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>
127 #include <linux/vs_socket.h>
128 #include <linux/vs_limit.h>
129 #include <linux/vs_context.h>
136 * Each address family might have different locking rules, so we have
137 * one slock key per address family:
139 static struct lock_class_key af_family_keys[AF_MAX];
140 static struct lock_class_key af_family_slock_keys[AF_MAX];
142 #ifdef CONFIG_DEBUG_LOCK_ALLOC
144 * Make lock validator output more readable. (we pre-construct these
145 * strings build-time, so that runtime initialization of socket
148 static const char *af_family_key_strings[AF_MAX+1] = {
149 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
150 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
151 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
152 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
153 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
154 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
155 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
156 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
157 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
158 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
159 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
171 "slock-27" , "slock-28" , "slock-29" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX"
177 * sk_callback_lock locking rules are per-address-family,
178 * so split the lock classes by using a per-AF key:
180 static struct lock_class_key af_callback_keys[AF_MAX];
182 /* Take into consideration the size of the struct sk_buff overhead in the
183 * determination of these values, since that is non-constant across
184 * platforms. This makes socket queueing behavior and performance
185 * not depend upon such differences.
187 #define _SK_MEM_PACKETS 256
188 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
189 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
190 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
192 /* Run time adjustable parameters. */
193 __u32 sysctl_wmem_max = SK_WMEM_MAX;
194 __u32 sysctl_rmem_max = SK_RMEM_MAX;
195 __u32 sysctl_wmem_default = SK_WMEM_MAX;
196 __u32 sysctl_rmem_default = SK_RMEM_MAX;
198 /* Maximal space eaten by iovec or ancilliary data plus some space */
199 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
201 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
205 if (optlen < sizeof(tv))
207 if (copy_from_user(&tv, optval, sizeof(tv)))
210 *timeo_p = MAX_SCHEDULE_TIMEOUT;
211 if (tv.tv_sec == 0 && tv.tv_usec == 0)
213 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
214 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
218 static void sock_warn_obsolete_bsdism(const char *name)
221 static char warncomm[TASK_COMM_LEN];
222 if (strcmp(warncomm, current->comm) && warned < 5) {
223 strcpy(warncomm, current->comm);
224 printk(KERN_WARNING "process `%s' is using obsolete "
225 "%s SO_BSDCOMPAT\n", warncomm, name);
230 static void sock_disable_timestamp(struct sock *sk)
232 if (sock_flag(sk, SOCK_TIMESTAMP)) {
233 sock_reset_flag(sk, SOCK_TIMESTAMP);
234 net_disable_timestamp();
239 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
244 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
245 /* Silently drop if VNET is active (if INET bind() has been
246 * overridden) and the context is not entitled to read the
250 (int) sk->sk_xid > 0 && sk->sk_xid != skb->xid) {
256 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
257 number of warnings when compiling with -W --ANK
259 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
260 (unsigned)sk->sk_rcvbuf) {
265 /* It would be deadlock, if sock_queue_rcv_skb is used
266 with socket lock! We assume that users of this
267 function are lock free.
269 err = sk_filter(sk, skb, 1);
274 skb_set_owner_r(skb, sk);
276 /* Cache the SKB length before we tack it onto the receive
277 * queue. Once it is added it no longer belongs to us and
278 * may be freed by other threads of control pulling packets
283 skb_queue_tail(&sk->sk_receive_queue, skb);
285 if (!sock_flag(sk, SOCK_DEAD))
286 sk->sk_data_ready(sk, skb_len);
290 EXPORT_SYMBOL(sock_queue_rcv_skb);
292 int sk_receive_skb(struct sock *sk, struct sk_buff *skb)
294 int rc = NET_RX_SUCCESS;
296 if (sk_filter(sk, skb, 0))
297 goto discard_and_relse;
302 if (!sock_owned_by_user(sk)) {
304 * trylock + unlock semantics:
306 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
308 rc = sk->sk_backlog_rcv(sk, skb);
310 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
312 sk_add_backlog(sk, skb);
321 EXPORT_SYMBOL(sk_receive_skb);
323 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
325 struct dst_entry *dst = sk->sk_dst_cache;
327 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
328 sk->sk_dst_cache = NULL;
335 EXPORT_SYMBOL(__sk_dst_check);
337 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
339 struct dst_entry *dst = sk_dst_get(sk);
341 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
349 EXPORT_SYMBOL(sk_dst_check);
352 * This is meant for all protocols to use and covers goings on
353 * at the socket level. Everything here is generic.
356 int sock_setsockopt(struct socket *sock, int level, int optname,
357 char __user *optval, int optlen)
359 struct sock *sk=sock->sk;
360 struct sk_filter *filter;
367 * Options without arguments
370 #ifdef SO_DONTLINGER /* Compatibility item... */
371 if (optname == SO_DONTLINGER) {
373 sock_reset_flag(sk, SOCK_LINGER);
379 if(optlen<sizeof(int))
382 if (get_user(val, (int __user *)optval))
392 if(val && !capable(CAP_NET_ADMIN))
397 sock_set_flag(sk, SOCK_DBG);
399 sock_reset_flag(sk, SOCK_DBG);
402 sk->sk_reuse = valbool;
410 sock_set_flag(sk, SOCK_LOCALROUTE);
412 sock_reset_flag(sk, SOCK_LOCALROUTE);
415 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
418 /* Don't error on this BSD doesn't and if you think
419 about it this is right. Otherwise apps have to
420 play 'guess the biggest size' games. RCVBUF/SNDBUF
421 are treated in BSD as hints */
423 if (val > sysctl_wmem_max)
424 val = sysctl_wmem_max;
426 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
427 if ((val * 2) < SOCK_MIN_SNDBUF)
428 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
430 sk->sk_sndbuf = val * 2;
433 * Wake up sending tasks if we
436 sk->sk_write_space(sk);
440 if (!capable(CAP_NET_ADMIN)) {
447 /* Don't error on this BSD doesn't and if you think
448 about it this is right. Otherwise apps have to
449 play 'guess the biggest size' games. RCVBUF/SNDBUF
450 are treated in BSD as hints */
452 if (val > sysctl_rmem_max)
453 val = sysctl_rmem_max;
455 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
457 * We double it on the way in to account for
458 * "struct sk_buff" etc. overhead. Applications
459 * assume that the SO_RCVBUF setting they make will
460 * allow that much actual data to be received on that
463 * Applications are unaware that "struct sk_buff" and
464 * other overheads allocate from the receive buffer
465 * during socket buffer allocation.
467 * And after considering the possible alternatives,
468 * returning the value we actually used in getsockopt
469 * is the most desirable behavior.
471 if ((val * 2) < SOCK_MIN_RCVBUF)
472 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
474 sk->sk_rcvbuf = val * 2;
478 if (!capable(CAP_NET_ADMIN)) {
486 if (sk->sk_protocol == IPPROTO_TCP)
487 tcp_set_keepalive(sk, valbool);
489 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
493 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
497 sk->sk_no_check = valbool;
501 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
502 sk->sk_priority = val;
508 if(optlen<sizeof(ling)) {
509 ret = -EINVAL; /* 1003.1g */
512 if (copy_from_user(&ling,optval,sizeof(ling))) {
517 sock_reset_flag(sk, SOCK_LINGER);
519 #if (BITS_PER_LONG == 32)
520 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
521 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
524 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
525 sock_set_flag(sk, SOCK_LINGER);
530 sock_warn_obsolete_bsdism("setsockopt");
535 set_bit(SOCK_PASSCRED, &sock->flags);
537 clear_bit(SOCK_PASSCRED, &sock->flags);
540 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
546 if (val < 0 || val > MAX_S_CONTEXT) {
556 sock_set_flag(sk, SOCK_RCVTSTAMP);
557 sock_enable_timestamp(sk);
559 sock_reset_flag(sk, SOCK_RCVTSTAMP);
565 sk->sk_rcvlowat = val ? : 1;
569 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
573 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
576 #ifdef CONFIG_NETDEVICES
577 case SO_BINDTODEVICE:
579 char devname[IFNAMSIZ];
582 if (!capable(CAP_NET_RAW)) {
587 /* Bind this socket to a particular device like "eth0",
588 * as specified in the passed interface name. If the
589 * name is "" or the option length is zero the socket
594 sk->sk_bound_dev_if = 0;
596 if (optlen > IFNAMSIZ - 1)
597 optlen = IFNAMSIZ - 1;
598 memset(devname, 0, sizeof(devname));
599 if (copy_from_user(devname, optval, optlen)) {
604 /* Remove any cached route for this socket. */
607 if (devname[0] == '\0') {
608 sk->sk_bound_dev_if = 0;
610 struct net_device *dev = dev_get_by_name(devname);
615 sk->sk_bound_dev_if = dev->ifindex;
624 case SO_ATTACH_FILTER:
626 if (optlen == sizeof(struct sock_fprog)) {
627 struct sock_fprog fprog;
630 if (copy_from_user(&fprog, optval, sizeof(fprog)))
633 ret = sk_attach_filter(&fprog, sk);
637 case SO_DETACH_FILTER:
638 spin_lock_bh(&sk->sk_lock.slock);
639 filter = sk->sk_filter;
641 sk->sk_filter = NULL;
642 spin_unlock_bh(&sk->sk_lock.slock);
643 sk_filter_release(sk, filter);
646 spin_unlock_bh(&sk->sk_lock.slock);
652 set_bit(SOCK_PASSSEC, &sock->flags);
654 clear_bit(SOCK_PASSSEC, &sock->flags);
657 /* We implement the SO_SNDLOWAT etc to
658 not be settable (1003.1g 5.3) */
668 int sock_getsockopt(struct socket *sock, int level, int optname,
669 char __user *optval, int __user *optlen)
671 struct sock *sk = sock->sk;
680 unsigned int lv = sizeof(int);
683 if(get_user(len,optlen))
691 v.val = sock_flag(sk, SOCK_DBG);
695 v.val = sock_flag(sk, SOCK_LOCALROUTE);
699 v.val = !!sock_flag(sk, SOCK_BROADCAST);
703 v.val = sk->sk_sndbuf;
707 v.val = sk->sk_rcvbuf;
711 v.val = sk->sk_reuse;
715 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
723 v.val = -sock_error(sk);
725 v.val = xchg(&sk->sk_err_soft, 0);
729 v.val = !!sock_flag(sk, SOCK_URGINLINE);
733 v.val = sk->sk_no_check;
737 v.val = sk->sk_priority;
742 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
743 v.ling.l_linger = sk->sk_lingertime / HZ;
747 sock_warn_obsolete_bsdism("getsockopt");
751 v.val = sock_flag(sk, SOCK_RCVTSTAMP);
755 lv=sizeof(struct timeval);
756 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
760 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
761 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
766 lv=sizeof(struct timeval);
767 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
771 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
772 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
777 v.val = sk->sk_rcvlowat;
785 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
789 if (len > sizeof(sk->sk_peercred))
790 len = sizeof(sk->sk_peercred);
791 if (copy_to_user(optval, &sk->sk_peercred, len))
799 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
803 if (copy_to_user(optval, address, len))
808 /* Dubious BSD thing... Probably nobody even uses it, but
809 * the UNIX standard wants it for whatever reason... -DaveM
812 v.val = sk->sk_state == TCP_LISTEN;
816 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
820 return security_socket_getpeersec_stream(sock, optval, optlen, len);
823 return(-ENOPROTOOPT);
827 if (copy_to_user(optval, &v, len))
830 if (put_user(len, optlen))
836 * Initialize an sk_lock.
838 * (We also register the sk_lock with the lock validator.)
840 static void inline sock_lock_init(struct sock *sk)
842 sock_lock_init_class_and_name(sk,
843 af_family_slock_key_strings[sk->sk_family],
844 af_family_slock_keys + sk->sk_family,
845 af_family_key_strings[sk->sk_family],
846 af_family_keys + sk->sk_family);
850 * sk_alloc - All socket objects are allocated here
851 * @family: protocol family
852 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
853 * @prot: struct proto associated with this new sock instance
854 * @zero_it: if we should zero the newly allocated sock
856 struct sock *sk_alloc(int family, gfp_t priority,
857 struct proto *prot, int zero_it)
859 struct sock *sk = NULL;
860 kmem_cache_t *slab = prot->slab;
863 sk = kmem_cache_alloc(slab, priority);
865 sk = kmalloc(prot->obj_size, priority);
869 memset(sk, 0, prot->obj_size);
870 sk->sk_family = family;
872 * See comment in struct sock definition to understand
873 * why we need sk_prot_creator -acme
875 sk->sk_prot = sk->sk_prot_creator = prot;
881 if (security_sk_alloc(sk, family, priority))
884 if (!try_module_get(prot->owner))
891 kmem_cache_free(slab, sk);
897 void sk_free(struct sock *sk)
899 struct sk_filter *filter;
900 struct module *owner = sk->sk_prot_creator->owner;
905 filter = sk->sk_filter;
907 sk_filter_release(sk, filter);
908 sk->sk_filter = NULL;
911 sock_disable_timestamp(sk);
913 if (atomic_read(&sk->sk_omem_alloc))
914 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
915 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
917 security_sk_free(sk);
919 clr_vx_info(&sk->sk_vx_info);
921 clr_nx_info(&sk->sk_nx_info);
923 if (sk->sk_prot_creator->slab != NULL)
924 kmem_cache_free(sk->sk_prot_creator->slab, sk);
930 struct sock *sk_clone(struct sock *sk, const gfp_t priority)
932 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
935 struct sk_filter *filter;
937 memcpy(newsk, sk, sk->sk_prot->obj_size);
942 sk_node_init(&newsk->sk_node);
943 sock_lock_init(newsk);
946 atomic_set(&newsk->sk_rmem_alloc, 0);
947 atomic_set(&newsk->sk_wmem_alloc, 0);
948 atomic_set(&newsk->sk_omem_alloc, 0);
949 skb_queue_head_init(&newsk->sk_receive_queue);
950 skb_queue_head_init(&newsk->sk_write_queue);
951 #ifdef CONFIG_NET_DMA
952 skb_queue_head_init(&newsk->sk_async_wait_queue);
955 rwlock_init(&newsk->sk_dst_lock);
956 rwlock_init(&newsk->sk_callback_lock);
957 lockdep_set_class(&newsk->sk_callback_lock,
958 af_callback_keys + newsk->sk_family);
960 newsk->sk_dst_cache = NULL;
961 newsk->sk_wmem_queued = 0;
962 newsk->sk_forward_alloc = 0;
963 newsk->sk_send_head = NULL;
964 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
965 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
967 sock_reset_flag(newsk, SOCK_DONE);
968 skb_queue_head_init(&newsk->sk_error_queue);
970 filter = newsk->sk_filter;
972 sk_filter_charge(newsk, filter);
974 if (sk->sk_create_child)
975 sk->sk_create_child(sk, newsk);
977 if (unlikely(xfrm_sk_clone_policy(newsk))) {
978 /* It is still raw copy of parent, so invalidate
979 * destructor and make plain sk_free() */
980 newsk->sk_destruct = NULL;
987 newsk->sk_priority = 0;
988 atomic_set(&newsk->sk_refcnt, 2);
990 set_vx_info(&newsk->sk_vx_info, sk->sk_vx_info);
991 newsk->sk_xid = sk->sk_xid;
993 set_nx_info(&newsk->sk_nx_info, sk->sk_nx_info);
994 newsk->sk_nid = sk->sk_nid;
997 * Increment the counter in the same struct proto as the master
998 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
999 * is the same as sk->sk_prot->socks, as this field was copied
1002 * This _changes_ the previous behaviour, where
1003 * tcp_create_openreq_child always was incrementing the
1004 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1005 * to be taken into account in all callers. -acme
1007 sk_refcnt_debug_inc(newsk);
1008 newsk->sk_socket = NULL;
1009 newsk->sk_sleep = NULL;
1011 if (newsk->sk_prot->sockets_allocated)
1012 atomic_inc(newsk->sk_prot->sockets_allocated);
1018 EXPORT_SYMBOL_GPL(sk_clone);
1020 void __init sk_init(void)
1022 if (num_physpages <= 4096) {
1023 sysctl_wmem_max = 32767;
1024 sysctl_rmem_max = 32767;
1025 sysctl_wmem_default = 32767;
1026 sysctl_rmem_default = 32767;
1027 } else if (num_physpages >= 131072) {
1028 sysctl_wmem_max = 131071;
1029 sysctl_rmem_max = 131071;
1034 * Simple resource managers for sockets.
1039 * Write buffer destructor automatically called from kfree_skb.
1041 void sock_wfree(struct sk_buff *skb)
1043 struct sock *sk = skb->sk;
1045 /* In case it might be waiting for more memory. */
1046 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1047 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1048 sk->sk_write_space(sk);
1053 * Read buffer destructor automatically called from kfree_skb.
1055 void sock_rfree(struct sk_buff *skb)
1057 struct sock *sk = skb->sk;
1059 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1063 int sock_i_uid(struct sock *sk)
1067 read_lock(&sk->sk_callback_lock);
1068 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1069 read_unlock(&sk->sk_callback_lock);
1073 unsigned long sock_i_ino(struct sock *sk)
1077 read_lock(&sk->sk_callback_lock);
1078 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1079 read_unlock(&sk->sk_callback_lock);
1084 * Allocate a skb from the socket's send buffer.
1086 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1089 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1090 struct sk_buff * skb = alloc_skb(size, priority);
1092 skb_set_owner_w(skb, sk);
1100 * Allocate a skb from the socket's receive buffer.
1102 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1105 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1106 struct sk_buff *skb = alloc_skb(size, priority);
1108 skb_set_owner_r(skb, sk);
1116 * Allocate a memory block from the socket's option memory buffer.
1118 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1120 if ((unsigned)size <= sysctl_optmem_max &&
1121 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1123 /* First do the add, to avoid the race if kmalloc
1126 atomic_add(size, &sk->sk_omem_alloc);
1127 mem = kmalloc(size, priority);
1130 atomic_sub(size, &sk->sk_omem_alloc);
1136 * Free an option memory block.
1138 void sock_kfree_s(struct sock *sk, void *mem, int size)
1141 atomic_sub(size, &sk->sk_omem_alloc);
1144 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1145 I think, these locks should be removed for datagram sockets.
1147 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1151 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1155 if (signal_pending(current))
1157 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1158 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1159 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1161 if (sk->sk_shutdown & SEND_SHUTDOWN)
1165 timeo = schedule_timeout(timeo);
1167 finish_wait(sk->sk_sleep, &wait);
1173 * Generic send/receive buffer handlers
1176 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1177 unsigned long header_len,
1178 unsigned long data_len,
1179 int noblock, int *errcode)
1181 struct sk_buff *skb;
1186 gfp_mask = sk->sk_allocation;
1187 if (gfp_mask & __GFP_WAIT)
1188 gfp_mask |= __GFP_REPEAT;
1190 timeo = sock_sndtimeo(sk, noblock);
1192 err = sock_error(sk);
1197 if (sk->sk_shutdown & SEND_SHUTDOWN)
1200 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1201 skb = alloc_skb(header_len, gfp_mask);
1206 /* No pages, we're done... */
1210 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1211 skb->truesize += data_len;
1212 skb_shinfo(skb)->nr_frags = npages;
1213 for (i = 0; i < npages; i++) {
1217 page = alloc_pages(sk->sk_allocation, 0);
1220 skb_shinfo(skb)->nr_frags = i;
1225 frag = &skb_shinfo(skb)->frags[i];
1227 frag->page_offset = 0;
1228 frag->size = (data_len >= PAGE_SIZE ?
1231 data_len -= PAGE_SIZE;
1234 /* Full success... */
1240 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1241 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1245 if (signal_pending(current))
1247 timeo = sock_wait_for_wmem(sk, timeo);
1250 skb_set_owner_w(skb, sk);
1254 err = sock_intr_errno(timeo);
1260 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1261 int noblock, int *errcode)
1263 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1266 static void __lock_sock(struct sock *sk)
1271 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1272 TASK_UNINTERRUPTIBLE);
1273 spin_unlock_bh(&sk->sk_lock.slock);
1275 spin_lock_bh(&sk->sk_lock.slock);
1276 if(!sock_owned_by_user(sk))
1279 finish_wait(&sk->sk_lock.wq, &wait);
1282 static void __release_sock(struct sock *sk)
1284 struct sk_buff *skb = sk->sk_backlog.head;
1287 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1291 struct sk_buff *next = skb->next;
1294 sk->sk_backlog_rcv(sk, skb);
1297 * We are in process context here with softirqs
1298 * disabled, use cond_resched_softirq() to preempt.
1299 * This is safe to do because we've taken the backlog
1302 cond_resched_softirq();
1305 } while (skb != NULL);
1308 } while((skb = sk->sk_backlog.head) != NULL);
1312 * sk_wait_data - wait for data to arrive at sk_receive_queue
1313 * @sk: sock to wait on
1314 * @timeo: for how long
1316 * Now socket state including sk->sk_err is changed only under lock,
1317 * hence we may omit checks after joining wait queue.
1318 * We check receive queue before schedule() only as optimization;
1319 * it is very likely that release_sock() added new data.
1321 int sk_wait_data(struct sock *sk, long *timeo)
1326 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1327 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1328 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1329 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1330 finish_wait(sk->sk_sleep, &wait);
1334 EXPORT_SYMBOL(sk_wait_data);
1337 * Set of default routines for initialising struct proto_ops when
1338 * the protocol does not support a particular function. In certain
1339 * cases where it makes no sense for a protocol to have a "do nothing"
1340 * function, some default processing is provided.
1343 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1348 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1354 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1359 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1364 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1370 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1375 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1380 int sock_no_listen(struct socket *sock, int backlog)
1385 int sock_no_shutdown(struct socket *sock, int how)
1390 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1391 char __user *optval, int optlen)
1396 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1397 char __user *optval, int __user *optlen)
1402 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1408 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1409 size_t len, int flags)
1414 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1416 /* Mirror missing mmap method error code */
1420 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1423 struct msghdr msg = {.msg_flags = flags};
1425 char *kaddr = kmap(page);
1426 iov.iov_base = kaddr + offset;
1428 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1434 * Default Socket Callbacks
1437 static void sock_def_wakeup(struct sock *sk)
1439 read_lock(&sk->sk_callback_lock);
1440 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1441 wake_up_interruptible_all(sk->sk_sleep);
1442 read_unlock(&sk->sk_callback_lock);
1445 static void sock_def_error_report(struct sock *sk)
1447 read_lock(&sk->sk_callback_lock);
1448 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1449 wake_up_interruptible(sk->sk_sleep);
1450 sk_wake_async(sk,0,POLL_ERR);
1451 read_unlock(&sk->sk_callback_lock);
1454 static void sock_def_readable(struct sock *sk, int len)
1456 read_lock(&sk->sk_callback_lock);
1457 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1458 wake_up_interruptible(sk->sk_sleep);
1459 sk_wake_async(sk,1,POLL_IN);
1460 read_unlock(&sk->sk_callback_lock);
1463 static void sock_def_write_space(struct sock *sk)
1465 read_lock(&sk->sk_callback_lock);
1467 /* Do not wake up a writer until he can make "significant"
1470 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1471 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1472 wake_up_interruptible(sk->sk_sleep);
1474 /* Should agree with poll, otherwise some programs break */
1475 if (sock_writeable(sk))
1476 sk_wake_async(sk, 2, POLL_OUT);
1479 read_unlock(&sk->sk_callback_lock);
1482 static void sock_def_destruct(struct sock *sk)
1484 kfree(sk->sk_protinfo);
1487 void sk_send_sigurg(struct sock *sk)
1489 if (sk->sk_socket && sk->sk_socket->file)
1490 if (send_sigurg(&sk->sk_socket->file->f_owner))
1491 sk_wake_async(sk, 3, POLL_PRI);
1494 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1495 unsigned long expires)
1497 if (!mod_timer(timer, expires))
1501 EXPORT_SYMBOL(sk_reset_timer);
1503 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1505 if (timer_pending(timer) && del_timer(timer))
1509 EXPORT_SYMBOL(sk_stop_timer);
1511 void sock_init_data(struct socket *sock, struct sock *sk)
1513 skb_queue_head_init(&sk->sk_receive_queue);
1514 skb_queue_head_init(&sk->sk_write_queue);
1515 skb_queue_head_init(&sk->sk_error_queue);
1516 #ifdef CONFIG_NET_DMA
1517 skb_queue_head_init(&sk->sk_async_wait_queue);
1520 sk->sk_send_head = NULL;
1522 init_timer(&sk->sk_timer);
1524 sk->sk_allocation = GFP_KERNEL;
1525 sk->sk_rcvbuf = sysctl_rmem_default;
1526 sk->sk_sndbuf = sysctl_wmem_default;
1527 sk->sk_state = TCP_CLOSE;
1528 sk->sk_socket = sock;
1530 sock_set_flag(sk, SOCK_ZAPPED);
1534 sk->sk_type = sock->type;
1535 sk->sk_sleep = &sock->wait;
1538 sk->sk_sleep = NULL;
1540 rwlock_init(&sk->sk_dst_lock);
1541 rwlock_init(&sk->sk_callback_lock);
1542 lockdep_set_class(&sk->sk_callback_lock,
1543 af_callback_keys + sk->sk_family);
1545 sk->sk_state_change = sock_def_wakeup;
1546 sk->sk_data_ready = sock_def_readable;
1547 sk->sk_write_space = sock_def_write_space;
1548 sk->sk_error_report = sock_def_error_report;
1549 sk->sk_destruct = sock_def_destruct;
1551 sk->sk_sndmsg_page = NULL;
1552 sk->sk_sndmsg_off = 0;
1554 sk->sk_peercred.pid = 0;
1555 sk->sk_peercred.uid = -1;
1556 sk->sk_peercred.gid = -1;
1557 sk->sk_write_pending = 0;
1558 sk->sk_rcvlowat = 1;
1559 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1560 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1562 sk->sk_stamp.tv_sec = -1L;
1563 sk->sk_stamp.tv_usec = -1L;
1565 set_vx_info(&sk->sk_vx_info, current->vx_info);
1566 sk->sk_xid = vx_current_xid();
1568 set_nx_info(&sk->sk_nx_info, current->nx_info);
1569 sk->sk_nid = nx_current_nid();
1570 atomic_set(&sk->sk_refcnt, 1);
1573 void fastcall lock_sock(struct sock *sk)
1576 spin_lock_bh(&sk->sk_lock.slock);
1577 if (sk->sk_lock.owner)
1579 sk->sk_lock.owner = (void *)1;
1580 spin_unlock(&sk->sk_lock.slock);
1582 * The sk_lock has mutex_lock() semantics here:
1584 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1588 EXPORT_SYMBOL(lock_sock);
1590 void fastcall release_sock(struct sock *sk)
1593 * The sk_lock has mutex_unlock() semantics:
1595 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1597 spin_lock_bh(&sk->sk_lock.slock);
1598 if (sk->sk_backlog.tail)
1600 sk->sk_lock.owner = NULL;
1601 if (waitqueue_active(&sk->sk_lock.wq))
1602 wake_up(&sk->sk_lock.wq);
1603 spin_unlock_bh(&sk->sk_lock.slock);
1605 EXPORT_SYMBOL(release_sock);
1607 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1609 if (!sock_flag(sk, SOCK_TIMESTAMP))
1610 sock_enable_timestamp(sk);
1611 if (sk->sk_stamp.tv_sec == -1)
1613 if (sk->sk_stamp.tv_sec == 0)
1614 do_gettimeofday(&sk->sk_stamp);
1615 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1618 EXPORT_SYMBOL(sock_get_timestamp);
1620 void sock_enable_timestamp(struct sock *sk)
1622 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1623 sock_set_flag(sk, SOCK_TIMESTAMP);
1624 net_enable_timestamp();
1627 EXPORT_SYMBOL(sock_enable_timestamp);
1630 * Get a socket option on an socket.
1632 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1633 * asynchronous errors should be reported by getsockopt. We assume
1634 * this means if you specify SO_ERROR (otherwise whats the point of it).
1636 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1637 char __user *optval, int __user *optlen)
1639 struct sock *sk = sock->sk;
1641 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1644 EXPORT_SYMBOL(sock_common_getsockopt);
1646 #ifdef CONFIG_COMPAT
1647 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1648 char __user *optval, int __user *optlen)
1650 struct sock *sk = sock->sk;
1652 if (sk->sk_prot->compat_setsockopt != NULL)
1653 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1655 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1657 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1660 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1661 struct msghdr *msg, size_t size, int flags)
1663 struct sock *sk = sock->sk;
1667 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1668 flags & ~MSG_DONTWAIT, &addr_len);
1670 msg->msg_namelen = addr_len;
1674 EXPORT_SYMBOL(sock_common_recvmsg);
1677 * Set socket options on an inet socket.
1679 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1680 char __user *optval, int optlen)
1682 struct sock *sk = sock->sk;
1684 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1687 EXPORT_SYMBOL(sock_common_setsockopt);
1689 #ifdef CONFIG_COMPAT
1690 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1691 char __user *optval, int optlen)
1693 struct sock *sk = sock->sk;
1695 if (sk->sk_prot->compat_setsockopt != NULL)
1696 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1698 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1700 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1703 void sk_common_release(struct sock *sk)
1705 if (sk->sk_prot->destroy)
1706 sk->sk_prot->destroy(sk);
1709 * Observation: when sock_common_release is called, processes have
1710 * no access to socket. But net still has.
1711 * Step one, detach it from networking:
1713 * A. Remove from hash tables.
1716 sk->sk_prot->unhash(sk);
1719 * In this point socket cannot receive new packets, but it is possible
1720 * that some packets are in flight because some CPU runs receiver and
1721 * did hash table lookup before we unhashed socket. They will achieve
1722 * receive queue and will be purged by socket destructor.
1724 * Also we still have packets pending on receive queue and probably,
1725 * our own packets waiting in device queues. sock_destroy will drain
1726 * receive queue, but transmitted packets will delay socket destruction
1727 * until the last reference will be released.
1732 xfrm_sk_free_policy(sk);
1734 sk_refcnt_debug_release(sk);
1738 EXPORT_SYMBOL(sk_common_release);
1740 static DEFINE_RWLOCK(proto_list_lock);
1741 static LIST_HEAD(proto_list);
1743 int proto_register(struct proto *prot, int alloc_slab)
1745 char *request_sock_slab_name = NULL;
1746 char *timewait_sock_slab_name;
1750 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1751 SLAB_HWCACHE_ALIGN, NULL, NULL);
1753 if (prot->slab == NULL) {
1754 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1759 if (prot->rsk_prot != NULL) {
1760 static const char mask[] = "request_sock_%s";
1762 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1763 if (request_sock_slab_name == NULL)
1764 goto out_free_sock_slab;
1766 sprintf(request_sock_slab_name, mask, prot->name);
1767 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1768 prot->rsk_prot->obj_size, 0,
1769 SLAB_HWCACHE_ALIGN, NULL, NULL);
1771 if (prot->rsk_prot->slab == NULL) {
1772 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1774 goto out_free_request_sock_slab_name;
1778 if (prot->twsk_prot != NULL) {
1779 static const char mask[] = "tw_sock_%s";
1781 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1783 if (timewait_sock_slab_name == NULL)
1784 goto out_free_request_sock_slab;
1786 sprintf(timewait_sock_slab_name, mask, prot->name);
1787 prot->twsk_prot->twsk_slab =
1788 kmem_cache_create(timewait_sock_slab_name,
1789 prot->twsk_prot->twsk_obj_size,
1790 0, SLAB_HWCACHE_ALIGN,
1792 if (prot->twsk_prot->twsk_slab == NULL)
1793 goto out_free_timewait_sock_slab_name;
1797 write_lock(&proto_list_lock);
1798 list_add(&prot->node, &proto_list);
1799 write_unlock(&proto_list_lock);
1803 out_free_timewait_sock_slab_name:
1804 kfree(timewait_sock_slab_name);
1805 out_free_request_sock_slab:
1806 if (prot->rsk_prot && prot->rsk_prot->slab) {
1807 kmem_cache_destroy(prot->rsk_prot->slab);
1808 prot->rsk_prot->slab = NULL;
1810 out_free_request_sock_slab_name:
1811 kfree(request_sock_slab_name);
1813 kmem_cache_destroy(prot->slab);
1818 EXPORT_SYMBOL(proto_register);
1820 void proto_unregister(struct proto *prot)
1822 write_lock(&proto_list_lock);
1823 list_del(&prot->node);
1824 write_unlock(&proto_list_lock);
1826 if (prot->slab != NULL) {
1827 kmem_cache_destroy(prot->slab);
1831 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1832 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1834 kmem_cache_destroy(prot->rsk_prot->slab);
1836 prot->rsk_prot->slab = NULL;
1839 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1840 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1842 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1844 prot->twsk_prot->twsk_slab = NULL;
1848 EXPORT_SYMBOL(proto_unregister);
1850 #ifdef CONFIG_PROC_FS
1851 static inline struct proto *__proto_head(void)
1853 return list_entry(proto_list.next, struct proto, node);
1856 static inline struct proto *proto_head(void)
1858 return list_empty(&proto_list) ? NULL : __proto_head();
1861 static inline struct proto *proto_next(struct proto *proto)
1863 return proto->node.next == &proto_list ? NULL :
1864 list_entry(proto->node.next, struct proto, node);
1867 static inline struct proto *proto_get_idx(loff_t pos)
1869 struct proto *proto;
1872 list_for_each_entry(proto, &proto_list, node)
1881 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1883 read_lock(&proto_list_lock);
1884 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1887 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1890 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1893 static void proto_seq_stop(struct seq_file *seq, void *v)
1895 read_unlock(&proto_list_lock);
1898 static char proto_method_implemented(const void *method)
1900 return method == NULL ? 'n' : 'y';
1903 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1905 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1906 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1909 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1910 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1911 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1913 proto->slab == NULL ? "no" : "yes",
1914 module_name(proto->owner),
1915 proto_method_implemented(proto->close),
1916 proto_method_implemented(proto->connect),
1917 proto_method_implemented(proto->disconnect),
1918 proto_method_implemented(proto->accept),
1919 proto_method_implemented(proto->ioctl),
1920 proto_method_implemented(proto->init),
1921 proto_method_implemented(proto->destroy),
1922 proto_method_implemented(proto->shutdown),
1923 proto_method_implemented(proto->setsockopt),
1924 proto_method_implemented(proto->getsockopt),
1925 proto_method_implemented(proto->sendmsg),
1926 proto_method_implemented(proto->recvmsg),
1927 proto_method_implemented(proto->sendpage),
1928 proto_method_implemented(proto->bind),
1929 proto_method_implemented(proto->backlog_rcv),
1930 proto_method_implemented(proto->hash),
1931 proto_method_implemented(proto->unhash),
1932 proto_method_implemented(proto->get_port),
1933 proto_method_implemented(proto->enter_memory_pressure));
1936 static int proto_seq_show(struct seq_file *seq, void *v)
1938 if (v == SEQ_START_TOKEN)
1939 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1948 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1950 proto_seq_printf(seq, v);
1954 static struct seq_operations proto_seq_ops = {
1955 .start = proto_seq_start,
1956 .next = proto_seq_next,
1957 .stop = proto_seq_stop,
1958 .show = proto_seq_show,
1961 static int proto_seq_open(struct inode *inode, struct file *file)
1963 return seq_open(file, &proto_seq_ops);
1966 static struct file_operations proto_seq_fops = {
1967 .owner = THIS_MODULE,
1968 .open = proto_seq_open,
1970 .llseek = seq_lseek,
1971 .release = seq_release,
1974 static int __init proto_init(void)
1976 /* register /proc/net/protocols */
1977 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1980 subsys_initcall(proto_init);
1982 #endif /* PROC_FS */
1984 EXPORT_SYMBOL(sk_alloc);
1985 EXPORT_SYMBOL(sk_free);
1986 EXPORT_SYMBOL(sk_send_sigurg);
1987 EXPORT_SYMBOL(sock_alloc_send_skb);
1988 EXPORT_SYMBOL(sock_init_data);
1989 EXPORT_SYMBOL(sock_kfree_s);
1990 EXPORT_SYMBOL(sock_kmalloc);
1991 EXPORT_SYMBOL(sock_no_accept);
1992 EXPORT_SYMBOL(sock_no_bind);
1993 EXPORT_SYMBOL(sock_no_connect);
1994 EXPORT_SYMBOL(sock_no_getname);
1995 EXPORT_SYMBOL(sock_no_getsockopt);
1996 EXPORT_SYMBOL(sock_no_ioctl);
1997 EXPORT_SYMBOL(sock_no_listen);
1998 EXPORT_SYMBOL(sock_no_mmap);
1999 EXPORT_SYMBOL(sock_no_poll);
2000 EXPORT_SYMBOL(sock_no_recvmsg);
2001 EXPORT_SYMBOL(sock_no_sendmsg);
2002 EXPORT_SYMBOL(sock_no_sendpage);
2003 EXPORT_SYMBOL(sock_no_setsockopt);
2004 EXPORT_SYMBOL(sock_no_shutdown);
2005 EXPORT_SYMBOL(sock_no_socketpair);
2006 EXPORT_SYMBOL(sock_rfree);
2007 EXPORT_SYMBOL(sock_setsockopt);
2008 EXPORT_SYMBOL(sock_wfree);
2009 EXPORT_SYMBOL(sock_wmalloc);
2010 EXPORT_SYMBOL(sock_i_uid);
2011 EXPORT_SYMBOL(sock_i_ino);
2012 EXPORT_SYMBOL(sysctl_optmem_max);
2013 #ifdef CONFIG_SYSCTL
2014 EXPORT_SYMBOL(sysctl_rmem_max);
2015 EXPORT_SYMBOL(sysctl_wmem_max);