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.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
114 #include <asm/uaccess.h>
115 #include <asm/system.h>
117 #include <linux/netdevice.h>
118 #include <net/protocol.h>
119 #include <linux/skbuff.h>
120 #include <net/net_namespace.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];
140 * Make lock validator output more readable. (we pre-construct these
141 * strings build-time, so that runtime initialization of socket
144 static const char *af_family_key_strings[AF_MAX+1] = {
145 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
146 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
147 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
148 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
149 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
150 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
151 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
152 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
153 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
154 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
155 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
156 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
159 static const char *af_family_slock_key_strings[AF_MAX+1] = {
160 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
161 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
162 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
163 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
164 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
165 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
166 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
167 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
168 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
169 "slock-27" , "slock-28" , "slock-AF_CAN" ,
170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
171 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
174 static const char *af_family_clock_key_strings[AF_MAX+1] = {
175 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
176 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
177 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
178 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
179 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
180 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
181 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
182 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
183 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
184 "clock-27" , "clock-28" , "clock-AF_CAN" ,
185 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
186 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
191 * sk_callback_lock locking rules are per-address-family,
192 * so split the lock classes by using a per-AF key:
194 static struct lock_class_key af_callback_keys[AF_MAX];
196 /* Take into consideration the size of the struct sk_buff overhead in the
197 * determination of these values, since that is non-constant across
198 * platforms. This makes socket queueing behavior and performance
199 * not depend upon such differences.
201 #define _SK_MEM_PACKETS 256
202 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
203 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
204 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
206 /* Run time adjustable parameters. */
207 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
208 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
209 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
210 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
212 /* Maximal space eaten by iovec or ancilliary data plus some space */
213 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
215 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
219 if (optlen < sizeof(tv))
221 if (copy_from_user(&tv, optval, sizeof(tv)))
223 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
227 static int warned __read_mostly;
230 if (warned < 10 && net_ratelimit()) {
232 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
233 "tries to set negative timeout\n",
234 current->comm, task_pid_nr(current));
238 *timeo_p = MAX_SCHEDULE_TIMEOUT;
239 if (tv.tv_sec == 0 && tv.tv_usec == 0)
241 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
242 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
246 static void sock_warn_obsolete_bsdism(const char *name)
249 static char warncomm[TASK_COMM_LEN];
250 if (strcmp(warncomm, current->comm) && warned < 5) {
251 strcpy(warncomm, current->comm);
252 printk(KERN_WARNING "process `%s' is using obsolete "
253 "%s SO_BSDCOMPAT\n", warncomm, name);
258 static void sock_disable_timestamp(struct sock *sk)
260 if (sock_flag(sk, SOCK_TIMESTAMP)) {
261 sock_reset_flag(sk, SOCK_TIMESTAMP);
262 net_disable_timestamp();
267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
272 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
273 number of warnings when compiling with -W --ANK
275 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
276 (unsigned)sk->sk_rcvbuf) {
281 err = sk_filter(sk, skb);
285 if (!sk_rmem_schedule(sk, skb->truesize)) {
291 skb_set_owner_r(skb, sk);
293 /* Cache the SKB length before we tack it onto the receive
294 * queue. Once it is added it no longer belongs to us and
295 * may be freed by other threads of control pulling packets
300 skb_queue_tail(&sk->sk_receive_queue, skb);
302 if (!sock_flag(sk, SOCK_DEAD))
303 sk->sk_data_ready(sk, skb_len);
307 EXPORT_SYMBOL(sock_queue_rcv_skb);
309 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
311 int rc = NET_RX_SUCCESS;
313 if (sk_filter(sk, skb))
314 goto discard_and_relse;
319 bh_lock_sock_nested(sk);
322 if (!sock_owned_by_user(sk)) {
324 * trylock + unlock semantics:
326 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
328 rc = sk_backlog_rcv(sk, skb);
330 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
332 sk_add_backlog(sk, skb);
341 EXPORT_SYMBOL(sk_receive_skb);
343 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
345 struct dst_entry *dst = sk->sk_dst_cache;
347 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
348 sk->sk_dst_cache = NULL;
355 EXPORT_SYMBOL(__sk_dst_check);
357 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
359 struct dst_entry *dst = sk_dst_get(sk);
361 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
369 EXPORT_SYMBOL(sk_dst_check);
371 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
373 int ret = -ENOPROTOOPT;
374 #ifdef CONFIG_NETDEVICES
375 struct net *net = sock_net(sk);
376 char devname[IFNAMSIZ];
381 if (!capable(CAP_NET_RAW))
388 /* Bind this socket to a particular device like "eth0",
389 * as specified in the passed interface name. If the
390 * name is "" or the option length is zero the socket
393 if (optlen > IFNAMSIZ - 1)
394 optlen = IFNAMSIZ - 1;
395 memset(devname, 0, sizeof(devname));
398 if (copy_from_user(devname, optval, optlen))
401 if (devname[0] == '\0') {
404 struct net_device *dev = dev_get_by_name(net, devname);
410 index = dev->ifindex;
415 sk->sk_bound_dev_if = index;
427 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
430 sock_set_flag(sk, bit);
432 sock_reset_flag(sk, bit);
436 * This is meant for all protocols to use and covers goings on
437 * at the socket level. Everything here is generic.
440 int sock_setsockopt(struct socket *sock, int level, int optname,
441 char __user *optval, int optlen)
443 struct sock *sk=sock->sk;
450 * Options without arguments
453 if (optname == SO_BINDTODEVICE)
454 return sock_bindtodevice(sk, optval, optlen);
456 if (optlen < sizeof(int))
459 if (get_user(val, (int __user *)optval))
468 if (val && !capable(CAP_NET_ADMIN)) {
471 sock_valbool_flag(sk, SOCK_DBG, valbool);
474 sk->sk_reuse = valbool;
481 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
484 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
487 /* Don't error on this BSD doesn't and if you think
488 about it this is right. Otherwise apps have to
489 play 'guess the biggest size' games. RCVBUF/SNDBUF
490 are treated in BSD as hints */
492 if (val > sysctl_wmem_max)
493 val = sysctl_wmem_max;
495 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
496 if ((val * 2) < SOCK_MIN_SNDBUF)
497 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
499 sk->sk_sndbuf = val * 2;
502 * Wake up sending tasks if we
505 sk->sk_write_space(sk);
509 if (!capable(CAP_NET_ADMIN)) {
516 /* Don't error on this BSD doesn't and if you think
517 about it this is right. Otherwise apps have to
518 play 'guess the biggest size' games. RCVBUF/SNDBUF
519 are treated in BSD as hints */
521 if (val > sysctl_rmem_max)
522 val = sysctl_rmem_max;
524 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
526 * We double it on the way in to account for
527 * "struct sk_buff" etc. overhead. Applications
528 * assume that the SO_RCVBUF setting they make will
529 * allow that much actual data to be received on that
532 * Applications are unaware that "struct sk_buff" and
533 * other overheads allocate from the receive buffer
534 * during socket buffer allocation.
536 * And after considering the possible alternatives,
537 * returning the value we actually used in getsockopt
538 * is the most desirable behavior.
540 if ((val * 2) < SOCK_MIN_RCVBUF)
541 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
543 sk->sk_rcvbuf = val * 2;
547 if (!capable(CAP_NET_ADMIN)) {
555 if (sk->sk_protocol == IPPROTO_TCP)
556 tcp_set_keepalive(sk, valbool);
558 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
562 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
566 sk->sk_no_check = valbool;
570 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
571 sk->sk_priority = val;
577 if (optlen < sizeof(ling)) {
578 ret = -EINVAL; /* 1003.1g */
581 if (copy_from_user(&ling,optval,sizeof(ling))) {
586 sock_reset_flag(sk, SOCK_LINGER);
588 #if (BITS_PER_LONG == 32)
589 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
590 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
593 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
594 sock_set_flag(sk, SOCK_LINGER);
599 sock_warn_obsolete_bsdism("setsockopt");
604 set_bit(SOCK_PASSCRED, &sock->flags);
606 clear_bit(SOCK_PASSCRED, &sock->flags);
612 if (optname == SO_TIMESTAMP)
613 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
615 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
616 sock_set_flag(sk, SOCK_RCVTSTAMP);
617 sock_enable_timestamp(sk);
619 sock_reset_flag(sk, SOCK_RCVTSTAMP);
620 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
627 sk->sk_rcvlowat = val ? : 1;
631 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
635 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
638 case SO_ATTACH_FILTER:
640 if (optlen == sizeof(struct sock_fprog)) {
641 struct sock_fprog fprog;
644 if (copy_from_user(&fprog, optval, sizeof(fprog)))
647 ret = sk_attach_filter(&fprog, sk);
651 case SO_DETACH_FILTER:
652 ret = sk_detach_filter(sk);
657 set_bit(SOCK_PASSSEC, &sock->flags);
659 clear_bit(SOCK_PASSSEC, &sock->flags);
662 if (!capable(CAP_NET_ADMIN))
669 /* We implement the SO_SNDLOWAT etc to
670 not be settable (1003.1g 5.3) */
680 int sock_getsockopt(struct socket *sock, int level, int optname,
681 char __user *optval, int __user *optlen)
683 struct sock *sk = sock->sk;
691 unsigned int lv = sizeof(int);
694 if (get_user(len, optlen))
701 v.val = sock_flag(sk, SOCK_DBG);
705 v.val = sock_flag(sk, SOCK_LOCALROUTE);
709 v.val = !!sock_flag(sk, SOCK_BROADCAST);
713 v.val = sk->sk_sndbuf;
717 v.val = sk->sk_rcvbuf;
721 v.val = sk->sk_reuse;
725 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
733 v.val = -sock_error(sk);
735 v.val = xchg(&sk->sk_err_soft, 0);
739 v.val = !!sock_flag(sk, SOCK_URGINLINE);
743 v.val = sk->sk_no_check;
747 v.val = sk->sk_priority;
752 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
753 v.ling.l_linger = sk->sk_lingertime / HZ;
757 sock_warn_obsolete_bsdism("getsockopt");
761 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
762 !sock_flag(sk, SOCK_RCVTSTAMPNS);
766 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
770 lv=sizeof(struct timeval);
771 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
775 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
776 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
781 lv=sizeof(struct timeval);
782 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
786 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
787 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
792 v.val = sk->sk_rcvlowat;
800 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
804 if (len > sizeof(sk->sk_peercred))
805 len = sizeof(sk->sk_peercred);
806 if (copy_to_user(optval, &sk->sk_peercred, len))
814 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
818 if (copy_to_user(optval, address, len))
823 /* Dubious BSD thing... Probably nobody even uses it, but
824 * the UNIX standard wants it for whatever reason... -DaveM
827 v.val = sk->sk_state == TCP_LISTEN;
831 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
835 return security_socket_getpeersec_stream(sock, optval, optlen, len);
847 if (copy_to_user(optval, &v, len))
850 if (put_user(len, optlen))
856 * Initialize an sk_lock.
858 * (We also register the sk_lock with the lock validator.)
860 static inline void sock_lock_init(struct sock *sk)
862 sock_lock_init_class_and_name(sk,
863 af_family_slock_key_strings[sk->sk_family],
864 af_family_slock_keys + sk->sk_family,
865 af_family_key_strings[sk->sk_family],
866 af_family_keys + sk->sk_family);
869 static void sock_copy(struct sock *nsk, const struct sock *osk)
871 #ifdef CONFIG_SECURITY_NETWORK
872 void *sptr = nsk->sk_security;
875 memcpy(nsk, osk, osk->sk_prot->obj_size);
876 #ifdef CONFIG_SECURITY_NETWORK
877 nsk->sk_security = sptr;
878 security_sk_clone(osk, nsk);
882 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
886 struct kmem_cache *slab;
890 sk = kmem_cache_alloc(slab, priority);
892 sk = kmalloc(prot->obj_size, priority);
895 if (security_sk_alloc(sk, family, priority))
898 if (!try_module_get(prot->owner))
905 security_sk_free(sk);
908 kmem_cache_free(slab, sk);
914 static void sk_prot_free(struct proto *prot, struct sock *sk)
916 struct kmem_cache *slab;
917 struct module *owner;
922 security_sk_free(sk);
924 kmem_cache_free(slab, sk);
931 * sk_alloc - All socket objects are allocated here
932 * @net: the applicable net namespace
933 * @family: protocol family
934 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
935 * @prot: struct proto associated with this new sock instance
937 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
942 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
944 sk->sk_family = family;
946 * See comment in struct sock definition to understand
947 * why we need sk_prot_creator -acme
949 sk->sk_prot = sk->sk_prot_creator = prot;
951 sock_net_set(sk, get_net(net));
957 void sk_free(struct sock *sk)
959 struct sk_filter *filter;
964 filter = rcu_dereference(sk->sk_filter);
966 sk_filter_uncharge(sk, filter);
967 rcu_assign_pointer(sk->sk_filter, NULL);
970 sock_disable_timestamp(sk);
972 if (atomic_read(&sk->sk_omem_alloc))
973 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
974 __func__, atomic_read(&sk->sk_omem_alloc));
976 put_net(sock_net(sk));
977 sk_prot_free(sk->sk_prot_creator, sk);
981 * Last sock_put should drop referrence to sk->sk_net. It has already
982 * been dropped in sk_change_net. Taking referrence to stopping namespace
984 * Take referrence to a socket to remove it from hash _alive_ and after that
985 * destroy it in the context of init_net.
987 void sk_release_kernel(struct sock *sk)
989 if (sk == NULL || sk->sk_socket == NULL)
993 sock_release(sk->sk_socket);
994 release_net(sock_net(sk));
995 sock_net_set(sk, get_net(&init_net));
998 EXPORT_SYMBOL(sk_release_kernel);
1000 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1004 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1005 if (newsk != NULL) {
1006 struct sk_filter *filter;
1008 sock_copy(newsk, sk);
1011 get_net(sock_net(newsk));
1012 sk_node_init(&newsk->sk_node);
1013 sock_lock_init(newsk);
1014 bh_lock_sock(newsk);
1015 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1017 atomic_set(&newsk->sk_rmem_alloc, 0);
1018 atomic_set(&newsk->sk_wmem_alloc, 0);
1019 atomic_set(&newsk->sk_omem_alloc, 0);
1020 skb_queue_head_init(&newsk->sk_receive_queue);
1021 skb_queue_head_init(&newsk->sk_write_queue);
1022 #ifdef CONFIG_NET_DMA
1023 skb_queue_head_init(&newsk->sk_async_wait_queue);
1026 rwlock_init(&newsk->sk_dst_lock);
1027 rwlock_init(&newsk->sk_callback_lock);
1028 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1029 af_callback_keys + newsk->sk_family,
1030 af_family_clock_key_strings[newsk->sk_family]);
1032 newsk->sk_dst_cache = NULL;
1033 newsk->sk_wmem_queued = 0;
1034 newsk->sk_forward_alloc = 0;
1035 newsk->sk_send_head = NULL;
1036 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1038 sock_reset_flag(newsk, SOCK_DONE);
1039 skb_queue_head_init(&newsk->sk_error_queue);
1041 filter = newsk->sk_filter;
1043 sk_filter_charge(newsk, filter);
1045 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1046 /* It is still raw copy of parent, so invalidate
1047 * destructor and make plain sk_free() */
1048 newsk->sk_destruct = NULL;
1055 newsk->sk_priority = 0;
1056 atomic_set(&newsk->sk_refcnt, 2);
1059 * Increment the counter in the same struct proto as the master
1060 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1061 * is the same as sk->sk_prot->socks, as this field was copied
1064 * This _changes_ the previous behaviour, where
1065 * tcp_create_openreq_child always was incrementing the
1066 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1067 * to be taken into account in all callers. -acme
1069 sk_refcnt_debug_inc(newsk);
1070 sk_set_socket(newsk, NULL);
1071 newsk->sk_sleep = NULL;
1073 if (newsk->sk_prot->sockets_allocated)
1074 atomic_inc(newsk->sk_prot->sockets_allocated);
1080 EXPORT_SYMBOL_GPL(sk_clone);
1082 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1084 __sk_dst_set(sk, dst);
1085 sk->sk_route_caps = dst->dev->features;
1086 if (sk->sk_route_caps & NETIF_F_GSO)
1087 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1088 if (sk_can_gso(sk)) {
1089 if (dst->header_len) {
1090 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1092 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1093 sk->sk_gso_max_size = dst->dev->gso_max_size;
1097 EXPORT_SYMBOL_GPL(sk_setup_caps);
1099 void __init sk_init(void)
1101 if (num_physpages <= 4096) {
1102 sysctl_wmem_max = 32767;
1103 sysctl_rmem_max = 32767;
1104 sysctl_wmem_default = 32767;
1105 sysctl_rmem_default = 32767;
1106 } else if (num_physpages >= 131072) {
1107 sysctl_wmem_max = 131071;
1108 sysctl_rmem_max = 131071;
1113 * Simple resource managers for sockets.
1118 * Write buffer destructor automatically called from kfree_skb.
1120 void sock_wfree(struct sk_buff *skb)
1122 struct sock *sk = skb->sk;
1124 /* In case it might be waiting for more memory. */
1125 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1126 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1127 sk->sk_write_space(sk);
1132 * Read buffer destructor automatically called from kfree_skb.
1134 void sock_rfree(struct sk_buff *skb)
1136 struct sock *sk = skb->sk;
1138 skb_truesize_check(skb);
1139 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1140 sk_mem_uncharge(skb->sk, skb->truesize);
1144 int sock_i_uid(struct sock *sk)
1148 read_lock(&sk->sk_callback_lock);
1149 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1150 read_unlock(&sk->sk_callback_lock);
1154 unsigned long sock_i_ino(struct sock *sk)
1158 read_lock(&sk->sk_callback_lock);
1159 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1160 read_unlock(&sk->sk_callback_lock);
1165 * Allocate a skb from the socket's send buffer.
1167 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1170 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1171 struct sk_buff * skb = alloc_skb(size, priority);
1173 skb_set_owner_w(skb, sk);
1181 * Allocate a skb from the socket's receive buffer.
1183 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1186 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1187 struct sk_buff *skb = alloc_skb(size, priority);
1189 skb_set_owner_r(skb, sk);
1197 * Allocate a memory block from the socket's option memory buffer.
1199 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1201 if ((unsigned)size <= sysctl_optmem_max &&
1202 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1204 /* First do the add, to avoid the race if kmalloc
1207 atomic_add(size, &sk->sk_omem_alloc);
1208 mem = kmalloc(size, priority);
1211 atomic_sub(size, &sk->sk_omem_alloc);
1217 * Free an option memory block.
1219 void sock_kfree_s(struct sock *sk, void *mem, int size)
1222 atomic_sub(size, &sk->sk_omem_alloc);
1225 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1226 I think, these locks should be removed for datagram sockets.
1228 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1232 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1236 if (signal_pending(current))
1238 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1239 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1240 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1242 if (sk->sk_shutdown & SEND_SHUTDOWN)
1246 timeo = schedule_timeout(timeo);
1248 finish_wait(sk->sk_sleep, &wait);
1254 * Generic send/receive buffer handlers
1257 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1258 unsigned long header_len,
1259 unsigned long data_len,
1260 int noblock, int *errcode)
1262 struct sk_buff *skb;
1267 gfp_mask = sk->sk_allocation;
1268 if (gfp_mask & __GFP_WAIT)
1269 gfp_mask |= __GFP_REPEAT;
1271 timeo = sock_sndtimeo(sk, noblock);
1273 err = sock_error(sk);
1278 if (sk->sk_shutdown & SEND_SHUTDOWN)
1281 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1282 skb = alloc_skb(header_len, gfp_mask);
1287 /* No pages, we're done... */
1291 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1292 skb->truesize += data_len;
1293 skb_shinfo(skb)->nr_frags = npages;
1294 for (i = 0; i < npages; i++) {
1298 page = alloc_pages(sk->sk_allocation, 0);
1301 skb_shinfo(skb)->nr_frags = i;
1306 frag = &skb_shinfo(skb)->frags[i];
1308 frag->page_offset = 0;
1309 frag->size = (data_len >= PAGE_SIZE ?
1312 data_len -= PAGE_SIZE;
1315 /* Full success... */
1321 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1322 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1326 if (signal_pending(current))
1328 timeo = sock_wait_for_wmem(sk, timeo);
1331 skb_set_owner_w(skb, sk);
1335 err = sock_intr_errno(timeo);
1341 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1342 int noblock, int *errcode)
1344 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1347 static void __lock_sock(struct sock *sk)
1352 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1353 TASK_UNINTERRUPTIBLE);
1354 spin_unlock_bh(&sk->sk_lock.slock);
1356 spin_lock_bh(&sk->sk_lock.slock);
1357 if (!sock_owned_by_user(sk))
1360 finish_wait(&sk->sk_lock.wq, &wait);
1363 static void __release_sock(struct sock *sk)
1365 struct sk_buff *skb = sk->sk_backlog.head;
1368 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1372 struct sk_buff *next = skb->next;
1375 sk_backlog_rcv(sk, skb);
1378 * We are in process context here with softirqs
1379 * disabled, use cond_resched_softirq() to preempt.
1380 * This is safe to do because we've taken the backlog
1383 cond_resched_softirq();
1386 } while (skb != NULL);
1389 } while ((skb = sk->sk_backlog.head) != NULL);
1393 * sk_wait_data - wait for data to arrive at sk_receive_queue
1394 * @sk: sock to wait on
1395 * @timeo: for how long
1397 * Now socket state including sk->sk_err is changed only under lock,
1398 * hence we may omit checks after joining wait queue.
1399 * We check receive queue before schedule() only as optimization;
1400 * it is very likely that release_sock() added new data.
1402 int sk_wait_data(struct sock *sk, long *timeo)
1407 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1408 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1409 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1410 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1411 finish_wait(sk->sk_sleep, &wait);
1415 EXPORT_SYMBOL(sk_wait_data);
1418 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1420 * @size: memory size to allocate
1421 * @kind: allocation type
1423 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1424 * rmem allocation. This function assumes that protocols which have
1425 * memory_pressure use sk_wmem_queued as write buffer accounting.
1427 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1429 struct proto *prot = sk->sk_prot;
1430 int amt = sk_mem_pages(size);
1433 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1434 allocated = atomic_add_return(amt, prot->memory_allocated);
1437 if (allocated <= prot->sysctl_mem[0]) {
1438 if (prot->memory_pressure && *prot->memory_pressure)
1439 *prot->memory_pressure = 0;
1443 /* Under pressure. */
1444 if (allocated > prot->sysctl_mem[1])
1445 if (prot->enter_memory_pressure)
1446 prot->enter_memory_pressure(sk);
1448 /* Over hard limit. */
1449 if (allocated > prot->sysctl_mem[2])
1450 goto suppress_allocation;
1452 /* guarantee minimum buffer size under pressure */
1453 if (kind == SK_MEM_RECV) {
1454 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1456 } else { /* SK_MEM_SEND */
1457 if (sk->sk_type == SOCK_STREAM) {
1458 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1460 } else if (atomic_read(&sk->sk_wmem_alloc) <
1461 prot->sysctl_wmem[0])
1465 if (prot->memory_pressure) {
1466 if (!*prot->memory_pressure ||
1467 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1468 sk_mem_pages(sk->sk_wmem_queued +
1469 atomic_read(&sk->sk_rmem_alloc) +
1470 sk->sk_forward_alloc))
1474 suppress_allocation:
1476 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1477 sk_stream_moderate_sndbuf(sk);
1479 /* Fail only if socket is _under_ its sndbuf.
1480 * In this case we cannot block, so that we have to fail.
1482 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1486 /* Alas. Undo changes. */
1487 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1488 atomic_sub(amt, prot->memory_allocated);
1492 EXPORT_SYMBOL(__sk_mem_schedule);
1495 * __sk_reclaim - reclaim memory_allocated
1498 void __sk_mem_reclaim(struct sock *sk)
1500 struct proto *prot = sk->sk_prot;
1502 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1503 prot->memory_allocated);
1504 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1506 if (prot->memory_pressure && *prot->memory_pressure &&
1507 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1508 *prot->memory_pressure = 0;
1511 EXPORT_SYMBOL(__sk_mem_reclaim);
1515 * Set of default routines for initialising struct proto_ops when
1516 * the protocol does not support a particular function. In certain
1517 * cases where it makes no sense for a protocol to have a "do nothing"
1518 * function, some default processing is provided.
1521 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1526 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1532 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1537 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1542 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1548 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1553 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1558 int sock_no_listen(struct socket *sock, int backlog)
1563 int sock_no_shutdown(struct socket *sock, int how)
1568 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1569 char __user *optval, int optlen)
1574 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1575 char __user *optval, int __user *optlen)
1580 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1586 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1587 size_t len, int flags)
1592 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1594 /* Mirror missing mmap method error code */
1598 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1601 struct msghdr msg = {.msg_flags = flags};
1603 char *kaddr = kmap(page);
1604 iov.iov_base = kaddr + offset;
1606 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1612 * Default Socket Callbacks
1615 static void sock_def_wakeup(struct sock *sk)
1617 read_lock(&sk->sk_callback_lock);
1618 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1619 wake_up_interruptible_all(sk->sk_sleep);
1620 read_unlock(&sk->sk_callback_lock);
1623 static void sock_def_error_report(struct sock *sk)
1625 read_lock(&sk->sk_callback_lock);
1626 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1627 wake_up_interruptible(sk->sk_sleep);
1628 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1629 read_unlock(&sk->sk_callback_lock);
1632 static void sock_def_readable(struct sock *sk, int len)
1634 read_lock(&sk->sk_callback_lock);
1635 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1636 wake_up_interruptible_sync(sk->sk_sleep);
1637 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1638 read_unlock(&sk->sk_callback_lock);
1641 static void sock_def_write_space(struct sock *sk)
1643 read_lock(&sk->sk_callback_lock);
1645 /* Do not wake up a writer until he can make "significant"
1648 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1649 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1650 wake_up_interruptible_sync(sk->sk_sleep);
1652 /* Should agree with poll, otherwise some programs break */
1653 if (sock_writeable(sk))
1654 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1657 read_unlock(&sk->sk_callback_lock);
1660 static void sock_def_destruct(struct sock *sk)
1662 kfree(sk->sk_protinfo);
1665 void sk_send_sigurg(struct sock *sk)
1667 if (sk->sk_socket && sk->sk_socket->file)
1668 if (send_sigurg(&sk->sk_socket->file->f_owner))
1669 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1672 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1673 unsigned long expires)
1675 if (!mod_timer(timer, expires))
1679 EXPORT_SYMBOL(sk_reset_timer);
1681 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1683 if (timer_pending(timer) && del_timer(timer))
1687 EXPORT_SYMBOL(sk_stop_timer);
1689 void sock_init_data(struct socket *sock, struct sock *sk)
1691 skb_queue_head_init(&sk->sk_receive_queue);
1692 skb_queue_head_init(&sk->sk_write_queue);
1693 skb_queue_head_init(&sk->sk_error_queue);
1694 #ifdef CONFIG_NET_DMA
1695 skb_queue_head_init(&sk->sk_async_wait_queue);
1698 sk->sk_send_head = NULL;
1700 init_timer(&sk->sk_timer);
1702 sk->sk_allocation = GFP_KERNEL;
1703 sk->sk_rcvbuf = sysctl_rmem_default;
1704 sk->sk_sndbuf = sysctl_wmem_default;
1705 sk->sk_state = TCP_CLOSE;
1706 sk_set_socket(sk, sock);
1708 sock_set_flag(sk, SOCK_ZAPPED);
1711 sk->sk_type = sock->type;
1712 sk->sk_sleep = &sock->wait;
1715 sk->sk_sleep = NULL;
1717 rwlock_init(&sk->sk_dst_lock);
1718 rwlock_init(&sk->sk_callback_lock);
1719 lockdep_set_class_and_name(&sk->sk_callback_lock,
1720 af_callback_keys + sk->sk_family,
1721 af_family_clock_key_strings[sk->sk_family]);
1723 sk->sk_state_change = sock_def_wakeup;
1724 sk->sk_data_ready = sock_def_readable;
1725 sk->sk_write_space = sock_def_write_space;
1726 sk->sk_error_report = sock_def_error_report;
1727 sk->sk_destruct = sock_def_destruct;
1729 sk->sk_sndmsg_page = NULL;
1730 sk->sk_sndmsg_off = 0;
1732 sk->sk_peercred.pid = 0;
1733 sk->sk_peercred.uid = -1;
1734 sk->sk_peercred.gid = -1;
1735 sk->sk_write_pending = 0;
1736 sk->sk_rcvlowat = 1;
1737 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1738 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1740 sk->sk_stamp = ktime_set(-1L, 0);
1742 atomic_set(&sk->sk_refcnt, 1);
1743 atomic_set(&sk->sk_drops, 0);
1746 void lock_sock_nested(struct sock *sk, int subclass)
1749 spin_lock_bh(&sk->sk_lock.slock);
1750 if (sk->sk_lock.owned)
1752 sk->sk_lock.owned = 1;
1753 spin_unlock(&sk->sk_lock.slock);
1755 * The sk_lock has mutex_lock() semantics here:
1757 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1761 EXPORT_SYMBOL(lock_sock_nested);
1763 void release_sock(struct sock *sk)
1766 * The sk_lock has mutex_unlock() semantics:
1768 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1770 spin_lock_bh(&sk->sk_lock.slock);
1771 if (sk->sk_backlog.tail)
1773 sk->sk_lock.owned = 0;
1774 if (waitqueue_active(&sk->sk_lock.wq))
1775 wake_up(&sk->sk_lock.wq);
1776 spin_unlock_bh(&sk->sk_lock.slock);
1778 EXPORT_SYMBOL(release_sock);
1780 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1783 if (!sock_flag(sk, SOCK_TIMESTAMP))
1784 sock_enable_timestamp(sk);
1785 tv = ktime_to_timeval(sk->sk_stamp);
1786 if (tv.tv_sec == -1)
1788 if (tv.tv_sec == 0) {
1789 sk->sk_stamp = ktime_get_real();
1790 tv = ktime_to_timeval(sk->sk_stamp);
1792 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1794 EXPORT_SYMBOL(sock_get_timestamp);
1796 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1799 if (!sock_flag(sk, SOCK_TIMESTAMP))
1800 sock_enable_timestamp(sk);
1801 ts = ktime_to_timespec(sk->sk_stamp);
1802 if (ts.tv_sec == -1)
1804 if (ts.tv_sec == 0) {
1805 sk->sk_stamp = ktime_get_real();
1806 ts = ktime_to_timespec(sk->sk_stamp);
1808 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1810 EXPORT_SYMBOL(sock_get_timestampns);
1812 void sock_enable_timestamp(struct sock *sk)
1814 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1815 sock_set_flag(sk, SOCK_TIMESTAMP);
1816 net_enable_timestamp();
1821 * Get a socket option on an socket.
1823 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1824 * asynchronous errors should be reported by getsockopt. We assume
1825 * this means if you specify SO_ERROR (otherwise whats the point of it).
1827 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1828 char __user *optval, int __user *optlen)
1830 struct sock *sk = sock->sk;
1832 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1835 EXPORT_SYMBOL(sock_common_getsockopt);
1837 #ifdef CONFIG_COMPAT
1838 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1839 char __user *optval, int __user *optlen)
1841 struct sock *sk = sock->sk;
1843 if (sk->sk_prot->compat_getsockopt != NULL)
1844 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1846 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1848 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1851 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1852 struct msghdr *msg, size_t size, int flags)
1854 struct sock *sk = sock->sk;
1858 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1859 flags & ~MSG_DONTWAIT, &addr_len);
1861 msg->msg_namelen = addr_len;
1865 EXPORT_SYMBOL(sock_common_recvmsg);
1868 * Set socket options on an inet socket.
1870 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1871 char __user *optval, int optlen)
1873 struct sock *sk = sock->sk;
1875 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1878 EXPORT_SYMBOL(sock_common_setsockopt);
1880 #ifdef CONFIG_COMPAT
1881 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1882 char __user *optval, int optlen)
1884 struct sock *sk = sock->sk;
1886 if (sk->sk_prot->compat_setsockopt != NULL)
1887 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1889 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1891 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1894 void sk_common_release(struct sock *sk)
1896 if (sk->sk_prot->destroy)
1897 sk->sk_prot->destroy(sk);
1900 * Observation: when sock_common_release is called, processes have
1901 * no access to socket. But net still has.
1902 * Step one, detach it from networking:
1904 * A. Remove from hash tables.
1907 sk->sk_prot->unhash(sk);
1910 * In this point socket cannot receive new packets, but it is possible
1911 * that some packets are in flight because some CPU runs receiver and
1912 * did hash table lookup before we unhashed socket. They will achieve
1913 * receive queue and will be purged by socket destructor.
1915 * Also we still have packets pending on receive queue and probably,
1916 * our own packets waiting in device queues. sock_destroy will drain
1917 * receive queue, but transmitted packets will delay socket destruction
1918 * until the last reference will be released.
1923 xfrm_sk_free_policy(sk);
1925 sk_refcnt_debug_release(sk);
1929 EXPORT_SYMBOL(sk_common_release);
1931 static DEFINE_RWLOCK(proto_list_lock);
1932 static LIST_HEAD(proto_list);
1934 #ifdef CONFIG_PROC_FS
1935 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1937 int val[PROTO_INUSE_NR];
1940 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1942 #ifdef CONFIG_NET_NS
1943 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1945 int cpu = smp_processor_id();
1946 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1948 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1950 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1952 int cpu, idx = prot->inuse_idx;
1955 for_each_possible_cpu(cpu)
1956 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1958 return res >= 0 ? res : 0;
1960 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1962 static int sock_inuse_init_net(struct net *net)
1964 net->core.inuse = alloc_percpu(struct prot_inuse);
1965 return net->core.inuse ? 0 : -ENOMEM;
1968 static void sock_inuse_exit_net(struct net *net)
1970 free_percpu(net->core.inuse);
1973 static struct pernet_operations net_inuse_ops = {
1974 .init = sock_inuse_init_net,
1975 .exit = sock_inuse_exit_net,
1978 static __init int net_inuse_init(void)
1980 if (register_pernet_subsys(&net_inuse_ops))
1981 panic("Cannot initialize net inuse counters");
1986 core_initcall(net_inuse_init);
1988 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1990 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1992 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1994 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1996 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1998 int cpu, idx = prot->inuse_idx;
2001 for_each_possible_cpu(cpu)
2002 res += per_cpu(prot_inuse, cpu).val[idx];
2004 return res >= 0 ? res : 0;
2006 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2009 static void assign_proto_idx(struct proto *prot)
2011 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2013 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2014 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2018 set_bit(prot->inuse_idx, proto_inuse_idx);
2021 static void release_proto_idx(struct proto *prot)
2023 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2024 clear_bit(prot->inuse_idx, proto_inuse_idx);
2027 static inline void assign_proto_idx(struct proto *prot)
2031 static inline void release_proto_idx(struct proto *prot)
2036 int proto_register(struct proto *prot, int alloc_slab)
2039 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2040 SLAB_HWCACHE_ALIGN, NULL);
2042 if (prot->slab == NULL) {
2043 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2048 if (prot->rsk_prot != NULL) {
2049 static const char mask[] = "request_sock_%s";
2051 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2052 if (prot->rsk_prot->slab_name == NULL)
2053 goto out_free_sock_slab;
2055 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2056 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2057 prot->rsk_prot->obj_size, 0,
2058 SLAB_HWCACHE_ALIGN, NULL);
2060 if (prot->rsk_prot->slab == NULL) {
2061 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2063 goto out_free_request_sock_slab_name;
2067 if (prot->twsk_prot != NULL) {
2068 static const char mask[] = "tw_sock_%s";
2070 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2072 if (prot->twsk_prot->twsk_slab_name == NULL)
2073 goto out_free_request_sock_slab;
2075 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2076 prot->twsk_prot->twsk_slab =
2077 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2078 prot->twsk_prot->twsk_obj_size,
2079 0, SLAB_HWCACHE_ALIGN,
2081 if (prot->twsk_prot->twsk_slab == NULL)
2082 goto out_free_timewait_sock_slab_name;
2086 write_lock(&proto_list_lock);
2087 list_add(&prot->node, &proto_list);
2088 assign_proto_idx(prot);
2089 write_unlock(&proto_list_lock);
2092 out_free_timewait_sock_slab_name:
2093 kfree(prot->twsk_prot->twsk_slab_name);
2094 out_free_request_sock_slab:
2095 if (prot->rsk_prot && prot->rsk_prot->slab) {
2096 kmem_cache_destroy(prot->rsk_prot->slab);
2097 prot->rsk_prot->slab = NULL;
2099 out_free_request_sock_slab_name:
2100 kfree(prot->rsk_prot->slab_name);
2102 kmem_cache_destroy(prot->slab);
2108 EXPORT_SYMBOL(proto_register);
2110 void proto_unregister(struct proto *prot)
2112 write_lock(&proto_list_lock);
2113 release_proto_idx(prot);
2114 list_del(&prot->node);
2115 write_unlock(&proto_list_lock);
2117 if (prot->slab != NULL) {
2118 kmem_cache_destroy(prot->slab);
2122 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2123 kmem_cache_destroy(prot->rsk_prot->slab);
2124 kfree(prot->rsk_prot->slab_name);
2125 prot->rsk_prot->slab = NULL;
2128 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2129 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2130 kfree(prot->twsk_prot->twsk_slab_name);
2131 prot->twsk_prot->twsk_slab = NULL;
2135 EXPORT_SYMBOL(proto_unregister);
2137 #ifdef CONFIG_PROC_FS
2138 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2139 __acquires(proto_list_lock)
2141 read_lock(&proto_list_lock);
2142 return seq_list_start_head(&proto_list, *pos);
2145 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2147 return seq_list_next(v, &proto_list, pos);
2150 static void proto_seq_stop(struct seq_file *seq, void *v)
2151 __releases(proto_list_lock)
2153 read_unlock(&proto_list_lock);
2156 static char proto_method_implemented(const void *method)
2158 return method == NULL ? 'n' : 'y';
2161 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2163 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2164 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2167 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2168 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2169 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2171 proto->slab == NULL ? "no" : "yes",
2172 module_name(proto->owner),
2173 proto_method_implemented(proto->close),
2174 proto_method_implemented(proto->connect),
2175 proto_method_implemented(proto->disconnect),
2176 proto_method_implemented(proto->accept),
2177 proto_method_implemented(proto->ioctl),
2178 proto_method_implemented(proto->init),
2179 proto_method_implemented(proto->destroy),
2180 proto_method_implemented(proto->shutdown),
2181 proto_method_implemented(proto->setsockopt),
2182 proto_method_implemented(proto->getsockopt),
2183 proto_method_implemented(proto->sendmsg),
2184 proto_method_implemented(proto->recvmsg),
2185 proto_method_implemented(proto->sendpage),
2186 proto_method_implemented(proto->bind),
2187 proto_method_implemented(proto->backlog_rcv),
2188 proto_method_implemented(proto->hash),
2189 proto_method_implemented(proto->unhash),
2190 proto_method_implemented(proto->get_port),
2191 proto_method_implemented(proto->enter_memory_pressure));
2194 static int proto_seq_show(struct seq_file *seq, void *v)
2196 if (v == &proto_list)
2197 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2206 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2208 proto_seq_printf(seq, list_entry(v, struct proto, node));
2212 static const struct seq_operations proto_seq_ops = {
2213 .start = proto_seq_start,
2214 .next = proto_seq_next,
2215 .stop = proto_seq_stop,
2216 .show = proto_seq_show,
2219 static int proto_seq_open(struct inode *inode, struct file *file)
2221 return seq_open(file, &proto_seq_ops);
2224 static const struct file_operations proto_seq_fops = {
2225 .owner = THIS_MODULE,
2226 .open = proto_seq_open,
2228 .llseek = seq_lseek,
2229 .release = seq_release,
2232 static int __init proto_init(void)
2234 /* register /proc/net/protocols */
2235 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2238 subsys_initcall(proto_init);
2240 #endif /* PROC_FS */
2242 EXPORT_SYMBOL(sk_alloc);
2243 EXPORT_SYMBOL(sk_free);
2244 EXPORT_SYMBOL(sk_send_sigurg);
2245 EXPORT_SYMBOL(sock_alloc_send_skb);
2246 EXPORT_SYMBOL(sock_init_data);
2247 EXPORT_SYMBOL(sock_kfree_s);
2248 EXPORT_SYMBOL(sock_kmalloc);
2249 EXPORT_SYMBOL(sock_no_accept);
2250 EXPORT_SYMBOL(sock_no_bind);
2251 EXPORT_SYMBOL(sock_no_connect);
2252 EXPORT_SYMBOL(sock_no_getname);
2253 EXPORT_SYMBOL(sock_no_getsockopt);
2254 EXPORT_SYMBOL(sock_no_ioctl);
2255 EXPORT_SYMBOL(sock_no_listen);
2256 EXPORT_SYMBOL(sock_no_mmap);
2257 EXPORT_SYMBOL(sock_no_poll);
2258 EXPORT_SYMBOL(sock_no_recvmsg);
2259 EXPORT_SYMBOL(sock_no_sendmsg);
2260 EXPORT_SYMBOL(sock_no_sendpage);
2261 EXPORT_SYMBOL(sock_no_setsockopt);
2262 EXPORT_SYMBOL(sock_no_shutdown);
2263 EXPORT_SYMBOL(sock_no_socketpair);
2264 EXPORT_SYMBOL(sock_rfree);
2265 EXPORT_SYMBOL(sock_setsockopt);
2266 EXPORT_SYMBOL(sock_wfree);
2267 EXPORT_SYMBOL(sock_wmalloc);
2268 EXPORT_SYMBOL(sock_i_uid);
2269 EXPORT_SYMBOL(sock_i_ino);
2270 EXPORT_SYMBOL(sysctl_optmem_max);