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 <linux/net_tstamp.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
127 #include <linux/filter.h>
134 * Each address family might have different locking rules, so we have
135 * one slock key per address family:
137 static struct lock_class_key af_family_keys[AF_MAX];
138 static struct lock_class_key af_family_slock_keys[AF_MAX];
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 *const 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-AF_RDS" , "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_ISDN" , "sk_lock-AF_PHONET" ,
158 "sk_lock-AF_IEEE802154",
161 static const char *const 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-AF_RDS" , "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_ISDN" , "slock-AF_PHONET" ,
174 "slock-AF_IEEE802154",
177 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
178 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
179 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
180 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
181 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
182 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
183 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
184 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
185 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
186 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
187 "clock-27" , "clock-28" , "clock-AF_CAN" ,
188 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
189 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
190 "clock-AF_IEEE802154",
195 * sk_callback_lock locking rules are per-address-family,
196 * so split the lock classes by using a per-AF key:
198 static struct lock_class_key af_callback_keys[AF_MAX];
200 /* Take into consideration the size of the struct sk_buff overhead in the
201 * determination of these values, since that is non-constant across
202 * platforms. This makes socket queueing behavior and performance
203 * not depend upon such differences.
205 #define _SK_MEM_PACKETS 256
206 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
207 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
208 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
210 /* Run time adjustable parameters. */
211 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
212 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
213 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
214 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
216 /* Maximal space eaten by iovec or ancilliary data plus some space */
217 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
218 EXPORT_SYMBOL(sysctl_optmem_max);
220 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
224 if (optlen < sizeof(tv))
226 if (copy_from_user(&tv, optval, sizeof(tv)))
228 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
232 static int warned __read_mostly;
235 if (warned < 10 && net_ratelimit()) {
237 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
238 "tries to set negative timeout\n",
239 current->comm, task_pid_nr(current));
243 *timeo_p = MAX_SCHEDULE_TIMEOUT;
244 if (tv.tv_sec == 0 && tv.tv_usec == 0)
246 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
247 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
251 static void sock_warn_obsolete_bsdism(const char *name)
254 static char warncomm[TASK_COMM_LEN];
255 if (strcmp(warncomm, current->comm) && warned < 5) {
256 strcpy(warncomm, current->comm);
257 printk(KERN_WARNING "process `%s' is using obsolete "
258 "%s SO_BSDCOMPAT\n", warncomm, name);
263 static void sock_disable_timestamp(struct sock *sk, int flag)
265 if (sock_flag(sk, flag)) {
266 sock_reset_flag(sk, flag);
267 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
268 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
269 net_disable_timestamp();
275 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
280 struct sk_buff_head *list = &sk->sk_receive_queue;
282 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
283 number of warnings when compiling with -W --ANK
285 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
286 (unsigned)sk->sk_rcvbuf) {
287 atomic_inc(&sk->sk_drops);
291 err = sk_filter(sk, skb);
295 if (!sk_rmem_schedule(sk, skb->truesize)) {
296 atomic_inc(&sk->sk_drops);
301 skb_set_owner_r(skb, sk);
303 /* Cache the SKB length before we tack it onto the receive
304 * queue. Once it is added it no longer belongs to us and
305 * may be freed by other threads of control pulling packets
310 spin_lock_irqsave(&list->lock, flags);
311 skb->dropcount = atomic_read(&sk->sk_drops);
312 __skb_queue_tail(list, skb);
313 spin_unlock_irqrestore(&list->lock, flags);
315 if (!sock_flag(sk, SOCK_DEAD))
316 sk->sk_data_ready(sk, skb_len);
319 EXPORT_SYMBOL(sock_queue_rcv_skb);
321 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
323 int rc = NET_RX_SUCCESS;
325 if (sk_filter(sk, skb))
326 goto discard_and_relse;
330 if (sk_rcvqueues_full(sk, skb)) {
331 atomic_inc(&sk->sk_drops);
332 goto discard_and_relse;
335 bh_lock_sock_nested(sk);
338 if (!sock_owned_by_user(sk)) {
340 * trylock + unlock semantics:
342 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
344 rc = sk_backlog_rcv(sk, skb);
346 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
347 } else if (sk_add_backlog(sk, skb)) {
349 atomic_inc(&sk->sk_drops);
350 goto discard_and_relse;
361 EXPORT_SYMBOL(sk_receive_skb);
363 void sk_reset_txq(struct sock *sk)
365 sk_tx_queue_clear(sk);
367 EXPORT_SYMBOL(sk_reset_txq);
369 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
371 struct dst_entry *dst = __sk_dst_get(sk);
373 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
374 sk_tx_queue_clear(sk);
375 rcu_assign_pointer(sk->sk_dst_cache, NULL);
382 EXPORT_SYMBOL(__sk_dst_check);
384 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
386 struct dst_entry *dst = sk_dst_get(sk);
388 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
396 EXPORT_SYMBOL(sk_dst_check);
398 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
400 int ret = -ENOPROTOOPT;
401 #ifdef CONFIG_NETDEVICES
402 struct net *net = sock_net(sk);
403 char devname[IFNAMSIZ];
408 if (!capable(CAP_NET_RAW))
415 /* Bind this socket to a particular device like "eth0",
416 * as specified in the passed interface name. If the
417 * name is "" or the option length is zero the socket
420 if (optlen > IFNAMSIZ - 1)
421 optlen = IFNAMSIZ - 1;
422 memset(devname, 0, sizeof(devname));
425 if (copy_from_user(devname, optval, optlen))
429 if (devname[0] != '\0') {
430 struct net_device *dev;
433 dev = dev_get_by_name_rcu(net, devname);
435 index = dev->ifindex;
443 sk->sk_bound_dev_if = index;
455 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
458 sock_set_flag(sk, bit);
460 sock_reset_flag(sk, bit);
464 * This is meant for all protocols to use and covers goings on
465 * at the socket level. Everything here is generic.
468 int sock_setsockopt(struct socket *sock, int level, int optname,
469 char __user *optval, unsigned int optlen)
471 struct sock *sk = sock->sk;
478 * Options without arguments
481 if (optname == SO_BINDTODEVICE)
482 return sock_bindtodevice(sk, optval, optlen);
484 if (optlen < sizeof(int))
487 if (get_user(val, (int __user *)optval))
490 valbool = val ? 1 : 0;
496 if (val && !capable(CAP_NET_ADMIN))
499 sock_valbool_flag(sk, SOCK_DBG, valbool);
502 sk->sk_reuse = valbool;
511 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
514 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
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_wmem_max)
523 val = sysctl_wmem_max;
525 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
526 if ((val * 2) < SOCK_MIN_SNDBUF)
527 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
529 sk->sk_sndbuf = val * 2;
532 * Wake up sending tasks if we
535 sk->sk_write_space(sk);
539 if (!capable(CAP_NET_ADMIN)) {
546 /* Don't error on this BSD doesn't and if you think
547 about it this is right. Otherwise apps have to
548 play 'guess the biggest size' games. RCVBUF/SNDBUF
549 are treated in BSD as hints */
551 if (val > sysctl_rmem_max)
552 val = sysctl_rmem_max;
554 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
556 * We double it on the way in to account for
557 * "struct sk_buff" etc. overhead. Applications
558 * assume that the SO_RCVBUF setting they make will
559 * allow that much actual data to be received on that
562 * Applications are unaware that "struct sk_buff" and
563 * other overheads allocate from the receive buffer
564 * during socket buffer allocation.
566 * And after considering the possible alternatives,
567 * returning the value we actually used in getsockopt
568 * is the most desirable behavior.
570 if ((val * 2) < SOCK_MIN_RCVBUF)
571 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
573 sk->sk_rcvbuf = val * 2;
577 if (!capable(CAP_NET_ADMIN)) {
585 if (sk->sk_protocol == IPPROTO_TCP)
586 tcp_set_keepalive(sk, valbool);
588 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
592 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
596 sk->sk_no_check = valbool;
600 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
601 sk->sk_priority = val;
607 if (optlen < sizeof(ling)) {
608 ret = -EINVAL; /* 1003.1g */
611 if (copy_from_user(&ling, optval, sizeof(ling))) {
616 sock_reset_flag(sk, SOCK_LINGER);
618 #if (BITS_PER_LONG == 32)
619 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
620 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
623 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
624 sock_set_flag(sk, SOCK_LINGER);
629 sock_warn_obsolete_bsdism("setsockopt");
634 set_bit(SOCK_PASSCRED, &sock->flags);
636 clear_bit(SOCK_PASSCRED, &sock->flags);
642 if (optname == SO_TIMESTAMP)
643 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
645 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
646 sock_set_flag(sk, SOCK_RCVTSTAMP);
647 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
649 sock_reset_flag(sk, SOCK_RCVTSTAMP);
650 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
654 case SO_TIMESTAMPING:
655 if (val & ~SOF_TIMESTAMPING_MASK) {
659 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
660 val & SOF_TIMESTAMPING_TX_HARDWARE);
661 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
662 val & SOF_TIMESTAMPING_TX_SOFTWARE);
663 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
664 val & SOF_TIMESTAMPING_RX_HARDWARE);
665 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
666 sock_enable_timestamp(sk,
667 SOCK_TIMESTAMPING_RX_SOFTWARE);
669 sock_disable_timestamp(sk,
670 SOCK_TIMESTAMPING_RX_SOFTWARE);
671 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
672 val & SOF_TIMESTAMPING_SOFTWARE);
673 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
674 val & SOF_TIMESTAMPING_SYS_HARDWARE);
675 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
676 val & SOF_TIMESTAMPING_RAW_HARDWARE);
682 sk->sk_rcvlowat = val ? : 1;
686 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
690 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
693 case SO_ATTACH_FILTER:
695 if (optlen == sizeof(struct sock_fprog)) {
696 struct sock_fprog fprog;
699 if (copy_from_user(&fprog, optval, sizeof(fprog)))
702 ret = sk_attach_filter(&fprog, sk);
706 case SO_DETACH_FILTER:
707 ret = sk_detach_filter(sk);
712 set_bit(SOCK_PASSSEC, &sock->flags);
714 clear_bit(SOCK_PASSSEC, &sock->flags);
717 if (!capable(CAP_NET_ADMIN))
723 /* We implement the SO_SNDLOWAT etc to
724 not be settable (1003.1g 5.3) */
727 sock_set_flag(sk, SOCK_RXQ_OVFL);
729 sock_reset_flag(sk, SOCK_RXQ_OVFL);
738 EXPORT_SYMBOL(sock_setsockopt);
741 int sock_getsockopt(struct socket *sock, int level, int optname,
742 char __user *optval, int __user *optlen)
744 struct sock *sk = sock->sk;
752 int lv = sizeof(int);
755 if (get_user(len, optlen))
760 memset(&v, 0, sizeof(v));
764 v.val = sock_flag(sk, SOCK_DBG);
768 v.val = sock_flag(sk, SOCK_LOCALROUTE);
772 v.val = !!sock_flag(sk, SOCK_BROADCAST);
776 v.val = sk->sk_sndbuf;
780 v.val = sk->sk_rcvbuf;
784 v.val = sk->sk_reuse;
788 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
796 v.val = sk->sk_protocol;
800 v.val = sk->sk_family;
804 v.val = -sock_error(sk);
806 v.val = xchg(&sk->sk_err_soft, 0);
810 v.val = !!sock_flag(sk, SOCK_URGINLINE);
814 v.val = sk->sk_no_check;
818 v.val = sk->sk_priority;
823 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
824 v.ling.l_linger = sk->sk_lingertime / HZ;
828 sock_warn_obsolete_bsdism("getsockopt");
832 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
833 !sock_flag(sk, SOCK_RCVTSTAMPNS);
837 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
840 case SO_TIMESTAMPING:
842 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
843 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
844 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
845 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
846 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
847 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
848 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
849 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
850 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
851 v.val |= SOF_TIMESTAMPING_SOFTWARE;
852 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
853 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
854 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
855 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
859 lv = sizeof(struct timeval);
860 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
864 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
865 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
870 lv = sizeof(struct timeval);
871 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
875 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
876 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
881 v.val = sk->sk_rcvlowat;
889 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
893 if (len > sizeof(sk->sk_peercred))
894 len = sizeof(sk->sk_peercred);
895 if (copy_to_user(optval, &sk->sk_peercred, len))
903 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
907 if (copy_to_user(optval, address, len))
912 /* Dubious BSD thing... Probably nobody even uses it, but
913 * the UNIX standard wants it for whatever reason... -DaveM
916 v.val = sk->sk_state == TCP_LISTEN;
920 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
924 return security_socket_getpeersec_stream(sock, optval, optlen, len);
931 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
940 if (copy_to_user(optval, &v, len))
943 if (put_user(len, optlen))
949 * Initialize an sk_lock.
951 * (We also register the sk_lock with the lock validator.)
953 static inline void sock_lock_init(struct sock *sk)
955 sock_lock_init_class_and_name(sk,
956 af_family_slock_key_strings[sk->sk_family],
957 af_family_slock_keys + sk->sk_family,
958 af_family_key_strings[sk->sk_family],
959 af_family_keys + sk->sk_family);
963 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
964 * even temporarly, because of RCU lookups. sk_node should also be left as is.
966 static void sock_copy(struct sock *nsk, const struct sock *osk)
968 #ifdef CONFIG_SECURITY_NETWORK
969 void *sptr = nsk->sk_security;
971 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
972 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
973 sizeof(osk->sk_tx_queue_mapping));
974 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
975 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
976 #ifdef CONFIG_SECURITY_NETWORK
977 nsk->sk_security = sptr;
978 security_sk_clone(osk, nsk);
982 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
986 struct kmem_cache *slab;
990 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
993 if (priority & __GFP_ZERO) {
995 * caches using SLAB_DESTROY_BY_RCU should let
996 * sk_node.next un-modified. Special care is taken
997 * when initializing object to zero.
999 if (offsetof(struct sock, sk_node.next) != 0)
1000 memset(sk, 0, offsetof(struct sock, sk_node.next));
1001 memset(&sk->sk_node.pprev, 0,
1002 prot->obj_size - offsetof(struct sock,
1007 sk = kmalloc(prot->obj_size, priority);
1010 kmemcheck_annotate_bitfield(sk, flags);
1012 if (security_sk_alloc(sk, family, priority))
1015 if (!try_module_get(prot->owner))
1017 sk_tx_queue_clear(sk);
1023 security_sk_free(sk);
1026 kmem_cache_free(slab, sk);
1032 static void sk_prot_free(struct proto *prot, struct sock *sk)
1034 struct kmem_cache *slab;
1035 struct module *owner;
1037 owner = prot->owner;
1040 security_sk_free(sk);
1042 kmem_cache_free(slab, sk);
1049 * sk_alloc - All socket objects are allocated here
1050 * @net: the applicable net namespace
1051 * @family: protocol family
1052 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1053 * @prot: struct proto associated with this new sock instance
1055 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1060 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1062 sk->sk_family = family;
1064 * See comment in struct sock definition to understand
1065 * why we need sk_prot_creator -acme
1067 sk->sk_prot = sk->sk_prot_creator = prot;
1069 sock_net_set(sk, get_net(net));
1070 atomic_set(&sk->sk_wmem_alloc, 1);
1075 EXPORT_SYMBOL(sk_alloc);
1077 static void __sk_free(struct sock *sk)
1079 struct sk_filter *filter;
1081 if (sk->sk_destruct)
1082 sk->sk_destruct(sk);
1084 filter = rcu_dereference_check(sk->sk_filter,
1085 atomic_read(&sk->sk_wmem_alloc) == 0);
1087 sk_filter_uncharge(sk, filter);
1088 rcu_assign_pointer(sk->sk_filter, NULL);
1091 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1092 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1094 if (atomic_read(&sk->sk_omem_alloc))
1095 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1096 __func__, atomic_read(&sk->sk_omem_alloc));
1098 put_net(sock_net(sk));
1099 sk_prot_free(sk->sk_prot_creator, sk);
1102 void sk_free(struct sock *sk)
1105 * We substract one from sk_wmem_alloc and can know if
1106 * some packets are still in some tx queue.
1107 * If not null, sock_wfree() will call __sk_free(sk) later
1109 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1112 EXPORT_SYMBOL(sk_free);
1115 * Last sock_put should drop referrence to sk->sk_net. It has already
1116 * been dropped in sk_change_net. Taking referrence to stopping namespace
1118 * Take referrence to a socket to remove it from hash _alive_ and after that
1119 * destroy it in the context of init_net.
1121 void sk_release_kernel(struct sock *sk)
1123 if (sk == NULL || sk->sk_socket == NULL)
1127 sock_release(sk->sk_socket);
1128 release_net(sock_net(sk));
1129 sock_net_set(sk, get_net(&init_net));
1132 EXPORT_SYMBOL(sk_release_kernel);
1134 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1138 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1139 if (newsk != NULL) {
1140 struct sk_filter *filter;
1142 sock_copy(newsk, sk);
1145 get_net(sock_net(newsk));
1146 sk_node_init(&newsk->sk_node);
1147 sock_lock_init(newsk);
1148 bh_lock_sock(newsk);
1149 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1150 newsk->sk_backlog.len = 0;
1152 atomic_set(&newsk->sk_rmem_alloc, 0);
1154 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1156 atomic_set(&newsk->sk_wmem_alloc, 1);
1157 atomic_set(&newsk->sk_omem_alloc, 0);
1158 skb_queue_head_init(&newsk->sk_receive_queue);
1159 skb_queue_head_init(&newsk->sk_write_queue);
1160 #ifdef CONFIG_NET_DMA
1161 skb_queue_head_init(&newsk->sk_async_wait_queue);
1164 spin_lock_init(&newsk->sk_dst_lock);
1165 rwlock_init(&newsk->sk_callback_lock);
1166 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1167 af_callback_keys + newsk->sk_family,
1168 af_family_clock_key_strings[newsk->sk_family]);
1170 newsk->sk_dst_cache = NULL;
1171 newsk->sk_wmem_queued = 0;
1172 newsk->sk_forward_alloc = 0;
1173 newsk->sk_send_head = NULL;
1174 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1176 sock_reset_flag(newsk, SOCK_DONE);
1177 skb_queue_head_init(&newsk->sk_error_queue);
1179 filter = newsk->sk_filter;
1181 sk_filter_charge(newsk, filter);
1183 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1184 /* It is still raw copy of parent, so invalidate
1185 * destructor and make plain sk_free() */
1186 newsk->sk_destruct = NULL;
1193 newsk->sk_priority = 0;
1195 * Before updating sk_refcnt, we must commit prior changes to memory
1196 * (Documentation/RCU/rculist_nulls.txt for details)
1199 atomic_set(&newsk->sk_refcnt, 2);
1202 * Increment the counter in the same struct proto as the master
1203 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1204 * is the same as sk->sk_prot->socks, as this field was copied
1207 * This _changes_ the previous behaviour, where
1208 * tcp_create_openreq_child always was incrementing the
1209 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1210 * to be taken into account in all callers. -acme
1212 sk_refcnt_debug_inc(newsk);
1213 sk_set_socket(newsk, NULL);
1214 newsk->sk_sleep = NULL;
1216 if (newsk->sk_prot->sockets_allocated)
1217 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1219 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1220 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1221 net_enable_timestamp();
1226 EXPORT_SYMBOL_GPL(sk_clone);
1228 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1230 __sk_dst_set(sk, dst);
1231 sk->sk_route_caps = dst->dev->features;
1232 if (sk->sk_route_caps & NETIF_F_GSO)
1233 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1234 if (sk_can_gso(sk)) {
1235 if (dst->header_len) {
1236 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1238 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1239 sk->sk_gso_max_size = dst->dev->gso_max_size;
1243 EXPORT_SYMBOL_GPL(sk_setup_caps);
1245 void __init sk_init(void)
1247 if (totalram_pages <= 4096) {
1248 sysctl_wmem_max = 32767;
1249 sysctl_rmem_max = 32767;
1250 sysctl_wmem_default = 32767;
1251 sysctl_rmem_default = 32767;
1252 } else if (totalram_pages >= 131072) {
1253 sysctl_wmem_max = 131071;
1254 sysctl_rmem_max = 131071;
1259 * Simple resource managers for sockets.
1264 * Write buffer destructor automatically called from kfree_skb.
1266 void sock_wfree(struct sk_buff *skb)
1268 struct sock *sk = skb->sk;
1269 unsigned int len = skb->truesize;
1271 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1273 * Keep a reference on sk_wmem_alloc, this will be released
1274 * after sk_write_space() call
1276 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1277 sk->sk_write_space(sk);
1281 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1282 * could not do because of in-flight packets
1284 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1287 EXPORT_SYMBOL(sock_wfree);
1290 * Read buffer destructor automatically called from kfree_skb.
1292 void sock_rfree(struct sk_buff *skb)
1294 struct sock *sk = skb->sk;
1296 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1297 sk_mem_uncharge(skb->sk, skb->truesize);
1299 EXPORT_SYMBOL(sock_rfree);
1302 int sock_i_uid(struct sock *sk)
1306 read_lock(&sk->sk_callback_lock);
1307 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1308 read_unlock(&sk->sk_callback_lock);
1311 EXPORT_SYMBOL(sock_i_uid);
1313 unsigned long sock_i_ino(struct sock *sk)
1317 read_lock(&sk->sk_callback_lock);
1318 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1319 read_unlock(&sk->sk_callback_lock);
1322 EXPORT_SYMBOL(sock_i_ino);
1325 * Allocate a skb from the socket's send buffer.
1327 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1330 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1331 struct sk_buff *skb = alloc_skb(size, priority);
1333 skb_set_owner_w(skb, sk);
1339 EXPORT_SYMBOL(sock_wmalloc);
1342 * Allocate a skb from the socket's receive buffer.
1344 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1347 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1348 struct sk_buff *skb = alloc_skb(size, priority);
1350 skb_set_owner_r(skb, sk);
1358 * Allocate a memory block from the socket's option memory buffer.
1360 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1362 if ((unsigned)size <= sysctl_optmem_max &&
1363 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1365 /* First do the add, to avoid the race if kmalloc
1368 atomic_add(size, &sk->sk_omem_alloc);
1369 mem = kmalloc(size, priority);
1372 atomic_sub(size, &sk->sk_omem_alloc);
1376 EXPORT_SYMBOL(sock_kmalloc);
1379 * Free an option memory block.
1381 void sock_kfree_s(struct sock *sk, void *mem, int size)
1384 atomic_sub(size, &sk->sk_omem_alloc);
1386 EXPORT_SYMBOL(sock_kfree_s);
1388 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1389 I think, these locks should be removed for datagram sockets.
1391 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1395 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1399 if (signal_pending(current))
1401 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1402 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1403 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1405 if (sk->sk_shutdown & SEND_SHUTDOWN)
1409 timeo = schedule_timeout(timeo);
1411 finish_wait(sk_sleep(sk), &wait);
1417 * Generic send/receive buffer handlers
1420 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1421 unsigned long data_len, int noblock,
1424 struct sk_buff *skb;
1429 gfp_mask = sk->sk_allocation;
1430 if (gfp_mask & __GFP_WAIT)
1431 gfp_mask |= __GFP_REPEAT;
1433 timeo = sock_sndtimeo(sk, noblock);
1435 err = sock_error(sk);
1440 if (sk->sk_shutdown & SEND_SHUTDOWN)
1443 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1444 skb = alloc_skb(header_len, gfp_mask);
1449 /* No pages, we're done... */
1453 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1454 skb->truesize += data_len;
1455 skb_shinfo(skb)->nr_frags = npages;
1456 for (i = 0; i < npages; i++) {
1460 page = alloc_pages(sk->sk_allocation, 0);
1463 skb_shinfo(skb)->nr_frags = i;
1468 frag = &skb_shinfo(skb)->frags[i];
1470 frag->page_offset = 0;
1471 frag->size = (data_len >= PAGE_SIZE ?
1474 data_len -= PAGE_SIZE;
1477 /* Full success... */
1483 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1484 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1488 if (signal_pending(current))
1490 timeo = sock_wait_for_wmem(sk, timeo);
1493 skb_set_owner_w(skb, sk);
1497 err = sock_intr_errno(timeo);
1502 EXPORT_SYMBOL(sock_alloc_send_pskb);
1504 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1505 int noblock, int *errcode)
1507 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1509 EXPORT_SYMBOL(sock_alloc_send_skb);
1511 static void __lock_sock(struct sock *sk)
1516 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1517 TASK_UNINTERRUPTIBLE);
1518 spin_unlock_bh(&sk->sk_lock.slock);
1520 spin_lock_bh(&sk->sk_lock.slock);
1521 if (!sock_owned_by_user(sk))
1524 finish_wait(&sk->sk_lock.wq, &wait);
1527 static void __release_sock(struct sock *sk)
1529 struct sk_buff *skb = sk->sk_backlog.head;
1532 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1536 struct sk_buff *next = skb->next;
1539 sk_backlog_rcv(sk, skb);
1542 * We are in process context here with softirqs
1543 * disabled, use cond_resched_softirq() to preempt.
1544 * This is safe to do because we've taken the backlog
1547 cond_resched_softirq();
1550 } while (skb != NULL);
1553 } while ((skb = sk->sk_backlog.head) != NULL);
1556 * Doing the zeroing here guarantee we can not loop forever
1557 * while a wild producer attempts to flood us.
1559 sk->sk_backlog.len = 0;
1563 * sk_wait_data - wait for data to arrive at sk_receive_queue
1564 * @sk: sock to wait on
1565 * @timeo: for how long
1567 * Now socket state including sk->sk_err is changed only under lock,
1568 * hence we may omit checks after joining wait queue.
1569 * We check receive queue before schedule() only as optimization;
1570 * it is very likely that release_sock() added new data.
1572 int sk_wait_data(struct sock *sk, long *timeo)
1577 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1578 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1579 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1580 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1581 finish_wait(sk_sleep(sk), &wait);
1584 EXPORT_SYMBOL(sk_wait_data);
1587 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1589 * @size: memory size to allocate
1590 * @kind: allocation type
1592 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1593 * rmem allocation. This function assumes that protocols which have
1594 * memory_pressure use sk_wmem_queued as write buffer accounting.
1596 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1598 struct proto *prot = sk->sk_prot;
1599 int amt = sk_mem_pages(size);
1602 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1603 allocated = atomic_add_return(amt, prot->memory_allocated);
1606 if (allocated <= prot->sysctl_mem[0]) {
1607 if (prot->memory_pressure && *prot->memory_pressure)
1608 *prot->memory_pressure = 0;
1612 /* Under pressure. */
1613 if (allocated > prot->sysctl_mem[1])
1614 if (prot->enter_memory_pressure)
1615 prot->enter_memory_pressure(sk);
1617 /* Over hard limit. */
1618 if (allocated > prot->sysctl_mem[2])
1619 goto suppress_allocation;
1621 /* guarantee minimum buffer size under pressure */
1622 if (kind == SK_MEM_RECV) {
1623 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1625 } else { /* SK_MEM_SEND */
1626 if (sk->sk_type == SOCK_STREAM) {
1627 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1629 } else if (atomic_read(&sk->sk_wmem_alloc) <
1630 prot->sysctl_wmem[0])
1634 if (prot->memory_pressure) {
1637 if (!*prot->memory_pressure)
1639 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1640 if (prot->sysctl_mem[2] > alloc *
1641 sk_mem_pages(sk->sk_wmem_queued +
1642 atomic_read(&sk->sk_rmem_alloc) +
1643 sk->sk_forward_alloc))
1647 suppress_allocation:
1649 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1650 sk_stream_moderate_sndbuf(sk);
1652 /* Fail only if socket is _under_ its sndbuf.
1653 * In this case we cannot block, so that we have to fail.
1655 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1659 /* Alas. Undo changes. */
1660 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1661 atomic_sub(amt, prot->memory_allocated);
1664 EXPORT_SYMBOL(__sk_mem_schedule);
1667 * __sk_reclaim - reclaim memory_allocated
1670 void __sk_mem_reclaim(struct sock *sk)
1672 struct proto *prot = sk->sk_prot;
1674 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1675 prot->memory_allocated);
1676 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1678 if (prot->memory_pressure && *prot->memory_pressure &&
1679 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1680 *prot->memory_pressure = 0;
1682 EXPORT_SYMBOL(__sk_mem_reclaim);
1686 * Set of default routines for initialising struct proto_ops when
1687 * the protocol does not support a particular function. In certain
1688 * cases where it makes no sense for a protocol to have a "do nothing"
1689 * function, some default processing is provided.
1692 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1696 EXPORT_SYMBOL(sock_no_bind);
1698 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1703 EXPORT_SYMBOL(sock_no_connect);
1705 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1709 EXPORT_SYMBOL(sock_no_socketpair);
1711 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1715 EXPORT_SYMBOL(sock_no_accept);
1717 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1722 EXPORT_SYMBOL(sock_no_getname);
1724 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1728 EXPORT_SYMBOL(sock_no_poll);
1730 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1734 EXPORT_SYMBOL(sock_no_ioctl);
1736 int sock_no_listen(struct socket *sock, int backlog)
1740 EXPORT_SYMBOL(sock_no_listen);
1742 int sock_no_shutdown(struct socket *sock, int how)
1746 EXPORT_SYMBOL(sock_no_shutdown);
1748 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1749 char __user *optval, unsigned int optlen)
1753 EXPORT_SYMBOL(sock_no_setsockopt);
1755 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1756 char __user *optval, int __user *optlen)
1760 EXPORT_SYMBOL(sock_no_getsockopt);
1762 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1767 EXPORT_SYMBOL(sock_no_sendmsg);
1769 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1770 size_t len, int flags)
1774 EXPORT_SYMBOL(sock_no_recvmsg);
1776 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1778 /* Mirror missing mmap method error code */
1781 EXPORT_SYMBOL(sock_no_mmap);
1783 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1786 struct msghdr msg = {.msg_flags = flags};
1788 char *kaddr = kmap(page);
1789 iov.iov_base = kaddr + offset;
1791 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1795 EXPORT_SYMBOL(sock_no_sendpage);
1798 * Default Socket Callbacks
1801 static void sock_def_wakeup(struct sock *sk)
1803 read_lock(&sk->sk_callback_lock);
1804 if (sk_has_sleeper(sk))
1805 wake_up_interruptible_all(sk_sleep(sk));
1806 read_unlock(&sk->sk_callback_lock);
1809 static void sock_def_error_report(struct sock *sk)
1811 read_lock(&sk->sk_callback_lock);
1812 if (sk_has_sleeper(sk))
1813 wake_up_interruptible_poll(sk_sleep(sk), POLLERR);
1814 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1815 read_unlock(&sk->sk_callback_lock);
1818 static void sock_def_readable(struct sock *sk, int len)
1820 read_lock(&sk->sk_callback_lock);
1821 if (sk_has_sleeper(sk))
1822 wake_up_interruptible_sync_poll(sk_sleep(sk), POLLIN |
1823 POLLRDNORM | POLLRDBAND);
1824 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1825 read_unlock(&sk->sk_callback_lock);
1828 static void sock_def_write_space(struct sock *sk)
1830 read_lock(&sk->sk_callback_lock);
1832 /* Do not wake up a writer until he can make "significant"
1835 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1836 if (sk_has_sleeper(sk))
1837 wake_up_interruptible_sync_poll(sk_sleep(sk), POLLOUT |
1838 POLLWRNORM | POLLWRBAND);
1840 /* Should agree with poll, otherwise some programs break */
1841 if (sock_writeable(sk))
1842 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1845 read_unlock(&sk->sk_callback_lock);
1848 static void sock_def_destruct(struct sock *sk)
1850 kfree(sk->sk_protinfo);
1853 void sk_send_sigurg(struct sock *sk)
1855 if (sk->sk_socket && sk->sk_socket->file)
1856 if (send_sigurg(&sk->sk_socket->file->f_owner))
1857 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1859 EXPORT_SYMBOL(sk_send_sigurg);
1861 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1862 unsigned long expires)
1864 if (!mod_timer(timer, expires))
1867 EXPORT_SYMBOL(sk_reset_timer);
1869 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1871 if (timer_pending(timer) && del_timer(timer))
1874 EXPORT_SYMBOL(sk_stop_timer);
1876 void sock_init_data(struct socket *sock, struct sock *sk)
1878 skb_queue_head_init(&sk->sk_receive_queue);
1879 skb_queue_head_init(&sk->sk_write_queue);
1880 skb_queue_head_init(&sk->sk_error_queue);
1881 #ifdef CONFIG_NET_DMA
1882 skb_queue_head_init(&sk->sk_async_wait_queue);
1885 sk->sk_send_head = NULL;
1887 init_timer(&sk->sk_timer);
1889 sk->sk_allocation = GFP_KERNEL;
1890 sk->sk_rcvbuf = sysctl_rmem_default;
1891 sk->sk_sndbuf = sysctl_wmem_default;
1892 sk->sk_state = TCP_CLOSE;
1893 sk_set_socket(sk, sock);
1895 sock_set_flag(sk, SOCK_ZAPPED);
1898 sk->sk_type = sock->type;
1899 sk->sk_sleep = &sock->wait;
1902 sk->sk_sleep = NULL;
1904 spin_lock_init(&sk->sk_dst_lock);
1905 rwlock_init(&sk->sk_callback_lock);
1906 lockdep_set_class_and_name(&sk->sk_callback_lock,
1907 af_callback_keys + sk->sk_family,
1908 af_family_clock_key_strings[sk->sk_family]);
1910 sk->sk_state_change = sock_def_wakeup;
1911 sk->sk_data_ready = sock_def_readable;
1912 sk->sk_write_space = sock_def_write_space;
1913 sk->sk_error_report = sock_def_error_report;
1914 sk->sk_destruct = sock_def_destruct;
1916 sk->sk_sndmsg_page = NULL;
1917 sk->sk_sndmsg_off = 0;
1919 sk->sk_peercred.pid = 0;
1920 sk->sk_peercred.uid = -1;
1921 sk->sk_peercred.gid = -1;
1922 sk->sk_write_pending = 0;
1923 sk->sk_rcvlowat = 1;
1924 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1925 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1927 sk->sk_stamp = ktime_set(-1L, 0);
1930 * Before updating sk_refcnt, we must commit prior changes to memory
1931 * (Documentation/RCU/rculist_nulls.txt for details)
1934 atomic_set(&sk->sk_refcnt, 1);
1935 atomic_set(&sk->sk_drops, 0);
1937 EXPORT_SYMBOL(sock_init_data);
1939 void lock_sock_nested(struct sock *sk, int subclass)
1942 spin_lock_bh(&sk->sk_lock.slock);
1943 if (sk->sk_lock.owned)
1945 sk->sk_lock.owned = 1;
1946 spin_unlock(&sk->sk_lock.slock);
1948 * The sk_lock has mutex_lock() semantics here:
1950 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1953 EXPORT_SYMBOL(lock_sock_nested);
1955 void release_sock(struct sock *sk)
1958 * The sk_lock has mutex_unlock() semantics:
1960 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1962 spin_lock_bh(&sk->sk_lock.slock);
1963 if (sk->sk_backlog.tail)
1965 sk->sk_lock.owned = 0;
1966 if (waitqueue_active(&sk->sk_lock.wq))
1967 wake_up(&sk->sk_lock.wq);
1968 spin_unlock_bh(&sk->sk_lock.slock);
1970 EXPORT_SYMBOL(release_sock);
1972 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1975 if (!sock_flag(sk, SOCK_TIMESTAMP))
1976 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1977 tv = ktime_to_timeval(sk->sk_stamp);
1978 if (tv.tv_sec == -1)
1980 if (tv.tv_sec == 0) {
1981 sk->sk_stamp = ktime_get_real();
1982 tv = ktime_to_timeval(sk->sk_stamp);
1984 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1986 EXPORT_SYMBOL(sock_get_timestamp);
1988 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1991 if (!sock_flag(sk, SOCK_TIMESTAMP))
1992 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1993 ts = ktime_to_timespec(sk->sk_stamp);
1994 if (ts.tv_sec == -1)
1996 if (ts.tv_sec == 0) {
1997 sk->sk_stamp = ktime_get_real();
1998 ts = ktime_to_timespec(sk->sk_stamp);
2000 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2002 EXPORT_SYMBOL(sock_get_timestampns);
2004 void sock_enable_timestamp(struct sock *sk, int flag)
2006 if (!sock_flag(sk, flag)) {
2007 sock_set_flag(sk, flag);
2009 * we just set one of the two flags which require net
2010 * time stamping, but time stamping might have been on
2011 * already because of the other one
2014 flag == SOCK_TIMESTAMP ?
2015 SOCK_TIMESTAMPING_RX_SOFTWARE :
2017 net_enable_timestamp();
2022 * Get a socket option on an socket.
2024 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2025 * asynchronous errors should be reported by getsockopt. We assume
2026 * this means if you specify SO_ERROR (otherwise whats the point of it).
2028 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2029 char __user *optval, int __user *optlen)
2031 struct sock *sk = sock->sk;
2033 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2035 EXPORT_SYMBOL(sock_common_getsockopt);
2037 #ifdef CONFIG_COMPAT
2038 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2039 char __user *optval, int __user *optlen)
2041 struct sock *sk = sock->sk;
2043 if (sk->sk_prot->compat_getsockopt != NULL)
2044 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2046 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2048 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2051 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2052 struct msghdr *msg, size_t size, int flags)
2054 struct sock *sk = sock->sk;
2058 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2059 flags & ~MSG_DONTWAIT, &addr_len);
2061 msg->msg_namelen = addr_len;
2064 EXPORT_SYMBOL(sock_common_recvmsg);
2067 * Set socket options on an inet socket.
2069 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2070 char __user *optval, unsigned int optlen)
2072 struct sock *sk = sock->sk;
2074 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2076 EXPORT_SYMBOL(sock_common_setsockopt);
2078 #ifdef CONFIG_COMPAT
2079 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2080 char __user *optval, unsigned int optlen)
2082 struct sock *sk = sock->sk;
2084 if (sk->sk_prot->compat_setsockopt != NULL)
2085 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2087 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2089 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2092 void sk_common_release(struct sock *sk)
2094 if (sk->sk_prot->destroy)
2095 sk->sk_prot->destroy(sk);
2098 * Observation: when sock_common_release is called, processes have
2099 * no access to socket. But net still has.
2100 * Step one, detach it from networking:
2102 * A. Remove from hash tables.
2105 sk->sk_prot->unhash(sk);
2108 * In this point socket cannot receive new packets, but it is possible
2109 * that some packets are in flight because some CPU runs receiver and
2110 * did hash table lookup before we unhashed socket. They will achieve
2111 * receive queue and will be purged by socket destructor.
2113 * Also we still have packets pending on receive queue and probably,
2114 * our own packets waiting in device queues. sock_destroy will drain
2115 * receive queue, but transmitted packets will delay socket destruction
2116 * until the last reference will be released.
2121 xfrm_sk_free_policy(sk);
2123 sk_refcnt_debug_release(sk);
2126 EXPORT_SYMBOL(sk_common_release);
2128 static DEFINE_RWLOCK(proto_list_lock);
2129 static LIST_HEAD(proto_list);
2131 #ifdef CONFIG_PROC_FS
2132 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2134 int val[PROTO_INUSE_NR];
2137 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2139 #ifdef CONFIG_NET_NS
2140 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2142 int cpu = smp_processor_id();
2143 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2145 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2147 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2149 int cpu, idx = prot->inuse_idx;
2152 for_each_possible_cpu(cpu)
2153 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2155 return res >= 0 ? res : 0;
2157 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2159 static int __net_init sock_inuse_init_net(struct net *net)
2161 net->core.inuse = alloc_percpu(struct prot_inuse);
2162 return net->core.inuse ? 0 : -ENOMEM;
2165 static void __net_exit sock_inuse_exit_net(struct net *net)
2167 free_percpu(net->core.inuse);
2170 static struct pernet_operations net_inuse_ops = {
2171 .init = sock_inuse_init_net,
2172 .exit = sock_inuse_exit_net,
2175 static __init int net_inuse_init(void)
2177 if (register_pernet_subsys(&net_inuse_ops))
2178 panic("Cannot initialize net inuse counters");
2183 core_initcall(net_inuse_init);
2185 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2187 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2189 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2191 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2193 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2195 int cpu, idx = prot->inuse_idx;
2198 for_each_possible_cpu(cpu)
2199 res += per_cpu(prot_inuse, cpu).val[idx];
2201 return res >= 0 ? res : 0;
2203 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2206 static void assign_proto_idx(struct proto *prot)
2208 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2210 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2211 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2215 set_bit(prot->inuse_idx, proto_inuse_idx);
2218 static void release_proto_idx(struct proto *prot)
2220 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2221 clear_bit(prot->inuse_idx, proto_inuse_idx);
2224 static inline void assign_proto_idx(struct proto *prot)
2228 static inline void release_proto_idx(struct proto *prot)
2233 int proto_register(struct proto *prot, int alloc_slab)
2236 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2237 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2240 if (prot->slab == NULL) {
2241 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2246 if (prot->rsk_prot != NULL) {
2247 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2248 if (prot->rsk_prot->slab_name == NULL)
2249 goto out_free_sock_slab;
2251 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2252 prot->rsk_prot->obj_size, 0,
2253 SLAB_HWCACHE_ALIGN, NULL);
2255 if (prot->rsk_prot->slab == NULL) {
2256 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2258 goto out_free_request_sock_slab_name;
2262 if (prot->twsk_prot != NULL) {
2263 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2265 if (prot->twsk_prot->twsk_slab_name == NULL)
2266 goto out_free_request_sock_slab;
2268 prot->twsk_prot->twsk_slab =
2269 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2270 prot->twsk_prot->twsk_obj_size,
2272 SLAB_HWCACHE_ALIGN |
2275 if (prot->twsk_prot->twsk_slab == NULL)
2276 goto out_free_timewait_sock_slab_name;
2280 write_lock(&proto_list_lock);
2281 list_add(&prot->node, &proto_list);
2282 assign_proto_idx(prot);
2283 write_unlock(&proto_list_lock);
2286 out_free_timewait_sock_slab_name:
2287 kfree(prot->twsk_prot->twsk_slab_name);
2288 out_free_request_sock_slab:
2289 if (prot->rsk_prot && prot->rsk_prot->slab) {
2290 kmem_cache_destroy(prot->rsk_prot->slab);
2291 prot->rsk_prot->slab = NULL;
2293 out_free_request_sock_slab_name:
2295 kfree(prot->rsk_prot->slab_name);
2297 kmem_cache_destroy(prot->slab);
2302 EXPORT_SYMBOL(proto_register);
2304 void proto_unregister(struct proto *prot)
2306 write_lock(&proto_list_lock);
2307 release_proto_idx(prot);
2308 list_del(&prot->node);
2309 write_unlock(&proto_list_lock);
2311 if (prot->slab != NULL) {
2312 kmem_cache_destroy(prot->slab);
2316 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2317 kmem_cache_destroy(prot->rsk_prot->slab);
2318 kfree(prot->rsk_prot->slab_name);
2319 prot->rsk_prot->slab = NULL;
2322 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2323 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2324 kfree(prot->twsk_prot->twsk_slab_name);
2325 prot->twsk_prot->twsk_slab = NULL;
2328 EXPORT_SYMBOL(proto_unregister);
2330 #ifdef CONFIG_PROC_FS
2331 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2332 __acquires(proto_list_lock)
2334 read_lock(&proto_list_lock);
2335 return seq_list_start_head(&proto_list, *pos);
2338 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2340 return seq_list_next(v, &proto_list, pos);
2343 static void proto_seq_stop(struct seq_file *seq, void *v)
2344 __releases(proto_list_lock)
2346 read_unlock(&proto_list_lock);
2349 static char proto_method_implemented(const void *method)
2351 return method == NULL ? 'n' : 'y';
2354 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2356 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2357 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2360 sock_prot_inuse_get(seq_file_net(seq), proto),
2361 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2362 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2364 proto->slab == NULL ? "no" : "yes",
2365 module_name(proto->owner),
2366 proto_method_implemented(proto->close),
2367 proto_method_implemented(proto->connect),
2368 proto_method_implemented(proto->disconnect),
2369 proto_method_implemented(proto->accept),
2370 proto_method_implemented(proto->ioctl),
2371 proto_method_implemented(proto->init),
2372 proto_method_implemented(proto->destroy),
2373 proto_method_implemented(proto->shutdown),
2374 proto_method_implemented(proto->setsockopt),
2375 proto_method_implemented(proto->getsockopt),
2376 proto_method_implemented(proto->sendmsg),
2377 proto_method_implemented(proto->recvmsg),
2378 proto_method_implemented(proto->sendpage),
2379 proto_method_implemented(proto->bind),
2380 proto_method_implemented(proto->backlog_rcv),
2381 proto_method_implemented(proto->hash),
2382 proto_method_implemented(proto->unhash),
2383 proto_method_implemented(proto->get_port),
2384 proto_method_implemented(proto->enter_memory_pressure));
2387 static int proto_seq_show(struct seq_file *seq, void *v)
2389 if (v == &proto_list)
2390 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2399 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2401 proto_seq_printf(seq, list_entry(v, struct proto, node));
2405 static const struct seq_operations proto_seq_ops = {
2406 .start = proto_seq_start,
2407 .next = proto_seq_next,
2408 .stop = proto_seq_stop,
2409 .show = proto_seq_show,
2412 static int proto_seq_open(struct inode *inode, struct file *file)
2414 return seq_open_net(inode, file, &proto_seq_ops,
2415 sizeof(struct seq_net_private));
2418 static const struct file_operations proto_seq_fops = {
2419 .owner = THIS_MODULE,
2420 .open = proto_seq_open,
2422 .llseek = seq_lseek,
2423 .release = seq_release_net,
2426 static __net_init int proto_init_net(struct net *net)
2428 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2434 static __net_exit void proto_exit_net(struct net *net)
2436 proc_net_remove(net, "protocols");
2440 static __net_initdata struct pernet_operations proto_net_ops = {
2441 .init = proto_init_net,
2442 .exit = proto_exit_net,
2445 static int __init proto_init(void)
2447 return register_pernet_subsys(&proto_net_ops);
2450 subsys_initcall(proto_init);
2452 #endif /* PROC_FS */