
1186 sk->sk_peercred.gid = -1;
1187 sk->sk_rcvlowat = 1;
1188 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1189 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1190 sk->sk_owner = NULL;
1192 sk->sk_stamp.tv_sec = -1L;
1193 sk->sk_stamp.tv_usec = -1L;
1195 sk->sk_vx_info = NULL;
1197 sk->sk_nx_info = NULL;
1200 atomic_set(&sk->sk_refcnt, 1);
1203 void fastcall lock_sock(struct sock *sk)
1206 spin_lock_bh(&(sk->sk_lock.slock));
1207 if (sk->sk_lock.owner)
1209 sk->sk_lock.owner = (void *)1;
1210 spin_unlock_bh(&(sk->sk_lock.slock));
1213 EXPORT_SYMBOL(lock_sock);
1215 void fastcall release_sock(struct sock *sk)
1217 spin_lock_bh(&(sk->sk_lock.slock));
1218 if (sk->sk_backlog.tail)
1220 sk->sk_lock.owner = NULL;
1221 if (waitqueue_active(&(sk->sk_lock.wq)))
1222 wake_up(&(sk->sk_lock.wq));
1223 spin_unlock_bh(&(sk->sk_lock.slock));
1225 EXPORT_SYMBOL(release_sock);
1227 /* When > 0 there are consumers of rx skb time stamps */
1228 atomic_t netstamp_needed = ATOMIC_INIT(0);
1230 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1232 if (!sock_flag(sk, SOCK_TIMESTAMP))
1233 sock_enable_timestamp(sk);
1234 if (sk->sk_stamp.tv_sec == -1)
1236 if (sk->sk_stamp.tv_sec == 0)
1237 do_gettimeofday(&sk->sk_stamp);
1238 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1241 EXPORT_SYMBOL(sock_get_timestamp);
1243 void sock_enable_timestamp(struct sock *sk)
1245 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1246 sock_set_flag(sk, SOCK_TIMESTAMP);
1247 atomic_inc(&netstamp_needed);
1250 EXPORT_SYMBOL(sock_enable_timestamp);
1252 void sock_disable_timestamp(struct sock *sk)
1254 if (sock_flag(sk, SOCK_TIMESTAMP)) {
1255 sock_reset_flag(sk, SOCK_TIMESTAMP);
1256 atomic_dec(&netstamp_needed);
1259 EXPORT_SYMBOL(sock_disable_timestamp);
1261 EXPORT_SYMBOL(__lock_sock);
1262 EXPORT_SYMBOL(__release_sock);
1263 EXPORT_SYMBOL(sk_alloc);
1264 EXPORT_SYMBOL(sk_free);
1265 EXPORT_SYMBOL(sk_send_sigurg);
1266 EXPORT_SYMBOL(sock_alloc_send_pskb);
1267 EXPORT_SYMBOL(sock_alloc_send_skb);
1268 EXPORT_SYMBOL(sock_getsockopt);
1269 EXPORT_SYMBOL(sock_init_data);
1270 EXPORT_SYMBOL(sock_kfree_s);
1271 EXPORT_SYMBOL(sock_kmalloc);
1272 EXPORT_SYMBOL(sock_no_accept);
1273 EXPORT_SYMBOL(sock_no_bind);
1274 EXPORT_SYMBOL(sock_no_connect);
1275 EXPORT_SYMBOL(sock_no_getname);
1276 EXPORT_SYMBOL(sock_no_getsockopt);
1277 EXPORT_SYMBOL(sock_no_ioctl);
1278 EXPORT_SYMBOL(sock_no_listen);
1279 EXPORT_SYMBOL(sock_no_mmap);
1280 EXPORT_SYMBOL(sock_no_poll);
1281 EXPORT_SYMBOL(sock_no_recvmsg);
1282 EXPORT_SYMBOL(sock_no_release);
1283 EXPORT_SYMBOL(sock_no_sendmsg);
1284 EXPORT_SYMBOL(sock_no_sendpage);
1285 EXPORT_SYMBOL(sock_no_setsockopt);
1286 EXPORT_SYMBOL(sock_no_shutdown);
1287 EXPORT_SYMBOL(sock_no_socketpair);
1288 EXPORT_SYMBOL(sock_rfree);
1289 EXPORT_SYMBOL(sock_rmalloc);
1290 EXPORT_SYMBOL(sock_setsockopt);
1291 EXPORT_SYMBOL(sock_wfree);
1292 EXPORT_SYMBOL(sock_wmalloc);
1293 #ifdef CONFIG_SYSCTL
1294 EXPORT_SYMBOL(sysctl_optmem_max);
1295 EXPORT_SYMBOL(sysctl_rmem_max);
1296 EXPORT_SYMBOL(sysctl_wmem_max);