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/capability.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>
133 * Each address family might have different locking rules, so we have
134 * one slock key per address family:
136 struct lock_class_key af_family_keys[AF_MAX];
139 * sk_callback_lock locking rules are per-address-family,
140 * so split the lock classes by using a per-AF key:
142 static struct lock_class_key af_callback_keys[AF_MAX];
144 /* Take into consideration the size of the struct sk_buff overhead in the
145 * determination of these values, since that is non-constant across
146 * platforms. This makes socket queueing behavior and performance
147 * not depend upon such differences.
149 #define _SK_MEM_PACKETS 256
150 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
151 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
152 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
154 /* Run time adjustable parameters. */
155 __u32 sysctl_wmem_max = SK_WMEM_MAX;
156 __u32 sysctl_rmem_max = SK_RMEM_MAX;
157 __u32 sysctl_wmem_default = SK_WMEM_MAX;
158 __u32 sysctl_rmem_default = SK_RMEM_MAX;
160 /* Maximal space eaten by iovec or ancilliary data plus some space */
161 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
163 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
167 if (optlen < sizeof(tv))
169 if (copy_from_user(&tv, optval, sizeof(tv)))
172 *timeo_p = MAX_SCHEDULE_TIMEOUT;
173 if (tv.tv_sec == 0 && tv.tv_usec == 0)
175 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
176 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
180 static void sock_warn_obsolete_bsdism(const char *name)
183 static char warncomm[TASK_COMM_LEN];
184 if (strcmp(warncomm, current->comm) && warned < 5) {
185 strcpy(warncomm, current->comm);
186 printk(KERN_WARNING "process `%s' is using obsolete "
187 "%s SO_BSDCOMPAT\n", warncomm, name);
192 static void sock_disable_timestamp(struct sock *sk)
194 if (sock_flag(sk, SOCK_TIMESTAMP)) {
195 sock_reset_flag(sk, SOCK_TIMESTAMP);
196 net_disable_timestamp();
201 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
206 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
207 number of warnings when compiling with -W --ANK
209 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
210 (unsigned)sk->sk_rcvbuf) {
215 /* It would be deadlock, if sock_queue_rcv_skb is used
216 with socket lock! We assume that users of this
217 function are lock free.
219 err = sk_filter(sk, skb, 1);
224 skb_set_owner_r(skb, sk);
226 /* Cache the SKB length before we tack it onto the receive
227 * queue. Once it is added it no longer belongs to us and
228 * may be freed by other threads of control pulling packets
233 skb_queue_tail(&sk->sk_receive_queue, skb);
235 if (!sock_flag(sk, SOCK_DEAD))
236 sk->sk_data_ready(sk, skb_len);
240 EXPORT_SYMBOL(sock_queue_rcv_skb);
242 int sk_receive_skb(struct sock *sk, struct sk_buff *skb)
244 int rc = NET_RX_SUCCESS;
246 if (sk_filter(sk, skb, 0))
247 goto discard_and_relse;
252 if (!sock_owned_by_user(sk))
253 rc = sk->sk_backlog_rcv(sk, skb);
255 sk_add_backlog(sk, skb);
264 EXPORT_SYMBOL(sk_receive_skb);
266 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
268 struct dst_entry *dst = sk->sk_dst_cache;
270 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
271 sk->sk_dst_cache = NULL;
278 EXPORT_SYMBOL(__sk_dst_check);
280 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
282 struct dst_entry *dst = sk_dst_get(sk);
284 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
292 EXPORT_SYMBOL(sk_dst_check);
295 * This is meant for all protocols to use and covers goings on
296 * at the socket level. Everything here is generic.
299 int sock_setsockopt(struct socket *sock, int level, int optname,
300 char __user *optval, int optlen)
302 struct sock *sk=sock->sk;
303 struct sk_filter *filter;
310 * Options without arguments
313 #ifdef SO_DONTLINGER /* Compatibility item... */
314 if (optname == SO_DONTLINGER) {
316 sock_reset_flag(sk, SOCK_LINGER);
322 if(optlen<sizeof(int))
325 if (get_user(val, (int __user *)optval))
335 if(val && !capable(CAP_NET_ADMIN))
340 sock_set_flag(sk, SOCK_DBG);
342 sock_reset_flag(sk, SOCK_DBG);
345 sk->sk_reuse = valbool;
353 sock_set_flag(sk, SOCK_LOCALROUTE);
355 sock_reset_flag(sk, SOCK_LOCALROUTE);
358 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
361 /* Don't error on this BSD doesn't and if you think
362 about it this is right. Otherwise apps have to
363 play 'guess the biggest size' games. RCVBUF/SNDBUF
364 are treated in BSD as hints */
366 if (val > sysctl_wmem_max)
367 val = sysctl_wmem_max;
369 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
370 if ((val * 2) < SOCK_MIN_SNDBUF)
371 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
373 sk->sk_sndbuf = val * 2;
376 * Wake up sending tasks if we
379 sk->sk_write_space(sk);
383 if (!capable(CAP_NET_ADMIN)) {
390 /* Don't error on this BSD doesn't and if you think
391 about it this is right. Otherwise apps have to
392 play 'guess the biggest size' games. RCVBUF/SNDBUF
393 are treated in BSD as hints */
395 if (val > sysctl_rmem_max)
396 val = sysctl_rmem_max;
398 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
400 * We double it on the way in to account for
401 * "struct sk_buff" etc. overhead. Applications
402 * assume that the SO_RCVBUF setting they make will
403 * allow that much actual data to be received on that
406 * Applications are unaware that "struct sk_buff" and
407 * other overheads allocate from the receive buffer
408 * during socket buffer allocation.
410 * And after considering the possible alternatives,
411 * returning the value we actually used in getsockopt
412 * is the most desirable behavior.
414 if ((val * 2) < SOCK_MIN_RCVBUF)
415 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
417 sk->sk_rcvbuf = val * 2;
421 if (!capable(CAP_NET_ADMIN)) {
429 if (sk->sk_protocol == IPPROTO_TCP)
430 tcp_set_keepalive(sk, valbool);
432 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
436 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
440 sk->sk_no_check = valbool;
444 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
445 sk->sk_priority = val;
451 if(optlen<sizeof(ling)) {
452 ret = -EINVAL; /* 1003.1g */
455 if (copy_from_user(&ling,optval,sizeof(ling))) {
460 sock_reset_flag(sk, SOCK_LINGER);
462 #if (BITS_PER_LONG == 32)
463 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
464 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
467 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
468 sock_set_flag(sk, SOCK_LINGER);
473 sock_warn_obsolete_bsdism("setsockopt");
478 set_bit(SOCK_PASSCRED, &sock->flags);
480 clear_bit(SOCK_PASSCRED, &sock->flags);
485 sock_set_flag(sk, SOCK_RCVTSTAMP);
486 sock_enable_timestamp(sk);
488 sock_reset_flag(sk, SOCK_RCVTSTAMP);
494 sk->sk_rcvlowat = val ? : 1;
498 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
502 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
505 #ifdef CONFIG_NETDEVICES
506 case SO_BINDTODEVICE:
508 char devname[IFNAMSIZ];
511 if (!capable(CAP_NET_RAW)) {
516 /* Bind this socket to a particular device like "eth0",
517 * as specified in the passed interface name. If the
518 * name is "" or the option length is zero the socket
523 sk->sk_bound_dev_if = 0;
525 if (optlen > IFNAMSIZ - 1)
526 optlen = IFNAMSIZ - 1;
527 memset(devname, 0, sizeof(devname));
528 if (copy_from_user(devname, optval, optlen)) {
533 /* Remove any cached route for this socket. */
536 if (devname[0] == '\0') {
537 sk->sk_bound_dev_if = 0;
539 struct net_device *dev = dev_get_by_name(devname);
544 sk->sk_bound_dev_if = dev->ifindex;
553 case SO_ATTACH_FILTER:
555 if (optlen == sizeof(struct sock_fprog)) {
556 struct sock_fprog fprog;
559 if (copy_from_user(&fprog, optval, sizeof(fprog)))
562 ret = sk_attach_filter(&fprog, sk);
566 case SO_DETACH_FILTER:
567 spin_lock_bh(&sk->sk_lock.slock);
568 filter = sk->sk_filter;
570 sk->sk_filter = NULL;
571 spin_unlock_bh(&sk->sk_lock.slock);
572 sk_filter_release(sk, filter);
575 spin_unlock_bh(&sk->sk_lock.slock);
581 set_bit(SOCK_PASSSEC, &sock->flags);
583 clear_bit(SOCK_PASSSEC, &sock->flags);
586 /* We implement the SO_SNDLOWAT etc to
587 not be settable (1003.1g 5.3) */
597 int sock_getsockopt(struct socket *sock, int level, int optname,
598 char __user *optval, int __user *optlen)
600 struct sock *sk = sock->sk;
609 unsigned int lv = sizeof(int);
612 if(get_user(len,optlen))
620 v.val = sock_flag(sk, SOCK_DBG);
624 v.val = sock_flag(sk, SOCK_LOCALROUTE);
628 v.val = !!sock_flag(sk, SOCK_BROADCAST);
632 v.val = sk->sk_sndbuf;
636 v.val = sk->sk_rcvbuf;
640 v.val = sk->sk_reuse;
644 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
652 v.val = -sock_error(sk);
654 v.val = xchg(&sk->sk_err_soft, 0);
658 v.val = !!sock_flag(sk, SOCK_URGINLINE);
662 v.val = sk->sk_no_check;
666 v.val = sk->sk_priority;
671 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
672 v.ling.l_linger = sk->sk_lingertime / HZ;
676 sock_warn_obsolete_bsdism("getsockopt");
680 v.val = sock_flag(sk, SOCK_RCVTSTAMP);
684 lv=sizeof(struct timeval);
685 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
689 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
690 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
695 lv=sizeof(struct timeval);
696 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
700 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
701 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
706 v.val = sk->sk_rcvlowat;
714 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
718 if (len > sizeof(sk->sk_peercred))
719 len = sizeof(sk->sk_peercred);
720 if (copy_to_user(optval, &sk->sk_peercred, len))
728 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
732 if (copy_to_user(optval, address, len))
737 /* Dubious BSD thing... Probably nobody even uses it, but
738 * the UNIX standard wants it for whatever reason... -DaveM
741 v.val = sk->sk_state == TCP_LISTEN;
745 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
749 return security_socket_getpeersec_stream(sock, optval, optlen, len);
752 return(-ENOPROTOOPT);
756 if (copy_to_user(optval, &v, len))
759 if (put_user(len, optlen))
765 * sk_alloc - All socket objects are allocated here
766 * @family: protocol family
767 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
768 * @prot: struct proto associated with this new sock instance
769 * @zero_it: if we should zero the newly allocated sock
771 struct sock *sk_alloc(int family, gfp_t priority,
772 struct proto *prot, int zero_it)
774 struct sock *sk = NULL;
775 kmem_cache_t *slab = prot->slab;
778 sk = kmem_cache_alloc(slab, priority);
780 sk = kmalloc(prot->obj_size, priority);
784 memset(sk, 0, prot->obj_size);
785 sk->sk_family = family;
787 * See comment in struct sock definition to understand
788 * why we need sk_prot_creator -acme
790 sk->sk_prot = sk->sk_prot_creator = prot;
794 if (security_sk_alloc(sk, family, priority))
797 if (!try_module_get(prot->owner))
804 kmem_cache_free(slab, sk);
810 void sk_free(struct sock *sk)
812 struct sk_filter *filter;
813 struct module *owner = sk->sk_prot_creator->owner;
818 filter = sk->sk_filter;
820 sk_filter_release(sk, filter);
821 sk->sk_filter = NULL;
824 sock_disable_timestamp(sk);
826 if (atomic_read(&sk->sk_omem_alloc))
827 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
828 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
830 security_sk_free(sk);
831 if (sk->sk_prot_creator->slab != NULL)
832 kmem_cache_free(sk->sk_prot_creator->slab, sk);
838 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
840 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
843 struct sk_filter *filter;
845 memcpy(newsk, sk, sk->sk_prot->obj_size);
848 sk_node_init(&newsk->sk_node);
849 sock_lock_init(newsk);
852 atomic_set(&newsk->sk_rmem_alloc, 0);
853 atomic_set(&newsk->sk_wmem_alloc, 0);
854 atomic_set(&newsk->sk_omem_alloc, 0);
855 skb_queue_head_init(&newsk->sk_receive_queue);
856 skb_queue_head_init(&newsk->sk_write_queue);
857 #ifdef CONFIG_NET_DMA
858 skb_queue_head_init(&newsk->sk_async_wait_queue);
861 rwlock_init(&newsk->sk_dst_lock);
862 rwlock_init(&newsk->sk_callback_lock);
863 lockdep_set_class(&newsk->sk_callback_lock,
864 af_callback_keys + newsk->sk_family);
866 newsk->sk_dst_cache = NULL;
867 newsk->sk_wmem_queued = 0;
868 newsk->sk_forward_alloc = 0;
869 newsk->sk_send_head = NULL;
870 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
871 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
873 sock_reset_flag(newsk, SOCK_DONE);
874 skb_queue_head_init(&newsk->sk_error_queue);
876 filter = newsk->sk_filter;
878 sk_filter_charge(newsk, filter);
880 if (unlikely(xfrm_sk_clone_policy(newsk))) {
881 /* It is still raw copy of parent, so invalidate
882 * destructor and make plain sk_free() */
883 newsk->sk_destruct = NULL;
890 newsk->sk_priority = 0;
891 atomic_set(&newsk->sk_refcnt, 2);
894 * Increment the counter in the same struct proto as the master
895 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
896 * is the same as sk->sk_prot->socks, as this field was copied
899 * This _changes_ the previous behaviour, where
900 * tcp_create_openreq_child always was incrementing the
901 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
902 * to be taken into account in all callers. -acme
904 sk_refcnt_debug_inc(newsk);
905 newsk->sk_socket = NULL;
906 newsk->sk_sleep = NULL;
908 if (newsk->sk_prot->sockets_allocated)
909 atomic_inc(newsk->sk_prot->sockets_allocated);
915 EXPORT_SYMBOL_GPL(sk_clone);
917 void __init sk_init(void)
919 if (num_physpages <= 4096) {
920 sysctl_wmem_max = 32767;
921 sysctl_rmem_max = 32767;
922 sysctl_wmem_default = 32767;
923 sysctl_rmem_default = 32767;
924 } else if (num_physpages >= 131072) {
925 sysctl_wmem_max = 131071;
926 sysctl_rmem_max = 131071;
931 * Simple resource managers for sockets.
936 * Write buffer destructor automatically called from kfree_skb.
938 void sock_wfree(struct sk_buff *skb)
940 struct sock *sk = skb->sk;
942 /* In case it might be waiting for more memory. */
943 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
944 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
945 sk->sk_write_space(sk);
950 * Read buffer destructor automatically called from kfree_skb.
952 void sock_rfree(struct sk_buff *skb)
954 struct sock *sk = skb->sk;
956 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
960 int sock_i_uid(struct sock *sk)
964 read_lock(&sk->sk_callback_lock);
965 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
966 read_unlock(&sk->sk_callback_lock);
970 unsigned long sock_i_ino(struct sock *sk)
974 read_lock(&sk->sk_callback_lock);
975 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
976 read_unlock(&sk->sk_callback_lock);
981 * Allocate a skb from the socket's send buffer.
983 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
986 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
987 struct sk_buff * skb = alloc_skb(size, priority);
989 skb_set_owner_w(skb, sk);
997 * Allocate a skb from the socket's receive buffer.
999 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1002 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1003 struct sk_buff *skb = alloc_skb(size, priority);
1005 skb_set_owner_r(skb, sk);
1013 * Allocate a memory block from the socket's option memory buffer.
1015 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1017 if ((unsigned)size <= sysctl_optmem_max &&
1018 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1020 /* First do the add, to avoid the race if kmalloc
1023 atomic_add(size, &sk->sk_omem_alloc);
1024 mem = kmalloc(size, priority);
1027 atomic_sub(size, &sk->sk_omem_alloc);
1033 * Free an option memory block.
1035 void sock_kfree_s(struct sock *sk, void *mem, int size)
1038 atomic_sub(size, &sk->sk_omem_alloc);
1041 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1042 I think, these locks should be removed for datagram sockets.
1044 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1048 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1052 if (signal_pending(current))
1054 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1055 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1056 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1058 if (sk->sk_shutdown & SEND_SHUTDOWN)
1062 timeo = schedule_timeout(timeo);
1064 finish_wait(sk->sk_sleep, &wait);
1070 * Generic send/receive buffer handlers
1073 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1074 unsigned long header_len,
1075 unsigned long data_len,
1076 int noblock, int *errcode)
1078 struct sk_buff *skb;
1083 gfp_mask = sk->sk_allocation;
1084 if (gfp_mask & __GFP_WAIT)
1085 gfp_mask |= __GFP_REPEAT;
1087 timeo = sock_sndtimeo(sk, noblock);
1089 err = sock_error(sk);
1094 if (sk->sk_shutdown & SEND_SHUTDOWN)
1097 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1098 skb = alloc_skb(header_len, sk->sk_allocation);
1103 /* No pages, we're done... */
1107 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1108 skb->truesize += data_len;
1109 skb_shinfo(skb)->nr_frags = npages;
1110 for (i = 0; i < npages; i++) {
1114 page = alloc_pages(sk->sk_allocation, 0);
1117 skb_shinfo(skb)->nr_frags = i;
1122 frag = &skb_shinfo(skb)->frags[i];
1124 frag->page_offset = 0;
1125 frag->size = (data_len >= PAGE_SIZE ?
1128 data_len -= PAGE_SIZE;
1131 /* Full success... */
1137 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1138 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1142 if (signal_pending(current))
1144 timeo = sock_wait_for_wmem(sk, timeo);
1147 skb_set_owner_w(skb, sk);
1151 err = sock_intr_errno(timeo);
1157 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1158 int noblock, int *errcode)
1160 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1163 static void __lock_sock(struct sock *sk)
1168 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1169 TASK_UNINTERRUPTIBLE);
1170 spin_unlock_bh(&sk->sk_lock.slock);
1172 spin_lock_bh(&sk->sk_lock.slock);
1173 if(!sock_owned_by_user(sk))
1176 finish_wait(&sk->sk_lock.wq, &wait);
1179 static void __release_sock(struct sock *sk)
1181 struct sk_buff *skb = sk->sk_backlog.head;
1184 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1188 struct sk_buff *next = skb->next;
1191 sk->sk_backlog_rcv(sk, skb);
1194 * We are in process context here with softirqs
1195 * disabled, use cond_resched_softirq() to preempt.
1196 * This is safe to do because we've taken the backlog
1199 cond_resched_softirq();
1202 } while (skb != NULL);
1205 } while((skb = sk->sk_backlog.head) != NULL);
1209 * sk_wait_data - wait for data to arrive at sk_receive_queue
1210 * @sk: sock to wait on
1211 * @timeo: for how long
1213 * Now socket state including sk->sk_err is changed only under lock,
1214 * hence we may omit checks after joining wait queue.
1215 * We check receive queue before schedule() only as optimization;
1216 * it is very likely that release_sock() added new data.
1218 int sk_wait_data(struct sock *sk, long *timeo)
1223 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1224 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1225 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1226 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1227 finish_wait(sk->sk_sleep, &wait);
1231 EXPORT_SYMBOL(sk_wait_data);
1234 * Set of default routines for initialising struct proto_ops when
1235 * the protocol does not support a particular function. In certain
1236 * cases where it makes no sense for a protocol to have a "do nothing"
1237 * function, some default processing is provided.
1240 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1245 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1251 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1256 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1261 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1267 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1272 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1277 int sock_no_listen(struct socket *sock, int backlog)
1282 int sock_no_shutdown(struct socket *sock, int how)
1287 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1288 char __user *optval, int optlen)
1293 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1294 char __user *optval, int __user *optlen)
1299 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1305 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1306 size_t len, int flags)
1311 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1313 /* Mirror missing mmap method error code */
1317 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1320 struct msghdr msg = {.msg_flags = flags};
1322 char *kaddr = kmap(page);
1323 iov.iov_base = kaddr + offset;
1325 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1331 * Default Socket Callbacks
1334 static void sock_def_wakeup(struct sock *sk)
1336 read_lock(&sk->sk_callback_lock);
1337 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1338 wake_up_interruptible_all(sk->sk_sleep);
1339 read_unlock(&sk->sk_callback_lock);
1342 static void sock_def_error_report(struct sock *sk)
1344 read_lock(&sk->sk_callback_lock);
1345 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1346 wake_up_interruptible(sk->sk_sleep);
1347 sk_wake_async(sk,0,POLL_ERR);
1348 read_unlock(&sk->sk_callback_lock);
1351 static void sock_def_readable(struct sock *sk, int len)
1353 read_lock(&sk->sk_callback_lock);
1354 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1355 wake_up_interruptible(sk->sk_sleep);
1356 sk_wake_async(sk,1,POLL_IN);
1357 read_unlock(&sk->sk_callback_lock);
1360 static void sock_def_write_space(struct sock *sk)
1362 read_lock(&sk->sk_callback_lock);
1364 /* Do not wake up a writer until he can make "significant"
1367 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1368 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1369 wake_up_interruptible(sk->sk_sleep);
1371 /* Should agree with poll, otherwise some programs break */
1372 if (sock_writeable(sk))
1373 sk_wake_async(sk, 2, POLL_OUT);
1376 read_unlock(&sk->sk_callback_lock);
1379 static void sock_def_destruct(struct sock *sk)
1381 kfree(sk->sk_protinfo);
1384 void sk_send_sigurg(struct sock *sk)
1386 if (sk->sk_socket && sk->sk_socket->file)
1387 if (send_sigurg(&sk->sk_socket->file->f_owner))
1388 sk_wake_async(sk, 3, POLL_PRI);
1391 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1392 unsigned long expires)
1394 if (!mod_timer(timer, expires))
1398 EXPORT_SYMBOL(sk_reset_timer);
1400 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1402 if (timer_pending(timer) && del_timer(timer))
1406 EXPORT_SYMBOL(sk_stop_timer);
1408 void sock_init_data(struct socket *sock, struct sock *sk)
1410 skb_queue_head_init(&sk->sk_receive_queue);
1411 skb_queue_head_init(&sk->sk_write_queue);
1412 skb_queue_head_init(&sk->sk_error_queue);
1413 #ifdef CONFIG_NET_DMA
1414 skb_queue_head_init(&sk->sk_async_wait_queue);
1417 sk->sk_send_head = NULL;
1419 init_timer(&sk->sk_timer);
1421 sk->sk_allocation = GFP_KERNEL;
1422 sk->sk_rcvbuf = sysctl_rmem_default;
1423 sk->sk_sndbuf = sysctl_wmem_default;
1424 sk->sk_state = TCP_CLOSE;
1425 sk->sk_socket = sock;
1427 sock_set_flag(sk, SOCK_ZAPPED);
1431 sk->sk_type = sock->type;
1432 sk->sk_sleep = &sock->wait;
1435 sk->sk_sleep = NULL;
1437 rwlock_init(&sk->sk_dst_lock);
1438 rwlock_init(&sk->sk_callback_lock);
1439 lockdep_set_class(&sk->sk_callback_lock,
1440 af_callback_keys + sk->sk_family);
1442 sk->sk_state_change = sock_def_wakeup;
1443 sk->sk_data_ready = sock_def_readable;
1444 sk->sk_write_space = sock_def_write_space;
1445 sk->sk_error_report = sock_def_error_report;
1446 sk->sk_destruct = sock_def_destruct;
1448 sk->sk_sndmsg_page = NULL;
1449 sk->sk_sndmsg_off = 0;
1451 sk->sk_peercred.pid = 0;
1452 sk->sk_peercred.uid = -1;
1453 sk->sk_peercred.gid = -1;
1454 sk->sk_write_pending = 0;
1455 sk->sk_rcvlowat = 1;
1456 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1457 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1459 sk->sk_stamp.tv_sec = -1L;
1460 sk->sk_stamp.tv_usec = -1L;
1462 atomic_set(&sk->sk_refcnt, 1);
1465 void fastcall lock_sock(struct sock *sk)
1468 spin_lock_bh(&(sk->sk_lock.slock));
1469 if (sk->sk_lock.owner)
1471 sk->sk_lock.owner = (void *)1;
1472 spin_unlock_bh(&(sk->sk_lock.slock));
1475 EXPORT_SYMBOL(lock_sock);
1477 void fastcall release_sock(struct sock *sk)
1479 spin_lock_bh(&(sk->sk_lock.slock));
1480 if (sk->sk_backlog.tail)
1482 sk->sk_lock.owner = NULL;
1483 if (waitqueue_active(&(sk->sk_lock.wq)))
1484 wake_up(&(sk->sk_lock.wq));
1485 spin_unlock_bh(&(sk->sk_lock.slock));
1487 EXPORT_SYMBOL(release_sock);
1489 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1491 if (!sock_flag(sk, SOCK_TIMESTAMP))
1492 sock_enable_timestamp(sk);
1493 if (sk->sk_stamp.tv_sec == -1)
1495 if (sk->sk_stamp.tv_sec == 0)
1496 do_gettimeofday(&sk->sk_stamp);
1497 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1500 EXPORT_SYMBOL(sock_get_timestamp);
1502 void sock_enable_timestamp(struct sock *sk)
1504 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1505 sock_set_flag(sk, SOCK_TIMESTAMP);
1506 net_enable_timestamp();
1509 EXPORT_SYMBOL(sock_enable_timestamp);
1512 * Get a socket option on an socket.
1514 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1515 * asynchronous errors should be reported by getsockopt. We assume
1516 * this means if you specify SO_ERROR (otherwise whats the point of it).
1518 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1519 char __user *optval, int __user *optlen)
1521 struct sock *sk = sock->sk;
1523 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1526 EXPORT_SYMBOL(sock_common_getsockopt);
1528 #ifdef CONFIG_COMPAT
1529 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1530 char __user *optval, int __user *optlen)
1532 struct sock *sk = sock->sk;
1534 if (sk->sk_prot->compat_setsockopt != NULL)
1535 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1537 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1539 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1542 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1543 struct msghdr *msg, size_t size, int flags)
1545 struct sock *sk = sock->sk;
1549 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1550 flags & ~MSG_DONTWAIT, &addr_len);
1552 msg->msg_namelen = addr_len;
1556 EXPORT_SYMBOL(sock_common_recvmsg);
1559 * Set socket options on an inet socket.
1561 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1562 char __user *optval, int optlen)
1564 struct sock *sk = sock->sk;
1566 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1569 EXPORT_SYMBOL(sock_common_setsockopt);
1571 #ifdef CONFIG_COMPAT
1572 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1573 char __user *optval, int optlen)
1575 struct sock *sk = sock->sk;
1577 if (sk->sk_prot->compat_setsockopt != NULL)
1578 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1580 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1582 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1585 void sk_common_release(struct sock *sk)
1587 if (sk->sk_prot->destroy)
1588 sk->sk_prot->destroy(sk);
1591 * Observation: when sock_common_release is called, processes have
1592 * no access to socket. But net still has.
1593 * Step one, detach it from networking:
1595 * A. Remove from hash tables.
1598 sk->sk_prot->unhash(sk);
1601 * In this point socket cannot receive new packets, but it is possible
1602 * that some packets are in flight because some CPU runs receiver and
1603 * did hash table lookup before we unhashed socket. They will achieve
1604 * receive queue and will be purged by socket destructor.
1606 * Also we still have packets pending on receive queue and probably,
1607 * our own packets waiting in device queues. sock_destroy will drain
1608 * receive queue, but transmitted packets will delay socket destruction
1609 * until the last reference will be released.
1614 xfrm_sk_free_policy(sk);
1616 sk_refcnt_debug_release(sk);
1620 EXPORT_SYMBOL(sk_common_release);
1622 static DEFINE_RWLOCK(proto_list_lock);
1623 static LIST_HEAD(proto_list);
1625 int proto_register(struct proto *prot, int alloc_slab)
1627 char *request_sock_slab_name = NULL;
1628 char *timewait_sock_slab_name;
1632 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1633 SLAB_HWCACHE_ALIGN, NULL, NULL);
1635 if (prot->slab == NULL) {
1636 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1641 if (prot->rsk_prot != NULL) {
1642 static const char mask[] = "request_sock_%s";
1644 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1645 if (request_sock_slab_name == NULL)
1646 goto out_free_sock_slab;
1648 sprintf(request_sock_slab_name, mask, prot->name);
1649 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1650 prot->rsk_prot->obj_size, 0,
1651 SLAB_HWCACHE_ALIGN, NULL, NULL);
1653 if (prot->rsk_prot->slab == NULL) {
1654 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1656 goto out_free_request_sock_slab_name;
1660 if (prot->twsk_prot != NULL) {
1661 static const char mask[] = "tw_sock_%s";
1663 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1665 if (timewait_sock_slab_name == NULL)
1666 goto out_free_request_sock_slab;
1668 sprintf(timewait_sock_slab_name, mask, prot->name);
1669 prot->twsk_prot->twsk_slab =
1670 kmem_cache_create(timewait_sock_slab_name,
1671 prot->twsk_prot->twsk_obj_size,
1672 0, SLAB_HWCACHE_ALIGN,
1674 if (prot->twsk_prot->twsk_slab == NULL)
1675 goto out_free_timewait_sock_slab_name;
1679 write_lock(&proto_list_lock);
1680 list_add(&prot->node, &proto_list);
1681 write_unlock(&proto_list_lock);
1685 out_free_timewait_sock_slab_name:
1686 kfree(timewait_sock_slab_name);
1687 out_free_request_sock_slab:
1688 if (prot->rsk_prot && prot->rsk_prot->slab) {
1689 kmem_cache_destroy(prot->rsk_prot->slab);
1690 prot->rsk_prot->slab = NULL;
1692 out_free_request_sock_slab_name:
1693 kfree(request_sock_slab_name);
1695 kmem_cache_destroy(prot->slab);
1700 EXPORT_SYMBOL(proto_register);
1702 void proto_unregister(struct proto *prot)
1704 write_lock(&proto_list_lock);
1705 list_del(&prot->node);
1706 write_unlock(&proto_list_lock);
1708 if (prot->slab != NULL) {
1709 kmem_cache_destroy(prot->slab);
1713 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1714 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1716 kmem_cache_destroy(prot->rsk_prot->slab);
1718 prot->rsk_prot->slab = NULL;
1721 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1722 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1724 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1726 prot->twsk_prot->twsk_slab = NULL;
1730 EXPORT_SYMBOL(proto_unregister);
1732 #ifdef CONFIG_PROC_FS
1733 static inline struct proto *__proto_head(void)
1735 return list_entry(proto_list.next, struct proto, node);
1738 static inline struct proto *proto_head(void)
1740 return list_empty(&proto_list) ? NULL : __proto_head();
1743 static inline struct proto *proto_next(struct proto *proto)
1745 return proto->node.next == &proto_list ? NULL :
1746 list_entry(proto->node.next, struct proto, node);
1749 static inline struct proto *proto_get_idx(loff_t pos)
1751 struct proto *proto;
1754 list_for_each_entry(proto, &proto_list, node)
1763 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1765 read_lock(&proto_list_lock);
1766 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1769 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1772 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1775 static void proto_seq_stop(struct seq_file *seq, void *v)
1777 read_unlock(&proto_list_lock);
1780 static char proto_method_implemented(const void *method)
1782 return method == NULL ? 'n' : 'y';
1785 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1787 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1788 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1791 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1792 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1793 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1795 proto->slab == NULL ? "no" : "yes",
1796 module_name(proto->owner),
1797 proto_method_implemented(proto->close),
1798 proto_method_implemented(proto->connect),
1799 proto_method_implemented(proto->disconnect),
1800 proto_method_implemented(proto->accept),
1801 proto_method_implemented(proto->ioctl),
1802 proto_method_implemented(proto->init),
1803 proto_method_implemented(proto->destroy),
1804 proto_method_implemented(proto->shutdown),
1805 proto_method_implemented(proto->setsockopt),
1806 proto_method_implemented(proto->getsockopt),
1807 proto_method_implemented(proto->sendmsg),
1808 proto_method_implemented(proto->recvmsg),
1809 proto_method_implemented(proto->sendpage),
1810 proto_method_implemented(proto->bind),
1811 proto_method_implemented(proto->backlog_rcv),
1812 proto_method_implemented(proto->hash),
1813 proto_method_implemented(proto->unhash),
1814 proto_method_implemented(proto->get_port),
1815 proto_method_implemented(proto->enter_memory_pressure));
1818 static int proto_seq_show(struct seq_file *seq, void *v)
1820 if (v == SEQ_START_TOKEN)
1821 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1830 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1832 proto_seq_printf(seq, v);
1836 static struct seq_operations proto_seq_ops = {
1837 .start = proto_seq_start,
1838 .next = proto_seq_next,
1839 .stop = proto_seq_stop,
1840 .show = proto_seq_show,
1843 static int proto_seq_open(struct inode *inode, struct file *file)
1845 return seq_open(file, &proto_seq_ops);
1848 static struct file_operations proto_seq_fops = {
1849 .owner = THIS_MODULE,
1850 .open = proto_seq_open,
1852 .llseek = seq_lseek,
1853 .release = seq_release,
1856 static int __init proto_init(void)
1858 /* register /proc/net/protocols */
1859 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1862 subsys_initcall(proto_init);
1864 #endif /* PROC_FS */
1866 EXPORT_SYMBOL(sk_alloc);
1867 EXPORT_SYMBOL(sk_free);
1868 EXPORT_SYMBOL(sk_send_sigurg);
1869 EXPORT_SYMBOL(sock_alloc_send_skb);
1870 EXPORT_SYMBOL(sock_init_data);
1871 EXPORT_SYMBOL(sock_kfree_s);
1872 EXPORT_SYMBOL(sock_kmalloc);
1873 EXPORT_SYMBOL(sock_no_accept);
1874 EXPORT_SYMBOL(sock_no_bind);
1875 EXPORT_SYMBOL(sock_no_connect);
1876 EXPORT_SYMBOL(sock_no_getname);
1877 EXPORT_SYMBOL(sock_no_getsockopt);
1878 EXPORT_SYMBOL(sock_no_ioctl);
1879 EXPORT_SYMBOL(sock_no_listen);
1880 EXPORT_SYMBOL(sock_no_mmap);
1881 EXPORT_SYMBOL(sock_no_poll);
1882 EXPORT_SYMBOL(sock_no_recvmsg);
1883 EXPORT_SYMBOL(sock_no_sendmsg);
1884 EXPORT_SYMBOL(sock_no_sendpage);
1885 EXPORT_SYMBOL(sock_no_setsockopt);
1886 EXPORT_SYMBOL(sock_no_shutdown);
1887 EXPORT_SYMBOL(sock_no_socketpair);
1888 EXPORT_SYMBOL(sock_rfree);
1889 EXPORT_SYMBOL(sock_setsockopt);
1890 EXPORT_SYMBOL(sock_wfree);
1891 EXPORT_SYMBOL(sock_wmalloc);
1892 EXPORT_SYMBOL(sock_i_uid);
1893 EXPORT_SYMBOL(sock_i_ino);
1894 EXPORT_SYMBOL(sysctl_optmem_max);
1895 #ifdef CONFIG_SYSCTL
1896 EXPORT_SYMBOL(sysctl_rmem_max);
1897 EXPORT_SYMBOL(sysctl_wmem_max);