2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
128 #include <linux/filter.h>
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
138 static struct lock_class_key af_family_keys[AF_MAX];
139 static struct lock_class_key af_family_slock_keys[AF_MAX];
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
147 static const char *af_family_key_strings[AF_MAX+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
171 "slock-27" , "slock-28" , "slock-AF_CAN" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-29" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_MAX"
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
195 static struct lock_class_key af_callback_keys[AF_MAX];
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
220 if (optlen < sizeof(tv))
222 if (copy_from_user(&tv, optval, sizeof(tv)))
224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
228 static int warned __read_mostly;
231 if (warned < 10 && net_ratelimit()) {
233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
235 current->comm, task_pid_nr(current));
239 *timeo_p = MAX_SCHEDULE_TIMEOUT;
240 if (tv.tv_sec == 0 && tv.tv_usec == 0)
242 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
243 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
247 static void sock_warn_obsolete_bsdism(const char *name)
250 static char warncomm[TASK_COMM_LEN];
251 if (strcmp(warncomm, current->comm) && warned < 5) {
252 strcpy(warncomm, current->comm);
253 printk(KERN_WARNING "process `%s' is using obsolete "
254 "%s SO_BSDCOMPAT\n", warncomm, name);
259 static void sock_disable_timestamp(struct sock *sk)
261 if (sock_flag(sk, SOCK_TIMESTAMP)) {
262 sock_reset_flag(sk, SOCK_TIMESTAMP);
263 net_disable_timestamp();
268 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
273 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
274 number of warnings when compiling with -W --ANK
276 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
277 (unsigned)sk->sk_rcvbuf) {
282 err = sk_filter(sk, skb);
286 if (!sk_rmem_schedule(sk, skb->truesize)) {
292 skb_set_owner_r(skb, sk);
294 /* Cache the SKB length before we tack it onto the receive
295 * queue. Once it is added it no longer belongs to us and
296 * may be freed by other threads of control pulling packets
301 skb_queue_tail(&sk->sk_receive_queue, skb);
303 if (!sock_flag(sk, SOCK_DEAD))
304 sk->sk_data_ready(sk, skb_len);
308 EXPORT_SYMBOL(sock_queue_rcv_skb);
310 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
312 int rc = NET_RX_SUCCESS;
314 if (sk_filter(sk, skb))
315 goto discard_and_relse;
320 bh_lock_sock_nested(sk);
323 if (!sock_owned_by_user(sk)) {
325 * trylock + unlock semantics:
327 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
329 rc = sk->sk_backlog_rcv(sk, skb);
331 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
333 sk_add_backlog(sk, skb);
342 EXPORT_SYMBOL(sk_receive_skb);
344 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
346 struct dst_entry *dst = sk->sk_dst_cache;
348 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
349 sk->sk_dst_cache = NULL;
356 EXPORT_SYMBOL(__sk_dst_check);
358 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
360 struct dst_entry *dst = sk_dst_get(sk);
362 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
370 EXPORT_SYMBOL(sk_dst_check);
372 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
374 int ret = -ENOPROTOOPT;
375 #ifdef CONFIG_NETDEVICES
376 struct net *net = sock_net(sk);
377 char devname[IFNAMSIZ];
382 if (!capable(CAP_NET_RAW))
389 /* Bind this socket to a particular device like "eth0",
390 * as specified in the passed interface name. If the
391 * name is "" or the option length is zero the socket
394 if (optlen > IFNAMSIZ - 1)
395 optlen = IFNAMSIZ - 1;
396 memset(devname, 0, sizeof(devname));
399 if (copy_from_user(devname, optval, optlen))
402 if (devname[0] == '\0') {
405 struct net_device *dev = dev_get_by_name(net, devname);
411 index = dev->ifindex;
416 sk->sk_bound_dev_if = index;
428 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
431 sock_set_flag(sk, bit);
433 sock_reset_flag(sk, bit);
437 * This is meant for all protocols to use and covers goings on
438 * at the socket level. Everything here is generic.
441 int sock_setsockopt(struct socket *sock, int level, int optname,
442 char __user *optval, int optlen)
444 struct sock *sk=sock->sk;
451 * Options without arguments
454 if (optname == SO_BINDTODEVICE)
455 return sock_bindtodevice(sk, optval, optlen);
457 if (optlen < sizeof(int))
460 if (get_user(val, (int __user *)optval))
469 if (val && !capable(CAP_NET_ADMIN)) {
472 sock_valbool_flag(sk, SOCK_DBG, valbool);
475 sk->sk_reuse = valbool;
482 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
485 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
488 /* Don't error on this BSD doesn't and if you think
489 about it this is right. Otherwise apps have to
490 play 'guess the biggest size' games. RCVBUF/SNDBUF
491 are treated in BSD as hints */
493 if (val > sysctl_wmem_max)
494 val = sysctl_wmem_max;
496 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
497 if ((val * 2) < SOCK_MIN_SNDBUF)
498 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
500 sk->sk_sndbuf = val * 2;
503 * Wake up sending tasks if we
506 sk->sk_write_space(sk);
510 if (!capable(CAP_NET_ADMIN)) {
517 /* Don't error on this BSD doesn't and if you think
518 about it this is right. Otherwise apps have to
519 play 'guess the biggest size' games. RCVBUF/SNDBUF
520 are treated in BSD as hints */
522 if (val > sysctl_rmem_max)
523 val = sysctl_rmem_max;
525 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
527 * We double it on the way in to account for
528 * "struct sk_buff" etc. overhead. Applications
529 * assume that the SO_RCVBUF setting they make will
530 * allow that much actual data to be received on that
533 * Applications are unaware that "struct sk_buff" and
534 * other overheads allocate from the receive buffer
535 * during socket buffer allocation.
537 * And after considering the possible alternatives,
538 * returning the value we actually used in getsockopt
539 * is the most desirable behavior.
541 if ((val * 2) < SOCK_MIN_RCVBUF)
542 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
544 sk->sk_rcvbuf = val * 2;
548 if (!capable(CAP_NET_ADMIN)) {
556 if (sk->sk_protocol == IPPROTO_TCP)
557 tcp_set_keepalive(sk, valbool);
559 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
563 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
567 sk->sk_no_check = valbool;
571 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
572 sk->sk_priority = val;
578 if (optlen < sizeof(ling)) {
579 ret = -EINVAL; /* 1003.1g */
582 if (copy_from_user(&ling,optval,sizeof(ling))) {
587 sock_reset_flag(sk, SOCK_LINGER);
589 #if (BITS_PER_LONG == 32)
590 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
591 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
594 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
595 sock_set_flag(sk, SOCK_LINGER);
600 sock_warn_obsolete_bsdism("setsockopt");
605 set_bit(SOCK_PASSCRED, &sock->flags);
607 clear_bit(SOCK_PASSCRED, &sock->flags);
613 if (optname == SO_TIMESTAMP)
614 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
616 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
617 sock_set_flag(sk, SOCK_RCVTSTAMP);
618 sock_enable_timestamp(sk);
620 sock_reset_flag(sk, SOCK_RCVTSTAMP);
621 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
628 sk->sk_rcvlowat = val ? : 1;
632 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
636 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
639 case SO_ATTACH_FILTER:
641 if (optlen == sizeof(struct sock_fprog)) {
642 struct sock_fprog fprog;
645 if (copy_from_user(&fprog, optval, sizeof(fprog)))
648 ret = sk_attach_filter(&fprog, sk);
652 case SO_DETACH_FILTER:
653 ret = sk_detach_filter(sk);
658 set_bit(SOCK_PASSSEC, &sock->flags);
660 clear_bit(SOCK_PASSSEC, &sock->flags);
663 if (!capable(CAP_NET_ADMIN))
670 /* We implement the SO_SNDLOWAT etc to
671 not be settable (1003.1g 5.3) */
681 int sock_getsockopt(struct socket *sock, int level, int optname,
682 char __user *optval, int __user *optlen)
684 struct sock *sk = sock->sk;
692 unsigned int lv = sizeof(int);
695 if (get_user(len, optlen))
702 v.val = sock_flag(sk, SOCK_DBG);
706 v.val = sock_flag(sk, SOCK_LOCALROUTE);
710 v.val = !!sock_flag(sk, SOCK_BROADCAST);
714 v.val = sk->sk_sndbuf;
718 v.val = sk->sk_rcvbuf;
722 v.val = sk->sk_reuse;
726 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
734 v.val = -sock_error(sk);
736 v.val = xchg(&sk->sk_err_soft, 0);
740 v.val = !!sock_flag(sk, SOCK_URGINLINE);
744 v.val = sk->sk_no_check;
748 v.val = sk->sk_priority;
753 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
754 v.ling.l_linger = sk->sk_lingertime / HZ;
758 sock_warn_obsolete_bsdism("getsockopt");
762 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
763 !sock_flag(sk, SOCK_RCVTSTAMPNS);
767 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
771 lv=sizeof(struct timeval);
772 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
776 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
777 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
782 lv=sizeof(struct timeval);
783 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
787 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
788 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
793 v.val = sk->sk_rcvlowat;
801 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
805 if (len > sizeof(sk->sk_peercred))
806 len = sizeof(sk->sk_peercred);
807 if (copy_to_user(optval, &sk->sk_peercred, len))
815 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
819 if (copy_to_user(optval, address, len))
824 /* Dubious BSD thing... Probably nobody even uses it, but
825 * the UNIX standard wants it for whatever reason... -DaveM
828 v.val = sk->sk_state == TCP_LISTEN;
832 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
836 return security_socket_getpeersec_stream(sock, optval, optlen, len);
848 if (copy_to_user(optval, &v, len))
851 if (put_user(len, optlen))
857 * Initialize an sk_lock.
859 * (We also register the sk_lock with the lock validator.)
861 static inline void sock_lock_init(struct sock *sk)
863 sock_lock_init_class_and_name(sk,
864 af_family_slock_key_strings[sk->sk_family],
865 af_family_slock_keys + sk->sk_family,
866 af_family_key_strings[sk->sk_family],
867 af_family_keys + sk->sk_family);
870 static void sock_copy(struct sock *nsk, const struct sock *osk)
872 #ifdef CONFIG_SECURITY_NETWORK
873 void *sptr = nsk->sk_security;
876 memcpy(nsk, osk, osk->sk_prot->obj_size);
877 #ifdef CONFIG_SECURITY_NETWORK
878 nsk->sk_security = sptr;
879 security_sk_clone(osk, nsk);
883 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
887 struct kmem_cache *slab;
891 sk = kmem_cache_alloc(slab, priority);
893 sk = kmalloc(prot->obj_size, priority);
896 if (security_sk_alloc(sk, family, priority))
899 if (!try_module_get(prot->owner))
906 security_sk_free(sk);
909 kmem_cache_free(slab, sk);
915 static void sk_prot_free(struct proto *prot, struct sock *sk)
917 struct kmem_cache *slab;
918 struct module *owner;
923 security_sk_free(sk);
925 kmem_cache_free(slab, sk);
932 * sk_alloc - All socket objects are allocated here
933 * @net: the applicable net namespace
934 * @family: protocol family
935 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
936 * @prot: struct proto associated with this new sock instance
938 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
943 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
945 sk->sk_family = family;
947 * See comment in struct sock definition to understand
948 * why we need sk_prot_creator -acme
950 sk->sk_prot = sk->sk_prot_creator = prot;
952 sock_net_set(sk, get_net(net));
958 void sk_free(struct sock *sk)
960 struct sk_filter *filter;
965 filter = rcu_dereference(sk->sk_filter);
967 sk_filter_uncharge(sk, filter);
968 rcu_assign_pointer(sk->sk_filter, NULL);
971 sock_disable_timestamp(sk);
973 if (atomic_read(&sk->sk_omem_alloc))
974 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
975 __func__, atomic_read(&sk->sk_omem_alloc));
977 put_net(sock_net(sk));
978 sk_prot_free(sk->sk_prot_creator, sk);
982 * Last sock_put should drop referrence to sk->sk_net. It has already
983 * been dropped in sk_change_net. Taking referrence to stopping namespace
985 * Take referrence to a socket to remove it from hash _alive_ and after that
986 * destroy it in the context of init_net.
988 void sk_release_kernel(struct sock *sk)
990 if (sk == NULL || sk->sk_socket == NULL)
994 sock_release(sk->sk_socket);
995 release_net(sock_net(sk));
996 sock_net_set(sk, get_net(&init_net));
999 EXPORT_SYMBOL(sk_release_kernel);
1001 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1005 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1006 if (newsk != NULL) {
1007 struct sk_filter *filter;
1009 sock_copy(newsk, sk);
1012 get_net(sock_net(newsk));
1013 sk_node_init(&newsk->sk_node);
1014 sock_lock_init(newsk);
1015 bh_lock_sock(newsk);
1016 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1018 atomic_set(&newsk->sk_rmem_alloc, 0);
1019 atomic_set(&newsk->sk_wmem_alloc, 0);
1020 atomic_set(&newsk->sk_omem_alloc, 0);
1021 skb_queue_head_init(&newsk->sk_receive_queue);
1022 skb_queue_head_init(&newsk->sk_write_queue);
1023 #ifdef CONFIG_NET_DMA
1024 skb_queue_head_init(&newsk->sk_async_wait_queue);
1027 rwlock_init(&newsk->sk_dst_lock);
1028 rwlock_init(&newsk->sk_callback_lock);
1029 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1030 af_callback_keys + newsk->sk_family,
1031 af_family_clock_key_strings[newsk->sk_family]);
1033 newsk->sk_dst_cache = NULL;
1034 newsk->sk_wmem_queued = 0;
1035 newsk->sk_forward_alloc = 0;
1036 newsk->sk_send_head = NULL;
1037 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1039 sock_reset_flag(newsk, SOCK_DONE);
1040 skb_queue_head_init(&newsk->sk_error_queue);
1042 filter = newsk->sk_filter;
1044 sk_filter_charge(newsk, filter);
1046 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1047 /* It is still raw copy of parent, so invalidate
1048 * destructor and make plain sk_free() */
1049 newsk->sk_destruct = NULL;
1056 newsk->sk_priority = 0;
1057 atomic_set(&newsk->sk_refcnt, 2);
1060 * Increment the counter in the same struct proto as the master
1061 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1062 * is the same as sk->sk_prot->socks, as this field was copied
1065 * This _changes_ the previous behaviour, where
1066 * tcp_create_openreq_child always was incrementing the
1067 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1068 * to be taken into account in all callers. -acme
1070 sk_refcnt_debug_inc(newsk);
1071 newsk->sk_socket = NULL;
1072 newsk->sk_sleep = NULL;
1074 if (newsk->sk_prot->sockets_allocated)
1075 atomic_inc(newsk->sk_prot->sockets_allocated);
1081 EXPORT_SYMBOL_GPL(sk_clone);
1083 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1085 __sk_dst_set(sk, dst);
1086 sk->sk_route_caps = dst->dev->features;
1087 if (sk->sk_route_caps & NETIF_F_GSO)
1088 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1089 if (sk_can_gso(sk)) {
1090 if (dst->header_len) {
1091 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1093 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1094 sk->sk_gso_max_size = dst->dev->gso_max_size;
1098 EXPORT_SYMBOL_GPL(sk_setup_caps);
1100 void __init sk_init(void)
1102 if (num_physpages <= 4096) {
1103 sysctl_wmem_max = 32767;
1104 sysctl_rmem_max = 32767;
1105 sysctl_wmem_default = 32767;
1106 sysctl_rmem_default = 32767;
1107 } else if (num_physpages >= 131072) {
1108 sysctl_wmem_max = 131071;
1109 sysctl_rmem_max = 131071;
1114 * Simple resource managers for sockets.
1119 * Write buffer destructor automatically called from kfree_skb.
1121 void sock_wfree(struct sk_buff *skb)
1123 struct sock *sk = skb->sk;
1125 /* In case it might be waiting for more memory. */
1126 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1127 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1128 sk->sk_write_space(sk);
1133 * Read buffer destructor automatically called from kfree_skb.
1135 void sock_rfree(struct sk_buff *skb)
1137 struct sock *sk = skb->sk;
1139 skb_truesize_check(skb);
1140 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1141 sk_mem_uncharge(skb->sk, skb->truesize);
1145 int sock_i_uid(struct sock *sk)
1149 read_lock(&sk->sk_callback_lock);
1150 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1151 read_unlock(&sk->sk_callback_lock);
1155 unsigned long sock_i_ino(struct sock *sk)
1159 read_lock(&sk->sk_callback_lock);
1160 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1161 read_unlock(&sk->sk_callback_lock);
1166 * Allocate a skb from the socket's send buffer.
1168 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1171 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1172 struct sk_buff * skb = alloc_skb(size, priority);
1174 skb_set_owner_w(skb, sk);
1182 * Allocate a skb from the socket's receive buffer.
1184 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1187 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1188 struct sk_buff *skb = alloc_skb(size, priority);
1190 skb_set_owner_r(skb, sk);
1198 * Allocate a memory block from the socket's option memory buffer.
1200 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1202 if ((unsigned)size <= sysctl_optmem_max &&
1203 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1205 /* First do the add, to avoid the race if kmalloc
1208 atomic_add(size, &sk->sk_omem_alloc);
1209 mem = kmalloc(size, priority);
1212 atomic_sub(size, &sk->sk_omem_alloc);
1218 * Free an option memory block.
1220 void sock_kfree_s(struct sock *sk, void *mem, int size)
1223 atomic_sub(size, &sk->sk_omem_alloc);
1226 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1227 I think, these locks should be removed for datagram sockets.
1229 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1233 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1237 if (signal_pending(current))
1239 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1240 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1241 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1243 if (sk->sk_shutdown & SEND_SHUTDOWN)
1247 timeo = schedule_timeout(timeo);
1249 finish_wait(sk->sk_sleep, &wait);
1255 * Generic send/receive buffer handlers
1258 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1259 unsigned long header_len,
1260 unsigned long data_len,
1261 int noblock, int *errcode)
1263 struct sk_buff *skb;
1268 gfp_mask = sk->sk_allocation;
1269 if (gfp_mask & __GFP_WAIT)
1270 gfp_mask |= __GFP_REPEAT;
1272 timeo = sock_sndtimeo(sk, noblock);
1274 err = sock_error(sk);
1279 if (sk->sk_shutdown & SEND_SHUTDOWN)
1282 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1283 skb = alloc_skb(header_len, gfp_mask);
1288 /* No pages, we're done... */
1292 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1293 skb->truesize += data_len;
1294 skb_shinfo(skb)->nr_frags = npages;
1295 for (i = 0; i < npages; i++) {
1299 page = alloc_pages(sk->sk_allocation, 0);
1302 skb_shinfo(skb)->nr_frags = i;
1307 frag = &skb_shinfo(skb)->frags[i];
1309 frag->page_offset = 0;
1310 frag->size = (data_len >= PAGE_SIZE ?
1313 data_len -= PAGE_SIZE;
1316 /* Full success... */
1322 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1323 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1327 if (signal_pending(current))
1329 timeo = sock_wait_for_wmem(sk, timeo);
1332 skb_set_owner_w(skb, sk);
1336 err = sock_intr_errno(timeo);
1342 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1343 int noblock, int *errcode)
1345 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1348 static void __lock_sock(struct sock *sk)
1353 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1354 TASK_UNINTERRUPTIBLE);
1355 spin_unlock_bh(&sk->sk_lock.slock);
1357 spin_lock_bh(&sk->sk_lock.slock);
1358 if (!sock_owned_by_user(sk))
1361 finish_wait(&sk->sk_lock.wq, &wait);
1364 static void __release_sock(struct sock *sk)
1366 struct sk_buff *skb = sk->sk_backlog.head;
1369 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1373 struct sk_buff *next = skb->next;
1376 sk->sk_backlog_rcv(sk, skb);
1379 * We are in process context here with softirqs
1380 * disabled, use cond_resched_softirq() to preempt.
1381 * This is safe to do because we've taken the backlog
1384 cond_resched_softirq();
1387 } while (skb != NULL);
1390 } while ((skb = sk->sk_backlog.head) != NULL);
1394 * sk_wait_data - wait for data to arrive at sk_receive_queue
1395 * @sk: sock to wait on
1396 * @timeo: for how long
1398 * Now socket state including sk->sk_err is changed only under lock,
1399 * hence we may omit checks after joining wait queue.
1400 * We check receive queue before schedule() only as optimization;
1401 * it is very likely that release_sock() added new data.
1403 int sk_wait_data(struct sock *sk, long *timeo)
1408 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1409 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1410 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1411 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1412 finish_wait(sk->sk_sleep, &wait);
1416 EXPORT_SYMBOL(sk_wait_data);
1419 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1421 * @size: memory size to allocate
1422 * @kind: allocation type
1424 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1425 * rmem allocation. This function assumes that protocols which have
1426 * memory_pressure use sk_wmem_queued as write buffer accounting.
1428 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1430 struct proto *prot = sk->sk_prot;
1431 int amt = sk_mem_pages(size);
1434 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1435 allocated = atomic_add_return(amt, prot->memory_allocated);
1438 if (allocated <= prot->sysctl_mem[0]) {
1439 if (prot->memory_pressure && *prot->memory_pressure)
1440 *prot->memory_pressure = 0;
1444 /* Under pressure. */
1445 if (allocated > prot->sysctl_mem[1])
1446 if (prot->enter_memory_pressure)
1447 prot->enter_memory_pressure();
1449 /* Over hard limit. */
1450 if (allocated > prot->sysctl_mem[2])
1451 goto suppress_allocation;
1453 /* guarantee minimum buffer size under pressure */
1454 if (kind == SK_MEM_RECV) {
1455 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1457 } else { /* SK_MEM_SEND */
1458 if (sk->sk_type == SOCK_STREAM) {
1459 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1461 } else if (atomic_read(&sk->sk_wmem_alloc) <
1462 prot->sysctl_wmem[0])
1466 if (prot->memory_pressure) {
1467 if (!*prot->memory_pressure ||
1468 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1469 sk_mem_pages(sk->sk_wmem_queued +
1470 atomic_read(&sk->sk_rmem_alloc) +
1471 sk->sk_forward_alloc))
1475 suppress_allocation:
1477 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1478 sk_stream_moderate_sndbuf(sk);
1480 /* Fail only if socket is _under_ its sndbuf.
1481 * In this case we cannot block, so that we have to fail.
1483 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1487 /* Alas. Undo changes. */
1488 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1489 atomic_sub(amt, prot->memory_allocated);
1493 EXPORT_SYMBOL(__sk_mem_schedule);
1496 * __sk_reclaim - reclaim memory_allocated
1499 void __sk_mem_reclaim(struct sock *sk)
1501 struct proto *prot = sk->sk_prot;
1503 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1504 prot->memory_allocated);
1505 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1507 if (prot->memory_pressure && *prot->memory_pressure &&
1508 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1509 *prot->memory_pressure = 0;
1512 EXPORT_SYMBOL(__sk_mem_reclaim);
1516 * Set of default routines for initialising struct proto_ops when
1517 * the protocol does not support a particular function. In certain
1518 * cases where it makes no sense for a protocol to have a "do nothing"
1519 * function, some default processing is provided.
1522 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1527 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1533 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1538 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1543 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1549 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1554 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1559 int sock_no_listen(struct socket *sock, int backlog)
1564 int sock_no_shutdown(struct socket *sock, int how)
1569 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1570 char __user *optval, int optlen)
1575 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1576 char __user *optval, int __user *optlen)
1581 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1587 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1588 size_t len, int flags)
1593 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1595 /* Mirror missing mmap method error code */
1599 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1602 struct msghdr msg = {.msg_flags = flags};
1604 char *kaddr = kmap(page);
1605 iov.iov_base = kaddr + offset;
1607 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1613 * Default Socket Callbacks
1616 static void sock_def_wakeup(struct sock *sk)
1618 read_lock(&sk->sk_callback_lock);
1619 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1620 wake_up_interruptible_all(sk->sk_sleep);
1621 read_unlock(&sk->sk_callback_lock);
1624 static void sock_def_error_report(struct sock *sk)
1626 read_lock(&sk->sk_callback_lock);
1627 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1628 wake_up_interruptible(sk->sk_sleep);
1629 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1630 read_unlock(&sk->sk_callback_lock);
1633 static void sock_def_readable(struct sock *sk, int len)
1635 read_lock(&sk->sk_callback_lock);
1636 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1637 wake_up_interruptible_sync(sk->sk_sleep);
1638 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1639 read_unlock(&sk->sk_callback_lock);
1642 static void sock_def_write_space(struct sock *sk)
1644 read_lock(&sk->sk_callback_lock);
1646 /* Do not wake up a writer until he can make "significant"
1649 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1650 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1651 wake_up_interruptible_sync(sk->sk_sleep);
1653 /* Should agree with poll, otherwise some programs break */
1654 if (sock_writeable(sk))
1655 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1658 read_unlock(&sk->sk_callback_lock);
1661 static void sock_def_destruct(struct sock *sk)
1663 kfree(sk->sk_protinfo);
1666 void sk_send_sigurg(struct sock *sk)
1668 if (sk->sk_socket && sk->sk_socket->file)
1669 if (send_sigurg(&sk->sk_socket->file->f_owner))
1670 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1673 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1674 unsigned long expires)
1676 if (!mod_timer(timer, expires))
1680 EXPORT_SYMBOL(sk_reset_timer);
1682 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1684 if (timer_pending(timer) && del_timer(timer))
1688 EXPORT_SYMBOL(sk_stop_timer);
1690 void sock_init_data(struct socket *sock, struct sock *sk)
1692 skb_queue_head_init(&sk->sk_receive_queue);
1693 skb_queue_head_init(&sk->sk_write_queue);
1694 skb_queue_head_init(&sk->sk_error_queue);
1695 #ifdef CONFIG_NET_DMA
1696 skb_queue_head_init(&sk->sk_async_wait_queue);
1699 sk->sk_send_head = NULL;
1701 init_timer(&sk->sk_timer);
1703 sk->sk_allocation = GFP_KERNEL;
1704 sk->sk_rcvbuf = sysctl_rmem_default;
1705 sk->sk_sndbuf = sysctl_wmem_default;
1706 sk->sk_state = TCP_CLOSE;
1707 sk->sk_socket = sock;
1709 sock_set_flag(sk, SOCK_ZAPPED);
1712 sk->sk_type = sock->type;
1713 sk->sk_sleep = &sock->wait;
1716 sk->sk_sleep = NULL;
1718 rwlock_init(&sk->sk_dst_lock);
1719 rwlock_init(&sk->sk_callback_lock);
1720 lockdep_set_class_and_name(&sk->sk_callback_lock,
1721 af_callback_keys + sk->sk_family,
1722 af_family_clock_key_strings[sk->sk_family]);
1724 sk->sk_state_change = sock_def_wakeup;
1725 sk->sk_data_ready = sock_def_readable;
1726 sk->sk_write_space = sock_def_write_space;
1727 sk->sk_error_report = sock_def_error_report;
1728 sk->sk_destruct = sock_def_destruct;
1730 sk->sk_sndmsg_page = NULL;
1731 sk->sk_sndmsg_off = 0;
1733 sk->sk_peercred.pid = 0;
1734 sk->sk_peercred.uid = -1;
1735 sk->sk_peercred.gid = -1;
1736 sk->sk_write_pending = 0;
1737 sk->sk_rcvlowat = 1;
1738 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1739 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1741 sk->sk_stamp = ktime_set(-1L, 0);
1743 atomic_set(&sk->sk_refcnt, 1);
1744 atomic_set(&sk->sk_drops, 0);
1747 void lock_sock_nested(struct sock *sk, int subclass)
1750 spin_lock_bh(&sk->sk_lock.slock);
1751 if (sk->sk_lock.owned)
1753 sk->sk_lock.owned = 1;
1754 spin_unlock(&sk->sk_lock.slock);
1756 * The sk_lock has mutex_lock() semantics here:
1758 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1762 EXPORT_SYMBOL(lock_sock_nested);
1764 void release_sock(struct sock *sk)
1767 * The sk_lock has mutex_unlock() semantics:
1769 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1771 spin_lock_bh(&sk->sk_lock.slock);
1772 if (sk->sk_backlog.tail)
1774 sk->sk_lock.owned = 0;
1775 if (waitqueue_active(&sk->sk_lock.wq))
1776 wake_up(&sk->sk_lock.wq);
1777 spin_unlock_bh(&sk->sk_lock.slock);
1779 EXPORT_SYMBOL(release_sock);
1781 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1784 if (!sock_flag(sk, SOCK_TIMESTAMP))
1785 sock_enable_timestamp(sk);
1786 tv = ktime_to_timeval(sk->sk_stamp);
1787 if (tv.tv_sec == -1)
1789 if (tv.tv_sec == 0) {
1790 sk->sk_stamp = ktime_get_real();
1791 tv = ktime_to_timeval(sk->sk_stamp);
1793 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1795 EXPORT_SYMBOL(sock_get_timestamp);
1797 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1800 if (!sock_flag(sk, SOCK_TIMESTAMP))
1801 sock_enable_timestamp(sk);
1802 ts = ktime_to_timespec(sk->sk_stamp);
1803 if (ts.tv_sec == -1)
1805 if (ts.tv_sec == 0) {
1806 sk->sk_stamp = ktime_get_real();
1807 ts = ktime_to_timespec(sk->sk_stamp);
1809 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1811 EXPORT_SYMBOL(sock_get_timestampns);
1813 void sock_enable_timestamp(struct sock *sk)
1815 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1816 sock_set_flag(sk, SOCK_TIMESTAMP);
1817 net_enable_timestamp();
1822 * Get a socket option on an socket.
1824 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1825 * asynchronous errors should be reported by getsockopt. We assume
1826 * this means if you specify SO_ERROR (otherwise whats the point of it).
1828 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1829 char __user *optval, int __user *optlen)
1831 struct sock *sk = sock->sk;
1833 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1836 EXPORT_SYMBOL(sock_common_getsockopt);
1838 #ifdef CONFIG_COMPAT
1839 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1840 char __user *optval, int __user *optlen)
1842 struct sock *sk = sock->sk;
1844 if (sk->sk_prot->compat_getsockopt != NULL)
1845 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1847 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1849 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1852 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1853 struct msghdr *msg, size_t size, int flags)
1855 struct sock *sk = sock->sk;
1859 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1860 flags & ~MSG_DONTWAIT, &addr_len);
1862 msg->msg_namelen = addr_len;
1866 EXPORT_SYMBOL(sock_common_recvmsg);
1869 * Set socket options on an inet socket.
1871 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1872 char __user *optval, int optlen)
1874 struct sock *sk = sock->sk;
1876 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1879 EXPORT_SYMBOL(sock_common_setsockopt);
1881 #ifdef CONFIG_COMPAT
1882 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1883 char __user *optval, int optlen)
1885 struct sock *sk = sock->sk;
1887 if (sk->sk_prot->compat_setsockopt != NULL)
1888 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1890 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1892 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1895 void sk_common_release(struct sock *sk)
1897 if (sk->sk_prot->destroy)
1898 sk->sk_prot->destroy(sk);
1901 * Observation: when sock_common_release is called, processes have
1902 * no access to socket. But net still has.
1903 * Step one, detach it from networking:
1905 * A. Remove from hash tables.
1908 sk->sk_prot->unhash(sk);
1911 * In this point socket cannot receive new packets, but it is possible
1912 * that some packets are in flight because some CPU runs receiver and
1913 * did hash table lookup before we unhashed socket. They will achieve
1914 * receive queue and will be purged by socket destructor.
1916 * Also we still have packets pending on receive queue and probably,
1917 * our own packets waiting in device queues. sock_destroy will drain
1918 * receive queue, but transmitted packets will delay socket destruction
1919 * until the last reference will be released.
1924 xfrm_sk_free_policy(sk);
1926 sk_refcnt_debug_release(sk);
1930 EXPORT_SYMBOL(sk_common_release);
1932 static DEFINE_RWLOCK(proto_list_lock);
1933 static LIST_HEAD(proto_list);
1935 #ifdef CONFIG_PROC_FS
1936 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1938 int val[PROTO_INUSE_NR];
1941 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1943 #ifdef CONFIG_NET_NS
1944 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1946 int cpu = smp_processor_id();
1947 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1949 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1951 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1953 int cpu, idx = prot->inuse_idx;
1956 for_each_possible_cpu(cpu)
1957 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1959 return res >= 0 ? res : 0;
1961 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1963 static int sock_inuse_init_net(struct net *net)
1965 net->core.inuse = alloc_percpu(struct prot_inuse);
1966 return net->core.inuse ? 0 : -ENOMEM;
1969 static void sock_inuse_exit_net(struct net *net)
1971 free_percpu(net->core.inuse);
1974 static struct pernet_operations net_inuse_ops = {
1975 .init = sock_inuse_init_net,
1976 .exit = sock_inuse_exit_net,
1979 static __init int net_inuse_init(void)
1981 if (register_pernet_subsys(&net_inuse_ops))
1982 panic("Cannot initialize net inuse counters");
1987 core_initcall(net_inuse_init);
1989 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1991 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1993 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1995 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1997 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1999 int cpu, idx = prot->inuse_idx;
2002 for_each_possible_cpu(cpu)
2003 res += per_cpu(prot_inuse, cpu).val[idx];
2005 return res >= 0 ? res : 0;
2007 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2010 static void assign_proto_idx(struct proto *prot)
2012 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2014 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2015 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2019 set_bit(prot->inuse_idx, proto_inuse_idx);
2022 static void release_proto_idx(struct proto *prot)
2024 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2025 clear_bit(prot->inuse_idx, proto_inuse_idx);
2028 static inline void assign_proto_idx(struct proto *prot)
2032 static inline void release_proto_idx(struct proto *prot)
2037 int proto_register(struct proto *prot, int alloc_slab)
2039 char *request_sock_slab_name = NULL;
2040 char *timewait_sock_slab_name;
2043 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2044 SLAB_HWCACHE_ALIGN, NULL);
2046 if (prot->slab == NULL) {
2047 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2052 if (prot->rsk_prot != NULL) {
2053 static const char mask[] = "request_sock_%s";
2055 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2056 if (request_sock_slab_name == NULL)
2057 goto out_free_sock_slab;
2059 sprintf(request_sock_slab_name, mask, prot->name);
2060 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
2061 prot->rsk_prot->obj_size, 0,
2062 SLAB_HWCACHE_ALIGN, NULL);
2064 if (prot->rsk_prot->slab == NULL) {
2065 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2067 goto out_free_request_sock_slab_name;
2071 if (prot->twsk_prot != NULL) {
2072 static const char mask[] = "tw_sock_%s";
2074 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2076 if (timewait_sock_slab_name == NULL)
2077 goto out_free_request_sock_slab;
2079 sprintf(timewait_sock_slab_name, mask, prot->name);
2080 prot->twsk_prot->twsk_slab =
2081 kmem_cache_create(timewait_sock_slab_name,
2082 prot->twsk_prot->twsk_obj_size,
2083 0, SLAB_HWCACHE_ALIGN,
2085 if (prot->twsk_prot->twsk_slab == NULL)
2086 goto out_free_timewait_sock_slab_name;
2090 write_lock(&proto_list_lock);
2091 list_add(&prot->node, &proto_list);
2092 assign_proto_idx(prot);
2093 write_unlock(&proto_list_lock);
2096 out_free_timewait_sock_slab_name:
2097 kfree(timewait_sock_slab_name);
2098 out_free_request_sock_slab:
2099 if (prot->rsk_prot && prot->rsk_prot->slab) {
2100 kmem_cache_destroy(prot->rsk_prot->slab);
2101 prot->rsk_prot->slab = NULL;
2103 out_free_request_sock_slab_name:
2104 kfree(request_sock_slab_name);
2106 kmem_cache_destroy(prot->slab);
2112 EXPORT_SYMBOL(proto_register);
2114 void proto_unregister(struct proto *prot)
2116 write_lock(&proto_list_lock);
2117 release_proto_idx(prot);
2118 list_del(&prot->node);
2119 write_unlock(&proto_list_lock);
2121 if (prot->slab != NULL) {
2122 kmem_cache_destroy(prot->slab);
2126 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2127 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2129 kmem_cache_destroy(prot->rsk_prot->slab);
2131 prot->rsk_prot->slab = NULL;
2134 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2135 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
2137 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2139 prot->twsk_prot->twsk_slab = NULL;
2143 EXPORT_SYMBOL(proto_unregister);
2145 #ifdef CONFIG_PROC_FS
2146 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2147 __acquires(proto_list_lock)
2149 read_lock(&proto_list_lock);
2150 return seq_list_start_head(&proto_list, *pos);
2153 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2155 return seq_list_next(v, &proto_list, pos);
2158 static void proto_seq_stop(struct seq_file *seq, void *v)
2159 __releases(proto_list_lock)
2161 read_unlock(&proto_list_lock);
2164 static char proto_method_implemented(const void *method)
2166 return method == NULL ? 'n' : 'y';
2169 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2171 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2172 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2175 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2176 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2177 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2179 proto->slab == NULL ? "no" : "yes",
2180 module_name(proto->owner),
2181 proto_method_implemented(proto->close),
2182 proto_method_implemented(proto->connect),
2183 proto_method_implemented(proto->disconnect),
2184 proto_method_implemented(proto->accept),
2185 proto_method_implemented(proto->ioctl),
2186 proto_method_implemented(proto->init),
2187 proto_method_implemented(proto->destroy),
2188 proto_method_implemented(proto->shutdown),
2189 proto_method_implemented(proto->setsockopt),
2190 proto_method_implemented(proto->getsockopt),
2191 proto_method_implemented(proto->sendmsg),
2192 proto_method_implemented(proto->recvmsg),
2193 proto_method_implemented(proto->sendpage),
2194 proto_method_implemented(proto->bind),
2195 proto_method_implemented(proto->backlog_rcv),
2196 proto_method_implemented(proto->hash),
2197 proto_method_implemented(proto->unhash),
2198 proto_method_implemented(proto->get_port),
2199 proto_method_implemented(proto->enter_memory_pressure));
2202 static int proto_seq_show(struct seq_file *seq, void *v)
2204 if (v == &proto_list)
2205 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2214 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2216 proto_seq_printf(seq, list_entry(v, struct proto, node));
2220 static const struct seq_operations proto_seq_ops = {
2221 .start = proto_seq_start,
2222 .next = proto_seq_next,
2223 .stop = proto_seq_stop,
2224 .show = proto_seq_show,
2227 static int proto_seq_open(struct inode *inode, struct file *file)
2229 return seq_open(file, &proto_seq_ops);
2232 static const struct file_operations proto_seq_fops = {
2233 .owner = THIS_MODULE,
2234 .open = proto_seq_open,
2236 .llseek = seq_lseek,
2237 .release = seq_release,
2240 static int __init proto_init(void)
2242 /* register /proc/net/protocols */
2243 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2246 subsys_initcall(proto_init);
2248 #endif /* PROC_FS */
2250 EXPORT_SYMBOL(sk_alloc);
2251 EXPORT_SYMBOL(sk_free);
2252 EXPORT_SYMBOL(sk_send_sigurg);
2253 EXPORT_SYMBOL(sock_alloc_send_skb);
2254 EXPORT_SYMBOL(sock_init_data);
2255 EXPORT_SYMBOL(sock_kfree_s);
2256 EXPORT_SYMBOL(sock_kmalloc);
2257 EXPORT_SYMBOL(sock_no_accept);
2258 EXPORT_SYMBOL(sock_no_bind);
2259 EXPORT_SYMBOL(sock_no_connect);
2260 EXPORT_SYMBOL(sock_no_getname);
2261 EXPORT_SYMBOL(sock_no_getsockopt);
2262 EXPORT_SYMBOL(sock_no_ioctl);
2263 EXPORT_SYMBOL(sock_no_listen);
2264 EXPORT_SYMBOL(sock_no_mmap);
2265 EXPORT_SYMBOL(sock_no_poll);
2266 EXPORT_SYMBOL(sock_no_recvmsg);
2267 EXPORT_SYMBOL(sock_no_sendmsg);
2268 EXPORT_SYMBOL(sock_no_sendpage);
2269 EXPORT_SYMBOL(sock_no_setsockopt);
2270 EXPORT_SYMBOL(sock_no_shutdown);
2271 EXPORT_SYMBOL(sock_no_socketpair);
2272 EXPORT_SYMBOL(sock_rfree);
2273 EXPORT_SYMBOL(sock_setsockopt);
2274 EXPORT_SYMBOL(sock_wfree);
2275 EXPORT_SYMBOL(sock_wmalloc);
2276 EXPORT_SYMBOL(sock_i_uid);
2277 EXPORT_SYMBOL(sock_i_ino);
2278 EXPORT_SYMBOL(sysctl_optmem_max);