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 static struct lock_class_key af_family_keys[AF_MAX];
137 static struct lock_class_key af_family_slock_keys[AF_MAX];
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
141 * Make lock validator output more readable. (we pre-construct these
142 * strings build-time, so that runtime initialization of socket
145 static const char *af_family_key_strings[AF_MAX+1] = {
146 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
147 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
148 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
149 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
150 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
151 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
152 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
153 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
155 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
158 static const char *af_family_slock_key_strings[AF_MAX+1] = {
159 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
160 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
161 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
162 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
163 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
164 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
165 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
166 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
167 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
168 "slock-27" , "slock-28" , "slock-29" ,
169 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX"
174 * sk_callback_lock locking rules are per-address-family,
175 * so split the lock classes by using a per-AF key:
177 static struct lock_class_key af_callback_keys[AF_MAX];
179 /* Take into consideration the size of the struct sk_buff overhead in the
180 * determination of these values, since that is non-constant across
181 * platforms. This makes socket queueing behavior and performance
182 * not depend upon such differences.
184 #define _SK_MEM_PACKETS 256
185 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
186 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
187 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
189 /* Run time adjustable parameters. */
190 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
191 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
192 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
193 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
195 /* Maximal space eaten by iovec or ancilliary data plus some space */
196 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
198 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
202 if (optlen < sizeof(tv))
204 if (copy_from_user(&tv, optval, sizeof(tv)))
207 *timeo_p = MAX_SCHEDULE_TIMEOUT;
208 if (tv.tv_sec == 0 && tv.tv_usec == 0)
210 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
211 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
215 static void sock_warn_obsolete_bsdism(const char *name)
218 static char warncomm[TASK_COMM_LEN];
219 if (strcmp(warncomm, current->comm) && warned < 5) {
220 strcpy(warncomm, current->comm);
221 printk(KERN_WARNING "process `%s' is using obsolete "
222 "%s SO_BSDCOMPAT\n", warncomm, name);
227 static void sock_disable_timestamp(struct sock *sk)
229 if (sock_flag(sk, SOCK_TIMESTAMP)) {
230 sock_reset_flag(sk, SOCK_TIMESTAMP);
231 net_disable_timestamp();
236 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
241 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
242 number of warnings when compiling with -W --ANK
244 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
245 (unsigned)sk->sk_rcvbuf) {
250 err = sk_filter(sk, skb);
255 skb_set_owner_r(skb, sk);
257 /* Cache the SKB length before we tack it onto the receive
258 * queue. Once it is added it no longer belongs to us and
259 * may be freed by other threads of control pulling packets
264 skb_queue_tail(&sk->sk_receive_queue, skb);
266 if (!sock_flag(sk, SOCK_DEAD))
267 sk->sk_data_ready(sk, skb_len);
271 EXPORT_SYMBOL(sock_queue_rcv_skb);
273 int sk_receive_skb(struct sock *sk, struct sk_buff *skb)
275 int rc = NET_RX_SUCCESS;
277 if (sk_filter(sk, skb))
278 goto discard_and_relse;
283 if (!sock_owned_by_user(sk)) {
285 * trylock + unlock semantics:
287 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
289 rc = sk->sk_backlog_rcv(sk, skb);
291 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
293 sk_add_backlog(sk, skb);
302 EXPORT_SYMBOL(sk_receive_skb);
304 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
306 struct dst_entry *dst = sk->sk_dst_cache;
308 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
309 sk->sk_dst_cache = NULL;
316 EXPORT_SYMBOL(__sk_dst_check);
318 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
320 struct dst_entry *dst = sk_dst_get(sk);
322 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
330 EXPORT_SYMBOL(sk_dst_check);
333 * This is meant for all protocols to use and covers goings on
334 * at the socket level. Everything here is generic.
337 int sock_setsockopt(struct socket *sock, int level, int optname,
338 char __user *optval, int optlen)
340 struct sock *sk=sock->sk;
341 struct sk_filter *filter;
348 * Options without arguments
351 #ifdef SO_DONTLINGER /* Compatibility item... */
352 if (optname == SO_DONTLINGER) {
354 sock_reset_flag(sk, SOCK_LINGER);
360 if(optlen<sizeof(int))
363 if (get_user(val, (int __user *)optval))
373 if(val && !capable(CAP_NET_ADMIN))
378 sock_set_flag(sk, SOCK_DBG);
380 sock_reset_flag(sk, SOCK_DBG);
383 sk->sk_reuse = valbool;
391 sock_set_flag(sk, SOCK_LOCALROUTE);
393 sock_reset_flag(sk, SOCK_LOCALROUTE);
396 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
399 /* Don't error on this BSD doesn't and if you think
400 about it this is right. Otherwise apps have to
401 play 'guess the biggest size' games. RCVBUF/SNDBUF
402 are treated in BSD as hints */
404 if (val > sysctl_wmem_max)
405 val = sysctl_wmem_max;
407 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
408 if ((val * 2) < SOCK_MIN_SNDBUF)
409 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
411 sk->sk_sndbuf = val * 2;
414 * Wake up sending tasks if we
417 sk->sk_write_space(sk);
421 if (!capable(CAP_NET_ADMIN)) {
428 /* Don't error on this BSD doesn't and if you think
429 about it this is right. Otherwise apps have to
430 play 'guess the biggest size' games. RCVBUF/SNDBUF
431 are treated in BSD as hints */
433 if (val > sysctl_rmem_max)
434 val = sysctl_rmem_max;
436 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
438 * We double it on the way in to account for
439 * "struct sk_buff" etc. overhead. Applications
440 * assume that the SO_RCVBUF setting they make will
441 * allow that much actual data to be received on that
444 * Applications are unaware that "struct sk_buff" and
445 * other overheads allocate from the receive buffer
446 * during socket buffer allocation.
448 * And after considering the possible alternatives,
449 * returning the value we actually used in getsockopt
450 * is the most desirable behavior.
452 if ((val * 2) < SOCK_MIN_RCVBUF)
453 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
455 sk->sk_rcvbuf = val * 2;
459 if (!capable(CAP_NET_ADMIN)) {
467 if (sk->sk_protocol == IPPROTO_TCP)
468 tcp_set_keepalive(sk, valbool);
470 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
474 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
478 sk->sk_no_check = valbool;
482 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
483 sk->sk_priority = val;
489 if(optlen<sizeof(ling)) {
490 ret = -EINVAL; /* 1003.1g */
493 if (copy_from_user(&ling,optval,sizeof(ling))) {
498 sock_reset_flag(sk, SOCK_LINGER);
500 #if (BITS_PER_LONG == 32)
501 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
502 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
505 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
506 sock_set_flag(sk, SOCK_LINGER);
511 sock_warn_obsolete_bsdism("setsockopt");
516 set_bit(SOCK_PASSCRED, &sock->flags);
518 clear_bit(SOCK_PASSCRED, &sock->flags);
523 sock_set_flag(sk, SOCK_RCVTSTAMP);
524 sock_enable_timestamp(sk);
526 sock_reset_flag(sk, SOCK_RCVTSTAMP);
532 sk->sk_rcvlowat = val ? : 1;
536 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
540 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
543 #ifdef CONFIG_NETDEVICES
544 case SO_BINDTODEVICE:
546 char devname[IFNAMSIZ];
549 if (!capable(CAP_NET_RAW)) {
554 /* Bind this socket to a particular device like "eth0",
555 * as specified in the passed interface name. If the
556 * name is "" or the option length is zero the socket
561 sk->sk_bound_dev_if = 0;
563 if (optlen > IFNAMSIZ - 1)
564 optlen = IFNAMSIZ - 1;
565 memset(devname, 0, sizeof(devname));
566 if (copy_from_user(devname, optval, optlen)) {
571 /* Remove any cached route for this socket. */
574 if (devname[0] == '\0') {
575 sk->sk_bound_dev_if = 0;
577 struct net_device *dev = dev_get_by_name(devname);
582 sk->sk_bound_dev_if = dev->ifindex;
591 case SO_ATTACH_FILTER:
593 if (optlen == sizeof(struct sock_fprog)) {
594 struct sock_fprog fprog;
597 if (copy_from_user(&fprog, optval, sizeof(fprog)))
600 ret = sk_attach_filter(&fprog, sk);
604 case SO_DETACH_FILTER:
606 filter = rcu_dereference(sk->sk_filter);
608 rcu_assign_pointer(sk->sk_filter, NULL);
609 sk_filter_release(sk, filter);
610 rcu_read_unlock_bh();
613 rcu_read_unlock_bh();
619 set_bit(SOCK_PASSSEC, &sock->flags);
621 clear_bit(SOCK_PASSSEC, &sock->flags);
624 /* We implement the SO_SNDLOWAT etc to
625 not be settable (1003.1g 5.3) */
635 int sock_getsockopt(struct socket *sock, int level, int optname,
636 char __user *optval, int __user *optlen)
638 struct sock *sk = sock->sk;
647 unsigned int lv = sizeof(int);
650 if(get_user(len,optlen))
658 v.val = sock_flag(sk, SOCK_DBG);
662 v.val = sock_flag(sk, SOCK_LOCALROUTE);
666 v.val = !!sock_flag(sk, SOCK_BROADCAST);
670 v.val = sk->sk_sndbuf;
674 v.val = sk->sk_rcvbuf;
678 v.val = sk->sk_reuse;
682 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
690 v.val = -sock_error(sk);
692 v.val = xchg(&sk->sk_err_soft, 0);
696 v.val = !!sock_flag(sk, SOCK_URGINLINE);
700 v.val = sk->sk_no_check;
704 v.val = sk->sk_priority;
709 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
710 v.ling.l_linger = sk->sk_lingertime / HZ;
714 sock_warn_obsolete_bsdism("getsockopt");
718 v.val = sock_flag(sk, SOCK_RCVTSTAMP);
722 lv=sizeof(struct timeval);
723 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
727 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
728 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
733 lv=sizeof(struct timeval);
734 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
738 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
739 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
744 v.val = sk->sk_rcvlowat;
752 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
756 if (len > sizeof(sk->sk_peercred))
757 len = sizeof(sk->sk_peercred);
758 if (copy_to_user(optval, &sk->sk_peercred, len))
766 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
770 if (copy_to_user(optval, address, len))
775 /* Dubious BSD thing... Probably nobody even uses it, but
776 * the UNIX standard wants it for whatever reason... -DaveM
779 v.val = sk->sk_state == TCP_LISTEN;
783 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
787 return security_socket_getpeersec_stream(sock, optval, optlen, len);
790 return(-ENOPROTOOPT);
794 if (copy_to_user(optval, &v, len))
797 if (put_user(len, optlen))
803 * Initialize an sk_lock.
805 * (We also register the sk_lock with the lock validator.)
807 static void inline sock_lock_init(struct sock *sk)
809 spin_lock_init(&sk->sk_lock.slock);
810 sk->sk_lock.owner = NULL;
811 init_waitqueue_head(&sk->sk_lock.wq);
813 * Make sure we are not reinitializing a held lock:
815 debug_check_no_locks_freed((void *)&sk->sk_lock, sizeof(sk->sk_lock));
818 * Mark both the sk_lock and the sk_lock.slock as a
819 * per-address-family lock class:
821 lockdep_set_class_and_name(&sk->sk_lock.slock,
822 af_family_slock_keys + sk->sk_family,
823 af_family_slock_key_strings[sk->sk_family]);
824 lockdep_init_map(&sk->sk_lock.dep_map,
825 af_family_key_strings[sk->sk_family],
826 af_family_keys + sk->sk_family);
830 * sk_alloc - All socket objects are allocated here
831 * @family: protocol family
832 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
833 * @prot: struct proto associated with this new sock instance
834 * @zero_it: if we should zero the newly allocated sock
836 struct sock *sk_alloc(int family, gfp_t priority,
837 struct proto *prot, int zero_it)
839 struct sock *sk = NULL;
840 kmem_cache_t *slab = prot->slab;
843 sk = kmem_cache_alloc(slab, priority);
845 sk = kmalloc(prot->obj_size, priority);
849 memset(sk, 0, prot->obj_size);
850 sk->sk_family = family;
852 * See comment in struct sock definition to understand
853 * why we need sk_prot_creator -acme
855 sk->sk_prot = sk->sk_prot_creator = prot;
859 if (security_sk_alloc(sk, family, priority))
862 if (!try_module_get(prot->owner))
869 kmem_cache_free(slab, sk);
875 void sk_free(struct sock *sk)
877 struct sk_filter *filter;
878 struct module *owner = sk->sk_prot_creator->owner;
883 filter = rcu_dereference(sk->sk_filter);
885 sk_filter_release(sk, filter);
886 rcu_assign_pointer(sk->sk_filter, NULL);
889 sock_disable_timestamp(sk);
891 if (atomic_read(&sk->sk_omem_alloc))
892 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
893 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
895 security_sk_free(sk);
896 if (sk->sk_prot_creator->slab != NULL)
897 kmem_cache_free(sk->sk_prot_creator->slab, sk);
903 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
905 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
908 struct sk_filter *filter;
910 sock_copy(newsk, sk);
913 sk_node_init(&newsk->sk_node);
914 sock_lock_init(newsk);
917 atomic_set(&newsk->sk_rmem_alloc, 0);
918 atomic_set(&newsk->sk_wmem_alloc, 0);
919 atomic_set(&newsk->sk_omem_alloc, 0);
920 skb_queue_head_init(&newsk->sk_receive_queue);
921 skb_queue_head_init(&newsk->sk_write_queue);
922 #ifdef CONFIG_NET_DMA
923 skb_queue_head_init(&newsk->sk_async_wait_queue);
926 rwlock_init(&newsk->sk_dst_lock);
927 rwlock_init(&newsk->sk_callback_lock);
928 lockdep_set_class(&newsk->sk_callback_lock,
929 af_callback_keys + newsk->sk_family);
931 newsk->sk_dst_cache = NULL;
932 newsk->sk_wmem_queued = 0;
933 newsk->sk_forward_alloc = 0;
934 newsk->sk_send_head = NULL;
935 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
936 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
938 sock_reset_flag(newsk, SOCK_DONE);
939 skb_queue_head_init(&newsk->sk_error_queue);
941 filter = newsk->sk_filter;
943 sk_filter_charge(newsk, filter);
945 if (unlikely(xfrm_sk_clone_policy(newsk))) {
946 /* It is still raw copy of parent, so invalidate
947 * destructor and make plain sk_free() */
948 newsk->sk_destruct = NULL;
955 newsk->sk_priority = 0;
956 atomic_set(&newsk->sk_refcnt, 2);
959 * Increment the counter in the same struct proto as the master
960 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
961 * is the same as sk->sk_prot->socks, as this field was copied
964 * This _changes_ the previous behaviour, where
965 * tcp_create_openreq_child always was incrementing the
966 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
967 * to be taken into account in all callers. -acme
969 sk_refcnt_debug_inc(newsk);
970 newsk->sk_socket = NULL;
971 newsk->sk_sleep = NULL;
973 if (newsk->sk_prot->sockets_allocated)
974 atomic_inc(newsk->sk_prot->sockets_allocated);
980 EXPORT_SYMBOL_GPL(sk_clone);
982 void __init sk_init(void)
984 if (num_physpages <= 4096) {
985 sysctl_wmem_max = 32767;
986 sysctl_rmem_max = 32767;
987 sysctl_wmem_default = 32767;
988 sysctl_rmem_default = 32767;
989 } else if (num_physpages >= 131072) {
990 sysctl_wmem_max = 131071;
991 sysctl_rmem_max = 131071;
996 * Simple resource managers for sockets.
1001 * Write buffer destructor automatically called from kfree_skb.
1003 void sock_wfree(struct sk_buff *skb)
1005 struct sock *sk = skb->sk;
1007 /* In case it might be waiting for more memory. */
1008 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1009 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1010 sk->sk_write_space(sk);
1015 * Read buffer destructor automatically called from kfree_skb.
1017 void sock_rfree(struct sk_buff *skb)
1019 struct sock *sk = skb->sk;
1021 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1025 int sock_i_uid(struct sock *sk)
1029 read_lock(&sk->sk_callback_lock);
1030 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1031 read_unlock(&sk->sk_callback_lock);
1035 unsigned long sock_i_ino(struct sock *sk)
1039 read_lock(&sk->sk_callback_lock);
1040 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1041 read_unlock(&sk->sk_callback_lock);
1046 * Allocate a skb from the socket's send buffer.
1048 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1051 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1052 struct sk_buff * skb = alloc_skb(size, priority);
1054 skb_set_owner_w(skb, sk);
1062 * Allocate a skb from the socket's receive buffer.
1064 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1067 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1068 struct sk_buff *skb = alloc_skb(size, priority);
1070 skb_set_owner_r(skb, sk);
1078 * Allocate a memory block from the socket's option memory buffer.
1080 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1082 if ((unsigned)size <= sysctl_optmem_max &&
1083 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1085 /* First do the add, to avoid the race if kmalloc
1088 atomic_add(size, &sk->sk_omem_alloc);
1089 mem = kmalloc(size, priority);
1092 atomic_sub(size, &sk->sk_omem_alloc);
1098 * Free an option memory block.
1100 void sock_kfree_s(struct sock *sk, void *mem, int size)
1103 atomic_sub(size, &sk->sk_omem_alloc);
1106 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1107 I think, these locks should be removed for datagram sockets.
1109 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1113 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1117 if (signal_pending(current))
1119 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1120 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1121 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1123 if (sk->sk_shutdown & SEND_SHUTDOWN)
1127 timeo = schedule_timeout(timeo);
1129 finish_wait(sk->sk_sleep, &wait);
1135 * Generic send/receive buffer handlers
1138 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1139 unsigned long header_len,
1140 unsigned long data_len,
1141 int noblock, int *errcode)
1143 struct sk_buff *skb;
1148 gfp_mask = sk->sk_allocation;
1149 if (gfp_mask & __GFP_WAIT)
1150 gfp_mask |= __GFP_REPEAT;
1152 timeo = sock_sndtimeo(sk, noblock);
1154 err = sock_error(sk);
1159 if (sk->sk_shutdown & SEND_SHUTDOWN)
1162 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1163 skb = alloc_skb(header_len, sk->sk_allocation);
1168 /* No pages, we're done... */
1172 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1173 skb->truesize += data_len;
1174 skb_shinfo(skb)->nr_frags = npages;
1175 for (i = 0; i < npages; i++) {
1179 page = alloc_pages(sk->sk_allocation, 0);
1182 skb_shinfo(skb)->nr_frags = i;
1187 frag = &skb_shinfo(skb)->frags[i];
1189 frag->page_offset = 0;
1190 frag->size = (data_len >= PAGE_SIZE ?
1193 data_len -= PAGE_SIZE;
1196 /* Full success... */
1202 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1203 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1207 if (signal_pending(current))
1209 timeo = sock_wait_for_wmem(sk, timeo);
1212 skb_set_owner_w(skb, sk);
1216 err = sock_intr_errno(timeo);
1222 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1223 int noblock, int *errcode)
1225 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1228 static void __lock_sock(struct sock *sk)
1233 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1234 TASK_UNINTERRUPTIBLE);
1235 spin_unlock_bh(&sk->sk_lock.slock);
1237 spin_lock_bh(&sk->sk_lock.slock);
1238 if(!sock_owned_by_user(sk))
1241 finish_wait(&sk->sk_lock.wq, &wait);
1244 static void __release_sock(struct sock *sk)
1246 struct sk_buff *skb = sk->sk_backlog.head;
1249 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1253 struct sk_buff *next = skb->next;
1256 sk->sk_backlog_rcv(sk, skb);
1259 * We are in process context here with softirqs
1260 * disabled, use cond_resched_softirq() to preempt.
1261 * This is safe to do because we've taken the backlog
1264 cond_resched_softirq();
1267 } while (skb != NULL);
1270 } while((skb = sk->sk_backlog.head) != NULL);
1274 * sk_wait_data - wait for data to arrive at sk_receive_queue
1275 * @sk: sock to wait on
1276 * @timeo: for how long
1278 * Now socket state including sk->sk_err is changed only under lock,
1279 * hence we may omit checks after joining wait queue.
1280 * We check receive queue before schedule() only as optimization;
1281 * it is very likely that release_sock() added new data.
1283 int sk_wait_data(struct sock *sk, long *timeo)
1288 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1289 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1290 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1291 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1292 finish_wait(sk->sk_sleep, &wait);
1296 EXPORT_SYMBOL(sk_wait_data);
1299 * Set of default routines for initialising struct proto_ops when
1300 * the protocol does not support a particular function. In certain
1301 * cases where it makes no sense for a protocol to have a "do nothing"
1302 * function, some default processing is provided.
1305 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1310 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1316 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1321 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1326 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1332 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1337 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1342 int sock_no_listen(struct socket *sock, int backlog)
1347 int sock_no_shutdown(struct socket *sock, int how)
1352 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1353 char __user *optval, int optlen)
1358 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1359 char __user *optval, int __user *optlen)
1364 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1370 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1371 size_t len, int flags)
1376 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1378 /* Mirror missing mmap method error code */
1382 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1385 struct msghdr msg = {.msg_flags = flags};
1387 char *kaddr = kmap(page);
1388 iov.iov_base = kaddr + offset;
1390 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1396 * Default Socket Callbacks
1399 static void sock_def_wakeup(struct sock *sk)
1401 read_lock(&sk->sk_callback_lock);
1402 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1403 wake_up_interruptible_all(sk->sk_sleep);
1404 read_unlock(&sk->sk_callback_lock);
1407 static void sock_def_error_report(struct sock *sk)
1409 read_lock(&sk->sk_callback_lock);
1410 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1411 wake_up_interruptible(sk->sk_sleep);
1412 sk_wake_async(sk,0,POLL_ERR);
1413 read_unlock(&sk->sk_callback_lock);
1416 static void sock_def_readable(struct sock *sk, int len)
1418 read_lock(&sk->sk_callback_lock);
1419 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1420 wake_up_interruptible(sk->sk_sleep);
1421 sk_wake_async(sk,1,POLL_IN);
1422 read_unlock(&sk->sk_callback_lock);
1425 static void sock_def_write_space(struct sock *sk)
1427 read_lock(&sk->sk_callback_lock);
1429 /* Do not wake up a writer until he can make "significant"
1432 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1433 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1434 wake_up_interruptible(sk->sk_sleep);
1436 /* Should agree with poll, otherwise some programs break */
1437 if (sock_writeable(sk))
1438 sk_wake_async(sk, 2, POLL_OUT);
1441 read_unlock(&sk->sk_callback_lock);
1444 static void sock_def_destruct(struct sock *sk)
1446 kfree(sk->sk_protinfo);
1449 void sk_send_sigurg(struct sock *sk)
1451 if (sk->sk_socket && sk->sk_socket->file)
1452 if (send_sigurg(&sk->sk_socket->file->f_owner))
1453 sk_wake_async(sk, 3, POLL_PRI);
1456 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1457 unsigned long expires)
1459 if (!mod_timer(timer, expires))
1463 EXPORT_SYMBOL(sk_reset_timer);
1465 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1467 if (timer_pending(timer) && del_timer(timer))
1471 EXPORT_SYMBOL(sk_stop_timer);
1473 void sock_init_data(struct socket *sock, struct sock *sk)
1475 skb_queue_head_init(&sk->sk_receive_queue);
1476 skb_queue_head_init(&sk->sk_write_queue);
1477 skb_queue_head_init(&sk->sk_error_queue);
1478 #ifdef CONFIG_NET_DMA
1479 skb_queue_head_init(&sk->sk_async_wait_queue);
1482 sk->sk_send_head = NULL;
1484 init_timer(&sk->sk_timer);
1486 sk->sk_allocation = GFP_KERNEL;
1487 sk->sk_rcvbuf = sysctl_rmem_default;
1488 sk->sk_sndbuf = sysctl_wmem_default;
1489 sk->sk_state = TCP_CLOSE;
1490 sk->sk_socket = sock;
1492 sock_set_flag(sk, SOCK_ZAPPED);
1496 sk->sk_type = sock->type;
1497 sk->sk_sleep = &sock->wait;
1500 sk->sk_sleep = NULL;
1502 rwlock_init(&sk->sk_dst_lock);
1503 rwlock_init(&sk->sk_callback_lock);
1504 lockdep_set_class(&sk->sk_callback_lock,
1505 af_callback_keys + sk->sk_family);
1507 sk->sk_state_change = sock_def_wakeup;
1508 sk->sk_data_ready = sock_def_readable;
1509 sk->sk_write_space = sock_def_write_space;
1510 sk->sk_error_report = sock_def_error_report;
1511 sk->sk_destruct = sock_def_destruct;
1513 sk->sk_sndmsg_page = NULL;
1514 sk->sk_sndmsg_off = 0;
1516 sk->sk_peercred.pid = 0;
1517 sk->sk_peercred.uid = -1;
1518 sk->sk_peercred.gid = -1;
1519 sk->sk_write_pending = 0;
1520 sk->sk_rcvlowat = 1;
1521 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1522 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1524 sk->sk_stamp.tv_sec = -1L;
1525 sk->sk_stamp.tv_usec = -1L;
1527 atomic_set(&sk->sk_refcnt, 1);
1530 void fastcall lock_sock(struct sock *sk)
1533 spin_lock_bh(&sk->sk_lock.slock);
1534 if (sk->sk_lock.owner)
1536 sk->sk_lock.owner = (void *)1;
1537 spin_unlock(&sk->sk_lock.slock);
1539 * The sk_lock has mutex_lock() semantics here:
1541 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1545 EXPORT_SYMBOL(lock_sock);
1547 void fastcall release_sock(struct sock *sk)
1550 * The sk_lock has mutex_unlock() semantics:
1552 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1554 spin_lock_bh(&sk->sk_lock.slock);
1555 if (sk->sk_backlog.tail)
1557 sk->sk_lock.owner = NULL;
1558 if (waitqueue_active(&sk->sk_lock.wq))
1559 wake_up(&sk->sk_lock.wq);
1560 spin_unlock_bh(&sk->sk_lock.slock);
1562 EXPORT_SYMBOL(release_sock);
1564 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1566 if (!sock_flag(sk, SOCK_TIMESTAMP))
1567 sock_enable_timestamp(sk);
1568 if (sk->sk_stamp.tv_sec == -1)
1570 if (sk->sk_stamp.tv_sec == 0)
1571 do_gettimeofday(&sk->sk_stamp);
1572 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1575 EXPORT_SYMBOL(sock_get_timestamp);
1577 void sock_enable_timestamp(struct sock *sk)
1579 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1580 sock_set_flag(sk, SOCK_TIMESTAMP);
1581 net_enable_timestamp();
1584 EXPORT_SYMBOL(sock_enable_timestamp);
1587 * Get a socket option on an socket.
1589 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1590 * asynchronous errors should be reported by getsockopt. We assume
1591 * this means if you specify SO_ERROR (otherwise whats the point of it).
1593 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1594 char __user *optval, int __user *optlen)
1596 struct sock *sk = sock->sk;
1598 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1601 EXPORT_SYMBOL(sock_common_getsockopt);
1603 #ifdef CONFIG_COMPAT
1604 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1605 char __user *optval, int __user *optlen)
1607 struct sock *sk = sock->sk;
1609 if (sk->sk_prot->compat_setsockopt != NULL)
1610 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1612 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1614 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1617 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1618 struct msghdr *msg, size_t size, int flags)
1620 struct sock *sk = sock->sk;
1624 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1625 flags & ~MSG_DONTWAIT, &addr_len);
1627 msg->msg_namelen = addr_len;
1631 EXPORT_SYMBOL(sock_common_recvmsg);
1634 * Set socket options on an inet socket.
1636 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1637 char __user *optval, int optlen)
1639 struct sock *sk = sock->sk;
1641 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1644 EXPORT_SYMBOL(sock_common_setsockopt);
1646 #ifdef CONFIG_COMPAT
1647 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1648 char __user *optval, int optlen)
1650 struct sock *sk = sock->sk;
1652 if (sk->sk_prot->compat_setsockopt != NULL)
1653 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1655 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1657 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1660 void sk_common_release(struct sock *sk)
1662 if (sk->sk_prot->destroy)
1663 sk->sk_prot->destroy(sk);
1666 * Observation: when sock_common_release is called, processes have
1667 * no access to socket. But net still has.
1668 * Step one, detach it from networking:
1670 * A. Remove from hash tables.
1673 sk->sk_prot->unhash(sk);
1676 * In this point socket cannot receive new packets, but it is possible
1677 * that some packets are in flight because some CPU runs receiver and
1678 * did hash table lookup before we unhashed socket. They will achieve
1679 * receive queue and will be purged by socket destructor.
1681 * Also we still have packets pending on receive queue and probably,
1682 * our own packets waiting in device queues. sock_destroy will drain
1683 * receive queue, but transmitted packets will delay socket destruction
1684 * until the last reference will be released.
1689 xfrm_sk_free_policy(sk);
1691 sk_refcnt_debug_release(sk);
1695 EXPORT_SYMBOL(sk_common_release);
1697 static DEFINE_RWLOCK(proto_list_lock);
1698 static LIST_HEAD(proto_list);
1700 int proto_register(struct proto *prot, int alloc_slab)
1702 char *request_sock_slab_name = NULL;
1703 char *timewait_sock_slab_name;
1707 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1708 SLAB_HWCACHE_ALIGN, NULL, NULL);
1710 if (prot->slab == NULL) {
1711 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1716 if (prot->rsk_prot != NULL) {
1717 static const char mask[] = "request_sock_%s";
1719 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1720 if (request_sock_slab_name == NULL)
1721 goto out_free_sock_slab;
1723 sprintf(request_sock_slab_name, mask, prot->name);
1724 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1725 prot->rsk_prot->obj_size, 0,
1726 SLAB_HWCACHE_ALIGN, NULL, NULL);
1728 if (prot->rsk_prot->slab == NULL) {
1729 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1731 goto out_free_request_sock_slab_name;
1735 if (prot->twsk_prot != NULL) {
1736 static const char mask[] = "tw_sock_%s";
1738 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1740 if (timewait_sock_slab_name == NULL)
1741 goto out_free_request_sock_slab;
1743 sprintf(timewait_sock_slab_name, mask, prot->name);
1744 prot->twsk_prot->twsk_slab =
1745 kmem_cache_create(timewait_sock_slab_name,
1746 prot->twsk_prot->twsk_obj_size,
1747 0, SLAB_HWCACHE_ALIGN,
1749 if (prot->twsk_prot->twsk_slab == NULL)
1750 goto out_free_timewait_sock_slab_name;
1754 write_lock(&proto_list_lock);
1755 list_add(&prot->node, &proto_list);
1756 write_unlock(&proto_list_lock);
1760 out_free_timewait_sock_slab_name:
1761 kfree(timewait_sock_slab_name);
1762 out_free_request_sock_slab:
1763 if (prot->rsk_prot && prot->rsk_prot->slab) {
1764 kmem_cache_destroy(prot->rsk_prot->slab);
1765 prot->rsk_prot->slab = NULL;
1767 out_free_request_sock_slab_name:
1768 kfree(request_sock_slab_name);
1770 kmem_cache_destroy(prot->slab);
1775 EXPORT_SYMBOL(proto_register);
1777 void proto_unregister(struct proto *prot)
1779 write_lock(&proto_list_lock);
1780 list_del(&prot->node);
1781 write_unlock(&proto_list_lock);
1783 if (prot->slab != NULL) {
1784 kmem_cache_destroy(prot->slab);
1788 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1789 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1791 kmem_cache_destroy(prot->rsk_prot->slab);
1793 prot->rsk_prot->slab = NULL;
1796 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1797 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1799 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1801 prot->twsk_prot->twsk_slab = NULL;
1805 EXPORT_SYMBOL(proto_unregister);
1807 #ifdef CONFIG_PROC_FS
1808 static inline struct proto *__proto_head(void)
1810 return list_entry(proto_list.next, struct proto, node);
1813 static inline struct proto *proto_head(void)
1815 return list_empty(&proto_list) ? NULL : __proto_head();
1818 static inline struct proto *proto_next(struct proto *proto)
1820 return proto->node.next == &proto_list ? NULL :
1821 list_entry(proto->node.next, struct proto, node);
1824 static inline struct proto *proto_get_idx(loff_t pos)
1826 struct proto *proto;
1829 list_for_each_entry(proto, &proto_list, node)
1838 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1840 read_lock(&proto_list_lock);
1841 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1844 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1847 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1850 static void proto_seq_stop(struct seq_file *seq, void *v)
1852 read_unlock(&proto_list_lock);
1855 static char proto_method_implemented(const void *method)
1857 return method == NULL ? 'n' : 'y';
1860 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1862 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1863 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1866 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1867 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1868 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1870 proto->slab == NULL ? "no" : "yes",
1871 module_name(proto->owner),
1872 proto_method_implemented(proto->close),
1873 proto_method_implemented(proto->connect),
1874 proto_method_implemented(proto->disconnect),
1875 proto_method_implemented(proto->accept),
1876 proto_method_implemented(proto->ioctl),
1877 proto_method_implemented(proto->init),
1878 proto_method_implemented(proto->destroy),
1879 proto_method_implemented(proto->shutdown),
1880 proto_method_implemented(proto->setsockopt),
1881 proto_method_implemented(proto->getsockopt),
1882 proto_method_implemented(proto->sendmsg),
1883 proto_method_implemented(proto->recvmsg),
1884 proto_method_implemented(proto->sendpage),
1885 proto_method_implemented(proto->bind),
1886 proto_method_implemented(proto->backlog_rcv),
1887 proto_method_implemented(proto->hash),
1888 proto_method_implemented(proto->unhash),
1889 proto_method_implemented(proto->get_port),
1890 proto_method_implemented(proto->enter_memory_pressure));
1893 static int proto_seq_show(struct seq_file *seq, void *v)
1895 if (v == SEQ_START_TOKEN)
1896 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1905 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1907 proto_seq_printf(seq, v);
1911 static struct seq_operations proto_seq_ops = {
1912 .start = proto_seq_start,
1913 .next = proto_seq_next,
1914 .stop = proto_seq_stop,
1915 .show = proto_seq_show,
1918 static int proto_seq_open(struct inode *inode, struct file *file)
1920 return seq_open(file, &proto_seq_ops);
1923 static struct file_operations proto_seq_fops = {
1924 .owner = THIS_MODULE,
1925 .open = proto_seq_open,
1927 .llseek = seq_lseek,
1928 .release = seq_release,
1931 static int __init proto_init(void)
1933 /* register /proc/net/protocols */
1934 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1937 subsys_initcall(proto_init);
1939 #endif /* PROC_FS */
1941 EXPORT_SYMBOL(sk_alloc);
1942 EXPORT_SYMBOL(sk_free);
1943 EXPORT_SYMBOL(sk_send_sigurg);
1944 EXPORT_SYMBOL(sock_alloc_send_skb);
1945 EXPORT_SYMBOL(sock_init_data);
1946 EXPORT_SYMBOL(sock_kfree_s);
1947 EXPORT_SYMBOL(sock_kmalloc);
1948 EXPORT_SYMBOL(sock_no_accept);
1949 EXPORT_SYMBOL(sock_no_bind);
1950 EXPORT_SYMBOL(sock_no_connect);
1951 EXPORT_SYMBOL(sock_no_getname);
1952 EXPORT_SYMBOL(sock_no_getsockopt);
1953 EXPORT_SYMBOL(sock_no_ioctl);
1954 EXPORT_SYMBOL(sock_no_listen);
1955 EXPORT_SYMBOL(sock_no_mmap);
1956 EXPORT_SYMBOL(sock_no_poll);
1957 EXPORT_SYMBOL(sock_no_recvmsg);
1958 EXPORT_SYMBOL(sock_no_sendmsg);
1959 EXPORT_SYMBOL(sock_no_sendpage);
1960 EXPORT_SYMBOL(sock_no_setsockopt);
1961 EXPORT_SYMBOL(sock_no_shutdown);
1962 EXPORT_SYMBOL(sock_no_socketpair);
1963 EXPORT_SYMBOL(sock_rfree);
1964 EXPORT_SYMBOL(sock_setsockopt);
1965 EXPORT_SYMBOL(sock_wfree);
1966 EXPORT_SYMBOL(sock_wmalloc);
1967 EXPORT_SYMBOL(sock_i_uid);
1968 EXPORT_SYMBOL(sock_i_ino);
1969 EXPORT_SYMBOL(sysctl_optmem_max);
1970 #ifdef CONFIG_SYSCTL
1971 EXPORT_SYMBOL(sysctl_rmem_max);
1972 EXPORT_SYMBOL(sysctl_wmem_max);