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];
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
141 * Make lock validator output more readable. (we pre-construct these
142 * strings build-time, so that runtime initialization of socket
145 static const char *af_family_key_strings[AF_MAX+1] = {
146 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
147 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
148 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
149 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
150 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
151 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
152 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
153 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
155 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
157 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
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_MAX"
173 static const char *af_family_clock_key_strings[AF_MAX+1] = {
174 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
175 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
176 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
177 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
178 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
179 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
180 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
181 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
182 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
183 "clock-27" , "clock-28" , "clock-AF_CAN" ,
184 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
185 "clock-AF_RXRPC" , "clock-AF_MAX"
190 * sk_callback_lock locking rules are per-address-family,
191 * so split the lock classes by using a per-AF key:
193 static struct lock_class_key af_callback_keys[AF_MAX];
195 /* Take into consideration the size of the struct sk_buff overhead in the
196 * determination of these values, since that is non-constant across
197 * platforms. This makes socket queueing behavior and performance
198 * not depend upon such differences.
200 #define _SK_MEM_PACKETS 256
201 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
202 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
203 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 /* Run time adjustable parameters. */
206 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
207 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
208 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
211 /* Maximal space eaten by iovec or ancilliary data plus some space */
212 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
214 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
218 if (optlen < sizeof(tv))
220 if (copy_from_user(&tv, optval, sizeof(tv)))
222 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
226 static int warned __read_mostly;
229 if (warned < 10 && net_ratelimit()) {
231 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
232 "tries to set negative timeout\n",
233 current->comm, task_pid_nr(current));
237 *timeo_p = MAX_SCHEDULE_TIMEOUT;
238 if (tv.tv_sec == 0 && tv.tv_usec == 0)
240 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
241 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
245 static void sock_warn_obsolete_bsdism(const char *name)
248 static char warncomm[TASK_COMM_LEN];
249 if (strcmp(warncomm, current->comm) && warned < 5) {
250 strcpy(warncomm, current->comm);
251 printk(KERN_WARNING "process `%s' is using obsolete "
252 "%s SO_BSDCOMPAT\n", warncomm, name);
257 static void sock_disable_timestamp(struct sock *sk)
259 if (sock_flag(sk, SOCK_TIMESTAMP)) {
260 sock_reset_flag(sk, SOCK_TIMESTAMP);
261 net_disable_timestamp();
266 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
271 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
272 number of warnings when compiling with -W --ANK
274 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
275 (unsigned)sk->sk_rcvbuf) {
280 err = sk_filter(sk, skb);
284 if (!sk_rmem_schedule(sk, skb->truesize)) {
290 skb_set_owner_r(skb, sk);
292 /* Cache the SKB length before we tack it onto the receive
293 * queue. Once it is added it no longer belongs to us and
294 * may be freed by other threads of control pulling packets
299 skb_queue_tail(&sk->sk_receive_queue, skb);
301 if (!sock_flag(sk, SOCK_DEAD))
302 sk->sk_data_ready(sk, skb_len);
306 EXPORT_SYMBOL(sock_queue_rcv_skb);
308 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
310 int rc = NET_RX_SUCCESS;
312 if (sk_filter(sk, skb))
313 goto discard_and_relse;
318 bh_lock_sock_nested(sk);
321 if (!sock_owned_by_user(sk)) {
323 * trylock + unlock semantics:
325 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
327 rc = sk->sk_backlog_rcv(sk, skb);
329 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
331 sk_add_backlog(sk, skb);
340 EXPORT_SYMBOL(sk_receive_skb);
342 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
344 struct dst_entry *dst = sk->sk_dst_cache;
346 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
347 sk->sk_dst_cache = NULL;
354 EXPORT_SYMBOL(__sk_dst_check);
356 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
358 struct dst_entry *dst = sk_dst_get(sk);
360 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
368 EXPORT_SYMBOL(sk_dst_check);
370 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
372 int ret = -ENOPROTOOPT;
373 #ifdef CONFIG_NETDEVICES
374 struct net *net = sock_net(sk);
375 char devname[IFNAMSIZ];
380 if (!capable(CAP_NET_RAW))
387 /* Bind this socket to a particular device like "eth0",
388 * as specified in the passed interface name. If the
389 * name is "" or the option length is zero the socket
392 if (optlen > IFNAMSIZ - 1)
393 optlen = IFNAMSIZ - 1;
394 memset(devname, 0, sizeof(devname));
397 if (copy_from_user(devname, optval, optlen))
400 if (devname[0] == '\0') {
403 struct net_device *dev = dev_get_by_name(net, devname);
409 index = dev->ifindex;
414 sk->sk_bound_dev_if = index;
426 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
429 sock_set_flag(sk, bit);
431 sock_reset_flag(sk, bit);
435 * This is meant for all protocols to use and covers goings on
436 * at the socket level. Everything here is generic.
439 int sock_setsockopt(struct socket *sock, int level, int optname,
440 char __user *optval, int optlen)
442 struct sock *sk=sock->sk;
449 * Options without arguments
452 if (optname == SO_BINDTODEVICE)
453 return sock_bindtodevice(sk, optval, optlen);
455 if (optlen < sizeof(int))
458 if (get_user(val, (int __user *)optval))
467 if (val && !capable(CAP_NET_ADMIN)) {
470 sock_valbool_flag(sk, SOCK_DBG, valbool);
473 sk->sk_reuse = valbool;
480 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
483 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
486 /* Don't error on this BSD doesn't and if you think
487 about it this is right. Otherwise apps have to
488 play 'guess the biggest size' games. RCVBUF/SNDBUF
489 are treated in BSD as hints */
491 if (val > sysctl_wmem_max)
492 val = sysctl_wmem_max;
494 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
495 if ((val * 2) < SOCK_MIN_SNDBUF)
496 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
498 sk->sk_sndbuf = val * 2;
501 * Wake up sending tasks if we
504 sk->sk_write_space(sk);
508 if (!capable(CAP_NET_ADMIN)) {
515 /* Don't error on this BSD doesn't and if you think
516 about it this is right. Otherwise apps have to
517 play 'guess the biggest size' games. RCVBUF/SNDBUF
518 are treated in BSD as hints */
520 if (val > sysctl_rmem_max)
521 val = sysctl_rmem_max;
523 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
525 * We double it on the way in to account for
526 * "struct sk_buff" etc. overhead. Applications
527 * assume that the SO_RCVBUF setting they make will
528 * allow that much actual data to be received on that
531 * Applications are unaware that "struct sk_buff" and
532 * other overheads allocate from the receive buffer
533 * during socket buffer allocation.
535 * And after considering the possible alternatives,
536 * returning the value we actually used in getsockopt
537 * is the most desirable behavior.
539 if ((val * 2) < SOCK_MIN_RCVBUF)
540 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
542 sk->sk_rcvbuf = val * 2;
546 if (!capable(CAP_NET_ADMIN)) {
554 if (sk->sk_protocol == IPPROTO_TCP)
555 tcp_set_keepalive(sk, valbool);
557 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
561 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
565 sk->sk_no_check = valbool;
569 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
570 sk->sk_priority = val;
576 if (optlen < sizeof(ling)) {
577 ret = -EINVAL; /* 1003.1g */
580 if (copy_from_user(&ling,optval,sizeof(ling))) {
585 sock_reset_flag(sk, SOCK_LINGER);
587 #if (BITS_PER_LONG == 32)
588 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
589 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
592 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
593 sock_set_flag(sk, SOCK_LINGER);
598 sock_warn_obsolete_bsdism("setsockopt");
603 set_bit(SOCK_PASSCRED, &sock->flags);
605 clear_bit(SOCK_PASSCRED, &sock->flags);
611 if (optname == SO_TIMESTAMP)
612 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
614 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
615 sock_set_flag(sk, SOCK_RCVTSTAMP);
616 sock_enable_timestamp(sk);
618 sock_reset_flag(sk, SOCK_RCVTSTAMP);
619 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
626 sk->sk_rcvlowat = val ? : 1;
630 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
634 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
637 case SO_ATTACH_FILTER:
639 if (optlen == sizeof(struct sock_fprog)) {
640 struct sock_fprog fprog;
643 if (copy_from_user(&fprog, optval, sizeof(fprog)))
646 ret = sk_attach_filter(&fprog, sk);
650 case SO_DETACH_FILTER:
651 ret = sk_detach_filter(sk);
656 set_bit(SOCK_PASSSEC, &sock->flags);
658 clear_bit(SOCK_PASSSEC, &sock->flags);
661 if (!capable(CAP_NET_ADMIN))
668 /* We implement the SO_SNDLOWAT etc to
669 not be settable (1003.1g 5.3) */
679 int sock_getsockopt(struct socket *sock, int level, int optname,
680 char __user *optval, int __user *optlen)
682 struct sock *sk = sock->sk;
690 unsigned int lv = sizeof(int);
693 if (get_user(len, optlen))
700 v.val = sock_flag(sk, SOCK_DBG);
704 v.val = sock_flag(sk, SOCK_LOCALROUTE);
708 v.val = !!sock_flag(sk, SOCK_BROADCAST);
712 v.val = sk->sk_sndbuf;
716 v.val = sk->sk_rcvbuf;
720 v.val = sk->sk_reuse;
724 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
732 v.val = -sock_error(sk);
734 v.val = xchg(&sk->sk_err_soft, 0);
738 v.val = !!sock_flag(sk, SOCK_URGINLINE);
742 v.val = sk->sk_no_check;
746 v.val = sk->sk_priority;
751 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
752 v.ling.l_linger = sk->sk_lingertime / HZ;
756 sock_warn_obsolete_bsdism("getsockopt");
760 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
761 !sock_flag(sk, SOCK_RCVTSTAMPNS);
765 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
769 lv=sizeof(struct timeval);
770 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
774 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
775 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
780 lv=sizeof(struct timeval);
781 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
785 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
786 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
791 v.val = sk->sk_rcvlowat;
799 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
803 if (len > sizeof(sk->sk_peercred))
804 len = sizeof(sk->sk_peercred);
805 if (copy_to_user(optval, &sk->sk_peercred, len))
813 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
817 if (copy_to_user(optval, address, len))
822 /* Dubious BSD thing... Probably nobody even uses it, but
823 * the UNIX standard wants it for whatever reason... -DaveM
826 v.val = sk->sk_state == TCP_LISTEN;
830 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
834 return security_socket_getpeersec_stream(sock, optval, optlen, len);
846 if (copy_to_user(optval, &v, len))
849 if (put_user(len, optlen))
855 * Initialize an sk_lock.
857 * (We also register the sk_lock with the lock validator.)
859 static inline void sock_lock_init(struct sock *sk)
861 sock_lock_init_class_and_name(sk,
862 af_family_slock_key_strings[sk->sk_family],
863 af_family_slock_keys + sk->sk_family,
864 af_family_key_strings[sk->sk_family],
865 af_family_keys + sk->sk_family);
868 static void sock_copy(struct sock *nsk, const struct sock *osk)
870 #ifdef CONFIG_SECURITY_NETWORK
871 void *sptr = nsk->sk_security;
874 memcpy(nsk, osk, osk->sk_prot->obj_size);
875 #ifdef CONFIG_SECURITY_NETWORK
876 nsk->sk_security = sptr;
877 security_sk_clone(osk, nsk);
881 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
885 struct kmem_cache *slab;
889 sk = kmem_cache_alloc(slab, priority);
891 sk = kmalloc(prot->obj_size, priority);
894 if (security_sk_alloc(sk, family, priority))
897 if (!try_module_get(prot->owner))
904 security_sk_free(sk);
907 kmem_cache_free(slab, sk);
913 static void sk_prot_free(struct proto *prot, struct sock *sk)
915 struct kmem_cache *slab;
916 struct module *owner;
921 security_sk_free(sk);
923 kmem_cache_free(slab, sk);
930 * sk_alloc - All socket objects are allocated here
931 * @net: the applicable net namespace
932 * @family: protocol family
933 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
934 * @prot: struct proto associated with this new sock instance
936 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
941 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
943 sk->sk_family = family;
945 * See comment in struct sock definition to understand
946 * why we need sk_prot_creator -acme
948 sk->sk_prot = sk->sk_prot_creator = prot;
950 sock_net_set(sk, get_net(net));
956 void sk_free(struct sock *sk)
958 struct sk_filter *filter;
963 filter = rcu_dereference(sk->sk_filter);
965 sk_filter_uncharge(sk, filter);
966 rcu_assign_pointer(sk->sk_filter, NULL);
969 sock_disable_timestamp(sk);
971 if (atomic_read(&sk->sk_omem_alloc))
972 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
973 __func__, atomic_read(&sk->sk_omem_alloc));
975 put_net(sock_net(sk));
976 sk_prot_free(sk->sk_prot_creator, sk);
980 * Last sock_put should drop referrence to sk->sk_net. It has already
981 * been dropped in sk_change_net. Taking referrence to stopping namespace
983 * Take referrence to a socket to remove it from hash _alive_ and after that
984 * destroy it in the context of init_net.
986 void sk_release_kernel(struct sock *sk)
988 if (sk == NULL || sk->sk_socket == NULL)
992 sock_release(sk->sk_socket);
993 release_net(sock_net(sk));
994 sock_net_set(sk, get_net(&init_net));
997 EXPORT_SYMBOL(sk_release_kernel);
999 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1003 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1004 if (newsk != NULL) {
1005 struct sk_filter *filter;
1007 sock_copy(newsk, sk);
1010 get_net(sock_net(newsk));
1011 sk_node_init(&newsk->sk_node);
1012 sock_lock_init(newsk);
1013 bh_lock_sock(newsk);
1014 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1016 atomic_set(&newsk->sk_rmem_alloc, 0);
1017 atomic_set(&newsk->sk_wmem_alloc, 0);
1018 atomic_set(&newsk->sk_omem_alloc, 0);
1019 skb_queue_head_init(&newsk->sk_receive_queue);
1020 skb_queue_head_init(&newsk->sk_write_queue);
1021 #ifdef CONFIG_NET_DMA
1022 skb_queue_head_init(&newsk->sk_async_wait_queue);
1025 rwlock_init(&newsk->sk_dst_lock);
1026 rwlock_init(&newsk->sk_callback_lock);
1027 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1028 af_callback_keys + newsk->sk_family,
1029 af_family_clock_key_strings[newsk->sk_family]);
1031 newsk->sk_dst_cache = NULL;
1032 newsk->sk_wmem_queued = 0;
1033 newsk->sk_forward_alloc = 0;
1034 newsk->sk_send_head = NULL;
1035 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1037 sock_reset_flag(newsk, SOCK_DONE);
1038 skb_queue_head_init(&newsk->sk_error_queue);
1040 filter = newsk->sk_filter;
1042 sk_filter_charge(newsk, filter);
1044 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1045 /* It is still raw copy of parent, so invalidate
1046 * destructor and make plain sk_free() */
1047 newsk->sk_destruct = NULL;
1054 newsk->sk_priority = 0;
1055 atomic_set(&newsk->sk_refcnt, 2);
1058 * Increment the counter in the same struct proto as the master
1059 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1060 * is the same as sk->sk_prot->socks, as this field was copied
1063 * This _changes_ the previous behaviour, where
1064 * tcp_create_openreq_child always was incrementing the
1065 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1066 * to be taken into account in all callers. -acme
1068 sk_refcnt_debug_inc(newsk);
1069 sk_set_socket(newsk, NULL);
1070 newsk->sk_sleep = NULL;
1072 if (newsk->sk_prot->sockets_allocated)
1073 atomic_inc(newsk->sk_prot->sockets_allocated);
1079 EXPORT_SYMBOL_GPL(sk_clone);
1081 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1083 __sk_dst_set(sk, dst);
1084 sk->sk_route_caps = dst->dev->features;
1085 if (sk->sk_route_caps & NETIF_F_GSO)
1086 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1087 if (sk_can_gso(sk)) {
1088 if (dst->header_len) {
1089 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1091 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1092 sk->sk_gso_max_size = dst->dev->gso_max_size;
1096 EXPORT_SYMBOL_GPL(sk_setup_caps);
1098 void __init sk_init(void)
1100 if (num_physpages <= 4096) {
1101 sysctl_wmem_max = 32767;
1102 sysctl_rmem_max = 32767;
1103 sysctl_wmem_default = 32767;
1104 sysctl_rmem_default = 32767;
1105 } else if (num_physpages >= 131072) {
1106 sysctl_wmem_max = 131071;
1107 sysctl_rmem_max = 131071;
1112 * Simple resource managers for sockets.
1117 * Write buffer destructor automatically called from kfree_skb.
1119 void sock_wfree(struct sk_buff *skb)
1121 struct sock *sk = skb->sk;
1123 /* In case it might be waiting for more memory. */
1124 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1125 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1126 sk->sk_write_space(sk);
1131 * Read buffer destructor automatically called from kfree_skb.
1133 void sock_rfree(struct sk_buff *skb)
1135 struct sock *sk = skb->sk;
1137 skb_truesize_check(skb);
1138 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1139 sk_mem_uncharge(skb->sk, skb->truesize);
1143 int sock_i_uid(struct sock *sk)
1147 read_lock(&sk->sk_callback_lock);
1148 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1149 read_unlock(&sk->sk_callback_lock);
1153 unsigned long sock_i_ino(struct sock *sk)
1157 read_lock(&sk->sk_callback_lock);
1158 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1159 read_unlock(&sk->sk_callback_lock);
1164 * Allocate a skb from the socket's send buffer.
1166 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1169 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1170 struct sk_buff * skb = alloc_skb(size, priority);
1172 skb_set_owner_w(skb, sk);
1180 * Allocate a skb from the socket's receive buffer.
1182 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1185 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1186 struct sk_buff *skb = alloc_skb(size, priority);
1188 skb_set_owner_r(skb, sk);
1196 * Allocate a memory block from the socket's option memory buffer.
1198 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1200 if ((unsigned)size <= sysctl_optmem_max &&
1201 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1203 /* First do the add, to avoid the race if kmalloc
1206 atomic_add(size, &sk->sk_omem_alloc);
1207 mem = kmalloc(size, priority);
1210 atomic_sub(size, &sk->sk_omem_alloc);
1216 * Free an option memory block.
1218 void sock_kfree_s(struct sock *sk, void *mem, int size)
1221 atomic_sub(size, &sk->sk_omem_alloc);
1224 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1225 I think, these locks should be removed for datagram sockets.
1227 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1231 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1235 if (signal_pending(current))
1237 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1238 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1239 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1241 if (sk->sk_shutdown & SEND_SHUTDOWN)
1245 timeo = schedule_timeout(timeo);
1247 finish_wait(sk->sk_sleep, &wait);
1253 * Generic send/receive buffer handlers
1256 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1257 unsigned long header_len,
1258 unsigned long data_len,
1259 int noblock, int *errcode)
1261 struct sk_buff *skb;
1266 gfp_mask = sk->sk_allocation;
1267 if (gfp_mask & __GFP_WAIT)
1268 gfp_mask |= __GFP_REPEAT;
1270 timeo = sock_sndtimeo(sk, noblock);
1272 err = sock_error(sk);
1277 if (sk->sk_shutdown & SEND_SHUTDOWN)
1280 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1281 skb = alloc_skb(header_len, gfp_mask);
1286 /* No pages, we're done... */
1290 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1291 skb->truesize += data_len;
1292 skb_shinfo(skb)->nr_frags = npages;
1293 for (i = 0; i < npages; i++) {
1297 page = alloc_pages(sk->sk_allocation, 0);
1300 skb_shinfo(skb)->nr_frags = i;
1305 frag = &skb_shinfo(skb)->frags[i];
1307 frag->page_offset = 0;
1308 frag->size = (data_len >= PAGE_SIZE ?
1311 data_len -= PAGE_SIZE;
1314 /* Full success... */
1320 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1321 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1325 if (signal_pending(current))
1327 timeo = sock_wait_for_wmem(sk, timeo);
1330 skb_set_owner_w(skb, sk);
1334 err = sock_intr_errno(timeo);
1340 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1341 int noblock, int *errcode)
1343 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1346 static void __lock_sock(struct sock *sk)
1351 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1352 TASK_UNINTERRUPTIBLE);
1353 spin_unlock_bh(&sk->sk_lock.slock);
1355 spin_lock_bh(&sk->sk_lock.slock);
1356 if (!sock_owned_by_user(sk))
1359 finish_wait(&sk->sk_lock.wq, &wait);
1362 static void __release_sock(struct sock *sk)
1364 struct sk_buff *skb = sk->sk_backlog.head;
1367 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1371 struct sk_buff *next = skb->next;
1374 sk->sk_backlog_rcv(sk, skb);
1377 * We are in process context here with softirqs
1378 * disabled, use cond_resched_softirq() to preempt.
1379 * This is safe to do because we've taken the backlog
1382 cond_resched_softirq();
1385 } while (skb != NULL);
1388 } while ((skb = sk->sk_backlog.head) != NULL);
1392 * sk_wait_data - wait for data to arrive at sk_receive_queue
1393 * @sk: sock to wait on
1394 * @timeo: for how long
1396 * Now socket state including sk->sk_err is changed only under lock,
1397 * hence we may omit checks after joining wait queue.
1398 * We check receive queue before schedule() only as optimization;
1399 * it is very likely that release_sock() added new data.
1401 int sk_wait_data(struct sock *sk, long *timeo)
1406 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1407 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1408 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1409 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1410 finish_wait(sk->sk_sleep, &wait);
1414 EXPORT_SYMBOL(sk_wait_data);
1417 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1419 * @size: memory size to allocate
1420 * @kind: allocation type
1422 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1423 * rmem allocation. This function assumes that protocols which have
1424 * memory_pressure use sk_wmem_queued as write buffer accounting.
1426 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1428 struct proto *prot = sk->sk_prot;
1429 int amt = sk_mem_pages(size);
1432 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1433 allocated = atomic_add_return(amt, prot->memory_allocated);
1436 if (allocated <= prot->sysctl_mem[0]) {
1437 if (prot->memory_pressure && *prot->memory_pressure)
1438 *prot->memory_pressure = 0;
1442 /* Under pressure. */
1443 if (allocated > prot->sysctl_mem[1])
1444 if (prot->enter_memory_pressure)
1445 prot->enter_memory_pressure(sk);
1447 /* Over hard limit. */
1448 if (allocated > prot->sysctl_mem[2])
1449 goto suppress_allocation;
1451 /* guarantee minimum buffer size under pressure */
1452 if (kind == SK_MEM_RECV) {
1453 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1455 } else { /* SK_MEM_SEND */
1456 if (sk->sk_type == SOCK_STREAM) {
1457 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1459 } else if (atomic_read(&sk->sk_wmem_alloc) <
1460 prot->sysctl_wmem[0])
1464 if (prot->memory_pressure) {
1465 if (!*prot->memory_pressure ||
1466 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1467 sk_mem_pages(sk->sk_wmem_queued +
1468 atomic_read(&sk->sk_rmem_alloc) +
1469 sk->sk_forward_alloc))
1473 suppress_allocation:
1475 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1476 sk_stream_moderate_sndbuf(sk);
1478 /* Fail only if socket is _under_ its sndbuf.
1479 * In this case we cannot block, so that we have to fail.
1481 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1485 /* Alas. Undo changes. */
1486 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1487 atomic_sub(amt, prot->memory_allocated);
1491 EXPORT_SYMBOL(__sk_mem_schedule);
1494 * __sk_reclaim - reclaim memory_allocated
1497 void __sk_mem_reclaim(struct sock *sk)
1499 struct proto *prot = sk->sk_prot;
1501 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1502 prot->memory_allocated);
1503 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1505 if (prot->memory_pressure && *prot->memory_pressure &&
1506 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1507 *prot->memory_pressure = 0;
1510 EXPORT_SYMBOL(__sk_mem_reclaim);
1514 * Set of default routines for initialising struct proto_ops when
1515 * the protocol does not support a particular function. In certain
1516 * cases where it makes no sense for a protocol to have a "do nothing"
1517 * function, some default processing is provided.
1520 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1525 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1531 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1536 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1541 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1547 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1552 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1557 int sock_no_listen(struct socket *sock, int backlog)
1562 int sock_no_shutdown(struct socket *sock, int how)
1567 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1568 char __user *optval, int optlen)
1573 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1574 char __user *optval, int __user *optlen)
1579 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1585 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1586 size_t len, int flags)
1591 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1593 /* Mirror missing mmap method error code */
1597 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1600 struct msghdr msg = {.msg_flags = flags};
1602 char *kaddr = kmap(page);
1603 iov.iov_base = kaddr + offset;
1605 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1611 * Default Socket Callbacks
1614 static void sock_def_wakeup(struct sock *sk)
1616 read_lock(&sk->sk_callback_lock);
1617 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1618 wake_up_interruptible_all(sk->sk_sleep);
1619 read_unlock(&sk->sk_callback_lock);
1622 static void sock_def_error_report(struct sock *sk)
1624 read_lock(&sk->sk_callback_lock);
1625 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1626 wake_up_interruptible(sk->sk_sleep);
1627 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1628 read_unlock(&sk->sk_callback_lock);
1631 static void sock_def_readable(struct sock *sk, int len)
1633 read_lock(&sk->sk_callback_lock);
1634 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1635 wake_up_interruptible_sync(sk->sk_sleep);
1636 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1637 read_unlock(&sk->sk_callback_lock);
1640 static void sock_def_write_space(struct sock *sk)
1642 read_lock(&sk->sk_callback_lock);
1644 /* Do not wake up a writer until he can make "significant"
1647 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1648 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1649 wake_up_interruptible_sync(sk->sk_sleep);
1651 /* Should agree with poll, otherwise some programs break */
1652 if (sock_writeable(sk))
1653 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1656 read_unlock(&sk->sk_callback_lock);
1659 static void sock_def_destruct(struct sock *sk)
1661 kfree(sk->sk_protinfo);
1664 void sk_send_sigurg(struct sock *sk)
1666 if (sk->sk_socket && sk->sk_socket->file)
1667 if (send_sigurg(&sk->sk_socket->file->f_owner))
1668 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1671 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1672 unsigned long expires)
1674 if (!mod_timer(timer, expires))
1678 EXPORT_SYMBOL(sk_reset_timer);
1680 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1682 if (timer_pending(timer) && del_timer(timer))
1686 EXPORT_SYMBOL(sk_stop_timer);
1688 void sock_init_data(struct socket *sock, struct sock *sk)
1690 skb_queue_head_init(&sk->sk_receive_queue);
1691 skb_queue_head_init(&sk->sk_write_queue);
1692 skb_queue_head_init(&sk->sk_error_queue);
1693 #ifdef CONFIG_NET_DMA
1694 skb_queue_head_init(&sk->sk_async_wait_queue);
1697 sk->sk_send_head = NULL;
1699 init_timer(&sk->sk_timer);
1701 sk->sk_allocation = GFP_KERNEL;
1702 sk->sk_rcvbuf = sysctl_rmem_default;
1703 sk->sk_sndbuf = sysctl_wmem_default;
1704 sk->sk_state = TCP_CLOSE;
1705 sk_set_socket(sk, sock);
1707 sock_set_flag(sk, SOCK_ZAPPED);
1710 sk->sk_type = sock->type;
1711 sk->sk_sleep = &sock->wait;
1714 sk->sk_sleep = NULL;
1716 rwlock_init(&sk->sk_dst_lock);
1717 rwlock_init(&sk->sk_callback_lock);
1718 lockdep_set_class_and_name(&sk->sk_callback_lock,
1719 af_callback_keys + sk->sk_family,
1720 af_family_clock_key_strings[sk->sk_family]);
1722 sk->sk_state_change = sock_def_wakeup;
1723 sk->sk_data_ready = sock_def_readable;
1724 sk->sk_write_space = sock_def_write_space;
1725 sk->sk_error_report = sock_def_error_report;
1726 sk->sk_destruct = sock_def_destruct;
1728 sk->sk_sndmsg_page = NULL;
1729 sk->sk_sndmsg_off = 0;
1731 sk->sk_peercred.pid = 0;
1732 sk->sk_peercred.uid = -1;
1733 sk->sk_peercred.gid = -1;
1734 sk->sk_write_pending = 0;
1735 sk->sk_rcvlowat = 1;
1736 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1737 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1739 sk->sk_stamp = ktime_set(-1L, 0);
1741 atomic_set(&sk->sk_refcnt, 1);
1742 atomic_set(&sk->sk_drops, 0);
1745 void lock_sock_nested(struct sock *sk, int subclass)
1748 spin_lock_bh(&sk->sk_lock.slock);
1749 if (sk->sk_lock.owned)
1751 sk->sk_lock.owned = 1;
1752 spin_unlock(&sk->sk_lock.slock);
1754 * The sk_lock has mutex_lock() semantics here:
1756 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1760 EXPORT_SYMBOL(lock_sock_nested);
1762 void release_sock(struct sock *sk)
1765 * The sk_lock has mutex_unlock() semantics:
1767 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1769 spin_lock_bh(&sk->sk_lock.slock);
1770 if (sk->sk_backlog.tail)
1772 sk->sk_lock.owned = 0;
1773 if (waitqueue_active(&sk->sk_lock.wq))
1774 wake_up(&sk->sk_lock.wq);
1775 spin_unlock_bh(&sk->sk_lock.slock);
1777 EXPORT_SYMBOL(release_sock);
1779 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1782 if (!sock_flag(sk, SOCK_TIMESTAMP))
1783 sock_enable_timestamp(sk);
1784 tv = ktime_to_timeval(sk->sk_stamp);
1785 if (tv.tv_sec == -1)
1787 if (tv.tv_sec == 0) {
1788 sk->sk_stamp = ktime_get_real();
1789 tv = ktime_to_timeval(sk->sk_stamp);
1791 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1793 EXPORT_SYMBOL(sock_get_timestamp);
1795 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1798 if (!sock_flag(sk, SOCK_TIMESTAMP))
1799 sock_enable_timestamp(sk);
1800 ts = ktime_to_timespec(sk->sk_stamp);
1801 if (ts.tv_sec == -1)
1803 if (ts.tv_sec == 0) {
1804 sk->sk_stamp = ktime_get_real();
1805 ts = ktime_to_timespec(sk->sk_stamp);
1807 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1809 EXPORT_SYMBOL(sock_get_timestampns);
1811 void sock_enable_timestamp(struct sock *sk)
1813 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1814 sock_set_flag(sk, SOCK_TIMESTAMP);
1815 net_enable_timestamp();
1820 * Get a socket option on an socket.
1822 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1823 * asynchronous errors should be reported by getsockopt. We assume
1824 * this means if you specify SO_ERROR (otherwise whats the point of it).
1826 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1827 char __user *optval, int __user *optlen)
1829 struct sock *sk = sock->sk;
1831 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1834 EXPORT_SYMBOL(sock_common_getsockopt);
1836 #ifdef CONFIG_COMPAT
1837 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1838 char __user *optval, int __user *optlen)
1840 struct sock *sk = sock->sk;
1842 if (sk->sk_prot->compat_getsockopt != NULL)
1843 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1845 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1847 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1850 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1851 struct msghdr *msg, size_t size, int flags)
1853 struct sock *sk = sock->sk;
1857 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1858 flags & ~MSG_DONTWAIT, &addr_len);
1860 msg->msg_namelen = addr_len;
1864 EXPORT_SYMBOL(sock_common_recvmsg);
1867 * Set socket options on an inet socket.
1869 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1870 char __user *optval, int optlen)
1872 struct sock *sk = sock->sk;
1874 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1877 EXPORT_SYMBOL(sock_common_setsockopt);
1879 #ifdef CONFIG_COMPAT
1880 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1881 char __user *optval, int optlen)
1883 struct sock *sk = sock->sk;
1885 if (sk->sk_prot->compat_setsockopt != NULL)
1886 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1888 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1890 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1893 void sk_common_release(struct sock *sk)
1895 if (sk->sk_prot->destroy)
1896 sk->sk_prot->destroy(sk);
1899 * Observation: when sock_common_release is called, processes have
1900 * no access to socket. But net still has.
1901 * Step one, detach it from networking:
1903 * A. Remove from hash tables.
1906 sk->sk_prot->unhash(sk);
1909 * In this point socket cannot receive new packets, but it is possible
1910 * that some packets are in flight because some CPU runs receiver and
1911 * did hash table lookup before we unhashed socket. They will achieve
1912 * receive queue and will be purged by socket destructor.
1914 * Also we still have packets pending on receive queue and probably,
1915 * our own packets waiting in device queues. sock_destroy will drain
1916 * receive queue, but transmitted packets will delay socket destruction
1917 * until the last reference will be released.
1922 xfrm_sk_free_policy(sk);
1924 sk_refcnt_debug_release(sk);
1928 EXPORT_SYMBOL(sk_common_release);
1930 static DEFINE_RWLOCK(proto_list_lock);
1931 static LIST_HEAD(proto_list);
1933 #ifdef CONFIG_PROC_FS
1934 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1936 int val[PROTO_INUSE_NR];
1939 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1941 #ifdef CONFIG_NET_NS
1942 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1944 int cpu = smp_processor_id();
1945 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1947 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1949 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1951 int cpu, idx = prot->inuse_idx;
1954 for_each_possible_cpu(cpu)
1955 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1957 return res >= 0 ? res : 0;
1959 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1961 static int sock_inuse_init_net(struct net *net)
1963 net->core.inuse = alloc_percpu(struct prot_inuse);
1964 return net->core.inuse ? 0 : -ENOMEM;
1967 static void sock_inuse_exit_net(struct net *net)
1969 free_percpu(net->core.inuse);
1972 static struct pernet_operations net_inuse_ops = {
1973 .init = sock_inuse_init_net,
1974 .exit = sock_inuse_exit_net,
1977 static __init int net_inuse_init(void)
1979 if (register_pernet_subsys(&net_inuse_ops))
1980 panic("Cannot initialize net inuse counters");
1985 core_initcall(net_inuse_init);
1987 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1989 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1991 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1993 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1995 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1997 int cpu, idx = prot->inuse_idx;
2000 for_each_possible_cpu(cpu)
2001 res += per_cpu(prot_inuse, cpu).val[idx];
2003 return res >= 0 ? res : 0;
2005 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2008 static void assign_proto_idx(struct proto *prot)
2010 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2012 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2013 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2017 set_bit(prot->inuse_idx, proto_inuse_idx);
2020 static void release_proto_idx(struct proto *prot)
2022 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2023 clear_bit(prot->inuse_idx, proto_inuse_idx);
2026 static inline void assign_proto_idx(struct proto *prot)
2030 static inline void release_proto_idx(struct proto *prot)
2035 int proto_register(struct proto *prot, int alloc_slab)
2037 char *request_sock_slab_name = NULL;
2038 char *timewait_sock_slab_name;
2041 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2042 SLAB_HWCACHE_ALIGN, NULL);
2044 if (prot->slab == NULL) {
2045 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2050 if (prot->rsk_prot != NULL) {
2051 static const char mask[] = "request_sock_%s";
2053 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2054 if (request_sock_slab_name == NULL)
2055 goto out_free_sock_slab;
2057 sprintf(request_sock_slab_name, mask, prot->name);
2058 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
2059 prot->rsk_prot->obj_size, 0,
2060 SLAB_HWCACHE_ALIGN, NULL);
2062 if (prot->rsk_prot->slab == NULL) {
2063 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2065 goto out_free_request_sock_slab_name;
2069 if (prot->twsk_prot != NULL) {
2070 static const char mask[] = "tw_sock_%s";
2072 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2074 if (timewait_sock_slab_name == NULL)
2075 goto out_free_request_sock_slab;
2077 sprintf(timewait_sock_slab_name, mask, prot->name);
2078 prot->twsk_prot->twsk_slab =
2079 kmem_cache_create(timewait_sock_slab_name,
2080 prot->twsk_prot->twsk_obj_size,
2081 0, SLAB_HWCACHE_ALIGN,
2083 if (prot->twsk_prot->twsk_slab == NULL)
2084 goto out_free_timewait_sock_slab_name;
2088 write_lock(&proto_list_lock);
2089 list_add(&prot->node, &proto_list);
2090 assign_proto_idx(prot);
2091 write_unlock(&proto_list_lock);
2094 out_free_timewait_sock_slab_name:
2095 kfree(timewait_sock_slab_name);
2096 out_free_request_sock_slab:
2097 if (prot->rsk_prot && prot->rsk_prot->slab) {
2098 kmem_cache_destroy(prot->rsk_prot->slab);
2099 prot->rsk_prot->slab = NULL;
2101 out_free_request_sock_slab_name:
2102 kfree(request_sock_slab_name);
2104 kmem_cache_destroy(prot->slab);
2110 EXPORT_SYMBOL(proto_register);
2112 void proto_unregister(struct proto *prot)
2114 write_lock(&proto_list_lock);
2115 release_proto_idx(prot);
2116 list_del(&prot->node);
2117 write_unlock(&proto_list_lock);
2119 if (prot->slab != NULL) {
2120 kmem_cache_destroy(prot->slab);
2124 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2125 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2127 kmem_cache_destroy(prot->rsk_prot->slab);
2129 prot->rsk_prot->slab = NULL;
2132 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2133 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
2135 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2137 prot->twsk_prot->twsk_slab = NULL;
2141 EXPORT_SYMBOL(proto_unregister);
2143 #ifdef CONFIG_PROC_FS
2144 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2145 __acquires(proto_list_lock)
2147 read_lock(&proto_list_lock);
2148 return seq_list_start_head(&proto_list, *pos);
2151 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2153 return seq_list_next(v, &proto_list, pos);
2156 static void proto_seq_stop(struct seq_file *seq, void *v)
2157 __releases(proto_list_lock)
2159 read_unlock(&proto_list_lock);
2162 static char proto_method_implemented(const void *method)
2164 return method == NULL ? 'n' : 'y';
2167 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2169 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2170 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2173 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2174 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2175 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2177 proto->slab == NULL ? "no" : "yes",
2178 module_name(proto->owner),
2179 proto_method_implemented(proto->close),
2180 proto_method_implemented(proto->connect),
2181 proto_method_implemented(proto->disconnect),
2182 proto_method_implemented(proto->accept),
2183 proto_method_implemented(proto->ioctl),
2184 proto_method_implemented(proto->init),
2185 proto_method_implemented(proto->destroy),
2186 proto_method_implemented(proto->shutdown),
2187 proto_method_implemented(proto->setsockopt),
2188 proto_method_implemented(proto->getsockopt),
2189 proto_method_implemented(proto->sendmsg),
2190 proto_method_implemented(proto->recvmsg),
2191 proto_method_implemented(proto->sendpage),
2192 proto_method_implemented(proto->bind),
2193 proto_method_implemented(proto->backlog_rcv),
2194 proto_method_implemented(proto->hash),
2195 proto_method_implemented(proto->unhash),
2196 proto_method_implemented(proto->get_port),
2197 proto_method_implemented(proto->enter_memory_pressure));
2200 static int proto_seq_show(struct seq_file *seq, void *v)
2202 if (v == &proto_list)
2203 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2212 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2214 proto_seq_printf(seq, list_entry(v, struct proto, node));
2218 static const struct seq_operations proto_seq_ops = {
2219 .start = proto_seq_start,
2220 .next = proto_seq_next,
2221 .stop = proto_seq_stop,
2222 .show = proto_seq_show,
2225 static int proto_seq_open(struct inode *inode, struct file *file)
2227 return seq_open(file, &proto_seq_ops);
2230 static const struct file_operations proto_seq_fops = {
2231 .owner = THIS_MODULE,
2232 .open = proto_seq_open,
2234 .llseek = seq_lseek,
2235 .release = seq_release,
2238 static int __init proto_init(void)
2240 /* register /proc/net/protocols */
2241 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2244 subsys_initcall(proto_init);
2246 #endif /* PROC_FS */
2248 EXPORT_SYMBOL(sk_alloc);
2249 EXPORT_SYMBOL(sk_free);
2250 EXPORT_SYMBOL(sk_send_sigurg);
2251 EXPORT_SYMBOL(sock_alloc_send_skb);
2252 EXPORT_SYMBOL(sock_init_data);
2253 EXPORT_SYMBOL(sock_kfree_s);
2254 EXPORT_SYMBOL(sock_kmalloc);
2255 EXPORT_SYMBOL(sock_no_accept);
2256 EXPORT_SYMBOL(sock_no_bind);
2257 EXPORT_SYMBOL(sock_no_connect);
2258 EXPORT_SYMBOL(sock_no_getname);
2259 EXPORT_SYMBOL(sock_no_getsockopt);
2260 EXPORT_SYMBOL(sock_no_ioctl);
2261 EXPORT_SYMBOL(sock_no_listen);
2262 EXPORT_SYMBOL(sock_no_mmap);
2263 EXPORT_SYMBOL(sock_no_poll);
2264 EXPORT_SYMBOL(sock_no_recvmsg);
2265 EXPORT_SYMBOL(sock_no_sendmsg);
2266 EXPORT_SYMBOL(sock_no_sendpage);
2267 EXPORT_SYMBOL(sock_no_setsockopt);
2268 EXPORT_SYMBOL(sock_no_shutdown);
2269 EXPORT_SYMBOL(sock_no_socketpair);
2270 EXPORT_SYMBOL(sock_rfree);
2271 EXPORT_SYMBOL(sock_setsockopt);
2272 EXPORT_SYMBOL(sock_wfree);
2273 EXPORT_SYMBOL(sock_wmalloc);
2274 EXPORT_SYMBOL(sock_i_uid);
2275 EXPORT_SYMBOL(sock_i_ino);
2276 EXPORT_SYMBOL(sysctl_optmem_max);