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 * release dst right now while its hot
295 dst_release(skb->dst);
297 /* Cache the SKB length before we tack it onto the receive
298 * queue. Once it is added it no longer belongs to us and
299 * may be freed by other threads of control pulling packets
304 skb_queue_tail(&sk->sk_receive_queue, skb);
306 if (!sock_flag(sk, SOCK_DEAD))
307 sk->sk_data_ready(sk, skb_len);
311 EXPORT_SYMBOL(sock_queue_rcv_skb);
313 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
315 int rc = NET_RX_SUCCESS;
317 if (sk_filter(sk, skb))
318 goto discard_and_relse;
323 bh_lock_sock_nested(sk);
326 if (!sock_owned_by_user(sk)) {
328 * trylock + unlock semantics:
330 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
332 rc = sk_backlog_rcv(sk, skb);
334 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
336 sk_add_backlog(sk, skb);
345 EXPORT_SYMBOL(sk_receive_skb);
347 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
349 struct dst_entry *dst = sk->sk_dst_cache;
351 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
352 sk->sk_dst_cache = NULL;
359 EXPORT_SYMBOL(__sk_dst_check);
361 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
363 struct dst_entry *dst = sk_dst_get(sk);
365 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
373 EXPORT_SYMBOL(sk_dst_check);
375 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
377 int ret = -ENOPROTOOPT;
378 #ifdef CONFIG_NETDEVICES
379 struct net *net = sock_net(sk);
380 char devname[IFNAMSIZ];
385 if (!capable(CAP_NET_RAW))
392 /* Bind this socket to a particular device like "eth0",
393 * as specified in the passed interface name. If the
394 * name is "" or the option length is zero the socket
397 if (optlen > IFNAMSIZ - 1)
398 optlen = IFNAMSIZ - 1;
399 memset(devname, 0, sizeof(devname));
402 if (copy_from_user(devname, optval, optlen))
405 if (devname[0] == '\0') {
408 struct net_device *dev = dev_get_by_name(net, devname);
414 index = dev->ifindex;
419 sk->sk_bound_dev_if = index;
431 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
434 sock_set_flag(sk, bit);
436 sock_reset_flag(sk, bit);
440 * This is meant for all protocols to use and covers goings on
441 * at the socket level. Everything here is generic.
444 int sock_setsockopt(struct socket *sock, int level, int optname,
445 char __user *optval, int optlen)
447 struct sock *sk=sock->sk;
454 * Options without arguments
457 if (optname == SO_BINDTODEVICE)
458 return sock_bindtodevice(sk, optval, optlen);
460 if (optlen < sizeof(int))
463 if (get_user(val, (int __user *)optval))
472 if (val && !capable(CAP_NET_ADMIN)) {
475 sock_valbool_flag(sk, SOCK_DBG, valbool);
478 sk->sk_reuse = valbool;
485 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
488 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
491 /* Don't error on this BSD doesn't and if you think
492 about it this is right. Otherwise apps have to
493 play 'guess the biggest size' games. RCVBUF/SNDBUF
494 are treated in BSD as hints */
496 if (val > sysctl_wmem_max)
497 val = sysctl_wmem_max;
499 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
500 if ((val * 2) < SOCK_MIN_SNDBUF)
501 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
503 sk->sk_sndbuf = val * 2;
506 * Wake up sending tasks if we
509 sk->sk_write_space(sk);
513 if (!capable(CAP_NET_ADMIN)) {
520 /* Don't error on this BSD doesn't and if you think
521 about it this is right. Otherwise apps have to
522 play 'guess the biggest size' games. RCVBUF/SNDBUF
523 are treated in BSD as hints */
525 if (val > sysctl_rmem_max)
526 val = sysctl_rmem_max;
528 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
530 * We double it on the way in to account for
531 * "struct sk_buff" etc. overhead. Applications
532 * assume that the SO_RCVBUF setting they make will
533 * allow that much actual data to be received on that
536 * Applications are unaware that "struct sk_buff" and
537 * other overheads allocate from the receive buffer
538 * during socket buffer allocation.
540 * And after considering the possible alternatives,
541 * returning the value we actually used in getsockopt
542 * is the most desirable behavior.
544 if ((val * 2) < SOCK_MIN_RCVBUF)
545 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
547 sk->sk_rcvbuf = val * 2;
551 if (!capable(CAP_NET_ADMIN)) {
559 if (sk->sk_protocol == IPPROTO_TCP)
560 tcp_set_keepalive(sk, valbool);
562 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
566 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
570 sk->sk_no_check = valbool;
574 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
575 sk->sk_priority = val;
581 if (optlen < sizeof(ling)) {
582 ret = -EINVAL; /* 1003.1g */
585 if (copy_from_user(&ling,optval,sizeof(ling))) {
590 sock_reset_flag(sk, SOCK_LINGER);
592 #if (BITS_PER_LONG == 32)
593 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
594 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
597 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
598 sock_set_flag(sk, SOCK_LINGER);
603 sock_warn_obsolete_bsdism("setsockopt");
608 set_bit(SOCK_PASSCRED, &sock->flags);
610 clear_bit(SOCK_PASSCRED, &sock->flags);
616 if (optname == SO_TIMESTAMP)
617 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
619 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
620 sock_set_flag(sk, SOCK_RCVTSTAMP);
621 sock_enable_timestamp(sk);
623 sock_reset_flag(sk, SOCK_RCVTSTAMP);
624 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
631 sk->sk_rcvlowat = val ? : 1;
635 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
639 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
642 case SO_ATTACH_FILTER:
644 if (optlen == sizeof(struct sock_fprog)) {
645 struct sock_fprog fprog;
648 if (copy_from_user(&fprog, optval, sizeof(fprog)))
651 ret = sk_attach_filter(&fprog, sk);
655 case SO_DETACH_FILTER:
656 ret = sk_detach_filter(sk);
661 set_bit(SOCK_PASSSEC, &sock->flags);
663 clear_bit(SOCK_PASSSEC, &sock->flags);
666 if (!capable(CAP_NET_ADMIN))
673 /* We implement the SO_SNDLOWAT etc to
674 not be settable (1003.1g 5.3) */
684 int sock_getsockopt(struct socket *sock, int level, int optname,
685 char __user *optval, int __user *optlen)
687 struct sock *sk = sock->sk;
695 unsigned int lv = sizeof(int);
698 if (get_user(len, optlen))
705 v.val = sock_flag(sk, SOCK_DBG);
709 v.val = sock_flag(sk, SOCK_LOCALROUTE);
713 v.val = !!sock_flag(sk, SOCK_BROADCAST);
717 v.val = sk->sk_sndbuf;
721 v.val = sk->sk_rcvbuf;
725 v.val = sk->sk_reuse;
729 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
737 v.val = -sock_error(sk);
739 v.val = xchg(&sk->sk_err_soft, 0);
743 v.val = !!sock_flag(sk, SOCK_URGINLINE);
747 v.val = sk->sk_no_check;
751 v.val = sk->sk_priority;
756 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
757 v.ling.l_linger = sk->sk_lingertime / HZ;
761 sock_warn_obsolete_bsdism("getsockopt");
765 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
766 !sock_flag(sk, SOCK_RCVTSTAMPNS);
770 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
774 lv=sizeof(struct timeval);
775 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
779 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
780 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
785 lv=sizeof(struct timeval);
786 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
790 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
791 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
796 v.val = sk->sk_rcvlowat;
804 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
808 if (len > sizeof(sk->sk_peercred))
809 len = sizeof(sk->sk_peercred);
810 if (copy_to_user(optval, &sk->sk_peercred, len))
818 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
822 if (copy_to_user(optval, address, len))
827 /* Dubious BSD thing... Probably nobody even uses it, but
828 * the UNIX standard wants it for whatever reason... -DaveM
831 v.val = sk->sk_state == TCP_LISTEN;
835 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
839 return security_socket_getpeersec_stream(sock, optval, optlen, len);
851 if (copy_to_user(optval, &v, len))
854 if (put_user(len, optlen))
860 * Initialize an sk_lock.
862 * (We also register the sk_lock with the lock validator.)
864 static inline void sock_lock_init(struct sock *sk)
866 sock_lock_init_class_and_name(sk,
867 af_family_slock_key_strings[sk->sk_family],
868 af_family_slock_keys + sk->sk_family,
869 af_family_key_strings[sk->sk_family],
870 af_family_keys + sk->sk_family);
873 static void sock_copy(struct sock *nsk, const struct sock *osk)
875 #ifdef CONFIG_SECURITY_NETWORK
876 void *sptr = nsk->sk_security;
879 memcpy(nsk, osk, osk->sk_prot->obj_size);
880 #ifdef CONFIG_SECURITY_NETWORK
881 nsk->sk_security = sptr;
882 security_sk_clone(osk, nsk);
886 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
890 struct kmem_cache *slab;
894 sk = kmem_cache_alloc(slab, priority);
896 sk = kmalloc(prot->obj_size, priority);
899 if (security_sk_alloc(sk, family, priority))
902 if (!try_module_get(prot->owner))
909 security_sk_free(sk);
912 kmem_cache_free(slab, sk);
918 static void sk_prot_free(struct proto *prot, struct sock *sk)
920 struct kmem_cache *slab;
921 struct module *owner;
926 security_sk_free(sk);
928 kmem_cache_free(slab, sk);
935 * sk_alloc - All socket objects are allocated here
936 * @net: the applicable net namespace
937 * @family: protocol family
938 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
939 * @prot: struct proto associated with this new sock instance
941 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
946 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
948 sk->sk_family = family;
950 * See comment in struct sock definition to understand
951 * why we need sk_prot_creator -acme
953 sk->sk_prot = sk->sk_prot_creator = prot;
955 sock_net_set(sk, get_net(net));
961 void sk_free(struct sock *sk)
963 struct sk_filter *filter;
968 filter = rcu_dereference(sk->sk_filter);
970 sk_filter_uncharge(sk, filter);
971 rcu_assign_pointer(sk->sk_filter, NULL);
974 sock_disable_timestamp(sk);
976 if (atomic_read(&sk->sk_omem_alloc))
977 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
978 __func__, atomic_read(&sk->sk_omem_alloc));
980 put_net(sock_net(sk));
981 sk_prot_free(sk->sk_prot_creator, sk);
985 * Last sock_put should drop referrence to sk->sk_net. It has already
986 * been dropped in sk_change_net. Taking referrence to stopping namespace
988 * Take referrence to a socket to remove it from hash _alive_ and after that
989 * destroy it in the context of init_net.
991 void sk_release_kernel(struct sock *sk)
993 if (sk == NULL || sk->sk_socket == NULL)
997 sock_release(sk->sk_socket);
998 release_net(sock_net(sk));
999 sock_net_set(sk, get_net(&init_net));
1002 EXPORT_SYMBOL(sk_release_kernel);
1004 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1008 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1009 if (newsk != NULL) {
1010 struct sk_filter *filter;
1012 sock_copy(newsk, sk);
1015 get_net(sock_net(newsk));
1016 sk_node_init(&newsk->sk_node);
1017 sock_lock_init(newsk);
1018 bh_lock_sock(newsk);
1019 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1021 atomic_set(&newsk->sk_rmem_alloc, 0);
1022 atomic_set(&newsk->sk_wmem_alloc, 0);
1023 atomic_set(&newsk->sk_omem_alloc, 0);
1024 skb_queue_head_init(&newsk->sk_receive_queue);
1025 skb_queue_head_init(&newsk->sk_write_queue);
1026 #ifdef CONFIG_NET_DMA
1027 skb_queue_head_init(&newsk->sk_async_wait_queue);
1030 rwlock_init(&newsk->sk_dst_lock);
1031 rwlock_init(&newsk->sk_callback_lock);
1032 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1033 af_callback_keys + newsk->sk_family,
1034 af_family_clock_key_strings[newsk->sk_family]);
1036 newsk->sk_dst_cache = NULL;
1037 newsk->sk_wmem_queued = 0;
1038 newsk->sk_forward_alloc = 0;
1039 newsk->sk_send_head = NULL;
1040 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1042 sock_reset_flag(newsk, SOCK_DONE);
1043 skb_queue_head_init(&newsk->sk_error_queue);
1045 filter = newsk->sk_filter;
1047 sk_filter_charge(newsk, filter);
1049 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1050 /* It is still raw copy of parent, so invalidate
1051 * destructor and make plain sk_free() */
1052 newsk->sk_destruct = NULL;
1059 newsk->sk_priority = 0;
1060 atomic_set(&newsk->sk_refcnt, 2);
1063 * Increment the counter in the same struct proto as the master
1064 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1065 * is the same as sk->sk_prot->socks, as this field was copied
1068 * This _changes_ the previous behaviour, where
1069 * tcp_create_openreq_child always was incrementing the
1070 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1071 * to be taken into account in all callers. -acme
1073 sk_refcnt_debug_inc(newsk);
1074 sk_set_socket(newsk, NULL);
1075 newsk->sk_sleep = NULL;
1077 if (newsk->sk_prot->sockets_allocated)
1078 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1084 EXPORT_SYMBOL_GPL(sk_clone);
1086 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1088 __sk_dst_set(sk, dst);
1089 sk->sk_route_caps = dst->dev->features;
1090 if (sk->sk_route_caps & NETIF_F_GSO)
1091 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1092 if (sk_can_gso(sk)) {
1093 if (dst->header_len) {
1094 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1096 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1097 sk->sk_gso_max_size = dst->dev->gso_max_size;
1101 EXPORT_SYMBOL_GPL(sk_setup_caps);
1103 void __init sk_init(void)
1105 if (num_physpages <= 4096) {
1106 sysctl_wmem_max = 32767;
1107 sysctl_rmem_max = 32767;
1108 sysctl_wmem_default = 32767;
1109 sysctl_rmem_default = 32767;
1110 } else if (num_physpages >= 131072) {
1111 sysctl_wmem_max = 131071;
1112 sysctl_rmem_max = 131071;
1117 * Simple resource managers for sockets.
1122 * Write buffer destructor automatically called from kfree_skb.
1124 void sock_wfree(struct sk_buff *skb)
1126 struct sock *sk = skb->sk;
1128 /* In case it might be waiting for more memory. */
1129 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1130 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1131 sk->sk_write_space(sk);
1136 * Read buffer destructor automatically called from kfree_skb.
1138 void sock_rfree(struct sk_buff *skb)
1140 struct sock *sk = skb->sk;
1142 skb_truesize_check(skb);
1143 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1144 sk_mem_uncharge(skb->sk, skb->truesize);
1148 int sock_i_uid(struct sock *sk)
1152 read_lock(&sk->sk_callback_lock);
1153 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1154 read_unlock(&sk->sk_callback_lock);
1158 unsigned long sock_i_ino(struct sock *sk)
1162 read_lock(&sk->sk_callback_lock);
1163 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1164 read_unlock(&sk->sk_callback_lock);
1169 * Allocate a skb from the socket's send buffer.
1171 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1174 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1175 struct sk_buff * skb = alloc_skb(size, priority);
1177 skb_set_owner_w(skb, sk);
1185 * Allocate a skb from the socket's receive buffer.
1187 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1190 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1191 struct sk_buff *skb = alloc_skb(size, priority);
1193 skb_set_owner_r(skb, sk);
1201 * Allocate a memory block from the socket's option memory buffer.
1203 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1205 if ((unsigned)size <= sysctl_optmem_max &&
1206 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1208 /* First do the add, to avoid the race if kmalloc
1211 atomic_add(size, &sk->sk_omem_alloc);
1212 mem = kmalloc(size, priority);
1215 atomic_sub(size, &sk->sk_omem_alloc);
1221 * Free an option memory block.
1223 void sock_kfree_s(struct sock *sk, void *mem, int size)
1226 atomic_sub(size, &sk->sk_omem_alloc);
1229 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1230 I think, these locks should be removed for datagram sockets.
1232 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1236 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1240 if (signal_pending(current))
1242 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1243 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1244 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1246 if (sk->sk_shutdown & SEND_SHUTDOWN)
1250 timeo = schedule_timeout(timeo);
1252 finish_wait(sk->sk_sleep, &wait);
1258 * Generic send/receive buffer handlers
1261 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1262 unsigned long header_len,
1263 unsigned long data_len,
1264 int noblock, int *errcode)
1266 struct sk_buff *skb;
1271 gfp_mask = sk->sk_allocation;
1272 if (gfp_mask & __GFP_WAIT)
1273 gfp_mask |= __GFP_REPEAT;
1275 timeo = sock_sndtimeo(sk, noblock);
1277 err = sock_error(sk);
1282 if (sk->sk_shutdown & SEND_SHUTDOWN)
1285 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1286 skb = alloc_skb(header_len, gfp_mask);
1291 /* No pages, we're done... */
1295 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1296 skb->truesize += data_len;
1297 skb_shinfo(skb)->nr_frags = npages;
1298 for (i = 0; i < npages; i++) {
1302 page = alloc_pages(sk->sk_allocation, 0);
1305 skb_shinfo(skb)->nr_frags = i;
1310 frag = &skb_shinfo(skb)->frags[i];
1312 frag->page_offset = 0;
1313 frag->size = (data_len >= PAGE_SIZE ?
1316 data_len -= PAGE_SIZE;
1319 /* Full success... */
1325 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1326 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1330 if (signal_pending(current))
1332 timeo = sock_wait_for_wmem(sk, timeo);
1335 skb_set_owner_w(skb, sk);
1339 err = sock_intr_errno(timeo);
1345 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1346 int noblock, int *errcode)
1348 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1351 static void __lock_sock(struct sock *sk)
1356 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1357 TASK_UNINTERRUPTIBLE);
1358 spin_unlock_bh(&sk->sk_lock.slock);
1360 spin_lock_bh(&sk->sk_lock.slock);
1361 if (!sock_owned_by_user(sk))
1364 finish_wait(&sk->sk_lock.wq, &wait);
1367 static void __release_sock(struct sock *sk)
1369 struct sk_buff *skb = sk->sk_backlog.head;
1372 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1376 struct sk_buff *next = skb->next;
1379 sk_backlog_rcv(sk, skb);
1382 * We are in process context here with softirqs
1383 * disabled, use cond_resched_softirq() to preempt.
1384 * This is safe to do because we've taken the backlog
1387 cond_resched_softirq();
1390 } while (skb != NULL);
1393 } while ((skb = sk->sk_backlog.head) != NULL);
1397 * sk_wait_data - wait for data to arrive at sk_receive_queue
1398 * @sk: sock to wait on
1399 * @timeo: for how long
1401 * Now socket state including sk->sk_err is changed only under lock,
1402 * hence we may omit checks after joining wait queue.
1403 * We check receive queue before schedule() only as optimization;
1404 * it is very likely that release_sock() added new data.
1406 int sk_wait_data(struct sock *sk, long *timeo)
1411 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1412 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1413 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1414 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1415 finish_wait(sk->sk_sleep, &wait);
1419 EXPORT_SYMBOL(sk_wait_data);
1422 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1424 * @size: memory size to allocate
1425 * @kind: allocation type
1427 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1428 * rmem allocation. This function assumes that protocols which have
1429 * memory_pressure use sk_wmem_queued as write buffer accounting.
1431 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1433 struct proto *prot = sk->sk_prot;
1434 int amt = sk_mem_pages(size);
1437 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1438 allocated = atomic_add_return(amt, prot->memory_allocated);
1441 if (allocated <= prot->sysctl_mem[0]) {
1442 if (prot->memory_pressure && *prot->memory_pressure)
1443 *prot->memory_pressure = 0;
1447 /* Under pressure. */
1448 if (allocated > prot->sysctl_mem[1])
1449 if (prot->enter_memory_pressure)
1450 prot->enter_memory_pressure(sk);
1452 /* Over hard limit. */
1453 if (allocated > prot->sysctl_mem[2])
1454 goto suppress_allocation;
1456 /* guarantee minimum buffer size under pressure */
1457 if (kind == SK_MEM_RECV) {
1458 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1460 } else { /* SK_MEM_SEND */
1461 if (sk->sk_type == SOCK_STREAM) {
1462 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1464 } else if (atomic_read(&sk->sk_wmem_alloc) <
1465 prot->sysctl_wmem[0])
1469 if (prot->memory_pressure) {
1472 if (!*prot->memory_pressure)
1474 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1475 if (prot->sysctl_mem[2] > alloc *
1476 sk_mem_pages(sk->sk_wmem_queued +
1477 atomic_read(&sk->sk_rmem_alloc) +
1478 sk->sk_forward_alloc))
1482 suppress_allocation:
1484 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1485 sk_stream_moderate_sndbuf(sk);
1487 /* Fail only if socket is _under_ its sndbuf.
1488 * In this case we cannot block, so that we have to fail.
1490 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1494 /* Alas. Undo changes. */
1495 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1496 atomic_sub(amt, prot->memory_allocated);
1500 EXPORT_SYMBOL(__sk_mem_schedule);
1503 * __sk_reclaim - reclaim memory_allocated
1506 void __sk_mem_reclaim(struct sock *sk)
1508 struct proto *prot = sk->sk_prot;
1510 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1511 prot->memory_allocated);
1512 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1514 if (prot->memory_pressure && *prot->memory_pressure &&
1515 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1516 *prot->memory_pressure = 0;
1519 EXPORT_SYMBOL(__sk_mem_reclaim);
1523 * Set of default routines for initialising struct proto_ops when
1524 * the protocol does not support a particular function. In certain
1525 * cases where it makes no sense for a protocol to have a "do nothing"
1526 * function, some default processing is provided.
1529 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1534 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1540 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1545 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1550 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1556 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1561 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1566 int sock_no_listen(struct socket *sock, int backlog)
1571 int sock_no_shutdown(struct socket *sock, int how)
1576 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1577 char __user *optval, int optlen)
1582 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1583 char __user *optval, int __user *optlen)
1588 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1594 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1595 size_t len, int flags)
1600 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1602 /* Mirror missing mmap method error code */
1606 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1609 struct msghdr msg = {.msg_flags = flags};
1611 char *kaddr = kmap(page);
1612 iov.iov_base = kaddr + offset;
1614 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1620 * Default Socket Callbacks
1623 static void sock_def_wakeup(struct sock *sk)
1625 read_lock(&sk->sk_callback_lock);
1626 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1627 wake_up_interruptible_all(sk->sk_sleep);
1628 read_unlock(&sk->sk_callback_lock);
1631 static void sock_def_error_report(struct sock *sk)
1633 read_lock(&sk->sk_callback_lock);
1634 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1635 wake_up_interruptible(sk->sk_sleep);
1636 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1637 read_unlock(&sk->sk_callback_lock);
1640 static void sock_def_readable(struct sock *sk, int len)
1642 read_lock(&sk->sk_callback_lock);
1643 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1644 wake_up_interruptible_sync(sk->sk_sleep);
1645 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1646 read_unlock(&sk->sk_callback_lock);
1649 static void sock_def_write_space(struct sock *sk)
1651 read_lock(&sk->sk_callback_lock);
1653 /* Do not wake up a writer until he can make "significant"
1656 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1657 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1658 wake_up_interruptible_sync(sk->sk_sleep);
1660 /* Should agree with poll, otherwise some programs break */
1661 if (sock_writeable(sk))
1662 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1665 read_unlock(&sk->sk_callback_lock);
1668 static void sock_def_destruct(struct sock *sk)
1670 kfree(sk->sk_protinfo);
1673 void sk_send_sigurg(struct sock *sk)
1675 if (sk->sk_socket && sk->sk_socket->file)
1676 if (send_sigurg(&sk->sk_socket->file->f_owner))
1677 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1680 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1681 unsigned long expires)
1683 if (!mod_timer(timer, expires))
1687 EXPORT_SYMBOL(sk_reset_timer);
1689 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1691 if (timer_pending(timer) && del_timer(timer))
1695 EXPORT_SYMBOL(sk_stop_timer);
1697 void sock_init_data(struct socket *sock, struct sock *sk)
1699 skb_queue_head_init(&sk->sk_receive_queue);
1700 skb_queue_head_init(&sk->sk_write_queue);
1701 skb_queue_head_init(&sk->sk_error_queue);
1702 #ifdef CONFIG_NET_DMA
1703 skb_queue_head_init(&sk->sk_async_wait_queue);
1706 sk->sk_send_head = NULL;
1708 init_timer(&sk->sk_timer);
1710 sk->sk_allocation = GFP_KERNEL;
1711 sk->sk_rcvbuf = sysctl_rmem_default;
1712 sk->sk_sndbuf = sysctl_wmem_default;
1713 sk->sk_state = TCP_CLOSE;
1714 sk_set_socket(sk, sock);
1716 sock_set_flag(sk, SOCK_ZAPPED);
1719 sk->sk_type = sock->type;
1720 sk->sk_sleep = &sock->wait;
1723 sk->sk_sleep = NULL;
1725 rwlock_init(&sk->sk_dst_lock);
1726 rwlock_init(&sk->sk_callback_lock);
1727 lockdep_set_class_and_name(&sk->sk_callback_lock,
1728 af_callback_keys + sk->sk_family,
1729 af_family_clock_key_strings[sk->sk_family]);
1731 sk->sk_state_change = sock_def_wakeup;
1732 sk->sk_data_ready = sock_def_readable;
1733 sk->sk_write_space = sock_def_write_space;
1734 sk->sk_error_report = sock_def_error_report;
1735 sk->sk_destruct = sock_def_destruct;
1737 sk->sk_sndmsg_page = NULL;
1738 sk->sk_sndmsg_off = 0;
1740 sk->sk_peercred.pid = 0;
1741 sk->sk_peercred.uid = -1;
1742 sk->sk_peercred.gid = -1;
1743 sk->sk_write_pending = 0;
1744 sk->sk_rcvlowat = 1;
1745 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1746 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1748 sk->sk_stamp = ktime_set(-1L, 0);
1750 atomic_set(&sk->sk_refcnt, 1);
1751 atomic_set(&sk->sk_drops, 0);
1754 void lock_sock_nested(struct sock *sk, int subclass)
1757 spin_lock_bh(&sk->sk_lock.slock);
1758 if (sk->sk_lock.owned)
1760 sk->sk_lock.owned = 1;
1761 spin_unlock(&sk->sk_lock.slock);
1763 * The sk_lock has mutex_lock() semantics here:
1765 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1769 EXPORT_SYMBOL(lock_sock_nested);
1771 void release_sock(struct sock *sk)
1774 * The sk_lock has mutex_unlock() semantics:
1776 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1778 spin_lock_bh(&sk->sk_lock.slock);
1779 if (sk->sk_backlog.tail)
1781 sk->sk_lock.owned = 0;
1782 if (waitqueue_active(&sk->sk_lock.wq))
1783 wake_up(&sk->sk_lock.wq);
1784 spin_unlock_bh(&sk->sk_lock.slock);
1786 EXPORT_SYMBOL(release_sock);
1788 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1791 if (!sock_flag(sk, SOCK_TIMESTAMP))
1792 sock_enable_timestamp(sk);
1793 tv = ktime_to_timeval(sk->sk_stamp);
1794 if (tv.tv_sec == -1)
1796 if (tv.tv_sec == 0) {
1797 sk->sk_stamp = ktime_get_real();
1798 tv = ktime_to_timeval(sk->sk_stamp);
1800 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1802 EXPORT_SYMBOL(sock_get_timestamp);
1804 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1807 if (!sock_flag(sk, SOCK_TIMESTAMP))
1808 sock_enable_timestamp(sk);
1809 ts = ktime_to_timespec(sk->sk_stamp);
1810 if (ts.tv_sec == -1)
1812 if (ts.tv_sec == 0) {
1813 sk->sk_stamp = ktime_get_real();
1814 ts = ktime_to_timespec(sk->sk_stamp);
1816 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1818 EXPORT_SYMBOL(sock_get_timestampns);
1820 void sock_enable_timestamp(struct sock *sk)
1822 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1823 sock_set_flag(sk, SOCK_TIMESTAMP);
1824 net_enable_timestamp();
1829 * Get a socket option on an socket.
1831 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1832 * asynchronous errors should be reported by getsockopt. We assume
1833 * this means if you specify SO_ERROR (otherwise whats the point of it).
1835 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1836 char __user *optval, int __user *optlen)
1838 struct sock *sk = sock->sk;
1840 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1843 EXPORT_SYMBOL(sock_common_getsockopt);
1845 #ifdef CONFIG_COMPAT
1846 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1847 char __user *optval, int __user *optlen)
1849 struct sock *sk = sock->sk;
1851 if (sk->sk_prot->compat_getsockopt != NULL)
1852 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1854 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1856 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1859 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1860 struct msghdr *msg, size_t size, int flags)
1862 struct sock *sk = sock->sk;
1866 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1867 flags & ~MSG_DONTWAIT, &addr_len);
1869 msg->msg_namelen = addr_len;
1873 EXPORT_SYMBOL(sock_common_recvmsg);
1876 * Set socket options on an inet socket.
1878 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1879 char __user *optval, int optlen)
1881 struct sock *sk = sock->sk;
1883 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1886 EXPORT_SYMBOL(sock_common_setsockopt);
1888 #ifdef CONFIG_COMPAT
1889 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1890 char __user *optval, int optlen)
1892 struct sock *sk = sock->sk;
1894 if (sk->sk_prot->compat_setsockopt != NULL)
1895 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1897 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1899 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1902 void sk_common_release(struct sock *sk)
1904 if (sk->sk_prot->destroy)
1905 sk->sk_prot->destroy(sk);
1908 * Observation: when sock_common_release is called, processes have
1909 * no access to socket. But net still has.
1910 * Step one, detach it from networking:
1912 * A. Remove from hash tables.
1915 sk->sk_prot->unhash(sk);
1918 * In this point socket cannot receive new packets, but it is possible
1919 * that some packets are in flight because some CPU runs receiver and
1920 * did hash table lookup before we unhashed socket. They will achieve
1921 * receive queue and will be purged by socket destructor.
1923 * Also we still have packets pending on receive queue and probably,
1924 * our own packets waiting in device queues. sock_destroy will drain
1925 * receive queue, but transmitted packets will delay socket destruction
1926 * until the last reference will be released.
1931 xfrm_sk_free_policy(sk);
1933 sk_refcnt_debug_release(sk);
1937 EXPORT_SYMBOL(sk_common_release);
1939 static DEFINE_RWLOCK(proto_list_lock);
1940 static LIST_HEAD(proto_list);
1942 #ifdef CONFIG_PROC_FS
1943 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1945 int val[PROTO_INUSE_NR];
1948 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1950 #ifdef CONFIG_NET_NS
1951 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1953 int cpu = smp_processor_id();
1954 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1956 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1958 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1960 int cpu, idx = prot->inuse_idx;
1963 for_each_possible_cpu(cpu)
1964 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1966 return res >= 0 ? res : 0;
1968 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1970 static int sock_inuse_init_net(struct net *net)
1972 net->core.inuse = alloc_percpu(struct prot_inuse);
1973 return net->core.inuse ? 0 : -ENOMEM;
1976 static void sock_inuse_exit_net(struct net *net)
1978 free_percpu(net->core.inuse);
1981 static struct pernet_operations net_inuse_ops = {
1982 .init = sock_inuse_init_net,
1983 .exit = sock_inuse_exit_net,
1986 static __init int net_inuse_init(void)
1988 if (register_pernet_subsys(&net_inuse_ops))
1989 panic("Cannot initialize net inuse counters");
1994 core_initcall(net_inuse_init);
1996 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1998 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2000 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2002 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2004 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2006 int cpu, idx = prot->inuse_idx;
2009 for_each_possible_cpu(cpu)
2010 res += per_cpu(prot_inuse, cpu).val[idx];
2012 return res >= 0 ? res : 0;
2014 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2017 static void assign_proto_idx(struct proto *prot)
2019 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2021 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2022 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2026 set_bit(prot->inuse_idx, proto_inuse_idx);
2029 static void release_proto_idx(struct proto *prot)
2031 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2032 clear_bit(prot->inuse_idx, proto_inuse_idx);
2035 static inline void assign_proto_idx(struct proto *prot)
2039 static inline void release_proto_idx(struct proto *prot)
2044 int proto_register(struct proto *prot, int alloc_slab)
2047 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2048 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2051 if (prot->slab == NULL) {
2052 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2057 if (prot->rsk_prot != NULL) {
2058 static const char mask[] = "request_sock_%s";
2060 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2061 if (prot->rsk_prot->slab_name == NULL)
2062 goto out_free_sock_slab;
2064 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2065 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2066 prot->rsk_prot->obj_size, 0,
2067 SLAB_HWCACHE_ALIGN, NULL);
2069 if (prot->rsk_prot->slab == NULL) {
2070 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2072 goto out_free_request_sock_slab_name;
2076 if (prot->twsk_prot != NULL) {
2077 static const char mask[] = "tw_sock_%s";
2079 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2081 if (prot->twsk_prot->twsk_slab_name == NULL)
2082 goto out_free_request_sock_slab;
2084 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2085 prot->twsk_prot->twsk_slab =
2086 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2087 prot->twsk_prot->twsk_obj_size,
2089 SLAB_HWCACHE_ALIGN |
2092 if (prot->twsk_prot->twsk_slab == NULL)
2093 goto out_free_timewait_sock_slab_name;
2097 write_lock(&proto_list_lock);
2098 list_add(&prot->node, &proto_list);
2099 assign_proto_idx(prot);
2100 write_unlock(&proto_list_lock);
2103 out_free_timewait_sock_slab_name:
2104 kfree(prot->twsk_prot->twsk_slab_name);
2105 out_free_request_sock_slab:
2106 if (prot->rsk_prot && prot->rsk_prot->slab) {
2107 kmem_cache_destroy(prot->rsk_prot->slab);
2108 prot->rsk_prot->slab = NULL;
2110 out_free_request_sock_slab_name:
2111 kfree(prot->rsk_prot->slab_name);
2113 kmem_cache_destroy(prot->slab);
2119 EXPORT_SYMBOL(proto_register);
2121 void proto_unregister(struct proto *prot)
2123 write_lock(&proto_list_lock);
2124 release_proto_idx(prot);
2125 list_del(&prot->node);
2126 write_unlock(&proto_list_lock);
2128 if (prot->slab != NULL) {
2129 kmem_cache_destroy(prot->slab);
2133 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2134 kmem_cache_destroy(prot->rsk_prot->slab);
2135 kfree(prot->rsk_prot->slab_name);
2136 prot->rsk_prot->slab = NULL;
2139 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2140 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2141 kfree(prot->twsk_prot->twsk_slab_name);
2142 prot->twsk_prot->twsk_slab = NULL;
2146 EXPORT_SYMBOL(proto_unregister);
2148 #ifdef CONFIG_PROC_FS
2149 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2150 __acquires(proto_list_lock)
2152 read_lock(&proto_list_lock);
2153 return seq_list_start_head(&proto_list, *pos);
2156 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2158 return seq_list_next(v, &proto_list, pos);
2161 static void proto_seq_stop(struct seq_file *seq, void *v)
2162 __releases(proto_list_lock)
2164 read_unlock(&proto_list_lock);
2167 static char proto_method_implemented(const void *method)
2169 return method == NULL ? 'n' : 'y';
2172 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2174 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2175 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2178 sock_prot_inuse_get(seq_file_net(seq), proto),
2179 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2180 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2182 proto->slab == NULL ? "no" : "yes",
2183 module_name(proto->owner),
2184 proto_method_implemented(proto->close),
2185 proto_method_implemented(proto->connect),
2186 proto_method_implemented(proto->disconnect),
2187 proto_method_implemented(proto->accept),
2188 proto_method_implemented(proto->ioctl),
2189 proto_method_implemented(proto->init),
2190 proto_method_implemented(proto->destroy),
2191 proto_method_implemented(proto->shutdown),
2192 proto_method_implemented(proto->setsockopt),
2193 proto_method_implemented(proto->getsockopt),
2194 proto_method_implemented(proto->sendmsg),
2195 proto_method_implemented(proto->recvmsg),
2196 proto_method_implemented(proto->sendpage),
2197 proto_method_implemented(proto->bind),
2198 proto_method_implemented(proto->backlog_rcv),
2199 proto_method_implemented(proto->hash),
2200 proto_method_implemented(proto->unhash),
2201 proto_method_implemented(proto->get_port),
2202 proto_method_implemented(proto->enter_memory_pressure));
2205 static int proto_seq_show(struct seq_file *seq, void *v)
2207 if (v == &proto_list)
2208 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2217 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2219 proto_seq_printf(seq, list_entry(v, struct proto, node));
2223 static const struct seq_operations proto_seq_ops = {
2224 .start = proto_seq_start,
2225 .next = proto_seq_next,
2226 .stop = proto_seq_stop,
2227 .show = proto_seq_show,
2230 static int proto_seq_open(struct inode *inode, struct file *file)
2232 return seq_open_net(inode, file, &proto_seq_ops,
2233 sizeof(struct seq_net_private));
2236 static const struct file_operations proto_seq_fops = {
2237 .owner = THIS_MODULE,
2238 .open = proto_seq_open,
2240 .llseek = seq_lseek,
2241 .release = seq_release_net,
2244 static __net_init int proto_init_net(struct net *net)
2246 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2252 static __net_exit void proto_exit_net(struct net *net)
2254 proc_net_remove(net, "protocols");
2258 static __net_initdata struct pernet_operations proto_net_ops = {
2259 .init = proto_init_net,
2260 .exit = proto_exit_net,
2263 static int __init proto_init(void)
2265 return register_pernet_subsys(&proto_net_ops);
2268 subsys_initcall(proto_init);
2270 #endif /* PROC_FS */
2272 EXPORT_SYMBOL(sk_alloc);
2273 EXPORT_SYMBOL(sk_free);
2274 EXPORT_SYMBOL(sk_send_sigurg);
2275 EXPORT_SYMBOL(sock_alloc_send_skb);
2276 EXPORT_SYMBOL(sock_init_data);
2277 EXPORT_SYMBOL(sock_kfree_s);
2278 EXPORT_SYMBOL(sock_kmalloc);
2279 EXPORT_SYMBOL(sock_no_accept);
2280 EXPORT_SYMBOL(sock_no_bind);
2281 EXPORT_SYMBOL(sock_no_connect);
2282 EXPORT_SYMBOL(sock_no_getname);
2283 EXPORT_SYMBOL(sock_no_getsockopt);
2284 EXPORT_SYMBOL(sock_no_ioctl);
2285 EXPORT_SYMBOL(sock_no_listen);
2286 EXPORT_SYMBOL(sock_no_mmap);
2287 EXPORT_SYMBOL(sock_no_poll);
2288 EXPORT_SYMBOL(sock_no_recvmsg);
2289 EXPORT_SYMBOL(sock_no_sendmsg);
2290 EXPORT_SYMBOL(sock_no_sendpage);
2291 EXPORT_SYMBOL(sock_no_setsockopt);
2292 EXPORT_SYMBOL(sock_no_shutdown);
2293 EXPORT_SYMBOL(sock_no_socketpair);
2294 EXPORT_SYMBOL(sock_rfree);
2295 EXPORT_SYMBOL(sock_setsockopt);
2296 EXPORT_SYMBOL(sock_wfree);
2297 EXPORT_SYMBOL(sock_wmalloc);
2298 EXPORT_SYMBOL(sock_i_uid);
2299 EXPORT_SYMBOL(sock_i_ino);
2300 EXPORT_SYMBOL(sysctl_optmem_max);