2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
94 #include <linux/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/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <net/xfrm.h>
124 #include <linux/ipsec.h>
126 #include <linux/filter.h>
132 /* Take into consideration the size of the struct sk_buff overhead in the
133 * determination of these values, since that is non-constant across
134 * platforms. This makes socket queueing behavior and performance
135 * not depend upon such differences.
137 #define _SK_MEM_PACKETS 256
138 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
139 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
140 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
142 /* Run time adjustable parameters. */
143 __u32 sysctl_wmem_max = SK_WMEM_MAX;
144 __u32 sysctl_rmem_max = SK_RMEM_MAX;
145 __u32 sysctl_wmem_default = SK_WMEM_MAX;
146 __u32 sysctl_rmem_default = SK_RMEM_MAX;
148 /* Maximal space eaten by iovec or ancilliary data plus some space */
149 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
151 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
155 if (optlen < sizeof(tv))
157 if (copy_from_user(&tv, optval, sizeof(tv)))
160 *timeo_p = MAX_SCHEDULE_TIMEOUT;
161 if (tv.tv_sec == 0 && tv.tv_usec == 0)
163 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
164 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
168 static void sock_warn_obsolete_bsdism(const char *name)
171 static char warncomm[TASK_COMM_LEN];
172 if (strcmp(warncomm, current->comm) && warned < 5) {
173 strcpy(warncomm, current->comm);
174 printk(KERN_WARNING "process `%s' is using obsolete "
175 "%s SO_BSDCOMPAT\n", warncomm, name);
180 static void sock_disable_timestamp(struct sock *sk)
182 if (sock_flag(sk, SOCK_TIMESTAMP)) {
183 sock_reset_flag(sk, SOCK_TIMESTAMP);
184 net_disable_timestamp();
190 * This is meant for all protocols to use and covers goings on
191 * at the socket level. Everything here is generic.
194 int sock_setsockopt(struct socket *sock, int level, int optname,
195 char __user *optval, int optlen)
197 struct sock *sk=sock->sk;
198 struct sk_filter *filter;
205 * Options without arguments
208 #ifdef SO_DONTLINGER /* Compatibility item... */
211 sock_reset_flag(sk, SOCK_LINGER);
216 if(optlen<sizeof(int))
219 if (get_user(val, (int __user *)optval))
229 if(val && !capable(CAP_NET_ADMIN))
234 sock_set_flag(sk, SOCK_DBG);
236 sock_reset_flag(sk, SOCK_DBG);
239 sk->sk_reuse = valbool;
247 sock_set_flag(sk, SOCK_LOCALROUTE);
249 sock_reset_flag(sk, SOCK_LOCALROUTE);
252 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
255 /* Don't error on this BSD doesn't and if you think
256 about it this is right. Otherwise apps have to
257 play 'guess the biggest size' games. RCVBUF/SNDBUF
258 are treated in BSD as hints */
260 if (val > sysctl_wmem_max)
261 val = sysctl_wmem_max;
263 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
264 if ((val * 2) < SOCK_MIN_SNDBUF)
265 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
267 sk->sk_sndbuf = val * 2;
270 * Wake up sending tasks if we
273 sk->sk_write_space(sk);
277 /* Don't error on this BSD doesn't and if you think
278 about it this is right. Otherwise apps have to
279 play 'guess the biggest size' games. RCVBUF/SNDBUF
280 are treated in BSD as hints */
282 if (val > sysctl_rmem_max)
283 val = sysctl_rmem_max;
285 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
286 /* FIXME: is this lower bound the right one? */
287 if ((val * 2) < SOCK_MIN_RCVBUF)
288 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
290 sk->sk_rcvbuf = val * 2;
295 if (sk->sk_protocol == IPPROTO_TCP)
296 tcp_set_keepalive(sk, valbool);
298 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
302 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
306 sk->sk_no_check = valbool;
310 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
311 sk->sk_priority = val;
317 if(optlen<sizeof(ling)) {
318 ret = -EINVAL; /* 1003.1g */
321 if (copy_from_user(&ling,optval,sizeof(ling))) {
326 sock_reset_flag(sk, SOCK_LINGER);
328 #if (BITS_PER_LONG == 32)
329 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
330 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
333 sk->sk_lingertime = ling.l_linger * HZ;
334 sock_set_flag(sk, SOCK_LINGER);
339 sock_warn_obsolete_bsdism("setsockopt");
344 set_bit(SOCK_PASSCRED, &sock->flags);
346 clear_bit(SOCK_PASSCRED, &sock->flags);
351 sock_set_flag(sk, SOCK_RCVTSTAMP);
352 sock_enable_timestamp(sk);
354 sock_reset_flag(sk, SOCK_RCVTSTAMP);
360 sk->sk_rcvlowat = val ? : 1;
364 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
368 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
371 #ifdef CONFIG_NETDEVICES
372 case SO_BINDTODEVICE:
374 char devname[IFNAMSIZ];
377 if (!capable(CAP_NET_RAW)) {
382 /* Bind this socket to a particular device like "eth0",
383 * as specified in the passed interface name. If the
384 * name is "" or the option length is zero the socket
389 sk->sk_bound_dev_if = 0;
391 if (optlen > IFNAMSIZ)
393 if (copy_from_user(devname, optval, optlen)) {
398 /* Remove any cached route for this socket. */
401 if (devname[0] == '\0') {
402 sk->sk_bound_dev_if = 0;
404 struct net_device *dev = dev_get_by_name(devname);
409 sk->sk_bound_dev_if = dev->ifindex;
418 case SO_ATTACH_FILTER:
420 if (optlen == sizeof(struct sock_fprog)) {
421 struct sock_fprog fprog;
424 if (copy_from_user(&fprog, optval, sizeof(fprog)))
427 ret = sk_attach_filter(&fprog, sk);
431 case SO_DETACH_FILTER:
432 spin_lock_bh(&sk->sk_lock.slock);
433 filter = sk->sk_filter;
435 sk->sk_filter = NULL;
436 spin_unlock_bh(&sk->sk_lock.slock);
437 sk_filter_release(sk, filter);
440 spin_unlock_bh(&sk->sk_lock.slock);
444 /* We implement the SO_SNDLOWAT etc to
445 not be settable (1003.1g 5.3) */
455 int sock_getsockopt(struct socket *sock, int level, int optname,
456 char __user *optval, int __user *optlen)
458 struct sock *sk = sock->sk;
467 unsigned int lv = sizeof(int);
470 if(get_user(len,optlen))
478 v.val = sock_flag(sk, SOCK_DBG);
482 v.val = sock_flag(sk, SOCK_LOCALROUTE);
486 v.val = !!sock_flag(sk, SOCK_BROADCAST);
490 v.val = sk->sk_sndbuf;
494 v.val = sk->sk_rcvbuf;
498 v.val = sk->sk_reuse;
502 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
510 v.val = -sock_error(sk);
512 v.val = xchg(&sk->sk_err_soft, 0);
516 v.val = !!sock_flag(sk, SOCK_URGINLINE);
520 v.val = sk->sk_no_check;
524 v.val = sk->sk_priority;
529 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
530 v.ling.l_linger = sk->sk_lingertime / HZ;
534 sock_warn_obsolete_bsdism("getsockopt");
538 v.val = sock_flag(sk, SOCK_RCVTSTAMP);
542 lv=sizeof(struct timeval);
543 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
547 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
548 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
553 lv=sizeof(struct timeval);
554 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
558 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
559 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
564 v.val = sk->sk_rcvlowat;
572 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
576 if (len > sizeof(sk->sk_peercred))
577 len = sizeof(sk->sk_peercred);
578 if (copy_to_user(optval, &sk->sk_peercred, len))
586 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
590 if (copy_to_user(optval, address, len))
595 /* Dubious BSD thing... Probably nobody even uses it, but
596 * the UNIX standard wants it for whatever reason... -DaveM
599 v.val = sk->sk_state == TCP_LISTEN;
603 return security_socket_getpeersec(sock, optval, optlen, len);
606 return(-ENOPROTOOPT);
610 if (copy_to_user(optval, &v, len))
613 if (put_user(len, optlen))
619 * sk_alloc - All socket objects are allocated here
620 * @family: protocol family
621 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
622 * @prot: struct proto associated with this new sock instance
623 * @zero_it: if we should zero the newly allocated sock
625 struct sock *sk_alloc(int family, unsigned int __nocast priority,
626 struct proto *prot, int zero_it)
628 struct sock *sk = NULL;
629 kmem_cache_t *slab = prot->slab;
632 sk = kmem_cache_alloc(slab, priority);
634 sk = kmalloc(prot->obj_size, priority);
638 memset(sk, 0, prot->obj_size);
639 sk->sk_family = family;
641 * See comment in struct sock definition to understand
642 * why we need sk_prot_creator -acme
644 sk->sk_prot = sk->sk_prot_creator = prot;
648 if (security_sk_alloc(sk, family, priority)) {
650 kmem_cache_free(slab, sk);
655 __module_get(prot->owner);
660 void sk_free(struct sock *sk)
662 struct sk_filter *filter;
663 struct module *owner = sk->sk_prot_creator->owner;
668 filter = sk->sk_filter;
670 sk_filter_release(sk, filter);
671 sk->sk_filter = NULL;
674 sock_disable_timestamp(sk);
676 if (atomic_read(&sk->sk_omem_alloc))
677 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
678 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
680 security_sk_free(sk);
681 if (sk->sk_prot_creator->slab != NULL)
682 kmem_cache_free(sk->sk_prot_creator->slab, sk);
688 void __init sk_init(void)
690 if (num_physpages <= 4096) {
691 sysctl_wmem_max = 32767;
692 sysctl_rmem_max = 32767;
693 sysctl_wmem_default = 32767;
694 sysctl_rmem_default = 32767;
695 } else if (num_physpages >= 131072) {
696 sysctl_wmem_max = 131071;
697 sysctl_rmem_max = 131071;
702 * Simple resource managers for sockets.
707 * Write buffer destructor automatically called from kfree_skb.
709 void sock_wfree(struct sk_buff *skb)
711 struct sock *sk = skb->sk;
713 /* In case it might be waiting for more memory. */
714 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
715 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
716 sk->sk_write_space(sk);
721 * Read buffer destructor automatically called from kfree_skb.
723 void sock_rfree(struct sk_buff *skb)
725 struct sock *sk = skb->sk;
727 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
731 int sock_i_uid(struct sock *sk)
735 read_lock(&sk->sk_callback_lock);
736 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
737 read_unlock(&sk->sk_callback_lock);
741 unsigned long sock_i_ino(struct sock *sk)
745 read_lock(&sk->sk_callback_lock);
746 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
747 read_unlock(&sk->sk_callback_lock);
752 * Allocate a skb from the socket's send buffer.
754 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
755 unsigned int __nocast priority)
757 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
758 struct sk_buff * skb = alloc_skb(size, priority);
760 skb_set_owner_w(skb, sk);
768 * Allocate a skb from the socket's receive buffer.
770 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
771 unsigned int __nocast priority)
773 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
774 struct sk_buff *skb = alloc_skb(size, priority);
776 skb_set_owner_r(skb, sk);
784 * Allocate a memory block from the socket's option memory buffer.
786 void *sock_kmalloc(struct sock *sk, int size, unsigned int __nocast priority)
788 if ((unsigned)size <= sysctl_optmem_max &&
789 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
791 /* First do the add, to avoid the race if kmalloc
794 atomic_add(size, &sk->sk_omem_alloc);
795 mem = kmalloc(size, priority);
798 atomic_sub(size, &sk->sk_omem_alloc);
804 * Free an option memory block.
806 void sock_kfree_s(struct sock *sk, void *mem, int size)
809 atomic_sub(size, &sk->sk_omem_alloc);
812 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
813 I think, these locks should be removed for datagram sockets.
815 static long sock_wait_for_wmem(struct sock * sk, long timeo)
819 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
823 if (signal_pending(current))
825 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
826 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
827 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
829 if (sk->sk_shutdown & SEND_SHUTDOWN)
833 timeo = schedule_timeout(timeo);
835 finish_wait(sk->sk_sleep, &wait);
841 * Generic send/receive buffer handlers
844 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
845 unsigned long header_len,
846 unsigned long data_len,
847 int noblock, int *errcode)
850 unsigned int gfp_mask;
854 gfp_mask = sk->sk_allocation;
855 if (gfp_mask & __GFP_WAIT)
856 gfp_mask |= __GFP_REPEAT;
858 timeo = sock_sndtimeo(sk, noblock);
860 err = sock_error(sk);
865 if (sk->sk_shutdown & SEND_SHUTDOWN)
868 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
869 skb = alloc_skb(header_len, sk->sk_allocation);
874 /* No pages, we're done... */
878 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
879 skb->truesize += data_len;
880 skb_shinfo(skb)->nr_frags = npages;
881 for (i = 0; i < npages; i++) {
885 page = alloc_pages(sk->sk_allocation, 0);
888 skb_shinfo(skb)->nr_frags = i;
893 frag = &skb_shinfo(skb)->frags[i];
895 frag->page_offset = 0;
896 frag->size = (data_len >= PAGE_SIZE ?
899 data_len -= PAGE_SIZE;
902 /* Full success... */
908 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
909 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
913 if (signal_pending(current))
915 timeo = sock_wait_for_wmem(sk, timeo);
918 skb_set_owner_w(skb, sk);
922 err = sock_intr_errno(timeo);
928 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
929 int noblock, int *errcode)
931 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
934 static void __lock_sock(struct sock *sk)
939 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
940 TASK_UNINTERRUPTIBLE);
941 spin_unlock_bh(&sk->sk_lock.slock);
943 spin_lock_bh(&sk->sk_lock.slock);
944 if(!sock_owned_by_user(sk))
947 finish_wait(&sk->sk_lock.wq, &wait);
950 static void __release_sock(struct sock *sk)
952 struct sk_buff *skb = sk->sk_backlog.head;
955 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
959 struct sk_buff *next = skb->next;
962 sk->sk_backlog_rcv(sk, skb);
965 * We are in process context here with softirqs
966 * disabled, use cond_resched_softirq() to preempt.
967 * This is safe to do because we've taken the backlog
970 cond_resched_softirq();
973 } while (skb != NULL);
976 } while((skb = sk->sk_backlog.head) != NULL);
980 * sk_wait_data - wait for data to arrive at sk_receive_queue
981 * @sk: sock to wait on
982 * @timeo: for how long
984 * Now socket state including sk->sk_err is changed only under lock,
985 * hence we may omit checks after joining wait queue.
986 * We check receive queue before schedule() only as optimization;
987 * it is very likely that release_sock() added new data.
989 int sk_wait_data(struct sock *sk, long *timeo)
994 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
995 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
996 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
997 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
998 finish_wait(sk->sk_sleep, &wait);
1002 EXPORT_SYMBOL(sk_wait_data);
1005 * Set of default routines for initialising struct proto_ops when
1006 * the protocol does not support a particular function. In certain
1007 * cases where it makes no sense for a protocol to have a "do nothing"
1008 * function, some default processing is provided.
1011 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1016 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1022 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1027 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1032 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1038 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1043 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1048 int sock_no_listen(struct socket *sock, int backlog)
1053 int sock_no_shutdown(struct socket *sock, int how)
1058 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1059 char __user *optval, int optlen)
1064 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1065 char __user *optval, int __user *optlen)
1070 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1076 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1077 size_t len, int flags)
1082 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1084 /* Mirror missing mmap method error code */
1088 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1091 struct msghdr msg = {.msg_flags = flags};
1093 char *kaddr = kmap(page);
1094 iov.iov_base = kaddr + offset;
1096 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1102 * Default Socket Callbacks
1105 static void sock_def_wakeup(struct sock *sk)
1107 read_lock(&sk->sk_callback_lock);
1108 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1109 wake_up_interruptible_all(sk->sk_sleep);
1110 read_unlock(&sk->sk_callback_lock);
1113 static void sock_def_error_report(struct sock *sk)
1115 read_lock(&sk->sk_callback_lock);
1116 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1117 wake_up_interruptible(sk->sk_sleep);
1118 sk_wake_async(sk,0,POLL_ERR);
1119 read_unlock(&sk->sk_callback_lock);
1122 static void sock_def_readable(struct sock *sk, int len)
1124 read_lock(&sk->sk_callback_lock);
1125 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1126 wake_up_interruptible(sk->sk_sleep);
1127 sk_wake_async(sk,1,POLL_IN);
1128 read_unlock(&sk->sk_callback_lock);
1131 static void sock_def_write_space(struct sock *sk)
1133 read_lock(&sk->sk_callback_lock);
1135 /* Do not wake up a writer until he can make "significant"
1138 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1139 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1140 wake_up_interruptible(sk->sk_sleep);
1142 /* Should agree with poll, otherwise some programs break */
1143 if (sock_writeable(sk))
1144 sk_wake_async(sk, 2, POLL_OUT);
1147 read_unlock(&sk->sk_callback_lock);
1150 static void sock_def_destruct(struct sock *sk)
1152 if (sk->sk_protinfo)
1153 kfree(sk->sk_protinfo);
1156 void sk_send_sigurg(struct sock *sk)
1158 if (sk->sk_socket && sk->sk_socket->file)
1159 if (send_sigurg(&sk->sk_socket->file->f_owner))
1160 sk_wake_async(sk, 3, POLL_PRI);
1163 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1164 unsigned long expires)
1166 if (!mod_timer(timer, expires))
1170 EXPORT_SYMBOL(sk_reset_timer);
1172 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1174 if (timer_pending(timer) && del_timer(timer))
1178 EXPORT_SYMBOL(sk_stop_timer);
1180 void sock_init_data(struct socket *sock, struct sock *sk)
1182 skb_queue_head_init(&sk->sk_receive_queue);
1183 skb_queue_head_init(&sk->sk_write_queue);
1184 skb_queue_head_init(&sk->sk_error_queue);
1186 sk->sk_send_head = NULL;
1188 init_timer(&sk->sk_timer);
1190 sk->sk_allocation = GFP_KERNEL;
1191 sk->sk_rcvbuf = sysctl_rmem_default;
1192 sk->sk_sndbuf = sysctl_wmem_default;
1193 sk->sk_state = TCP_CLOSE;
1194 sk->sk_socket = sock;
1196 sock_set_flag(sk, SOCK_ZAPPED);
1200 sk->sk_type = sock->type;
1201 sk->sk_sleep = &sock->wait;
1204 sk->sk_sleep = NULL;
1206 rwlock_init(&sk->sk_dst_lock);
1207 rwlock_init(&sk->sk_callback_lock);
1209 sk->sk_state_change = sock_def_wakeup;
1210 sk->sk_data_ready = sock_def_readable;
1211 sk->sk_write_space = sock_def_write_space;
1212 sk->sk_error_report = sock_def_error_report;
1213 sk->sk_destruct = sock_def_destruct;
1215 sk->sk_sndmsg_page = NULL;
1216 sk->sk_sndmsg_off = 0;
1218 sk->sk_peercred.pid = 0;
1219 sk->sk_peercred.uid = -1;
1220 sk->sk_peercred.gid = -1;
1221 sk->sk_write_pending = 0;
1222 sk->sk_rcvlowat = 1;
1223 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1224 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1226 sk->sk_stamp.tv_sec = -1L;
1227 sk->sk_stamp.tv_usec = -1L;
1229 atomic_set(&sk->sk_refcnt, 1);
1232 void fastcall lock_sock(struct sock *sk)
1235 spin_lock_bh(&(sk->sk_lock.slock));
1236 if (sk->sk_lock.owner)
1238 sk->sk_lock.owner = (void *)1;
1239 spin_unlock_bh(&(sk->sk_lock.slock));
1242 EXPORT_SYMBOL(lock_sock);
1244 void fastcall release_sock(struct sock *sk)
1246 spin_lock_bh(&(sk->sk_lock.slock));
1247 if (sk->sk_backlog.tail)
1249 sk->sk_lock.owner = NULL;
1250 if (waitqueue_active(&(sk->sk_lock.wq)))
1251 wake_up(&(sk->sk_lock.wq));
1252 spin_unlock_bh(&(sk->sk_lock.slock));
1254 EXPORT_SYMBOL(release_sock);
1256 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1258 if (!sock_flag(sk, SOCK_TIMESTAMP))
1259 sock_enable_timestamp(sk);
1260 if (sk->sk_stamp.tv_sec == -1)
1262 if (sk->sk_stamp.tv_sec == 0)
1263 do_gettimeofday(&sk->sk_stamp);
1264 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1267 EXPORT_SYMBOL(sock_get_timestamp);
1269 void sock_enable_timestamp(struct sock *sk)
1271 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1272 sock_set_flag(sk, SOCK_TIMESTAMP);
1273 net_enable_timestamp();
1276 EXPORT_SYMBOL(sock_enable_timestamp);
1279 * Get a socket option on an socket.
1281 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1282 * asynchronous errors should be reported by getsockopt. We assume
1283 * this means if you specify SO_ERROR (otherwise whats the point of it).
1285 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1286 char __user *optval, int __user *optlen)
1288 struct sock *sk = sock->sk;
1290 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1293 EXPORT_SYMBOL(sock_common_getsockopt);
1295 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1296 struct msghdr *msg, size_t size, int flags)
1298 struct sock *sk = sock->sk;
1302 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1303 flags & ~MSG_DONTWAIT, &addr_len);
1305 msg->msg_namelen = addr_len;
1309 EXPORT_SYMBOL(sock_common_recvmsg);
1312 * Set socket options on an inet socket.
1314 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1315 char __user *optval, int optlen)
1317 struct sock *sk = sock->sk;
1319 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1322 EXPORT_SYMBOL(sock_common_setsockopt);
1324 void sk_common_release(struct sock *sk)
1326 if (sk->sk_prot->destroy)
1327 sk->sk_prot->destroy(sk);
1330 * Observation: when sock_common_release is called, processes have
1331 * no access to socket. But net still has.
1332 * Step one, detach it from networking:
1334 * A. Remove from hash tables.
1337 sk->sk_prot->unhash(sk);
1340 * In this point socket cannot receive new packets, but it is possible
1341 * that some packets are in flight because some CPU runs receiver and
1342 * did hash table lookup before we unhashed socket. They will achieve
1343 * receive queue and will be purged by socket destructor.
1345 * Also we still have packets pending on receive queue and probably,
1346 * our own packets waiting in device queues. sock_destroy will drain
1347 * receive queue, but transmitted packets will delay socket destruction
1348 * until the last reference will be released.
1353 xfrm_sk_free_policy(sk);
1355 #ifdef INET_REFCNT_DEBUG
1356 if (atomic_read(&sk->sk_refcnt) != 1)
1357 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1358 sk, atomic_read(&sk->sk_refcnt));
1363 EXPORT_SYMBOL(sk_common_release);
1365 static DEFINE_RWLOCK(proto_list_lock);
1366 static LIST_HEAD(proto_list);
1368 int proto_register(struct proto *prot, int alloc_slab)
1370 char *request_sock_slab_name;
1374 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1375 SLAB_HWCACHE_ALIGN, NULL, NULL);
1377 if (prot->slab == NULL) {
1378 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1383 if (prot->rsk_prot != NULL) {
1384 static const char mask[] = "request_sock_%s";
1386 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1387 if (request_sock_slab_name == NULL)
1388 goto out_free_sock_slab;
1390 sprintf(request_sock_slab_name, mask, prot->name);
1391 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1392 prot->rsk_prot->obj_size, 0,
1393 SLAB_HWCACHE_ALIGN, NULL, NULL);
1395 if (prot->rsk_prot->slab == NULL) {
1396 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1398 goto out_free_request_sock_slab_name;
1403 write_lock(&proto_list_lock);
1404 list_add(&prot->node, &proto_list);
1405 write_unlock(&proto_list_lock);
1409 out_free_request_sock_slab_name:
1410 kfree(request_sock_slab_name);
1412 kmem_cache_destroy(prot->slab);
1417 EXPORT_SYMBOL(proto_register);
1419 void proto_unregister(struct proto *prot)
1421 write_lock(&proto_list_lock);
1423 if (prot->slab != NULL) {
1424 kmem_cache_destroy(prot->slab);
1428 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1429 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1431 kmem_cache_destroy(prot->rsk_prot->slab);
1433 prot->rsk_prot->slab = NULL;
1436 list_del(&prot->node);
1437 write_unlock(&proto_list_lock);
1440 EXPORT_SYMBOL(proto_unregister);
1442 #ifdef CONFIG_PROC_FS
1443 static inline struct proto *__proto_head(void)
1445 return list_entry(proto_list.next, struct proto, node);
1448 static inline struct proto *proto_head(void)
1450 return list_empty(&proto_list) ? NULL : __proto_head();
1453 static inline struct proto *proto_next(struct proto *proto)
1455 return proto->node.next == &proto_list ? NULL :
1456 list_entry(proto->node.next, struct proto, node);
1459 static inline struct proto *proto_get_idx(loff_t pos)
1461 struct proto *proto;
1464 list_for_each_entry(proto, &proto_list, node)
1473 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1475 read_lock(&proto_list_lock);
1476 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1479 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1482 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1485 static void proto_seq_stop(struct seq_file *seq, void *v)
1487 read_unlock(&proto_list_lock);
1490 static char proto_method_implemented(const void *method)
1492 return method == NULL ? 'n' : 'y';
1495 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1497 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1498 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1501 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1502 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1503 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1505 proto->slab == NULL ? "no" : "yes",
1506 module_name(proto->owner),
1507 proto_method_implemented(proto->close),
1508 proto_method_implemented(proto->connect),
1509 proto_method_implemented(proto->disconnect),
1510 proto_method_implemented(proto->accept),
1511 proto_method_implemented(proto->ioctl),
1512 proto_method_implemented(proto->init),
1513 proto_method_implemented(proto->destroy),
1514 proto_method_implemented(proto->shutdown),
1515 proto_method_implemented(proto->setsockopt),
1516 proto_method_implemented(proto->getsockopt),
1517 proto_method_implemented(proto->sendmsg),
1518 proto_method_implemented(proto->recvmsg),
1519 proto_method_implemented(proto->sendpage),
1520 proto_method_implemented(proto->bind),
1521 proto_method_implemented(proto->backlog_rcv),
1522 proto_method_implemented(proto->hash),
1523 proto_method_implemented(proto->unhash),
1524 proto_method_implemented(proto->get_port),
1525 proto_method_implemented(proto->enter_memory_pressure));
1528 static int proto_seq_show(struct seq_file *seq, void *v)
1530 if (v == SEQ_START_TOKEN)
1531 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1540 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1542 proto_seq_printf(seq, v);
1546 static struct seq_operations proto_seq_ops = {
1547 .start = proto_seq_start,
1548 .next = proto_seq_next,
1549 .stop = proto_seq_stop,
1550 .show = proto_seq_show,
1553 static int proto_seq_open(struct inode *inode, struct file *file)
1555 return seq_open(file, &proto_seq_ops);
1558 static struct file_operations proto_seq_fops = {
1559 .owner = THIS_MODULE,
1560 .open = proto_seq_open,
1562 .llseek = seq_lseek,
1563 .release = seq_release,
1566 static int __init proto_init(void)
1568 /* register /proc/net/protocols */
1569 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1572 subsys_initcall(proto_init);
1574 #endif /* PROC_FS */
1576 EXPORT_SYMBOL(sk_alloc);
1577 EXPORT_SYMBOL(sk_free);
1578 EXPORT_SYMBOL(sk_send_sigurg);
1579 EXPORT_SYMBOL(sock_alloc_send_skb);
1580 EXPORT_SYMBOL(sock_init_data);
1581 EXPORT_SYMBOL(sock_kfree_s);
1582 EXPORT_SYMBOL(sock_kmalloc);
1583 EXPORT_SYMBOL(sock_no_accept);
1584 EXPORT_SYMBOL(sock_no_bind);
1585 EXPORT_SYMBOL(sock_no_connect);
1586 EXPORT_SYMBOL(sock_no_getname);
1587 EXPORT_SYMBOL(sock_no_getsockopt);
1588 EXPORT_SYMBOL(sock_no_ioctl);
1589 EXPORT_SYMBOL(sock_no_listen);
1590 EXPORT_SYMBOL(sock_no_mmap);
1591 EXPORT_SYMBOL(sock_no_poll);
1592 EXPORT_SYMBOL(sock_no_recvmsg);
1593 EXPORT_SYMBOL(sock_no_sendmsg);
1594 EXPORT_SYMBOL(sock_no_sendpage);
1595 EXPORT_SYMBOL(sock_no_setsockopt);
1596 EXPORT_SYMBOL(sock_no_shutdown);
1597 EXPORT_SYMBOL(sock_no_socketpair);
1598 EXPORT_SYMBOL(sock_rfree);
1599 EXPORT_SYMBOL(sock_setsockopt);
1600 EXPORT_SYMBOL(sock_wfree);
1601 EXPORT_SYMBOL(sock_wmalloc);
1602 EXPORT_SYMBOL(sock_i_uid);
1603 EXPORT_SYMBOL(sock_i_ino);
1604 #ifdef CONFIG_SYSCTL
1605 EXPORT_SYMBOL(sysctl_optmem_max);
1606 EXPORT_SYMBOL(sysctl_rmem_max);
1607 EXPORT_SYMBOL(sysctl_wmem_max);