2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/request_sock.h>
123 #include <net/sock.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];
140 #ifdef CONFIG_DEBUG_LOCK_ALLOC
142 * Make lock validator output more readable. (we pre-construct these
143 * strings build-time, so that runtime initialization of socket
146 static const char *af_family_key_strings[AF_MAX+1] = {
147 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
148 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
149 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
150 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
151 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
152 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
153 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
154 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
155 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
156 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "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-29" ,
170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX"
175 * sk_callback_lock locking rules are per-address-family,
176 * so split the lock classes by using a per-AF key:
178 static struct lock_class_key af_callback_keys[AF_MAX];
180 /* Take into consideration the size of the struct sk_buff overhead in the
181 * determination of these values, since that is non-constant across
182 * platforms. This makes socket queueing behavior and performance
183 * not depend upon such differences.
185 #define _SK_MEM_PACKETS 256
186 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
187 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
188 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
190 /* Run time adjustable parameters. */
191 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
192 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
193 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
194 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
196 /* Maximal space eaten by iovec or ancilliary data plus some space */
197 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
199 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
203 if (optlen < sizeof(tv))
205 if (copy_from_user(&tv, optval, sizeof(tv)))
208 *timeo_p = MAX_SCHEDULE_TIMEOUT;
209 if (tv.tv_sec == 0 && tv.tv_usec == 0)
211 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
212 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
216 static void sock_warn_obsolete_bsdism(const char *name)
219 static char warncomm[TASK_COMM_LEN];
220 if (strcmp(warncomm, current->comm) && warned < 5) {
221 strcpy(warncomm, current->comm);
222 printk(KERN_WARNING "process `%s' is using obsolete "
223 "%s SO_BSDCOMPAT\n", warncomm, name);
228 static void sock_disable_timestamp(struct sock *sk)
230 if (sock_flag(sk, SOCK_TIMESTAMP)) {
231 sock_reset_flag(sk, SOCK_TIMESTAMP);
232 net_disable_timestamp();
237 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
242 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
243 number of warnings when compiling with -W --ANK
245 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
246 (unsigned)sk->sk_rcvbuf) {
251 err = sk_filter(sk, skb);
256 skb_set_owner_r(skb, sk);
258 /* Cache the SKB length before we tack it onto the receive
259 * queue. Once it is added it no longer belongs to us and
260 * may be freed by other threads of control pulling packets
265 skb_queue_tail(&sk->sk_receive_queue, skb);
267 if (!sock_flag(sk, SOCK_DEAD))
268 sk->sk_data_ready(sk, skb_len);
272 EXPORT_SYMBOL(sock_queue_rcv_skb);
274 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
276 int rc = NET_RX_SUCCESS;
278 if (sk_filter(sk, skb))
279 goto discard_and_relse;
284 bh_lock_sock_nested(sk);
287 if (!sock_owned_by_user(sk)) {
289 * trylock + unlock semantics:
291 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
293 rc = sk->sk_backlog_rcv(sk, skb);
295 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
297 sk_add_backlog(sk, skb);
306 EXPORT_SYMBOL(sk_receive_skb);
308 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
310 struct dst_entry *dst = sk->sk_dst_cache;
312 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
313 sk->sk_dst_cache = NULL;
320 EXPORT_SYMBOL(__sk_dst_check);
322 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
324 struct dst_entry *dst = sk_dst_get(sk);
326 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
334 EXPORT_SYMBOL(sk_dst_check);
337 * This is meant for all protocols to use and covers goings on
338 * at the socket level. Everything here is generic.
341 int sock_setsockopt(struct socket *sock, int level, int optname,
342 char __user *optval, int optlen)
344 struct sock *sk=sock->sk;
345 struct sk_filter *filter;
352 * Options without arguments
355 #ifdef SO_DONTLINGER /* Compatibility item... */
356 if (optname == SO_DONTLINGER) {
358 sock_reset_flag(sk, SOCK_LINGER);
364 if (optlen < sizeof(int))
367 if (get_user(val, (int __user *)optval))
376 if (val && !capable(CAP_NET_ADMIN)) {
380 sock_set_flag(sk, SOCK_DBG);
382 sock_reset_flag(sk, SOCK_DBG);
385 sk->sk_reuse = valbool;
393 sock_set_flag(sk, SOCK_LOCALROUTE);
395 sock_reset_flag(sk, SOCK_LOCALROUTE);
398 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
401 /* Don't error on this BSD doesn't and if you think
402 about it this is right. Otherwise apps have to
403 play 'guess the biggest size' games. RCVBUF/SNDBUF
404 are treated in BSD as hints */
406 if (val > sysctl_wmem_max)
407 val = sysctl_wmem_max;
409 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
410 if ((val * 2) < SOCK_MIN_SNDBUF)
411 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
413 sk->sk_sndbuf = val * 2;
416 * Wake up sending tasks if we
419 sk->sk_write_space(sk);
423 if (!capable(CAP_NET_ADMIN)) {
430 /* Don't error on this BSD doesn't and if you think
431 about it this is right. Otherwise apps have to
432 play 'guess the biggest size' games. RCVBUF/SNDBUF
433 are treated in BSD as hints */
435 if (val > sysctl_rmem_max)
436 val = sysctl_rmem_max;
438 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
440 * We double it on the way in to account for
441 * "struct sk_buff" etc. overhead. Applications
442 * assume that the SO_RCVBUF setting they make will
443 * allow that much actual data to be received on that
446 * Applications are unaware that "struct sk_buff" and
447 * other overheads allocate from the receive buffer
448 * during socket buffer allocation.
450 * And after considering the possible alternatives,
451 * returning the value we actually used in getsockopt
452 * is the most desirable behavior.
454 if ((val * 2) < SOCK_MIN_RCVBUF)
455 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
457 sk->sk_rcvbuf = val * 2;
461 if (!capable(CAP_NET_ADMIN)) {
469 if (sk->sk_protocol == IPPROTO_TCP)
470 tcp_set_keepalive(sk, valbool);
472 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
476 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
480 sk->sk_no_check = valbool;
484 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
485 sk->sk_priority = val;
491 if (optlen < sizeof(ling)) {
492 ret = -EINVAL; /* 1003.1g */
495 if (copy_from_user(&ling,optval,sizeof(ling))) {
500 sock_reset_flag(sk, SOCK_LINGER);
502 #if (BITS_PER_LONG == 32)
503 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
504 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
507 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
508 sock_set_flag(sk, SOCK_LINGER);
513 sock_warn_obsolete_bsdism("setsockopt");
518 set_bit(SOCK_PASSCRED, &sock->flags);
520 clear_bit(SOCK_PASSCRED, &sock->flags);
525 sock_set_flag(sk, SOCK_RCVTSTAMP);
526 sock_enable_timestamp(sk);
528 sock_reset_flag(sk, SOCK_RCVTSTAMP);
534 sk->sk_rcvlowat = val ? : 1;
538 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
542 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
545 #ifdef CONFIG_NETDEVICES
546 case SO_BINDTODEVICE:
548 char devname[IFNAMSIZ];
551 if (!capable(CAP_NET_RAW)) {
556 /* Bind this socket to a particular device like "eth0",
557 * as specified in the passed interface name. If the
558 * name is "" or the option length is zero the socket
563 sk->sk_bound_dev_if = 0;
565 if (optlen > IFNAMSIZ - 1)
566 optlen = IFNAMSIZ - 1;
567 memset(devname, 0, sizeof(devname));
568 if (copy_from_user(devname, optval, optlen)) {
573 /* Remove any cached route for this socket. */
576 if (devname[0] == '\0') {
577 sk->sk_bound_dev_if = 0;
579 struct net_device *dev = dev_get_by_name(devname);
584 sk->sk_bound_dev_if = dev->ifindex;
593 case SO_ATTACH_FILTER:
595 if (optlen == sizeof(struct sock_fprog)) {
596 struct sock_fprog fprog;
599 if (copy_from_user(&fprog, optval, sizeof(fprog)))
602 ret = sk_attach_filter(&fprog, sk);
606 case SO_DETACH_FILTER:
608 filter = rcu_dereference(sk->sk_filter);
610 rcu_assign_pointer(sk->sk_filter, NULL);
611 sk_filter_release(sk, filter);
612 rcu_read_unlock_bh();
615 rcu_read_unlock_bh();
621 set_bit(SOCK_PASSSEC, &sock->flags);
623 clear_bit(SOCK_PASSSEC, &sock->flags);
626 /* We implement the SO_SNDLOWAT etc to
627 not be settable (1003.1g 5.3) */
637 int sock_getsockopt(struct socket *sock, int level, int optname,
638 char __user *optval, int __user *optlen)
640 struct sock *sk = sock->sk;
648 unsigned int lv = sizeof(int);
651 if (get_user(len, optlen))
658 v.val = sock_flag(sk, SOCK_DBG);
662 v.val = sock_flag(sk, SOCK_LOCALROUTE);
666 v.val = !!sock_flag(sk, SOCK_BROADCAST);
670 v.val = sk->sk_sndbuf;
674 v.val = sk->sk_rcvbuf;
678 v.val = sk->sk_reuse;
682 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
690 v.val = -sock_error(sk);
692 v.val = xchg(&sk->sk_err_soft, 0);
696 v.val = !!sock_flag(sk, SOCK_URGINLINE);
700 v.val = sk->sk_no_check;
704 v.val = sk->sk_priority;
709 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
710 v.ling.l_linger = sk->sk_lingertime / HZ;
714 sock_warn_obsolete_bsdism("getsockopt");
718 v.val = sock_flag(sk, SOCK_RCVTSTAMP);
722 lv=sizeof(struct timeval);
723 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
727 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
728 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
733 lv=sizeof(struct timeval);
734 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
738 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
739 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
744 v.val = sk->sk_rcvlowat;
752 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
756 if (len > sizeof(sk->sk_peercred))
757 len = sizeof(sk->sk_peercred);
758 if (copy_to_user(optval, &sk->sk_peercred, len))
766 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
770 if (copy_to_user(optval, address, len))
775 /* Dubious BSD thing... Probably nobody even uses it, but
776 * the UNIX standard wants it for whatever reason... -DaveM
779 v.val = sk->sk_state == TCP_LISTEN;
783 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
787 return security_socket_getpeersec_stream(sock, optval, optlen, len);
795 if (copy_to_user(optval, &v, len))
798 if (put_user(len, optlen))
804 * Initialize an sk_lock.
806 * (We also register the sk_lock with the lock validator.)
808 static inline void sock_lock_init(struct sock *sk)
810 sock_lock_init_class_and_name(sk,
811 af_family_slock_key_strings[sk->sk_family],
812 af_family_slock_keys + sk->sk_family,
813 af_family_key_strings[sk->sk_family],
814 af_family_keys + sk->sk_family);
818 * sk_alloc - All socket objects are allocated here
819 * @family: protocol family
820 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
821 * @prot: struct proto associated with this new sock instance
822 * @zero_it: if we should zero the newly allocated sock
824 struct sock *sk_alloc(int family, gfp_t priority,
825 struct proto *prot, int zero_it)
827 struct sock *sk = NULL;
828 struct kmem_cache *slab = prot->slab;
831 sk = kmem_cache_alloc(slab, priority);
833 sk = kmalloc(prot->obj_size, priority);
837 memset(sk, 0, prot->obj_size);
838 sk->sk_family = family;
840 * See comment in struct sock definition to understand
841 * why we need sk_prot_creator -acme
843 sk->sk_prot = sk->sk_prot_creator = prot;
847 if (security_sk_alloc(sk, family, priority))
850 if (!try_module_get(prot->owner))
857 kmem_cache_free(slab, sk);
863 void sk_free(struct sock *sk)
865 struct sk_filter *filter;
866 struct module *owner = sk->sk_prot_creator->owner;
871 filter = rcu_dereference(sk->sk_filter);
873 sk_filter_release(sk, filter);
874 rcu_assign_pointer(sk->sk_filter, NULL);
877 sock_disable_timestamp(sk);
879 if (atomic_read(&sk->sk_omem_alloc))
880 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
881 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
883 security_sk_free(sk);
884 if (sk->sk_prot_creator->slab != NULL)
885 kmem_cache_free(sk->sk_prot_creator->slab, sk);
891 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
893 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
896 struct sk_filter *filter;
898 sock_copy(newsk, sk);
901 sk_node_init(&newsk->sk_node);
902 sock_lock_init(newsk);
904 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
906 atomic_set(&newsk->sk_rmem_alloc, 0);
907 atomic_set(&newsk->sk_wmem_alloc, 0);
908 atomic_set(&newsk->sk_omem_alloc, 0);
909 skb_queue_head_init(&newsk->sk_receive_queue);
910 skb_queue_head_init(&newsk->sk_write_queue);
911 #ifdef CONFIG_NET_DMA
912 skb_queue_head_init(&newsk->sk_async_wait_queue);
915 rwlock_init(&newsk->sk_dst_lock);
916 rwlock_init(&newsk->sk_callback_lock);
917 lockdep_set_class(&newsk->sk_callback_lock,
918 af_callback_keys + newsk->sk_family);
920 newsk->sk_dst_cache = NULL;
921 newsk->sk_wmem_queued = 0;
922 newsk->sk_forward_alloc = 0;
923 newsk->sk_send_head = NULL;
924 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
926 sock_reset_flag(newsk, SOCK_DONE);
927 skb_queue_head_init(&newsk->sk_error_queue);
929 filter = newsk->sk_filter;
931 sk_filter_charge(newsk, filter);
933 if (unlikely(xfrm_sk_clone_policy(newsk))) {
934 /* It is still raw copy of parent, so invalidate
935 * destructor and make plain sk_free() */
936 newsk->sk_destruct = NULL;
943 newsk->sk_priority = 0;
944 atomic_set(&newsk->sk_refcnt, 2);
947 * Increment the counter in the same struct proto as the master
948 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
949 * is the same as sk->sk_prot->socks, as this field was copied
952 * This _changes_ the previous behaviour, where
953 * tcp_create_openreq_child always was incrementing the
954 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
955 * to be taken into account in all callers. -acme
957 sk_refcnt_debug_inc(newsk);
958 newsk->sk_socket = NULL;
959 newsk->sk_sleep = NULL;
961 if (newsk->sk_prot->sockets_allocated)
962 atomic_inc(newsk->sk_prot->sockets_allocated);
968 EXPORT_SYMBOL_GPL(sk_clone);
970 void __init sk_init(void)
972 if (num_physpages <= 4096) {
973 sysctl_wmem_max = 32767;
974 sysctl_rmem_max = 32767;
975 sysctl_wmem_default = 32767;
976 sysctl_rmem_default = 32767;
977 } else if (num_physpages >= 131072) {
978 sysctl_wmem_max = 131071;
979 sysctl_rmem_max = 131071;
984 * Simple resource managers for sockets.
989 * Write buffer destructor automatically called from kfree_skb.
991 void sock_wfree(struct sk_buff *skb)
993 struct sock *sk = skb->sk;
995 /* In case it might be waiting for more memory. */
996 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
997 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
998 sk->sk_write_space(sk);
1003 * Read buffer destructor automatically called from kfree_skb.
1005 void sock_rfree(struct sk_buff *skb)
1007 struct sock *sk = skb->sk;
1009 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1013 int sock_i_uid(struct sock *sk)
1017 read_lock(&sk->sk_callback_lock);
1018 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1019 read_unlock(&sk->sk_callback_lock);
1023 unsigned long sock_i_ino(struct sock *sk)
1027 read_lock(&sk->sk_callback_lock);
1028 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1029 read_unlock(&sk->sk_callback_lock);
1034 * Allocate a skb from the socket's send buffer.
1036 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1039 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1040 struct sk_buff * skb = alloc_skb(size, priority);
1042 skb_set_owner_w(skb, sk);
1050 * Allocate a skb from the socket's receive buffer.
1052 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1055 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1056 struct sk_buff *skb = alloc_skb(size, priority);
1058 skb_set_owner_r(skb, sk);
1066 * Allocate a memory block from the socket's option memory buffer.
1068 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1070 if ((unsigned)size <= sysctl_optmem_max &&
1071 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1073 /* First do the add, to avoid the race if kmalloc
1076 atomic_add(size, &sk->sk_omem_alloc);
1077 mem = kmalloc(size, priority);
1080 atomic_sub(size, &sk->sk_omem_alloc);
1086 * Free an option memory block.
1088 void sock_kfree_s(struct sock *sk, void *mem, int size)
1091 atomic_sub(size, &sk->sk_omem_alloc);
1094 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1095 I think, these locks should be removed for datagram sockets.
1097 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1101 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1105 if (signal_pending(current))
1107 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1108 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1109 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1111 if (sk->sk_shutdown & SEND_SHUTDOWN)
1115 timeo = schedule_timeout(timeo);
1117 finish_wait(sk->sk_sleep, &wait);
1123 * Generic send/receive buffer handlers
1126 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1127 unsigned long header_len,
1128 unsigned long data_len,
1129 int noblock, int *errcode)
1131 struct sk_buff *skb;
1136 gfp_mask = sk->sk_allocation;
1137 if (gfp_mask & __GFP_WAIT)
1138 gfp_mask |= __GFP_REPEAT;
1140 timeo = sock_sndtimeo(sk, noblock);
1142 err = sock_error(sk);
1147 if (sk->sk_shutdown & SEND_SHUTDOWN)
1150 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1151 skb = alloc_skb(header_len, gfp_mask);
1156 /* No pages, we're done... */
1160 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1161 skb->truesize += data_len;
1162 skb_shinfo(skb)->nr_frags = npages;
1163 for (i = 0; i < npages; i++) {
1167 page = alloc_pages(sk->sk_allocation, 0);
1170 skb_shinfo(skb)->nr_frags = i;
1175 frag = &skb_shinfo(skb)->frags[i];
1177 frag->page_offset = 0;
1178 frag->size = (data_len >= PAGE_SIZE ?
1181 data_len -= PAGE_SIZE;
1184 /* Full success... */
1190 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1191 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1195 if (signal_pending(current))
1197 timeo = sock_wait_for_wmem(sk, timeo);
1200 skb_set_owner_w(skb, sk);
1204 err = sock_intr_errno(timeo);
1210 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1211 int noblock, int *errcode)
1213 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1216 static void __lock_sock(struct sock *sk)
1221 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1222 TASK_UNINTERRUPTIBLE);
1223 spin_unlock_bh(&sk->sk_lock.slock);
1225 spin_lock_bh(&sk->sk_lock.slock);
1226 if (!sock_owned_by_user(sk))
1229 finish_wait(&sk->sk_lock.wq, &wait);
1232 static void __release_sock(struct sock *sk)
1234 struct sk_buff *skb = sk->sk_backlog.head;
1237 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1241 struct sk_buff *next = skb->next;
1244 sk->sk_backlog_rcv(sk, skb);
1247 * We are in process context here with softirqs
1248 * disabled, use cond_resched_softirq() to preempt.
1249 * This is safe to do because we've taken the backlog
1252 cond_resched_softirq();
1255 } while (skb != NULL);
1258 } while ((skb = sk->sk_backlog.head) != NULL);
1262 * sk_wait_data - wait for data to arrive at sk_receive_queue
1263 * @sk: sock to wait on
1264 * @timeo: for how long
1266 * Now socket state including sk->sk_err is changed only under lock,
1267 * hence we may omit checks after joining wait queue.
1268 * We check receive queue before schedule() only as optimization;
1269 * it is very likely that release_sock() added new data.
1271 int sk_wait_data(struct sock *sk, long *timeo)
1276 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1277 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1278 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1279 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1280 finish_wait(sk->sk_sleep, &wait);
1284 EXPORT_SYMBOL(sk_wait_data);
1287 * Set of default routines for initialising struct proto_ops when
1288 * the protocol does not support a particular function. In certain
1289 * cases where it makes no sense for a protocol to have a "do nothing"
1290 * function, some default processing is provided.
1293 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1298 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1304 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1309 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1314 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1320 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1325 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1330 int sock_no_listen(struct socket *sock, int backlog)
1335 int sock_no_shutdown(struct socket *sock, int how)
1340 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1341 char __user *optval, int optlen)
1346 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1347 char __user *optval, int __user *optlen)
1352 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1358 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1359 size_t len, int flags)
1364 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1366 /* Mirror missing mmap method error code */
1370 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1373 struct msghdr msg = {.msg_flags = flags};
1375 char *kaddr = kmap(page);
1376 iov.iov_base = kaddr + offset;
1378 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1384 * Default Socket Callbacks
1387 static void sock_def_wakeup(struct sock *sk)
1389 read_lock(&sk->sk_callback_lock);
1390 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1391 wake_up_interruptible_all(sk->sk_sleep);
1392 read_unlock(&sk->sk_callback_lock);
1395 static void sock_def_error_report(struct sock *sk)
1397 read_lock(&sk->sk_callback_lock);
1398 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1399 wake_up_interruptible(sk->sk_sleep);
1400 sk_wake_async(sk,0,POLL_ERR);
1401 read_unlock(&sk->sk_callback_lock);
1404 static void sock_def_readable(struct sock *sk, int len)
1406 read_lock(&sk->sk_callback_lock);
1407 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1408 wake_up_interruptible(sk->sk_sleep);
1409 sk_wake_async(sk,1,POLL_IN);
1410 read_unlock(&sk->sk_callback_lock);
1413 static void sock_def_write_space(struct sock *sk)
1415 read_lock(&sk->sk_callback_lock);
1417 /* Do not wake up a writer until he can make "significant"
1420 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1421 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1422 wake_up_interruptible(sk->sk_sleep);
1424 /* Should agree with poll, otherwise some programs break */
1425 if (sock_writeable(sk))
1426 sk_wake_async(sk, 2, POLL_OUT);
1429 read_unlock(&sk->sk_callback_lock);
1432 static void sock_def_destruct(struct sock *sk)
1434 kfree(sk->sk_protinfo);
1437 void sk_send_sigurg(struct sock *sk)
1439 if (sk->sk_socket && sk->sk_socket->file)
1440 if (send_sigurg(&sk->sk_socket->file->f_owner))
1441 sk_wake_async(sk, 3, POLL_PRI);
1444 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1445 unsigned long expires)
1447 if (!mod_timer(timer, expires))
1451 EXPORT_SYMBOL(sk_reset_timer);
1453 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1455 if (timer_pending(timer) && del_timer(timer))
1459 EXPORT_SYMBOL(sk_stop_timer);
1461 void sock_init_data(struct socket *sock, struct sock *sk)
1463 skb_queue_head_init(&sk->sk_receive_queue);
1464 skb_queue_head_init(&sk->sk_write_queue);
1465 skb_queue_head_init(&sk->sk_error_queue);
1466 #ifdef CONFIG_NET_DMA
1467 skb_queue_head_init(&sk->sk_async_wait_queue);
1470 sk->sk_send_head = NULL;
1472 init_timer(&sk->sk_timer);
1474 sk->sk_allocation = GFP_KERNEL;
1475 sk->sk_rcvbuf = sysctl_rmem_default;
1476 sk->sk_sndbuf = sysctl_wmem_default;
1477 sk->sk_state = TCP_CLOSE;
1478 sk->sk_socket = sock;
1480 sock_set_flag(sk, SOCK_ZAPPED);
1483 sk->sk_type = sock->type;
1484 sk->sk_sleep = &sock->wait;
1487 sk->sk_sleep = NULL;
1489 rwlock_init(&sk->sk_dst_lock);
1490 rwlock_init(&sk->sk_callback_lock);
1491 lockdep_set_class(&sk->sk_callback_lock,
1492 af_callback_keys + sk->sk_family);
1494 sk->sk_state_change = sock_def_wakeup;
1495 sk->sk_data_ready = sock_def_readable;
1496 sk->sk_write_space = sock_def_write_space;
1497 sk->sk_error_report = sock_def_error_report;
1498 sk->sk_destruct = sock_def_destruct;
1500 sk->sk_sndmsg_page = NULL;
1501 sk->sk_sndmsg_off = 0;
1503 sk->sk_peercred.pid = 0;
1504 sk->sk_peercred.uid = -1;
1505 sk->sk_peercred.gid = -1;
1506 sk->sk_write_pending = 0;
1507 sk->sk_rcvlowat = 1;
1508 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1509 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1511 sk->sk_stamp = ktime_set(-1L, -1L);
1513 atomic_set(&sk->sk_refcnt, 1);
1516 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1519 spin_lock_bh(&sk->sk_lock.slock);
1520 if (sk->sk_lock.owner)
1522 sk->sk_lock.owner = (void *)1;
1523 spin_unlock(&sk->sk_lock.slock);
1525 * The sk_lock has mutex_lock() semantics here:
1527 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1531 EXPORT_SYMBOL(lock_sock_nested);
1533 void fastcall release_sock(struct sock *sk)
1536 * The sk_lock has mutex_unlock() semantics:
1538 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1540 spin_lock_bh(&sk->sk_lock.slock);
1541 if (sk->sk_backlog.tail)
1543 sk->sk_lock.owner = NULL;
1544 if (waitqueue_active(&sk->sk_lock.wq))
1545 wake_up(&sk->sk_lock.wq);
1546 spin_unlock_bh(&sk->sk_lock.slock);
1548 EXPORT_SYMBOL(release_sock);
1550 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1553 if (!sock_flag(sk, SOCK_TIMESTAMP))
1554 sock_enable_timestamp(sk);
1555 tv = ktime_to_timeval(sk->sk_stamp);
1556 if (tv.tv_sec == -1)
1558 if (tv.tv_sec == 0) {
1559 sk->sk_stamp = ktime_get_real();
1560 tv = ktime_to_timeval(sk->sk_stamp);
1562 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1564 EXPORT_SYMBOL(sock_get_timestamp);
1566 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1569 if (!sock_flag(sk, SOCK_TIMESTAMP))
1570 sock_enable_timestamp(sk);
1571 ts = ktime_to_timespec(sk->sk_stamp);
1572 if (ts.tv_sec == -1)
1574 if (ts.tv_sec == 0) {
1575 sk->sk_stamp = ktime_get_real();
1576 ts = ktime_to_timespec(sk->sk_stamp);
1578 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1580 EXPORT_SYMBOL(sock_get_timestampns);
1582 void sock_enable_timestamp(struct sock *sk)
1584 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1585 sock_set_flag(sk, SOCK_TIMESTAMP);
1586 net_enable_timestamp();
1589 EXPORT_SYMBOL(sock_enable_timestamp);
1592 * Get a socket option on an socket.
1594 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1595 * asynchronous errors should be reported by getsockopt. We assume
1596 * this means if you specify SO_ERROR (otherwise whats the point of it).
1598 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1599 char __user *optval, int __user *optlen)
1601 struct sock *sk = sock->sk;
1603 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1606 EXPORT_SYMBOL(sock_common_getsockopt);
1608 #ifdef CONFIG_COMPAT
1609 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1610 char __user *optval, int __user *optlen)
1612 struct sock *sk = sock->sk;
1614 if (sk->sk_prot->compat_getsockopt != NULL)
1615 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1617 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1619 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1622 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1623 struct msghdr *msg, size_t size, int flags)
1625 struct sock *sk = sock->sk;
1629 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1630 flags & ~MSG_DONTWAIT, &addr_len);
1632 msg->msg_namelen = addr_len;
1636 EXPORT_SYMBOL(sock_common_recvmsg);
1639 * Set socket options on an inet socket.
1641 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1642 char __user *optval, int optlen)
1644 struct sock *sk = sock->sk;
1646 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1649 EXPORT_SYMBOL(sock_common_setsockopt);
1651 #ifdef CONFIG_COMPAT
1652 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1653 char __user *optval, int optlen)
1655 struct sock *sk = sock->sk;
1657 if (sk->sk_prot->compat_setsockopt != NULL)
1658 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1660 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1662 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1665 void sk_common_release(struct sock *sk)
1667 if (sk->sk_prot->destroy)
1668 sk->sk_prot->destroy(sk);
1671 * Observation: when sock_common_release is called, processes have
1672 * no access to socket. But net still has.
1673 * Step one, detach it from networking:
1675 * A. Remove from hash tables.
1678 sk->sk_prot->unhash(sk);
1681 * In this point socket cannot receive new packets, but it is possible
1682 * that some packets are in flight because some CPU runs receiver and
1683 * did hash table lookup before we unhashed socket. They will achieve
1684 * receive queue and will be purged by socket destructor.
1686 * Also we still have packets pending on receive queue and probably,
1687 * our own packets waiting in device queues. sock_destroy will drain
1688 * receive queue, but transmitted packets will delay socket destruction
1689 * until the last reference will be released.
1694 xfrm_sk_free_policy(sk);
1696 sk_refcnt_debug_release(sk);
1700 EXPORT_SYMBOL(sk_common_release);
1702 static DEFINE_RWLOCK(proto_list_lock);
1703 static LIST_HEAD(proto_list);
1705 int proto_register(struct proto *prot, int alloc_slab)
1707 char *request_sock_slab_name = NULL;
1708 char *timewait_sock_slab_name;
1712 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1713 SLAB_HWCACHE_ALIGN, NULL, NULL);
1715 if (prot->slab == NULL) {
1716 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1721 if (prot->rsk_prot != NULL) {
1722 static const char mask[] = "request_sock_%s";
1724 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1725 if (request_sock_slab_name == NULL)
1726 goto out_free_sock_slab;
1728 sprintf(request_sock_slab_name, mask, prot->name);
1729 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1730 prot->rsk_prot->obj_size, 0,
1731 SLAB_HWCACHE_ALIGN, NULL, NULL);
1733 if (prot->rsk_prot->slab == NULL) {
1734 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1736 goto out_free_request_sock_slab_name;
1740 if (prot->twsk_prot != NULL) {
1741 static const char mask[] = "tw_sock_%s";
1743 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1745 if (timewait_sock_slab_name == NULL)
1746 goto out_free_request_sock_slab;
1748 sprintf(timewait_sock_slab_name, mask, prot->name);
1749 prot->twsk_prot->twsk_slab =
1750 kmem_cache_create(timewait_sock_slab_name,
1751 prot->twsk_prot->twsk_obj_size,
1752 0, SLAB_HWCACHE_ALIGN,
1754 if (prot->twsk_prot->twsk_slab == NULL)
1755 goto out_free_timewait_sock_slab_name;
1759 write_lock(&proto_list_lock);
1760 list_add(&prot->node, &proto_list);
1761 write_unlock(&proto_list_lock);
1765 out_free_timewait_sock_slab_name:
1766 kfree(timewait_sock_slab_name);
1767 out_free_request_sock_slab:
1768 if (prot->rsk_prot && prot->rsk_prot->slab) {
1769 kmem_cache_destroy(prot->rsk_prot->slab);
1770 prot->rsk_prot->slab = NULL;
1772 out_free_request_sock_slab_name:
1773 kfree(request_sock_slab_name);
1775 kmem_cache_destroy(prot->slab);
1780 EXPORT_SYMBOL(proto_register);
1782 void proto_unregister(struct proto *prot)
1784 write_lock(&proto_list_lock);
1785 list_del(&prot->node);
1786 write_unlock(&proto_list_lock);
1788 if (prot->slab != NULL) {
1789 kmem_cache_destroy(prot->slab);
1793 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1794 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1796 kmem_cache_destroy(prot->rsk_prot->slab);
1798 prot->rsk_prot->slab = NULL;
1801 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1802 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1804 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1806 prot->twsk_prot->twsk_slab = NULL;
1810 EXPORT_SYMBOL(proto_unregister);
1812 #ifdef CONFIG_PROC_FS
1813 static inline struct proto *__proto_head(void)
1815 return list_entry(proto_list.next, struct proto, node);
1818 static inline struct proto *proto_head(void)
1820 return list_empty(&proto_list) ? NULL : __proto_head();
1823 static inline struct proto *proto_next(struct proto *proto)
1825 return proto->node.next == &proto_list ? NULL :
1826 list_entry(proto->node.next, struct proto, node);
1829 static inline struct proto *proto_get_idx(loff_t pos)
1831 struct proto *proto;
1834 list_for_each_entry(proto, &proto_list, node)
1843 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1845 read_lock(&proto_list_lock);
1846 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1849 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1852 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1855 static void proto_seq_stop(struct seq_file *seq, void *v)
1857 read_unlock(&proto_list_lock);
1860 static char proto_method_implemented(const void *method)
1862 return method == NULL ? 'n' : 'y';
1865 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1867 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1868 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1871 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1872 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1873 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1875 proto->slab == NULL ? "no" : "yes",
1876 module_name(proto->owner),
1877 proto_method_implemented(proto->close),
1878 proto_method_implemented(proto->connect),
1879 proto_method_implemented(proto->disconnect),
1880 proto_method_implemented(proto->accept),
1881 proto_method_implemented(proto->ioctl),
1882 proto_method_implemented(proto->init),
1883 proto_method_implemented(proto->destroy),
1884 proto_method_implemented(proto->shutdown),
1885 proto_method_implemented(proto->setsockopt),
1886 proto_method_implemented(proto->getsockopt),
1887 proto_method_implemented(proto->sendmsg),
1888 proto_method_implemented(proto->recvmsg),
1889 proto_method_implemented(proto->sendpage),
1890 proto_method_implemented(proto->bind),
1891 proto_method_implemented(proto->backlog_rcv),
1892 proto_method_implemented(proto->hash),
1893 proto_method_implemented(proto->unhash),
1894 proto_method_implemented(proto->get_port),
1895 proto_method_implemented(proto->enter_memory_pressure));
1898 static int proto_seq_show(struct seq_file *seq, void *v)
1900 if (v == SEQ_START_TOKEN)
1901 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1910 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1912 proto_seq_printf(seq, v);
1916 static struct seq_operations proto_seq_ops = {
1917 .start = proto_seq_start,
1918 .next = proto_seq_next,
1919 .stop = proto_seq_stop,
1920 .show = proto_seq_show,
1923 static int proto_seq_open(struct inode *inode, struct file *file)
1925 return seq_open(file, &proto_seq_ops);
1928 static const struct file_operations proto_seq_fops = {
1929 .owner = THIS_MODULE,
1930 .open = proto_seq_open,
1932 .llseek = seq_lseek,
1933 .release = seq_release,
1936 static int __init proto_init(void)
1938 /* register /proc/net/protocols */
1939 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1942 subsys_initcall(proto_init);
1944 #endif /* PROC_FS */
1946 EXPORT_SYMBOL(sk_alloc);
1947 EXPORT_SYMBOL(sk_free);
1948 EXPORT_SYMBOL(sk_send_sigurg);
1949 EXPORT_SYMBOL(sock_alloc_send_skb);
1950 EXPORT_SYMBOL(sock_init_data);
1951 EXPORT_SYMBOL(sock_kfree_s);
1952 EXPORT_SYMBOL(sock_kmalloc);
1953 EXPORT_SYMBOL(sock_no_accept);
1954 EXPORT_SYMBOL(sock_no_bind);
1955 EXPORT_SYMBOL(sock_no_connect);
1956 EXPORT_SYMBOL(sock_no_getname);
1957 EXPORT_SYMBOL(sock_no_getsockopt);
1958 EXPORT_SYMBOL(sock_no_ioctl);
1959 EXPORT_SYMBOL(sock_no_listen);
1960 EXPORT_SYMBOL(sock_no_mmap);
1961 EXPORT_SYMBOL(sock_no_poll);
1962 EXPORT_SYMBOL(sock_no_recvmsg);
1963 EXPORT_SYMBOL(sock_no_sendmsg);
1964 EXPORT_SYMBOL(sock_no_sendpage);
1965 EXPORT_SYMBOL(sock_no_setsockopt);
1966 EXPORT_SYMBOL(sock_no_shutdown);
1967 EXPORT_SYMBOL(sock_no_socketpair);
1968 EXPORT_SYMBOL(sock_rfree);
1969 EXPORT_SYMBOL(sock_setsockopt);
1970 EXPORT_SYMBOL(sock_wfree);
1971 EXPORT_SYMBOL(sock_wmalloc);
1972 EXPORT_SYMBOL(sock_i_uid);
1973 EXPORT_SYMBOL(sock_i_ino);
1974 EXPORT_SYMBOL(sysctl_optmem_max);
1975 #ifdef CONFIG_SYSCTL
1976 EXPORT_SYMBOL(sysctl_rmem_max);
1977 EXPORT_SYMBOL(sysctl_wmem_max);